Form Approved
OMB NO. 0704-0188
REPORT DOCUMENTATION PAGE
Public reporting burden for this collection of information is estimated to average 1 hour per response, including the time for reviewino instructions search
exist
data sources
,-T9 a"dJnalnta.'nin9 *• *"« neede* and completing and reviewing thl collection of information. Send 'comment regS^S?s bSrden ^n!S«r Sm Ä«Sct <rf Ws'
9a
uavis Highway, buite 1ZQ4, Arlington, VA 22202-4302, and to the Office of Management and Budget, Paperwork Reduction Project (0704-0188), Washington, DC 20503.
1. AGENCY USE ONLY (Leave blank)
2. REPORT DATE
3. REPORT TYPE AND DATES COVERED
Book
4. TITLE AND SUBTITLE
Wide-Bändgap Semiconductors for High-Power, High-Frequency and
High-Temperature Applications — 1999
MRS Symposium Proceedings Volume 572
6. AUTHOR(S)
5. FUNDING NUMBERS
DAAD 19-99-1-0147
Steven C. Binari, Albert A. Burk, Michael R. Melloch, Chanh Nguyen
(editors)
° J
7. PERFORMING ORGANIZATION NAMES(S) AND ADDRESS(ES)
PERFORMING ORGANIZATION
REPORT NUMBER
Materials Research Society
Warrendale, PA 15086
9.
SPONSORING / MONITORING AGENCY NAME(S) AND ADDRESS(ES)
10. SPONSORING / MONITORING
AGENCY REPORT NUMBER
U.S. Army Research Office
P.O. Box 12211
Research Triangle Park, NC 27709-2211
ARO 39847.2-MS-CF
11. SUPPLEMENTARY NOTES
The views, opinions and/or findings contained in this report are those of the author(s) and should not be construed as
an official Department of the Army position, policy or decision, unless so designated by other documentation.
12a. DISTRIBUTION / AVAILABILITY STATEMENT
12 b. DISTRIBUTION CODE
Approved for public release; distribution unlimited.
13. ABSTRACT (Maximum 200 words)
ABSTRACT IN PREFACE
14. SUBJECT TERMS
15. NUMBER IF PAGES
16. PRICE CODE
17. SECURITY CLASSIFICATION
OR REPORT
18. SECURITY CLASSIFICATION
OF THIS PAGE
19. SECURITY CLASSIFICATION
OF ABSTRACT
UNCLASSIFIED
UNCLASSIFIED
UNCLASSIFIED
NSN 7540-01-280-5500
20. LIMITATION OF ABSTRACT
UL
Standard Form 298 (Rev. 2-89)
Prescribed by ANSI Std. 239-18
298-102
"
;XJl.:U;iT
'
mmmm
r>YMi\j:MM m mm HNGS
Volume 572
^ij-.'ie^rtV '
Wide-Bandgap
Semiconductors
for High-Power,
High-Frequency and
High-Temperature
Applications—1999
p
r
~'%
»
St:
lit
EDITORS
Steven C. Binari
Wi
"'IS
I w
1«
Albert A. Burk
.i 9
|
i •
Michael R. Melloch
1
Chanh Mguyen
f
mM
;
afc;';^
- ".
*. . ,J»)
J
" v, '
.yk
'fe'^SJSS
-.
US
!:«S-*i
r-;^5*r^;
Wide-Bandgap Semiconductors
for High-Power, High-Frequency
and High-Temperature
Applications—1999
19991101 134
Preceding PageJ Blank
MATERIALS RESEARCH SOCIETY
SYMPOSRJM PROCEEDINGS VOLUME 572
Wide-Bandgap Semiconductors
for High-Power, High-Frequency
and High-Temperature
Applications—1999
Symposium held April 5-8,1999, San Francisco, California, U.S.A.
EDITORS:
Steven C. Binari
Naval Research Laboratory
Washington, D.C., U.S.A.
Albert A. Burk
Cree Research Inc.
Durham, North Carolina, U.S.A.
Michael R. Melloch
Purdue University
West Lafayette, Indiana, U.S.A.
Chanh Nguyen
HRL Laboratories
Malibu, California, U.S.A.
iMlRlsl
Materials Research Society
Warrendale, Pennsylvania
This work was supported in part by the Army Research Office under Qrant Number
ARO: DAAD19-99-1-0147. The views, opinions, and/or findings contained in this report
are those of the author(s) and should not be construed as an official Department of the
Army position, policy, or decision, unless so designated by other documentation.
Single article reprints from this publication are available through
University Microfilms Inc., 300 North Zeeb Road, Ann Arbor, Michigan 48106
CODEN: MRSPDH
Copyright 1999 by Materials Research Society.
All rights reserved.
This book has been registered with Copyright Clearance Center, Inc. Tor further
information, please contact the Copyright Clearance Center, Salem, Massachusetts.
Published by:
Materials Research Society
506 Keystone Drive
Warrendale, PA 15086
Telephone (724) 779-3003
Fax (724) 779-8313
Web site: http://www.mrs.org/
Library of Congress Cataloging-in-Publication Data
Wide bandgap semiconductors for high-power, high-frequency and high-temperature
applications—1999 : symposium held April 5-8, 1999, San Francisco, California,
U.S.A. / editors, Steven C. Binari, Albert A. Burk, Michael R. Melloch, Chanh Nguyen
p.cm.—(Materials Research Society symposium proceedings,
ISSN 0272-9172 ; v. 572)
Includes bibliographical references and index.
ISBN 1-55899-479-3
1. Wide gap semiconductors—Congresses. 2. Silicon carbide—Congresses.
3. Gallium nitride—Congresses. I. Binari, Steven C. II. Burk, Albert A.
III. Melloch, Michael R. IV. Nguyen, Chanh V. Series: Materials Research Society
symposium proceedings ; v. 572
TK7871.85.W52 1999
99-40454
621.3815—dc21
CIP
Manufactured in the United States of America
CONTENTS
Preface
xiii
Materials Research Society Symposium Proceedings
xiv
PART I: SiC DEVICES AND PROCESSING
*SiC Power Electronic Devices, MOSFETs and Rectifiers
JA. Cooper, S-H. Ryu, Y. Li, M. Matin, J. Spitz, D.T. Morisette,
HM. McGlothlin, M.K. Das, M.R. Melloch, MA. Capano,
andJ.N. Vloodall
• Recent Progress in SiC Microwave MESFETs
S.T. Mien, S.T. Sheppard, W.L. Pribble, RA. Sadler,
T.S. Alcom, Z. Ring, and J.W. Palmour
3
15
'Current Status of SiC Power Switching Devices: Diodes
and GTOs
S. Seshadri, Aft. Agarwal, W.B. Hall, S.S. Mani, M.F. MacMillan,
R. Rodrigues, T. Hanson, S. Khatri, and PA. Sanger
23
The Effects of Damage on Hydrogen-Implant-Induced
Thin-Film Separation From Bulk Silicon Carbide
R.B. Gregory, O.W. Holland, D.ft. Thomas, TA. Wetteroth,
and S.R. Wilson
33
Characterization of Si02/SiC Samples Using Photoelectron
Spectroscopy
L.I. Johansson, P-A. Glans, Q. Wahab, T.M. Grehk,
Th. Eickhoff, and W. Drube
39
Annealing of Ion Implantation Damage in SiC Using a
Graphite Mask
C. Thomas, C. Taylor. J. Griffin, W.L. Rose, M.G. Spencer,
M. Capano, S. Rendakova, and K. Komegay
45
Effect of Varying Oxidation Parameters on the Generation
of C-Dangling Bond Centers in Oxidized SiC
PJ. Macfarlane and M.E. Zvanut
51
Thick Oxide Layers on N and P SiC Wafers by a
Depo-Conversion Technique
Q. Zhang, V. Madangarli, I. Khlebnikov, S. Soloviev,
and T.S. Sudarshan
57
Bias-Temperature-Stress Induced Mobility Improvement
in 4H-SiC MOSFETs
ft. Chatty, T.P. Chow, RJ. Gutmann, E. Arnold, and D. Alok
63
Invited Paper
Full Band Monte Carlo Simulation of Short Channel MOSFETs
in 4H and 6H-SiC
69
M. HJelm, li-E. liilsson, E. Dubaric, C. Persson, P. Käckell,
and CS. Petersson
High Voltage Schottky Barrier Diodes on p-Type SiC Using
Metal-Overlap on a Thick Oxide Layer as Edge Termination
75
Q. Zhang, V. Madangarli, S. Soloviev, and T.S. Sudarshan
High Voltage P-N Junction Diodes in Silicon Carbide Using
Field Plate Edge Termination
81
R.K. Chilukuri, P. Ananthanarayanan, V. Nagapudi, and
BJ. Baliga
Carbon and Silicon Related Surface Compounds of Palladium
Ultrathin Films on SiC After Different Annealing Temperatures
87
WJ. Lu, D.T. Shi, T. Crenshaw, A. Burger, and W.E. Collins
A Materials Investigation of Nickel Based Contacts to
n-SiC Subjected to Operational Thermal Stresses
Characteristic of High Power Switching
93
M.W. Cole, C.W. tlubbard, CO. Fountzoulas, DJ. Demaree,
and F. Ren
Preparation of Conductive Tungsten Carbide Layers for SiC
High-Temperature Applications
99
tl. Romanus, V. Cimalla, S.I. Ahmed, JA. Schaefer, O. Ecke,
R. Avci, and L. Spiess
A Formation of Si02/4H-SiC Interface by Oxidizing Deposited
Poly-Si and High-Temperature Hydrogen Annealing
105
H. Fukuda, K. Sakamoto, K. Nagai, T. Sekigawa, S. Yoshida,
and K. Aral
High-Temperature Stable WSi2-Contacts on
p-6H-Silicon Carbide
Ill
F. Erler, H. Romanus, J.K.N. Lindner, and L. Spiess
Structural and Electrical Properties of Beryllium Implanted
Silicon Carbide
117
T. Henkel, Y. Tanaka, N. Kobayashi, H. Tanoue, M. Gong,
X.D. Chen, S. Fung, and CD. Beling
Elevated Temperature Silicon Carbide Chemical Sensors
123
MA. George, MA. Ayoub, D. Ha, and DJ. Larkin
The Effect of Annealing on Argon Implanted Edge Terminations
for 4H-SiC Schottky Diodes
129
A.P. Knights, DJ. Morrison, N.G. Wright, CM. Johnson,
A.G. O'Tieill, S. Ortolland, K.P. Homewood, MA. Lourenco,
R.M. Gwilliam, and P.G. Coleman
Oxidation Modeling for SiC
135
n.G. Wright, CM. Johnson, andAG. O'lieill
VI
Annealing Effects of Schottky Contacts on the Characteristics
of 4H-SiC Schottky Barrier Diodes
S.C. /fang, B.ti. Kum, SJ. Do, J.H. Je, and M.W. Shin
141
PART II: SiC EPITAXY AND CHARACTERIZATION
'Epitaxial Growth of SiC in a Vertical Multi-Wafer CVD System:
Already Suited as Production Process?
R. Rupp, C. Hecht, A. Wiedenhofer, and D. Stephani
149
Multi-Wafer VPE Growth of Highly Uniform SiC Epitaxial Layers
MJ. O'Loughlin, li.D. Tiordby, Jr., and A A. Burk, Jr.
161
Characterization of Thick 4H-SiC Hot-Wall CVD Layers
MJ. Paisley, K.G. Irvine, O. Kordina, R. Singh, J.W. Palmow,
and C.H. Carter, Jr.
167
Homo-Epitaxial and Selective Area Growth of 4H and 6H
Silicon Carbide Using a Resistively Heated Vertical Reactor
E. Eshun, C. Taylor, M.O. Spencer, K. Komegay, I. Ferguson,
A. Qurray, and R. Stall
173
Properties of 4H-SiC by Sublimation Close Space Technique
S. Nishino, K. Matsumoto, Y. Chen, and Y. Nishio
179
Effect of Ge on SiC Film Morphology in SiC/Si Films Grown
by MOCVD
Vl.L. Sarney, L. Salamanca-Riba, P. Zhou, M.O. Spencer,
C. Taylor, R.P. Sharma, and KA. Jones
185
Properties of Heteroepitaxial 3C-SiC Layer on Si Using
Si2(CH3)6 by CVD
191
Characterization of p-Type Buffer Layers for SiC Microwave
Device Applications
A.O. Konstantinov, S. Karlsson, P-Ä. Nilsson, A-M. Saroukhan,
J-O. Svedberg, N. Nordell, C.I. Harris, J. Eriksson, and N. Rorsman
197
Optical Characterization of SiC Wafers
J.C. Burton, M. Pophristic, F.H. Long, and I. Ferguson
201
Growth of SiC Thin Films on (100) and (111) Silicon by
Pulsed Laser Deposition Combined With a Vacuum
Annealing Process
J. Huang, L. Wang, J. Wen, Y. Wang, C. Lin, C-M. Zetterling,
and M. Östling
207
On the Role of Foreign Atoms in the Optimization of 3C-SiC/Si
Heterointerfaces
P. Masri, N. Moreaud, M. Averous, Th. Stauden, T. Wöhner,
andJ. Pezoldt
213
Y. Chen, Y. Masuda, Y. Nishio, K. Matsumoto, and S. Nishino
"Invited Paper
vii
3C-SiC Buffer Layers Converted From Si at a Low Temperature
219
H.M. Liaw, S.Q. Hong, P. Fejes, D. Werho, H. Tompkins,
S. Zollner, S.R. Wilson, KJ. Linthicum, and R.F. Davis
Time Resolved Photoluminescence of Cubic Mg Doped GaN
225
R. Seitz, C. Oaspar, T. Monteiro, E. Pereira, B. Schoettker,
T. Frey, DJ. As, D. Schikora, and K. Lischka
Dielectric Function of AIN Grown on Si (111) by MBE
231
S. Zollner, A. Konkar, R.B. Gregory, S.R. Wilson, SA. Hikishin,
and H. Temkin
The Comparative Studies of Chemical Vapor Deposition
Grown Epitaxial Layers and of Sublimation Sandwich Method
Grown 4H-SiC Samples
237
A.O. Evwaraye, S.R. Smith, and W.C. Mitchel
PART III: SiC BULK GROWTH AND CHARACTERIZATION
• ImpurityEffects in the Growth of 4H-SiC Crystals by
Physical Vapor Transport
245
V. Balakrishna, G. Augustine, and R.ti. Hopkins
Characterization of Vanadium-Doped 4H-SiC Using Optical
Admittance Spectroscopy
253
S.R. Smith, A.O. Evwaraye, W.C. Mitchel, J.S. Solomon,
andJ. Goldstein
On-Line Monitoring of PVT SiC Bulk Crystal Growth Using
Digital X-ray Imaging
259
PJ. Weltmann, M. Bickermann, M. Grau, D. Hofmann,
T.L. Straubinger, and A. Winnacker
Polytype Stability and Defect Reduction in 4H-SiC Crystals
Grown Via Sublimation Technique
265
7?. Yakimova, T. Iakimov, M. Syväjärvi, H. Jacobsson,
P. Räback, A. Vehanen, and E. Janz6n
Growth and Characterization of 2" 6H-Silicon Carbide
271
E. Schmitt, R. Eckstein, and M. Kölbl
Experimental and Theoretical Analysis of the Hall-Mobility
in n-Type Bulk 6H- and 4H-SiC
275
St.G. Müller, D. Hofmann, and A. Winnacker
Mid-Infrared Photoconductivity Spectra of Donor Impurities
in Hexagonal Silicon Carbide
R J. Linville, GJ. Brown, W.C. Mitchel, A. Saxler, and R. Perrin
Invited Paper
viii
281
PART IV: GaN GROWTH AND CHARACTERIZATION
The Influence of the Sapphire Substrate on the Temperature
Dependence of the GaN Bandgap
J. Krüger, li. Shapiro, S. Subramanya, Y. Kim, ti. Siegle,
P. Perlin, E.R. Weber, W.S. Wong, T. Sands, li.W. Cheung,
and RJ. Molnar
289
Effect of N/Ga Flux Ratio in GaN Buffer Layer Growth by MBE
on (0001) Sapphire on Defect Formation in the GaN Main Layer
S. Ruvimov, Z. Liliental-Weber, J. Washburn, Y. Kim,
Q.S. Sudhir, J. Krueger, and E.R. Weber
295
Enhanced Optical Emission From GaN Film Grown on
Composite Intermediate Layers
X. Zhang, S-J. Chua, P. Li, and K-B. Chong
301
Pendeo-Epitaxial Growth of GaN on SiC and Silicon Substrates
Via Metalorganic Chemical Vapor Deposition
KJ. Linthicum, T. Qehrke, D. Thomson, C. Ronning, E.P. Carlson,
CA. Zorman, M. Nehregany, and R.F. Davis
307
Maskless Lateral Epitaxial Overgrowth of GaN on Sapphire
P. Pint, H. Marchand, J.P. Ibbetson, B. Moran, L. Zhao,
S.P. DenBaars, J.S. Speck, and U.K. Mishra
315
Reproducibility and Uniformity of MOVPE Planetary Reactors®
for the Growth of GaN Based Materials
M. tieuken. It. Protzmann, O. Schoen, M. Luenenbuerger,
H. Juergensen, M. Bremser, and E. Woelk
321
Synchrotron X-ray Topography Studies of Epitaxial Lateral
Overgrowth of GaN on Sapphire
PJ. McTially, T. Tuomi, R. Rantamäki, K. Jacobs, L. Considine,
M. O'Hare, D. Lowney, and A.n. Danilewsky
327
Conducting (Si-Doped) Aluminum Nitride Epitaxial Films
Grown by Molecular Beam Epitaxy
J.O. Kim, M. Moorthy, and R.M. Park
333
Investigation of the Morphology of AIN Films Grown on
Sapphire by MOCVD Using Transmission Electron Microscopy
W.L. Sarney, L. Salamanca-Riba, P. Zhou, S. Wilson,
M.O. Spencer, and KA. Jones
339
Temperature Dependent Morphology Transition of GaN Films
A.RA. Zauner, F.K. De Theije, P.R. Hageman, WJ.P. Van Enckevort,
JJ Schermer, and P.K. Larsen
Comparative Study of Emission From Highly Excited (In, Al)
GaN Thin Films and Heterostructures
B.D. Little, S. Bidnyk, TJ. Schmidt, J.B. Lam, Y.ti. Kwon,
JJ. Song, S. Keller, U.K. Mishra, S.P. DenBaars, and W. Yang
345
351
Atomic Scale Analysis of InGaN Multi-Quantum Wells
357
M. Benamara, Z. Liliental-Weber, W. Swider, J. Washbum,
R.D. Dupuis, PA. Qrudowski, CJ. Eiting, J.W. Yang, and MA. Khan
TEM Study of Mg-Doped Bulk GaN Crystals
363
Z. Liliental-Weber, M. Benamara, S. Ruvimov, J.ti. Mazur,
J. Washbum, I. Orzegory, and S. Porowski
Deformation-Induced Dislocations in 4H-SiC and GaN
369
M.tl. Hong, A.V. Samant, V. Orlov, B. Färber, C. Kisielowski,
and P. Pirouz
Ca Dopant Site Within Ion Implanted GaN Lattice
377
H. Kobayashi and W.M. Gibson
Growth and Characterization of InGaN/GaN Heterostructures
Using Plasma-Assisted Molecular Beam Epitaxy
383
K.H. Shim, S.E. Hong, K.H. Kim, M.C Paek, and K.I. Cho
Piezoelectric Coefficients of Aluminum Nitride and
Gallium Nitride
389
CM. Lueng, H.L.W. Chan, W.K. Fong, C. Surya, and C.L. Choy
Fast and Slow UV-Phoforesponse in n-Type GaN
395
R. Rocha, S. Koynov, P. Brogueira, R. Schwarz, V. Chu,
M. Topf, D. Meister, and B.K. Meyer
Epitaxial Growth of GaN Thin Films Using a Hybrid Pulsed
Laser Deposition System
401
P. M6rel, M. Chaker, H. Pepin, and M. Tabbal
Epitaxial Growth of AIN on Si Substrates With Intermediate
3C-SiC as Buffer Layers
407
S.Q. Hong, H.M. Liaw, K. Linthicum, R.F. Davis, P. Fejes,
S. Zollner, M. Kottke, and S.R. Wilson
SIMS and CL Characterization of Manganese-Doped Aluminum
Nitride Films
413
R.C. Tucceri, CD. Bland, M.L. Caldwell, M.H. Ervin,
ti.P. Magtoto, CM. Spalding, MA. Wood, and H.H. Richardson
Photoluminescence Between 3.36 eV and 3.41 eV From
GaN Epitaxial Layers
419
R. Seitz, C. Oaspar, T. Monteiro, E. Perexra, MA. Poisson,
and B. Beaumont
Disorder Induced IR Anomaly in Hexagonal AIGaN Short-Period
Superlattices and Alloys
427
A.M. Mintairov, A.S. Vlasov, J.L. Merz, D. Korakakis,
T.D. Moustakas, A.O. Osinsky, R. Gaska, and M.B. Smirnov
Nondegenerate Optical Pump-Probe Spectroscopy of Highly
Excited Group III Nitrides
TJ. Schmidt, JJ. Song, S. Keller, U.K. Mishra, S.P. DenBaars,
and W. Yang
433
Study of Near-Threshold Gain Mechanisms in MOCVD-Grown
GaN Epilayers and InGaN/GaN Heterostructures
S. Bidnyk, TJ. Schmidt, B.D. Little, andJJ. Song
439
Electron Transport in the lll-V Nitride Alloys
B.E. Foutz, S.K. O'Leary, M.S. Shur, and L.F. Eastman
445
High-Quality GaN Grown by Molecular Beam Epitaxy
on Ge(OOl)
ti. Siegle, Y. Kim, Q.S. Sudhir, J. Krüger, F. Perlin, J.W. Ager III,
C. Kisielowski, and E.R. Weber
451
Carrier Recombination Dynamics of AlxGai.xN Epilayers
Grown by MOCVD
Y-fl. Cho, G.H. Gainer, J.B. ham, JJ. Song, W. Yang, and
SA. McPherson
457
Comparative Study of GaN Growth Process by MOVPE
J. Sun, J.M. Redwing, and T.F. Kuech
463
PART V: GaN DEVICES AND PROCESSING
*AIGaN Microwave Power HFETs on Insulating SiC Substrates
G. Sullivan, E. Gertner, R. Pittman, M. Chen, R. Pierson, A. Higgins,
Q. Chen, J-W. Yang, R.P. Smith, R. Perez, A. Khan, J. Redwing,
B. McDermott, and K. Boutros
471
Recessed Gate GaN MESFETs Fabricated by the
Photoelectrochemical Etching Process
W.S. Lee, Y.H. Choi, K.W. Chung, M.W. Shin, and D.C. Moon
481
Current-Voltage Characteristics of Ungated AIGaN/GaN
Heterostructures
J.D. Albrecht, P.P. Ruden, S.C. Binari, K. Ikossi-Anastasiou,
M.G. Ancona, R.L. Henry, D.D. Koleske, andA.E. Wickenden
489
Hydrostatic and Uniaxial Stress Dependence and Photo-Induced
Effects on the Channel Conductance of n-AIGaN/GaN
Heterostructures Grown on Sapphire Substrates
A.K. Fung, C. Cat, P.P. Ruden, M.I. Nathan, M.Y. Chen, B.T. McDermott,
QJ. Sullivan, J.M. Van Hove, K. Boutros, J. Redwing, J.W. Yang,
Q. Chen, MA. Khan, W. Schaff, and M. Murphy
495
The Influence of Spontaneous and Piezoelectric
Polarization on Novel AIGaN/GaN/lnGaN Device
Structures
B.E. Foutz, MJ. Murphy, O. Ambacher, V. Tilak, JA. Smart,
J.R. Shealy, WJ. Schaff, and L.F. Eastman
501
Piezoelectric Scattering in Large-Bandgap Semiconductors
and Low-Dimensional Heterostructures
B.K. Ridley, NA. Zakhleniuk, C.R. Bennett, M. Babiker,
and D.R. Anderson
507
Invited Paper
Activation Characteristics of Donor and Acceptor Implants
in GaN
XA. Cao, SJ. Pearton, R.K. Singh, R.Q. Wilson, JA. Sekhar,
J.C. Zolper, J. Han, DJ. Rieger, RJ. Shul, HJ. Quo, SJ. Permycook,
and J.M. Zavada
513
Transmutation Doping of Ill-Nitrides
G. Popovici
519
High Barrier Height n-GaN Schottky Diodes With a Barrier
Height of 1.3 eV by Using Sputtered Copper Metal
W.C. Lai, M. Yokoyama, C.Y. Chang, J.D. Quo, J.S. Tsang,
S.H. Chan, and S.M. Sze
523
IIIB-Nitride Semiconductors for High-Temperature Electronic
Applications
X. Bai, D.M. Hill, and M.E. Kordesch
529
Photo-Assisted RIE of GaN in BCI3/CI2/N2
Ti. Medelci, A. Tempez, I. Berishev, D. Starikov, and A. Bensaoula
535
Correlation of Drain Current Pulsed Response With Microwave
Power Output in AIGaN/GaN HEMTs
S.C. Binari, K. Ikossi-Anastasiou, W. Kruppa, H.B. Dietrich,
O. Keiner, R.L. Henry, D.D. Koleske, andA.E. Wickenden
541
Photoionization Spectra of Traps Responsible for Current
Collapse in GaN MESFETs
P.B. Klein, JA. Freitas, Jr., and S.C. Binari
547
Author Index
553
Subject Index
557
XII
PREFACE
The introduction of the SiC substrate and the demonstration of bright III-N
light-emitting diodes were catalysts for a large increase in research and
development of wide-bandgap semiconductor materials and devices during the
nineties. This symposium, "Wide-Bandgap Semiconductors for High-Power, HighFrequency and High-Temperature Applications—1999," from the 1999 MRS Spring
Meeting in San Francisco, California, focused on high-power, high-frequency and
high-temperature applications of these wide-bandgap semiconductors. The
symposium attracted a wide range of researchers who presented 120 papers in
nine different sessions on topics such as bulk crystal growth, epitaxy, materials
characterization, processing, and devices.
We would like to thank our invited speakers, J.A. Cooper, S.T. Allen, R. Rupp,
V. Balakrishna, S. Seshadri, J.M. Redwing, and G. Sullivan, and the many
dedicated anonymous reviewers who made this publication possible.
Generous financial support from the Air Force Research Laboratories, Army
Research Office, DARPA, and ODDR&E(R) contributed to the success of this
symposium.
Steven C. Binari
Albert A. Burk
Michael R. Melloch
Chanh Nguyen
xii!
MATERIALS RESEARCH SOCIETY SYMPOSIUM PROCEEDINGS
Volume 535— III-V and IV-IV Materials and Processing Challenges for Highly Integrated
Microelectonics and Optoelectronics, S.A. Ringel, E.A. Fitzgerald, I. Adesida,
D. Houghton, 1999, ISBN: 1-55899-441-6
Volume 536— Microcrystalline and Nanocrystalline Semiconductors—1998, L.T. Canham,
M.J. Sailor, K. Tanaka, C-C. Tsai, 1999, ISBN: 1-55899-442-4
Volume 537— QaN and Related Alloys, S.J. Pearton, C. Kuo, A.F. Wright, T. Uenoyama, 1999,
ISBN: 1-55899-443-2
Volume 538— Multiscale Modelling of Materials, V.V. Bulatov, T. Diaz de la Rubia, R. Phillips,
E. Kaxiras, N. Qhoniem, 1999, ISBN: 1-55899-444-0
Volume 539— Fracture and Ductile vs. Brittle Behavior—Theory, Modelling and
Experiment, Q.E. Beltz, R.L. Blumberg Selinger, K-S. Kim, M.P. Marder, 1999,
ISBN: 1-55899-445-9
Volume 540— Microstructural Processes in Irradiated Materials, S.J. Zinkle, Q.E. Lucas,
R.C. Ewing, J.S. Williams, 1999, ISBN: 1-55899-446-7
Volume 541— Ferroelectric Thin Films VII, R.E. Jones, R.W. Schwartz, S.R. Summerfeit, I.K. Yoo,
1999, ISBN: 1-55899-447-5
Volume 542— Solid Freeform and Additive Fabrication, D. Dimos, S.C. Danforth, M.J. Cima, 1999,
ISBN: 1-55899-448-3
Volume 543— Dynamics in Small Confining Systems IV, J.M. Drake, Q.S. Qrest, J. Klafter,
R. Kopelman, 1999, ISBN: 1-55899-449-1
Volume 544— Plasma Deposition and Treatment of Polymers, W.W. Lee, R. d'Agostino,
M.R. Wertheimer, B.D. Ratner, 1999, ISBN: 1-55899-450-5
Volume 545— Thermoelectric Materials 1998—The Next Generation Materials for Small-Scale
Refrigeration and Power Generation Applications, T.M. Tritt, M.Q. Kanatzidis,
Q.D. Mahan, H.B. Lyon, Jr., 1999, ISBN: 1-55899-451-3
Volume 546— Materials Science of Microelectromechanical Systems (MEMS) Devices,
A.H. Heuer, S.J. Jacobs, 1999, ISBN: 1-55899-452-1
Volume 547— Solid-State Chemistry of Inorganic Materials II, S.M. Kauzlarich,
E.M. McCarron III, A.W. Sleight, H-C. zur Loye, 1999, ISBN: 1-55899-453-X
Volume 548— Solid-State Ionics V, Q-A. Nazri, C. Julien, A. Rougier, 1999,
ISBN: 1-55899-454-8
Volume 549— Advanced Catalytic Materials—1998, P.W. Lednor, D.A. Nagaki, L.T. Thompson,
1999, ISBN: 1-55899-455-6
Volume 550— Biomedical Materials—Drug Delivery, Implants and Tissue Engineering,
T. Neenan, M. Marcolongo, R.F. Valentini, 1999, ISBN: 1-55899-456-4
Volume 551— Materials in Space—Science, Technology and Exploration, A.F. Hepp,
J.M. Prahl, T.Q. Keith, S.O. Bailey, J.R. Fowler, 1999, ISBN: 1-55899-457-2
Volume 552— High-Temperature Ordered Intermetallic Alloys VIII, E.P. Qeorge, M.J. Mills,
M. Yamaguchi, 1999, ISBN: 1-55899-458-0
Volume 553— Quasicrystals, J-M. Dubois, P.A. Thiel, A-P. Tsai, K. Urban, 1999,
ISBN: 1-55899-459-9
Volume 554— Bulk Metallic Glasses, W.L. Johnson, C.T. Liu, A. Inoue, 1999,
ISBN: 1-55899-460-2
Volume 555— Properties and Processing of Vapor-Deposited Coatings, R.N. Johnson,
W.Y. Lee, M.A. Pickering, B.W. Sheldon, 1999, ISBN: 1-55899-461-0
Volume 556— Scientific Basis for Nuclear Waste Management XXII, D.J. Wronkiewicz,
J.H. Lee, 1999, ISBN: 1-55899-462-9
Volume 557— Amorphous and Heterogeneous Silicon Thin Films: Fundamentals to
Devices—1999, H.M. Branz, R.W. Collins, H. Okamoto, S. Quha, R. Schropp,
1999, ISBN: 1-55899-464-5
MATERIALS RESEARCH SOCIETY SYMPOSIUM PROCEEDINGS
Volume 558— Flat-Panel Displays and Sensors—Principles, Materials and Processes, F.R. Libsch,
B. Chalamala, R. Friend, T. Jackson, H. Ohshima, 1999, ISBN: 1-55899-465-3
Volume 559— Liquid Crystal Materials and Devices, T.J. Bunning, S.H. Chen, L.C. Chien,
S-C.A. Lien, T. Kajiyama, N. Koide, 1999, ISBN: 1-55899-466-1
Volume 560— Luminescent Materials, J. McKittrick, A. Kitai, K. Mishra, C. Ronda, B. DiBartolo,
1999, ISBN: 1-55899-467-X
Volume 561— Organic nonlinear Optical Materials and Devices, B. Kippelen, H. Lackritz,
R. Claus, 1999, ISBN: 1-55899-468-8
Volume 562— Polycrystalline Metal and Magnetic Thin Films, K.P. Rodbell, D. Laughlin,
O. Thomas, B. Zhang, 1999, ISBN: 1-55899-469-6
Volume 563— Materials Reliability in Microelectronics IX, C.A. Volkert, A.H. Verbruggen, D. Brown,
1999, ISBN: 1-55899-470-X
Volume 564— Advanced Interconnects and Contacts, D.C. Edelstein, T. Kikkawa, M. Özturk,
K-N. Tu, E. Weitzman, 1999, ISBN: 1-55899-471-8
Volume 565— Low-Dielectric Constant Materials V, J.P. Hummel, K. Endo, W.W. Lee, M.E. Mills,
S-Q. Wang, 1999, ISBN: 1-55899-472-6
Volume 566— Chemical-Mechanical Polishing—Fundamentals and Challenges, S.V. Babu,
S. Danyluk, M.I. Krishnan, M. Tsujimura, 1999, ISBN: 1-55899-473-4
Volume 567— Ultrathin Si02 and High-K Materials for ULSI Gate Dielectrics, H.R. Huff, M.L. Green,
T. Hattori, Q. Lucovsky, CA. Richter, 1999, ISBN: 1-55899-474-2
Volume 568— Silicon Front-End Processing—Physics and Technology of Dopant-Defect
Interactions, H-J.L. Qossmann, T.E. Haynes, M.E. Law, A.N. Larsen, S. Odanaka,
1999, ISBN: 1-55899-475-0
Volume 569— In Situ Process Diagnostics and Modelling, O. Auciello, A.R. Krauss, E.A. Irene,
J.A. Schultz, 1999, ISBN: 1-55899-476-9
Volume 570— Epitaxial Qrowth, T.P. Pearsall, A-L. Barabasi, F. Liu, Q.N. Maracas, 1999,
ISBN: 1-55899-477-7
Volume 571— Semiconductor Quantum Dots, D. Ila, H. Lee, S. Moss, D. Norris, 1999,
ISBN: 1-55899-478-5
Volume 572— Wide-Bandgap Semiconductors for High-Power, High-Frequency and
High-Temperature Applications—1999, S. Binari, A. Burk, M. Melloch,
C. Nguyen, 1999, ISBN: 1-55899-479-3
Volume 573— Compound Semiconductor Surface Passivation and Novel Device Processing,
H. Hasegawa, M. Hong, Z.H. Lu, S.J. Pearton, 1999, ISBN: 1-55899-480-7
Volume 574— Multicomponent Oxide Films for Electronics, M.E. Hawley, D.H. Blank, C-B. Eom,
S.K. Streiffer, D.O. Schlom, 1999, ISBN: 1-55899-481-5
Volume 575— New Materials for Batteries and Fuel Cells, D.H. Doughty, H-P. Brack, K. Naoi,
L.F. Nazar, 1999, ISBN: 1-55899-482-3
Volume 576— Organic/Inorganic Hybrid Materials II, L.C. Klein, L. Francis, M.R. DeQuire,
J.E. Mark, 1999, ISBN: 1-55899-483-1
Volume 577— Advanced Hard and Soft Magnetic Materials, L.H. Lewis, J.M.D. Coey, B-M. Ma,
T. Schrefl, L. Schultz, M.E. McHenry, V.Q. Harris, J. Fidler, R. Hasegawa, A. Inoue,
1999, ISBN: 1-55899-485-8
Prior Materials Research Society Symposium Proceedings available by contacting Materials Research Society
Part I
SiC Devices and
Processing
SiC Power Electronic Devices, MOSFETs arid Rectifiers
J. A. COOPER, S-H. RYU, Y. LI, M. MATIN, J. SPITZ, D. T. MORISETTE, H. M.
McGLOTHLIN, M. K. DAS, M. R. MELLOCH, M. A. CAPANO, AND J. M.
WOODALL
School of Electrical and Computer Engineering, Purdue University, West Lafayette, IN
cooperj @ecn.purdue.edu
ABSTRACT
SiC power switching devices have demonstrated performance figures of merit far
beyond the silicon theoretical limits, but much work remains before commercial
production will be feasible. A significant fraction of the remaining work centers on
materials science issues. This paper reviews the current status of unipolar power
switching devices in SiC and identifies the major technological and material science
barriers that need to be overcome.
1. INTRODUCTION
Since the development of the modified sublimation process for growth of SiC
crystals during the 1980's [1, 2] and the commercial availability of single-crystal 6H and
4H SiC wafers during the early 1990's [3], interest and enthusiasm for SiC electronic
devices has grown exponentially. SiC is attractive for several reasons: (i) its extreme
thermal stability offers the possibility of reliable high-temperature operation; (ii) its high
breakdown electric field makes it possible to build power switching devices with
resistance-area products 400x lower than silicon [4] and microwave power devices with
power densities lOOx higher than GaAs; and (iii) its native oxide (Si02) enables the
whole range of MOS devices and integrated circuits known in silicon [5]. These exciting
properties have given impetus to significant device development programs in Europe,
Japan, and the United States.
The current device development activities are taking place in an environment where
many basic fabrication and material science issues are still unresolved. The situation is
reminiscent of the early days of silicon technology, and indeed the SiC industry today is
in many ways comparable to the silicon industry of the 1950's. The prospects are
exciting, the problems are real, and the future is uncertain. In this review, we focus on
one aspect of SiC device and technology development, unipolar (or majority carrier)
power switching devices, specifically power MOSFETs and Schottky rectifiers. These
devices are expected to be among the first SiC devices to enter commercial production
early in the next decade. In the sections that follow, we will outline the basic device
designs, review the present status of device development, indicate the relationship
between material science issues and device performance, and identify the most critical
performance and material science issues still to be resolved.
Mat. Res. Soc. Symp. Proc. Vol. 572 ® 1999 Materials Research Society
2. MOTIVATION FOR POWER MOSFETs AND SCHOTTKY RECTIFIERS
Unipolar devices such as the power MOSFET and the Schottky rectifier are attractive
for electronic power switching applications for several reasons. Unlike bipolar devices
such as the insulated-gate bipolar transistor (IGBT) or the pin diode, the on-state current
in unipolar devices does not flow through a forward-biased pn junction. The voltage
drop across a forward-biased pn junction in 4H-SiC is about 2.8 V. Assuming a current
density of 200 A/cm2, SiC IGBTs and pin diodes will dissipate 560 W/cm2 just to
establish current flow. This dissipation is absent in power MOSFETs and Schottky
diodes. Second, since unipolar devices do not store minority carrier charge in the
conducting state, they exhibit minimal reverse recovery transients during switching. In
high-frequency switching applications, power dissipated during switching transients can
be the dominant power loss in the system.
3. SCHOTTKY RECTIFIERS
3.1 BASIC DESIGN
Figure 1 shows the cross section of a Schottky rectifier in SiC [6]. The starting
wafer is 4H-SiC, cut approximately 8° off axis to enable step-controlled epitaxy [7], and
polished on the (0001) silicon face. The n-type substrate is doped with nitrogen, with
resistivity about 10-20 mQ-cm and thickness of 300 - 350 (im. A lightly-doped n-type
spilayer is grown on the substrate with doping and thickness chosen to provide the
desired blocking voltage while minimizing on-resistance. For diodes designed for 1500
V operation, the epilayer is about 10 Jim thick, doped 4 - 8xl015 cm"3.
Implanted Edge
Termination
Anode
2^^
N- Ipilayer"
Figure 1. Cross section of a high-voltage SiC Schottky diode.
3.2 SELECTION OF SCHOTTKY METAL
The Schottky metal is chosen to provide the desired barrier height relative to 4H-SiC.
Typical Schottky metals are Ni (barrier height 1.3 eV) and Ti (barrier height 0.8 eV).
Figure 2 shows I-V characteristics of Ni and Ti Schottky diodes on 4H-SiC. As seen, the
lower barrier height of Ti results in a lower turn-on voltage in the conducting state
(desirable), but higher leakage current in the blocking state. The reverse leakage current
is due to emission of electrons from the Schottky metal into the semiconductor, a process
that depends exponentially on barrier height.
The design goal is to obtain the highest possible on current and blocking voltage
while minimizing power dissipation in the device. As a general guideline, one would like
to support a forward current of at least 100 A/cm2 in the on-state with a forward drop less
than 2 V. This results in a power dissipation of 200 W/cm2 in the diode. In the blocking
state one would like the reverse leakage current to be less than 1 mA/cm2 at the specified
blocking voltage. Although the Ti diode has a lower forward drop than the Ni diode, it
has orders of magnitude higher reverse leakage current. Taking 1 mA/cm2 as the
maximum reverse leakage current, the Ti diode could be specified for a maximum reverse
voltage of 450 V, while the Ni diode could be rated to 1200 V. For this reason, Ni
appears to be the better choice for a Schottky metal on 4H-SiC.
II 1 1 1 1 1 1 1 1 I 1 1 1 1 F"]
\jdr\
r~jS
*W
*?
A
1
.TijJL
IN 1
•M
i
¥,„
-2000
-1000
0
Voltage (V)
1
2
Figure 2. Current-voltage characteristics of Ni and Ti Schottky diodes on 4H-SiC.
3.3 EDGE TERMINATION
The diodes shown in Fig. 2 are fabricated on a 13 Jim n-type epilayer that has a
theoretical plane-junction breakdown voltage of about 2000 V [6]. The Ni diode exhibits
a maximum blocking voltage of 1720 V, about 87% of the theoretical plane-junction
breakdown voltage for this epilayer. In order to achieve this result, it is necessary to
provide edge termination along the periphery of the Schottky metal to minimize twodimensional field crowding. The edge termination employed in these diodes makes use
of a boron implant under the edge of the Schottky metal, as shown in Fig. 1. This
implant is annealed at 1000 °C, a temperature high enough to anneal most of the implant
damage without activating the dopants [8]. The effect is to form a resistive layer that
carries a small leakage current under reverse bias. This leakage current produces a lateral
voltage drop that smoothly tapers the electric field, reducing field crowding at the edges
of the Schottky metal. The leakage current is noticeable in the Ni Schottky diode at
reverse voltages below about 800 V (see Fig. 2), but is obscured in the Ti diode by the
higher Schottky barrier leakage. Other edge terminations have also been used
successfully [9, 10].
3.4 YIELD-LIMITING MATERIAL DEFECTS
Although the diodes in Fig. 2 are small, recent work has focused on scaling to larger
devices to obtain on-state currents in the range of several A to tens of A [11]. These
efforts have made it possible to estimate the density of yield-limiting defects in the
material. In one experiment, a 35 mm "low-micropipe" (micropipe density < 30 cm"2)
4H-SiC wafer was fabricated with circular diodes of diameters 1.25, 2, and 3 mm. The
yield of good devices for these diodes was 58, 31, and 22 %, respectively, established by
testing to 200 V reverse bias. Fitting to a simple yield model indicates a defect density of
about 20 cm"2, consistent with the expected density of micropipes in the wafer.
Micropipes are open-core screw dislocations with large Burgers vectors that extend
through the substrate and subsequently-grown epilayers. Such defects are obviously fatal
to high-voltage power devices. Fortunately, the density of micropipes in commerciallygrown wafers has been declining steadily in recent years, and the best reported wafers
now have micropipe densities below 1 cm"2 [12]. Such wafers would permit fabrication
of 50 A Schottky diodes with acceptable yield.
In addition to micropipes, SiC wafers have a much higher density of single-screw
dislocations, typically several thousand per cm2 [13]. Although these defects are not
immediately fatal to power devices, they appear to be responsible for "soft breakdown"
characteristics, and may possibly give rise to a negative temperature coefficient of
breakdown in some devices [14]. A negative temperature coefficient of breakdown is
undesirable, since it leads to filamentation and destructive breakdown of the device.
Another problem that becomes apparent when Schottky diodes are scaled to larger
diameters is the effect of surface imperfections and irregularities. Since the maximum
electric field occurs at the semiconductor surface, the Schottky diode is especially
sensitive to surface defects or irregularities. Any macroscopic surface imperfection, such
as a pit or asperity, will cause field crowding and premature breakdown. Nomarski
microscopy reveals a variety of pits, asperities, and gouges resulting from polishing
damage on commercially available SiC wafers. This is an area where the materials
science community could make an important contribution, since it is critical to obtaining
the large-area high-current Schottky diodes demanded by customers.
3.5 CURRENT STATUS OF EXPERIMENTAL DEVICES
SiC Schottky diodes have achieved impressive performance figures compared to
silicon devices. A useful figure of merit for unipolar devices is 9 = VB2 / RON.SP. where
VB is the maximum blocking voltage and RON.SP is tne specific on-resistance in fi-cm2
tne
(RON SP is
product of on-resistance and device area). This figure of merit is
appropriate because the minimum achievable RQN SP in a unipolar device theoretically
scales as the square of the blocking voltage. The Ni Schottky diode in Fig. 2, fabricated
on a 13 (xm 4H-SiC epilayer, has a blocking voltage of 1,720 V and a specific onresistance of 5.6 mil-cm2, yielding a 6 of 528 MW/cm2 [6]. By comparison, the
theoretical limit for silicon unipolar devices is a 9 of 4 MW/cm2. Thus, the diode of Fig.
2 has a figure of merit approximately 130x higher than the theoretical limit for silicon
devices. Schottky diodes fabricated in our laboratory on 50 um epilayers of 4H-SiC have
demonstrated blocking voltages as high as 4,200 V [15].
4. POWER MOSFETs
4.1 BASIC DESIGNS
Figure 3 shows cross sections of DMOS and UMOS power transistors in SiC. These
MOSFETs are vertical devices with the n+ substrate serving as the drain contact. A
lightly-doped n-type epilayer is grown on the substrate to form the drain drift region. As
with the Schottky diode, the doping and thickness of the drift region are chosen to obtain
the desired blocking voltage while minimizing the on-resistance. In the UMOS structure,
a moderately-doped p-type epilayer is also grown on top of the n-type epilayer. This ptype layer is grounded, and acts as the base of the MOSFET (the base is equivalent to the
p-type substrate of a planar MOSFET). In the DMOS structure, the p-base is formed by
ion implantation of aluminum or boron. N+ source regions are created in selected areas
by implantation of nitrogen or phosphorus, and the implants are activated by a high
temperature anneal. In the UMOS structure, a trench is defined by reactive ion etching
following implant activation. The device is then oxidized to form the gate insulator.
Polysilicon is deposited by LPCVD, doped by diffusion of a spin-on dopant, and
patterned to form the gate electrode. Ni source contacts and Al base contacts are
deposited, defined by liftoff, and annealed.
Gate
Source
SiOs
T
Polysilicon
Source
<
Gate
Source
)
p
Si02
Ni
Fl
w3ase^
^7NV"
Source
Polysilicon
r
I
N* • '_• ,
• • P'Bas
N- Dritt Region
i\
- Drift Reg on
N+ Substrate
... _v ...^Substrate
vV
,
r
Drain
Figure 3. Cross sections of UMOS (left) and DMOS power transistors.
Ni
4.2 OPTIMIZATION OF THE MOS INTERFACE
The electrical quality of the oxide/semiconductor interface is of critical importance
to the operation of all MOS-based power devices. Oxidation of SiC is both similar to and
different from oxidation of silicon [5]. Since the SiC crystal consists of alternating
planes of silicon and carbon atoms, the oxidation process involves two reactions:
oxidation of silicon to produce silicon dioxide, and oxidation of carbon to produce carbon
monoxide and carbon dioxide. The Si02 forms the passivating oxide while the carbon
byproducts escape in the gas phase.
Oxidation of SiC is much slower than oxidation of silicon, and the oxidation rate
depends upon the surface orientation. The fastest oxidation occurs on the (0004-) carbon
face, while the slowest oxidation occurs on the (0001) silicon face. The difference in
oxidation rates is related to the bonding structure, in which successive planes of silicon
and carbon atoms are alternately singly bonded and triply bonded. Oxidation of the
carbon face involves oxidation of carbon planes triply bonded to silicon planes, followed
by oxidation of silicon planes singly bonded to carbon planes. In contrast, oxidation of
the silicon face consists of oxidation of silicon planes triply bonded to carbon planes,
followed by oxidation of carbon planes singly bonded to silicon planes. As might be
expected, these two processes exhibit significantly different reaction kinetics.
The Si02 - SiC interface is more complex than the Si02 - silicon interface, and the
role of carbon species in determining the fixed interface charge QF and the density of
interface states D^ are subjects of current research. At the present time, the optimum
oxidation conditions appear to be thermal oxidation in wet 02 or steam at 1150 °C,
followed by a 30 min. in-situ anneal at 1150 °C in argon [16] and a subsequent reoxidation anneal at 950 °C for 60 min. in wet 02 [17]. The argon anneal permits carbon
species to diffuse out of the oxide under non-oxidizing conditions. The subsequent low
temperature re-oxidation forms the final interface, and appears to be beneficial in
reducing interface "state density. Pre-oxidation cleaning and loading steps are also
important. Samples are cleaned using a standard "RCA" clean and loaded into the
oxidation tube at 850 °C under flowing 02. The furnace temperature is then slowly
ramped to the 1150 °C oxidation temperature. This procedure minimizes degradation of
the surface due to loss of silicon during the furnace insertion [16].
Using the above-described oxidation procedure, MOS interfaces on the (0001)
silicon face of 4H and 6H SiC typically have fixed charge densities from 5 - 8x10" cm"2
and interface state densities that range from around 1.5x10" cm"2 eV"' at 0.5 eV above the
valence band to around 5xl010 cm"2 eV"1 near midgap [18]. These values are 2 - 3x higher
than currently found in silicon devices, but are not high enough to cause problems with
MOSFET operation. As a general guideline, a MOSFET in strong inversion is operating
with about 6 - 8xl0'2 inversion electrons per cm2, determined by the maximum electric
field in the gate oxide (3-4 MV/cm). As can be seen, the fixed charge and interface
state density numbers quoted above are small compared to the density of inversion
electrons in strong inversion.
MOS interfaces formed on the a-axis surfaces of SiC are definitely inferior to those
formed on the silicon face. Measurements on the (1400) and 1130) surfaces indicate
interface state densities in the range 5 - 7x10" cm'2 eV, 5 - lOx higher than on the
silicon face [19]. MOS interfaces formed on the (0004-) carbon face are even worse, with
interface state densities so high as to prevent full modulation of the surface potential.
The differences are thought to be due to the presence of carbon bonds at the interface.
On the carbon face, essentially all the interface bonds are associated with carbon atoms,
while on the a-axis surfaces, approximately half the bonds are associated with carbon.
On the silicon face, the vast majority of bonds are associated with silicon atoms, and on
an atomic scale this interface resembles the Si02 - Si interface.
The breakdown electric field of oxides on 4H and 6H-SiC are comparable to oxides
on silicon, typically in the range 8-10 MV/cm, although the spread in breakdown
voltage across a SiC wafer is larger than for silicon. This is an indication of the relative
immaturity of the technology, and probably reflects the difficulty in polishing SiC wafers
due to the extreme hardness of the material. As the case with Schottky diodes, any
surface imperfection or irregularity will concentrate the surface electric field, leading to
premature local breakdown of the oxide.
4.3 MOS OXIDE RELIABILITY
Power switching devices operate at high electric fields, and SiC devices are capable
of sustaining much higher internal fields than silicon devices. Indeed, the main
advantage of SiC for power devices is the fact that the critical field for avalanche
breakdown is 8 - lOx higher than in silicon. In MOS devices, a serious problem occurs
at the oxide/semiconductor interface. Because the ratio of dielectric constants between
Si02 and SiC is about 2.5, the electric field in the oxide is 2.5x higher than the surface
electric field in the SiC. Since the critical field for avalanche breakdown in SiC is 2 - 4
MV/cm (depending on doping), the field in the oxide is in the range 5-12 MV/cm.
These fields are uncomfortably close to the breakdown field of the oxide. Note that if the
same oxide were used on a silicon device, the fields in the silicon (and the oxide) would
be about lOx lower, safely below the breakdown field of the oxide.
Oxides on SiC are also more susceptible to electron injection from the
semiconductor, since the conduction band discontinuity between SiC and Si02 is lower
than that between silicon and Si02, and injection increases exponentially as barrier height
is lowered. For these reasons, the long-term reliability of oxides on SiC must be studied
carefully to determine the tradeoffs between mean-time-to-failure and oxide field.
Although this investigation is still in its early stages, some preliminary conclusions can
be drawn. Figure 4 shows mean-time-to-failure of thermal oxides on n-type 6H-SiC as a
function of oxide field for three temperatures [20]. Also shown in the figure are the
mean-time-to-failure for oxides on silicon at the same temperatures. At high fields, the
failure times for oxides on SiC are shorter than for oxides on silicon. However, the field
dependence of failure time is stronger for oxides on SiC, with the result that at fields in
the range 3-4 MV/cm the mean-time-to-failures of SiC and silicon are comparable.
These results suggest that acceptable oxide reliability on SiC can be obtained at
temperatures up to 150 °C if the field in the oxide is kept below 4 - 5 MV/cm. This can
be done without compromising device operation if precautions are taken in the device
design to limit the maximum fields in the oxide.
>
ß
10°
0)
E
tc
C0
CD
5
5
6
7
8
9
10
Oxide Field (MV/cm)
Figure 4. Mean-time-to-failure for oxides on n-type 6H-SiC and on p-type silicon as
a function of oxide field and temperature.
4.4 INVERSION LAYER MOBILITY AND IMPLANT ACTIVATION
Since the avalanche breakdown field in SiC is 8 - lOx higher than in silicon, the
specific on-resistance (Q - cm2) of the drain drift region in power MOSFETs will be
about 400x lower than for silicon devices of the same blocking voltage. However, to date
no SiC power MOSFET has achieved this ideal. This is because for blocking voltages
below about 5,000 V, the resistance of SiC MOSFETs is actually dominated by the
channel resistance rather than the drift region resistance. The channel resistance is
limited by the inversion layer electron mobility. In silicon, the inversion layer mobility is
about half the bulk mobility, but in SiC the inversion layer mobility is only about 10 20% of the bulk value. The reason for this is not well understood. Moreover, the
inversion layer mobility on 4H-SiC is typically much lower than on 6H, in spite of the
fact that the bulk mobility in 4H is higher than in 6H. Typical values for inversion layer
mobility reported in the literature are in the range 25 - 50 cmVVs on 4H-SiC and 70 - 90
cmVVs on 6H-SiC. These mobility differences cannot be attributed to differences in
fixed oxide charge density or interface state density between 4H and 6H-SiC, since these
are fairly comparable between the two interfaces.
It has been recently observed that the inversion layer mobility on 4H-SiC (and
probably to a lesser extent on 6H-SiC) can be severely degraded by thermal processing
before gate oxide formation [21]. This is attributed to step bunching on the surface.
Figure 5 shows severe step bunching resulting from a 1700 °C anneal used to activate the
boron p-well implant in 4H-SiC [22]. To enable step-controlled epitaxy, 6H and 4H-SiC
wafers are routinely cut at 3.5° and 8° off-axis, respectively. During the hightemperature implant anneal, considerable motion of these surface steps occurs and the
surface seeks a lower energy state by reducing the density of steps while increasing their
height. If this surface is subsequently oxidized to form a MOSFET, the resulting oxidesemiconductor interface can have macroscopic steps whose height is a significant fraction
of the oxide thickness. In such a case it may be appropriate to visualize this surface as
10
consisting of alternating horizontal
and vertical interface regions, or
equivalently, as a series connection
of horizontal and vertical
MOSFETs. When the surface is
brought into inversion by the gate
bias, the horizontal regions will
typically go into strong inversion
while the vertical surfaces are only
weakly inverted, or not inverted at
all. This occurs because of the
relative geometry of the field lines
from the gate to the semiconductor,
because the oxide on the vertical
sidewall is 2-3x thicker than on the
horizontal surface, or because the
fixed charge and interface state
Figure 5. AFM image of the surface of 4H-SiC,
densities on the vertical sidewall
H
2
implanted with boron to a dose of 4x10 cm" ,
are higher than on the horizontal
typical of a p-well implant, and annealed at 1700
surface. In any event, the Steps
°C for 40 min. in argon. Dimensions of the
create potential barriers to
image are K) xlO (im.
electrons flowing from source to
drain, significantly degrading the
effective surface mobility of the MOSFET. In extreme cases the effective mobility can
have fractional values. These extremely low mobilities are correlated with high
temperature anneals (1600 - 1700 °C) required to activate p-well implants prior to gate
oxidation. Table 1 illustrates the range of inversion layer mobilities observed on 4H-SiC
as a function of implant anneal conditions prior to gate oxidation. As seen, in order to
obtain high inversion layer mobilities on 4H-SiC it is necessary to keep the maximum
implant anneal temperature below about 1200 °C. This temperature is insufficient to
activate boron or aluminum used for p-well implants in DMOSFETs, but is high enough
to activate nitrogen or phosphorus used for the source implants.
N-type dopants in SiC are nitrogen and phosphorus, and p-type dopants are
aluminum and boron. Since diffusion of impurities is impractical in SiC due to the low
diffusion coefficient of impurities at any reasonable temperature, ion implantation is used
for selective doping. Figure 6 shows sheet resistivity for nitrogen and phosphorus source
Anneal Temperature
No p-well implant
1200 °C
25 cmWs
1300 °C
7 cm2/Vs
Boron p-well implant
1400 °C '
5 cmVVs
1700 °C
9 cm2/Vs *
0.32 cmWs *
0.06 cnr/Vs
Table 1. Inversion layer electron mobility on 4H-SiC as a function of implant anneal
conditions prior to gate oxidation. Asterisks (*) indicate that the implant anneal was
conducted under a silane overpressure; all other anneals were conducted in argon.
11
implants and electron inversion layer mobility in 4H-SiC as a function of implant anneal
temperature [23]. As seen, to obtain inversion layer mobilities greater than 10, it is
necessary to keep the implant anneal temperatures below about 1250 °C. This can be
done successfully if phosphorus is used instead of nitrogen as the source dopant.
The problem becomes more severe if either aluminum or boron implantation is
required in the fabrication process, as for the DMOS structure of Fig. 3. Figure 7 shows
the activation percentage of boron in 4H-SiC as a function of anneal temperature [22].
To obtain good activation, it is necessary to anneal at temperatures in excess of 1600 °C.
Comparison with Fig. 6 indicates that the expected MOS inversion layer mobility
following such an anneal will be in the single digits, at best. The actual results on
DMOSFETs are even worse, since the MOS inversion channel is formed in the region
damaged by the boron implant. The second line in Table 1 shows that inversion layer
mobilities in implanted regions annealed at 1700 °C are fractional. As described in the
previous section, the low effective inversion layer mobility is caused by the dramatic step
bunching that occurs during implant anneal (c.f. Fig. 5). One solution is to convert the
near-surface region of the semiconductor to n-type, forming an accumulation-layer
MOSFET, or ACCUFET [24, 25]. The density of donors in the ACCUFET layer is kept
low so that the layer is completely depleted at zero gate bias, but there the layer has
enough conductivity under positive gate bias that a conductive path is established around
the potential barriers associated with the surface steps, significantly enhancing the
effective mobility of the MOSFET. This solution, while workable, is undesirable
because of the difficulty of controlling the threshold voltage precisely when it is close to
zero, and the likelihood that the threshold voltage will go negative (creating a normallyon MOSFET) as temperature is raised.
1 1 1 1 I 1
I l-r-i-i i | i i i i j i
i i [ i i i i
CO
>
~
^**,*"S**I_J
^Mobility
|
10
o
CD
>.
cd
Argon Anneal
^sJ
Resistivity
N7 ogen
:
y
1000
Im slant
CD
en
'
'!
I
Iji
1 ll vity
co
.. ..
' .... i ... .
1200
l 100
:
Phosph )rus
j
Impla it
1400
1300
1500
Anneal Temperature (°C)
Figure 6. Inversion layer electron mobility and source implant sheet resistivity for
nitrogen and phosphorus in 4H-SiC as a function of implant anneal temperature. The
highest inversion layer mobility (25 cmVVs) is obtained for a 1200 °C anneal, where the
phosphorus source implant sheet resistivity is an acceptable 500 Q per square.
12
100
V' •
Figure 7. Activation percentage of
boron in 4H-SiC as a function of
anneal temperature. All anneals are
performed for 40 min. in argon.
10 -
1
1450 1500 1550 1600 1650 1700 17!
Anneal Temperature (°C)
4.5 CURRENT STATUS OF EXPERIMENTAL DEVICES
In spite of the material and processing difficulties described above, progress in the
development of SiC power MOSFETs has been dramatic. Currently, both DMOS and
UMOS devices are being developed. UMOSFETs fabricated on 10 Jim 4H-SiC epilayers
have achieved blocking voltages of 1400 V at a specific on-resistance of 15.7 mQ-cm2,
for a figure of merit 9 = VB2 / R0N of 125 MW/cm2, approximately 25x higher than the'
theoretical limit for silicon MOSFETs [25]. Lateral DMOSFETs on insulating 4H-SiC
substrates have achieved blocking voltages of 2600 V [26]. With the availability of
thicker epilayers (up to 50 |im are currently available), it is anticipated that MOSFETs
with blocking voltages in the 3 - 5 kV range will soon be reported.
It is worth noting that the performance of the best reported SiC MOSFETs are still
well below the theoretical limit expected for 4H-SiC. For example, the 1400 V
UMOSFET reported above with a 9 of 125 MW/cm2 is still a factor of 16 below the 4HSiC theoretical limit. The main reason for the shortfall is the specific on-resistance RQN,
which is dominated by the resistance of the MOSFET channel. In order to reach the SiC
theoretical limit, the MOSFET channel resistance needs to be reduced by about an order
of magnitude. This is the biggest challenge facing MOSFET device developers today.
5. SUMMARY AND CONCLUSIONS
During the past several years, the development of prototype devices in SiC has been
taking place in parallel with the optimization of unit fabrication processes and research
on fundamental material science issues. This leads to mistakes and false starts in device
development, but it also tends to bring into sharp focus the critical material science
issues. At the present time, from a device development perspective, the most critical
material science issues are: (1) size and cost of SiC wafers, (2) surface morphology of asreceived wafers, (3) density of micropipes and single screw dislocations in the material,
(4) activation of implants (particularly aluminum and boron p-type implants), (5) low
inversion layer electron mobility at the SiQ2 - SiC interface and its relationship to
13
previous high-temperature steps, (6) high ohmic contact resistance (particularly to p-type
SiC), and (7) the stability and reliability of SiO, on SiC under high-field and hightemperature stress. Continued close communication between device developers and
material scientists is needed to focus attention on these critical problems and verify their
solution.
ACKNOWLEDGEMENTS
This work is supported by ONR under MURI grant, no. N00014-95-1-1302.
REFERENCES
1. G. Ziegler, P. Lanig, D. Theis, and C. Weyrich, IEEE Trans. Electr. Dev. 30, 227 (1983).
2. C. H. Carter, Jr., L. Tang, and R. F. Davis, 4th National Review Meeting on the Growth and
Characterization of SiC, Raleigh, NC, USA, 1987; U.S. Patent No. 4,866,005 (Sept. 12, 1989) R.
F. Davis, C. H. Carter, Jr., and C. E. Hunter.
3. Cree Research, Inc., Durham, NC.
4. B. J. Baliga, IEEE Electron Device Lett. 10, 455 (1989).
5. J. A. Cooper, Jr., Physica Stat. Solidi (a), 162, 305 (1997).
6. K. J. Schoen, J. M. Woodall, J. A. Cooper, Jr., and M. R. Melloch, IEEE Trans. Electr. Dev.
45, 1595 (1998).
7. H. Matsunami, T. Ueda, and H. Nishino, Proc. Mat. Res. Soc. Symp. 162, 397 (1990).
8. A. Itoh, T. Kimoto, and H. Matsunami, IEEE Electron Device Lett. 17, 139 (1996).
9. D. Alok, B. J. Baliga, and P. K. McLarty, IEEE Electron Device Lett. 15, 394 (1994).
10. R. Singh and J. W. Palmour, Proc. Int'l. Symp. on Power Semi. Devices and ICs, Weimar,
Germany, May 26-29, 1997.
11. G. M. Dolny, D. T. Morisette, P. M. Shenoy, M. Zafrani, J. Gladish, J. M. Woodall, J. A.
Cooper, Jr., and M. R. Melloch, IEEE Device Res. Con/., Charlottesville, VA, June 22 - 24, 1998.
12. C. H. Carter, Jr., Cree Research, Inc., Durham, NC, private communication.
13. S. Wang, M. Dudley. C. H. Carter, Jr., and H. S. Hong, Mat. Res. Soc. Symp. Proc, 339, 735
(1994).
14. P. G. Neudeck and C. Fazi, IEEE Electron Device Lett. 18, 96 (1997).
15. H. M. McGlothlin, D. T. Morisette, J. A. Cooper, and M. R. Melloch, unpublished.
16. J. N. Shenoy, G. L. Chindalore, M. R. Melloch, J. A. Cooper, Jr., J. W. Palmour, and K. G.
Irvine, J. Electronic Mat'Is, 24, 303 (1995).
17. L. A. Lipkin and J. W. Palmour, J. Electr. Mafls, 25, 909 (1996).
18. M. K. Das, J. A. Cooper, Jr., and M. R. Melloch, /. Electr. Marts, 27, 353 (1998).
19. J. N. Shenoy, M. K. Das, J. A. Cooper, Jr., M. R. Melloch, and J. W. Palmour, J. Appl.
Physics, 79, 3042 (1996).
20. M. M. Maranowski and J. A. Cooper, Jr., IEEE Trans. Electr. Dev. 46, (1999).
21. M. K. Das, J. A. Cooper, Jr., M. R. Melloch, and M. A. Capano, Semi. Interface Specialists
Conf., San Diego, CA, December 3 - 5, 1998.
22. M. A. Capano, S-H. Ryu, M. R. Melloch, and J. A. Cooper, Jr., J. Electr. Mafls, 27, 370
(1998).
23. R. Santharumar, M. K. Das, M. A. Capano, J. A. Cooper, and M. R. Melloch, unpublished.
24. P. M. Shenoy and B. J. Baliga, IEEE Electron Device Lett., 18, 589 (1997).
25. J. Tan, J. A. Cooper, Jr., and M. R. Melloch, IEEE Electron Device Lett., 19,487 (1998).
26. J. Spitz, M. R. Melloch, J. A. Cooper, Jr., and M. A. Capano, IEEE Electron Device Lett.,
19, 100(1998).
14
RECENT PROGRESS IN SiC MICROWAVE MESFETs
S.T. ALLEN, S.T. SHEPPARD, W.L. PRIBBLE, RA. SADLER, T.S. ALCORN, Z. RING, and
J.W. PALMOUR
Cree Research, Inc., 4600 Silicon Drive, Durham, NC, 27703; (919) 361-5709
ABSTRACT
SiC MESFET's have shown an RF power density of 4.6 W/mm at 3.5 GHz and a power
added efficiency of 60% with 3 W/mm at 800 MHz, demonstrating that SiC devices are capable
of very high power densities and high efficiencies. Single devices with 48 mm of gate periphery
were mounted in a hybrid circuit and achieved a maximum RF power of 80 watts CW at 3.1 GHz
with 38% PAE.
INTRODUCTION
Silicon carbide (SiC) MESFET's are emerging as a promising technology for high- power
microwave applications due to a combination of superior properties of SiC, including a high
breakdown electric field, high saturated electron velocity and high thermal conductivity. In this
paper we report on the substantial progress that has recently been made in microwave SiC
MESFET technology. Improvements to device design and fabrication have led to increased
breakdown voltages of greater than 150 V for FET's with a channel doping of 3x10" cm"3.
Improvements in process control and repeatability have resulted in the ability to yield devices
with up to 48 mm of gate periphery. Cree also recently began fabricating MESFET's on 2-inch
diameter semi-insulating SiC substrates, a substantial increase in wafer area over the previous 13/8" substrates, preparing the way for producing these devices in large quantity.
ADVANTAGEOUS PROPERTIES OF SILICON CARBIDE
SiC occurs in over 200 different crystal structures, or polytypes, but for semiconductor
applications the 6H and 4H polytypes have received the most attention due to the availability of
high-quality single crystalline substrates. For microwave MESFET's the 4H-SiC polytype is
preferable because it has a larger bandgap and higher electron mobility than 6H-SiC. It is the
wide bandgap of 3.2 eV, compared to 1.1 eV for Si and 1.4 eV for GaAs, that gives SiC its
primary advantage for high-power devices. This wide bandgap gives rise to a breakdown electric
field that is 10 times higher than in GaAs or Si. This is illustrated in Figure 1, which shows the
measured breakdown voltage of 4H-SiC p-n junction diodes as well as the theoretical curves for
Si and GaAs. This high breakdown field has been exploited to fabricate sub-micron SiC
MESFET's with gate-to-drain breakdown voltages exceeding 200 V.
The one drawback to SiC for use in microwave devices is its poor low-field electron
mobility, which is in the range of 300 - 500 cm2/V-sec. for doping levels of interest for
MESFET's, i.e., 1x10" cm"3 < ND < 5x10" cm'3. This results in a larger source resistance and
lower transconductance than is typical of GaAs MESFET's, but is partially offset by the ability
to operate the SiC devices under extremely high electric fields. The saturated electron velocity in
6H-SiC is 2xl07 cm/s and has been predicted by Monte Carlo simulations to be 2.7xl07 cm/s in
4H-SiC [2], almost 3 times higher than in GaAs at high fields. Although the knee voltage of SiC
MESFET's is relatively high (typically » 9 V), the drain efficiency of the devices is still high
because the breakdown voltage is over 100 V. The channel current density is reasonably large,
typically around 300 mA/mm for 0.7 um gate length devices, due to the high saturated velocity.
When combined with the high breakdown voltage, this results in the large RF power density of
over 4 W/mm that has been measured for SiC MESFET's.
15
Mat. Res. Soc. Symp. Proc. Vol. 572 ® 1999 Materials Research Society
!s\s^
i
i
1 * Mt«*ur«d 4H-SIC Data |
^SiC
*v
^
s^ ^GaAs
;
^
^
:
^L
1016
1017
1018
Doping Level (cnr3)
10«
10la
Figure 1: Measured breakdown voltage of 4H-SiC p-n junction
diodes as a function of doping, and the theoretical maximums for
Si and GaAs.
The other property of SiC that gives it a significant advantage over other semiconductors is
its very high thermal conductivity. In order to take advantage of the high electrical power
densities available in this material, it is also necessary to dissipate very high thermal power
densities, making the thermal conductivity an extremely important parameter. Measured thermal
conductivity data for SiC is shown in Table I for low doped n- and p-type, doped n-type and
semi-insulating material. The thermal conductivity was measured by creating a temperature
difference across a piece of SiC. Thermocouples were inserted into holes drilled 1 cm apart and
AT and applied power were used to calculate the thermal conductivity at 25°C and 100°C.
Copper and Al were used as calibration standards. Thermal conductivity is the product of a
material's density, heat capacity and its thermal diffusivity: the latter being dependent on the
doping and quality of a material. The very low doped material exhibits a-axis thermal
conductivity roughly the same as that reported by Slack [2] for pure Lely platelets. The doped
materials and c-axis direction have significantly lower thermal conductivity values. Even the
lowest measured value of 3.3 W/cm-K is 7 times higher than the thermal conductivity of GaAs,
implying that not only is the power density of SiC high in terms of W/mm of gate periphery but
SiC also has extremely high power handling capability in terms of W/mm of die area. For high
power, high frequency applications, this is the more important figure of merit since die size
becomes constrained by wavelength, and all power devices are operated at their thermal limit.
Table I
Measured Thermal Conductivity Data for SiC
Sample
Tvpe
4H Semi
Ins.
4Hn
4Hn
6Hn
6Hn
6Hn
6Hp
Slack[2]
Direction
He
He
±c
lie
He
1c
J.C
±C
Carrier
(cm"3)
S.I.
2.0el8n
5eI5n
1.5el8n
3.5eI7n
3.5el7n
1.4el6p
~le!7
16
Thermal Conductivity
373 K
298 K
(W/cm-K) (W/cm-K)
2.6
3.3
3.3
4.8
3.0
3.2
3.8
4.0
~5
2.5
2.9
2.3
2.3
2.8
3.2
~3
RECENT ADVANCES IN SIC SUBSTRATES
The development of SiC electronic devices has been limited in the past by the lack of
availability of large, high quality SiC substrates. The primary defects in bulk SiC are
micropipes, which are superscrew dislocations in the crystal that have open cores, resulting in
pinholes in the wafer. Recently Cree has made advancements in crystal growth technology that
resulted in producing 4H-SiC wafers with a micropipe density of < 1 cm , which is more than
two orders of magnitude less than it was three years ago. Because the active area of microwave
MESFET's is very small, limited to the source- drain separation of 4 um, a micropipe density of
< 10 cm"2 has a negligible impact on yield. Therefore, with these recent reductions in micropipe
defect densities and the increase in wafer diameter size (2-inch in production and 3-inch
demonstrated) SiC crystal quality has improved to the point where it would be viable to fabricate
SiC MESFET's in production quantities.
The semi-insulating material has been characterized extensively by W.C. Mitchell and R.
Perrin at the Air Force Wright Patterson Laboratory. Figure 2 is an Arrhenius plot of the high
temperature resistivity of one of these wafers determined from Hall-effect measurements. The
extracted activation energy of 1.7 eV is consistent with deep-level compensation in SiC with a
bandgap of 3.2 eV and leads to an extrapolated room temperature resistivity of 1020 Q-cm. At a
temperature of 500°C, the resistivity exceeds 107 fi-cm, making the semi-insulating properties of
the substrate consistent with a technology capable of operating at extremely high temperatures.
Room Temp.
2.0
2.5
3.0
3.5
1AT(1000/K)
Figure 2: Plot of resistivity vs. 1/T for semi-insulating 4H-SiC as
determined with Hall-effect measurements.
MICOWAVE POWER RESULTS
Cree's optimized S-band power FET's have a gate length of 0.7 urn and employ a channel
doping of 3x10" cm"3. The details of the fabrication sequence have been discussed previously
[3]. The FET's are designed to have a threshold voltage of V = -10 V, resulting in an Idss of 300
mA/mm at Vds = 10 V and a peak transconductance of 45 mS/mm. With this device structure, 1mm FET's typically have a three-terminal breakdown voltage Vds in the range of 150 - 200 V,
defined at the 1 mA/mm point. As determined from small-signal S-parameter measurements,
average values for frequency response are fT = 9 GHz and fmax = 20 GHz.
From these devices a maximum power density of 4.6 W/mm at 3.5 GHz has been measured
using an on-wafer load pull system. As shown in Figure 3, a 0.25-mm FET operating at a drain
bias of 60 V had a peak power of 1.15 W, a Class A PAE of 39% and an associated power gain
of 12.5 dB. A similar device operating at 800 MHz had a power density of 3.0 W/mm and an
improved PAE of 60%, as shown in Figure 4, demonstrating that because of the high operating
voltages, the intrinsic efficiency of SiC MESFET's is high.
17
435 £
o
30 >
Q.
25 |
20 |
o
15 m
w,= 0.25 mm
v„ = 60 V
io £
Freq = 3.5 GHz
12
14
16
18
Input Power (dBm)
20
22
Figure 3: 0.25-mm SiC MESFET with a peak power density of
4.6 W/mm at 3.5 GHz with a drain bias of 60 V.
36-,
34-
70
Power = 3 W/mm
60
33-
PAE = 60%
32-
Lin. Gain = 19.6 dB
50 1
'
31-
>
PAE
40 £
Q
30-
30 m
292827-
ld, = 90 mA
264
20 |
Vj, = 50 V
10-g
Freq. = 800 MHz
6
8
10
12
14
16
Input Po wer dBm)
18
0
20
Figure 4: Power sweep of a 0.84-mm SiC MESFET with a PAE
of 60% and 3.0 W/mm.
A single device with 48-mm of gate periphery was packaged in a hybrid circuit with input
and output matching networks fabricated from alumina, and produced 80 watts CW at 3.1 GHz
with 38% PAE, as shown in Figure 5. This is the highest CW power level ever reported for a
single device operating at this frequency, and demonstrates the potential of SiC to have a
substantial impact on solid state microwave power amplifiers.
M>,
49.
48-
60
, . , , ! , . , , . i , . i i
CW Power = 80 Watts
45
Assoc. Gain = 7.6 dB
40
PAE = 38%
35
30
47-
25
46-
20
15
4544-
</
10
V„ = 58 V
Freq. = 3.1 GHz
33
34
35
:
36 37 3i
9 40
Input Powe r(d Bm)
41
42
.5
43
Figure 5: Power sweep of a 48-mm SiC MESFET at 3.1 GHz
showing a CW power level of 80 watts.
18
With an increase in the channel doping and a reduction of the gate length to less than 0.45
um, SiC MESFET's have shown excellent power performance up to 10 GHz. As illustrated by
the power sweep in Figure 6 a power density of 4.3 W/mm was measured from a 0.25-mm
MESFET at 10 GHz, with a peak power of 1.1 W, a Class A PAE of 20% and an associated gain
of 9 dB. Cree has also recently developed a via hole process for the semi-insulating SiC
substrates, making the fabrication of MMIC's possible. Combined with the high power densities
shown at X-band, this makes SiC MESFET's an attractive MMIC technology for future systems.
E
m
28
Power Density = 4.3 W/mm
Assoc. Gain = 9.0 dB
PAE = 20 %
r 26
POUT
/ —* :
S- 22
15 *
CD
— — —*-%£+— —
/
o
10
Vos = 60V
Freq = 10.0 GHz :
Wo = 0.25 mm
12
14
16
18
Input Power (dBm)
20
Figure 6: On-wafer load pull measurement of a SiC MESFET
showing 4.3 W/mm at 10 GHz.
GAN/ALGAN HEMT's ON SiC SUBSTRATES
Another exciting technology for high power microwave applications that leverages the
advantages of semi-insulating SiC is GaN/AlGaN High Electron Mobility Transistors (HEMT's).
The close lattice match between SiC and GaN results in a lower defect density epilayer than
typically obtained with growth of GaN on sapphire substrates. Additionally, the much high
thermal conductivity of the SiC substrate is critical for dissipating the very high power densities
possible with the GaN system. GaN devices fabricated on sapphire substrates have achieved
relatively high power densities, but the junction temperatures at which these levels are achieved
are typically in excess of 300°C due to the very high thermal impedance of the substrate. The
structure of the GaN/AlGaN HEMT devices fabricated on SiC substrates is shown in Figure 7.
The structure contains an A1N buffer layer, 2 urn of undoped GaN and approximately 27 nm of
Al0,4Ga,,i86N. The AlGaN cap layer comprises a 5 nm undoped spacer layer, a 12 nm, 2xl018/cm3
Si-doped donor layer and a 10 nm undoped barrier layer. Device isolation is achieved with
shallow mesas dry etched in a chlorine-based plasma.
Typical dc output characteristics of a 1-mm-wide HEMT with L0 = 0.45, IGS =1.0 and IGD =
1.5 |im show a peak current of 680 mA/mm at Vas = +2 V and a maximum extrinsic
transconductance near Vas = -0.5 V of 200 mS/mm. Typical three-terminal gate-drain breakdown
voltages range between 60-70 V. Small-signal gain measurements at Vm = 20 V and Vcs = -1 V
show an extrapolated unity gain frequency fj of 28 GHz [4]. The maximum available gain
(MAG) remained high up to the maximum frequency of the network analyzer, so the_4iAX of this
device is estimated to be 114 GHz by modeling the power gain above 35 GHz. The effective
channel electron velocity, as determined from the slope of the fT vs. 1/IG data from many devices,
lies in the range 6-8x10 cm/s.
19
Gate
Drain
Source
undoped AIGaN (14%) 10 nm
2e18/cm3 AIGaN (14%) 12 nm
undoped AIGaN (14%) 5 nm
insulating GaN 2 urn
AIN Buffer Layer
Semi-Insulating 4H-SiC
Figure 7: Cross-sectional view of the GaN/AIGaN HEMT
structure grown on a semi-insulating 4H-SiC substrate.
On-wafer measurements were performed on a Maury load-pull system at 10 GHz and a
drain bias of 30 V. A power sweep for a 0.125 mm HEMT (LG = 0.45, LGS = 0.5, and L0D = 1.0
urn) is plotted in Figure 8. The most significant result was a record RF power density of 6.9
watts/mm with a PAE of 51 % and an associated gain of 9 dB, that was achieved at -6.0 dB of
compression (-4.3 W/mm at -1 dB compression). This power density is about 8 times higher
than typical GaAs devices, and is more than twice as high as any other GaN HEMT grown on a
sapphire substrate, affirming the potential for AlGaN/GaN HEMT's on SI 4H-SiC substrates for
use in high power amplifiers at X-band. When tested at 16 GHz, this device showed a CW
power density of 4.4 W/mm with a PAE of 27% and an associated gain of 6.9 dB at the -3 dB
compression point, which also demonstrates their advantage for high-power Ku-band operation.
0.125 mm GaN/AIGaN HEMT on SI 4H-SiC
LG = 0.45 nm
Los = 0.5 |im
L„ = 1.0nm
15
20
Power In [dBm]
25
Figure 8: A 10 GHz CW power sweep for a 0.125-mm GaN/AIGaN
HEMT on SI 4H-SiC (performed at the Air Force Research Laboratory,
SNDM) that shows a record power density of 6.9 W/mm. The device was
biased at Vm = 30 V and FGS = -2.25 V.
20
In order to measure total RF power on larger parts, HEMT's with 3 mm of gate width were
packaged with ceramic input and output matching networks. After thinning the SI 4H-SiC
substrate to approximately 4 mils, the HEMT was packaged into a hybrid matching network on a
carrier that was maintained at 18°C. The power sweep shown in Figure 9 represents a single 3mm device tested at 7.4 GHz. During the part of the sweep up to 30 dBm input power, the device
was biased at VDS = 28.4 V and Vas = -1.8 V. The drain bias was changed to 31 V for the last
three data points to achieve a maximum output power of 9.1 watts, a PAE of 29.6 % and an
associated gain of 7.1 dB (-2.1 W/mm at -1 dB compression). The high parasitic source
inductance introduced into the hybrid circuit by the Au ribbon bonds reduced the gain and
frequency response, and the characterization of the 3-mm HEMT at 10 GHz was unwarranted.
When source via holes are incorporated in the GaN-on-SiC HEMT process, the total RF power is
expected to improve, even at 10 GHz.
3.0 mm GaN/AIGaN HEMT on SI 4H-SiC
„P„=9.IW .
LU
<D.
25
30
35
Power In [dBm]
Figure 9: A 7.4 GHz CW power sweep for a 3-mm GaN/AIGaN HEMT
on SI 4H-SiC in a hybrid matching circuit. For the initial part of the
sweep, the device was biased at VDS = 28.4 V and Vos = -1.8 V. The drain
bias was changed to 31 V for the last three data points to achieve a
maximum output power of 9.1 watts.
CONCLUSION
The progress in the power performance of SiC MESFET's has been rapid, and these devices
are now generating considerable attention for both military and commercial applications. The
combination of high electrical power density of the devices and the high thermal conductivity of
the SiC substrates enables power levels that are not achievable with other solid state
technologies, particularly for CW communication applications. At power levels of 80 W per
chip, SiC MESFET's would have a significant power advantage over Si LDMOS or GaAs
MESFET's in base station power amplifiers. As SiC MESFET's mature into a MMIC
technology, they should have an impact on solid state power applications through X-band.
Additionally, recent progress with GaN/AIGaN HEMT's fabricated on semi-insulating 4H-SiC
substrates shows this to be an exciting technology with tremendous potential for high power
applications at X-band and above.
21
ACKNOWLEDGMENTS
The SiC material development was sponsored by DARPA, monitored by AFWL on contract
no. F33615-95-C-5426. The SiC MESFET work was sponsored by the Naval Research
Laboratory, contract no. N00014-96-C-2152, and by a DARPA MAFET Thrust 3 program,
contract no. F33615-96-C-1967. The GaN/AlGaN HEMTs were funded on a Phase II SBIR
from the Army Research Laboratories, contract no. DAAL01-97-C-0108. The load-pull
measurements of Fig. 8 were performed by Lois Kehias and Thomas Jenkins of AFRL.
REFERENCES
[1] R.P. Joshi, J. Appl. Phys. 78, 5518 (1995).
[2] G.A Slack, J. Appl. Phys. 35,3460 (1964).
[3] C.E. Weitzel, J.W. Palmour, C.H. Carter, Jr., K.J. Nordquist, Elect. Dev. Let. 15, 406
(1994).
[4] S. T. Sheppard, K. Doverspike, W. L. Pribble, S. T. Allen, J. W. Palmour, L. T. Kehias, T. J.
Jenkins, 56th Annual Device Research Conference, 1998.
22
CURRENT STATUS OF SIC POWER SWITCHING DEVICES: DIODES & GTOS
S. SESHADRI, A.K. AGARWAL1, W.B. HALL2, S.S. MANI3, M.F. MACMILLAN, R.
RODRIGUES4, T. HANSON4, S. KHATRI4 AND P.A. SÄNGER
ESSS, STC Northrop Grumman Corporation, 1350 Beulah Road, Pittsburgh, PA 15235-5098,
sseshadr@aeslcad.essd.pa.northgrum.com
'Currently at Cree Research, Inc. 4600 Silicon Drive, Durham, NC 27703
2
Northrop Grumman Corporation, Baltimore, MD
'Currently at Sandia National Laboratory, Albuquerque, NM
4
Silicon Power Corporation, Malvern, PA
ABSTRACT
The progress that has been made in SiC diodes and GTOs is reviewed. A 100 A/1000 V SiC pi-n diode package, the highest current rating reported for any SiC device, a 69 A conduction/11
A turn-off of a SiC GTO and MTO , as well as the first all-SiC, 3 phase Pulse Width Modulated
(PWM) inverter are reported, herein, for the first time. The inverter achieves voltage controlled
turn off with a high temperature capable, hybrid SiC JFET. Material and process technology
issues that will need to be addressed before device commercialization can be realized are
discussed.
INTRODUCTION
The potential of SiC devices for power switching applications is well recognized.1'2 Several
markets have been identified for these devices (see Table 1). It is easily seen that there is a greater
chance for commercial success of SiC power switching devices over the relatively more mature
SiC RF devices. Demonstration of 5.9 kV p-i-n and 1700 V Schottky diodes, 2.6 kV MOSFETs
and 1.4 kV Gate Turn-Off (GTO) thyristors are illustrative of the significant advances that have
been made in the power area. However, these successes on small, low current devices now need
to be followed by demonstration of scaled up devices. While applications of devices with voltage
ratings > 9 kV currently appears to be the realm of SiC devices, lower voltage applications face
Table 1
Potential markets for SiC devices (in order of relative size)
Application area
Typical device
Motor controllers for Electric Vehicles
Schottky diode, MOSFET
Engine and airflow controls for aviation electronics
Schottky diode, MOSFET,
sensors
Electric Utility Power applications
P-I-N diode, Thyristors
Industrial High voltage power electronics
P-I-N diode, MOSFET/IGBT
Ballasts for flourescent lamps
MOSFET
Conventional ICE automobile sensors/electronics
Sensors
HDTV transmitters
SIT
Surveillance andtactical radars
MESFET, SIT
. 23
Mat. Res. Soc. Symp. Proc. Vol. 572 ® 1999 Materials Research Society
significant challenges from the concurrent development of alternative technologies. The expected
improvement of silicon technology, illustrated by developments in silicon IGBT and MOSFET
devices, as well as interest in other materials for power device applications3 are cases in point of
the need for timely identification and resolution of outstanding technical issues to obtain
continued performance and reliability improvements of SiC power devices. This paper focuses
only on issues related to the development of SiC diodes and thyristors.
Material quality
Material quality is an important issue for the fabrication of high power devices. Much of the
research being performed on high voltage devices are presently aimed at devices having ratings of
at least 1-10 A/~l kV, increasing to 100-1000A/5 kV. The need to minimize on-state voltages
limits practical devices to current densities of ~ 500 to 1000 A/cm2. Such a rating is sufficient to
enable conductivity modulation, without significant carrier-carrier and Auger scattering induced
resistance and will require device sizes between 5 and 20 mm2. However, present devices are
limited by catastrophic voltage degradation due to hollow-core screw dislocation ("micropipe")
bulk defects which exist in densities of as low as 30 cm"2. Thus, device size is limited to ~1 mm2.
Progress in this area is incremental. Because, it is well known that the distribution is NOT uniform
one approach to device scaling has been to selectively place devices in sufficiently large
micropipe-free areas of the wafer. This approach neglects the contribution of other defects.
Another approach would be to parallel many smaller devices. While possible for unipolar devices,
the success of this strategy will have to be demonstrated for bipolar devices since current hogging
in the on-state and uniform switching to prevent catastrophic destruction are potential problems,
especially forGTOs.
Fig. 1 shows the breakdown voltage of devices of various sizes. It is clear from the figure that
devices much smaller than 1 mm2 have severely degraded breakdown characteristics. The
observed degradation is consistent not with micropipe defects, but rather with ~103 to 105 cm"2
defect densities associated with closed core screw dislocations in SiC substrates. Such defects
have been found to result in higher pre-breakdown reverse leakage currents, softer reverse
breakdown I-V curves, and highly localized microplasma breakdown current filaments with spacecharge limited conduction in small diameter, low voltage (<250 V) p-i-n diodes. Encouragingly, a
positive temperature coefficient of breakdown voltage (PTCBV) — important to prevent localized
self-heating of device to destructive breakdown — has been observed even for devices with a few
screw dislocations. A failure power
density of ~2 MW/cm2, ~5x greater
100
than for silicon rectifiers have been
£ 80 FHrrlF
reported for those devices.5 However,
60
it is not yet clear whether these results
apply to higher voltage/large area
S! 40
-»-NGC
devices where large numbers of
I 20 :.:„|..U:.||)||V • • • • Kimotoet.al.
dislocations or the higher power
" 0
density could prevent true avalanche
1
10
100
1000
breakdown. This is important because,
Device Area (x!0A5 cmA2)
if true, it would result in a detrimental,
effective
negative
temperature
Fig. 1: Dependence of breakdown voltage on device
breakdown
voltage
coefficient,
area. Kimoto, et al. Data from ref. 4
leading to current filamentation and
• 'v.
I
•
:
;•:»
24
catastrophic breakdown. Preliminary measurements on devices made at Northrop Grumman
indicate that a positive temperature coefficients can be achieved for multi-kV devices. However,
in accord with the data in figure 1, many devices have a negative coefficient, making the
paralleling approach a more viable approach than the selective placing of large area devices
because smaller devices can be used. Nevertheless, questions about how these dislocations affect
diode and bipolar gain devices such as the Bipolar Junction Transistor (BJT), Insulated Gate
Bipolar Transistor (IGBT) and GTO thyristors still remain.
Devices with 5 kV ratings will require -25 to 50 \nm thick epitaxial drift regions with doping
concentrations < lxlO15 cm"3. While such parameters are reasonably achievable with good
morphology using current epitaxial growth processes, epitaxial defects such as "tetrahedral pits'1,
"comet tails" and "carrots" do exist and can also degrade device performance. The geometry of
these devices locally reduces drift layer thickness, enhancing the peak electric field. Conventional
wisdom places their origin at bulk defects or other surface features such as polishing marks. As in
the case of bulk defects, de-rating the maximum field (i.e. voltage) rating of a given epitaxially
grown drift region generally increases the on-resistance and stored charge resulting in increased
forward drop and switching time , respectively, of a device. Moreover, the still uncontrollable
minority carrier lifetime of these layers affects the on-state characteristics of high voltage diodes
and GTOs. This, combined with the certainty of higher than desirable dislocation densities for the
forseeable future makes the reduction of epitaxial defect densities an imperative.
Minority carrier lifetime is an important parameter for bipolar devices. While some
measurements have been made on representative material, its value on typical grown layers is not
generally well known and more attention is needed in this area. Measurements performed on
epitaxially-grown low voltage p-i-n mesa diodes indicates that lifetime is perimeter governed,
ostensibly due to poor device passivation. This has important implications for bipolar devices
including GTOs where the necessity for high turn-off gains places importance on narrow anode
fingers. Most of this problem is undoubtedly due to the poor quality of mesa side walls which
forced the abandonment of SiC UMOSFETs in favor of planar DMOS geometries. Planar devices
have the advantage of eliminating oxide growth on surfaces containing both Si and C atoms. Since
most wafers are grown along the c-axis and prepared with the Si face (as opposed to the C face)
as the surface, oxide quality on this surface is expected to more closely resemble the silicon
system. However, such an approach results in the need for ion implantation technology.
Activation of ion implants (N and more recently P for n-type, Al or B for p-type) has been
shown to occur at temperatures above 1500 C. Low dose (~1016 - 1017 cm"3) implants are needed
for MOSFET, IGBT and GTO gate regions, as well as for the Junction Termination Extension
(JTE) edge termination technique favored for high voltage devices. Activation is most critical for
JTE, since its implementation requires control of the total charge in the implanted layer. A
reduced activation temperature is important to prevent step bunching. In this regard, P
implantation shows promise, having been recently shown to result in lower sheet resistance.
However, both P and Al have relatively higher masses than their counterparts, resulting in
shallower implants with more crystal damage precisely where one does not want it - the device
active junction. Indeed, some identified problems related to poor activation include single
injection-induced voltage snapback in implanted diodes, poor ohmic contacts and poor turn-on
behavior of GTOs. Nevertheless, Al is preferred over B for applications such as guard ring
terminations where the greater B diffusion coefficient may result in poor lateral definition. A good
ohmic contact is important for optimizing device current and on-state voltage rating. Once again,
25
p-type contacts remain the main problem. High dose implants using AVC co-implants have been
shown to yield sufficiently low contact resistances.
Devices
PIN diodes
SiC p-i-n diodes are expected to be
the diode of choice above ~2 kV and in
l.OE+03
:M&;:
applications requiring temperatures above
1.0E+02
-150 C because of the soft breakdown
«U.OE+Ol
/£--*'
and lower barrier height of Schottky
| 1.0E+00
| 1.0E-01
devices. Cree Research has reported an
u
1.0E-02
impressive 5.9 kV JTE-terminated
1.0E-03
implanted diode, the highest blocking
100
1000
10000
voltage yet achieved for this device.
Breakdown Voltage (V)
While the on-resistance was not reported,
this device had a 5.4 V forward drop at Fig. 2: Current and breakdown voltage rating for some
100 A/cm2. However, the device size was recent SiC p-I-n diodes.
only 100 |Xm dia. and the authors were
not able to test the breakdown stability. As with the Schottky diodes, large area devices suffer
degraded voltage breakdown. Other workers have reported 1 mm2 planar, Al-implanted JTEterminated p-i-n diodes of 6 A/3.3 kV (Ec ~ 2.5 MV/cm) and 5 A/2 kV (Ec ~ 2.1 MV/cm) with
PTCBV to 150 °C, sustainable avalanche current density of 20 mA/cm2, 4.8 V @ 70 A/cm2 and
4.0 V @ 500 A/cm2, corresponding to on-resistances of 3 mQ-cm2 and 2.2 mQ-cm2,
respectively.6 The authors estimated a rather low temperature insensitive, lifetime of < 100 ns
from turn-off measurements. The results of these and other recent work in the literature are
shown in Fig. 2. Also shown in the figure are our most recent results (described below).
Fig. 3 illustrates the first demonstration of a 100 A/1000 V diode package, the highest power
rating yet reported for a SiC diode, using a parallel arrangement of 55 400 & 600 um dia. planar,
Al-implanted guard ring terminated diodes with 10 ^m drift layers. Measurements performed in a
high temperature dielectric fluid showed that PTCBV could be obtained on individual devices up
to 100 °C with a sustainable avalanche breakdown current density of 16 mA/cm2. Additionally, 12
A/ 2.45 kV was achieved in a single (unpackaged) 1 mm dia. diode with 20 ^m drift layer.
However, in both instances, poor ohmic contacts formed by an oxygen leak during the sintering
process were responsible for excessively high on-state voltage at room temperature. The lack of
reduction in on-state voltage above 150 °C is thought to be due to non-ideal package and parasitic
device resistance. It should be mentioned that no evidence of single injection related snapback has
been observed in these devices.
Thyristors
The conductivity modulation possible with bipolar devices such as the Bipolar Junction
Transistor (BJT), Insulated Gate Bipolar Transistor (IGBT) and thyristor make these devices
more favored than unipolar devices such as the MOSFET for high voltage applications. High
power BJTs have not, to date, been developed in SiC. SiC IGBTs have the same temperature
26
VERT/DIV
20 A
CURSOR
(b)
(a)
Fig. 3: Temperature dependent (a) reverse and (b) forward I-V characteristic of a 100 A/ 1000
V SiC pin diode package consisting of 15 parallel 600(im dia. diodes.
reliability and high field sensitivity issues faced by conventional SiC MOSFETs.7'8 These facts
make the SiC thyristor presently the best option for high temperature, high voltage applications.
A thyristor is a four layer devices with the emitters of the upper (pnp) and lower (npn)
transistors form the anode and cathode contacts, respectively, with the n-base of the pnp forming
a gate contact. A common collector/base junction supports the large positive and negative
voltages during their blocking state. Thyristors are turned on by a small gate current applied to
the upper transistor in the off-state, resulting in low current gain transistor-like characteristics
until the sum of the common base gains of the two transistors exceeds 1, at which time
regenerative feedback turns the collector base into forward bias, leaving both transistors in
voltage saturation. This results in a low on-state voltage of approximately a single diode drop.
The devices turn off when the corresponding main current or voltage goes to zero. On-state
dissipation considerations due to the larger junction voltage of SiC ensures that thyristor switches
become more favorable only at very high voltages or in applications where the high temperature
reliability issues of the MOSFET cannot be adequately compensated. A lower voltage range is
also possible for applications where the greater current handling capability (with its attendant
smaller device size) are at a premium. Thyristors with high voltage (900 V), high power (700 V/6
A), high temperature (500 °C), and increased reliability (> 600 hr lifetime) have already been
demonstrated.9
Gate Turn-Off thyristors are more favored for DC power switching applications because of
their independent gate turn off capability. These devices, therefore, do not need to support large
negative voltages in this application. This simplifies their design, resulting in an asymmetric design
in which the drift region is made thinner by insertion of a higher doped buffer layer in the lower
base layer to prevent the occurrence of punch through breakdown. The problem of terminating
the lower junction is also eliminated as only the common base/collector junction need support any
voltage. The insertion of the buffer layer increases the turn-off gain by reducing the common base
gain of the npn transistor while simultaneously reducing the likliehood of open base breakdown in
the npn transistor which limits the off-state voltage. The thinner drift layer of such a structure
27
reduces the on-state voltage drop and the stored charge, decreasing the power dissipation or
increasing the switching capability of the device. SiC GTOs are inverted with respect to their
silicon counterparts because of the need to avoid using highly resistive p-type SiC substrates.
We have previously demonstrated GTOs with on-state voltages of 2.7 V at 390 C and 3.35 V
at room temperature for a current density of 500 A/cm2. The large decrease in voltage with
increasing temperature occurs because of the
!
VERT/CIV
j
10 A
decreasing SiC band gap, thereby lowering the
[-25C
CURSOR
knee voltage, and because increased
*
ionization of n- and p-type impurities of the
H0RIZ/DJV
±
2 V
anode and substrate lowers the on-resistance
CURSOR
I
of these regions. Room temperature contact
resistances for the best devices were a very
PER STEP
SIGaA
low 2.4 x 10"5 Q-cm2. Like the p-i-n diodes,
OFFSET
|
the robustness of the GTO and packaging
B or sm/OIV
1
approach was demonstrated by the conduction
20
&>A<*n\
X OF COLLECTOR
of 69 A (4300 A/cm2) through fifteen GTOs
PEAK VOLTS
-4300
A/an-'
33.7
connected in parallel in high power package
(see fig. 4). These devices did not show any
indication of current hogging for the current Fig. 4: On-state current through 14 SiC C;TOS
comlected in parallel.
densities measured.
1
""J"
~4QOMA
Turn-off gain (Wig) of GTOs is generally dependent on the exact geometry of the gate and
anode. Measurements shown in Fig. 5 on interdigitated devices demonstrate from a relatively high
value of 8 at just above the holding current density (-40 A/cm2), when the device is barely latched
on and there is very little minority carrier stored charge in the drift layer, to a minimum of just
under 2 at current densities >100 A/cm2, which occurs once conductivity modulation occurs and
the drift region is saturated with minority carriers. Thus, one expects that switching times will be
dependent on both cathode current and peak GTO gate current with the slowest turn-off
occurring for operation at high turnoff gain and high on-state current densities. The fastest turnoff is then achieved by operating under unity gain conditions with low on-state current density.
However, the desire for fast turn-off is
moderated by the desire to minimize
device area. The dependence of the turn5a 8
off current gain on the cathode current is
u 6
IS
the reason for the complexity of GTO
IU 4
gate drives. We have chosen a trade-off
in which we operate at a current density
of 500-1000 A/cm2. Our demonstration
100
1000
10000
10
of unity gain switching times of < 200 ns
Ew. *!**» Cathode current density (A/cm2)
for SiC GTOs at a current density of
1500 A/cm2, therefore, suggests very
Fig 5: Turn-off gain versus cathode current for a typical strongly that SiC GTOs will live up to
SiC GTO.
their expectations for fast switching.
Storage, rise and fall times of-113,43 and 100 ns, respectively, have previously been reported
for SiC thyristors at a very low current density of 30 A/cm2.10 Maximum turn-off current densities
> 5000 A/cm2 have been achieved both by us and reported by others.11 However, the inverse
28
relationship between this gain and anode finger width suggests that a trade-off must be made
between maximum turn-off gain, maximum current density (on-state voltage) and device foot
print within the previously prescribed range of -500 to 1000 A/cm2.12 We have previously
reported GTO operation at 325 °C with switching times increasing by a factor of 4 from room
temperature due to an increasing minority carrier lifetime. We have also demonstrated repetitive
GTO switching up to 350 V (3A) with 3.2 uJ switching losses consistent with the lower stored
charge in these devices compared to Si GTOs.
We have now extended the current rating of packaged GTOs to 11 A (@ 800 A/cm2) (see Fig.
6 (a)) with switching times < 100 ns using the same package as that demonstrating 69A
conduction without degradation in their turn-off current characteristics. This result represents the
largest turn-off current reported to date and demonstrates the viability of the paralleling approach
with respect to turn-off. Turn-off of 69 A was not expected to be achieved because the structure
employed had a peak turn-of current density of -1000-1500 A/cm2. Unfortunately, the devices
used had poor breakdown voltage so no high voltage measurements could be taken.
Fig. 6: Turn-off transients for 14 SiC GTOs connected in parallel, (a) GTO mode, (b) MTO' mode. 11A
corresponds to a current density of ~800 A/cm2.
We now examine the turn-off characteristics of the 69A GTO package in Fig. 6a shows typical.
Before the application of the turn-off gate pulse the device is in its on-state with all three junctions
forward biased. Thus, both transistors are in saturation. Application of the turn-off gate pulse
induces the gate-to-anode junction into reverse bias, causing VAK to climb towards zero almost
instantaneously. After -80 ns, the storage time (ts) of the device, the minority carrier
concentration at the middle junction decreases below its equilibrium value, reverse biasing it and
allowing the voltage to fall to its off-state value. Concurrently, both base and cathode currents
decrease rapidly with a characteristic fall time, tf. The absence of the tail current typically seen in
silicon bipolar switches greatly diminishes the switching losses of the SiC GTO and is indicative of
the fact that the residual minority carrier density in the base regions is very small, as expected
because the lOx larger critical field of SiC enables commensurately thinner drift region, leading to
lOx lower stored charge.
Voltage controlled GTO turn-off
Voltage control is especially important during turn-off because of the low turn-off current gain
of the GTO. A hybrid MTO package containing a silicon power MOSFET with a SiC GTO was
29
assembled to examine the MOS-gated turn-off characteristics of SiC thyristors. The device
operates with the MOSFET shunting the gate-to-anode junction in a manner analogous to a
removable cathode short in silicon GTOs. The MOSFET in this arrangement is external to the
device and thus can be independently optimized without having the oxide subjected to large
electric fields. Furthermore, it is only a low voltage device, since it will never see more than the
gate/anode breakdown in the worst case. Fig. 6 (b) illustrates the switching behavior of this hybrid
package. Voltage controlled turn-off of an equivalent current to that obtained in the GTO mode
switching is clearly demonstrated, indicating that sufficiently large current levels can be easily
switched using this device configuration. The turn-off is slower than for the GTO in part because
of the fact that in the GTO-mode turn-off, the gate-to-anode junction is strongly reverse biased,
whereas in the MTO-mode turn-off this junction is simply held near zero voltage during the
storage cycle. The fact that we see no noticeable degradation of turn-off capability while
paralleling these many devices under these more restrictive conditions also bodes well for the
paralleling approach to device scaling.
The dV/dt rating of a GTO represents the maximum rate of rise of the voltage before the onset
of unwanted retriggering of the device due to capacitive displacement currents. Thus, it partially
determines the maximum device operating frequency. A high rating also has the added advantage
of reducing or eliminating the need for snubber components. Typical silicon GTOs have ratings in
the 600-1000 V/^s range. We have measured ratings as high as 700 V/us in GTO mode,
increasing to >1400 V/(is with the use of resistive shunts (as high as 100 ß) such as the
MOSFETs connected across the gate and anode, sufficient to warrant snubberless operation of
inverters for motor drive control applications.
The high temperature reliability concerns of MOSFETs has been previously mentioned.
Therefore, a fhyristor with high temperature, voltage-controlled turn-off capability can be
achieved by simply replacing the MOSFET with a SiC JFET, resulting in a Junction Controlled
Thyristor. Like the MTO , this device need not be integrated into one chip, thereby allowing
both devices to be independently optimized. Moreover, the JFET, like the MOSFET, need only be
a low voltage device, since it is never subjected to voltages above the relatively low gate/anode
reverse breakdown voltage.
Turn-off transient was measured on a hybrid device constructed from a 4H-SiC GTO chip and
a vertical 4H-SiC JFET (Fig. 7). The JFET had an on-resistance of 2.5 ohm with a gate to source
bias, Vgs, of+3 V. The JFET can be turned
off with V^ = -1 V or less. Turn-on was
achieved by forward biasing the anode to gate
junction of the GTO while keeping the JFET
in off condition through an application of -7.0
V on the gate of the JFET. In order to record
the turn-off transient, the JFET was turned on
by an application of +3 V on the gate. This
presents a 2.5 ohm resistance between the
anode and the gate of the GTO thus diverting
all of the anode to cathode current through
the JFET. The turn-off time was less than Fig. 7: Turn-ofT characteristics of a SiC JCT at a
100 ns. These results clearly demonstrate the current density of 1000 A/cm2.
functionality and potential of the hybrid JCT
30
for a higher switching speed up to 50-100 kHz.
All switching tests to date have been performed under resistive loading conditions. However,
typical applications require switching under inductive loading conditions, where significant
voltage overshoot occurs. Devices have not yet been tested for their ruggedness at the limit of
breakdown. Preliminary switching comparisons suggested that SiC GTOs had fall times and turnoff energy losses ~10x lower than for comparable SiMCTs.13
SiC inverter
The rapid progress of SiC power device
technology has led to the development of
the first all-SiC inverter for PWM
applications. Fig. 8 shows a schematic of
this circuit, which consists of 3 phase
switching with each phase consisting of an
upper and lower switch containing a SiC
GTO with an anti-parallel SiC pin diode Fig. 8: Schematic diagram of PWM circuit. Three
and a low voltage SiC JFET connected phases each consisting of an upper and lower pole. Each
across the GTO gate and anode. In a pole contains contains SiC GTO, JFET and diode.
significant improvement over silicon
devices, no snubber circuits are used or expected to be necessary because of the high dV/dt of the
voltage controlled switch. Fig. 9 illustrates the entire PWM system, including the gate drive
circuits and the motor, used in the demonstration.
In the demonstration, a DC bus voltage of 45 V is applied across the load, a 0.25 hp motor of
the type used to re-circulate the coolant in an automobile. The GTOs conduct peak and average
currents of 1.3 A and 200 mA,
respectively, with the upper switches
open for a phase angle of 120° and the
lower switches being pulsed at a 4 kHz
frequency over the same interval. Fig. 10
shows the operation of the inverter,
with the phase current (i.e. current
through one of the three phases of the
motor) integrating to its maximum
value with the pulsed bottom switch.
The current and voltage levels are
presently too low to determine typical Fig. 9: Photograph of the first all-SiC PWM inverter for
switching characteristics
of the motor drive control.
individual SiC components. Tests are
currently underway to examine these issues up to the operational limits in voltage, current,
frequency and temperature. We look forward to reporting these results in the near future.
CONCLUSIONS
Progress in SiC p-I-n diodes and GTOs has been reviewed. Among the highlights were
demonstration of a 100A/1000V diode package, 69A conduction/11 A turn-off of a GTO
31
package, MOSFET and JFET controlled turn-off
of SiC thyristors and all-SiC PWM for motor
control applications. The above results are
indicative of the fast pace with which these
devices are scaling up to power levels needed in
typical power switching applications. Materials
limitations affecting further improvements were
also addressed.
,:W1I1
ACKNOWLEDGEMENTS
Fig. 10: Phase voltage and current waveforms of
The authors thank DARPA (contract # 786- the SiC PWM operating at 4 kHz. The narrow
DTD-96FEB01),
DARPA
(contract
# spikes in the voltage waveform represents the offF342/14)/EPRI (Order # WO8069-09) and Army state of on of the lower poles.
(contract # F33615-96-3-2608) for their support
of the work described in this paper. The support of V. Hegde and the technicians of the Advanced
Electronic Structures Lab, especially J.C. Kotvas, at STC is also gratefully acknowledged.
REFERENCES
'K. Shenai, et al. IEEE Trans. Electr. Dev., V. 36-9,1989, p. 18111.
M. Bhatnagar and B.J. Baliga, IEEE Trans. Electr. Dev., V.40-3, 1993, p.645.
3
T. P. Chow, N. Ramungul and M.Ghezzo, Mat. Res. Soc. Symp. Vol. 483,1998, p. 89.
4
T. Kimoto, N. Miyamoto and H. Matsunami, IEEE Trans. Electr. Dev. V.46-3, (1999) P. 471.
5
P. G. Neudeck, W. Huang and M. Dudley, IEEE Trans. Electr. Dev. V.46-3, (1999) p. 478.
See also P.G. Neudeck and C. Fazi, IEEE Trans. Electr. Dev. V.46-3, (1999) p. 485.
6
H. Mitlehener, et al., Proc. Of 19989 Int'l. Symp. On Power Semicond. Devices & ICs (SPSD
'98), Kyoto, Japan, June 3-6,1998, p. 127.
7
A.K. Agarwal, S. Seshadri andL.B. Rowland, IEEE Electr. Dev. Lett. V18, (1997) p. 592.
8
M.M. Maranowski and J.A. Cooper, Jr., IEEE Trans. Electr. Dev. V.46-3, (1999) p. 520.
9
J.W. Palmour, R. Singh, LA. Lipkin and D.G. Waltz, Trans. 3rd Int'l. Conf. on High Temp.
Elec. (HiTEC) Alb. N.M. Vol 2, XVI-9-14, Jun, 1996.
10
Z. Xie, J. Fleming and J.Zhao, IEEE Electr. Dev. Lett. 1995.
11
B. Li, L. Cao and J. Zhao, IEEE Electr. Dev. Lett. (1998).
12
K. Xie, J.H. Zhao, J.R. Flemish, T. Burke, W.R. Buchwald, G. Lorenzo and H. Singh, IEEE
Elec. Dev. Lett. 17,142 (1996).
13
J. Mooken, R. Lewis, J.L. Hudgins, A.K. Agarwal, J.B. Casady, S. Siergiej and S. Seshadri,
Proc. 32nd IEEE Industry Applications Society (IAS) Conference, Oct. 5-9, New Orleans,
LA 1997.
2
32
THE EFFECTS OF DAMAGE ON HYDROGEN-IMPLANT-INDUCED
THIN-FILM SEPARATION FROM BULK SILICON CARBIDE
R.B. GREGORY *, O.W. HOLLAND **, D.K. THOMAS **, T.A. WETTEROTH *,
S.R. WILSON*
Motorola Inc., Semiconductor Products Sector, Tempe, Arizona
**Oak Ridge National Laboratory, Solid State Division, Oak Ridge, TN. 37831-6048
ABSTRACT
Exfoliation of SiC by hydrogen implantation and subsequent annealing forms the basis for a
thin-film separation process which, when combined with hydrophilic wafer bonding, can be
exploited to produce silicon-carbide-on-insulator, SiCOI. SiC thin films produced by this
process exhibit unacceptably high resistivity because defects generated by the implant neutralize
electrical carriers. Separation occurs because of chemical interaction of hydrogen with dangling
bonds within microvoids created by the implant, and physical stresses due to gas-pressure
effects during post-implant anneal. Experimental results show that exfoliation of SiC is
dependent upon the concentration of implanted hydrogen, but the damage generated by the
implant approaches a point when exfoliation is, in fact, retarded. This is attributed to excessive
damage at the projected range of the implant which inhibits physical processes of implantinduced cleaving. Damage is controlled independently of hydrogen dosage by elevating the
temperature of the SiC during implant in order to promote dynamic annealing. The resulting
decrease in damage is thought to promote growth of micro-cracks which form a continuous
cleave. Channeled H+ implantation enhances the cleaving process while simultaneously
minimizing residual damage within the separated film. It is shown that high-temperature
irradiation and channeling each reduces the hydrogen fluence required to affect separation of a
thin film and results in a lower concentration of defects. This increases the potential for
producing SiCOI which is sufficiently free of defects and, thus, more easily electrically activated.
INTRODUCTION
Hydrogen implantation through an oxide film followed by hydrophilic wafer bonding and a
thermal cycle is a process developed to cleave a thin film of silicon-on-insulator (SOI).1 The
process has recently been applied to produce silicon carbide-on-insulator (SiCOI) films for
possible use as a wide bandgap semiconductor in power rf and switching devices.2
SiC thin films separated from bulk material using this process have measured too resistive, a
condition attributed to damage in the SiC thin film caused by the hydrogen implant itself. The
experiments described in this work are motivated by the desire to understand the implant damage
mechanisms in order to make the separation process more efficient and produce defect-free, lowresistivity SiC.
The problem is illustrated in Figure 1 which shows a schematic of the hydrogen-implantinduced separation process and a channeled RBS spectrum of SiCOI (-500 nm) produced by this
process. Backscattered counts from Si in the SiCOI (integrated over channels 540-640) measure
824 greater than similarly measured counts from virgin SiC. Calculating density using the RBS
data, one measures 1.27 xlO20 displaced atoms/cm3. These vacancies have the potential to cause
33
Mat. Res. Soc. Symp. Proc. Vol. 572 ®1999 Materials Research Society
RBS
500
550
600
Channel No.
Bonded and cleaved SiC
Figure 1. Schematic diagram of transferred and cleaved SiC (left) and RBS-channeling spectra of
the transferred SiC film (right)... H+ implant damage polished off before analysis.
the deactivation of electrical carriers in material with typical doping concentrations of 1017-1018
atoms/cm3.
Previous work (with Si) shows that the process to cleave a thin film by hydrogen
implantation followed by a thermal cycle is a combination of hydrogen chemistry and physical
processes. The implant results in the formation of platelet-like microvoids which, during
subsequent anneal, expand due to gas pressure of excess hydrogen. This links the microvoids
into a continuous fracture, cleaving a thin film from the bulk wafer.3 The use of H+-implantation
to affect transfer of a thin film from a bulk Si wafer was based upon observations of bubbling and
exfoliation of implanted Si wafers after annealing.4,5
The effects of ion-induced damage on the efficiency of the transfer process and its
dependence on H+ dose are demonstrated by observing exfoliation of SiC following H+implantation and anneal. Means to control damage independently of H dose are demonstrated
with elevated-temperature and channeled implantation. It is proposed that channeled
implantation generates less residual damage from the surface to at least half the projected range,
1/2RP, of the implant simply because crystalline axes of SiC are aligned with the H+ beam,
decreasing the cross section for ion-solid collisions. The elevated temperature implants affect insitu, dynamic annealing in order to control H+ implant damage in SiC.
EXPERIMENTS and RESULTS
Random vs. Channeled H+-im plantation
Experiments to measure damage and exfoliation of SiC as a function of H dose were
accomplished using bulk SiC samples, 4H polytype, supplied as research grade material by Cree
Research. They were implanted with 60 keV H+ to doses ranging from 2.5 xlO16 to 10.5 xlO16
atoms/cm2. Samples were tilted 7° from normal to affect random beam alignment. Additional
samples were implanted with the H+ beam aligned to [1000] axes to affect channeled implants
over the same dose range. Damage analyses were accomplished by Rutherford backscattering
(RBS)-channeling using a 2.3 He" ion beam aligned with [1000] axes normal to the surface of the
34
sample. Backscattered ions were detected at 160° relative to the incident beam using a solid
state, surface barrier detector. Samples were then annealed in order to cause exfoliation of thin
SiC from the bulk material. The amount of exfoliation was evaluated using optical microscopy.
Figure 2 shows RBS-channeled spectra for three of the samples from the set generated to
evaluate the effects of implant damage on exfoliation for random-implanted H+. The spectra
represent as-implanted samples at room - or ambient - temperature (RT). These spectra show
that damage to the SiC at the projected range, Rp, increases with the H+ implant dose. The
scattering yields near 1/2 Rp are also progressively greater (as a function of dose) than the yield
from the virgin reference sample indicating the presence of ion-induced, residual damage at this
location in all the samples. This is possibly due to displaced atoms that either dechannel or
directly backscatter the incident He"1^ ions. Analysis by positron annihilation spectroscopy (not
shown) indicates the presence of open volume defects, the result of displaced atoms at
concentrations below the sensitivity of RBS. Such defectivity may be responsible for
deactivating intrinsic carriers in SiC as previously reported for similarly implanted material.6
Following RBS characterization, all samples were subjected to 950°C, 15-minute anneals,
then optically imaged using a microscope with Nomarski contrast. Figure 3 shows a portion of a
series of optical micrographs produced to observe exfoliation of SiC as a function of dose for 60
keV H+ implants done at RT. During the 950°C, 15-minute anneal, bubble formation occurs as
the H+ dose approaches 4.5 xlO'Vcm2, as seen in the optical micrograph [Figure 3(a)], (Samples
implanted with small increments of dose between 2.5 xlO16 and 4.5 xlO'Vcm2 revealed that the
critical dose to produce exfoliation is very near 4.5 xlOl6/cm2.) Evidence for material removal or
exfoliation of the 4.5 xl016/cm2 sample is clearly seen in Figure 3(a) by the appearance of broken
bubbles. The amount of exfoliated surface material maximizes near a dose of 5.5 xlOl6/cm2
[Figure 3(b)], but at higher doses exfoliation decreases as seen in Figures 3(c) and (d), indicating a
retrograde effect of the 60 keV H+ implant to doses greater than 5.5 xl016/cm2.
The information conveyed by the images in Figure 3 is represented graphically in Figure 4
which shows the percentage of area that exfoliates following the 950°C anneal. Two sets of data
are graphed, one for the randomly implanted samples and one for channel-implanted samples.
One sees that channeled implants shift the onset of exfoliation (as well as maximum exfoliation)
to approximately 1 xlO16 lower dose than the random implants. Furthermore, the maximum
io3
X
:
~
Figure 2. RBS-channeled spectra for Si in
60 keV H+-implanted SiC. Three dosages
shown are 4.5 xlO16, 6.5 xlO16, and 10.5
x 10' 6/cm2. Reference spectra include the
aligned yield from nonimplanted (virgin) SiC
and the randomized yield from an implanted
sample.
RANDOM
10s
'AU-V.*'»" ,,«»*«Vt
1.00
1.10
1.20
1.30
BACKSCATTERED ENERGY (MeV)
35
100 um
Figure 3. Nomarski optical micrographs of SiC implanted with 60 keV H+ to dose
(a) 4.5 xlO16, (b) 5.5 xlO16, (c) 6.5 xlO16, and(d) 8.5 xlO'Vcm2, after furnace annealing at
950°C for 15 minutes.
exfoliated area increases from 37% for the random implant (dosed 5.5 xl 0l6/cm2) to 69% for the
channeled implant (dosed 4.5 xlO'Vcm2). The rate of retrograde behavior of exfoliation appears
the same for both random and channeled series.
SIMS depth profiles of hydrogen in random and channel-implanted samples are shown in
Figure 5. Each of the samples was implanted with 60 keV H+ to 2.0 xlO'Vcm2, then annealed.
The dose was held low enough to prevent exfoliation of the SiC during the anneal. The profiles
show that the channeled implant has slightly greater range than the random implant. More
significant, though, the retained hydrogen concentration measures almost three times greater for
the channel-implanted sample.
100a
0)
(B
o
X
HI
<U
<
Figure 4. Percentage of area of
exfoliated SiC as a function of
60 keV H implant dose. Two
sample series graphed, one which
was implanted in a random
direction and one implanted with
samples aligned to [1000] axes.
806040-_
200 -
5^
I
'
I
'
I
'
I
'
I
'
I
'
I
'
I
2
4
6
8
10
H* Implant Dose (x1016/cm2)
KT'J
10
c
0)
D)
O
•a
Channeled H
Figure 5. SIMS profiles of 60 keVimplanted hydrogen,
(7.37x1015/cm2)
2
dose=2xl0l6/cm2,
following anneal at 950CC for
15 minutes.
101
io18-3 RandomI
>.
x 101
HX«ifrn»(2.70x1015/cm")
l
'
I
'
0.4
l
'
l
'
0.8
I
'
l
1.2
'
l
'
l
'
l
1.6
Depth (microns)
36
Elevated Temperature Implantation
To learn the effects of elevated temperature implantation for controlling H+ implant damage,
two sets of 4H-SiC samples were implanted with 60 keV H+. One set was implanted at room
temperature to doses ranging from 3.25 xlO16 to 4.5 xlO'Vcm2. Samples of the second set were
heated to 600°C during implant to doses ranging from 2.25 xlO16 to 8.0 xlO'Vcm2.
Figure 6 shows RBS-channeling spectra for two samples H+-implanted to a dose of 2.0
xlO"Vcm2, one implanted at RT and the other with the temperature elevated to 600CC ("hot"
implant). It is clear from comparing the scattering yields in the respective samples that the hot
implant generated less damage at Rp as well as 1/2 Rp. Also evident is a slightly greater Rp for the
hot implant. Optical micrographs for the series of hot implants (not shown) indicate the
threshold dose for surface exfoliation of SiC during a 950°C anneal is -2.75 xlO'Vcm2.
Optical micrographs shown in Figures 7(a) and (b) compare the surface morphology after
annealing for hot and RT implants, respectively. The images show about the same degree of
bubbling and exfoliation although the H+ implant dose for the RT implant is much higher, 4.5
xl0"7cm2, compared with 2.75 xlO'Vcm2 for the 600°C implant. Previous work shows that the
critical fluence for exfoliation decreases almost linearly with irradiation temperature.7
0.6
0.5
Si DEPTH (microns)
0.4
0.3
0.2
0.1
RANDOM
\
\
£ 102
T-*-V«A.-"~*
S
w
p
0.0
w
-v«
\
•A
X
V +* + * •
10'
— VIRGIN
• RT IMPLANT
600» C IMPLANT
0.9
50 urn
\
P%
1.0
1.1
1.2
1.3
BACKSCATTERED ENERGY (MeV)
Figure 6. RBS-channeling spectra comparing
damage to SiC following 60 keV H+,
2 xlO'Vcm2 implants at R.T. and 600°C.
Figure 7. Nomarski optical micrographs
of 60 keV H+-implanted SiC showing
similar degrees of bubbling and exfoliation
for (a) 2.75 xl016/cm2 @ 600°C and
(b)4.5xlOl6/cm2@RT.
DISCUSSION and CONCLUSIONS
The dependence of surface exfoliation of SiC on H+ dose and the retrograde behavior of
exfoliation as damage increases beyond a specific dose supports the the following model. Both
the hydrogen concentration and the lattice damage affect the degree of exfoliation. Both increase
with implantation dose, but damage retards exfoliation. It is clear that more hydrogen available
within the lattice will lead to more bubbling and exfoliation, but the role of damage in suppressing
the effect is not obvious. As seen from previous work with H+-implanted Si, the formation of
extended defects (i.e., platelets) is critical to the formation of microcracks within the lattice.3
These microcracks and their ability to expand and interconnect yield the large macroscopic
regions within the lattice which become separated from the underlying substrate either during
37
exfoliation or thin-film transfer. One anticipates that substantial lattice damage may inhibit the
formation of such macroscopic regions by hindering or stopping the propagation of the
microcracks and thus preventing them from forming an interconnecting network.
The present work demonstrates the ability to control ion-induced damage independently from
the implant dose by elevating the temperature of a sample to 600°C during the implant in order
to dynamically anneal the SiC and potentially reduce the damage relative to an implant performed
at room temperature. The RBS data show a significant reduction in residual damage (from the
surface tol/2Rp) for the hot implant. The optical micrographs indicate that implanting hot also
allows a substantial reduction in critical H dose needed for cleaving the thin SiC film, resulting
with further decrease in damage.
Channeling the H+ implant dramatically enhances the process of exfoliation. Measurements
of exfoliated area from optical images indicate more robust exfoliation with lower dose relative to
random implantation. The SIMS results suggest less out-diffusion of hydrogen during anneal, but
increased diffusion into the bulk (below Rp) in the channel-implanted sample. This in turn
suggests unique damage morphology at Rp which is not entirely understood.
It appears possible for damage in SiC to reach a concentration great enough to disrupt the
formation of a continuous network of cracks. This conclusion is supported when damage is
controlled independently of hydrogen concentration, either by elevating the temperature of the
SiC during implant, or by channeling the hydrogen, or, quite possibly both. Each of these
methods allows a reduction in critical H+ fluence required to affect separation of a thin film and,
therefore, may provide high-quality SiCOI material.
REFERENCES
1. N. Bruel, Electron. Lett. 37, p. 1201 (1995).
2. L. Di Cioccio, Y. Le Tiec, C. Jaussaud, E. Hugonnard-Bruyere, and M. Bruel, Mat. Sei. Forum
264-268, p. 765(1998).
3. M. K. Weldon, V. E. Marsico, Y. J. Chabal, A. Agarwal, D. J. Eaglesham, J. Sapjeta, W. L.
Brown, D. C. Jacobson, Y. Caudano, S. B. Christman, and E. E. Chaban, J. Vac. Sei. Technol. B
15, p. 1065(1997).
4. W. K. Chu, R. H. Kastle, R. F. Lever, S. F. Mader, and B. J. Masters, Phys. Rev. B 16,
p.3851 (1977).
5. C. C. Griffioen, J. H. Evans, P. C. D. Jong, and A. van Veen, Nucl. Instrum. Methods Phys.
Res. B27, p. 417 (1987).
6. T. Dalibor, C. Peppermuller, G. Pensi, S. Sridhara, R. P. Devaty, W. J. Choyke, A. Itoh, T.
Kimoto, and H. Matsunami, Inst. Phys. Conf. Ser. 142, p. 517 (1996).
7. R. B. Gregory, T. A. Wetteroth, S. R. Wilson, O. W. Holland, and D. K. Thomas, to be
published.
38
CHARACTERIZATION OF Si02/SiC SAMPLES USING
PHOTOELECTRON SPECTROSCOPY
L. I. JOHANSSON*, P- A. GLANS*, Q. WAHAB*, T.M. GREHK**,
TH. EICKHOFF**, W. DRUBE**
*Department of Physics, Linköping University, S-58183 Linköping
**Hamburger Synchrotronstrahlungslabor HASYLAB am Deutschen ElektronenSynchrotron DESY, D-22603 Hamburg
ABSTRACT
The results of photoemission studies of Si02/SiC samples for the purpose of
revealing presence of any carbon containing by-products at the interface are
reported. Two components could be identified in recorded Si 2p and C Is core level
spectra. For Si 2p these were identified to originate from SiÜ2 and SiC while for
C Is they were interpreted to originate from graphite like carbon and SiC. The
variation in relative intensity of these components with emission angle was first
investigated. Thereafter the intensity of the different components were studied
after successive Ar+-sputtering cycles. Both experiments showed contribution
from graphite like carbon on top of the oxide but not at the interface.
INTRODUCTION
The defect density at the oxide/semiconductor interface is an important
factor for the performance of devices. For Si02/SiC the defect densities obtained to
date are relatively high and one limiting factor for the formation of a high quality
oxide is considered to be a carbon containing by-product at the interface. Studies
of oxide layers thermally grown on SiC have earlier been reported using Auger
Electron Spectroscopy (AES) [1] , transmission electron microscopy (XTEM) [2]
and X-ray Photoelectron Spectroscopy (XPS) [3]. The XTEM results showed a
homogeneous Si02 layer with a well defined interface and the AES results showed
an oxide layer free from carbon related compounds except for a region very close to
the interface. In the angle resolved XPS study [3] an interface Si4Ci.x02 (x<2)
oxide was revealed and also presence of Si4C404 at the surface and in the oxide.
These findings motivated our investigation of Si02/SiC samples using
synchrotron radiation.
By using a high photon energy (3.0 keV) a direct and simultaneous probing
of the SiC substrate, the interface and the oxide layer was first made. The probing
depth was varied by changing the electron emission angle. Recorded Si 2p and
C Is core level spectra showed two components. The relative intensities of these
were extracted and compared to calculated intensity variations. The analysis
showed a graphite like carbon layer on top of the oxide but not at the interface.
Secondly the composition in the surface region was studied after successive Ar+sputtering cycles using lower photon energies, giving a much smaller probing
depth.. The result obtained was the same, i.e. that contribution from graphite like
39
Mat. Res. Soc. Symp. Proc. Vol. 572 ® 1999 Materials Research Society
carbon on top of the oxide but not at the interface could be identified,
findings are presented and discussed below.
These
EXPERIMENT
Two samples with different oxide thicknesses were prepared on n-type Siterminated 4H-SiC substrates, obtained from CREE Research, with doping
concentrations of around 2 x 1018 cm"3. The SiC>2 layers were thermally grown via a
dry oxidation process at 1100 °C for respectively 30 (Sample A) and 45 minutes
(Sample B), followed by annealing in Ar for 10 minutes at the same temperature.
The Si02 layer thickness was measured by ellipsometry and determined to be 58
and 75 Ä for samples A and B respectively.
Two different beamlines were utilized for the photoemission experiments.
The X-ray wiggler beamline BW2 at HASYLAB [4, 5] and BL 22 at MAX-lab [6].
The end station consisted in both cases of a Scienta hemispherical electron
analyzer. At BW2 a photon energy of 3.0 keV was selected with an overall energy
resolution of 0.8 eV for the chosen beamline and analyzer settings. The electron
emission angle was varied by rotating the sample relative to the fixed analyzer.
At BL 22 photon energies of <800 eV were utilized and spectra were collected at
normal emission after successive Ar+-sputtering cycles. The binding energies were
in these latter experiments determined relative to the Fermi edge of a Ta foil
mounted on the sample holder.
RESULTS
Si 2p and C Is core level spectra recorded at 3.0 keV and at different
electron emission angles are shown in Fig. 1. Only two prominent peaks are
Si2p
hv = 3000 eV
i
A 8e=0°
Cls
h 3e=0°
[[
hv = 3000 eV
j
30°
30°
*-SiC
--SiC
45°
45°
55
1 °
1 Si02W/j|
Graphite like _ J^\ Sip)-,
—
carbon
*~i V//j'
60°
|60°
165°
|65°
\V70°
r
i
i
i
i
i
i
i
r"Tinfrr
,1,1,1,1
|
* Y* ' |" " f
286
282
BINDING ENERGY (eV)
104
100
BINDING ENERGY (eV)
Fig. 1
1 55°
Si 2p and C Is spectra recorded at different electron emission angles
using a photon energy of 3.0 keV.
40
clearly visible in both cases. The ones at lower binding energy correspond to the
bulk peaks of SiC. The ones at larger binding energy correspond respectively to
the Si 2p peak from SiC-2 and to a C Is peak from carbon in a form different from
SiC. The binding energy of this additional C Is peak correspond fairly well with
graphitic carbon [7,8] but the width of the peak is so large that it cannot originate
from an ordered graphite layer, therefore we refer to it as a graphite like peak.
Only relative intensities are shown in Fig. 1 since the spectra have been
normalized to the high binding energy peak. The variation in relative intensity
with emission angle can be utilized to determine from where in the sample the
graphite like carbon signal originates.
The intensities of the components in the Si 2p and C Is spectra were
extracted using a curve fit procedure [9]. The intensity ratio between the SiC-2 and
the SiC peak for Si 2p level (labeled Si below) and the ratio between the graphite
like and the SiC peak for C Is level (labeled C below) were then determined at
each emission angle [10]. The ratios obtained for sample A are shown in Fig. 2,
together with the C/Si ratio multiplied by a factor of 15. The C/Si ratio is seen to
increase monotonically and quite strongly with increasing emission angle. This
indicates that the graphite like carbon signal originates from a carbon containing
layer at the surface and not at the interface, as discussed below.
When applying a simple layer attenuation model to calculate the intensity
ratios assumptions concerning both the elemental distribution and the electron
attenuation length have to be made. Based on earlier reported values [11]
attenuation lengths of 45 A and 50 A were assumed for C Is and Si 2p
photoelectrons at a photon energy of 3.0 keV. Various different models for the
elemental distribution were tried [10] but the observed variations were only
Fig. 2
Experimentally extracted
intensity ratios C (graphite
like carbon/SiC), Si (Si02/SiC)
and C/Si (multiplied by 15) for
the sample with a nominal
oxide layer thickness of 58 A
are shown by filled diamonds,
crosses and open circles
respectively. The solid
lines show calculated
intensity ratios when
assuming the model of the
elemental distribution
shown by the insert.
20
40
EMISSION ANGLE
41
adequately reproduced when assuming a carbon layer on top of the oxide. The
expected intensity ratios are then given by;
c
Ac COSÖ
cc
cSi
Si
/ „ Xr
c
ye
COSÖ
" -1)
d0x
(e
XSj cosö
i)
csi
where c* represents the concentration of element x in matrix m, dg and dox the
thickness of the graphite like and oxide layers, Xc and X.si the electron attenuation
lengths and 9 the electron emission angle. When assuming elemental
= 1/3 and c =1.0 best agreement between
= l/2
concentrations of c.:
experimental and calculated intensity ratios for sample A was obtained for
dox = 39 Ä and dg = 3.0 A, which is illustrated by the solid lines in Fig. 2. For
sample B layer thicknesses of dox = 56 Ä and dg = 4.3 Ä produced best agreement.
These results thus indicate presence of a graphite like carbon layer at the surface
but not at the interface for both samples investigated.
We cannot exclude the possibility that the contribution from this layer
actually shadows a weaker contribution from graphite like carbon at the interface,
however. In order to check this possibility additional experiments were made
using lower photon energies in which the Si 2p and C Is components were
monitored after successive Ar+-sputtering cycles. Such spectra recorded from
sample A are shown in Fig. 3. The emitted photoelectrons have a kinetic energy of
around 250 eV in both cases, giving an electron attenuation length of ca. 10 A. An
ion energy of 1.0 keV was utilized which gave a sputter rate of about 1A per min
106
Fig. 3
102
BINDING ENERGY (eV)
288
98
284
BINDING ENERGY (eV)
Normal emission Si 2p and C Is spectra recorded after different
Ar+-sputtering times.
42
as determined from measurements on oxidized Si samples with known oxide
thickness values. In Fig. 3 the Si 2p component from Si02 is seen to decrease
with sputtering time while the contribution from SiC starts to appear after 30
min of sputtering and thereafter increases in relative intensity. For the C Is
components the graphite like carbon component completely dominates the
spectrum of the unsputtered surface but is not visible after 30 min of sputtering
when the SiC component starts to appear. A weak graphite like C Is component
is still visible after 15 min of sputtering but this should not be interpreted to
indicate that there is carbon in the oxide layer. Instead knock on effects during
sputtering of the surface carbon layer is believed to give rise to this weak
component since measurements made using several different photon energies
indicated that this weak component actually originated from the outermost
surface region. The important point is that no increased contribution from a
graphite like carbon component was possible to identify when the interface
between the oxide and SiC was probed. Thus also these experiments showed
contribution from graphite like carbon on top of the oxide but not at the interface.
A point worth noticing is that we could not identify more than two
components in either the C Is or Si 2p spectra so no contribution from an
interface Si4C4.x02 compound or presence of Si4C404 at the surface and in the
oxide, as proposed in an earlier investigation [3], could be identified. Another
point to be commented is the discrepancy in the oxide layer thickness values
extracted from photoemission and ellipsometry measurements. In this case we
believe the main reason to be the carbon layer on top of the oxide since such a
layer was not assumed in the analysis of the ellipsometry data.
CONCLUSIONS
The results of photoemission studies of two Si02/SiC samples have been
reported. Eecorded Si 2p and the C Is core level spectra each showed two
components which for the Si 2p level were identified as originating from SiC and
Si02 respectively while for the C Is level they were identified as originating from
SiC and graphite like carbon. Both angle resolved measurement made using a
photon energy of 3.0 keV and normal emission measurements made using lower
photon energies and after successive Ar+-sputtering cycles gave the same results.
A graphite like carbon layer on top of the oxide was identified. No contribution of
graphite like carbon at the Si02/SiC interface or presence of an interface
compound, as proposed in earlier studies, could be identified.
ACNOWLEDGEMENTS
Support from the Swedish Natural Science Research Council and SiCEP is
gratefully acknowledged. The authors wish to thank Prof. H. Arwin for the help
with the ellipsometry measurements.
43
REFERENCES
1
2
3
4
5
6
7
8
9
10
11
Q. Wahab, L. Hultman, M. Willander and J.E. Sundgren. J. Electron. Mater.
24, 1345 (1995).
Q. Wahab, R. Turan, L. Hultman, M. Willander and J.-E. Sundgren.
Thin Solid Films 287, 252 (1996).
B Hornetz, H.-J. Michel and J. Halbritter, J Mater. Res. 9, 3088 (1994).
W. Drube, H. Schulte-Schrepping, H.-G. Schmidt, R. Treusch and
G. Materlik, Rev. Sei. Instrum. 66, 1668 (1995).
W. Drube, T. M. Grehk, R. Treusch and G. Materlik, J. Electron Spect. Rel
Phenom. 88-89, 683 (1998).
J.N. Andersen, O. Björnholm, A. Sandell, R. Nyholm, J. Forsell, L. Thänell,
A. Nilsson and N. Märtensson, Synchrotron Radiation News 4, 15 (1991).
L. I. Johansson, Fredrik Owman and Per Märtensson. Phys. Rev. B 53,
13793 (1996).
F.Sette, G.K. Wertheim, Y. Ma, G. Meigs, S. Modesti and C.T. Chen, Phys.
Rev. B 41, 9766 (1990).
P.H. Mahowald, D.J. Friedman, G.P. Carey, K.A. Bertness and J.J.
Yeah, J. Vac. Sei. Technol. A 5, 2982 (1987).
L. I. Johansson, P.-A. Glans, Q. Wahab T.M, Grehk, Th. Eickhoff and
W. Drube, to be published.
P.J. Cumpson and M.P. Seah, Surf, and Interf. Analysis 25, 430 (1997).
44
ANNEALING OF ION IMPLANTATION DAMAGE IN SiC
USING A GRAPHITE MASK
Chris Thomas *, Crawford Taylor *, James Griffin *, William L. Rose *, M. G. Spencer *,
Mike Capano **, S. Rendakova ***, and Kevin Kornegay ****
*Materials Science Research Center of Excellence, Howard University, Washington, DC, 20059
** Purdue University, 1285 Electrical Engineering Bldg., West Lafayette, IN, 47907-1285
***TDI, Inc., Gaithersburg, MD, 20877
****Department of Electrical Engineering, Cornell University, Ithaca, New York, 14853
ABSTRACT
For p-type ion implanted SiC, temperatures in excess of 1600 °C are required to activate the
dopant atoms and to reduce the crystal damage inherent in the implantation process. At these
high temperatures, however, macrosteps (periodic welts) develop on the SiC surface. In this
work, we investigate the use of a graphite mask as an anneal cap to eliminate the formation of
macrosteps. N-type 4H- and 6H-SiC epilayers, both ion implanted with low energy (keV) Boron
(B) schedules at 600 °C, and 6H-SiC substrates, ion implanted with Aluminum (Al), were
annealed using a Graphite mask as a cap. The anneals were done at 1660 °C for 20 and 40
minutes. Atomic force microscopy (AFM), capacitance-voltage (C-V) and secondary ion mass
spectrometry (SMS) measurements were then taken to investigate the effects of the anneal on
the surface morphology and the substitutional activation of the samples. It is shown that, by using
the Graphite cap for the 1660 CC anneals, neither polytype developed macrosteps for any of the
dopant elements or anneal times. The substitutional activation of Boron in 6H-SiC was about
15%.
INTRODUCTION
The improvement of the material quality of SiC and the development of its device technology
have been, for the past several years, the focus of intense investigation by research groups from
around the world. It is projected that its large bandgap, high electron saturation velocity,
exceptional thermal conductivity (greater than copper), and large breakdown field strength will
improve the standard commercial benchmarks of high-power and high-frequency devices and
allow them to operate in caustic environments and at higher temperatures [1]. Of the three
methods of doping—ion implantation, thermal diffusion, and in situ doping—ion implantation will
play the most important part in the fabrication of these devices. This is because, unlike Silicon
technology, thermal diffusion, due to the low diffusion coefficients of the standard dopants below
1800 °C, is not a viable method for doping SiC [2]. In addition, the in situ doping method, while
it is the principal method of preparing doped device quality epitaxial material at present, cannot
be used for planar device fabrication and other applications where precision is required in small
areas. Still, despite its advantages, ion implantation of SiC is not as mature a technology as it is
for Silicon, where ion implantation is routinely used in device fabrication. One of the problems
associated with the ion implantation of SiC surrounds its post implant anneal. An anneal is
necessary to activate the dopant atoms and to reduce the inherent crystal damage created by the
ion implantation process. For p-type SiC, this anneal is usually done at temperatures up to 1700
°C [3]. One side effect of this anneal is that, at these high temperatures, the post anneal surface
45
Mat. Res. Soc. Symp. Proc. Vol. 572 ® 1999 Materials Research Society
morphology of the material is dominated by macrosteps. In this paper, we present a technique
that eliminates the formation of these macrosteps, and we also demonstrate Boron (p-type)
activation. AFM measurements were used to determine the morphology before and after
annealing, SIMS data was used to track the Boron ions through the steps of the experiment, and
C-V measurements were used to determine the substitutional activation.
EXPERIMENT
The samples used in this experiment were 4 um thick n-type (3.8 x 1015 cm"3) 6H-SiC and
(1.5 x 1015 cm"3) 4H-SiC epitaxial layers that were ion implanted at 600 °C with low energy
Boron. The specific energy/dose schedules for the Boron implants are as follows: 320 keV/3.6 x
1014 cm"2, 240 keV/2.8 x 1013 cm"2, 160 keV/2.0 x 10 13 cm"2, 100 keV/7.5 x 1012 cm"2, 50
keV/5.0 x 1012 cm"2, 30 keV/3.75 x 1012 cm"2. In addition, n-type (7 x 1018 cm"3) 6H-SiC
substrates that were ion implanted with low energy Aluminum were also used. The energy/dose
schedules for the Aluminum implants are as follows: 260 keV/1.0 x 1015, 150 keV/5.0 x 1014, 80
keV/3.0x 1014, 30 keV/2.0 x 1014.
The samples were first cleaned with a TCE/acetone/methanol degrease, followed by a 1
minute dip in 10:1 HF:H20 and a rinse in DI water. Random 5x5 urn2 AFM scans were taken to
determine the surface morphology of the samples before annealing.. The samples were again
cleaned in the manner described above, and a 0.8 urn thick carbon mask was fabricated on the
surface. For the anneals, an EMCORE resistive heated growth reactor was used. The reactor was
calibrated at one point by melting Silicon to verify the temperatures used for the anneals. While
Helium was the anneal ambient for most runs, the 6H-SiC substrates were annealed in Argon.
The anneal times used were 20 and 40 minutes and the anneal temperature was 1660 °C. There
was no data for the 40 minute anneal of the 6H-SiC substrates.
Upon completion of the anneals, the carbon mask was removed by oxidation at 800 °C for one
hour in a quartz furnace, and AFM measurements were done again to determine the effect of the
anneal on the morphology of the samples. The doping profile for the Boron implanted 6H-SiC
was determined from reverse capacitance-voltage measurements of Schottky diodes (Aluminum
ohmic and Schottky contacts). The measurements were taken at a frequency of 1MHz.
RESULTS AND DISCUSSIONS
The AFM micrographs shown in figure 1 detail the effect of a 40 minute anneal without a
Graphite cap at 1660 °C on the surface morphology of 6H-SiC. Figure 1(a) is the surface just
before the anneal is performed. The macrosteps shown in figure 1(b) were developed during the
anneal process. The effect is well known and has been reported by groups at Purdue University
[3] and Cree Research, Inc. [4]. These macrosteps extend across the entire surface of the sample
and are so pronounced that they can be viewed with an optical microscope. Similar results have
been reported for ion implanted 4H-SiC [5]. It is known that at about 1400 °C, Silicon containing
species (Si, Si2C, and SiC2) sublime from the surface layer of SiC [6]. It is this loss of Si atoms
and the ensuing redistribution of the Carbon rich surface layer that leads to the formation of the
macrosteps. Figure 2 shows the atomic force micrographs of the final surfaces of the samples that
were annealed for 20 and 40 minutes using our Graphite mask as a cap. From the micrographs, it
can be seen that no macrosteps were formed for any of the anneal times, even though each anneal
was done at 1660 °C. There is, however, a trend for each set of samples: the roughness increases
as the time of the anneal increases. This is seen in going from micrographs (b) to (c) and from the
46
Figure 1: AFM picture of Boron implanted (a) 6H-SiC epilayer before anneal (b) 6H-SiC epilayer after 40 minute
anneal at 1660 °C. Both micrographs are 5 (im by 5 (Xm in size.
micrographs (d) to (e), where each pair of micrographs corresponds to time increases from 20 to
40 minutes for 6H-SiC and 4H-SiC respectively. This result is consistent with those reported by
Capano et al [3].
Reports of SiC plates [6] and wafers [3] suspended above and almost touching the sample
during the annealing process are examples of techniques being used at the moment to reduce the
formation of macrosteps. These techniques attempt to create a Si overpressure that allows little to
no net movement of Si atoms from the surface layer of the sample, however, while the formation
of macrosteps is reduced, the surface roughness is still appreciable. It has also been reported that
a Si rich ambient gas such as Silane, if used during the anneal, will maintain the requisite Si
Figure 2: Atomic force micrographs of (a) Aluminum implanted 6H-SiC substrate annealed for 20 min, and Boron
implanted (b) 6H-SiC epi annealed for 20 min, (c) 6H-SiC epi annealed for 40 min, (d) 4H-SiC epi annealed for 20
min, (e) 4H-SiC epi annealed for 40 min. The edge of all micrographs correspond to 5 |im, and all anneals were done
atl660°C.
47
10» F
•
eo
53
0.15
0.2
Depth (fun)
M
004
0.044
0.046
0.048
0.05
Depth Qim)
Figure 3: Doping profile for (a) Aluminum implanted 6H-SiC substrate annealed for 20 min, (b) Boron implanted
6H-SiC epi annealed for 20 min., and (c) Boron implanted 6H-SiC epi annealed for 40 min.
overpressure and therefore suppress the formation of macrosteps [4]. Current results with the
silane anneal are encouraging, but, from our experiments, the success of the anneal is difficult to
reproduce. Our technique, the use of a Graphite mask as a cap, not only suppresses the
sublimation of Silicon from the surface, but it also prevents the formation of macrosteps by
preventing the redistribution of the surface layer. In addition, unlike A1N, which is another
material being used as an anneal cap for SiC, the Graphite mask can operate at temperatures up
tol850°C.
Along with to the AFM data, C-V measurements were taken to determine the substitutional
activation of the the 6H-SiC epilayers that were implanted with Boron. The results are shown in
48
H
1.0«20
1
_,
h
/
«?
E
,
h
»
1.0«19
^
y
g
^
700
S00
900
1000
Depth (nm)
Figure 4: Secondary ion mass spectrometry (SIMS) plot of the Boron implanted 6H-SiC epilayer samples annealed at
1660 °C for 40 minutes.
the doping profiles of figure 3. There is an increase in activation with anneal time as seen by
going from profiles (a) to (b) in figure 3, a trend that was also observed by Capano et al [3]. The
doping profile data is supported by the SIMS plot shown in figure 4, where up to the depth of 0.5
(Xm the movement of the Boron atoms produces a concentration of about 2 x 1018. This means
that for the 40 minute anneal corresponding to the doping profile in figure 3(b), the substitutional
activation was about 15%.
CONCLUSION
The use of a Graphite mask as an anneal cap was shown to prevent macrostep formation.
Temperatures of 1660 °C were used to anneal Aluminum implanted 6H-SiC substrates and Boron
implanted 6H- and 4H-SiC epitaxial layers. At these temperatures, macrosteps are usually formed
across the surface of the sample. None of the samples annealed with the Graphite cap developed
macrosteps during the anneal. There was a general trend of increased roughness for longer anneal
times, but the resulting roughness even after a 40 minutes anneal was not appreciable. There was
also a trend of increased activation with increased anneal time. Finally, p-type activation was
demonstrated for the Boron implanted 6H-SiC epilayer samples.
ACKNOWLEDGMENTS
This work was supported by MURI for manufacturable power switching devices. The contract
manager was John Zolper.
49
REFERENCES
[1]
M. Bhatnagar and B. J. Baliga, IEEE Trans. Electron Devices 40 (1993), p.645.
[2]
M. V. Rao, J. A. Gardner, P. H. Chi, O. W. Holland, G. Keiner, J. Kretchmer, and M.
Ghezzo, J. Appl. Phys. Vol. 81, No. 10, p. 6635 (1997).
[3]
M. A. Capano, S. Ryu, M. R. Melloch, J. A. Cooper, Jr., and M. R. Buss, J. Electronic
Materials, Vol. 27, No. 4, (1998).
[4]
CREE Research, Inc., Oral presentation at MURI review, (1999).
[5]
M. A. Capano et. al., Oral presentation at MURI review, (1999).
[6]
L. Ottaviani, D. Planson, M. L. Locatelli, J. P. Chante, B. Canut, and S. Ramos, Material
Science Forum, Vols. 264-268, p. 709 (1998).
50
EFFECT OF VARYING OXIDATION PARAMETERS ON THE GENERATION
OF C-DANGLING BOND CENTERS IN OXIDIZED SIC
P.J. Macfarlane and M.E. Zvanut, Department of Physics, University of Alabama at
Birmingham, 310 Campbell Hall, Birmingham, AL 35294-1170
Abstract
SiC is perhaps the most appropriate material to replace Si in power-metal-oxidesemiconductor-field-effect-transistors (MOSFETs), because, unlike the other wide band-gap
semiconductors, SiC can be thermally oxidized similarly to Si to form a SiC>2 insulating layer. In
our studies of oxidized SiC, we have used electron paramagnetic resonance (EPR) to identify Cdangling bonds generated by hydrogen release from C-H bonds. While hydrogen's effect on SiCbased MOSFETs is uncertain, studies of Si-based MOSFETs indicate that it is important to
minimize hydrogen in MOS structures. To examine the role of hydrogen, we have studied the
effects of SiC/Si02 fabrication on the density of C-related centers, which are made EPR active by
a dry heat-treatment. Here we examine the starting and ending procedures of our oxidation
routine. The parameter that appears to have the greatest effect on center density is the ending
step of our oxidation procedure. For example, samples that were removed from the furnace in
flowing O2 produced the smallest concentration of centers after dry heat-treatment. We report on
the details of these experiments and use our results to suggest an oxidation procedure that limits
center production.
Introduction
SiC has both a wide band-gap and high thermal conductivity that make it an attractive
replacement for Si in high power, high temperature microelectronic devices. For power metaloxide-field-effect transistors (MOSFETs) in particular, SiC is of interest because unlike the other
wide band-gap semiconductors, it can be thermally oxidized similarly to Si in order to create a
Si02 insulating layer.
In our previous studies of oxidized 3C-SiC, 4H-SiC, and 6H-SiC [1-2], we have observed
centers that can be activated by dry heat-treatments at temperatures greater than 800 °C. The gvalues of these centers range from 2.0025 to 2.0029, which is within the range of g-values
typical of C-related centers [3-5]. The temperatures at which these centers are generated are
significantly greater that those used to generate Si dangling bonds. Thus, we suggest that the
centers are unpaired electrons located on C atoms. We have also observed that these centers can
be activated by heat-treatment in an ambient that does not contain moisture and passivated in an
ambient that contains moisture. Therefore, we suggest that these centers are activated by release
of a hydrogenous species from C dangling bonds. Supporting the relation to hydrogen is an
experiment in which we heat-treated oxidized samples in dry (<0.6 ppm H2O) O2. The
concentration of centers activated is similar to that activated by dry heat-treatment in N2. In
addition, we have observed that for the 3C-SiC epilayer samples the density of centers can be
reduced by annealing in forming gas. Hydrofluoric acid etching studies of the samples indicated
that the centers are not located in the oxide.
In this study of the C-related center in oxidized, heat-treated 3C-SiC, 4H-SiC, and 6HSiC, we investigate the effects of altering the loading and unloading procedure of our oxidation
routine on the concentration of C-related centers produced by dry heat-treatment. In our
examinations of the removal steps, we unload the samples in wet O2 or standard N2 and compare
51
Mat. Res. Soc. Symp. Proc. Vol. 572 ® 1999 Materials Research Society
the results with slowly cooling the samples in air [1-2,6]. We also compare the effects of
inserting samples into the furnace in air, wet 02, and N2 ambients. The 3C-SiC epilayer samples
are included to gain a clear scientific understanding, although the hexagonal polytypes are more
applicable to current high power device research.
Experimental Information
The 4H-SiC and 6H-SiC materials are from double-side polished, p-type, 1.5" wafers
supplied by Cree Research. The polished surfaces of both wafers are cut approximately 3.5° off
the {0001} crystal planes. The 4H-SiC wafer has an epilayer deposited on its Si face. HOYA
Corporation supplied the 3C-SiC epilayer layer samples. The approximately 1 um thick, cubic
SiC film was deposited via chemical vapor deposition (CVD) on both the polished and
unpolished faces of a (100) oriented Si substrate. Samples are cut into strips 1.5 cm by 0.23 cm, a
size suitable for the EPR microwave cavity.
Prior to oxidation, we clean the samples by rinsing in trichloroethane, xylenes, acetone,
methanol and deionized water for 5 min time intervals. The samples are then etched for 1 min in
9:1 H20:HF (50%). The samples are oxidized for 6 hr at 1150° C in 02 bubbled through
deionized water. The loading and unloading steps of the oxidation procedure are described
below. After oxidation, the samples are heat-treated in a double-walled quartz tube furnace at
900 °C for 320 min. The heat-treatment is conducted in a dry (<0.3 ppm H20) N2 ambient. At the
end of the dry heat-treatment, the samples are quenched to approximately 100° C where they are
cooled sufficiently before they are removed from the dry ambient.
For the oxidation, five different insert and removal conditions are examined:
1) Insert in air, remove quickly in N2 ("Pull-out N2")
2) Insert in air, remove quickly in 02 ("Pull-out 02")
3) Insert in air, slowly cool in air ("Air Cooled")
4) Insert in 02, remove quickly in 02
5) Insert in N2, remove quickly in 02
The slow cool in air is accomplished by gradually pulling the sample to the edge of the furnace
tube where the temperature is about 940 °C. The sample remains there about 5 min before being
placed in room ambient. For the other removal conditions, the samples are pulled from the
furnace with either 02 or N2 flowing and immediately placed in room ambient. The insert
conditions consisted of a 30 min 02 or N2 purge of the furnace tube, followed by insertion of the
samples while the 02 or N2 gas is flowing. For the latter case, the N2 gas is terminated and 02 is
used for the oxidation. The oxygen is bubbled through deionized water for all 02 treatments, and
the N2 as standard grade.
All EPR measurements are conducted at room temperature using an X-band Bruker 200
spectrometer. EPR is a spectroscopic technique able to detect paramagnetic defects in solids. We
use the dry heat-treatment to activate the paramagnetic state of a defect already present in the
material. The activation process is similar to the one used for the Si Pb center, a Si dangling bond
at the interface between Si02 and Si [7-8]. The concentration of centers is found by double
integration of the EPR spectrum and comparison with the double integral of a spectrum obtained
from a known standard. Typically, we integrate the EPR spectrum with the largest amplitude and
determine the concentration from the other spectra by scaling their amplitudes. All spectra are
measured with 1 G peak-to-peak modulation amplitude and 1 mW microwave power.
52
Results
14
f
}
Electron paramagnetic resonance spectra were
12
measured for each of the polytypes after oxidation and
*
after dry heat-treatment. From these measurements, the
10
concentration of C-related centers after dry heatT
treatment is determined. In Figure 1, we plot the
8
concentration of the centers in each of the three
\
polytypes for three different unloading procedures. The
6
unfilled squares, filled circles, and unfilled triangles
4
represent the concentration of centers in the oxidized
*
3C-SiC epilayer, 4H-SiC and 6H-SiC samples,
2
respectively. Error in concentration is due to noise in
the EPR spectrum. While variation in concentration
Air Cooled Pull-Out02 Pull-OutN2
appears between the different SiC polytypes, a common
End of Oxidation Conditions
trend is observed. Unloading the samples in N2
produced the largest concentration of centers. In Figure 1: The concentration of centers
contrast, removing the samples from the furnace in wet produced as a function of the ending
O2 produced the smallest concentration of centers.
oxidation conditions. "Air Cooled"
In Figure 2, we plot the concentration of centers describes the condition in which samples
measured in the 6H-SiC samples as a function of their were slowly cooled in the furnace in air
insert ambients. "Air" is used to describe the condition after the heat was turned off- "Pull-out
in which the samples were inserted into the furnace in 02" and "Pull-Out N2" describe the
an air ambient. Similarly, "Wet O2" and "N2" are used conditions in which samples were directly
unloaded from the furnace in wet 0 and
to indicate the conditions in which the samples were standard N ambients, respectively.2 The
2
inserted into the furnace in flowing wet O2 and flowing
D, • ,and A indicate the concentration of
standard N2 ambients, respectively. The data point for centers in the oxidized 3C-SiC, 4H-SiC,
the sample inserted into air is the same as the point in and 6H-SiC samples.
the previous figure of the 6H-SiC sample was removed
in an flowing wet O2 ambient. The error in concentration arises from noise in the EPR spectrum.
The concentration of centers is the same for each of the different insert ambients.
i
i
, f
Discussion
While there is no direct evidence the C-related centers we observe are associated with
electrically active defects, centers detected by EPR in oxidized Si have been shown to be related
to electrical defects. For example, electrically active interface states were found to be related in
part to an interfacial Si dangling bond, called the Pb center [9], and some charge traps in the SiC>2
have been attributed to the E' center, a hole trapped at an O vacancy in SiC>2 [10]. Thus, a center
detected by EPR in oxidized SiC could also be related to electrical defects. We observe some
similarities between processing procedures that reduce the concentrations of interface states and
oxide charge traps and those that reduce the concentration of the centers in our samples. The
similar and contrasting results are described below.
In examining the beginning and ending steps of our oxidation procedure, we observe that
the final steps of the oxidation have the largest effect on the concentration of centers. Figure 1
indicates that removing the samples in O2 minimizes the concentration. Compared to removal in
N2, the center density is decreased by 70%, 50%, and 40% in 3C-, 4H- and 6H-SiC samples,
respectively. The reduction in center concentration may be related to the 5 min 940 °C wet O2
53
"anneal" that the samples receive before they are
removed from the furnace. Lipkin and Palmour have
14
demonstrated that both interface states and oxide
charge in oxidized 6H-SiC can be reduced by "re12
oxidizing" 6H-SiC samples in wet 02 at 950 °C for 1.5
10
to 3 hr [11]. Although Lipkin and Palmour's "reoxidation" anneal was for a significantly longer period
of time, it is interesting to note that a brief wet 02 heattreatment of our samples also considerably reduced the
6 C-related center concentration. Most likely, unloading
the samples in wet O2 stabilizes the interface between
the Si02 and the SiC preventing Si evaporation [12].
o
Assuming that N2 is primarily responsible for U
the affects seen in the air annealed samples,
comparison of the slowly air cooled and more rapidly
Wet02
N2 cooled samples suggests that thermal shock may
Insert Ambient
also affect the production of the C-related centers. In
Fig. 1, a lower concentration of centers is found for Figure 2: The concentration of centers in
samples that were slowly cooled in air versus samples the 6H-SiC samples as a function of the
that were more rapidly removed in N2. In particular, insert ambient. "Air," "02," and "N2"
we observe that the density of centers in 3C-SiC indicate that the samples were loaded into
samples that were slowly cooled was 35% lower than the furnace in air, wet 02, or standard N2
the density observed in samples that were rapidly ambients, respectively,
pulled in N2 to room ambient. Although the density of
the centers in the hexagonal polytypes were the same within sample to sample variation, on
average the concentration decreased by approximately 10%. The large reduction in center
concentration found in the 3C-SiC samples is consistent with the fact that film/substrate samples
such as the 3C-SiC epilayers, are more susceptible to thermal shock than bulk substrates.
Studying the effect of oxidation conditions on the electrical properties of hexagonal polytypes,
Shenoy and coworkers observed a similar behavior [13]. They compared the effects of removing
samples using a "slow pull" method, in which the temperature was gradually reduced to 900 °C
before the samples were unloaded, to using a "fast pull" technique in which samples were
withdrawn from the furnace over approximately 100 sec. Like we observed with the
concentration of our C-related center, they found that the density of interface states and oxides
charges traps could be reduced by 10% when unloading samples with the "slow pull" procedure
versus the "fast pull" method. While the EPR/electrical comparison is intriguing, we must
acknowledge that in our experiments, the oxygen in the air may contribute to the reduction of the
EPR defect concentration. Future experiments will be conducted in order to determine the role
that thermal shock plays in the production of the C-related centers.
Figure 2 indicates that the concentration of the C-related centers does not depend on
whether the insert ambient is oxidizing (02) or inert (N2). This is in contrast with electrical
studies which show that samples loaded in an inert gas (Ar) have a higher concentration of
electrically active defects [13] than samples loaded in 02. The increase in electrical defects were
attributed to Si evaporation from the SiC surface, leaving a C-rich layer on the 6H-SiC substrate.
We, however, do not observe this for samples that were inserted in the furnace in flowing N2
versus flowing 02. In fact, the concentration of centers indicated by Fig. 2 is approximately the
same regardless of whether the samples are inserted in air, N2 or wet 02. We speculate that these
results indicate that as far as our C-related centers are concerned any initial surface layer is likely
1
54
1
1
consumed during oxidation and converted into the oxidation products. Perhaps, because
oxidation occurs at the interface between the SiC substrate and the Si02 layer, the ending steps
of the oxidation would have more of an effect on the quality of the SiC/Si02 interface. This
agrees with our observations of sizeable changes in concentration produced in samples oxidized
with different ending steps.
Conclusion
In conclusion, the concentration of centers can be affected by altering the ending steps of
our oxidation procedure. The ending conditions, ordered from smallest center concentration
produced to largest, are unloading the samples in wet O2, slowly cooling the samples in air, and
unloading the samples in standard N2. The insert ambient did not affect the center concentration.
This may indicate that because oxidation occurs at the SiC/Si02 interface, any initial effects of
the oxidation procedure are effectively removed during the oxidation. Thus, the ending process
should have a larger effect on the concentration of centers produced.
Acknowledgements
This work is supported by ONR grant no. N00014-96-2-1238. P.J. Macfarlane is supported by a
fellowship from the Alabama Space Grant Consortium.
References
[I]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[II]
[12]
[13]
P.J. Macfarlane and M.E. Zvanut, Appl. Phys. Lett. 71,2148 (1997).
P.J. Macfarlane and M.E. Zvanut, in Hydrogen in Semiconductors and Metals, edited by
N.H. Nickel, W.B. Jackson, R.C. Bowman, and R. Leisure (Mater. Res. Soc. 513,
Pittsburgh, PA 1998), pp. 433-438.
X. Zhou, G. Watkins, K.M. McNamara Rutledge, R.P. Messmer, and S. Chawla, Phys.
Rev. B 54,7881 (1996).
G. Gerardi, E. Poindexter, and C. Young, Appl. Spectrosc. 50, 1427 (1996).
V.V. Afanas'ev and A. Stesmans, Appl. Phys. Lett. 69,2252 (1996).
P.J. Macfarlane and M.E. Zvanut, J. of Electron. Mater. 28,144 (1999).
K.L. Brower, Physical Review B42,3444 (1990).
J. Stathis, J. Appl. Phys. 77, 6205 (1995).
E. Poindexter, J. Non-Crystalline Solids 187, 257 (1995).
P.M. Lenahan and P.V. Dressendorfer, J. Appl. Phys. 55, 3495 (1984).
LA. Lipkin and J.W. Palmour, J. Electron. Mater. 25,909 (1996).
L. Muehloff, W.J. Choyke, M.J. Bozack, and J.T. Yates, J. Appl. Phys. 60, 2842 (1986).
J.N. Shenoy, G.L. Chindalore, M.R. Melloch, JA. Cooper, J.W. Palmour, and K.G. Irvine,
J. Electron. Mater. 24, 303 (1995).
55
THICK OXIDE LAYERS ON N AND P SiC WAFERS
BY A DEPO-CONVERSION TECHNIQUE
Q. Zhang, V. Madangarli, I. Khlebnikov, S. Soloviev and T. S. Sudarshan
Department of Electrical Engineering
University of South Carolina, SC 29208, U.S.A
Tel: 803-777-7302; Fax: 803-777-8045
E-mail: Zhang@engr.sc.edu
ABSTRACT
The electrical properties of thick oxide layers on n and/7-type 6H-SiC obtained by a depoconversion technique are presented. High frequency capacitance-voltage measurements on
MOS capacitors with a - 3000 Ä thick oxide indicates an effective charge density comparable
to that of MOS capacitors with thermal oxide. The breakdown field of the depo-converted
oxide obtained using a ramp response technique indicates a good quality oxide with average
values in excess of 6 MV/cm on .p-type SiC and 9 MV/cm on w-type SiC. The oxide
breakdown field was observed to decrease with increase in MOS capacitor diameter.
INTRODUCTION
Silicon Carbide (SiC) devices have been proposed for many industrial applications
requiring high power, high frequency and hard radiation. Recently, tremendous improvements
have been made in the crystal growth and processing techniques including oxidation which will
accelerate the commercialization of SiC devices. Oxide layers on SiC not only find application
in the fabrication of MOS devices but also as field oxides and passivation layers in high voltage
devices. While the oxide layer should have high breakdown strength, low leakage current, and
low effective charge density for satisfactory MOS device performance, when oxide layers are
used as field oxides in high voltage devices high breakdown strength is most essential. For
example, a metal-overlap onto an oxide layer can be used as an edge termination technique for
improving the breakdown voltage of high voltage devices by minimizing the electric field
enhancement at the contact periphery of devices. In order for this field oxide to be effective it
should have good dielectric properties and sufficient thickness to sustain the high breakdown
voltage. However, the SiC oxidation rate is too slow to obtain thick oxide layers via the
conventional thermal oxidation techniques currently in practice [1,2,3]. Even though alternate
techniques such as oxide deposition by LPCVD and poly-silicon conversion have been reported
[4,5], the high voltage characteristics of deposited oxide vs a conventional thermal oxide has
not been studied in detail. In this paper, we report the possibilities of using simple depoconverison technique for obtaining thick oxide layers with high breakdown strength on n m&ptype 6H-SiC wafers.
EXPERIMENT
The 6H-SiC wafers used in our experiments were from Cree Research (substrate doping
~ 1.6xl018cm"3) with a 10 um thick epilayer of- 6xl015cm"3 doping concentration forp-type
and ~ 3.9xl015 cm'3 for «-type SiC respectively. The 30 mm diameter wafer was cut into 10
mm X 10 mm square pieces to obtain several samples for experiments. The as-received SiC
samples were first cleaned using TCE at 85°C for 15 min. followed by ultrasonic cleaning in
acetone and methanol respectively. Then the native surface oxide was chemically etched using
a 20% HF solution prior to a modified RCA cleaning process. After the RCA cleaning process
the SiC wafers were etched in a 5% HF solution for 10 seconds to remove the surface oxide
57
Mat. Res. Soc. Symp. Proc. Vol. 572 ° 1999 Materials Research Society
caused by the RCA process. De-ionized water rinses were used after every step. In order to
obtain the depo-converted oxide, a Si film is deposited on the SiC wafer and converted to
silicon dioxide by oxidation. The Si films were RF sputtered from a single crystal Si target in a
vacuum chamber under Ar ambient and the sputtered Si films were converted to oxide by wet
oxidation at 1050°C for 3 hours to ensure total conversion of the deposited Si. All samples with
converted oxide had a very uniform surface morphology, indicating that depo-conversion
method can be used to obtain thick oxide films. MOS capacitor (MOS-C) structures were
fabricated, to characterize the electrical properties of the depo-converted oxide, by evaporating
Al on both sides of the sample. Moreover, the usefulness of the depo-converted oxide as a field
oxide was demonstrated by fabricating a lkV 6H-SiC Schottky diode with oxide layer edge
termination.
RESULTS AND DISCUSSION
C- V characteristics
Capacitance-voltage measurements were made on MOS capacitors with different oxide
thicknesses on n and/) type SiC at room temperature using a Kiethley 590 capacitance-voltage
(C-V) analyzer at 100 kHz. Voltage sweeps applied to the MOS capacitor were made from
accumulation to depletion/inversion. Typical C-V characteristics of MOS capacitors with
thermal and converted oxides fabricated on n and/»-type SiC are shown in Fig. 1.
Thermal oxide
0.5
lermal oxide
210A
-20
-5
0
Gate voltage (V)
(a)N-type6H-SiC
-10
0
Gate voltage (V)
(b) P-type 6H-SiC
Fig. 1. Comparison of C-V characteristics of MOS capacitors with depo-converted oxide vs
thermal oxide on n and p-type 6H-SiC wafers.
The oxide and SiC parameters obtained from C-V measurements are given in Table 1.
Table 1. Oxide and SiC wafer parameters from C-V measurements
SiC
wafer type
N-type
P-type
Oxide
thickness (Ä)
Doping
concentration (cm'3)
Flat-band
voltage (V)
210*
2500*
4000"
370*
3200*
6.5E15
4.4E15
5.1E15
4.6E15
2.1E15
-0.4
-2.4
-1.47
-7.25
-11.75
* Thermal oxide
# Depo-converted oxide
58
Effective charge
density
(cm"2)
7.327E+10
1.685E+11
6.114E+10
2.885E+12
4.156E+12
From Table 1, it is observed that the effective charge density of the depo-converted oxide is
in the same order of magnitude as that of the thermal oxide. As expected, the flat-band voltage
and effective charge density are lower for the oxides grown on N-type SiC compared to those
on P-type SiC [6].
Oxide breakdown measurements
In order to determine the breakdown strength of the oxide films, a non-destructive ramp
response technique [7] was used to measure the current-voltage characteristics in the
accumulation regime of the oxide films on n and p type SiC wafers. From the maximum
breakdown voltage and thickness of the oxide films, the critical breakdown field of the oxide
films was computed.
Experimental results show that for a given thickness of the oxide, the breakdown field of
the oxide reduces with increase in the diameter of the MOS capacitor structure fabricated on
the SiC wafer. A typical variation of the oxide breakdown field on a n type SiC with increasing
MOS capacitor diameter is illustrated in Fig. 2.
Thermal oxide ~ 250 A
Depo-converted oxide ~ 2500 A
Depo-converted oxide - 4000 A
SO
100
150
200
MOS Capacitor Diameter (um)
Fig. 2. Dependence of oxide breakdown field on MOS capacitor diameter («-type SiC).
As observed from Fig. 2, the breakdown field of the thermal oxide increases rapidly with
decrease in MOS capacitor diameter below ~ 150 urn. On the other hand both the 2500 A and
4000 Ä thick depo-converted oxides exhibit a gradual reduction in the breakdown field with
increase in MOS capacitor diameter upto about 150 um, after which there is no significant
reduction in the breakdown field with further increase in MOS-C diameter. The decrease in the
oxide breakdown field with increase in MOS-C diameter could be due to the following reasons.
We have observed that oxide breakdown in SiC wafers generally occurs at locations
corresponding to the edge of bulk structural defects in the SiC wafer such as polytype
inclusions, regions of crystallographic mis-orientation, or different doping concentration [8].
Due to poor SiC material quality the probability of enclosing such defects underneath the gate
contact of a MOS-C increases with increase in diameter. This in turn increases the probability
of oxide breakdown with increase in MOS-C diameter. Another plausible explanation could be
related to the variations in the electric field distribution in the oxide film with increase in MOS-
59
C diameter. The accumulation layer underneath the gate contact is probably more uniform for
small diameter MOS structures compared to large diameter structures due to contact edge
effects. Hence, the electric field distribution in the oxide could be more uniform for small
diameter structures thus enabling them to withstand higher electric stress compared to large
diameter structures where electric field enhancement around the gate contact periphery results
in lower oxide breakdown strengths.
-*-- *
2
7"
1000
2000
Oxide thickness (A)
3000
4000
Fig. 3. Dependence of breakdown field on oxide thickness (diameter of MOS-C : 240 um).
Fig. 3 shows the dependence of the breakdown field on oxide thickness for n and/? type SiC
wafers. The breakdown field of oxide films in the case of /»-type SiC wafers decreases
significantly from a value of more than 8 MV/cm for ~ 300 A thermal oxide to ~ 6 MV/cm for
a ~ 3000 Ä depo-converted oxide. On the other hand, the breakdown field of oxide films on ntype SiC wafers reduces slightly from ~ 9.6 MV/cm for ~ 300 Ä thermal oxide to ~ 8.7 MV/cm
for a ~ 4000 Ä thick depo-converted oxide. The reduction in oxide breakdown field with
increase in thickness is understandable considering the well known fact that the breakdown
strength of solid dielectrics decreases with increase in thickness due to a reduction in the heat
removal rate which promotes an electro-thermal breakdown at smaller fields [9]. The difference
in the extent of reduction of the oxide breakdown field on «-type vs p-type SiC, could be due to
poorer oxide quality on/>-type SiC with increasing thickness.
The fairly high breakdown strengths of the depo-converted oxide indicates that it is possible
to use thick oxide layers obtained by the depo-conversion technique for applications requiring
high oxide breakdown strength. For example, the depo-converted oxide can be effectively used
for field plate edge termination in Schottky diodes. In fact, we have successfully fabricated
high voltage (lkV) p-type SiC Schottky diodes with a ~ 6300 Ä depo-converted oxide as edge
termination [10].
CONCLUSION
In conclusion, thick oxide films have been successfully formed on n and /?-type SiC
substrates by converting Si to oxide. C-V measurement indicates that this converted oxide
film exhibits an effective charge density comparable to that of a thermal oxide. The depoconverted oxides also indicated fairly high breakdown strengths, in the range of 6-9 MV/cm.
The oxide breakdown strength was observed to decrease with increasing oxide thickness
especially on p-type SiC. Also, for a given oxide thickness the oxide breakdown strength
decreased with increase in MOS capacitor diameter. Thick oxide layers obtained by converting
60
Si to oxide exhibits promising characteristics for applications in power SiC devices, such as
field plate edge termination of Schottky diodes.
ACKNOWLEDGEMENTS
This work was supported by ARO (grant no DAA H04-96-1-0467) via a DEPSCoR program.
The authors are pleased to acknowledge many discussions with Dr. G. Gradinaru, and Mr. Y.
Gao.
REFERENCE
1.
2.
3.
4
5.
6.
7.
M. Bhatnagar and B.J. Baliga, IEEE Trans. ED 44, 645 (1993).
P.G Neudeck, J. Electron. Mater. 24, 283 (1995).
J. Schmitt and R. Helbig, J. Electrochem. Soc. 141, 2262 (1994).
J W. Palmour, U. S Patent No. 5 612 260, (18 March 1997).
J. Tan, M. K. Das, J. A. Cooper, Jr., and M. R. Melloch, Appl. Phys. Lett. 70, 2280 (1997).
S. Dimitrijev, H. F. Li, H.B. Harrison, and D. Sweatman, IEEE Trans. EDL 18, 175 (1997).
V. P. Madangarli and T.S. Sudarshan, Proc. of the 7th Int. Conf. on SiC, Ill-Nitrides, and
Related Materials (ICSCIII-N'97), 665 (1997).
8. S. Soloviev, I. Khlebnikov, V. Madangarli and T. S. Sudarshan, J. Electron. Mater. 27,
1124(1998).
9. B. Tareev, Physic ofDielectric Materials, (Mir Publishers, Moscow, 1979).
10. Q. Zhang, V. Madangarli, S. Soloviev and T. S. Sudarshan, to be presented at 1999 MRS
Spring Meeting (unpublished).
61
Bias-Temperature-Stress Induced Mobility
Improvement in 4H-SiC MOSFETs
K. Chattyf, T. P. Chowt, R- J- Gutmannf, E. Arnold*, and D. AlokJ
t Rensselaer Polytechnic Institute, Troy, NY 12180-3590, U.S.A.
Tel: 518-276-6044, Fax: 518-276-8761, e-mail: kchatty@unix.cie.rpi.edu
t Philips Research, Briarcliff Manor NY 10510, U.S.A.
ABSTRACT
In this work, we report on an instability which affects the field effect mobility in 4HSiC MOSFETs. The devices (MOSFETs and capacitors) were subjected to a biastemperature stress (BTS) for 30 minutes at 150°C at stress voltages corresponding
to oxide fields upto lMV/cm. Following a positive BTS(i.e. gate voltage positive),
the field effect mobility increased by upto two orders of magnitude from the original
value; upon application of a negative BTS to the MOSFET, the device characteristics
degraded to the unstressed state. The high mobility state could be recovered by a
positive BTS and was reversible with repeated bias stressing. An explanation of this
phenomenon is proposed based on the effect of interfacial ions on the dependence of
both trapped charge and inversion charge densities on gate bias.
INTRODUCTION
The electrical properties of the current state-of-the-art SiC-Si02 interfaces are inferior
to those of silicon. The densities of oxide charges and interface states are much higher
than those at the Si-Si02 interface[l]. The effect of the localized states is seen in the
degradation of the transconductance and the increase in the threshold voltage of
the MOSFETs[2]. The understanding and control of the characteristics of SiC-Si02
interface is crucial to the realization of practical SiC MOS devices. In this work, we
report the investigation of an instability which affects the field effect mobility in SiC
MOSFETs.
MOSFET FABRICATION
The 4H-SiC wafers used for the fabrication of the MOSFETs had an epitaxial thickness and doping of 10 fim and 4xl015 cm-3 respectively. The wafers were cleaned
before a 800nm thick plasma TEOS oxide (field oxide) was deposited on the wafers.
A lOOnm thick plasma TEOS oxide was deposited to act as a pad oxide during implantation of the source and drain. The source and drain were then implanted with
nitrogen (80keV, 2xl015 cm-2; 40keV, lxlO15 cm"2) at 650 °C. The samples were
then annealed at 1200 °C for 1 hour in an argon to electrically activate the implants.
63
Mat. Res. Soc. Symp. Proc. Vol. 572
e
1999 Materials Research Society
Source and Drain
- Contacts
-
SiO
Poly-Si gate
-t—
1
p-epi
p+ 4H-SiC
Fig. 1: Schematic of the 4H-SiC lateral MOSFET.
After the implant activation anneal, the field oxide on one of the samples (thin oxide
MOSFET) was etched back to 200nm. The samples underwent oxidation in a wet
ambient at 1100°C for 6 hours and 40 min, followed by an anneal for 1 hour at the oxidation temperature in argon. The oxide was then subjected to a re-oxidation anneal
in a wet ambient at 950°C for 3 hours. The annealing cycles were similar to the work
by Sridevan et al [3]. Subsequent to the oxidation, polysilicon was deposited and degenerately doped by phosphorus implantation. After definition of the gate, a 600nm
thick plasma TEOS oxide was deposited to serve as an interlevel dielectric. The source
and drain contacts (Al/Ni/Al) were denned using liftoff technique, and the contacts
(source, drain and the substrate) were annealed at 1000°C in argon. The gate contact
was patterned and etched, followed by (Ti/Mo) contact metallization(Figure 1).
RESULTS AND DISCUSSION
Experimental results were obtained on two MOSFETs with gate width-to-length
(W/L) ratios of 8, and gate oxide thicknesses of 200 nm (thin oxide) and 900 nm
(thick oxide). The field effect mobility was calculated from the transconductance
dlo/dVe at a drain voltage of 25-100mV:
ßFE '
dID/dVa
(1)
where IDS is the drain current, VG is the gate voltage and COT is the oxide capacitance
per unit area. The mobility values quoted below were taken at the steepest part of
the Ijn-Vß curve.
The initial field effect mobility of the thin-oxide MOSFET was 0.5 cm2/V.s.
Above room temperature, on repeated gate voltage sweeps, the Ip-Ve transfer characteristic varied. Assuming that these variations were due to mobile ions in the oxide,
a bias-temperature-stress(BTS) was applied to stabilize the transfer characteristics
64
2
6.E-07
After BTS
c
e
4.E-07
a
2.E-07
a
U
c
Before BTS
10
20
30
Gate VoltagofV)
Fig. 2: Transfer characteristics of the thin oxide MOSFET before and after
BTS(VD=25mV).
by drifting the mobile charge towards the semiconductor-insulator interface^].
The BTS of the thin-oxide MOSFET was done by applying a gate voltage corresponding to an oxide field of lMV/cm at 150°C for 30 minutes, with the source and
drain contacts floating. The device was cooled to room temperature while maintaining the bias voltage on the device. Figure 2 compares the transfer characteristics of
the thin-oxide MOSFET before and after BTS. The field effect mobility of the sample increased by a factor of 16, from 0.5 to 8 cm2/V.s after BTS. Similar instabilities
were observed in the transfer characteristics of the thick oxide MOSFET. The BTS
consisted of applying a gate voltage corresponding to an oxide field of 0.5MV/cm
at 150°C for 30 minutes, with figures 3 and 4 showing the transfer characteristics
before and after BTS. In this case, the field effect mobility of the sample increased
from 0.1 to 13cm2/V.s, a two orders-of-magnitude improvement.
Transfer characteristic of another thick-oxide MOSFET before and after BTS
is shown in Figure 5, along with the transfer characteristics one month after the
bias stressing. While the BTS improves the transconductance, but the characteristics
degrade back to the pre-bias-stressed values with time, as the ions diffuse from the
SiC-Si02 interface.
To further investigate this phenomenon, mobile-ion analysis experiments were
done on MOS capacitors located on the same wafer. After the initial room temperature capacitance-voltage characteristic was measured using a 1MHz Boonton capacitance meter, a BTS of +20V at 150°C for 30 minutes was applied. The capacitancevoltage characteristic was also measured after a negative BTS.(Following the negative
BTS, the MOSFET did not turn on until a subsequent positive BTS). Figure 6 shows
the results of the C-V measurements, from which a mobile-ion charge density of
3xl012 cm-2 was extracted.
Figure 7 compares the gate-voltage dependence of transconductance for a thick
65
1.E-0S
After BTS^*""
After BTS
V 8.0E-0S
I3
>^
Before BTS
o
| 1.E-10
1
Drain Cumi
ä1 1.E-0B
c
Before BTS
/
1.E-12
0
10
20
30
40
0
SO
10
f
20
30
40
50
Gate Vottago(V)
Gate VoHagefV)
Fig. 3: Transfer characteristics (linear
scale) of the thick oxide MOSFET before
and after bias stressing (VD=25mV)
Fig. 4: Transfer characteristics (log scale)
of the thick oxide MOSFET before and after bias stressing (Vr>=25mV)
oxide MOSFET before and after the BTS. Before BTS, the transconductance increases
monotonically with gate bias: after the BTS, the transconductance increases more
rapidly, reaches a peak, and decreases with a further increase in gate voltage, as would
normally be observed in a conventional MOSFET.
After positive
BTS
Xy
I 6.E-08
3
u
c
4.E-08
1 Month after
I positive BTS
y
Before
//BTS
0
20
40
60
80
Gate Voltage(V)
Fig. 5: Transfer characteristics for the thick oxide MOSFET before and after BTS
The unusual SiC MOSFET characteristics are attributed to mobile charges
within the gate oxide and their effect on the dependence of interface state occupancy and inversion charge on the gate bias voltage. A possible explanation for this
phenomenon is as follows: A change <5V<j in gate voltage results in a change 8Qt in
66
the charge trapped in the interface states and a change 5Qinv in the inversion charge:
5Va--
SQt + SQi,
(2)
while the drain current depends only on the inversion charge:
6ID = (W/L)VDpinvSQinv
(3)
where /**„„ is the mobility of electrons in the inversion layer. Therefore, pFE < (Mnv
(Eq. 1), as long as the magnitude of trapped charge increases with gate voltage.
The relative change in the inversion charge becomes larger in strong inversion, and
the field effect mobility increases with gate voltage. If the interface state density
is large, very large gate voltages would be needed for the field effect mobility to
approach the inversion layer mobility. With a large positive oxide charge pushed
toward the semiconductor interface during BTS, strong inversion can exist at a lower
gate bias. Consequently, the field effect mobility increases. The total mobile ion
charge is expected to be larger in the thick-oxide MOSPET, which results in a more
pronounced mobility improvement.
CONCLUSION
Lateral n-channel MOSPETs with two different gate oxide thicknesses (900nm and
200nm) were fabricated on 4H-SiC. Upon bias stressing the field-effect mobility of
the MOSFETs increased by over 2 orders(one order) of magnitude in the case of
the thick (thin) oxide MOSFET. The field-effect mobility was degraded upon the
application of a negative bias stress but the high mobility state could be recovered by
a subsequent positive BTS. The improvement of the transconductance and the field
effect mobility of the MOSFETs is attributed to the effect of positive oxide charges
on the dependence of interface state occupancy and inversion charge on gate bias.
After Positive BTS
-60
-40
VoltagefV)
Fig. 6: Capacitance-voltage characteristics before and after BTS.
67
0.03
ST 0.025
|
0.02
3a
0.015
After Positive
c
§ 0.01
c
E
Before BTS
i= 0.005
20
40
60
80
100
Gate VoKagefV)
Fig. 7: Gate voltage dependence of transconductance of the thick oxide MOSFET before
and after bias stressing (Vu=100mV)
ACKNOWLEDGEMENT
The authors from Rensselaer Polytechnic Institute, Troy, NY, gratefully acknowledge
the support of this work by Philips Research, Briarcliff Manor, NY, MURI of the
Office of Naval Research under contract no. N0014-95-1-1302 and the Center for
Power Electronic Systems (NSF Engineering Research Center under contract no. CR19229-427756)
REFERENCES
[1] T. Ouisse, Phys. Stat. Sol. (a) 162, 239 (1997).
[2] E. Arnold, N. Ramungul, T. P. Chow and M. Ghezzo, Proc. 7th Int. Conf. on SiC,
Ill-nitrides and related materials, Stockholm, 1013 (1997).
[3] S. Sridevan and B. J. Baliga, IEEE Electron Device Lett. 19, 228 (1998).
[4] E. H. Nicollian and J. R. Brews," MOS( Metal Oxide Semiconductor) Physics and
Technology, Ch 15, Wiley 1982.
FULL BAND MONTE CARLO SIMULATION OF SHORT CHANNEL MOSFETs
IN 4H AND 6H-S1C
M. HJELM1'2, H-E. NTLSSON1'2, E. DUBARIC1-2, C. PERSSON3, P. KÄCKELL4, C. S.
PETERSSON2
'Department of Information Technology, Mid-Sweden University, S-851 70 Sundsvall, Sweden,
Mats.Hjelm@ite.mh.se
department of Solid State Electronics, Kungl. Tekniska Högskolan (KTH), Elektrum 229, S-164
40 Kista, Sweden
department of Physics and Measurement Technology, Linköping University, S-581 83
Linköping, Sweden
4
Institut für Festkörpertheorie und Theoretische Optik, Max-Wien-Platz 1,07743 Jena, Germany
ABSTRACT
This is a presentation of a full band Monte Carlo (MC) study, which compares electron transport and device performance for 4H and 6H-SiC 100 nm n-channel MOSFETs. The model used
for the electrons is based on data from a full potential band structure calculation using the Local
Density Approximation (LDA) to the Density Functional Theory (DFT). For the holes the transport is based on a three band k-p model including spin orbit interaction. The two polytypes are
compared regarding surface mobilities obtained with the program, as well as transconductance,
unit current gain frequency, carrier velocity, I-V characteristics and energy distribution in the
channel for the MOSFETs.
INTRODUCTION
Silicon carbide (SiC) is considered to be a very promising material for high temperature and
high power applications due to its high breakdown voltage and high thermal conductivity. One
possibility is the fabrication of high speed integrated MOSFETs in 4H-SiC on a semi-insulating
substrate, with the expectation of a reliable and stable operation. An advantage for both 4H-SiC
and 6H-SiC is their discontinuous energy spectrum in the conduction band along the c-axis direction, which results in limited carrier heating by the electric field. Another factor tending to lower
the carrier energy is the strong polar optical scattering. CMOS integrated circuits have been fabricated on 6H-SiC [1,2], showing that commercial SiC MOSFETs will be available in the near future.
This paper presents a comparison, using a full band Monte Carlo model, of 4H-SiC and 6HSiC regarding the surface mobility and device properties in short channel MOSFETs.
MONTE CARLO MODEL
The full band MC program is based on a large lookup table, stored for the irreducible part of
the Brillouin zone and containing the energy and energy gradient versus k-vector from the band
structure. Between the points represented in the lookup table, the energy values are calculated using a second order cubic spline, and the energy gradient by linear interpolation. The band data in
the simulations presented here are based on the calculations in reference [3] and reference [4].
The following scattering processes are considered: acoustic phonon scattering, polar optical
phonon scattering, zero and first order optical intervalley phonon scattering and ionised impurity
scattering [5,6]. For the acoustic and optical phonons, the coupling constants are obtained by fitting data from bulk simulations to experimental data from references [7] and [8]. As a result the
transport properties in the simulations correspond to those in the material used in the referred experiments.
In the case of surface scattering, we are using the semi-empirical model proposed by E. Sangiorgi and M. R. Pinto [9], which is based on a combination of specular and diffusive reflection in
the interface. The diffusive scattering is obtained by the random selection of a k vector in such a
way that the energy is conserved and the movement is directed away from the surface. To perform
69
Mat. Res. Soc. Symp. Proc. Vol. 572 ® 1999 Materials Research Society
specular scattering, which is merely an elastic bounce, the normal component of the movement is
reversed and the component parallel to the interface is maintained. As both 4H and 6H-SiC are anisotropic the diffusive scattering is also anisotropic with the angular distribution corresponding to
the density of states in the k-space. Each time a carrier hits the oxide surface a random number is
used to select between the two types of scattering, with a constant (Q defining the probability of
diffusive scattering. According to reference [9] a C value of 0.06 results in a good correspondence
between MC simulations and experiments for good interfaces between Si and Si02. In our simulations we have used this value to simulate a high quality interface. For the simulation of a poor
interface we have used a C value of 0.50.
For the overlap integral of electron wave functions in 4H-SiC, we are using equation (1)
which is an expression often used for cubic semiconductors [10], where k and k' are the initial and
final states, a (= 0.323 for band 1 and 0.8 for band 2, 3 and 4) is the nonparbolicty parameter, e
and e' are the initial and final energies and 0 is the angle between the initial and final states.
A/2,,
,A/2
,
,A/2
p. , .,. _ [(1 + ae) (1+0x0
+«(EE)
' >~
(l+2ae)(l + 2ae')
1(
-,2
n
cosQ]
(l)
Due to the high degree of anisotropy, we use another approximate analytic expression, equation (2), for the overlap integral in 6H-SiC.
f(k,k') = \-{0.8732 + 0.0268 siny){l-exp[-1.01688 I02(l-0.855 siny)\k-kf)]} (2)
Here, y is the angle between the vector q = k - k' and the c-axis. The simulation is self consistent and uses a two dimensional solver for Poisson's equation. All other calculations are performed in three dimensions. There is no need to consider impact ionisation since its threshold
energy is very high. The model neither takes into account carrier generation and recombination
processes, nor does it consider the carriers at the oxide interface as a two-dimensional gas. All the
simulations in the study are made using the same version of the program.
SIMULATION RESULTS
Surface mobilities
The surface mobilities, which are shown in table I are calculated from the diffusion coefficients with the Einstein relation, where the carrier mean energies are sampled values from the simulation. The diffusion coefficients are also obtained from sampled data according to equation (3),
where u is the position along the x, y or z axis respectively and Du is the corresponding diffusion
coefficient.
Du = $((u-{«))2)
(3)
The simulations are made using a simplified device consisting of two regions with SiC and
Si02, and with the plane interface between the two materials.We have assumed an electron concentration of 4 x 1018 since the simplified device does not permit calculation of the electron concentration in the channel. The doping level was 1 x 1013 donors/cm3 and 1 x 1017 acceptors/cm3.
The field perpendicular to the interface was 500 kV/cm.
As can be seen from the table, 6H-SiC has a strong anisotropy with a ratio > 4.8. 4H-SiC is
also anisotropic with a ratio of about 0.8 - 0.9. Since the mobility is so low for transport in the caxis direction for 6H-SiC, we only consider electron transport perpendicular to the c-axis in the
remainder of this study. The surface mobilities with C=0 are superior for 4H-SiC. The values for
4H are better than the corresponding bulk mobilities which can be explained by a higher degree of
screening of impurities. For good interface qualities, 4H-SiC still has better mobility but with a
smaller advantage. In the case of inferior oxide quality, 6H-SiC has an advantage. We interpret
70
Table I: Surface mobilities in cm2/Vs. The texts "i c-axis" and "c-axis" refer to the transport
direction in relation to the c-axis.
6H-SiC
4H-SiC
Temp. K
C
± c-axis
i c-axis
c-axis
c-axis
300
0.00
862
968
358
74
300
0.06
371
454
308
50
300
0.50
74
88
148
14
500
0.00
218
260
115
25
500
0.06
169
193
108
21
500
0.50
61
66
80
11
these characteristics as the result of a higher probability for electrons in 6H-SiC to find final states
after scattering, with a large velocity component perpendicular to the c-axis. In figure 1 we show
the velocities for 1000 random points in k-space with the energy 0.05 eV for 4H and 6H-SiC. In
4H-SiC the velocities are distributed near the surface of a sphere, while the velocity distribution in
6H-SiC has a flattened form. It can be seen in the figure, that for 6H-SiC the majority of the electrons have a relatively small component parallel to the c-axis. In this context is it also important to
consider that the same C constant does not a priori correspond to the same interface quality in the
two polytypes.
4H-SJC
6H-SJC
20 -I
0-
-20
20
Fig. 1. Velocity distribution for 1000 random points i k-space with energy = 0.05 eV. The velocities are
given in cm/s x 106. The coordinates are oriented with z parallel to the c-axis.
MOSFET simulations
The simulated MOSFET is shown in figure 2, and in table II the simulation results are shown
for transconductance (gm), total gate capacitance (Cgtor) and unit current gain frequency (fT). These
values are for a source potential of 0 V, a drain potential (V&) of 2.0 V and represent the mean
value for gate potentials (Kgi) between 1.3 and 2.3 V, with an approximate threshold voltage (K,A)
of 0.8 V, assuming that the gate is made of aluminium and with no charges in the oxide or on the
71
100 nm
Fig. 2. Simulated MOSFET transistor structure. The width of the device is
1 |im. Doping values are in cm"3.
Table II: Simulation results for the MOSFETs.
6H-SiC
4H-SiC
Temp. K
C
S x 10-4
Cgtot
fTGHz
F x 1016
gmSx
lO"4
Cgtot
F x 1016
fTGHz
300
0.06
6.12
5.10
191
4.25
4.86
139
300
0.50
3.80
4.89
123
3.63
4.83
120
500
0.06
5.00
4.98
160
3.10
4.63
105
500
0.50
3.42
4.90
111
2.81
4.62
95
oxide-semiconductor interface. For the calculation of gm and CgM, the drain current (JD) as well as
the total charge on the gate (ß„to,) were sampled at equidistant V$s values with a difference ot 0.1
V. The gm was calculated as the quotient of difference in ID and Vgs. Similarly, Cgto, was obtained
as the quotient of difference in Qgtot and Vgs.
In contrast to the bulk mobilities, the MOSFET characteristics are better in all cases for the
4H polytype although the difference is so small for C = 0.50 that it may be considered insignificant This is not surprising as the MOSFETs work with high electric fields parallel to the channel
and the carrier mean velocity reaches saturation or near saturation, i.e. the low field mobilities are
not applicable, see figure 3. Furthermore, at the beginning and end of the channel, the electron velocity component perpendicular to the oxide interface is important. Consequently, the very low caxis mobility for 6H-SiC results in smaller current. An interesting fact, that can also be seen in figure 3 is that 6H-SiC besides lower velocity, has a much lower peak value for the mean energy in
72
(a)
0.4
v 4H, T=300
o 4H, T=500
5"0.3 - A 6H, T=300
o 6H, T=500
v 4H, T=300
- o 4H, T=500
A 6H, T=300
- o 6H, T=500
-
1
k
«0.2
(b)
Nxio
I
fR
u
o
| 0.1
3
^=-fi-#S
g/0
«$:
50
100
150
Distance [nm]
;
200
50
100
150
200
Distance [nm]
Fig. 3. Mean energy in the channel (a) and mean velocity parallel to the interface (b). Vgs=\.% V, Vds=2.Qt
V, C = 0.06. The distance is measured from the left border of the component.The mean energy is for a
distance up to 2.5 nm from the interface, and the mean velocity for all electrons in the channel.
the channel. The I-V characteristics for C=0.06, are shown in figure 4, together with the corresponding maximum mean energy in the channel. As can be seen, 6H-SiC has an advantage as it
has lower energy when Vds > 0.8 V, which increases with higher drain potential.
To get a comparison with a corresponding device in silicon, we have made a simulation with
the Medici program [11] using the energy balance model. The Si device was identical with the one
in SiC, with the exception that the doping in the channel region was increased to 1.3 x 1018 which
resulted in a threshold gate potential of 0.8 V. In this case the maximum carrier energy with
Fgs=1.8 V and ^=2.0 V, was 0.54 eV, which may be compared with the carrier energies 0.39 eV
and 0.25 eV for 4H-SiC and 6H-SiC respectively, obtained with the MC program. The approximate maximum fields were: 5.4 x 105 V/cm parallel to the interface in the Si device and 6.5 x 105
in the SiC device, 6.8 x 105 V/cm and 4.2 x 105 perpendicular to the interface in Si and SiC respectively.
(b)
x 10-4
(a )
^fr^
v4H,t=300k:
7
_
5 -o 4H, T=500 K^ ^^
6
0.4
0.3
4
^P^3^5^^ '
jr
23 '
2-
l nT^^^^^^
Q)
^fS0.2
<D
'
ciB^J3jl°^ r^
Id:
1 f/
JK
A
6H, T=300 K-
D
6H, T=500 K
v 4H, T=300
o 4H, T=500
A 6H, T=300
o 6H, T=500
0
I
// Jsf -
c
LU
0.1
I
n
1
1.5
2
"0
0.5
1
V
dsM
vds[V]
Fig. 4. I-V characteristics (a) and maximum mean energy (b). Vgs = 1.8 V, C = 0.06.
t
J
id
Jf
jy M
1
0.5
73
1.5
2
CONCLUSIONS
We have presented a full band Monte Carlo simulation, comparing both surface mobilities
and device performance for lOOnm n-MOSFETs in 4H and 6H-SiC. We have used the model of
Sangiorgi and Pinto [9] for the oxide semiconductor interface with a diffuse scattering factor (C)
modelling the oxide quality.
The surface mobility simulations show a difference in surface mobility in favour of 4H-SiC
for good interface quality (C=0.06), but for bad interfaces (C=0.50) 6H-SiC has higher mobility.
We consider this to be an effect of the strong anisotropy in 6H-S1C, with higher probabilities for
carriers to find final states after scattering in the channel direction.
The simulated MOSFETs show better transconductance and unit current gain frequency for
4H-SiC in all cases. For instance, gm was 6.12 x 10"4 S and/j-191 GHz for the 4H-SiC device with
a gate width of 1 urn at 300 K and C=0.06. The corresponding values for 6H were 4.25 x 10"4 S
and 139 GHz. However, for the poor interface quality, the difference is small and the/j-values are
123 and 120 GHz for the same conditions as above. This means that in order to benefit from the
higher mobility in 4H-SiC a good oxide quality is necessary. The electrons in the channel have
lower energies in the SiC MOSFETs than in a corresponding Si device. At high V& values, 6H-SiC
has an advantage having lower carrier energies than 4H-SiC. For instance with F^=1.8 V, ^=2.0
V, T= 300 K, and C=0.06, the maximum mean energy in the channel is 0.39 evfor 4H-SiC and
0.25 eV for 6H-SiC.
The difference in carrier energy in favour of 6H-SiC is a potential that could be utilized in
device designs, leading us to the conclusion that 6H-SiC may be as good as, if not better than, 4HSiC in MOSFET devices.
ACKNOWLEDGMENTS
Financial support from ISS Foundation and Mid-Sweden University is gratefully acknowledged.
REFERENCES
1. J. N. Shenoy, J. A. Cooper, Jr. and M. R. Melloch, IEEE Dev. Lett. 18, 93 (1997).
2. J. Spitz, M. R. Melloch, J. A. Cooper, Jr. and A. Capano, IEEE Dev. Lett 19, 100 (1998).
3. P. Käckell, B. Wenzien and F. Bechstedt, Phys. Rev. B 50 (15) 10761 (1994).
4. C. Persson and U. Lindefelt, J. Appl Phys. 82 (11), 5,496 (1997).
5. K. Tsukioka, D. Vasileska and D. K. Ferry, Physica B 185, 466 (1993).
6. H-E. Nilsson, U. Sannemo and C. S. Petersson, J. Appl. Phys. 80 (6), 3365-3369 (1996).
7. M. Schadt, G. Pensl, R. P. Devaty, W. J. Choyke, R. Stein and D. Stephani, Appl. Phys. Lett.
65, 3120 (1994).
8. W. J. Schaffer, G H. Negley, K. G. Irvine and J. W. Palmour in Diamond, Silicon Carbide and
Nitride Wide Bandgap Semiconductors, edited by C. H. Carter, G Gildenbalt, S. Nakamura, and
R. Nemanich, (Mater. Res. Soc. Proc. 339, Pittsburgh, PA 1994) pp. 595-600.
9. E. Sangiorgi and M. R. Pinto, IEEE Trans, on Elec. Dev. 39 (2), 356-361 (1992).
10. W. Fawcett, A. D. Boardman and S. Swain, J. Phys. Chem. Solids, 31, 1963 (1970).
11. Avant! Corporation, TCAD Business Unit, Fremont, California. Medici, Two-Dimensional
Device Simulation Program, Version 4.1, Users Manual, July 1998.
74
High Voltage Schottky Barrier Diodes on P-Type SiC using Metal-Overlap on a Thick
Oxide Layer as Edge Termination
Q. Zhang, V. Madangarli, S. Soloviev and T. S. Sudarshan
Department of Electrical Engineering
University of South Carolina, SC 29208, U.S.A
Tel: 803-777-7302; Fax: 803-777-8045
E-mail: Zhang@engr.sc.edu
ABSTRACT
P-type 6H SiC Schottky barrier diodes with good rectifying characteristics upto breakdown
voltage as high as 1000V have been successfully fabricated using metal-overlap over a thick
oxide layer (~ 6000 Ä) as edge termination and Al as the barrier metal. The influence of the
oxide layer edge termination in improving the reverse breakdown voltage as well as the
forward current - voltage characteristics is presented. The terminated Schottky diodes indicate
a factor of two higher breakdown voltage and 2-3 times larger forward current densities than
those without edge termination. The specific series resistance of the unterminated diodes was
-228 mfi-cm2, while that of the terminated diodes was -84 mil-cm2.
INTRODUCTION
Silicon carbide (SiC) has recently been given renewed attention because of its fine
characteristics such as stability at high temperatures, wide bandgap, high breakdown field and
high thermal conductivity. In the case of high-voltage devices, edge termination plays a very
critical role in determining the breakdown voltage. High breakdown voltage (~ 730 VJ SiC
Schottky barrier diodes have been reported using edge termination techniques [1,2]. However,
in these processes ion implantation is required to obtain a high resistivity layer on the surface at
the edges of the device [1]. A metal-overlap onto an oxide layer at the edge of the device can
be used to minimize the electric field enhancement so that the breakdown voltage can approach
the ideal plane parallel value of the SiC wafer [3]. But due to the slow oxidation rates on SiC it
is difficult to get a thick oxide layer on SiC by conventional thermal oxidation of SiC [4].
Moreover, although there have been a lot of publications on high voltage Schottky diodes on ntype SiC, there is very little work reported on high voltage Schottky contact on p-type SiC. To
the best of our knowledge the only significant study upto now on p-type SiC Schottky diodes
has been by R.Raghunathan and B.J.Baliga, who reported p-type 4H- and 6H-SiC Schottky
diodes with breakdown voltage upto 600V in 1998 [5], In this paper, we compare the forward
and reverse I-V characteristics of a conventional p-type 6H-SiC Schottky diode with no edge
termination to that of a 1 kV Schottky diode with a thick oxide layer edge termination.
EXPERIMENT
Schottky diodes were fabricated on a p-type 6H-SiC wafer from Cree Research (substrate
doping ~ 1.6xl018 cm"3) with a 10 micron thick epilayer of- 6xl015 cm"3 doping concentration,
and ideal parallel plane breakdown voltage of- 1200 V. The 30 mm diameter wafer was cut
into 10 mm x 10 mm square pieces to obtain several samples for experiments. All the samples
were cleaned by RCA procedure to obtain a SiC surface with minimal surface contamination
prior to diode fabrication. Unterminated Schottky diodes of approximately 140 urn diameter
75
Mat. Res. Soc. Symp. Proc. Vol. 572
@
1999 Materials Research Society
were fabricated on few samples following a conventional photolithographic process (Fig. 1(a)),
while Schottky diodes with oxide layer edge termination were fabricated as follows.
In order to obtain a thick oxide layer for edge termination, instead of sputter depositing
Si02 on top of the thermal oxide as reported by other groups [3], we have adopted a depoconversion technique which is discussed in a companion paper [6]. This technique essentially
involves sputter deposition of a thick Si layer on the SiC wafer and conversion of Si to Si02 by
oxidation. After conversion, the sample was annealed in Ar for half an hour in order to improve
the oxide quality.
The thick oxide layer was then selectively etched using a first mask to form the Schottky
contact window with a diameter of 140 microns. After a cleaning process, Al was evaporated
onto both sides of the sample in high vacuum (<10"5 Torr). A Schottky contact was formed on
the polished epilayer while an ohmic contact is obtained on the roughened backside of the
sample. The Al on the epilayer side was selectively etched using a second mask with a diameter
of 250 microns to fabricate individual Schottky diode structures with a metal overlap length of
55 micron over the thick oxide layer as shown in Fig. 1 (b). It has to be noted that, even though
generally it is sufficient to have an overlap approximately equal to the epilayer thickness [3],
we have used a larger overlap because of the availability of a photomask with 250 um diameter
circular features.
Al Schottky contact _
Al ohmic contact
6H SiC P-substrate
(1E18 cm'3)
6H-SiC P-epi (6E15 cm'3; 10 |am)
6H SiC P-substrate
(1E18 cm"3)
lU Oxide layer ~ 6000 A
(b)
00
Fig. 1. Cross section of Al/6H-SiC Schottky diode (a) without edge termination, and (b) with
thick oxide edge termination.
After the fabrication of the diodes, their forward and reverse I-V characteristics were
measured using a DC voltage source (Kiethley 237 High Voltage SMU) as well a pulse
measurement system. The pulse measurement system [7] comprised of a Tektronix FG540
function generator inputting a ramp pulse of- 500 us duration into a Trek 50/750 High Voltage
Pulse Amplifier, and a Tektronix TDS 540 Digitizing oscilloscope used for recording the
applied voltage and measuring the current through the diode structure.
76
RESULTS AND DISCUSSION
(a) Forward characteristics
Typical forward current density vs voltage (J-V) characteristics of Al/6H-SiC Schottky
diodes with and without edge termination under DC bias conditions are shown in Fig. 2. The
current density through the diodes with edge termination was observed to be approximately 2-3
times higher than that through the diodes without edge termination. The ideality factor (n) for
both the diodes were calculated to be between 1.1 and 1.8. The forward voltage drop at a
current density of 50 A/cm2 was ~ 11.7 V for the diodes without edge termination, while it was
~ 6.5 V for the diodes with edge termination. The series resistance (Ron,,sp) calculated from a
plot of IdV/dl vs I was found to be -228 mfi-cm2 for diodes without edge termination, while it
was ~ 84 mfi-cm2 for diodes with edge termination. These values are significantly higher than
the ideal series resistance value of ~ 13 mfl-cm2 calculated using the doping concentration and
thickness of the epitaxial layer and substrate provided to us by the manufacturer and the
mobility of holes in the epi-layer and substrate [5], The discrepancy between the ideal and
experimental Ron„sp values could be due to a significantly lower substrate doping concentration
and / or a large series resistance of the backside ohmic contact. Also, inherently the epitaxial
and substrate region series resistance is high for p-type SiC due to the large ionization energy
of the dopant atom (Al) in SiC [5,8], The rather high forward voltage drop of the p-type SiC
Schottky diodes could be attributed to the large series resistance values. The barrier height
calculated from the forward I-V measurement was between 1.2-1.7 eV, which is comparable to
that reported by Raghunathan et al [5] for P-type 6H-SiC Schottky diodes.
Forward Bias: D.C
-130
—
I
T"
-5
-10
-15
Voltage (V)
Fig. 2. Typical forward current density vs voltage (J-V) characteristics of Al/6H-SiC Schottky
diodes with and without edge termination under DC excitation.
The higher forward current density through the diode with termination could be attributed
to the formation of a low resistivity accumulation layer in the semiconductor below the MOS
structure surrounding the Schottky contact. This can result in an effective increase in the
current conduction area (Fig. 3), which will lead to a lower series resistance, and hence higher
currents and a lower forward voltage drop.
77
It has to be noted that the current density in Fig. 2 was calculated assuming uniform current
conduction across the 140 um dia. (Dl) Schottky contact area, while the effective current
conduction area in case of the terminated diodes could be larger (D2 >140 um) due to the
presence of a low resistivity accumulation layer surrounding the Schottky contact area. We are
in the process verifying this hypothesis by 2D numerical simulation using ATLAS, and a
detailed analysis of this phenomena will be presented in a future paper.
Low resistivity
accumulation layer
Current Flow
Fig. 3. Schematic of expected current flow pattern through the Al/6H-SiC Schottky diode (a)
without edge termination, and (b) with thick oxide edge termination.
(b) Reverse characteristics
A comparison of reverse bias current density vs voltage characteristics of the Schottky
diodes with and without edge termination under DC bias conditions is shown in Fig. 4.
Reverse Bias: D.C
100
e
With edge termination I
(Vbr~900V)
50
a
Without edge termination
(Vbr~S00V)f"
._.' _
_L
0
500
1000
Voltage (V)
Fig. 4. Typical reverse current density vs voltage (J-V) characteristics of Al/6H-SiC Schottky
diodes with and without edge termination under DC excitation.
The reverse breakdown voltage of the Schottky diodes with edge termination is observed to
be approximately two times larger than that of the diodes without edge termination, while the
reverse leakage current densities of both diodes were < 100 uA/cm2 upto 400 V. Both the
78
diodes also indicated a reasonably high ON/OFF ratio (h,sv I JR,20OV) of about 4 X 10 .
Extensive 2D numerical simulation using ATLAS (to be presented in a future paper) confirms
that the increase in the breakdown voltage of the diodes with edge termination is due to an ~
30% reduction in the peak electric field at the Schottky contact periphery as a result of the field
plate surrounding the Schottky contact. In fact, the numerical simulation clearly indicates the
expansion of the depletion region below the MOS region surrounding the Schottky contact, and
hence the high field region is shifted away from the Schottky contact edge to inside the oxide
layer corresponding to the edge of the metal overlap on top of the oxide layer, resulting in a
higher breakdown voltage.
(c) Effect of oxide thickness on reverse breakdown voltage
Under reverse bias conditions of the terminated Schottky diodes, the oxide layer will
sustain a portion of total applied voltage. As mentioned earlier, the high field region now shifts
away from the Schottky contact edge into the oxide layer, near the edge of the metal overlap. If
the electric field in the oxide layer exceeds the breakdown field of SiC>2, the Schottky diode
will fail due to breakdown of the oxide layer. In order to investigate the influence of the
thickness of the oxide layer on the Schottky diode breakdown voltage, Schottky diodes with
different oxide thicknesses were fabricated. Fig. 5 shows the normalized pulse J-V
characteristics of Schottky diodes, with different oxide thicknesses, under reverse bias.
I
Deposited oxide
3000 A : V, = 5S3 V
Deposited oxide
1700 A :V, = 380V
Deposited oxide
6000 A : V. = 1000 V
Unterminated'
*kr I *327 V !
Thermal oxide
300A:V„ = 263V
i_,
0
100
200
300
400
,
I
500
1
600
700
800
900
1000
Voltage (V)
Fig. 5. Normalized pulse J-V characteristics of Schottky diodes with different oxide
thicknesses, under reverse bias.
It is interesting to note that Schottky diodes with a thin thermal oxide (300 A) failed at a
lower voltage than the unterminated Schottky diode, indicating possible oxide breakdown when
the oxide layer is too thin. But it has to be noted, however, that due to poor SiC material
quality, a large scatter was observed in the breakdown voltage of both the unterminated and
terminated diodes (> 20% difference between Vbr,min and Vbr,max). Hence the anomalous
reduction in the diode breakdown voltage with a thin oxide could also be due to poor material
quality. A detailed measurement of the breakdown voltage of several unterminated and thin
79
oxide terminated Schottky diodes is necessary to pin point the exact reason for the anomalous
breakdown of a thin oxide terminated diode. Still, from Fig. 5 it is clearly evident that the
breakdown voltage increases with increase in the oxide thickness. ATLAS simulation results
(to be presented in a future paper) indicates that if the oxide thickness less than a minimum
value, the field plate edge termination is not effective, and a thin oxide terminated diode
structure can possibly fail at a lower voltage than an unterminated diode. Also, while the
breakdown voltage is observed to increase with increase in the oxide thickness, theoretically
beyond a certain value of oxide thickness there is no significant increase the breakdown
voltage. For the Schottky diodes investigated in our laboratory, increasing the oxide thickness
beyond 6000 A did not result in further improvement in the breakdown voltage (up to 10000 A
thick oxide layers were investigated in the present experiment).
CONCLUSION
High voltage p-type 6H SiC Schottky diodes up to 1000 V have been successfully
fabricated using a thick oxide layer for edge termination. The forward current density of the
Schottky diodes with edge termination was 2-3 times larger than those for diodes without edge
termination, while the reverse leakage current density of both diodes were < 100 uA/cm2 upto
400V. The difference in forward current densities could be due to a larger effective conduction
area for diodes with edge termination, as a result of an accumulation layer underneath the MOS
region surrounding the Schottky contact. Also, the Schottky diodes with edge termination
indicated a factor of two higher breakdown voltages than those without edge termination as a
result of electric field relief at the Schottky contact edge. A minimum oxide thickness of- 6000
A was necessary to attain 1 kV reverse breakdown voltage for the Schottky diodes fabricated
on p-type 6H-SiC in this experiment. Further increase in oxide thickness did not result in any
improvement in the breakdown voltage.
ACKNOWLEDGEMENTS
This work was supported by ARO (grant no DAA H04-96-1-0467) via a DEPSCoR program.
The authors also wish to acknowledge Dr. I. Khlebnikov for his valuable suggestions and
Mr. Marc Tarplee for performing the ATLAS simulation.
REFERENCE
1.
2.
3.
4.
5.
6.
D. Alok and B.J. Baliga, IEEE Trans. ED 44, 1013 (1997).
A. Itoh, T. Kimoto and H. Matsunami, IEEE Trans. EDL 17, 130 (1996).
V. Saxena, J. N. Su, and A.J. Steckl, IEEE Trans. ED 46, 456 (1998).
M. Bhatnagar and B.J. Baliga, IEEE Trans. ED 40, 645 (1993).
R. Raghunathan and B.J. Baliga, IEEE Trans. EDL 19, 71 (1998).
Q. Zhang, V. Madangarli, I. Khlebnikov, S. Soloviev and T. S. Sudarshan, to be presented
at 1999 MRS Spring Meeting (unpublished).
7. T. S. Sudarshan, V.P. Madangarli, G. Gradinaru, C.C. Tin, R. Hu, and T. Isaacs-Smith,
MRS Proc. 423, 99 (1996).
8. S.M. Sze, Physics of Semiconductor Devices, 2nd. Ed. (John Wiley & Sons Publishers, New
York, 1981).
80
HIGH VOLTAGE P-N JUNCTION DIODES IN SILICON CARBIDE
USING FIELD PLATE EDGE TERMINATION
R.K. CPDLUKURI, P. ANANTHANARAYANAN, V. NAGAPUDI, B.J. BALIGA
Power Semiconductor Research Center, North Carolina State University
Raleigh, NC 27606
ravi@apollo.psrc.ncsu.edu
ABSTRACT
In this paper, we report the successful use of field plates as planar edge terminations for
P+-N as well as N*-P planar ion implanted junction diodes on 6H- and 4H-SiC. Process splits
were done to vary the dielectric material (SiC«2 vs. Si3N4), the N-type implant (nitrogen vs.
phosphorous), the P-type implant (aluminum vs. boron), and the post-implantation anneal
temperature. The nitrogen implanted diodes on 4H-SiC with field plates using S1O2 as the
dielectric, exhibited a breakdown voltage of 1100 V, which is the highest ever reported measured
breakdown voltage for any planar ion implanted junction diode and is nearly 70% of the ideal
breakdown voltage. The reverse leakage current of this diode was less than lxlO"5 A/cm2 even at
breakdown. The unterminated nitrogen implanted diodes blocked lower voltages (-840V). In
contrast, the unterminated aluminum implanted diodes exhibited higher breakdown voltages
(-800V) than the terminated diodes (-275 V). This is attributed to formation of a high resistivity
layer at the surface near the edges of the diode by the P-type ion implant, acting as a junction
termination extension. Diodes on 4H-SiC showed higher breakdown than those on 6H-SiC.
Breakdown voltages were independent of temperature in the range of 25 °C to 150 °C, while the
leakage currents increased slowly with temperature, indicating surface dominated components.
INTRODUCTION
SiC has become an attractive semiconductor in recent years for high speed, high power
and high temperature applications because of its wide band-gap, high critical electric field and
high thermal conductivity. However, its widespread application has been limited due to the fact
that the device fabrication technology for SiC is still in its stage of infancy. Due to low impurity
diffusion coefficients, doping in SiC is usually obtained either by epitaxial growth or by ion
implantation. The former is not attractive because it requires mesa etching for edge termination,
and hence the device topology becomes non-planar making passivation difficult. Hence, in order
to improve the commercial viability of SiC devices, the successful development of planar ion
implantation technology for SiC for microelectronic technology is of major importance. Dopant
activation, surface morphology control and ohmic contacts have been the major obstacles in the
development of a reproducible ion implantation technology. Until recently, P-N junctions in SiC
have been formed using multiple epitaxial layers or unmasked ion implantation. However, these
approaches compromise on surface planarity and are not preferred [1-3].
Although there have been quite a few reports on ion implantation, 6H-SiC has received
most of the attention [2-6]. In most cases, the dopants used have been nitrogen, aluminum and
boron. There have been only a couple of reports on other dopants like phosphorous and
beryllium [7,8] where breakdown voltages of 675 V were reported for phosphorous implanted
non-planar junctions. The diodes fabricated in the above work have been mesa-etch terminated
which are not compatible with IC technology. Various planar edge terminations such as floating
metal rings and resistive Schottky barrier field plates have been explored for 6H-SiC devices [9]
but only 50% of ideal breakdown voltage was achieved. A planar, near ideal, edge termination
81
Mat. Res. Soc. Symp. Proc. Vol. 572 ° 1999 Materials Research Society
using Argon implantation [10,11] has been reported for 6H-SiC Schottky barrier diodes but the
leakage current is greatly increased by the implanted region. This method was also demonstrated
for 4H-SiC with breakdown voltages exceeding those reported for 4H-SiC mesa-etch terminated
diodes [12]. Most recently, a 3.4 kV ion implanted PIN-rectifier has been implemented on 4HSiC with low leakage currents obtained by using junction termination extension created by boron
ion implantation [13]. In this paper, reverse blocking characteristics of diodes made using
phosphorous, nitrogen, aluminum and boron implanted layers and having different edge
terminations are discussed.
EXPERIMENT
The starting wafers were research grade 6H- and 4H-SiC wafers (both n-type and p-type)
consisting of a 10 um thick epitaxial layer (0.85-1.0xl016 cm"3) on a 300 ^m thick substrate
(lxlO18 cm"3), obtained from CREE. Devices were fabricated using two process sequences
differing mainly in the dielectric used to mask the implants, the temperature of post implantation
anneal and the final dielectric used in device termination.
The first fabrication process was a 2-mask process in which the post-implantation anneal
was at <1400 °C and the dielectric used for passivation was silicon dioxide. In this process, after
cleaning the wafers by standard techniques, a 7000 A thick oxide was deposited using Low
Pressure Chemical Vapor Deposition (LPCVD). Next, windows for ion implantation were
opened in the oxide, and a 500 A pad oxide was deposited by LPCVD. This was to ensure that
the peak implant doping occurs at the surface, which would result in good ohmic contacts. High
temperature (1000 °C) nitrogen and aluminum implantations were performed using multiple
implant sequences with energies 25, 50, 75 and 120 keV to give box-profiles with varying total
doses. Monte Carlo simulations performed using SUPREM showed that these implant conditions
would yield depths of uniform concentration or junction depths in the range of 0.25-0.3 um in the
case of the nitrogen implants and 0:17-0.2 urn in the case of the aluminum implants. Postimplantation proximity anneals for dopant activation were done at 1300 C and 1400 C for the
nitrogen and aluminum implanted samples, respectively, in Argon for 30 minutes. Aluminum
contacts were formed by lift-off on the device side, after etching the pad oxide. Blanket
evaporation of aluminum was done on the back side to form ohmic contacts to the substrate.
For post-implantation anneals at temperatures >1600 °C, which is desirable for high
dopant activation, the above process was not feasible due to reflow of the oxide at such
temperatures. Hence, a 4-mask process was designed for device fabrication. In this process, the
silicon nitride was used as the dielectric for passivation. Diodes were also made from
phosphorous and deep high energy boron implants, and planar edge terminations such as field
rings were also incorporated. After cleaning the wafers, the first mask level was used to etch
alignment marks on bare SiC surface using RIE. Following an RCA clean, 1 um each of oxide
and polysilicon were deposited by LPCVD. Such a thick stack of polysilicon and oxide was
found, based on simulations, to be required to mask the high energy boron implants. Windows
were opened in the stack of polysilicon and oxide for implantation and a 500 A of pad oxide was
deposited by LPCVD on both wafers for obtaining peak doping of the implantation profile at the
surface. Based on simulations, the implantation profiles yielded uniform concentration upto a
depth of 0.3, 0.2, 0.2 and 0.6 urn in case of nitrogen, phosphorous, aluminum and deep
boron/aluminum implants, respectively, with net doses of 4.6xl015, 4.6xl015, 3.4xl016 and
6.8xl014 cm"2. After ion implantation, blanket etch was done to remove both the oxide and the
polysilicon. Next, post-implantation proximity anneals were performed for all wafers at 1600 °C
in Argon for 30 min in a SiC crucible. The SiC crucible helped maintain a vapor pressure of SiC
82
10 um field plate
Dose- 2.6el5
4H-SiC
6H-S1C
(00
800
1000
u
1200
V (volts)
Mu
'wu
***
1200
V(\folts)
(a)
(b)
Fig. 1. Reverse J-V characteristics of nitrogen implanted diodes with SiC>2 as dielectric : (a) comparison of
6H- and 4H-SiC diodes with field plates; (b) effect of field plate length.
and hence completely prevented surface degradation and pitting [4]. Following the post
implantation anneal, a 0.55 ^m thick nitride layer was deposited by LPCVD. A 0.4 |im LPCVD
oxide was used as the mask to open contact windows in the nitride film. Aluminum contacts were
formed by lift-off for the device side, and by blanket evaporation on the back side.
RESULTS & DISCUSSION
A. Diodes with S1O2 as dielectric (post-implantation anneal at < 1400 °C)
The diode reverse I-V measurements were performed using Keithley 251 I-V setup and
the system leakage was found to be a few pico-amperes. It is known that edge termination plays
an important role in determining the breakdown voltage of a p-n junction diode. In the case of
nitrogen implanted diodes on 6H-SiC, field plates served to increase the breakdown voltage from
350 ± 50 V to 500 ± 20 V. The breakdown voltage of nitrogen implanted diodes on 4H-SiC
increased from 800 ± 40 V to 1100 ± 60 V with field plates. This indicates that a simple planar
field plate edge termination can be used to obtain higher breakdown voltages. Extremely low
leakage current densities in the range of 1 x 10'5 A/cm2 were obtained in both 6H- and 4H-SiC
nitrogen implanted diodes terminated with field plates even just before breakdown. A comparison
of the reverse I-V characteristics of these nitrogen implanted diodes on 6H- and 4H-SiC with field
plates are shown in Fig. 1(a). It can be observed that much higher breakdown voltages were
obtained in 4H-SiC than in 6H-SiC. The variation of the breakdown characteristics with the dose
of nitrogen implant was studied in 6H-SiC and no significant trend or variation was found. The
nitrogen implanted junction diodes without field plates on 4H-SiC also showed good reverse
blocking characteristics with a maximum breakdown voltage of 840 V and average leakage
currents in the range of lxlO"4 A/cm2 prior to breakdown as shown in Fig. 1(b). A breakdown
voltage of 1100 V obtained in this study on 4H-SiC is the highest ever reported measured
breakdown voltage for any planar ion implanted junction diode on SiC. The breakdown voltage is
nearly 70 % of the theoretical parallel plane ideal breakdown voltage of about 1600 V for a diode
with the given epitaxial layer specifications. A total of 30 devices were measured, 15 with field
plates and 15 without field plates, and it was found that the presence of field plate improved the
reverse I-V characteristics of the diodes consistently and as expected from theory [14]. The
breakdown voltage was found to be almost independent of the of field plate length as shown in
Fig. 1(b). The distribution of the leakage currents just before breakdown for the nitrogen
implanted diodes on 4H SiC across the sample is shown in Fig. 2. The average leakage current
for diodes with field plates, reduced by an order of magnitude when compared to that of diodes
without field plates, to a value as low as lxlO"5 A/cm2 just before breakdown.
83
4H-SIC
«tsc
E
i
WthoutlteH
avtfhMd
10"'
10*
200
Leaka^ GOTO* Density (A/arf)
400
600
800 1000
Reverse Voltage (Volts)
1200
Fig. 3. Comparison of reverse J-V characteristics
of nitrogen and aluminum implanted diodes.
Fig. 2. Reverse leakage current density histogram
of nitrogen implanted diodes near breakdown.
The aluminum implanted junction diodes on 4H-SiC exhibited higher breakdown voltages
without field plates than with field plates (Fig. 3). For a dose of 6.8xl015 cm-2, while the
aluminum implanted diodes without field plates showed breakdown voltages of 800+10 V, the
ones with field plates supported only 275±40 V. This is attributed to the poor activation of the
aluminum implanted region due to which, the actual doping in the ion implanted region is much
lower than the implanted aluminum concentration. The sheet resistance of the aluminum implanted
region, measured using Kelvin test elements on the same wafers, and was found to be 25
kfl/square which supports the above argument. This results in the presence of a high resistivity
layer at the surface near the edges of the diode (between points A and B in Fig. 4(a)) in case of
the diodes without field plates. It is believed that this region then acts like a junction termination
extension and promotes the spreading of the potential along the surface laterally which results in
reduced electric field [13] at a given voltage for the diodes without field plates. Hence, a higher
breakdown voltage is observed for diodes without field plates when compared to diodes with field
plates. The breakdown voltage of aluminum implanted diodes was found to be higher for a higher
dose. However, like in the nitrogen implanted diodes, no significant variation of breakdown
voltage with the length of field plate was observed.
B. Diodes with S13N4 as dielectric (post-implantation anneal at 1600 °C)
The reverse I-V characteristics were measured for all diodes with different edge
terminations and the results are summarized in Fig. 5. The unterminated diodes gave the lowest
Oxide
metal contact
k 0.7pm
implanted layer
10pm A
300pm
^
epi layer (1x10" cm-3)
epi layer (1x10" cm-3)
substrate (1x10" cm"3)
300pm
substrate (1x10" cm"3)
Fig. 4. Cross-section of fabricated planar diodes with Si02 as dielectric : (a) without field plate (b) with
field plate.
84
Implanted
Species
N
No edge
termination
300 +/-25 V
Single field
plate
550 +/-20 V
Single field
ring
350 +/-100 V
1 field ring +
1 field plate
350 +/- 30 V
2 field rings +
1 field plate
500 +/-60 V
P
400 +/-S0 V
700 +/-40 V
400 +/-100 V
400 +/- SO V
475 +/- 100 V
Al
ISO +/-20 V
600 +/-50 V
260 +/-40 V
300+/- 30 V
300+/- 100 V
B and Al
7S+/-50 V
350+/-50 V
200 +/-20 V
200 +/- 50
175+/-2S V
Fig. 5. Variation of breakdown voltages with edge terminations of 4H-SiC diodes using Si3N4as dielectric.
breakdown voltages in all cases. The highest breakdown voltages were obtained for the diodes
with a single field plate for all the different dopants. The least variation of breakdown voltages
across the wafer was observed for these diodes. This confirmed that a simple planar field plate
edge termination can be used to obtain higher breakdown voltages. However, no other trend
could be established for the variation of breakdown voltage with the edge terminations. The
higher effectiveness of the field plate termination than the guard ring termination has also been
reported with 4H-SiC Schottky diodes [15]. Further, in our study, it was observed that
phosphorous is a better choice than nitrogen for making donor implanted diodes in SiC. Not only
did phosphorous implanted diodes show higher breakdown voltages, but they also exhibited lower
leakage currents for the same voltage when compared to nitrogen implanted diodes. Leakage
currents were found to be lxlO"4, lxlO"3, 1x10"* and lxlO'7 Amperes for phosphorous, nitrogen,
aluminum and boron/aluminum implanted diodes, respectively, just before breakdown. The
aluminum implanted diodes consistently supported higher voltages than boron/aluminum
implanted diodes. The acceptor implanted diodes showed much lower leakage currents than the
donor implanted diodes. The leakage currents obtained for nitrogen implanted diodes on 4H-SiC
were found to be higher in this process when compared to the previous process. The high leakage
currents in the diodes was attributed to excessive leakage at the periphery of the junction due to
residual ion implantation damage [4,7] caused by the higher dose. Hot implantation is also known
to bring about inferior junction characteristics due to formation of dislocation loops [5].
High temperature (25 °C to 150 °C) characterization was also performed for all the diodes.
The variation of breakdown voltage with temperature is shown in Fig. 6. The breakdown voltage
was found to be almost invariant with temperature which is an encouraging result when compared
to previously reported negative temperature coefficient for the breakdown voltage [1]. The
leakage currents also remained in the same order of magnitude when the temperature was
increased from room temperature to 150 °C (-423 K), which is again highly desirable. We believe
that the leakage currents increased very slowly with temperature because the leakage current was
predominantly due to surface leakage components and not due to diffusion or space charge
generation components which are extremely low in SiC.
CONCLUSIONS
Electrical properties of P+-N as well as N*-? planar ion implanted junction diodes on 6Hand 4H-SiC with different edge terminations were studied. The single field plate edge termination
resulted in higher breakdown voltages than those with other planar edge terminations such as the
floating field rings. The nitrogen implanted diodes on 4H-SiC with field plates using Si02 as the
dielectric, exhibited a breakdown voltage of 1100 V, which is the highest ever reported measured
breakdown voltage for any planar ion implanted junction diode and is nearly 70% of the ideal
breakdown voltage. The reverse leakage current of this diode was less than lxlO"5 A/cm2 even at
breakdown. Diodes on 4H-SiC showed higher breakdown than those on 6H-SiC. Breakdown
85
~800
a, 700
4H-SJC
P • Al —
£ 600
a
I 500
N »-
j> 400 B + Al —
n
300
20
40
60
80
100
120
Temperature (degree C)
140
160
Fig. 6. Variation of breakdown voltages with temperature of field plate terminated 4H-SiC diodes using
Si^as dielectric.
voltages were independent of temperature in the range of 25 °C to 150 °C; while, the leakage
currents increased slowly with temperature, but stayed within the same order of magnitude.
These results are of relevance to design and fabrication of high voltage devices such as
MOSFETS containing reverse blocking junctions.
ACKNOWLEDGEMENTS
The authors would like to acknowledge the industrial sponsors of Power Semiconductor
Research Center for supporting this work.
REFERENCES
1. J.W. Palmour and L.A. Lipkin, Trans, of 2nd International High Temperature Electronic
Conf., 1, XI-3 (1994).
2. L.G. Matus, J. A. Powell, and C.S. Salupo, Appl. Phys. Lett., 59 (14) 1770 (1991).
3. P.G. Neudeck, D.J. Larkin, J. A. Powell, C.S. Salupo, and L.G. Matus, IEEE Trans. Electron
Devices, 41 (5), 826 (1994).
4. M.V. Rao, J. Gardner, P. Griffiths, O.W. Holland, G. Keiner, P.H. Chi, and D.S. Simons,
Nucl. Instr. and Meth. in Phys. Res., B (106), 333 (1995).
5. T.Kimoto, A. Itoh, H. Matsunami, T. Nakata, and M. Watanabe, J. of Electronic Materials,
24 (4), 235 (1995).
6. M. Ghezzo, D.M. Brown, E. Downey, and J. Kretchmer, Appl. Phys. Lett., 63 (9), 1206
(1993).
7. Y. Zheng, N. Ramungul, R. Patel, V. Khemka, and T.P. Chow, Proc. of International
Conference for SiC and Related Materials, MoP-06 (1997).
8. R. Patel, V. Khemka, N. Ramungul, T.P. Chow, M. Ghezzo, and J. Kretchmer, Proc. of
International Symposium on Power Semiconductor Devices & ICs, 122 (1998).
9. M. Bhatnagar, H. Nakanishi, S. Bothra, P.K. McLarty, and B.J. Baliga, Proc. of International
Symposium on Power Semiconductor Devices and ICs, 89 (1993).
10. D. Alok, B.J. Baliga, and P.K. McLarty, IEEE Electron Device Lett., 15, 394 (1994).
11. D. Alok and B.J. Baliga, IEEE Trans, on Electron Devices, 44 (6), 1013 (1997).
12. D. Alok, R. Raghunathan, and B.J. Baliga, IEEE Trans, on Electron Devices, 43 (8), 1315
(1996).
13. K. Rottner, Proc. of International Conference for SiC and Related Materials, 136 (1997).
14. B.J. Baliga in Power Semiconductor Devices, (PWS publishing company, 1995).
15. R. Singh and J.W. Palmour, Proc. of International Symposium on Power Semiconductor
Devices & ICs, 157(1997).
86
CARBON AND SILICON RELATED SURFACE COMPOUNDS OF PALLADIUM
ULTRATHIN FILMS ON SIC AFTER DIFFERENT ANNEALING TEMPERATURES
W.J. LU*, D.T. SHI, T. CRENSHAW, A. BURGER, WE. COLLINS
Department of Physics and the Center for Photonic Materials and Devices
Fisk University
Nashville, TN 37208
ABSTRACT
Pd/SiC Schottky diode has triggered interest as a chemical sensor to be operated at high
temperatures. Various surface compounds formed at high temperatures are known to alter the device
performance. In this work, the carbon and silicon related compounds and morphology ofPd ultra-thin
film on 6H-SiC and 4H-SiC are investigated after thermal annealing using X-ray photoelectron
spectroscopy (XPS) and atomic force microscopy (AFM). The Pd ultra-thin films of about 3 nm in
thickness are deposited by RF sputtering. The XPS analysis reveals the presence of silicon
oxycarbides (SiCxOy) as deposited. After being annealed above 300°C, the atomic ratio of C to 0 in
SiCxOy decreases with increasing the annealing temperatures, and the Pd film becomes a Pd silicide
nanofeatured layer on SiC. When the annealing temperature is at 500°C, the majority of the SiCxOy
is converted into Si02. An amorphous Si phase exists after annealing at 200 to 400°C, which indicates
that the Si-C bonds in SiC are broken at lower temperatures due to the presence of Pd. Graphite and
C=0 are found on the as deposited samples and also after annealing at temperatures up to 600°C. The
formations of the carbon and silicon related compounds on Pd/4H-SiC are very similar to those on
Pd/6H-SiC
INTRODUCTION
SiC-based device developments require that the metal contacts and interconnects are
physically, chemically, and electrically stable under severe conditions, such as at high temperatures.
Metal contact studies on SiC have resulted in commercially available SiC devices which can be
operated at high temperatures [1,2]. The diffusions and reactions between metal thin films and the
SiC substrate result in the formation of various interfacial compounds, and alter the electrical
properties. Therefore, it is one of the most critical issues to investigate the interfacial compositions
of metal/SiC at elevated temperatures in SiC based device research.
Pd/SiC Schottky diodes have been successfully demonstrated as a chemical sensor for
hydrogen and hydrocarbon [3,4], which can be operated at high temperatures. The heat treatment
significantly promotes interfacial diffusion and chemical reactions, and broadens the interface region.
The two-dimensional surface diffusion and surface segregation of Si from dissociated SiC result in
a thin silicon oxide layer on the top of the Pd film [3], The Pd chemical states are various co-existing
palladium silicides (P(LSi, x = 1,2,3,4) for the Pd thickness of about 400Ä [4] after annealing at
425°C. Lu et al. [5] found the Pd exists as PdSi and Pd2Si for the ultra-thin Pd film (-30Ä thickness)
at the elevated temperatures. To the best of our knowledge, the formations of the silicon and carbon
related compounds in Pd/SiC after annealing at high temperatures have not been systematically
87
Mat. Res. Soc. Symp. Proc. Vol. 572 ° 1999 Materials Research Society
investigated.
In this study, the interfacial composition and the morphological features of Pd ultra-thin films
on 6H-SiC and 4H-SiC at different annealing temperatures were investigated using XPS and AFM.
The Pd ultra-thin films were prepared by the RF sputtering method, and the thickness was about 30Ä.
The silicon and carbon related compounds in Pd/SiC were determined by XPS after annealing.
EXPERIMENTAL
1. Samples and Pd Ultra-thin Film Fabrication
n-type, Si-face 6H-SiC and 4H-SiC wafers with 3.5° off-axis on Si (0001) substrates were
purchased from Cree Research Inc. The doping concentration was 2.6 x 1018 cm3. An RF sputtering
system (Kurt J. Lesker Company) was used for the Pd ultra-thin film preparation. The SiC wafer
cleaning procedure and the Pd ultra-thin preparation procedure are in Ref.[5]. The thickness of Pd
thin film on SiC was about 30 A, so that XPS can effectively detect the Pd/SiC interfacial
compositions.
The Pd/SiC samples were analyzed and annealed consecutively from 100°C to 600°C in 100°C
increments in air for 30 minutes each time. The Pd/6H-SiC and Pd/4H-SiC samples were prepared
and annealed at the same time.
2. Characterization
The XPS experiments were performed on a Kratos X-SAM 800 spectrometer with an energy
resolution of 0.1 eV. The base pressure was at 10'9 Torr. The Mg Ka (M^1253.6 eV) radiation was
used. The energy scale of the analyzer was calibrated by the Cu (2p3/2) XPS peak at 932.67 eV, and
the Au (4f7/2) XPS peak at 84.00 eV.
The exit angle of the photoelectron was 90° to the sample surface. The data were smoothed
and satellite peaks were subtracted before background subtraction and deconvolution were
performed. The background of the XPS was subtracted using the Sherley algorithm. The GaussianLorentzian method was applied to deconvolute the XPS.
ANanoScope E (Digital Instruments, Inc.) was used for AFM measurements. The standard
silicon nitride cantilever supplied by Digital Instruments, Inc. was employed. The force constant of
the silicon nitride cantilever is 0.12 N/m. At least three different regions on the surface were measured
by AFM for each sample.
RESULTS
1. Carbon related compounds in Pd/6H-SiC
XPS is used to examine the surface composition changes on Pd/6H-SiC and Pd/4H-SiC
samples from the as deposited and after each annealing step. The surface species formed on the
Pd/SiC surface are quite complex, and two kinds of surface compounds are co-existing on Pd/SiC
samples after annealing; (a) the Pd suicides and Pd element, (b) the silicon and carbon related
compounds. In this paper, we present the results for the silicon and carbon related compounds on 6HSiC and 4H-SiC substrates.
The studies on the Pd-related compounds on Pd/SiC samples after different annealing
temperatures has been presented at another report [5]. There are two reactions between Pd and SiC
in the range from room temperature to 600°C; (a) at 300°C, the Pd reacts to SiC to form Pd2Si on
both 6H-SiC and 4H-SiC substrates, and (b) Pd2Si reacts with SiC to form PdSi at 500°C for Pd/4H-
SiC, and at 600°C for Pd/6H-SiC.
Table I lists the XPS peaks of the C( 1 s) and Si(2p3/2) after deconvolution for Pd/6H-SiC and
Pd/4H-SiC as deposited and after each annealing step. Figure 1 shows the deconvolution of the C(l s)
XPS for Pd/6H-SiC as deposited and after consecutive annealing at 300°C and 600°C. As shown,
the strong peak at the binding energy of 282.9 eV represents the C(ls) XPS from the Si-C bonds in
SiC substrate [5-8]. The peak at 284.1 eV exists in the Pd/SiC samples as deposited, and it is assigned
to the carbon oxycarbide (SiCxOy)compounds [6,7]. The SiCxOy is formed during the Pd film
deposition. The 285.1 eV peak is from the C-C bond (graphite), which is commonly found in almost
any sample due to the contaminants. The broad peaks at 287.2-287.8 eV and -286.5 eV are assigned
to the C=0 and C-0 bonds, respectively [9]. Table II shows the chemical shifts in XPS of various
silicon and carbon compounds compared with the XPS for the Si-C bond in SiC.
I"
S"
zt
260
Binding Energy (eV)
(A)
2äa
lie
21U
2fc
Binding Energy (eV)
Binding Energy (eV)
(B)
(C)
Figure 1. Deconvolution of the C (Is) XPS for Pd/6H-SiC as deposited (A) and
after consecutive annealing at 300°C (B) and at 600°C (C).
TABLE I. Deconvolution of the XPS peaks of the C (Is) and Si (2p3/2) on Pd/SiC
as deposited and after consecutive annealing in air for 30 minutes.
Samples
C (Is) peaks
Si (2p3/2) peaks
as deposited
annealed at 100°C
annealed at 200°C
annealed at 300°C
annealed at 400°C
annealed at 500°C
annealed at 600°C
282.9,284.1,285.1,287.2
282.9,284.1,285.1,287.3
282.9,284.1,285.2,287.4
282.9, 284.1, 285.2, 286.5, 287.8
282.9,284.1,285.2,287.8
282.9,284.2,285.1
282.9,285.1,287.6
100.8, 102.3, 103.6
100.8,102.3,
99.0, 100.8, 102.3
99.0, 100.8, 102.3
99.0, 100.8, 102.6
100.8, 103.2,103.6
100.8, 103.6
The concentrations of the carbon related compounds on Pd/SiC change with increasing
annealing temperatures. The C(ls) XPS peak of the C-C bond also increases with increasing
annealing temperatures. The surface graphite comes not only from the contaminants, but results from
the products of the reactions of the Pd and the SiC substrate [5]. The SiCxOy and CO groups on the
89
surface decrease with increasing the annealing temperatures. A small amount ofthe C-0 group is also
found after annealing at 300°C. At the annealing temperature of 600°C, the main carbon compound
is graphite although small amounts of SiCxOy and C=0 still exist on the surface. The C=0 group is
also found on the substrate at room temperature, and the concentration of the C=0 is low through
the annealing temperature up to 600°C. The C=0 group is difficult to remove, and the thermal
cleaning of the SiC substrates usually requires at least 800°C to obtain the oxygen-free SiC surface
under ultrahigh vacuum (UHV) environment [10]. It seems that the oxygen in air directly attacks the
surface C of the SiC substrate to form C=0.
TABLE II. Chemical shifts in XPS of various silicon and carbon compounds compared
with the Si-C bond in SiC.
Species
C (Is) in SiC (282.9 eV)
c-c
2.0 [6,7], 1.6 [9]
Si-Si
Si-Pd
SiCxOy
Si02
C-0
C=0
Si (2p3/2) in SiC (100.8 eV)
-1.8 [11]
0[4]
0.5 - 2.4 [6,7]
>2.4[6,7],>1.8[9],2.5[8]
1.2 [6,7]
3.6 [9]
4.9 [9]
2. Silicon related compounds in Pd/6H-SiC
Figure 2 shows the deconvolution of the Si (2p3/2) XPS for Pd/6H-SiC after consecutive
annealing at 300°C, and 600°C. As shown, the strong peak at 100.8 eV is assigned to the Si (2p3/2)
XPS from the SiC substrate and the Pd suicides [4]. The chemical shifts of the Si (2p3/2) peak from
the silicon oxycarbides (SiCxOy) vary with the stoichiometry of C and O [6,7]. The binding energy
of SiCxOy is 102.3 eV for the sample as deposited. It increases to 102.7 eV after consecutive
annealing at 400°C, and 103.2 eV after annealing at 500°C. After being annealed at 600°C, the
majority of the silicon compounds on Pd76H-SiC is Si02 instead of silicon oxycarbides.
ifa
ito '
ite
lfti
iti
Binding Energy (eV)
Binding Energy (eV)
(A)
(B)
(C)
Figure 2. Deconvolution of the Si ^p^) XPS for Pd/6H-SiC as deposited (A)
and after consecutive annealing at 300°C (B) and at 600°C (C).
90
After annealing at 200°C, a small XPS peak at 99.0 eV appears. This peak is assigned to Si-Si
bond [11], and diminishes after annealing at 400°C. As in our previous work [5], Pd starts to react
with SiC to form PdjSi and C at 300°C. The presence of Si indicates that the Si-C bonds in SiC break
in to C and Si phases on Pd/SiC at 200°C. Obviously, the existing of Pd seems to act as a catalyst to
break the Si-C bonds. At higher temperatures, Si is consumed by the reactions with Pd suicides [5].
Table m summarizes the various carbon and silicon compounds on Pd/SiC after consecutive
annealing from 100 to 600°C.
TABLE m. Carbon and silicon related compounds on Pd/SiC as deposited and
after consecutive annealing in air for 30 minutes.
Carbon and silicon related compounds
Annealing Temperatures (°C)
C, SiCxOy, c=o
as deposited
C, SiCxOy, C=0
100°C
Si, C, SiCxOy, CO
200°C
Si, C, SiCxOy*, Pd2Si, CO, C-0
300°C
Si, C, SiCxOy*, Pd2Si, CO
400°C
C, SiCxOy*, Pd2Si, CO, Si02
500°C
C, SiO„ PdSi, CO
600°C
* the atomic ratio of C to O decreases with increasing annealing temperatures.
Figure 3 shows the morphological changes forPd/6H-SiC as deposited and after consecutive
annealing at 400CC and 600°C using AFM. As shown, the surface of the Pd/SiC sample as deposited
has a good uniformity (Figure 3A). After annealing at 300°C, the Pd film becomes a nanofeatured
layer. The average size of the nanofeatures is about 20-30 nm (Figure 3B). With increasing the
annealing temperature to 600°C, the nanofeatured layer has been broken into nanosize rounded
clusters on 6H-SiC surface (Figure 3C). The changes in the morphological features on the Pd/SiC
surface are related to the surface compositions. At the annealing temperature of 300°C, Pd2Si is
formed and the nanofeatured layer appears [5], Meanwhile, the atomic ratio of C to O in the SiCxOy
decreases. When the annealing temperature is 500 to 600°C, the majority of the SiCxOy is converted
into Si02, and Pd2Si reacts with SiC to form PdSi [5].
(A)
(B)
(C)
Figure 3. AFM images for Pd/6H-SiC as deposited (A, (1 x 1 urn2 x 5 nm)), after consecutive
annealing at 300°C (B, (1 x 1 um2x 10 nm)), and 600°C (C, (1 xlnm2x40 nm)).
91
3. Comparison of Pd/6H-SiC and Pd/4H-SiC
6H-SiC and 4H-SiC have very similar structures and reactivity with Pd at high temperatures.
The formations of Pd silicides on Pd/4H-SiC is slightly enhanced compared to those on Pd/6H-SiC
[5]. However, the formations of the carbon and silicon related compounds on 6H-SiC and 4H-SiC
substrates are almost identical.
CONCLUSIONS
This study reports that various carbon and silicon compounds are formed with the presence
of Pd ultra-thin film on SiC surface after different annealing temperatures. Silicon oxycarbides
(SiCxOy) are found and the atomic ratio of C to O decreases with increasing annealing temperatures.
Si02 is formed after annealing at 500°C and above. The amorphous Si is formed after annealing in
the temperature region of 200-400°C, and Pd seems to act as a catalyst to break the Si-C bonds in
SiC into Si and C phases. A low concentration of the C=0 group exists on Pd/SiC through the
annealing temperature up to 600°C. We found no significant differences in the formation of carbon
and silicon compounds between the Pd/6H-SiC and Pd/4H-SiC samples.
ACKNOWLEDGMENTS
The work was supported by NASA grant Nos NAG3-2126 and NCC8-133. The authors are
thankful for the discussion with Dr. Gary W. Hunter at NASA Lewis Research Center.
REFERENCES
[I]
V. Saxena, and AJ. Steckl, "SiC Materials and Devices", Semiconductors and Semimetals,
Vol. 52, edited by Y.S. Park, Academic Press, San Diego, CA, USA 1998, Chapter 3, p77.
[2] L.M. Porter, and R.F. Davis, Mater. Sei. and Engin., B34, 83 (1995).
[3] L.-Y. Chen, G.W. Hunter, P.G. Neudeck, G. Bansal, J. B. Petit, and D. Knight, J. Vac. Sei.
Technol. A 15(3), 1228 (1997).
[4] L.-Y. Chen, G.W. Hunter, P.G. Neudeck, and D. Knight, J. Vac. Sei. Technol. A 16(5), 2890
(1998).
[5] W.J. Lu, D.T. Shi, A. Burger, and W.E. Collins, accepted by J. Vac. Sei. Technol, A for
publication in 1999.
[6] C. Önneby, and CG Pantano, J. Vac. Sei. Technol, A, 15(3), 1597 (1997).
[7] C. Önneby, and CG Pantano, J. Vac. Sei. Technol, A, 16(4), 2742 (1998).
[8] S. Contarini, S.P. Howlett, C. Rizzo, and B. A. De Angelis, Applied Surf. Sei., 51, 177(1991).
[9] B. Hornetz, H.-J. Michel, and J. Halbritter, J. Vac. Sei. Technol, A, 13(3), 767 (1995).
[10] S. Tanaka, R. S. Kern, R.F. Davis, J. F. Wendelken, and J. Xu, Surf. Sei., 350, 247 (1996).
[II] J. R. Shallenberger, J. Vac. Sei. Technol., A, 14(3), 693 (1996).
92
A MATERIALS INVESTIGATION OF NICKEL BASED CONTACTS TO n-SiC
SUBJECTED TO OPERATIONAL THERMAL STRESSES CHARACTERISTIC
OF HIGH POWER SWITCHING
M.W. COLE*, C.W. HUBBARD*, CG. FOUNTZOULAS*, D.J. DEMAREE*, F. REN**
* US Army Research Laboratory, WMRD, Aberdeen Proving Ground, MD 21005
** Dcpt. of Chem. Eng., University of FL, Gainesville, FL 21213
ABSTRACT
This study developed and performed Laboratory experiments which mimic the acute cyclic
thermal loading characteristic of pulsed power device switching operation. Ni contacts to n-SiC
were the device components selected for cyclic thermal testing. Modifications of the contactSiC materials properties in response to cyclic thermal fatigue were quantitatively assessed via
Rutherford backscattering spectrometry (RBS), scanning electron microscopy (SEM), atomic
force microscopy (AFM), surface profilometry, transmission electron microscopy (TEM),
nanoindentation testing and current-voltage measurements. Decreases in nanohardness and
elastic modulus were observed in response to thermal fatigue. No compositional modifications
were observed at the metal-semiconductor interface. Our results demonstrated that the majority
of the material changes were initiated after the first thermal pulse and that the effects of
subsequent thermal cycling (up to 10 pulses) were negligible. The stability of the metalsemiconductor interface after exposure to repeated pulsed thermal cycling lends support for the
utilization of Ni as a contact metallization for pulsed power switching applications.
INTRODUCTION
Much attention has focused on SiC as a material for high power, high temperature and high
radiation tolerance device applications!; 1-3] It is the exceptional properties of SiC, such as
high breakdown field, large bandgap, high thermal conductivity and high electron saturation
velocity, which are responsible for these device application interests [1]. To date, SiC
electronic materials research efforts have focused predominantly on growth, processing science
and packaging issues. However, in order to promote, design and realize reliable SiC power
devices it is important to assess the performance of device components under the influence of
their potential operational stress regimes. This is particularly critical for pulsed power device
applications, namely palpitated high power switching, where the operational environment is
dominated by acute cyclic pulsed power actions which ultimately translate into severe thermal,
electrical and mechanical cyclic stresses in the device materials. In order to fully explore SiC's
utilization for pulsed power switching applications it is necessary to determine the effects of
such cyclic stress regimes, both individually and as combined effects, on the fundamental
pulsed power device components. In this paper we report results of a unidimensional, that is,
non-combined effects, investigation which evaluated the reliability of Ni-SiC ohmic contact
device components in response to acute cyclic thermal loading. Our results demonstrate that the
electrical, compositional and structural integrity of the metal-SiC interface strongly influences
the reliability of the Ni-SiC ohmic contacts under acute cyclic thermal stress.
It is well documented that device performance is often limited by the electrical and materials
integrity of the ohmic contacts [4,5]. Since ohmic contacts are a fundamental component of all
pulsed power devices the ohmic contact-SiC device structure was selected for cyclic thermal
testing. A number of different metals have been proposed as suitable ohmic contacts to n-SiC.
Specifically, metals such as Ni, Al/Ni/Al, Cr, Al, Au-Ta, TaSi2, W, Ta, Ti, Ti/Au, TiSi2, Co
and WSi have been studied with the Ni based metallization systems suggested as superior
candidates due to their low specific contact resistance, (pc), less than 5.0x10" ohms-cm [6-9].
Additionally, published data suggests that annealed Ni contacts, which react^with SiC to form
Ni2Si, exhibit excellent static thermal stability at temperatures as high as 500°C [7]. Based on
this information Ni was selected as the contact metallization for cyclic thermal fatigue testing.
93
Mat. Res. Soc. Symp. Proc. Vol. 572 ®1999 Materials Research Society
EXPERIMENTAL
200 nm of Ni was deposited on 4H n-type SiC substrates purchased from CREE Research.
The SiC substrates were non-research grade with a micropipe density of greater than 100 cm"2.
The substrates were Si faced and the donor density was 2.0 x 1019 cm"3. Prior to the metal
deposition the wafers were cleaned in warm electronic grade trichoroethane (TCA), boiling
acetone and methanol followed by a rinse in deionized water. The Ni deposition was
accomplished via electron beam evaporation with a base pressure of 5x10~7 Torr. The Ni on
SiC samples, were annealed at 950°C for 5 min. in a N2 ambient in order to produce ohmic
behavior. Cyclic thermal fatigue experiments were conducted using a 10.6 urn IR pulsed C02
laser. The pulsed thermal fatigue design was configured for a 3 second heating interval
followed by a 60 second cooling interval. The heating and cooling intervals were chosen to
mimic military switching requirements. Laser power levels were tailored to maintain a
temperature of 650° C for 1 cycle and 10 consecutive cycles. Temperature verification was
obtained with the aid of both a thermocouple and pyrometer.
Current-voltage measurements, were performed on the as-deposited, annealed and thermally
fatigued Ni-SiC samples. The electrical measurements were internally consistent and were used
solely to assess the electrical changes resulting from thermal fatigue relative to the non-fatigued
sample. All samples were analyzed for surface morphology modification via atomic force
microscopy (AFM) and surface profilometry. A Topometrix Discoverer Scanning Probe system
was used to obtain the AFM data and the profilometry measurements were obtained via a
Tencore P-2 low scan profiler. Elemental diffusion and phase formation were tracked using
Rutherford backscattering spectroscopy (RBS). RBS measurements were obtained on a
National Electronics Corp. 55DH-2 Accelerator using 2 MeV He+ ions with a scattering angle
of 170° and a solid angle of 5.5 msr. Experiments were performed at a tilt angle of 10° away
from the detector in IBM geometry (beam, surface normal and detected beam are coplanar).
Simulations were produced using the computer code RUMP [10]. The structural durability of
the thin film metallization on SiC was evaluated via nanoindentation testing. The
nanoindentation measurements were made using a Nano Instruments XP nanoindenter, with a
Berkovich 3-sided pyramid diamond indenter with controlled penetration depths of 300 nm.
The instrument allowed a indenter penetration vs. force curve to be determined allowing the
determination of both nanohardness and the effective elastic modulus from the slope of the
hysteresis curve. Microstructural analyses were obtained via cross-sectional TEM using a JEOL
3010 STEM operated at 300 keV.
RESULTS AND DISCUSSION
The I-V characteristics of the as-deposited Ni films displayed rectifying behavior suggestive
of a large barrier height, typically 1 eV or greater. The RBS analysis (figure 1) revealed no
reaction between the Ni and SiC substrate and showed no evidence of an interfacial oxide upon
Ni deposition. The I-V curve for the annealed e-beam deposited Ni on SiC possessed nearlinear characteristics and was symmetric with reversal of the voltage polarity. Due to lack of dry
etching facilities, no values for specific contact resistance were measured, thus a quantitative
value for specific contact resistance is not available. However, the Ni-SiC annealing protocol
used in this study is well documented in the literature to produce good ohmic behavior with
reproducible results,[7-9] thus, it is not unreasonable to assume that our samples also possess
good ohmic behavior, that is, low specific contact resistance . The fact that the I-V curve for the
annealed Ni-SiC structure possessed linear characteristics and was symmetric with reversal of
the voltage polarity, is indicative of ohmic contact formation [11]. The RBS analysis of the ebeam deposited Ni on SiC (figure 2) showed a thin layer of NiO present at the contacts surface
and Ni suicide formation adjacent to the SiC. The NiO is most likely due to the presence of
94
oxygen during the annealing cycle. However, the near surface nature of the NiO layer combined
with the linear I-V characteristics suggests-that the oxygen contamination was not severe enough
to impede ohmic contact formation. The RBS depth profile (figure 2) and TEM analyses (not
shown) revealed a non-abrupt contact-SiC interface in which the thickness of the contact
increased from 200 nm to 400 nm as a result of interfacial reactions during the anneal. This
large increase in contact thickness has been documented in the literature for Ni contacts on SiC
[6, 8, 12]. For Ni contacts to SiC, formation of Ni silicides upon annealing appears to be a
requirement for ohmic behavior. The critical step in suicide formation requires the continual
supply of Si atoms through breaking bonds in the substrate. The Si-C bonds can be broken in
several ways; sufficient thermal energy to break the SiC bonds (high temperatures) and/or
rapid interstitial migration of the metal through the SiC lattice which assists bond breaking and
aids suicide formation. Since our samples were annealed at a fairly high temperature (950°C)
and the fact that Ni is very small (.69 A ionic radius) compared to that of Si (2.71 A) and C
(2.60) suggests that both these mechanisms may have contributed to the formation of the silicide
phase detected by RBS.
Ni
SiC substrate
ft. -4
Ni silicide
NiO
<u
rV^*M
y
SiC substrate
'
.-
^
Silicon
B
o
.,
<
Mlelwl
JV—
S
Depth (nm)
Depth (hm)
Fig. 2. RBS depth profile for annealed Ni
contact to SiC.
Fig. 1. RBS depth profile for the e-beam
as-deposited Ni on SiC.
The I-V curve for the 10 cycle thermal fatigued contact on SiC was notably similar to that of
the unfatigued sample, that is, linear, symmetric with reversal of voltage polarity and possessed
no deviation in the slope of the I-V curve relative to that of the unfatigued sample. The fact that
ten cycles of thermal fatigue did not significantly degrade the I-V curve speaks well for the
electrical integrity of this contact metallization in response to pulsed thermal stress. Figure 3
displays the RBS depth profile for this sample. The results revealed no compositional changes
at the metal semiconductor interface. However, oxygen appears to have penetrated a bit deeper
into the sample. It is speculated that attenuated nanofractures, resultant from thermal shock,
confined to the upper portion of the contact may be responsible for this «diffusion of oxygen
[13]. It must be kept in mind that the interfacial properties of the metal-semiconductor contact
strongly influences electrical performance [4,5]. Thus, the fact that the metal-semiconductor
interfacial region remained compositionally and electrically unchanged lends support to the
excellent reliability of this contact in response to acute pulsed thermal stress.
95
120
NiO
Ni suicide
SiC substrate
6H
as-deposited annealed
1 cycle TF
10 cycles TF
Sample Treatment
Depth (nm)
Fig. 4. Surface roughness as a function
of sample treatment. Atomic force
microscopy (AFM) & profilometry
(P) data parallel one another.
Fig.3. RBS depth profile for the annealed
Ni contact to SiC after 10 cycles of
thermal fatigue.
Optical and SEM analysis showed the surfaces of the as-deposited e-beam evaporated Ni
thin film on SiC to be smooth and mirror like in appearance. However, after annealing at
950°C the surface changed drastically. The mirror-like metallic luster changed to a lusterless
dull gray color and the film was no longer smooth. The surface morphological changes were
quantified via profilometry and AFM analyses, and are displayed in figure 4. The magnitude of
the profilometry data is larger than that of the AFM data but the general trends parallel one
another. It is evident from figure 4 that annealing augmented the surface roughness of the metal
film. Specifically, the surface roughness increased by an order of magnitude. The films
surface remained lusterless after the thermal fatigue however the surface became smoother.
This surface smoothening occurred during the first cycle of the pulsed thermal fatigue and
maintained at steady state through the 10th cycle, that is, the rms roughness value remained
constant throughout the duration of the thermal cycling. We suggest that the initial thermal
shock of the first laser pulse caused the removal of a thin layer of loosely bound particles from
the films surface which resulted in a lower rms roughness value. The fact that the film did not
delaminate and the rms roughness value remained constant throughout the thermal cycling
bodes well for the films strong adhesion and cohesion properties.
Mechanical properties such as nanohardness and Young's modulus were used to assess the
changes in the physical durability of the metal-SiC component in response to thermal fatigue.
Simply defined the hardness of a thin film is the resistance of the film to penetration of its
surface, that is, resistance to local plastic deformation [14]. Hardness is a complex macroscopic
property related to the strength of interatomic forces and depends on several variables at the
nanoscopic level. Grain size, area and grain boundary structure, film composition (new phase
formation), impurities, defects and film texture all influence and/or control the nanohardness or
durability of thin films [15]. Thus, there is a strong relationship between a films microstructure
and its mechanical properties.
The magnitude of Young's (elastic) modulus is determined by the strength of the atomic
bonds in the film. The stronger the atomic bonding, the greater the stress required to increase
the interatomic spacing and thus the larger the value of the modulus of elasticity and the more
durable the film [14]. Like hardness, the modulus is a macroscopic property which depends on
many different variables at the nanoscopic-atomistic level. Specifically, modification of the
films chemical composition strongly influences the elastic modulus value since the various
phases are composed of different atomic species with different bond strengths. Thus, changes
96
in the contacts mircrostructure and composition will be reflected in the values of the contacts
nanohardness and elastic modulus.
Figure 5 displays the nanohardness and Young's modulus values for the as-deposited,
annealed and thermal fatigued contact metallization on SiC. The nanohardness and modulus
values changed significantly with sample treatment, showing a decrease in magnitude in
response to annealing and cyclic thermal fatigue. The largest change in both the hardness and
modulus occurs after the first thermal pulse. Additional thermal cycling caused negligible
changes in these properties. The compositional change from Ni to Ni-siHcide (and the thin
surfacial NiO layer) accounts for the change in the durability values in response to annealing.
TEM results on Ni-SiC ohmic contacts has shown the grain size of the Ni-Silicide to be smaller
that that of the as-deposited Ni on SiC [13]. Thus, the diminishment of the nanohardness after
annealing is easily explained by the increase of grain boundary area of the Ni-silicide with
respect to that of Ni. The decrease in durability values in response to the first thermal pulse is
most likely due to the promotion of thermally induced nanofractures (figure 6) within the upper
portion of the Ni-Silicide contact layer [13]. The fact that the hardness and modulus do not
change significantly after 10 cycles suggests that multiple or repeated thermal cycling has a
negligible affect on the films durability. This speaks well for the reliability of the Ni contacts
endurance to repeated pulsed thermal fatigue. In addition, the fact that the metal film did not
delaminate in response to the cyclic thermal stress bodes well for the films strong adhesion and
cohesion properties.
Sample Treatment
Fig. 5. Nanohardness and Young's modulus values as a function of sample treatment.
(a) as deposited
(b) annealed 950° C
(c) 10 cycles thermal fatigue
Fig. 6. Schematic representation of the microstructural changes as function of sample treatment.
The dark black areas in the annealed sample represent voids (voids were determined via
X-TEM) and the dark lines at the surface of the thermally fatigued sample represent
nanofractures.
97
CONCLUSIONS
We have developed and performed laboratory experiments which simulate the acute thermal
cyclic fatigue incured as a result of pulsed power switching operation. Evaluation of the
reliability of Ni contacts to SiC in response to cyclic thermal fatigue was investigated. Our
results demonstrated that most of the material changes occurred in response to the first thermal
pulse and that further pulsing (up to 10 pulses) inflected negligible changes in the contact-SiC
durability, compositional and electrical properties. The stability of the metal-semiconductor
interface after acute thermal cyclic fatigue lends support for for the utilization of Ni as a contact
metallization for pulsed power applications.
REFERENCES
1. S.J. Pearton, F. Ren, RJ. Shul and J.C. Zolper, Proceedings of The Electrochemical
Society, 97-1, 138 (1997).
2. Philip G. Neudeck, J. of Electronic Materials, 24, 283 (1995).
3. J.W. Palmour and C.H. Carter, Proceedings of 1993 International Semiconductor Device
Research Symposium, 695 (1993).
4. W.Y. Han, Y. Lu, H.S. Lee, M.W. Cole, L.M. Casas, A. DeAnni and K.A. Jones, J.
Appl. Phys. 74, 754 (1993).
5. M.W. Cole, W.Y. Han, L.M. Casas, D.W. Eckart and K.A. Jones, J. Vac. Soc. Technol.
A 12, 1904 (1994).
6. C. Hallin, R. Yakimova, B. Pecz, A. Georgieva,T.S. Marinova, L. Kasamakova, R.
Kakanakov, E.Janzen, J. of Electronic Materials, 26, 119 (1997).
7. J. Crofton, P.G. McMullin, J.R. Williams, M.J. Bozack, J. Appl. Phys., 77, 1317
(1995).
8. J. Crofton, L.Beyer, T. Hogue, R.R. Siergiej, S. Mani, J.B. Casady, T.N. Oder, J.R.
Williams, E.D. Luckowski, T. Isaacs-Smith, V.R. Iyer and S.E. Mohney, Proceedings of
The Fourth International High Temperature Electronics Conference, 84 (1998).
9. M.R. Melloch and J.A. Cooper, MRS Bulletin, 23, 42 (1997).
10. L.R. Doolittle, Nucl. Instrum. Methods B9, 344 (1985).
11. S.M. Sez, Semiconductor Devices Physics and Technology, (John Wiley & Sons, New
York, (1985).
12. M.I. Chaudhry, W.B. Berry and M.V. Zeller, Mat. Res. Soc. Proa, 162, 507 (1990).
13. M.W. Cole, C. Hubbard, D. Demaree, CG. Fountzoulas, D. Harris, A. Natarajan, P.
Searson, R.A. Miller and D. Zhu, Proceedings of the Army Science Conference, 22, 35
(1998).
14. W.D. Nix, Metallurgical Transactions A, 20A, 2217 (1989).
15. J.E. Sundgren and H.T. G. Hentzell, J. Vac Sei. Tech. A, 4, 2259 (1986).
PREPARATION OF CONDUCTIVE TUNGSTEN CARBIDE LAYERS
FOR SIC HIGH TEMPERATURE APPLICATIONS
H. ROMANUS*, V. CIMALLA*, S.I. AHMED*, JA. SCHAEFER*, G. ECKE**, R. AVCf,
L. SPIESS***
Institute of Physics, Technical University of Ilmenau, Germany, D-98693 Ilmenau
Institute of Solid State Electronics, Technical University of Ilmenau, Germany
Institute of Materials Engineering, Technical University of Ilmenau, Germany
+
Department of Physics, Montana State University, Bozeman, Montana 59717
ABSTRACT
Thin tungsten carbide films of different compositions were prepared by DC magnetron
sputtering of tungsten and carbon and subsequent annealing in different environments. The
onset of carbide formation was around 800°C. Annealing in a pure hydrogen ambient generally
results in carbon depletion in the layers with the formation of a dominant W2C phase. Adding
propane enhances the carbon content in the layers and stimulates the formation of the WC
phase. On silicon nitride substrates, variation of the propane concentration in an annealing
environment allows a continuous alteration of the layer structure between polycrystalline single
phase WC and a mixed layer with dominant W2C and with it, the adjustment of different values
of the electrical resistance. In contrast, on thin (100)SiC layers a textured W2C phase was
grown after annealing in propane/hydrogen at 900°C whereas at higher temperatures the
formation of suicides was observed. In addition, the chemical composition and the temperature
dependence of the electrical specific resistance were investigated and are also discussed.
INTRODUCTION
In recent years, increasing interest has been directed to wide band-gap semiconductors like
silicon carbide (SiC) due to its potential for applications in power and high temperature
electronics [1]. SiC devices show very good high temperature, chemical and mechanical
stability and high breakdown voltages. Recent improvements in crystal growth techniques now
provide industry and research with high quality bulk and epilayer material. However, there are
still a variety of factors limiting the commercialization of devices in this area [2]. One of the
most important factors with respect to high temperature devices is the requirement of a
metallization, which should maintain a low contact resistivity, have good adhesion to the
substrate material and have high stability at elevated temperatures, particularly at the interface.
Several attempts have been made to achieve good Ohmic contacts to SiC. However, at elevated
temperatures most metals are not stable on SiC and form rather good rectifying contacts with a
barrier height > 1 eV [3].
The formation of good stable Ohmic contacts to p-type SiC remains an important technological
problem that is hindering the industrial production of high temperature and power devices. A
theoretical study based on an ideal band bending model of a semiconductor-metal interface has
shown the impossibility of forming enhancement contacts to p-type SiC [4], because no contact
material exists with the required work function. In all cases, to form Ohmic contacts the width
of the depletion zone has to be minimized to increase thermionic emission. This can be attained
99
Mat. Res. Soc. Symp. Proc. Vol. 572 ®1999 Materials Research Society
by a high doping of the surface region. Nevertheless, a high value of the contact material work
function is a precondition for the formation of Ohmic contacts.
Metal carbides are promising contact materials that could fulfill the above mentioned
requirements. Transition-metal carbides like tungsten carbide possess an unusual combination
of physical properties: They are hard and very refractory as many ceramics and have good
electrical conductivity that is comparable to the parent transition metals [5]. The extreme
stability and hardness are already being exploited in many industrial applications like wearresistant coatings, cutting and drilling tools. However, the electrical properties make tungsten
carbide also an attractive material for contacts in high temperature electronics or as conductive
protective layers in sensor applications. Tungsten carbide exists in different phases, most
important are the WC and W2C phases. The close match of the lattice constants of the
hexagonal phase to the (OOOl)SiC plane and the high work function of 3.6 eV and 4.58 eV,
respectively, [5] favors them in applications as highly stable epitaxial contacts to SiC.
Several attempts have been made to prepare thin tungsten carbide layers on different substrates
including sputtering, chemical vapor deposition (CVD), solid-phase reaction, laser ablation and
ion beam synthesis. In most cases a mixed phase or amorphous tungsten carbide was grown,
mainly for hard and wear resistant coatings. The only techniques that have successfully
prepared single phase WC layers are the solid state reaction of tungsten on diamond [6] and
CVD on tantalum substrates [7] at high temperatures. However, both methods are strongly
influenced by the substrate material and CVD was shown to be inappropriate for SiC substrates.
Up to now neither have the contact properties of thin tungsten carbide layers to SiC been
investigated nor has epitaxial growth been possible.
EXPERIMENTAL
The tungsten-carbon layers were deposited by DC magnetron sputtering of a sintered
stoichiometric WC target in argon at a pressure of 3x10~2 Torr at room temperature or in situ
annealing at 500°C, respectively. The substrates were silicon wafers covered by 250 nm silicon
nitride or by thin cubic (100)SiC formed by carbonization [8]. Prior to sputter deposition they
were ultrasonically cleaned in methanol. The tungsten-carbon layer thickness was varied
between 50 and 300 nm. Annealing experiments were performed in situ or ex situ in a
conventional horizontal quartz tube in argon for 20 min or in a rapid thermal processing (RTP)
equipment [9] between 10 and 60 s in argon/ hydrogen/ propane mixtures with a propane
content that was varied between 0 and 5%. The layer structure was analyzed by X-ray
diffraction (XRD) with CuK„ radiation, the surface morphology by atomic force microscopy
(AFM), the electrical conductivity by a linear four-point method at room temperature and at
elevated temperatures up to 470°C and the chemical composition by Auger electron
spectroscopy (AES) and X-ray photoelectron spectroscopy (XPS).
RESULTS
The as-deposited layers were amorphous or micro crystalline with a homogeneous, smooth
surface without any distinct features. A compositional analysis revealed a carbon depletion
around 10 %. Annealing in situ resulted in micro crystalline thin films. Ex situ annealing in
pure hydrogen by RTP lead to strong carbon depletion via reaction to hydrocarbons, and the
formation of the W2C and other high temperature phases (Fig. 1). This reaction can be
100
suppressed by adding propane to the ambient. With an enhancement of the propane
concentration up to 0.025% on the silicon nitride covered substrates, the content of the W2C
phase decreased while content of the WC phase increased (Fig. 2). Only WC was detected by
XRD above 0.025% at 1200°C and in the AFM images small grains appeared at the surface. In
contrary, annealing of the tungsten carbon layers on the (100)SiC covered substrates at 900°C
under the same conditions enhances the formation of the W2C phase. Only one tungsten carbide
diffraction peak was detected indicating the formation of a texture. At higher temperatures (i.e.
1200°C in Fig. 3) tungsten suicide was additionally observed. This phase is probably formed by
the reaction of tungsten with silicon from the underlying substrate which diffuses through the
thin SiC layer.
30
40
2-Theta Scale
50
Fig. 1 XRD patterns, RTP annealing in pure hydrogen between 1000CC and 1350CC
40
2-Theta Scale
50
Fig. 2 XRD patterns, RTP annealing at 1200°C, of 0.01% to 0.05% propane in hydrogen
101
* &
»...up.ax.,0J3..
SiC substrate
12pO°C,0.1%
|:^vjL^V
,*A$i!t! substrate
gopoc,r%
SiC substrate
«M^^l^A^^w»«^^^*v* lV^-^^^^ ^fcl^^*w l^^**'^
-u^^JLwr^ [VA^YS—T
40
2-Theta Scale
Fig. 3 XRD patterns, RTP annealing S.3N4 and SiC substrates
90P°C,1%
SijN, substrate
30
50
The chemical analysis of the thin films was accomplished by AES depth profiling using a
sputter rate of around 1 to 2 nm/min. For the as-deposited films as well as for the RTP (Fig. 4)
and in situ annealed samples the tungsten and carbon amount is constant over the entire
thickness, independent of the annealing time (Table I). A low oxygen content of approximately
1% to 3 % was detected. A higher concentration was found only after annealing in the quartz
tube where already low traces of oxygen or water are sufficient to oxydize the layers. Similar
results were found by XPS depth profiling. XPS spectra indicate the formation of carbides. The
binding energy was estimated to be 32.1 eV and 31.36 eV for the WC and W2C, respectively,
where the latter is located very close to the value for elemental tungsten (31.32 eV [10]). The
calculated content of the two carbidic phases is in good agreement with XRD observations.
IUU
80
>
1
•
i
T
'—
.
•
-
I
|
1
W(179eV)
C (275 eV)
0 (383 eV)
•
|,0 1...
".
."
.".
20
....
.
•
""
"..
44.3%
H 40
Ü
C
O
-
54.4 %
.
.'
charging
(
J1
\1 :
50
100
150
200
250
sputter time [min]
Fig. 4 AES depth profiling, RTP at 1200°C and 0.02% propane (The changing concentrations
near the interface are due to charging effects at the isolating substrate.)
102
Table I Comparison between XRD, AES and XPS
Sample
XRD
AES
W
C
in situ annealed micro crystalline 53.3% 43.3%
Annealed in
mixture of
40..77% 2..24%
quartz tube
W/W2C and a
low content WC
RTA, 1200°C,
mixture of WC
54.4% 44.3%
0.02%, 60s
and W2C
RTA, 1200°C,
mixture of WC
55.2% 43.2%
0.02%, 10s
andW2C
RTA, 1200°C,
W2C and a low
65.5% 32.8%
0.05%, 60s
content WC
XPS
O
<3.2%
59..5%
1.35%
approximately 10% WC,
10% W2C, 80% W, oxide
only near surface
approximately 40%WC,
60% W2C
1.58%
1.66%
approximately 10% WC
and 90% W2C
At room temperature the as-deposited tungsten-carbon layers had a specific resistance of around
200 ulicm. With increasing content of the W2C phase at lower propane concentrations the
specific resistance decreased continuously to 100 nDcm (Fig. 5). As a result of the temperature
dependent measurements the as-deposited and in situ annealed layers revealed a negative
temperature coefficient which is well known for amorphous or microcrystalline metal films.
The specific resistivity of the crystalline tungsten carbide layers showed a temperature behavior
depending on their composition (Fig. 6): a linear increase with the temperature for pure WC
layers similar to pure metals and a non linear behavior in the case of temperature dependence of
the WC / W2C mixtures. Above 450°C surface oxidation affected the measurements.
mixture of W,C and WC
-%
0
0.01
0.
c[%]
10
Fig. 5 Electrical resistivity on propane concentration measured at room temperature, untreated
and rapid thermal processing at 1200°C
103
n
350
WC
7
„^ eiw-i
H?*"£
300
f^
150
100
m ixture o fWCa id W2C *»^ TT^
250
300
350
T[°C]
Fig. 6 Electrical specific resistance on the measurement temperature
50
100
150
200
400
450
CONCLUSIONS
Single phase polycrystalline WC layers were prepared by sputtering of WC targets and
subsequent short time annealing in a propane-hydrogen ambient. Thermal treatment in pure
hydrogen resulted in carbon depletion in the layers and in the formation of W2C. Propane
diluted in the annealing ambient stimulated a transformation of the tungsten-carbon layers to a
stoichiometric WC phase. On amorphous silicon nitride polycrystalline tungsten carbide layers
with adjustable ratio between the WC and the W2C phase were formed and therefore different
values of the electrical resistance in dependence on the propane concentration emerged. In
contrary on thin cubic (100)SiC layers a preferred formation of a textured W2C phase was
observed after annealing in propane/hydrogen at 900°C. At higher temperatures the presence of
additional silicon resulted in the formation of a tungsten suicide by diffusion of silicon through
the thin SiC layer or by an interface reaction. The origin of this silicon as well as the influence
of orientation of the SiC layer ((100) vs. (111)) will be the subject of further investigations.
REFERENCES
[1] J. W. Palmour, L. A. Lipkin, R. Singh, D. B. Slater, A.V. Suvorov, C. H. Carter, jr., Diam. Rel. Mater. 6,
1400-4 (1997).
[2] V. E. Chelnokov, A. L. Syrkin, V. A. Dmitriev, Diam. Rel. Mater. 6,1480-4 (1997).
[3] L. M. Porter, R. F. Davis, Mater. Sei. Eng. B (Solid-State Materials for Advanced Technology) 34 (2-3),
83-105 (1995).
[4] Spieß, L.; Nennewitz, O.; Weishart, H.; Lindner, J.; Skorupa, W.; Romanus, H.; Erler, F.; Pezoldt, J.:
Aluminum implantation of p-SiC for Ohmic contacts; Diam. Rel. Mater. 6, 1414-8 (1997).
[5] Carbide, Nitride and Boride Materials - Synthesis and Processing, edited by A. W. Weimer, Chapman &
Hall, London, (1997).
[6] A. Bächli, J. S. Chen, R.P. Ruiz, M.A. Nicolet, MRS Symp. Proc. 339,247-52 (1994).
[7] P. Tägtström, H. Högberg, U. Jansson, J. O. Carlsson, J. de Phys. TV, 5 (C5, pt.2) 967-74 (1995).
[8] V. Cimalla, J.K. Karagodina, J. Pezoldt, G. Eichhorn, Mater. Sei. Eng. B29,170-175 (1994).
[9] G. Leitz, J. Pezoldt, I. Patzschke, J.-P. Zöllner, G. Eichhorn, MRS Symp. Proc. 303,171-176 (1993).
[10] Practical Surface Analysis, edited by D. Briggs and M.P. Seah, John Wiley & Sons, New York, (1990).
104
A FORMATION OF SiO^H-SiC INTERFACE BY OXIDIZING DEPOSITED POLY-SI
AND HIGH TEMPERATURE HYDROGEN ANNEALING
K. Fukuda ( NEDO industrial researcher )*, K. Sakamoto**, K. Nagai**, T. Sekigawa*'**
S. Yoshida*'**, and K. Arai***
*Ultra-Low Loss Power Device Technologies Research Body, Electrotechnical Laboratory, 1-1-4,
Umezono, Tsukuba, Ibaraki, 305-8568 Japan, kfukuda@etl.go.jp
**Electrotechnical Laboratory, 1-1-4, Umezono, Tsukuba, Ibaraki, 305-8568 Japan
ABSTRACT
A formation of Si02/4H-SiC interfaces by oxidizing deposited poly-Si on a 4H-SiC
substrate and high temperature hydrogen annealing at low pressure ( 8.5xl02 Pa ) has been
investigated. The oxidation rate of deposited poly-Si was approximately 100 times faster than
that of a SiC. Hydrogen annealing more effectively reduced the flat band voltage shift ( AVj, )
of the 4H-SiC MOS structure than argon and vacuum annealing. Moreover, the good Si02/4HSiC interface was formed because AVfc decreased as the oxidation temperature increased.
INTRODUCTION
Recently, many studies on high-temperature, high-power and high-frequency electronic
devices fabricated from 6H- and 4H-SiC have been reported because SiC has excellent physical
properties such as high electric field breakdown strength, high saturated electron velocity and
high thermal conductivity [1-3]. Metal-oxide-semiconductor field-effect-transistors ( MOSFETs )
are as important in power SiC devices as in power Si devices. However, SiC MOSFETs have not
been realized yet for practical use because of two main problems. One is the difficulty of
thermally oxidizing SiC due to the anisotropy and very small value of the oxidation rate. Another
is a large amount of Si02/SiC interface state. Especially, the latter is thought to be one of the
origins of low channel mobility of SiC MOSFETs.
In Si MOS technology, the Si02 film formed by oxidizing poly-Si is used, for example, as
an insulator film between a floating gate and a control gate in erasable programmable read only
memory ( EPROM ). The oxidation rate of poly-Si is faster than that of single crystal Si and
exhibits no anisotropy. It is also well known that hydrogen annealing terminates dangling
bonds of Si at the Si02/Si interface and decreases the interface state density ( Djt).
However, Afanasev et al. reported that hydrogen annealing at the pressure of 1.1 X 10s Pa
occurred positive charge at the Si02/SiC interface, resulting in the shift of C-V characteristics of
6H-SiC MOS structures toward negative voltage from the ideal C-V characteristics, and that
hydrogen annealing seemed to be insignificant for the decrease of dangling bonds [4,5]. In
105
Mat. Res. Soc. Symp. Proc. Vol. 572 ® 1999 Materials Research Society
contrast, we found that hydrogen annealing at low pressure ( 8.5xl02 Pa ) did not shift C-V
characteristics of 4H-SiC MOS structures toward negative voltage from the ideal C-V
characteristics and decreased Du [6,7]. The positive charge which occurs at the Si02/4H-SiC
interface in hydrogen annealing at the pressure of 1.1 X105 Pa is considered to be due to the
excess hydrogen. Therefore, if the combination of oxidizing deposited poly-Si and high
temperature hydrogen annealing at low pressure ( 8.5xl02 Pa ) is applied to the fabrication of
SiC MOSFETs, two problems as mentioned above will be solved.
In this study, we have investigated a formation of the Si02/4H-SiC interface by oxidizing
deposited poly-Si and high temperature hydrogen annealing at low pressure ( 8.5xl02 Pa ) and
effect of the oxidation temperature on high-frequency C-V characteristics.
EXPERIMENTAL
The
8° off-angled n-type
4H-SiC (0001) substrates from Cree Research with a
4.9-
ß m -thick n-type epitaxial layer were used for this study. The n-type dopant was nitrogen.
The effective carrier density ( Nd-N2 ) was 1.7 X1016 cm"3. After the standard RCA cleaning,
10 nm thick sacrifice oxide films were grown at 1100°C in dry 02, and then they were removed
by 5% HF solution before loading in ultra-high-vacuum ( UHV ). The 22 nm thick poly-Si
layers were deposited on 4H-SiC substrates in a Si molecular-beam-epitaxy ( MBE ) apparatus.
The poly-Si layers were thermally oxidized at 1050°C, 1150°C and 1250°C for 30 min in dry 02.
The samples were moved from the hot zone to the cool zone and cooled down to 20°C rapidly.
The post-oxidation annealing ( POA ) of samples oxidized at 1050°C and 1150°C were
performed in hydrogen ( 8.5 X102 Pa ) at 1000°C for 30 min and POA of samples oxidized at
1250°C were performed in hydrogen ( 8.5 X102 Pa ) , argon ( 8.5 X102 Pa ) and vacuum (1X \QA
Pa ) at 1000°C for 30 min. Aluminum on top of oxide films and on the backs of the samples was
evaporated to make gate electrodes and ohmic contacts of MOS structures, respectively. Highfrequency C-V measurements were performed using an HP 4274 LCR meter in a shielded dark
box at room temperature. The thickness of Si02 films grown on 4H-SiC substrates by oxidizing
deposited poly-Si as shown in Fig. 1 were measured using the surface profilometer ( Dektak
IIA ). The thickness of Si02 films of MOS structures, estimated from capacitance-voltage
( C-V ) characteristics, was 60 ± 5nm.
RESULTS AND DISCUSSION
Oxidation rate
Figure 1 shows the oxidation time dependence of the Si02 thickness formed by oxidizing
deposited poly-Si on SiC substrates, single crystal Si [8] and single crystal SiC [9] at 1050CC.
The oxidation rate of a poly-Si is approximately 100 times faster than that of single crystal SiC
106
[9] and approximately 1.2 times faster than that of single crystal Si as expected.
This value is
available for practical use.
120
1
i'
i
'
i
i'''
^100
i
o
I 80
a
o
i'''
T=1050°C
60 r
O
40 r
20 I
"10
I
I
I I
I I I I
20
I I
30
I I
I I
I I
I
40
I
I
I
I
I
50
W
70
Oxidation Time(min)
Figure 1 Oxidation time dependence of the thickness of Si02 films formed by oxidizing deposited poly-Si
( closed circles ) on 4H-SiC substrates, single crystal Si [8]( open circles ) ,and single crystal SiC [9]
( closed square) at 1050°C.
POA effect on high-frequency CV characteristics of 4H-SJC MOS structures
Figure 2 shows effect of POA on high-frequency C-V characteristics of 4H-SiC MOS
structures. The dotted line was calculated using the oxide capacitance ( Cra ) and N^-N, of sample
with hydrogen annealing. The measured capacitance at Vg<-3V are lower than the calculated
values due to a small amount of minority carriers generated at room temperature because of the
wide-gap of 4H-SiC.
POA condition dependence of AVffi and the surface state density ( Nss) +
the fixed charge density ( Nf ) is summarized in Table 1. The values of AVft and Nss+Nf of
sample without any annealing are 14V and -5.0xl012cm"2. Even the vacuum and argon annealing
decrease AVj, and Nss+Nf. Especially, AVft and Nss+Nf were reduced to 2.2V and -7.4x10" cm"2
by argon annealing, respectively. Hydrogen annealing decreases more effectively AVft and
Nss+Nt. C-V characteristics of the sample with hydrogen annealing is very close to the ideal C-V
characteristics. We have already revealed that the density of hydrogen which accumulates at the
Si02/SiC interface formed by thermal oxidation of an SiC substrate increases as hydrogen
annealing temperature increases using secondary ion mass spectroscopy ( SIMS ) [6,7]. As it is
thought that the same phenomena occurs in this experiment, hydrogen terminates dangling bonds
of Si or C atoms at the Si02/SiC interface more strongly than oxygen because the binding energy
of Si-H is lower than that of Si-O. As a result, the values of AVj, and Nss+Nf of the sample
with hydrogen annealing are smaller than those of samples with argon or vacuum annealing,
which are 1.1 V and -4.0x10" cm"2, respectively. Hydrogen annealing is the most available for
the good Si02/4H-SiC interface.
107
I I I I I J I I I I I I I I I I I I I I I I I I I I I I I I
Vaccum
f=100kHz
Without
i
15
-10
-5
'
i
10
0
<
* '
15
20
Vg(V)
Figure 2 POA effect on high-frequency C-V characteristics of 4H-SiC MOS structures. The gate oxide
films were grown at 1250°C. POA in hydrogen ( 8.5xl02 Pa), argon( 8.5xl02 Pa) and vacuum( lxlO4
Pa ) was performed at 1000°C for 30min after oxidation.
Table 1 POA condition dependence of AV,,, and Nss+Nf
Nss+Nf(cm-2)
POA condition
AVm(V)
14
Without
-5.0xl012
4
7.4
-2.8xl012
Vacuum ( 1x10" Pa)
2.2
-7.4x10"
Argon (8.5xl02Pa)
1.1
-4.0x10"
Hydrogen ( 8.5xl02 Pa )
Oxidation temperature effect on high-frequency CV characteristics of 4H-SJC MOS structures
Figure 3 shows oxidation temperature effect on high-frequency C-V characteristics of 4HSiC MOS structures. All samples were annealed in hydrogen ( 8.5xl02 Pa ) at 1000°C for 30 min
after oxidation. The dotted line and C-V characteristics at 1250°C are the same as those in Fig.2.
AV^and Nf decrease with increasing temperature as shown in Fig.4. This suggests that dangling
bonds of Si or C atoms are terminated by oxygen atoms strongly as temperature increases. The
Si02/4H-SiC interface becomes better as oxidation temperature increases.
108
—1—1—
f=100kHz
-2
0 2 4
Vg(V)
6
8
10
Fig. 3 Effect of oxidation temperature on high-frequency C-V characteristics of 4H-SiC MOS structures.
They were annealed in hydrogen (8.5xl02 Pa) at 1000°C for 30 min after oxidation.
1000 1050 1100 1150 1200 1250 1300
Oxidation temperature (°C)
Fig.4 Oxidation temperature dependence of AV„, and NK+N, estimated from C-V characteristics of 4H-SiC
MOS structures as shown in Fig. 3.
CONCLUSION
The oxidation rate of deposited poly-Si is approximately 100 times faster than that of single
crystal SiC, which is available for practical use. Hydrogen annealing at low pressure ( 8.5xl02
Pa ) more effectively reduced AVj, and Nss+N, than argon and vacuum annealing, resulting in
AVj, and NS5+Nf of 1.1V and -4.0x10" cm"2, respectively. The value of AVfc is also reduced as
oxidation temperature increases. Finally, the problems of slow oxidation rate of single crystal
SiC and the bad Si02/4H-SiC interface can be solved using these technologies.
109
ACKNOWLEDGEMENTS
This work was carried out by the Ultra-Low Loss Power Device Technologies Project
under the management of the R&D Association for Future Electron Devices ( FED ) as a part of
the Ministry of International Trade and Industry ( MITI ) R&D of Industrial Science and
Technology Frontier Program supported by New Energy and Industrial Technology Development
Organization ( NEDO).
REFERENCES
1.
2.
3.
4.
5.
6.
7.
8.
9.
J. A. Cooper, Jr., M. R. Melloch, J. M. Woodall, J. Spitz, K. J. Schoen and J. P. Henning :
Ill-Nitrides and Related Materials, eds. G. Pensil, H. Morkoc, B. Monemar and E. Janzen
(Trans Tech Publications Ltd, Switzerland-Germany-UK-USA, 1998), p. 895.
K. Hara : Ill-Nitrides and Related Materials, eds. G. Pensil, H. Morkoc, B. Monemar and E.
Janzen (Trans Tech Publications Ltd, Switzerland-Germany-UK-USA, 1998), p. 901.
C. E. Weitzel : Ill-Nitrides and Related Materials, eds. G. Pensil, H. Morkoc, B. Monemar
and E. Janzen (Trans Tech Publications Ltd, Switzerland-Germany-UK-USA, 1998),
p. 907.
V. V. Afanasev, M. Bassler, G. Pensl and M. Schulz : Silicon Carbide, eds. W. J. Choyke,
H. Matsunami and G. Pensl (Akademie Verlag, Berlin, 1997) Vol. 2, p. 321.
V. V. Afanasev, A. Stesmans and C. I. Harris : Ill-Nitrides and Related Materials, eds.
G. Pensil, H. MorkoQ, B. Monemar and E. Janzen ( Trans Tech Publications Ltd,
Switzerland • Germany • UK • USA, 1998 ), p. 857.
K. Fukuda, K. Nagai, T Sekigawa, S. Yoshida, K. Arai and M. Yoshikawa : Extended
Abstract of the 1998 International Conference on Solid State Device and Materials ( the
Japan Society of Applied Physics, Tokyo, 1998), plOO.
K. Fukuda, K. Nagai, T. Sekigawa, S. Yoshida, K. Arai and M. Yoshikawa :
to be published in Jpn. J. Appl. Phys.
Y. Kamigaki and Y. Itoh : J. Appl. Phys.,48, 2891 (1977).
A. Gölz, G. Horstmann, E. Stein von Kamienski and H. Kurz : Silicon Carbide and Related
Materials, eds. S. Nakashima, H. Matsunami, S. Yoshida and H. Harima ( IOP Publishing
Ltd, London, 1995 ),p.634.
110
HIGH TEMPERATURE STABLE WSi2-CONTACTS
ON P-6H-SILICON CARBIDE
Frank ERLER *, Henry ROMANUS *, Jörg K.N. LINDNER **, Lothar SPIESS *
Technical University of Ilmenau, Institute of Applied Materials Science, Germany
University of Augsburg, Institute of Physics, Germany
ABSTRACT
Amorphous tungsten-silicon layers were deposited by DC co-sputtering and
subsequently annealed in an argon atmosphere up to 1325 K to form tetragonal crystalline WS12.
Al-implanted p-6H-SiC exhibits a small depletion area forming an ohmic contact with low
specific contact resistance. A modified Circular Transmission Line Model (CTLM), introduced
by Mario w & Das [1] and Reeves [2], was used to characterize the electrical properties of the
prepared contacts in the range between 300 K and 650 K. Deviations between calculated fieldemission contact resistances and measured contact resistances (pc=2-10~2 Qcm2, T=650 K)
could be explained by TEM-cross section investigations. These deviations are caused by
inhomogeneous contact interfaces originating from technological difficulties during contact
preparation.
INTRODUCTION
Silicon Carbide as a wide band gap semiconductor is suitable for high temperature
devices with different applications. For this purpose it is necessary to obtain low-resistivity
ohmic contacts on active semiconductor areas which do not degrade at high temperatures.
Therefore, the contact material must have a high melting point and should not allow chemical
reactions with SiC which may influence the electrical properties of the contact area.
Due to theoretical calculations of a p-depletion contact on 6H SiC [3], WS12 is
favourable as metallization material because of the high work function of 4.62 to 4.70 eV [4, 5].
The high acceptor activation energy (> 0.24 eV) causes an ionization of only approximately 1%
of available acceptors at room temperature [6]. Dopant incorporation via diffusion is not feasible
in SiC. However, ion implantation is a possibility to increase the acceptor concentration at the
surface. This causes a narrower depletion zone which allows for thermionic-field or field
emission. Fig. 1 demonstrates the influence of the hole concentration on the specific contact
resistance [7, 8], using Eq. (1).
Pc
qAT
exp-
4n^m*.ES
*fl
qh
,yfi7
tanh
qh
47tym*e
1
JNA-
\
kT
(1)
Thus an ohmic contact with a low specific contact resistance should be feasible.
Further research work [9] focused on the preparation and patterning of tungsten suicide
metallization and on measurement of contact resistivity, applying the Circular Transmission
Line Model [1,2].
111
Mat. Res. Soc. Symp. Proc. Vol. 572 ° 1999 Materials Research Society
EXPERIMENT
In order to increase the
acceptor concentration near the
surface, an Al-implantation was
performed using different doses
between 51014 and 61015 cm"2 at
room temperature and an
implantation energy of 50 keV. A
theoretical calculation of the
implantation profiles (TRIM) is
shown in Fig. 2. The
recrystallization of SiC was
achieved by annealing at 1925 K in
argon atmosphere for 20 min.
To obtain high temperature
stable ohmic contacts consisting of
WSJ2, tungsten and silicon were DC
co-sputtered in a stoichiometric ratio
Si:W of 2.1:1, resulting in an
amorphous layer of 400 nm
thickness. During the formation of
crystalline WSi2 at 1325 K in an
argon atmosphere for 20 min a
theoretical shrinkage to 300 nm is
expected to appear [10].
p [cm"
18
^* «17
H
10'° 7*10
I
T = 300 K
10"'
thermior lie emiss on
:
10"
/the rmionic-f eld emis sion
400 K
3mission
10"" =
E
o
a, 10"!
/
o
CL
10"'
"l\
i
A
500 K
A
700 K
900 K
10"' -
:
10"8
= 0.4 eV, A* = 96 A / K2cm2, m' = 0.8
r
|
, I
10
12
1/Sqrt(p)[10"1°cm 2'3]
Fig. 1 Calculated specific contact resistance
versus hole concentration
)
200
400
600
800
1000
1200
depth [A]
Fig. 2
Implantation profiles calculated using TRIM
112
1400
The formation of a crystalline WSi2-layer is demonstrated by x-ray diffraction using
different angles of incidence and Bragg-Brentano-geometry (Fig. 3).
25
,....,....,.... ,,V. ifij' i wwygAt |inn/i ii/. >n > | i y*i i f V>nn
30
35
40
45
50
55
60
65
70
75 ZU
Fig. 3
X-ray diffraction patterns of annealed WSi2
Measurement structures were patterned by a plasma enhanced chemical etching process
using a photo resist mask. The used gas fluxes were 20 seem CI2 and 20 seem SF6, the chamber
pressure amounted to 310 2 mbar and the RF-power was 50 W. Etching efficiency was
controlled by optical microscopy.
An example of contact resistance
measurement structures is shown in Fig. 4.
A modified Circular Transmission Line Model
according to Mariow & Das [1] and Reeves [2]
! )
was applied for obtaining specific contact
resistances at different temperatures up to 650 K.
Fig. 4 Contact resistance measurement
RESULTS
structures according to Marlow & Das (a)
and Reeves (b)
The summarized results for the measured
specific electrical resistivity and the temperature coefficient of the WSi2-layer are shown in
Fig. 5. The obtained resistivity of approximately 47 uQcm is slightly larger than the value of
40 ußem for WSi2-layers given in literature [4, 5]. This deviation indicates complete
transformation into crystalline \VSi2 with a small amount of pores.
Atomic Force Microscopy was used to determine the thickness of the annealed \VSi2metallization layer. It amounts to approximately 350 nm (Fig. 6), thus it is thicker than the
expected value of 300 nm. One possible explanation for this is the formation of pores in the
metallization layer during the annealing process.
Specific contact resistances of different samples were calculated from measured currentvoltage-plots. One example of such current-voltage-plots for different temperatures is shown in
Fig. 7. The equation systems, developed using CTLM and regression analysis, were solved by a
mathmatical software package. Table I shows some obtained values of specific contact
resistances for different temperatures, upper epi-layer concentrations and implantation doses.
o o
o
113
90
80
E
o
1
1
6H-SiC, 350 nm WSi2
annealed (1325 K, 20 min)
x
70
a
60
50
y
r
y
P20"C
= 47,3 poem
a
= 0,00203 K'
20*C
40
400
300
y
Regres; ion analysis
Y = n + T1*X
Param ValiiR
19,19343
n
3,09597
m
S*
?
y
'S
500
600
700
T[K]
Fig. 5
Table I
Specific electric resistance and temperature coefficient of WSi2-metallization
Measured specific contact resistivity of implanted and WSi2-coated p-6H SiC
upper epi-layer concentration, ND imp implanted dose)
(NAI
PC300K[^Cm ]
no WSi2-annealing
Pc300K[ßcm ]
annealed
Pc650K[ßcm2]
NA] = 3-10 cnf ,
NDirap = 5-1014cm-2
Schottkycontact
Schottkycontact
Schottkycontact
NAj = 7-1016cm-3,
NDimp=l-1015cm-2
Schottkycontact
Schottkycontact
0.2 - 0.8
NAi = 3-1018cm"3,
NDirap = 6-10l5cm"2
Schottkycontact
0.25-1.7
(2... 5)-10"2
p-6H-SiC
18
3
Scan Range: 56 pm
Resolution: 200 x 200
Z[nm]
Pointl: 42.1
Point2: 397.8
Diff.:
355.7
i
y~i~^~s—~^-~
CO
+-•
(0
Q
N
Distance [|jm]
Fig. 6 Atomic Force Microscope image
of an annealed and patterned ring structure;
depth-measurement
annealed
At higher temperature more of the
implanted acceptors are ionized. Therefore, a
lower ohmic contact resistivity was obtained.
A high-resolution cross-sectional TEM
(XTEM) micrograph of metallized 6H-SiC
(implantation: NDimP = 6 1015 cm"2, 50 keV) is
shown in Fig. 8. The grainy structure of the
metallization layer (denoted as M) is clearly
visible, the grain size being in the range of a
few hundred nanometers. The metallic layer
contains various pores, which are
characterized by bright areas in Fig. 8 and
which are most likely the result of the thermal
treatment within the metallization process
rather than of the sputtering technique used in
specimen preparation. The interface (small
arrow) between the metal layer and the SiC
114
substrate shows a wavy morphology with trenches reaching up to 100 nm down into the
substrate. These trenches are partially filled with metal grains.
-0,5
0,0
1,0
voltage [V]
Fig. 7 Measured current-voltage-plot of a 3-ring contact structure,
Reeves pattern, temperature as parameter
The SiC substrate is
subdivided into four regions, the
undamaged 6H substrate
followed by a 6H-layer with
small defects causing grainy
contrasts in Fig. 8. The 6H layers
are covered with epitaxially
grown 3C-SiC((lll)3C II
(0001)6H), the interface being
located approximately 130 nm
beneath the metal/
semiconductor interface. The
3C-SiC part is subdivided into
a defect rich, approximately 80
nm thick lower zone showing
stripy contrasts (planar defects)
in Fig. 8 and a 50 nm thick
Fig. 8 TEM image of the interface region
nearly defect free upper zone.
WSi2-metallization layer / p-6H Silicon Carbide
The results indicate that
(shows different defects, e.g. pores, 3C SiC)
the metal semiconductor contact
mainly consists of a metal/3C-SiC/6H-SiC contact, even though due to the trenches formed
there may be locally also some small metal/6H-SiC contact regions.
115
CONCLUSIONS
Doping of 6H-SiC by implantation of 50 keV aluminium ions using a dose of
6 1015 cm probably leads to the formation of an amorphous surface layer. This amorphous
layer may recrystallize as a twinned 3C-SiC layer during the subsequent 20 min anneal at
1925 K. In addition it has to be clarified whether the trenches observed are the result of the
annealing procedure, the subsequent HF and plasma cleaning or the reaction of deposited W or
Si with the SiC surface during metallization. Since the defect layers in SiC are flat, any
implantation related reasons can be ruled out.
Due to different difficulties during the implementation of implantation and metallization
process it was not possible to obtain contact resistances less then 10"3 Qcm . However, the high
implanted samples lead to an ohmic contact particularly at higher temperatures. This indicates a
field emission conducting process which is demonstrated in the left part of Fig. 1. Therefore,
WSi2 should be an available contact material for high temperature applications.
The aim for further research work has to be, firstly, the prevention of solid-state
transformation to 3C-SiC during the implantation annealing process. Secondly, to avoid the
growth of pores during the annealing for the formation of the tungsten disilicide layer.
ACKNOWLEDGMENTS
The authors gratefully acknowledge the advice and support of Olaf Nennewitz, Jörg Pezoldt,
Jakob Kriz and Wolfgang Skorupa.
Parts of this work were supported by the BMBF under contract no. 01 BM 303/8 (1993-1996).
REFERENCES
1.
2.
3.
G.S. Marlow, M. Das, Solid-State Electronics 25, 91-94 (1982).
G.K. Reeves, Solid State Electronics 23, 487-490 (1980).
L. Spiess, O. Nennewitz, H. Weishart, J. Lindner, W. Skorupa, H. Romanus, F. Erler,
J. Pezoldt, Diamond and Related Materials 6, 1414-1419 (1997).
4. S.P. Muraka, Metallization-77ieory and practice for VLSI and ULSI, (ButterworthHeinemann, Reed Publishing Inc., USA, 1993), pp. 41-64.
5. H.F. Hadamovsky, Werkstoffe der Halbleitertechnik, 1st ed. (Deutscher Verlag für
Grundstoffindustrie, Leipzig, DDR, 1985), pp. 262-291. [German]
6. L. Spiess, O. Nennwitz, J. Pezoldt, Inst. Phys. Conf. Ser. 142, 585-588 (1996).
7. Bergmann, Schäfer, Freyhardt, Festkörper - Lehrbuch der Experimentalphysik, Band 6
(Walter de Gruyter & Co., Berlin, Germany, 1992), pp. 531-539. [German]
8. S.S. Cotten, G.S. Gildenblat, Metal Semiconductor contacts and devices, in VLSI
Electronics, edited by N.G Einspruch (Academic Press Inc., London, 1986)
9. J. Kriz, K. Gottfried, C. Kaufmann, T. Gessner, Diamond and Related Materials 7,
pp. 77-80(1998).
10. A. Fabricius, O.Nennewitz, L. Spiess, V. Cimalla, J. Pezoldt, Mater. Res. Soc. Symp. Proc.
402, Boston, 1995, pp. 625-630
For further information, please feel free to contact:
Frank Erler, TU Ilmenau, PF 100565, 98684 Ilmenau, Germany
phone: +49 36 77 / 69 31 11, fax: +49 36 77 / 69 31 04
email: erler@e-technik.tu-ilmenau.de, www: http://phase.e-technik.tu-ilmenau.de
116
STRUCTURAL AND ELECTRICAL PROPERTIES OF BERYLLIUM
IMPLANTED SDLICON CARBIDE
T. HENKEL*, Y. TANAKA*, N. KOBAYASHP, H. TANOUE*,
M. GONG**, X.D. CHEN**, S. FUNG** and CD. BELING**
* Electrotechnical Laboratory, Tsukuba, Ibaraki 305-8568, Japan, henkel@etl.go.jp
** Physics Department, The University of Hong Kong, Hong Kong, China
ABSTRACT
Structural and electrical properties of beryllium implanted silicon carbide have been investigated
by secondary ion mass spectrometry, Rutherford backscattering as well as deep level transient
spectroscopy, resistivity and Hall measurements. Strong redistributions of the beryllium profiles
have been found after a short post-implantation anneal cycle at temperatures between 1500 °C
and 1700 °C. In particular, diffusion towards the surface has been observed which caused severe
depletion of beryllium in the surface region. The crystalline state of the implanted material is
well recovered already after annealing at 1450 °C. However, four deep levels induced by the
implantation process have been detected by deep level transient spectroscopy.
INTRODUCTION
Silicon carbide (SiC) is a promising material for high temperature, high frequency, and high
power device applications. Group DI elements are commonly used as acceptor dopants for this
semiconductor. Investigations on p-type SiC doped with these elements, however, revealed poor
electrical characteristics due to high acceptor ionization energies and low hole mobilities [1-5].
Therefore, alternative dopants with a higher electrical activation are highly desirable.
Due to the low mass and high solubility in the SiC lattice [6], beryllium (Be) can be used for the
production of thick p-type layers applying ion implantation. However, there have been very few
reports on Be doped SiC [7-12]. Be is known to be an electrically active impurity, i.e., a doubly
charged acceptor in SiC [10]. Two acceptor levels at 0.42 and 0.6 eV, respectively, were
determined by Hall measurements [8]. Further, one deep level at 0.38 eV was obtained by I-V
measurements on p-n junctions produced by Be implantation [12]. However, the precision of
these results is questionable because that data analysis is based on simplified model assumtions.
Despite the fact that Be has been successfully applied in the fabrication of diodes [11,12], much
is still unknown about the structure of this dopant in the SiC lattice.
In this paper, we report on structural and electrical properties of Be implanted SiC. Samples
were characterized by secondary ion mass spectrometry (SMS), Rutherford backscattering
spectrometry / channeling (RBS/C), deep level transient spectroscopy (DLTS), resistivity and
Hall measurements.
EXPERIMENT
Epitaxial layers ([0001] orientation, «-type, off-axis, thickness: 10 urn, carrier concentration:
lxlO16 cm"3) grown on 6H-SiC substrates as purchased from Cree Research [13] were used as
starting material. ^Be* was implanted applying energies between 50 and 590 keV (see Table I) in
order to obtain a box-shaped profile. Samples were maintained at room temperature (RT), and
117
Mat. Res. Soc. Symp. Proc. Vol. 572 ° 1999 Materials Research Society
were tilted 7° with respect to the ion beam to minimize channeling effects during implantation.
Ion range and nuclear energy distributions were obtained by Monte Carlo (MC) simulations
using the TRIM code (SRIM-98, full cascade) [14]. A mean displacement energy of 25 eV for
both Si and C atoms was applied. The concentration profile and the nuclear energy density
distribution are shown in Fig. 1. The critical energy density for the amorphization of SiC crystal
at RT (2xl021 keV/cm3 [15]) is indicated by the dashed line.
Ion energy (keV)
Ion fluence (10l4cm2)
50
75
100
130
170
210
260
320
400
490
590
0.61
0.62
0.63
0.84
0.74
0.85
0.80
1.00
1.03
0.94
1.52
2 „
«
ST 10"
18 1Q18
E
/
/
1 £•
10"
1016
300
600
900
1200
Depth (nm)
Fig. 1. Total nuclear energy distribution
tal dopant concentration for Be
mm
and total
implanted SiC as calculated using TRIM
Table I. Schedule used for Be implantation.
To repair the crystal damage and activate the implanted dopant, samples were annealed in
flowing argon (Ar) gas (1-2 atm) at temperatures between 1450 °C and 1700 °C for 1 min using
a rapid thermal annealing (RTA) system. The temperature rise and fall rates were about 50 °C/s
and 20 °C/s. Each sample was covered with another SiC crystal to protect the sample surface
from Si dissociation during annealing. Before the electrical measurements, a planar layer of
about 0.6 um was removed from the top of the epilayer by applying a combination of ion
implantation and wet chemical etching [16,17]. Ohmic contacts were prepared by metal
deposition (titanium on the top, nickel on the backside) using an e-beam evaporation system and
a post-deposition RTA cycle at 1100 °C for 5 min in flowing Ar gas.
To evaluate the thermal stability of the Be profile, SIMS measurements were conducted using a
CAMECA ims 4f instrument with a 14 keV N02" primary ion beam. The depth conversion was
performed by measuring the total crater depth with a surface profiler and assuming a constant
sputtering rate during depth profiling. The concentration calibration was performed using a
standard sample fabricated by a single energy 9Be+ implantation. A sensitivity factor was then
derived by comparing the area under the SIMS profile obtained from this sample with that one of
a profile obtained by a MC simulation.
Resistivity and Hall measurements employing the van der Pauw geometry were performed at
RT. To study deep level defects, DLTS measurements were conducted using an equipment
described elsewhere [18]. Ionization energies and capture cross-sections were evaluated from the
temperature dependence of the emission rates. Finally, damage distributions were determined by
3 MeV 4He+ RBS/C along the [0001] axis using a scattering angle of 150°. The depth scale given
in the RBS spectra below was calculated using the mean energy approximation [19] and the
density of the crystalline material (9.66xl022 at/cm3 [15]).
118
RESULTS
SIMS measurements were performed on both the as-implanted and the post-implantation
annealed samples to investigate the Be distribution before and after the high temperature
treatment. As can be seen in Fig. 2, the near surface tail and the plateau concentration of the asimplanted atom distribution are well reproduced by the MC simulations. However, the slope of
the tail towards the substrate is lower compared with the theoretical profile. Channeling effects
can be responsible for the observed discrepancy which are not considered in SRIM-98. Since
vacancy-type defects far beyond the nuclear energy deposition profile were detected by Positron
annihilation spectroscopy [20], defect-enhanced diffusion could be another reason.
1.0
Depth (|im)
Fig. 2.
SIMS 9Be depth profiles in SiC before and after annealing at the temperatures indicated.
A simulated depth profile as obtained using TRIM is shown for comparison.
Strong redistributions were found in the annealed samples. The dopant profile is already
thermally unstable after a short RTA cycle at 1500 °C. In particular, the heat treatment caused
severe depletion of Be in the surface region up to two orders of magnitude below the asimplanted atom concentration. The integral of the depth profile, which is a measure for the
implanted ion fluence, results in a fluence around 30 % lower compared with the as-implanted
profile. This indicates out-diffusion of the dopant which increases with increasing temperature,
i.e., 43 % and 92 % Be loss were found at 1600 °C and 1700 °C, respectively. Implantation
induced defects may govern the diffusion process. However, because of the small atomic size of
Be, migration via an interstitial mechanism is also anticipated.
Additionally, in-diffusion into the bulk of the epilayer, although less pronounced, was also
observed. The higher the anneal temperature the stronger is the redistribution in the tail region.
119
Moreover, after RTA at 1500 °C a peak in the Be profile can be seen at a depth of about 1.1 pm
which became more pronounced at higher anneal temperatures. It is assumed that Be atoms were
trapped by thermally stable defects formed during the anneal process in this (end-of-range)
region.
To remove the Be depleted surface layer, all samples (except the as-implanted specimen) to be
examined in the following by RBS and electrical measurements were etched as described above
and processed at temperatures <1600 °C.
Depth (nm)
600
900
1200
300
—i—
c
o
w
>.
CO
CD
rr
102 -
a
A
RTAat1450°C
RTAat1550°C
RTA at 1600 °C
D
°
random
as-implanted
virgin aligned
*o
ID
I ff
0.9
1.5
1.2
1.8
Energy (MeV)
Fig. 3.
RBS/C spectra (Si sublattice) of 3 MeV "He+ backscattered from Be implanted SiC before and
after annealing at the temperatures indicated
To characterize the as-implanted state and to evaluate the recovery of the crystal lattice, RBS/C
measurements were performed (see Fig. 3). A random and an aligned spectrum from virgin
unimplanted material are also shown for comparison. The scattering yield in the aligned
spectrum obtained from the as-implanted sample is far below the yield in the random spectrum.
This means that the crystal was not amorphized during implantation, in agreement with the
nuclear energy deposition calculated (see Fig. 1). This is a necessary condition for complete
annealing of damage induced by ion implantation since defect-free recrystallization of
amorphous SiC is not possible [15]. Further, the yield in the as-implanted sample approaches the
yield in the virgin material at about 1.2 um, i.e., the damaged region extends up to this depth
which is consistent with the calculated energy deposition profile. The peak observed at an energy
of 1.18 MeV is due to an oxide layer on the SiC surface which is confirmed by SIMS.
120
As can be further seen in Fig. 3, the scattering yield in the aligned spectra from the annealed
samples almost coincides with the yield in the aligned spectrum from the virgin sample at all
depths indicating a good lattice quality within the sensitivity limit of RBS. Thus, it can be
assumed that the crystalline state is well recovered after a RTA cycle at temperatures >1450 °C.
Resistivity and Hall measurements were performed at RT to obtain electrical properties of the
doped SiC layers. For comparison, a virgin unimplanted n-type sample was etched and provided
with contacts as described above. Although a weak p-type conduction was detected in the
implanted epilayers (free hole concentrations in the 1016 cm-3 range), well reproducible Hall
voltages could not be obtained. Obviously, the Hall measurements were strongly affected by the
substrate due to insufficient isolation of the p-type layer. However, resistivities were found to be
about 0.5 Qcm, i.e., one magnitude lower compared to the virgin sample. Within the limits of
these measurements, a dependence on the post-implantation anneal temperature in the range
from 1450 to 1600 °C was not detected.
Finally, a typical DLTS spectrum as obtained after RTA at 1600 °C is shown in Fig. 4. Four
peaks labeled BEi, BE2, BE3, and BE4, respectively, were observed in the temperature range
from -150 to +100 °C. Since no DLTS signals corresponding to deep levels were observed in
virgin unimplanted samples, all the deep levels were therefore introduced by Be implantation.
Ionization energies and capture cross-sections of these levels are given in Table EL For the
calculation of the capture cross-sections, these levels were assumed to be electron traps.
Since the free hole concentration of the p-type layer produced by Be implantation was found to
be of the same magnitude compared to the free electron concentration of the epilayer as stated
above, the width of the p-type depletion layer is assumed to be comparable to that of the n-type
one. Thus, the DLTS signals observed may arise from either electron traps at the n-side or hole
traps at the p-side of the p-n junction. The question, whether these levels are donor- or acceptorlike, cannot be answered from the present results but possibly by DLTS measurements on Be
implanted p-type material provided with Schottky contacts on top of the samples. Further
investigations are necessary in order to understand the origin of the deep levels and the structure
of the corresponding defects.
3 -Rate window: 136.4 ms
Deep level
BE,
Ionization Capture crossenergy (eV) section (cm2)
0.34
LL
5xl013
BE2
0.46
5xl014
BE3
0.52
5xl0"14
BE4
0.66
4xl0"16
*
x5
/
& 2
BE
1
3
CO
c
o>
"co
03
"*
ti 1
Q
"
/
;T /
0
-150
/
.
-100
-50
0
50
100
Temperature (°C)
Fig. 4. DLTS spectrum recorded on Be
implanted n-type SiC after annealing at
1600°C
Table n. Ionization energies and capture
cross-sections of the deep levels as determined from DLTS data
121
CONCLUSIONS
The doping behaviour of Be implanted into «-type 6H-SiC epitaxial layers was investigated by
SIMS, RBS/C, DLTS, resistivity and Hall measurements. Strong redistributions of the asimplanted Be profiles were found after RTA at temperatures >1500 °C. In particular, Be diffused
towards the surface on a higher level than into the epilayer. As a consequence, severe Be
depletion in the surface region occured. Moreover, it was shown by RBS/C that the crystalline
state is well recovered after a short RTA cycle at 1450 °C. However, four deep levels labeled
BE|, BE2, BE3, and BE4 were observed by DLTS which were generated by Be implantation. The
energy position of these levels as well as the nature of the corresponding defects is still an open
question.
ACKNOWLEDGMENTS
We thank Y. Ishida for the help during the Hall measurements. One of the authors (T.H.)
gratefully acknowledges the support of this work by Science and Technology Agency of Japan.
REFERENCES
1. T. Troffer, M. Schadt, T. Frank, H. Itoh, G. Pensl, J. Heindl, H.P. Strunk, M. Maier, phys. stat. sol.
(a) 162, 277 (1997).
2. M.V. Rao, J.A. Gardner, P.H. Chi, O.W. Holland, G. Keiner, J. Kretchmer and M. Ghezzo, J. Appl.
Phys. 81(10), 6635 (1997).
3. T. Henkel, Y. Tanaka, N. Kobayashi, I. Koutzarov, H. Okumura, S. Yoshida and T. Ohshima, Mat.
Res. Soc. Symp. Proc. 512, 163 (1998).
4. M. Gong, S. Fung, CD. Beling, G. Brauer, H. Wirth and W. Skorupa, J. Appl. Phys. 85(1) 105
(1998).
5. T. Troffer, G. Pensl, A. Schöner, A. Henry, C. Hallin, O. Kordina, E. Janzen, Mater. Sei. Forum
264-268, 557 (1998).
6. G.L. Harris, Properties of Silicon Carbide, (INSPEC, London, 1995), pp.153.
7. A.A. Kalnin, Y.M. Tairov and DA. Yaskov, Sov. Phys. - Solid State 8(3), 755 (1966).
8. Y.P. Maslakovets, E.N. Mokhov, Y.A. Vodakov and G.A. Lomakina, Sov. Phys. - Solid State 10(3),
634 (1968).
9. OJ. Marsh and H.L. Dunlap, Rad. Eff. 6, 301 (1970).
10. P.G. Baranov, Mater. Sei. Forum 264-268, 581 (1998).
11. N. Ramungul, Y. Zheng, R. Patel, V. Khemka and T.P. Chow, Mater. Sei. Forum 264-268, 1049
(1998).
12. N. Ramungul, V. Khemka, Y. Zheng, R. Patel and T.P. Chow, IEEE Trans. Electron Devices 46(3),
465 (1999).
13. Cree Research, Inc., 4600 Silicon Drive, Durham, NC 27703
14. J.F. Ziegler, J.P. Biersack, and U. Littmark, The Stopping and Range of Ions in Solids, (Pergamon,
New York, 1985), pp. 1.
15. V. Heera, W. Skorupa, Mat. Res. Soc. Symp. Proc. 438, 241 (1997).
16. J.A. Edmond, J.W. Palmour and R.F. Davis, J. Electrochem. Soc. 133(3), 650 (1986).
17. D. Alok and B.J. Baliga, J. Electron. Mater. 24(4), 311 (1995).
18. C.V. Reddy, S. Fung, and C. D. Beling, Rev. Sei. Instrum. 67(1), 257 (1996).
19. W. K. Chu, J. W. Mayer, and M. A. Nicolet, Backscattering Spectrometry, (Academic, New York,
1978), pp. 64.
20. H. Wirth, W. Anwand, G. Brauer, M. Voelskow, D. Panknin, W. Skorupa and P.G. Coleman, Mater.
Sei. Forum 264-268,729 (1998).
122
ELEVATED TEMPERATURE SILICON CARBIDE CHEMICAL SENSORS
M.A.GEORGE*, M. A. AYOUB*,
D. ILA** and D. J. LARKIN***
* Department of Chemistry, University ofAlabama inHuntsville
"Center for Irradiation ofMaterials, Alabama A&M University
'"NASA Lewis Research Center
ABSTRACT
In this study, the (I-V) properties of the sensors were measured as a function of
hydrogen, propylene and methane exposure at temperatures up to 400° C and sensor responses
were observed for each gas. The response to hydrogen and propylene had a rapid increase and
leveling off of the current followed by the subsequent decrease to the baseline when the gas was
switched off. However, exposure to methane resulted in a rapid spike in the current followed by
a gradual increase with continued exposure. X-ray photoelectron (XPS) studies of methane
exposed SiC sensors revealed that this behavior is attributed to the oxidation of methane at the
Pd surface.
INTRODUCTION
The Study of SiC has focused on methods to grow high quality SiC for a wide range of
applications including high temperature, high power devices [1] as well as optoelectronic
devices [2]. Recently, it has also been shown that SiC can be employed as both an oxygen and
a hydrogen sensor that operates in a temperature regime considerably higher than conventional
sensors such as tin oxide (Sn02) or silicon. Because of its outstanding thermal stability, silicon
carbide can be employed as a hydrogen and hydrocarbon sensor that can potentially operate at
temperatures up to 1000 °C [3-7]. Potential uses of elevated temperature SiC sensors include
automotive applications, process gas monitoring, aeronautics, and aerospace applications.
The deposition of a catalytic metal such as Palladium (Pd) onto silicon carbide (SiC) results
in Schottky diode behavior. The adsorbing gas changes the space charge region under the metal
clusters which in turn affect the conductivity of the crystal. This change in conductivity is
measured and can be correlated to surface concentrations and to the levels of the sampled gas in
the ambient. The high sensitivity for hydrogen containing combustible gases is enhanced by the
presence of catalytic metals. Hydrogen containing species dissociatively adsorb to the metal and
hydrogen atoms migrate to the Pd/SiC interface where they affect the current-voltage (I-V)
properties of the SiC [7].
EXPERIMENTAL
For this study of the 3
response of silicon carbide cd
sensors to various gases, 5x7 _IF_I
mm samples of epitaxial silicon n
<i>
carbide films deposited on bulk
3
silicon carbide substrates were Ü
examined. Both silicon face and
carbon face samples were
produced,
however
the
preliminary
studies
were
performed on silicon faced silicon
carbide films only. The sensors
were prepared by depositing 100
nm thick aluminum films to the
backside of the sample, while
palladium films were deposited to
\
c
s
£
£
C
*=
C
s
2
£
a
o
%
£
I l_
S
2
?
S
£
£
Time (sec)
Figure 1 Current response at 0.7 V to H2 and
N2 exposure.
123
Mat. Res. Soc. Symp. Proc. Vol. 572 ° 1999 Materials Research Society
the epitaxial silicon carbide film side. The Pd deposits ranged from 0.1 mm in diameter to 1.5
mm. Up to four palladium deposits were applied to one silicon carbide sample. The silicon
carbide sensors were then mounted in a Flatpax sample holder and contacts were attached using
conductive adhesive, colloidal silver.
In order to establish a baseline for the studies of silicon carbide chemical sensors, the
current-voltage response of the
sensors was examined upon exposure
to three specific gases, hydrogen,
«V\J4
propylene and methane. Experiments
involved exposing the sensors to the
respective gases in both air and
W~-N^
nitrogen carrier gases. The sensors
2
were kept in ambient room air during
all of me tests. This enabled the
*A«/
establishment of preliminary baseline
responses to the gases under ambient
conditions, the conditions that may be
\
Z
closer to the actual conditions of
operational sensors. The responses to
the gases were carried out at room
temperature, 100, 200, 300 and 400°
\ z1
C. Responses at room temperature
were not significant and therefore will
00
500
600
700
, 600
900
1000
Time (sec)
not be discussed in this paper. The
response to methane was measured at
Figure 2 Current response to nitrogen at 200 °C.
200, 300 and 400° C. Both currentvoltage and current-time (at constant
voltage) measurements were obtained using a Keithley 2400 Source-meter interfaced to a
Pentium computer. All of the experiments are presented as current-time plots where the sensor
has been exposed to the various gases of interest. All of these measurements were performed
with a forward bias at 0.7 V.
y
RESULTS
The sensor response to hydrogen and nitrogen at 200 ° C is shown in Figure 1. The
measurements were obtained using 1000 ppm H2 in N2. The figure contains two plots, one for
pure N2 (99.999%) and one for the H2-N2 mix. As can be seen in the plots, there is a good
response to the H2 upon turning the
O
o
H2-N2mix on, with a rapid decrease in
current upon turning the gas off.
Three on-off cycles are shown in the
plot demonstrating reproducibility for
the response. The same process was
performed for the pure N2. As can be 3
200°C
seen in the plot, the behavior of the Ä
4-»
pure N2 did not follow that of the H2- c
N2 mix. There was a change in 2
current observed for the pure N2.
However, there was a slight decrease Ü
in current rather then the increase that
occurs with the H2-N2 mix.
We
attribute this change to thermal
fluctuations of the sensor when the N2
gas is switched on and off. Both the
N2 and the H2-N2mix were done at the
Time (sec)
same flow rates, therefore the
magnitude of the decrease in current
Figure 3 Current response to H2 at 100 ° C, 200'
C and 300 ° C
124
from the thermal fluctuations due
c
0
to gas flow are insignificant when
£
compared to the increase in the
t=
o
current from the Schottky sensor
0.
i
response to H2. In Figure 2, the
sensor response to pure N, is
a.
^-^_^^
plotted and scaled by itself. The
plot shows the decrease in current
the N, is turned on and the -K
subsequent increase in current
when the gas is turned off. The
magnitude of the decrease is O
around 10 microamps, while the
magnitude of the H2 response is
200 °C
an increase of 30 miliamps.
In
order to establish the behavior of
1
F
1
the sensor H2 response to
temperature, I-t plots at 100, 200
Time (sec)
and 300 ° C are shown in Figure
Figure 4 Current response to propylene at 200 °C and
3. As can be seen in the figure,
the sensor shows an increase in
300 °C.
sensitivity
with
increasing
temperature consistent with published results by Hunter et-al [3,4].
The current-time response to propylene was measured at 100, 200 and 300° C.
Consistent with the results of Chen et-al [5], the measurement at 100 ° C was too low for a
sensor response. The plots at 200 °C and 300 °C are shown in Figure 4, and clearly exhibit a
temperature dependent increase in current with exposure to propylene. This may be attributed to
several factors including sensor preparation, contacts for the leads and the method of gas
dosing. Our experiments were carried out in ambient air, while the Chen work involved
systematic purging of the sensor-sampling chamber in air and nitrogen prior to exposure.
In order to understand the nature of the sensor response to methane, both current time
measurements and x-ray photoelectron spectroscopy experiments were performed on silicon
carbide sensors. As reported in the work by Chen, the response to methane did not follow the
behavior of H2 or the other two
5
hydrocarbons examined in their
__—-— tJ
work.
They attributed the
decreased response to poisoning
300° C
of thePd surface by the methane. ^ s
Pronounced
responses
to 3
propylene and ethylene were «
observed in their study for
temperatures ranging from 200- c
o
400 ° C, however the methane
had an initial increase in current 3
/
c
followed by a gradual decrease Ü
s
down to the original baseline.
2. /
*
u
The responses to methane
400°C
at 300 °C and 400 °C observed in
S
\"'"
>.
our study are shown in Figure 5.
i
1
'
i
i
The response at 300 ° C reveals
Time (sec)
that, upon turning the methane
on, there is an initial fast response
Figure 5 Current response to methane at 300° C and
followed by a decrease as with
400° C.
the Chen study. However in our
work, the decrease is immediately followed by a gradual increase in current over a period of a
few thousand seconds. In Figure 5, it is seen that the current continues to increase until at
I
SOU
OUU
__.
Propylene off
7 Propylene on
o
o
1
IUU.
.
«»•
turn
L
1
125
around 2400 seconds where the methane is turned off. Upon switching the methane flow off,
there is a rapid decrease as the current returns to near the original baseline.
Unlike the response to H2 where there is a rapid increase up to saturation, followed by
generally a constant current, with the methane, there is the initial response, decrease, and then a
gradual increase in current. Saturation does not occur immediately, but depending on the sensor
temperature may take several minutes to occur. This process appears to involve surface
reactions on the Pd metal, rather than the silicon carbide, and is dependent on the temperature
and the concentration of methane vapor that is near the surface that subsequently adsorbs and/or
reacts on the Pd
3. XPS Study of CH4 adsorption on Pd/silicon carbide
In our experiments, exposure to methane was carried out in ambient air, and as such, CO,
C02 and 02 were present to participate in surface reactions at the Pd surface. Several recent
studies have addressed the catalytic behavior of Pd upon methane adsorption and discuss the
reaction processes that occur in the presence of these gases [8-11].
In an effort to examine the surface
chemistry for CH4 interactions with Pd,
XPS was performed on Pd/silicon
carbide sensors prior to and after
exposure to CH4. Both the unexposed
and exposed samples were heated at
300°C for 30 min. In the case of the
exposed
sample,
methane
was
introduced during the heating process.
Figure 6 shows the XPS spectra of
these samples in the Cls region. The
Cls spectra are from the silicon carbide
surface of a CH4exposed sample,
shown in Figure 6(a), from the Pd
surface on a sample that was not
exposed to CH4, Figure 6(b) and from
the Pd surface for a CH4 exposed
sample, Figure 6(c). The peak shapes
on silicon carbide were the same for
both CH4 exposed and unexposed
samples.
The Cls XPS spectra in Figure 6
obtained from the silicon carbide
surface contains two peaks: silicon
carbide at around 282 eV and CHX at
around 284 eV.
The CHX peaks
observed on the silicon carbide surface
is not attributed to the CH4 sample gas;
rather it is due to pre-adsorbed carbon
Binding Energy feV]
from exposure to ambient conditions.
The Pd surface for both exposed and
Figure
6
XPS
spectra
in the Cls region for the (a)
unexposed samples contains at least
unexposed
surface,
(b)
unexposed
Pd/SiC Surface
three components: CHX at around 284
eV, CO at around 288 eV and another and (c) methane exposed Pd/SiC surface.
unassigned peak at around 292 eV.
The peak at 292 eV may correspond to C02, however, it has been reported that C02
dissociatively adsorbs on Pd in a temperature range of 200-400 °C [8]. Therefore, unless C02
adsorption on silicon carbide supported Pd results in a stable adsorbed C02 species, we will not
assign the 292 eV peak to C02. The spectra show the exposed sample with an obvious increase
in CO as well as the unassigned component at 292 eV. The heating process at 300° C in
126
ambient air therefore may have resulted in the dissociative adsorption of C02 and the resulting
CO peaks observed on the unexposed sample.
This process requires the presence of hydrogen and follows the reaction:
C02 + H2 ====>CO + H20
This implies that some adsorbed hydrogen was present on the unexposed sample for this
reaction to occur. The source of this hydrogen may be from the hydrocarbon layer that exists on
the sample from exposure to the ambient.
For the CH4 exposed sample, the occurrence of adsorbed CHX as a result of CH4
exposure is not evident in the XPS spectra since there was no obvious increase in the CHX XPS
peaks after exposure to CH4. There is however, a slight broadening of the Cls peak in the area
that CH, occurs and therefore some CHX is present. It has been reported that the adsorption of
CH4 on Pd when pre-adsorbed CO is present can result in a stabilized adsorbed CHX species
[9]. The CHX peaks observed in Figure 6 might also be due to the pre-adsorbed hydrocarbon
species observed on the unexposed sample that originates from the adsorption of carbon from
the ambient. The determination of the adsorption of a stable CHX species on Pd from methane
exposure requires more study in a controlled system where sample cleaning and gas dosing can
be performed in-situ.
It is evident in the XPS spectra that there is an increase in the occurrence of the oxidized
form of carbon, CO. The interaction of CH4 at the Pd surface is most likely then resulting in the
oxidation of the CH4 leaving the adsorbed carbon monoxide. This reaction has been reported to
occur as [8]:
C02 + CH4 =====> 2CO + 2H2
This reaction results in the format ion of hydrogen which should of course, in the absence of
competing processes, in turn dissociate and affect the surface potential at the Pd-SiC interface
leading to the increase in current
observed in the I-V behavior.
However,
the
experiment
as
performed occurs in ambient air with
an abundance of C02 present to induce
the oxidation of the CH4.
The
response to hydrogen therefore may
be inhibited due to the increase in
surface CO. This inhibited response
to the dissociated H2 from the CH4
molecule is seen as the spike followed
by the immediate decrease in current in
the I-t plot in Figure 5.
It is apparent that both the
carbon and oxygen XPS peaks should
give some indication of the overall
process. The XPS spectra for the
same samples involved in the carbon
XPS study were examined in the
oxygen region of the spectrum. These
are shown in Figure 7. Again, (a)
corresponds to the silicon carbide
region of the exposed sample while
Binding Energy |eV]
(b) is on the Pd of the unexposed
sample and (c) on the Pd of the CH4 Figure 7 XPS spectra in the Ols region for (a) the
methane exposed SiC, (b) the unexposed Pd/SiC
exposed sample.
The Ols peak occurs at around surface and (c) the exposed Pd/SiC surface.
533 eV while Pd has a peak
representing Pd3d very near the
127
oxygen peak at 537 eV. As would be expected, the spectrum on the silicon carbide shows no
Pd3d peak, while both of the samples on the Pd film has the Pd3d peak. There is a pronounced
difference in the peak shapes between the two Pd sample spectra. The Unexposed sample
shows both the Ols peak and the Pd3d peak, while the CH4 exposed sample has no Ols peak or
it has "vanished" below the background produced by the Pd3d peak. This indicates that
exposure to methane at 300 °C somehow involves the pre-adsorbed oxygen. In the absence of
CH4, heating the sample to 300 °C had no apparent affect on the Ols peak as seen with the
unexposed sample.
CONCLUSIONS
The surface reaction involving the adsorbed oxygen species is perhaps the mechanism that
leads to the gradual increase in current observed in methane-exposed SiC sensors over time.
This reaction is certainly temperature and time dependent and may follow kinetics that yields the
gradual increase in current. The oxidation of methane involving surface oxygen would result in
the increase in binding energy for Ols on the exposed films if the surface oxygen were due to
CO. This would bury the Ols peak in the Pd3d background. It is also reasonable to suggest
that the net effect of decreasing the surface oxygen from Pd-oxide may lead to the gradual
increase in current as seen in the I-t curve of Figure 5. Clearly, this is a bit speculative at this
point and a more detailed systematic study would provide the information needed to better
understand the mechanism for the observed behavior of the I-t curves for methane on the silicon
carbide sensors.
ACKNOWLEDGEMENTS
This work was supported by a grant from the NASA Lewis Research Center NAG3-2020.
REFERENCES
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
K. Shenai, R.S. Scott, B.J. Baliga, IEEE trans. Electron. Dev. 36(1989)1811.
J.R. Waldrop and R.W. Grant, Appl. Phys. Lett., 62(1993)2685.
G.W. Hunter, P.G. Neudeck, G.D. Jefferson, G.C. Madzsar, C.C. Liu and Q.H. Wu,
Report E-7773 NASA, (1993).
G.W. Hunter, P.G. Neudeck, C.C. Liu and Q.H. Wu, Conference on advanced EarthTo-Orbit Propulsion Technology, (1994).
L. -Y Chen, G.W. Hunter , P.G. Neudeck, D. Knight, C.C. Liu and Q.H. Wu,
Proceedings 190th Meeting of Electrochemical society (1996).
L. A. Spetz, A. Baranzahi, P. Tobias, I. Lundstrom, High temperature sensors based on
metal-insulator-silicon carbide devices, Physica Status Solidi (A) Applied Research, v
162, 1 (1997)493-511.
G. Müller, G. Krotz, E. Niemann, SiC for sensors and high-temperature electronics,
Sensors and Actuators, A, 43, 1-3 (1994) 259-268
A. Erdöhelyi, J. Cserenyi, E. Papp and F. Solymosi, Applied Catalysis A, 108 (1994)
205-219.
J. -J. Chen and N. Winograd, Surface Science 314 (1994) 188-200.
A.K. Bhattacharya, J.A. Breach, S. Chand, D. K. Ghorai, A. Hartridge, J. Keary and
K.K. Mallick, Applied Catalysis A: General, 80 (1992)L1-L5.
W. Lisowski, Surface Science, 312 (1994) 157-166.
128
THE EFFECT OF ANNEALING ON ARGON IMPLANTED EDGE TERMINATIONS
FOR 4H-SIC SCHOTTKY DIODES
A P KNIGHTS*, D J MORRISON**, N G WRIGHT**, C M JOHNSON**, A G O'NEILL**,
S ORTOLLAND** , K P HOMEWOOD*, M A LOURENCO*, R M GWILLIAM*, AND P G
COLEMAN***,
School of Electronic Engineering, Information Technology and Mathematics, University of
Surrey, Guildford GU2 5XH, United Kingdom, a.knights@ee.surrey.ac.uk
Department of Electrical and Electronic Engineering, University of Newcastle, Newcastleupon-Tyne, United Kingdom, NE1 7RU.
***School of Physics, University of East Anglia, Norwich, NR4 7TJ, United Kingdom.
ABSTRACT
The edge termination of SiC by the implantation of an inert ion species is used widely to
increase the breakdown voltage of high power devices. We report results of the edge termination
of Schottky barrier diodes using 30keV Ar+ ions with particular emphasis on the role of postimplant, relatively low temperature, annealing. The device leakage current measured at 100V is
increased from 2.5nA to 7uA by the implantation of 30keV Ar+ ions at a dose of lxl015cm"2. This
is reduced by two orders of magnitude following annealing at 600°C for 60 seconds, while a
breakdown voltage in excess of 750V is maintained. The thermal evolution of the defects
introduced by the implantation was monitored using positron annihilation spectroscopy (PAS) and
deep-level-transient spectroscopy (DLTS). While a concentration of open-volume defects in
excess of lxl019cm"3 is measured using PAS in all samples, electrically active trapping sites are
observed at concentrations ~lxl015cm3 using DLTS. The trap level is well-defined at Ec-E, =
0.9eV.
INTRODUCTION
SiC Schottky barrier diodes are ideal for power switching applications as they can be
operated at higher voltages and higher temperatures than equivalent Si or GaAs devices. The
requirements for a high power Schottky barrier diode are a low forward voltage drop, low reverse
leakage current and high breakdown voltage. Premature voltage breakdown often occurs because
of electric field crowding at the periphery of the device. In order to achieve high breakdown
voltages near to the theoretical limits expected for SiC, it is necessary to employ an edge
termination.
One technique used to terminate Schottky diodes is via the implantation of inert ions. The main
area of the device is protected using a mask and inert ions are implanted into the sample. This
process forms an area of high resistivity which allows the potential to spread across the surface of
the biased sample. Unfortunately, a significant and undesirable increase in leakage current
accompanies the increase in breakdown voltage. For example, devices exhibiting near-ideal
breakdown voltages on 6H-SiC have been reported using Ar+ ion implantation [1], however the
reverse leakage of these devices was 5xlO"2A/cm2 at <100V.
In this study we report an improvement in reverse leakage current of Ar+ implanted 4H-SiC
Schottky barrier diodes. Following implantation the devices are annealed at temperatures up to
600°C with a resulting decrease in the reverse leakage current of almost two orders of magnitude
while a breakdown voltage in excess of 750V is maintained.
129
Mat. Res. Soc. Symp. Proc. Vol. 572 ° 1999 Materials Research Society
In addition, the implantation induced defects have been monitored using positron annihilation
spectroscopy (PAS) and deep-level-transient spectroscopy (DLTS). These two complementary
techniques provide information on (1) the size and distribution of open-volume defects (such as
vacancies and voids), and (2) the position of the defect related mid-gap trap position respectively.
EXPERIMENTAL DETAILS
A ready-diced n-type 4H-SiC epitaxial wafer ( NM^ lxl018cm"3, NDepi=lxlO'6cm"3) supplied
by Cree Inc. was used in this study. Each 5mm by 5mm section was given a thorough solvent
clean using acetone and IPA. After a 5 minute soak in 10:90 HF:H20,1000Ä of Ni was deposited
on the highly doped backside and alloyed at 1000°C to form a large area backside ohmic contact.
Four samples were front-patterned with 312um diameter dots using standard photolithography
and were placed in 10:90 HF:H20 for 5 minutes immediately before the deposition of 1000A Ni
Schottky contacts. The samples were implanted at room temperature with lxlO1 cm" Ar+ ions at
30keV , with the contact pads acting as implant masks for the 30keV ions. Three samples were
subsequently annealed at either 400°C, 500°C, or 600°C for a duration of 60 seconds. The
remaining sample was left unannealed. The resulting diodes were characterised using forward and
reverse I-V measurements.
Four further samples were implanted and annealed under the same conditions, before the
fabrication of Schottky contact pads. These samples were first analysed using the entirely nondestructive technique of positron annihilation spectroscopy and subsequently, Ni Schottky contact
pads were deposited on top of the irradiated area and the samples were subjected to deep-leveltransient spectroscopy analysis.
RESULTS AND DISCUSSION
Current-Voltage Measurements
Figure 1 shows the I-V characteristics for
the Schottky diodes. The diodes exhibit
classic Schottky behavior with excellent
forward conduction. For the reversebiased, unimplanted sample, a leakage
current at 100V of 2.5nA is measured with
a subsequent increase to 7uA following
edge-termination via Ar+ implantation.
These values are consistent with previous
measurements [1]. Upon annealing, the
-60
-40
leakage current is seen to decrease
Applied bias (V)
significantly with increasing temperature.
Following an anneal at 600°C for 1 minute
it is reduced to a value of 0.09|iA At the
Figure 1 - I-V curves for reverse biased, edge-terminated
same time, the breakdown voltage for the Schottky diodes. The anneals were performed for 1 minute at
600°C device was found to be consistently
temperatures indicated on the plot.
greater than 750V, or -50% that of the
theoretical, ideal breakdown.
130
This improvement in leakage current is analogous to the situation found during the formation
of high resistivity regions in GaAs via implant isolation. Although high levels of resistivity can be
induced in doped GaAs following inert ion implantation, a low temperature anneal is required to
obtain maximum resistivity, close to the intrinsic value [2]. The anneal selectively removes
shallow levels responsible for hopping conduction leaving only the deep-levels required for
efficient isolation.
Positron Annihilation Spectroscopy (PAS)
The application of positron annihilation to the study of ion implantation induced defects has
been comprehensively described elsewhere [3]. Monoenergetic positrons were implanted into the
samples in the energy range 0-30keV and at each energy the Doppler-broadened annihilation
gamma photopeak linewidth was measured and described by the parameter S. Depth resolved
information is obtained by varying the energy of the incident positron beam with the mean depth
in Angstroms, z =130 E1-6, where E is the incident positron energy in keV. The implantation
profile becomes broader with increasing E; its FWHM is approximately equal to the mean
implantation depth.
The size of the samples (5mm x 5mm) required aperturing of the positron beam to 4mm
diameter immediately after leaving the source region. The samples were attached to thin tungsten
wires to minimise the surface area of
the backing material and thereby the
probability of detecting radiation from
positron annihilation at sites other than
1.10
o AS IMPLANTED
in the sample under study.
The
D 400°C
positron beam was centred on the
A 50O°C
1.08
V 600°C
target at each incident energy by direct
o UNIMPLANTED
observation of the beam and sample
tr 1.06
- FITS
in
shadow on a microchannelplateiphosphor screen-CCD camera assembly
| 1.04 L ^L ^L i»
at the end of the beam line. The total
1
<
photopeak count rate was ~ 800 s" and
*
<? %
*
V.
« 1.02
run times at each incident energy were
2-3000S.
1.00 Figure 2 shows the S parameter
L_L_L_
(representative of open-volume defect
0
0
0
0.98
size/concentration) versus incident
0 5 10 15 20 25 30 35 40
positron energy (depth) for the asINCIDENT POSITRON ENERGY (keV)
implanted and annealed samples,
together with the spectrum for an
Figure 2 - Positron S-parameter (open-volume defect
unimplanted
control
sample.
Information on the size and distribution size/concentration) versus incident positron energy (depth) for
implanted and annealed samples.
of the defects is obtained from fitted
models to the data shown as solid lines in the figure. All of the implanted samples show two
defected regions containing a large concentration of defects (>lxl019cm"3). The first defected
layer extends to a depth consistent with the implanted Ar+ ions of ~20nm, with a dominant
positron trap, saturated S parameter of >1.085. The second layer extends well beyond the range
of the ions to a depth of ~250nm. The saturated S parameter for these defects is -1.045. This type
131
of deep layer has also been observed in 200keV Ge+ implanted SiC [4] and probably results from
a combination of implanted ion channelling and defect diffusion.
In a recent publication Brauer et al [5] presented a plot illustrating the dependence of the S
parameter (normalised to its value, Sb, in undefected bulk SiC) on the size of the open-volume
defect in SiC in which a positron is trapped at the moment of annihilation. They found that for the
Si-C divacancy S/Sb ~ 1.05. This value rises to ~ 1.08 for a cluster of four divacancies, and
approaches an asymptotic value of ~ 1.15 for large clusters or voids. This would suggest that in
the present study the shallow layer comprises, in the main, vacancy clusters or voids, while the
deeper layer contains point-type defects such as divacancies.
Upon annealing, the larger sized (void) defects are observed to decrease in size/concentration.
However, the smaller, point-type defects are unchanged in either distribution, size or
concentration. If the open-volume defects are playing a significant role in the determination of the
electrical characteristics of the diodes it is likely that it is related to the break-up of the void
defects introduced up to the range of the implanted ions.
Deep-level-transient spectroscopy (DLTS)
Deep-level-transient spectroscopy (DLTS) experiments were performed under dark conditions
using a Bio-Rad DL4600 system. Samples were mounted on a stage in a liquid nitrogen cryostat.
The temperature was monitored using a platinum resistance thermometer attached directly to the
stage, giving an uncertainty of+0.5 K on the measured value. All DLTS spectra were taken twice
(ramping the temperature up and then down)
to account for temperature lag. Arrhenius
plots were obtained from the average values
1e-3
of the ramp-up and ramp-down peak
o As implanted
positions. Prior to the DLTS measurements
Annealed at 40O°C
capacitance-voltage (CV) and current-voltage
Annealed at 500°C
(IV) measurements were performed at
Annealed at 600°C
various temperatures, and standard Schottky
junction characteristics were obtained. The
IV and CV characteristics of all samples did
not show any significant differences.
DLTS spectra were obtained under
the same bias and pulse conditions (reverse
voltage, VR = -2.0 V; forward voltage, VF =
0.0 V; fill pulse, t = 2 ms) to enable direct
2.4
2.5
comparison between the different treated
1O00/T(K-1)
samples. Under these excitation conditions,
around 0.5 to 1 |J.m of bulk SiC layer is
sampled. For all samples, a typical DLTS
Figure 3 - Airhenius plots of the irradiation induced deep spectrum indicated the presence of a
level obtained from the as-implanted and annealed
dominant majority carrier trapping centre,
samples.
associated with a single exponential positive
peak present in the temperature range of 300 450 K, with activation energy around 900 meV
and apparent capture cross section of order of 10 s" K"
132
The samples in this study compare favourably with those from Alok et al. [6] where DLTS
measurements of similarly implanted SiC yielded traps between 0.2 and 0.8 leV below the
conduction band. A value of 900meV is consistent with efficient edge termination [6].
Figure 3 shows the Arrhenius plots of this deep level obtained from the as-implanted and
annealed samples. An estimation of the density of this trap gives a concentration of ~lxl015cnf3
for the as-implanted sample, reduced to 40% of this value following annealing. The same trap
concentration was obtained from all annealed samples and was independent of the annealing
temperature, within the experimental errors. Comparison with the high concentration of openvolume defects observed by PAS would suggest that if the carrier traps are vacancy-type, they
form a small subset of the open-volume defects present in the sample. The absence of a strong
dependence on annealing temperature of any shallow trapping sites makes it difficult to make a
direct correlation between the observed reduction in diode leakage current and the optimisation of
implant isolation of GaAs. Hence, it is implied that the removal of shallow defects by the relatively
low temperature anneal is not the explanation for the reduction in leakage current
CONCLUSIONS
We have described results of the edge termination of Schottky barrier diodes using 30keV Ar+
ions with particular emphasis on the role of post-implant, relatively low temperature, annealing.
The device leakage current measured at 100V was increased from 2.5nA to 7uA by the
implantation of 30keV Ar+ ions at a dose of lxl015cm"2, and reduced by two orders of magnitude
following annealing at 600°C for 60 seconds, while a breakdown voltage in excess of 750V was
maintained. Positron annihilation spectroscopy showed the concentration of open-volume defects
to be in excess of lxl019cm"3 in all samples. They are contained in two distinct defect bands. The
first is consistent with the implanted ion range and contains voids. The second extends to >10
times the ion range and is dominated by point-type defects. A reduction in void size/concentration
with annealing is correlated with the reduction in leakage current. Electrically active trapping sites
are observed at concentrations ~lxl015cm"3 using DLTS. The trap level is well-defined at Ec-Et =
0.9eV.
ACKNOWLEDGEMENTS
This work is supported as part of the SCEPTRE project under EPSRC grant no. GR/L62320.
REFERENCES
1. D Alok, B J Baliga, P K McLarty, IEEE Electron Device Letters, 15,394 (1994).
2. S J Pearton, International Journal of Modern Physics B., 7,4687 (1993).
3. P Asoka-Kumar, K G Lynn, and D O Welch, J. Appl. Phys., 76,4935 (1994).
4. G Brauer, W Anwand, P G Coleman, A P Knights, F Plazaola, Y Pacaud, W Skorupa, J
Stornier and P Willutski, Phys Rev B 54 3084 (1996)
133
5. G Brauer, W Anwand, P G Coleman, J Stoermer, F Plazaola, JM Campillo, Y Pacaud and W
Skorupa, J. Phys. Condens. Matter 10,1147 (1998)
6. D. Alok, B J Baliga, M Kothandaramam, and P K McLarty, Proceedings of Silicon Carbide
and related materials, Kyoto, Japan, (1995) pp.565-568.
134
OXIDATION MODELLING FOR SIC
N.G. WRIGHT, CM. JOHNSON AND A.G. O'NEILL
Dept. Of Electrical and Electronic Engineering, The University of Newcastle upon Tyne,
Newcastle UK, NE1 7RU, n.g.wright@ncl.ac.uk
ABSTRACT
A simple mechanistic model of the oxidation of SiC is presented and analysed using MonteCarlo simulation techniques. The model explains the observed anisotropic oxidation rate of
SiC in terms of the effect of weakening/strengthening of Si-C bonds arising from the ongoing incorporation of highly electronegative oxygen atoms into the crystal lattice. The
extraction of key process metrics (such as oxide thickness, interface roughness and oxide
defect density) from the Monte-Carlo simulations is discussed.
INTRODUCTION
High voltage MOS devices are one of the most attractive classes of switch in SiC
technology. Such devices offer the possibility of switching up to lOkV loads with low gate
drive requirements and low on-state losses. For good control and safety, normally-off
devices are generally required in power switching applications and so there has been
substantial international research into developing good quality SiÜ2 layers on SiC for use in
inversion/accumulation mode SiC power MOSFETs. To date attempts at producing such
oxide layers have not been particularly successful. Oxides grown on SiC (both the 4H and
6H polytypes) exhibit high fixed charge densities and poor oxide-semiconductor interfaces
with significant roughness [1]. Systematic study of oxidation of the 6H and 4H-SiC
polytypes by various groups has however produced a wealth of information about oxidation
processes in SiC [2]. For example, it is now well established that different crystal faces of
SiC oxidise at different rates resulting (for example) in uneven oxide thickness around an
etched trench [3, 4]. The lowest/highest oxidation rates are observed on the so-called
silicon/carbon faces (the 0001/0001 planes respectively) with a corresponding increase in
oxidation rate for planes between the two extremes. Dependency of oxidation rate on
crystallographic plane is also observed in silicon where it is often explained by arguments
based on the number of silicon-silicon bonds exposed to various crystal faces. As
0001/0001 planes have the same number of surface bonds but widely differing oxidation
rates, such an explanation is not sufficient for SiC. This paper presents recent work in the
development of a simple oxidation model for SiC proposed by the authors in an earlier
paper [5]. The consequences of the model are explored using Monte Carlo based simulation
techniques and new results on the oxidation of trench structures presented.
THEORY
Mechanistic Oxidation Model
As full details of the proposed oxidation model have been presented in an earlier paper [5],
we present only a brief discussion of the main issues here. The crystal structure and
proposed bonding of 4H-SiC is shown schematically in Figure 1 (with the z-axis scale
greatly exaggerated for clarity). From the point of view of the proposed oxidation model,
the important structural characteristic of all the hexagonal SiC structures is the absence of
135
Mat. Res. Soc. Symp. Proc. Vol. 572 ®1999 Materials Research Society
inversion symmetry along the z-axis of the chemical bonding between Si and C atoms. To
explore the consequences of this on oxidation, consider oxygen molecules interacting with
the crystal structure at the Si-face.
The first stage of oxidation is oxygen atoms bonding to the Si atoms outside the crystal. The
high electronegativity of the oxygen atoms will make the bond between the surface Si atoms
and the back-surface C atoms (labelled a in Figure 1) less ionic and hence lower the bond
energy (see Figure 2) [6]. At the second stage of oxidation, the a-bond will be broken by
the incoming oxygen resulting in a Si-O-C bond. The high electronegativity of the oxygen
atom
now in the a-bond position will increase the
ionic nature of C-Si ß-bonds (see Figure 2)
which will be consequently strengthened.
C-Face
Further oxidation proceeds via attack of these
strengthened ß-bonds - a process which will be
slower than the oxidation of the weakened abond. Once the bonds around the carbon atom
are fully oxidised, the release of a CO
molecule can then occur - with subsequent reordering of the Si-0 bonding structure. At the
Si-face then, there are one weakened and three
strengthened bonds resulting in an overall
C
increase in the amount of energy that must be
supplied by the oxidation process (i.e. the
Si
average activation energy of the process is .
increased). Subsequent oxidation of other
layers can proceed by the same mechanism
with CO being released from the structure as
the oxidation of the bonds around each C atom
is complete.
At the C-face however, such a mechanism
results in three weakened and one strengthened
bond resulting in a lowering of the activation
energy of the oxidation process. At other
crystal faces, the proportion of weakened to
strengthened bonds will be between these two
extremes producing consequent intermediate
oxidation rates. Adapting the model for other
oxidising ambients (e.g. wet oxidation) is
achieved by using chemical bond data to
modify the reactions assumed to occur during
oxidation
Monte Carlo Approach
The proposed oxidation model has been
implemented in the simulator, OXYSIM,
which is based on using a Monte Carlo type
approach to model the oxidation process at an
136
Si-Face
Figure 1: a schematic diagram of the
crystal structure of 4H-SiC in which the
z-axis (shown vertical in the diagram) is
greatly exaggerated. The layer nature of
the crystal is obvious with each layer
corresponding approximately to one of
the
three
stacking
positions
(traditionally
labelled
ABC)
of
hexagonal layer structures. The 4H-SiC
structure shown is given by the stacking
sequence ABCB with other polytypes
built up from different stacking
sequences (e.g. 6H-SiC = ABCACB).
atomic level. OXYSIM works by considering the interaction between the semiconductor
crystal lattice and any oxidising species (note: the model can be adapted to all common
oxidising reactants. For simplicity of explanation however, the description is given in terms
of dry oxidation, i.e. oxygen atoms reacting with the semiconductor). After the structure of
the lattice has been initialised by placing atoms in their correct position and setting relevant
bond energies, oxygen atoms (in this case) enter the lattice on calculated trajectories. The
oxidation temperature of the simulated process can be used to determine the number and
trajectories of the incoming oxygen atoms according to the well-known laws of statistical
mechanics governing gases. Those oxygen atoms that come within pre-defined distances of
lattice atoms are considered as possible candidates for interaction with the lattice atoms to
form bonds (and thus disrupt the original lattice). The program then uses a simple Monte
Carlo type decision process to decide whether the candidate interaction is allowed [7, 8]. The
probability of interaction, Pjnt, is calculated using the formula:
Pint=exp[-Ebond/kT]
(1)
where Ebond is the energy of the lattice bond, k is the Boltzmann constant and T is the
temperature (in Kelvin). The decision to allow or not allow the interaction is then made by
comparing this probability to a random number and acting according to the criterion:
Pint > random number => interaction allowed
(2)
Pint < random number => interaction not allowed
(3)
If the interaction is allowed then the oxygen atom enters the lattice and the bonding and
structure of the lattice modified according to pre-set structural rules (as specified in the model
described below). The process then continues for a specified time/number of incoming
oxygen atoms (similarly if the original interaction is not allowed).
Following the completion of the oxidation process, the resulting structure is analysed to
determine a number of simple process metrics. The thickness of the oxide layer at a given
point in the lattice is determined by simply calculating the perpendicular distance from the
lattice surface to the oxygen atom furthest from the surface. The mean roughness of the
oxide/lattice interface is defined to be the standard deviation of the thickness of the oxide at a
given point from the mean oxide thickness across the whole lattice. Information about the
defect density of the simulated oxide layer can be extracted by examining the bonding of both
the original lattice atoms and the incorporated oxygen atoms. The density and nature of misbonding can then be used to predict the density of electrically active defects from published
models of trap formation [9].
RESULTS
A quantitative test of the proposed model can be made by comparing the effect of the
proposed bond weakening with the activation energies for oxidation extracted from
experimental data using a linear rate model. Figure 2 shows such a comparison for wet
oxidation of SiC in which the surprisingly good agreement between the proposed model and
observed data can be clearly seen (despite the obvious simplifications of the model). Such
information could be of great importance in the process optimisation of SiC trench devices
137
*
such as UMOSFETs (in which a gate
/\ i
oxide is grown on the exposed
3.8 surfaces of a deep trench). Under the
3.6 high field conditions of a SiC
>
UMOSFET, the high electric field at
Ä 3.4>>
the trench corner can cause device
B
^ 32 breakdown in the oxide layer (c.f.
/
c
breakdown at the device periphery in
UJ
3
c
well designed silicon devices). This
2 2.8 I 1
/
can of course be alleviated by
IS
growing a thicker oxide but only at
£ 2.6 the expense of increasing the device
*2A.
threshold voltage (for a given doping
level in the SiC). As illustrated by
2.2 Figure 2, oxide thickness will be far
o
from uniform around an etched
trench and so device optimisation can
() 30 60 90 120 150 180
be hindered by the conflicting
off-angle from C-face
requirements of requiring a thick
oxide at the trench corner (to
increase device breakdown voltage) Figure 2: a graph illustrating a comparison
with the desire for a thin oxide on the between the observed dependency of activation
trench side-wall (for an acceptable energy (extracted from a linear oxidation rate
threshold voltage). In fact, this model) with angle of crystallographic plane
requirement is of key importance in (from the C-Face) and that predicted by the
deciding whether to fabricate proposed model (shown as solid line).
UMOSFETs on the silicon face Experimental data from [3] symbol and [4] •
(0001) or the carbon face (0001). symbol.
The carbon face produces a trench
oxide will the required thin/thick
characteristics on the side/bottom walls of the trench but generally results in a lower quality
oxide/SiC interface compared to oxides grown on the silicon face [10]. In order to examine
whether a trench oxidation process on the silicon face could be optimised to reduce the
differences in thickness between the side and bottom walls of the trench (and thus comer
closer to the desirable characteristics of the carbon face trench), we have examined the
relative oxidation rates of different crystallographic faces as a function of temperature.
A*
Figure 3 shows data from simulating the relative dry oxidation rates of the silicon and carbon
faces between 800K and 1500K (the approximate limits on practical oxidation processes).
The relative oxidation rates of the C- and Si-faces are clearly temperature dependent (in
agreement with experiment [3, 4]) - offering the potential for process optimisation according
to the desired criteria. Such data suggests the oxidation temperature plays an important role in
determining the relative thickness of the side-wall oxide compared to the oxide on the trench
bottom. The effect of process temperature on the predicted oxide shape in a 1.5um wide by
1.5um deep trench is explored in figure 4. For an oxide grown on a trench etched into a
silicon face wafer, the anisotropy in the oxide thickness of the side-wall and trench bottom is
predicted to fall with temperature (although the mean oxidation rate will of course rise). This
suggests that a short oxidation time at high temperature is the optimal process condition and
will produce the most even oxidation thickness around the trench in such a device.
138
Figure 3: a graph showing the simulated relative oxidation
rates of the C- and Si-faces (Re and RSJ respectively) as a
function of temperature (K).
1
:
I
2.5
2
1.5-
;_;...
1 -
0
1
2
3
Figure 4: Oxidation of a 1.5(im wide x 1.5p.m deep trench
etched into the silicon face of 4H-SiC. The solid line
represents the original trench shape, the dotted/dashed lines
the resulting oxide/SiC interfaces after oxidation at 1200K
and 1500K respectively.
CONCLUSION
A mechanistic model for the oxidation of 4H-SiC has been presented and explored using both
a simple analytic approach and the OXYSIM simulator. It has been shown that the model
reproduces well the anisotropic oxidation rates observed in 4H-SiC and, when used in
conjunction with OXYSIM, can predict important oxide quality metrics such as oxide
thickness, interface roughness and defect density. Such an approach can thus offer insights
into process optimisation and be useful in improving 4H-SiC oxide quality.
ACKNOWLEDGEMENTS
This work was supported by the UK Engineering and Physical Sciences Research Council.
The authors would also like to thank the sponsors of the SCEPTRE project for their generous
contributions to this work.
139
REFERENCES
1. J.A.Cooper Jr., Phys. Stat. Sol. 162, p. 305 (1997)
2. D. Alok, P. K. McLarty and B. J. Baliga, Appl. Phys Lett, 64, p. 2845 (194)
3. K. Ueno, Phys. Stat. Sol. 162, p. 290 (1997)
4. A. Rhys, N. Singh and M. Cameron, J. Electrochem. Soc 142, p. 1318 (1995)
5. N.G. Wright, C. M. Johnson and A.G. O'Neill, Mat. Sei. Eng. B56 (1999).
6. L. Pauling, "The Nature of the Chemical Bond" 3rd ed., Cornell University Press, New
York 1960
7. H.J. Herrman in "The Monte-Carlo Method in Condensed Matter Physics" (e.d. K.Binder),
Springer 1995
8 G.Bhanot,RepProgPhys51,p.429(1988)
9. V. F. Afanasev, M. Bassler, G. Pensl and M. Schulz, Phys. Stat. Sol. 162, p. 321 (1997)
10 S.Onda, R. Kumar and K. Hara, Phys. Stat. Sol. 162, p. 369 (1997)
140
ANNEALING EFFECTS OF SCHOTTKY CONTACTS ON THE
CHARACTERISTICS OF 4H-SIC SCHOTTKY
BARRIER DIODES
S.C. KANG, B.H. KUM, SJ. Do*, J.H. Je*, and M.W. SHIN
Department of Ceramic Materials Engineering, Myong Ji University 38-2 Nam-Dong, Yongin-Si,
Kyunggi-Do, Korea 449-728 Phone: 82-335-330-6465 Fax: 82-335-33-6457
Department of Materials Science and Engineering, Pohang Institute of Science and Tehnology P
O. BOX 125, POHANG 790-600, Korea. Tel: 82-562-79-2139 Fax: 82-562-79-2399
ABSTRACT
This paper reports on the relationship between the microstructure and the device performance
of Pt/4H-SiC schottky barrier diodes ( SBDs). The evolution of microstructure in the metal/SiC
interfaces annealed at different temperatures was characterized using X-ray scattering techniques.
The reverse characteristics of the devices were degraded with annealing temperatures. The
maximum breakdown voltages of as-deposited devices and 850 C annealed devices are 1300 V
and 626 V, respectively. However, the forward characteristics of the devices were found out to
improve with annealing temperatures. X-ray scattering analysis showed that Pt-silicides were
formed by annealing performed at or higher than 650 °C. The formation of suicides was shown
to increase the roughness of the Pt/SiC interface. It is believed that the forward characteristics of
the SBDs be strongly dependent on the crystallity of suicides formed in the Pt/SiC interface
during the annealing process.
INTRODUCTION
SiC has been given significant attention as a potential material for high-frequency, high-power,
and high-temperature applications due to its unique electrical and thermal properties. These
properties include a high electric field at breakdown (2 x 106 V/cm), a high electron velocity (2 x
107 cm/sec), a large band gap (2. 86 eV for 6H and 3.2 eV for 4H), and a high thermal
conductivity (4 W/K cm) [1]. In particular, the extremely high critical electric field of SiC makes
it a prime candidate for high-voltage applications, such as high-power rectifiers. Rectifiers utilize
SBDs to suppress high-voltage transients induced on the power line during current switching [2].
For a negligible dissipation of power during the switching, the reverse current transient of the
SBD must be suppressed, maintaining a high reverse voltage without breakdown. There have
been a lot of reports on the design and the fabrication of a SiC SBD to achieve its theoretical
breakdown voltage [3]. However, there are only a few reports found on the relationship between
141
Mat. Res. Soc. Symp. Proc. Vol. 572 • 1999 Materials Research Society
the microstructure and the device performance of 4H-SiC SBDs. In this study, we attempted to
establish the relationship between the current-voltage characteristics of Pt/4H-SiC SBDs and the
microstructure of the Pt/4H-SiC interface annealed at various temperatures. It was found out that
the annealing conditions for the schottky contact have a significant impact on the device
performance. The evolution of microstructure in the Pt/4H-SiC interface was characterized using
X-ray scattering techniques.
EXPERIMENT
Single crystal 4H-SiC wafer with a nitrogen doped (ND ~ 1.2 x 1016/cm3) epi-layer with 10 pm
thickness was used to fabricate the SBDs. The substrates were cleaned according to the standard
chemical cleaning procedure. The processing details for the fabrication of 4H-SiC SBDs are
found elsewhere [ Baliga ] except that Al shadow mask was employed as the shadow mask for
the Boron implantation [4]. Pt schottky contact ( t=3000 Ä ) was deposited by the sputtering
method through a metal shadow mask in a vacuum ( ~ 1 x 10"6 torr ). The implant dose and
energy were 1.0 xio15 cm"2 and 30 keV, respectively. Ohmic contacts were formed by
evaporation of Ni (t=3000 Ä ) in a vacuum( 7~9X 10"6 torr ) for the backside blanket. The
ohmic contact was annealed at around 1050 V, for 30 min in Ar ambient to remove the implant
damage. To investigate the annelaing effects on the device performance, the devices were
annealed at various temperature ( RT, 500 t, 650 t:, 750 "C, and 800 "C ) right after the
formation of schottky contact. The microstructure of the Pt/SiC interfaces was characterized
using X-ray scattering techniques. The I-V characteristics of device were measured using a Sony
tektronix 370 programmable curve tracer.
RESULTS AND DISCUSSIONS
The structure of the fabricated device and its typical current-voltage characteristics are shown
in Figure 1. The maximum blocking
Forward Voltage [ V ]
voltage obtained through this study
was 1300 V ( for samples with as_.
Boi or Implanted layer
deposited schottky contact).
1 "
ITTTT|
<
40
EDKIXIII
"
NI
C
2
110 urn
drift livir
N*5Ubstr»te(4H-SIC)
.
0
-
VB-!300V
-1 250
-1000
-750
-500
-250
0
Reverse Voltage [ V ]
142
Figure 1. Schematic device structure
of 4H-SiC SBD and its typical I-V
characteristics showing the maximum
breakdown voltage of 1300 V (device
with as-deposited schottky contact)
850 TC Annealed
A
750 Xi Annealed
•
As-deposited
V
650 t! Annealed
T
*
IE
T
E
\mrnt,
a
c
a
c
;»
u
As-deposited
:
75013
A
3
680 T):
»A
T
»A
850t;.
»A
»A
»A
I :
£
\\
•
LA.
5
r-
/
M»"
10
..-•
15
20
200
Forward Voltage[V]
•
;
'
•
400
•
/
'" '
•
600
i
'
800
i
1000
.
i
.
1200
i
1400
.
i
1600
Reverse Voltage [ V ]
(a)
(b)
Figure 2. The variation of forward and reverse bias I-V characteristics of Pt/SiC schottky barrier
diodes annealed at different temperatures; (a) forward and (b) reverse characteristics
Figures 2 (a) and (b) show the variation of forward and reverse bias I-V characteristics of Pt/SiC
schottky barrier diodes annealed at different temperatures. It is shown that the forward current
density of diodes is enhanced with the annealing temperature. The current density of devices
annealed at 850 t is as high as 420 mA/cm2 at 5 V. The current density of devices with the asdeposited schottky contact is about 140 times lower than the value for the samples that were
annealed at 850 1C. Apart from the forward characteristics, the breakdown voltage of devices is
found out to decrease with the annealing temperature. The ideality factor of the samples annealed
were shown to vary in a range 1.2 (850 "C) to 4.5 (as-deposited). The barrier heights are
expected to be higher for samples annealed at lower temperature. The degradation of breakdown
voltage of devices annealed at high temperature can be explained by the calculated specific-on
resistance. Figure 3 compares the specific-on resistance (R^) of as-deposited samples and of
samples annealed at 850 IC. The distribution of specific-on resistance of as-deposited samples is
shown to be about 1 order higher than that of annealed samples. It is known that the breakdown
voltage is reversely proportional to R^,, [5].
The electrical characteristics of devices are examined in view of the microstructure of the
interface between schottky contact and 4H-SiC. Figure 4 shows the intensity of reflected X-ray
143
from the surface and interface of Pt/4H-SiC layers annealed at different temperatures.
10-1
10
•
•
• •
•
Eo
o
c
a
•
• •
o
o
c
&
w
"5
102
«
Q>
9
o
o
IE
"5
"E
u
10J
'Ö
Before annealing
(As-deposited)
(0
100
a>
a
After annealing
(at 850 °C, 20 min, Ar)
(0
'
10"
100
1000
i—
1000
Breakdown voltage [ V ]
Breakdown voltage [ V ]
(b)
(a)
Figure 3. Breakdown dependence of specific-on resistance; (a) as-deposited and (b) annealed at
850 r.
It is evident that the roughness of the interface is
suddenly increased for the samples annealed at
temperatures higher than 650 °C. The higher
reflectivity for the samples annealed at 500 X^
compared to the as-deposited samples can be
attributed to the improvement of crystallity of Pt.
From X-ray scattering analysis it was shown that Pt
silicides form at temperatures higher than 650 °C,
while the crystallity of Pt itself is improved with
temperature below 650 V.
Figure 4. Intensity of reflected X-ray from the
surface and interface of Pt/4H-SiC layers annealed
Two-Theta[20]
at different temperatures.
144
FWHM=2.93
FWHM=2.45
o
e>
<P
dP
o
"cT
o
5
>«
a
4000
«c
°
o
%
e
°
a
°
%
o
As-deposited
o
o
o
o
25
26
27
o
!
o
o o
a
c
o
i
°o
— 2000000
§ °
o
o
o
o
o
o
o
o
o
o
a
«
X
J V
| at 6501=
III
24
o°
o °
[ at SOP'S
J
0
o
o °
o
B
8
2000
o
o
o o
°
°
o
o
rf>
o °
o
0
Oo
o
o
o
<&
o
o
o
o°
O
o o
o o
o °
c
3
O
o
°o
0
oo
o°
o
6000
%
oo
&
°S
BOOO
FWHM=3.29
FWHM=3.34
FWHM = 1.8S
o
°oo
o
o
o
Ju
§
| at 750 T5
o
n
at 850 T)
l"""""l^^^^
28212223242528272029302122232425262728293031
26
27
28
28 22 23 24 25 26 27 28 29 30 31
Theta(degrae)
Theta(degree)
(b)
(a)
Figure 5. Rocking curves of (a) Pt (111) diffraction when the Pt/4H-SiC layer are annealed at
different temperatures (as-deposited, 500 1C annealed, and 650 "C annealed) and (b) Pt2Si
annealed at 750 t and 850 "C.
Figure 5 (a) compares the rocking curves of Pt (111) diffraction when the Pt/4H-SiC layer are
annealed at different temperatures (as-deposited, 500 "C annealed, and 650 "C annealed). Fig.
5 (b) shows the rocking curves for Pt2Si. Long-scan X-ray diffraction exhibited that there is no Pt
phase left after annealing the sample at 850 "C [6]. It is concluded that the higher current
density of samples annealed at higher temperature stem from the better crystallity of Pt (below
650 t;) and Pt2Si (above 650 "C). It is worthwhile noting that the formation of silicides results
in the increase of roughness of Pt/4H-SiC interface [7]. This can be easily understood by
comparing Figure 4 and Figure 5. The forward and reverse characteristics in Pt/4H-SiC SBDs are
found to be dominated by the interface state which are controlled by the thermal annealing.
CONCLUSIONS
Electrical characteristics of Pt/4H-SiC SBDs were interpreted in light of the evolution of
microstructure of the Pt/4H-SiC interface by using X-ray scattering techniques. Devices with
the as-deposited Pt schottky contact exhibited the maximum breakdown voltage of 1300 V. The
reverse characteristics of the devices were degraded with annealing temperatures. It was shown
145
that the forward characteristics of the devices improve with annealing temperatures. X-ray
scattering analysis showed that Pt-silicides were formed by annealing at or higher than 650 "C.
The formation of silicides was shown to increase the roughness of the Pt/SiC interface. The
forward and reverse characteristics in Pt/4H-SiC SBDs are found to be dominated by the
interface state which are controlled by the thermal annealing.
ACKNOWLEDGMENTS
The authors wish to acknowledge the financial support of the Korea Research Foundation
made in the program year of 1998.
REFERENCES
1. Charles E. Weitzel, John W. Palmour, Calvin H. Carter, Jr., Karen Moore, Kevin J. Nordquist,
Scott Allen, Christine Thero, and Mohtt Bhatnagar, IEEE transactions on Electron Devices
41(10), 1732-1741, (1996)
2. Iver Lauermann, Rudiger Memming, and Dieter Meissner, J. Electronchem. Soc, 144(1),
p.73-80. (1997)
3. A. Itoh, T. Kimoto, and H. Matsunami., Member, IEEE, IEEE Electron Device letters, 16(6), p.
280-282. (1995)
4. Dev. Alok and B.J. Baliga, Proceeding of 1995 International Symposium on Power
Semiconductor Devices & Ics, p. 96-100, (1995)
5. J. Crofton, E. D. Luckowski, et al, Inst. Phys. Conf. Sen No. 142 : Chapter 3, Paper presented
at Silicon Carbide and Related Materials 1995 Conf. Kyoto, Japan, (1995)
6. L.M. Porter, RF. Davis, J.S. Bow, M.J. Kim, and R.W.Carpenter, Inst. Phys. Conf. Ser. No 137,
p. 581-584,(1993)
7. N. A. Papanicolaou, A. Christou, and M. L. Gipe, J. Appl. phys. 65(9), p. 3526-3530, (1989)
146
Part II
SiC Epitaxy and
Characterization
EPITAXIAL GROWTH OF SIC IN A VERTICAL MULTI-WAFER CVD
SYSTEM: ALREADY SUITED AS PRODUCTION PROCESS?
Roland Rupp,' Christian Hecht,2 Arno Wiedenhofer2, and Dietrich Stephani2
'Siemens AG, Semiconductor Components Group, HL PS E SiC, D-80312 Munich,
Germany
2
Siemens AG, Corporate Technology, Department ZT EN, Box 3220, D-91050 Erlangen,
Germany
ABSTRACT
Results about a new CVD system suited for epitaxial growth on six 2 inch SiC-wafers at a
time are presented. Excellent gas flow stability is achieved for this new reactor type as shown by
in- situ observations of the gas flow dynamics in the reactor chamber. These experimental results
agree favorably with numerical process simulation results.
The epitaxial layers grown in the multi-wafer system so far show a by an order of
magnitude higher background impurity level (< 1015 cm'3) as reported previously for layers
grown in single-wafer systems by the authors and other groups (< 1014 cm"3). On the other hand,
the doping homogeneity achieved until today is very encouraging. The variation on a 2 inch
wafer is less than ± 20% at about 1*1016 cm"3. The wafer to wafer variation of the average
doping value both within a run and from run to run is within 15 %. The reproducibility and
uniformity of the layer thickness is even better (total thickness variation < 5% on a 2 inch wafer).
The surface of the epitaxial layers is very smooth with a typical growth step height of 0.5 nm
(4H, 8° off orientation). First measurements on Schottky diodes build on these layers show low
leakage current values indicating low point defect density in the epitaxial layers.
INTRODUCTION
In the last 10 years significant progress in SiC epitaxial growth took place. Advancements
cover control, reproducibility and homogeneity of doping and thickness but also background
doping. This was enabled by better understanding of the deposition process (step control [1],
influence of graphite parts [2], site competition [3,4]) and by the development of commercially
available epitaxial equipment [5,6,7], which allows accurate control of the relevant process
parameters. Nevertheless the costs of the epitaxial process - apart from the still extremely high
wafer prices - are a major drawback for a wide range commercialization of SiC devices. This
would hold even in the case of zero micropipe density. A pragmatic way to achieve a substantial
potential in cost reduction is the use of multiple-wafer instead of single-wafer processing. On
the other hand, this leads to new challenges in direction of process control and homogeneity
adjustment.
Today there exist three major commercial suppliers for SiC epi systems, all using different
basic setups. These companies are EPIGRESS in Sweden (hot wall tube reactor [6]), AIXTRON
in Germany (planetary reactor with independent rotation of each wafer [5]) and EMCORE in
New Jersey, USA (vertical cold wall reactor [7]). To the knowledge of the authors multi-wafer
epitaxial processing of SiC is only reported either on AIXTRON ([8], seven 2" wafers) or on
EMCORE ([9], six 2" wafers) systems.
149
Mat. Res. Soc. Symp. Proc. Vol. 572 ® 1999 Materials Research Society
In the following paper we shall describe how and how far we have been able to solve the
difficulties of this technology and to report about growth results recently achieved with this tool.
A comparison will be made with results formerly achieved with a single-wafer system.
EXPERIMENTAL
The principal construction of the SiC multi-wafer CVD-system is based on the EMCORE
D180 type reactors besides a special chamber and heater design necessary for the high
temperature operation (see Fig. la,b), i. e. an RF-heating of the wafers is applied, which allows
wafer temperatures up to 1600°C.
A palladium cell is used to provide high purity hydrogen to the process and a loadlock
equipped with a turbo pump is attached to the process chamber to reduce the unproductive time
between processes. Silane and propane (2% / 5% diluted in hydrogen) are used as reactive gases
added to the main gas stream of hydrogen. Controlled n-type doping is achieved by feeding
nitrogen to the process gas. Intentional p-doping is not applied in the system to avoid memory
effects and to enable the growth of low nitrogen doped layers with compensation as small as
possible. The process temperature is measured by 3 optical pyrometers pointing at different
radial positions and allowing control of the temperature homogeneity across the wafer (AT
typically < 10 K across 2').
Nitrogen
Nitrogen
- gas diffusor
water cooled
" reactor wall
(stainless steel)
confined inlet
(see text
-Wafer
susceptor
* wafer fixation
tfÄäSs^
to the exhaust
electrical and rotating
feedthrough
electrical feedthrough
rotating feedthrough
to the exhaust
Fig. 1: Sketch of the EMCORE vertical epi reactor conception
a) single-wafer system [10,17]
b) multi-wafer system
The inner diameter of the reactor chamber is reduced in the upper part (see Fig. lb) by a
water cooled insert attached to the top flange („confined inlet"). In this way, the amount of
hydrogen flow necessary to adjust stable flow without recirculation can be decreased
significantly. This is mainly due to an increased flow velocity in this area with reduced cross
section.
The RF-coil is exposed to the process atmosphere and therefore a condensation of undefined
Si/C compounds from the process gas mixture occurs at the coil, primarily determining the
maintenance cycles by the need for coil cleaning. Presently the length of these cycles is 80-100
hours of growth time (corresponding to 300-400 jxm total growth).
150
The RF-susceptor / wafer holder assembly consists of a Mo-Plate with six SiC-coated graphite
pies as inserts, which have pockets for placing the wafers in it. The whole assembly is displayed
in Fig. 2.
"""
"'""''
Fig. 2:
RF-susceptor / wafer holder assembly
used in this study. The total diameter of
the Mo-plate is 185 mm.
SICco;
^graphltlffl
4H-SiC-wafers with a diameter of 35 and 51 mm purchased from Cree Res. Inc. (Durham
NC) were used as substrates for most of our growth experiments. These wafers are oriented in
the (000Indirection (Si-face) with an off-angle of 8 degrees towards (11-20).
Thickness and growth rate of the epitaxial layers were determined by weight difference [10] or spatially resolved - either optically (room temperature infrared reflectance measurement [11,12])
or electrically (capacitance-voltage-measurement; CV). This CV-method (contact size 1*1 mm2)
was also employed for determination of lateral and vertical doping profiles. Further
measurements were made at the University of Pittsburgh (low temperature photo luminescence;
LTPL) and at the Fraunhofer Institute for Integrated Circuits in Erlangen (atomic force
microscopy; AFM).
For the evaluation of inhomogeneities of properties (thickness and doping of layers) we
used the following scheme:
The measurements are performed on typically > 300 points equally spread on the wafer surface
without edge exclusion. The derived data are accumulated and the delta between the 10% and the
90 % value of the accumulated distribution is used as measure for the inhomogeneity. Percentage
values are then calculated by dividing this delta by the 50 % value of the accumulated
distribution (see Fig. 3). This technique is suited to eliminate the influence of measuring points
affected by crystal defects on the homogeneity assessment.
!
I
...• • • '
-
! /
Variation
<D
3
/
. = (seo-z10)/zso
C
0>
r\
= +17%
i' . . \:
Dg
Ü
Ü
•:
B 40-
40
j
jJi
8 20OJ
0»
80
.M-f
2,o
30
20
10
:::
i
ZB0
«J
'
i '
1 !"Vl : =
0
Zqr
151
Fig. 3:
Example of deriving a measure for the total
variation of a property (in this case doping)
from an accumulated occurrence plot. The
right axis belongs to the histogram (dashed).
RESULTS AND DISCUSSION
Flow dynamics
In the past we employed numerical process simulations to predict both flow stability and
deposition behavior in a single-wafer SiC-VPE system as reported elsewhere [13,14].
Encouraged by the good correlation between numerical and experimental results, these tools
were also applied during the design phase of the new multi- wafer reactor chamber to ensure that
stable gas flow conditions are achievable. The results showed that a total gas flow in excess of
100 slm (at 300 mbar chamber pressure) should be necessary to avoid gas recirculation in case of
a cylindrical reactor wall. On the other hand, with a confined inlet (see Fig. lb) the flow was
expected to be stable with less than half of this amount of hydrogen. A typical flow pattern and
temperature distribution as derived by the numerical simulation is shown in Fig. 4. This picture
shows completely laminar flow without any recirculation in the reaction chamber above the
wafer plate.
reactor axis
T(K)
0.25
=
0.15
0.05
1774
1696
1618
1541
1463
1386
1308
1231
1153
1076
998
921
843
765
688
610
533
455
377
300
Fig. 4:
Typical flow pattern and temperature
distribution in the reactor chamber as
derived by a 2-dimensional numerical
simulation. The susceptor is assumed to
be isothermal as a boundary condition.
For symmetry reasons only the right
half of the chamber is plotted.
The two axes give the coordinates in
meters.
0.2
As reported previously for a single-wafer reactor[13,14,15], it is possible to observe the
stability of gas flow in the reactor due to a silicon cluster formation. This holds also for the
multi-wafer system and Fig. 5 gives an example of the appearance of the resulting irradiant
layer. Therefore, we were able to test the numerical prediction after installation of the system by
observing the behavior of this layer. These observations confirmed that the flow remains stable
even at a hydrogen flow of only 40 slm (250 mbar, 350 rpm, 1500°C wafer temperature), i. e. no
swirls are visible moving upwards in the reactor chamber.
The formation of the Si clusters is caused by a local supersaturation of Si generated by the
complete dissociation of SiFLi and the relatively steep T-gradient directly above the wafer plate
as explained in [15].
152
Fig. 5:
Picture of an irradiant layer due to Si
supersaturation and cluster formation in
the gas phase. The picture is taken through
a viewport, which is normally closed by a
manual shutter.
A stable, not time dependent and strictly
localized layer is an indication for stable
flow conditions and no recirculation in the
gas phase.
Background doping:
The electrically active impurity concentration was mainly quantified by voltage-capacitance
measurements on undoped layers with Ti Schottky contacts. Usually the undoped layers are of ptype conductivity similar to what we reported earlier for single-wafer epitaxy [16,17].
Fig. 6 gives a comparison between the lateral impurity distribution in an epitaxial layer grown on
a 35 and a 51 mm diameter wafer. Both layers show a background level below 1015 cm" besides
an edge area with several mm width. The impurity distribution is not complete rotational
symmetry. It shows an eccentric minimum (shifted to the left) and a higher concentration on the
right.
f
'f
'?
2 inch
35 mm
M 0 - 5
N.-ND*1014cm-3
Fig. 6: Background impurity mapping (CV) of epitaxial layers upon wafers with different diameter. A
clear edge effect is visible. The axes just give relative coordinates of the metal contact points.
This can be explained by the impurity generation and transport mechanisms. The main
source for p-type impurities is outgasing of susceptor and wafer-holder materials during the
153
epitaxial growth process. These contaminants can be transported to the growing surface by
diffusion and forced convection due to the high-speed rotation. On the leading edge of the wafer
with respect to the direction of rotation, the convection acts in the same direction as the diffusion
(and against diffusion on the other edge). This leading edge corresponds to the right hand edge
of the wafer maps displayed in Fig. 6 which shows a wider zone of increased impurity
concentration. The average impurity concentration drops for increasing wafer diameter, due to a
reduced importance of those edge effects. In the center of the 2" wafer it has a value below 10
cm"3. The electrically determined impurity concentration steadily decreases during the growth
process, i. e. the impurity level is highest at the interface to the substrate wafer and lowest at the
surface. This decrease starts to saturate for layer thickness > 8-10 urn. A possible explanation for
this effect is that impurity-releasing surfaces in the neighborhood of the wafers become more and
more coated during the growth process and therefore the amount of impurities in the gas phase is
continuously decreasing. Comparing the electrically active impurity concentration between
different wafers processed within the same run, the background level can vary up to 30 %. This
also indicates that the direct surrounding of the wafer and probably mechanical tolerances of the
wafer in the respective pocket play the dominant role for the achievable purity.
The type of impurities acting as acceptors was determined by LTPL [18]. A typical
spectrum revealed by this technique from a nominally undoped epitaxial layer is displayed in
Fig. 7. It shows boron being the dominating impurity and some traces of Al. In addition, a well
developed nitrogen line is visible. The Ti-line (not displayed in Fig. 7) is very weak, intrinsic
point defect related lines (e. g. L-lines) are hardly detectable. On the other hand, a well
developed intrinsic line (I75) shows the good crystalline quality of the epitaxial layer.
PHOTON ENERGY CeW
3.23
3.??
3,25
I I I I II II
I I I I | I I I I | I I I I | I I I I | I I II I I
«
Fig. 7:
A typical LTPL spectrum
(excitation wavelength: 244 nm)
revealed from a nominally
undoped epitaxial layer with 10
Um thickness.
(intrinsic free exciton line I75;
impurity related bound exciton
lines: Po, Qo: nitrogen related;
4Alo: aluminum related; 4Bo:
boron related [18])
20 — T,
3800
3810
3820
3830
3810
3850
UAUElENGrH <A>
3860
3870
3880
3890
The purity described above can only be achieved as long as the SiC-coating of the graphite
pies is not damaged. Such a damage may take place by thermal stress induced crack formation or
via a reaction between the Mo plate and these parts and marks the end of their life time. In case
of such a damage an increase in acceptor impurity concentration of more than one order of
magnitude occurs. The length of the life time of the pies we worked with so far varied between
20 and 70 h of total process time. The reason of this big variation seems to be both quality and
thickness of the SiC coating. Further investigation on this topic is necessary because this life
time is a very important factor for process stability and total epi costs.
154
Homogeneity:
Contrary to typical vertical single-wafer systems [10] the wafers in the new equipment have
to be placed outside the center of rotation (see Fig. 1) obviously. Without additional precautions
this usually leads to a significant radial gradient in flow velocities and probably also in chemical
gas composition at the wafer location. Therefore our system allows control of the radial
distribution of silane, propane and dopants. Thereby we were able to achieve a reasonable spatial
uniformity for both thickness and doping as displayed in Figs. 8 and 9. The thickness mainly
varies in the radial direction, whereas the intentional nitrogen doping distribution shows a typical
maximum at the position of the trailing edge of the wafer. This distribution holds only for
optimized gas flow and growth conditions. For homogeneous gas introduction along the whole
diameter of our top flange we typically get a significantly higher variation of the nitrogen
incorporation in the range of about a factor of 2 even on a 35 mm wafer. In this case the
maximum of the doping is at the outmost part of the wafer with respect to the axis of the wafer
plate.
111 <i
111 \
Fig. 8: CV-map (contact size 1*1 mm2) and distribution function of intentional N-doping of an
epitaxial layer upon a 2 inch wafer. At the voltage necessary to measure in a depth of 3.5
urn (>150 V) all contact areas containing a defect have already failed (white areas). The
axes in the map give relative coordinates of the metal contact points.
The improved homogeneity shown in Fig. 8 was partly achieved by careful adjustment of
the above mentioned radial distribution of process gases but also by reducing the silicon
supersaturation, i. e. reduction of Si cluster formation (Fig. 5) by means of decreasing the silane
flow. This effect of Si cluster formation on the doping homogeneity probably can be explained
by a radially unequal reevaporation of these Si-clusters, leading to a radial variation of the local
Si/C ratio and - therefore - to a dopant incorporation efficiency which also depends on the radial
coordinate.
Unfortunately, this reduction of Si supersaturation additionally causes a reduction in growth
rate. Thus, the homogeneity shown in Fig. 8 can only be maintained at growth rates up to 3.5
um/h so far.
155
SAMPLE: 4H-SIC AD0W6-05 THICKNESS: (ilTl
2 inch
Fig. 9:
Thickness uniformity map of an epitaxial layer upon a
2" wafer measured by room temperature infrared
reflectance [11,12]. The total delta of the thickness
variation is < 5%.
The remaining mainly azimuthal orientated doping inhomogeneity can be dedicated to the
influence of the surrounding of the wafer: Usually the temperature of the wafer surface is about
30 degrees lower than the temperature of the surrounding coated graphite surface. This is due to
the additional heat transfer boundary between this graphite part and the wafer. Correlated to this
temperature difference there is a difference in the local chemical equilibria on the two surfaces.
In practice, one has to expect a higher Si vapor pressure above the hotter surrounding surfaces
than above the SiC wafer surface. Thus, the trailing edge of the wafer is always moving in this
area of increased Si vapor pressure, it faces a higher Si/C ratio in the gas phase leading to a
higher nitrogen incorporation. The resulting variation of doping in azimuthal direction of about ±
15 % can only be reduced by means of reducing the temperature difference between wafer and
surrounding material, i. e. a completely different growth set up would be necessary.
Besides the homogeneity of a singular epitaxial layer we also have to deal with the variation of
properties from wafer to wafer during one run (inter wafer homogeneity) and from run to run. In Fig.
10 the typical scatter of the doping concentration within one run is shown in dependence of the wafer
position (full load of the system with 6 two inch wafers). The data points give the average doping
concentration on each wafer, the size of the error bar corresponds to the inhomogeneity on each
wafer. It is clearly visible that there is not a random scatter of the average doping, but a sinusoidal
fluctuation along the azimuthal coordinate of the wafer plate (Fig. 2) with an amplitude of 10 to 15
%. This behavior is very reproducible and can be explained by a not completely leveled plane of
rotation of the wafer plate with respect to the coil plane. In this case temperature may vary slightly in
azimuthal direction leading to the observed variation of doping along this coordinate.
•
g 10
*
—,_
.
i
thickness
•
103
I
[.
f
i—
"
I
q
.—r—
.
doping
4 S
flj
i
1
2
*
A
5
£
position on susce ptor plate
156
Fig. 10:
Variation of thickness and
doping within one full loaded
run. For this plot the average
thickness is evaluated by weight
difference, doping is measured
byCV.
As also shown in Fig. 10 the layer thickness is varying in the same manner from wafer to
wafer but with a much smaller amplitude of less than 5 %. The thickness inhomogeneity on one
wafer usually is very small, typically about 4 to 8 % on two inch wafers like depicted in Fig. 9
(measured by room temperature infrared reflectance [11,12]).
Both thickness and doping seem to be reproducible with an alteration of less than 5 % from
one run to the next. This holds at least for the short time range, we do not have enough data to
make statements about the long term stability so far. In this case one has to take into account
aging effects of helping materials like the wafer plate assembly, which may cause systematic
shifts of epi-layer properties even in the case of macroscopic equal control parameters.
Surface morphology
Similar to what we have reported for a single-wafer epitaxial system formerly [16],
epilayers grown in the new multi-wafer system show very good surface morphology apart from
wafer related defects. An AFM image of a typical layer grown with 3.5 um/h is depicted in Fig.
11. The surface shows steps corresponding to its off orientation, which have a height of about
0.5 nm (twice the distance between two Si-C layers). This compares favorably with values
reported by other groups on 4H wafers with 8° off orientation [19]. A surface as smooth as
shown in Fig. 11 can be maintained over a wide range of process parameters and external Si/C
ratios (0.6 - 5 at 250 mbar, calculated from the macroscopic silane and propane flow).
Exceeding the limit of stable growth by further increase of propane flow leads to rough surface
formation first at the inner (pointing to the center of rotation) edge of the wafer. This is also an
indication for a increasing local Si/C ratio along the radial coordinate as discussed in the chapter
on doping homogeneity.
£k
Jfcr.
#• .-. ; V
Fig. 11:
AFM-image of a typical surface of an
epitaxial layer grown in the MWS
A linescan perpendicular to the steps
reveals an average step height of 0.5 nm.
Electrical characteristics and device yield
Recently we have started to use epitaxial layers grown in the multi-wafer system in our
device program. Typical electrical characteristics of edge-terminated [20,21] Schottky diodes
manufactured on such layers are shown in Fig. 10. At a reverse voltage level of 600 V these
diodes show a reverse current of« 10'3 A/cm2 with a yield up to 75 % (active area 1.2 mm2).
This yield is in good correlation with the specified micropipe density of the wafers used as
substrate (< 30 cm"2). Whereas this result was achieved by using 35 mm diameter wafers, similar
observations are made on 2 inch substrates but with a slightly lower yield.
157
Packaged diodes have a specific differential on-resistance of 1.2-1.5*10"3 Dem2, indicating
a carrier mobility in the epitaxial layer of about 700 - 900 Vs/cm2 (substrate resistivity: 0.02
ficm). This value together with the reasonably low leakage current density indicates a good
quality of the epitaxial layers and - specifically - very little defects generated in addition to the
wafer defects during the epi process and the subsequent process steps.
. 10"
JZ
X
-200
-100
device area : 1 mm2
Wafer: U241-06
epi layer: 1*1016cnr3, 8 pm -
io-!
- io-3
' IO"4
io-5
10*
10"'
10*
Fig. 12:
Typical reverse characteristics of
Ti-Schottky diodes with implanted
function terminated extension
[18,19] manufactured on an
epitaxial layer grown in the multi
wafer system.
(specific differential on-resistance:
1.2-1.5*10"3ncm2)
io-91
10'
-600
-400
-300
reverse voltage (V)
CONCLUSIONS
First results with a vertical multi-wafer VPE system showed that reasonable purity and
homogeneity of SiC epitaxial layers can be achieved. The attained background impurity level <
10 5 cm" is already sufficient for the manufacturing of devices with a reverse voltage up to 1000
V and there is potential for further improvement. The intra-wafer homogeneity of both doping
and thickness is acceptable, the variation of doping from wafer to wafer within one run and from
run to run is in a range of about 20 % and therefore suited for some starting applications of SiC
like Schottky diodes in the above mentioned voltage range.
A key issue for such devices are the process costs for the epi layer growth. The system
described in this paper allows growth on 6 wafers at a time and the necessary cooling, purging
and reloading time between two runs is less than one hour (i. e. time between end of growth and
start of growth on the next set of wafers). This is enabled by the avoidance of thermal isolations
in our reactor and by the use of a loadlock. A first estimation of possible manufacturing costs of
10 um thick epitaxial layers on two inch wafers show that a value of significantly below 200
US$ will be possible in a production environment (including capital usage, consumables, lab
cost, personal etc). It is a necessity to meet this value before SiC epitaxy can be named
"production suited", because otherwise it will be very difficult to identify volume applications,
which can accept the still very high price per unit area of SiC devices.
In our case we made a trade-off between quality and costs of the epitaxial layers: Compared
to results already reported for single-wafer epi systems [16,17,22,23], the properties of layers as
described above are not very impressive, but they can be achieved at quite low costs!
In summary, if we can stabilize and reproduce the results described above over a longer
time, we then have a production-suited epitaxial process at least for SiC devices like Schottky
diodes dedicated to the < 1000 V range.For higher blocking voltages and/or for applications less
tolerant towards doping variations further work on process and hardware optimization is
necessary.
158
ACKNOWLEDGEMENT
The authors would like to thank W. J. Choyke (Univ. Pittsburgh) for the photoluminescence
measurement on our epitaxial layers and C. Q. Chen (Univ. Erlangen) for the determination of epi
layer thickness by infrared reflectance. Further thanks go to Yu.N. Makarov (Univ. Erlangen) for
his work on numerical simulation of our epitaxial process. We also want to acknowledge the
engagement of Paul Fabiano, Alex Guarry, David Voorhees and Dennis Stucky (all of
EMCORE) in setting up the epi system and solving initial problems.
REFERENCES
1.
H. Matsunami, T. Ueda, H. Nishino, Mater. Res. Soc. Symp. Proc. 162 (1990) p. 397
2.
K. Rottner, R. Helbig, J. Cryst. Growth. 144 (1994) p. 258
3.
DJ. Larkin, Mat. Res. Soc. Symp. Proc, 410 (1996) p. 337
4.
DJ. Larkin, P.G. Neudeck, JA. Powell, L.G. Matus, Appl. Phys. Lett. 65 (1994) p. 1659
5.
Aixtron AG, Kackertstrasse 15-17, D52072 Aachen, Germany
6.
Epigress AB,Ideon Science & Technology Park, SE-223 70 Lund, Sweden
7.
EMCORE Corporation, 394 Elizabeth Avenue, Somerset, NJ 08873, USA
8.
A.A. Burk, MJ. O'Loughlin, S.S. Mani, Mat. Sei. Forum 264-268 (1998) p. 83
9.
R. Rupp, A. Wiedenhofer, D. Stephani, Proc. of the 2nd Europ. Conf. on SiC and Rel. Mat.
ECSCRM'98, Sept. 2-4 1998 Montpellier, France, in press
10. R. Rupp, P. Lanig, J. Voelkl, D. Stephani, J. Cryst. Growth, 146 (1995) p. 37
11. M.F. McMillan, U. Forsberg, P.O.Ä. Perssons, L. Hultman, E. Janzen, Material Science Forum,
264-268 (1997) p. 245
12. CQ. Chen, F. Engelbrecht, C. Pepperemüller, N. Schulze, R. Helbig, R. Rupp, Institute of Appl.
Physics, Univ. Erlangen-Ntirnberg, to be published
13. R. Rupp, Yu.N. Makarov, H. Behner, A. Wiedenhofer, phys stat sol (b), 202 (1997) p. 281
14. Yu.N. Makarov, Yu.E. Egorov, A.O. Galyukov, A.N. Vorob'ev, A.I. Zhmakin, R. Rupp,
Proc. of the 2nd Europ. Conf. on SiC and Rel. Mat. ECSCRM98, Sept. 2-4 1998
Montpellier, France, in press
15. A.N. Vorob'ev, S.Yu. Karpov, A.E. Komissarov, Yu.N. Makarov, A.S. Segal,
A.I. Zhamakin, R.Rupp, Proc. of the 2nd Europ. Conf. on SiC and Rel. Mat. ECSCRM-98,
Sept. 2-4 1998 Montpellier, France, in press
16. R. Rupp, A. Wiedenhofer, P. Friedrichs, D. Peters, R. Schörner, D. Stephani, Material
Science Forum, 264-268 (1998) p. 89
17. R. Rupp, P. Lanig, J. Völkl, D. Stephani, Mat. Res. Soc. Proc. 423 (1996) p. 253
18. R.P. Devaty, WJ. Choyke, phys stat sol (a), 162 (1997) p. 5
19. T. Kimoto, A. Itoh, H. Matsunami, Appl. Phys. Lett. 66 (1995) p. 3645
20. H. Mitlehner, W. Bartsch, M. Bruckmann, K.O. Dohnke, U. Weinert, Proc. of the ISPSD'97
Weimar 1997 IEEE p. 165
159
21. H. Mitlehner, P. Friedrichs, D. Peters, R. Schörner, U. Weinert, B. Weis, D. Stephani, Proc.
of the ISPSD'98 Kyoto 1998 IEEE p. 127
\
22. A.A. Burk, L.B. Rowland phys stat sol (b), 202 (1997) p. 263
\
23. O. Kordina, A. Henry, E. Janzen, C.H. Carter, Material Science Forum, 264-268 (1998) p. 97
160
MULTI-WAFER VPE GROWTH OF HIGHLY UNIFORM SiC EPITAXIAL LAYERS
M.J. O'Loughlin, H.D. Nordby, Jr., and A.A. Burk, Jr. *
Advanced Technology Laboratories, Northrop Grumman ES3, PO Box 1521 Baltimore, MD
21203, michaelJ_oloughlin @md.northgrum.com
current address, Cree Research, Durham, NC
ABSTRACT
A multi-wafer silicon carbide vapor phase epitaxy reactor is employed that features full
planetary motion and is capable of high quality epitaxy on seven, two-inch diameter substrates.
We are currently performing preproduction growths of static induction transistor (SIT) and metal
semiconductor field effect transistor (MESFET) active layers. On a 2-inch diameter substrate,
layer uniformity is typically +5% (standard deviation/mean) for both dopant concentration and
layer thickness (for 1 3/8-inch substrates, layer uniformity is around ±3%). For the seven wafers
within a run, interwafer uniformity has been dramatically improved to approximately +8% for
dopant concentration and ±3% for layer thickness. Process control charts will be presented
which exhibit that interrun (run-to-run) variation in both thickness and doping can be kept within
±10% of the desired values.
INTRODUCTION
Prototype silicon carbide (SiC) static induction transistors (SIT) and metal semiconductor
field effect transistors (MESFET) have been demonstrated with performance exceeding that
typically available from both silicon and gallium arsenide transistors [1,2,3]. To realize
production of SiC based devices, a multi-wafer SiC vapor phase epitaxy (VPE) reactor has been
developed in collaboration with Aixtron, AG [4]. The reactor, based on the design of Frijlink
and coworkers [5], features full planetary motion and, as currently configured, is capable of
simultaneous, high quality epitaxy on seven, two-inch diameter substrates.
To achieve satisfactory uniformity and reduce run-to-run variability, it was necessary to
modify many of the original reactor components [6]. The improvements in uniformity and
variability resulting from those modifications were sufficient to allow modest production goals
for device quality SiC epitaxial layers to be exceeded in the previous year. Significantly higher
volumes of epitaxial layers with more stringent specifications for dopant concentration and layer
thickness values and uniformities will be required to meet current and future production goals.
Indeed, production specifications are likely to require that all areas of all epitaxial layers be
within ±10% of target values for both dopant concentration and thickness. Several more reactor
modifications will be summarized which have allowed those production specifications to be
realized with high yield.
EXPERIMENT
The VPE reactor used to grow the SiC epitaxial layers discussed within has been described
in detail elsewhere [3,6,7]. A cross sectional view of the reactor is shown in figure 1. The
reactor consists of an inductively heated, mechanically rotated susceptor with seven wafer
holders (satellites). The satellites are themselves rotated on the susceptor by means of gas foil
levitation [5]. The satellite rotation is desirable to reduce upstream/downstream variation in
growth. The susceptor rotation averages inhomogeniety in the RF induction heating and
geometry of the non-rotating reactor components. However, susceptor rotation does not directly
161
Mat. Res. Soc. Symp. Proc. Vol. 572 ® 1999 Materials Research Society
affect inhomogenieties that result from
anisotropic conductivity in the graphite
susceptor, dimensional nonuniformity of the
machined
and
coated
susceptor,
and
misalignment, i.e. tilt or eccentricity, of the
susceptor on its support. All of those sources of
inhomogeniety can be minimized but are
difficult to eliminate.
A graphite foil-on-foam ceiling [8]
separates the growth region from the watercooled reactor chamber.
The ceiling is
passively heated by radiation from the ~1600°C
susceptor creating a pseudo hot-wall effect. The
hot ceiling reduces supersaturation and
Figure 1. Stylized cross section of VPE
condensation of silicon vapor in the reactor. In
reactor. A) Gas Injector, B) Ceiling, C)
a previous communication [6], it was reported
Gas Collector, D) Susceptor, E) Wafer
that the graphite foil-on-foam ceilings could
Holder,
F) Rotation Axis
only be used for tens of runs.
However,
recently, ceiling lifetime has been extended to
nearly 100 runs, which has increased epitaxial layer production efficiency.
Finally, a perforated gas collector terminates the outer diameter of the susceptor, the
growth region, and the ceiling. The gas collector, made from graphite foam or molybdenum,
ensures laminar flow of the gases in the growth chamber.
Typical conditions for SiC epitaxial layer growth in this reactor have been described in
detail elsewhere [6]. All of the growths reported in this study were performed at reduced
pressure, susceptor temperatures of ~1600°C, growth rates of ~3 |xm/hr, and silicon-to-carbon
ratios around 0.88. The mass of each wafer is measured with an analytical balance both before
and after each growth to determine epitaxial layer thickness, growth rate, and wafer-to-wafer
uniformity. After growth, all epitaxial layers, except highly doped layers used for ohmic
contacts, are characterized by capacitance versus voltage (CV) measurements on 200 Jim
diameter photolithographically defined Cr-Au Schottky diodes.
CV measurements are
performed at regular intervals along one radius of each wafer and used to extract relative
thickness and dopant concentration values and uniformities. The rotating wafers have been
found to be highly axially symmetric making the measurement of one radius sufficient. To
minimize the affect of long-term drift in growth rates and background impurity concentration,
the measurements from each run are used as a calibration for the next run.
RESULTS
All epitaxial layers were characterized by mass and CV measurement. Three levels of
uniformity were examined; intrawafer uniformity, interwafer (or intrarun) uniformity, and
interrun (run-to-run) uniformity.
A five-point overlay of dopant concentration as a function of depth, extracted from CV
measurements, is presented in figure 2. The 4.6 (im thick epitaxial layer, grown on a two-inch
diameter, low resistivity, n-type substrate, was intentionally doped with nitrogen to a
concentration of 4xl015 cm"3. For the epitaxial layer shown in figure 2, the uniformities
(standard deviation divided by the mean) of the dopant concentration and thickness were 4.9%
and 5.1% respectively. In both cases, the range (maximum-minimum) of measurements as a
percentage of the mean is approximately 12%. The uniformity exhibited by this layer does not
162
represent a "best-effort", but is
1E+19
representative of typical two-inch
epitaxial layers. For example, the
average uniformity for a recent series
of runs with two-inch wafers was
4.7% for dopant concentration and
5.2% for thickness.
Excellent
intrawafer uniformity for 1 3/8-inch
wafers has also been observed. For
those smaller diameter wafers, the
average uniformity, based on over
1E+14
200 wafers, was 3.6% for dopant
2
3
4
concentration and 2.4% for thickness.
Depth (ujn)
After extensive reduction of
susceptor induced inhomogeniety
Figure 2. CV dopant profile measured at five
[6,7], interwafer (wafer-to-wafer
positions along a radius of a 2" diameter wafer. The
within a run) uniformity had only
epitaxial layer is thinnest in the center. The thickest
been reduced to 10% at best. With
profile is measured at 4.6 mm from the edge.
that large of a standard deviation, at
least one-third of the wafers would
fall outside the ±10% absolute specifications for production epitaxial layers. However, by
replacing the graphite foam gas collector with a molybdenum one with smaller perforations, the
interwafer thickness uniformity improved to approximately 3%. The time evolution of
interwafer uniformity is presented in figure 3. The molybdenum gas collector was installed at
point A on that chart. It is evident
18%
that the improvement in thickness
uniformity was not matched by a
corresponding
improvement
in
dopant concentration uniformity. The
dopant concentration of the individual
wafers still exhibited a systematic
distribution, i.e. the highest doped
position was opposite the lowest
doped position. The highest doped
position correlated with the hottest
side of the susceptor. By rearranging
the susceptor hardware to improve
Run Number
the temperature balance (point B in
Figure 3. Inter wafer uniformity for 22 runs. Anew
figure 3), the dopant concentration
gas collector was installed at A. Susceptor hardware
uniformity has been reduced to 7%.
was rearranged at B to improve the temperature
Most of the remaining non-uniformity
balance.
is randomly distributed and arises
from small differences in local
temperature due to variations in; emissivity of components, thermal contact between wafer and
satellite, satellite levitation, etc. An overlay of the dopant profiles at five positions along the
radius of each of seven wafers in a single run is presented in figure 4. For the run represented by
that figure, both the thickness and dopant concentration uniformity at 2x10 cm"3 are
approximately 5%. At lxlO16 cm"3 the dopant concentration uniformity is slightly degraded to
7%.
163
Having demonstrated good
uniformity on a 2-inch diameter
wafer and for each wafer in a run,
the remaining obstacle to high yield
production epitaxy is interrun
variation.
A control chart
representing our ability to meet
target values for thickness and
dopant concentration as a function
of run number is presented in figure
5. The data points represent the
percentage deviation of the average
value for a run from the target
value. The error bars represent the
interwafer uniformity within each
run. Where the error bars (la) fall
outside the ±10% of target lines,
there will be a yield loss. It is
evident, from figure 5, that the
improvements that were effected to
decrease the interwafer variation
also permitted better control of runto-run variation. In fact, after point
B (corresponding to improved gas
collector geometry and susceptor
temperature
uniformity),
the
average absolute deviation from
target values were 2.8% for dopant
concentration
and
3.6%
for
thickness.
CONCLUSIONS
1E+18
8 1E+17
| 1E+16
o
1E+15 4
0.0
0.2
0.4
0.6
Depth (|im)
Figure 4. An overlay of the CV dopant profile
measured along the radius of each of seven wafers in a
run.
30%
[- -Doping -»-Thickness
20%
.
10%
/\
T
\^ \
BT
-10%
\ /•
1 ^/
A
\
/'
A
. p"1
/ J
-20%
-30%
Run Number
Figure 5. Control chart for doping and thickness.
Symbols represent percent deviation of the average
value for a run from the target value.
Sufficient uniformity has been
demonstrated for SiC epitaxy on seven at-a-time two-inch diameter wafers for high yield in
production. Intrawafer uniformity of approximately 5%, similar to what was previously reported
for 1 3/8-inch wafers, can be routinely achieved on two-inch diameter wafers. The most
significant improvement has been the reduction of interwafer uniformity to 3% for thickness and
7% for dopant concentration. The combination of good intrawafer and interwafer uniformity has
been coupled with better control of run-to-run variation such that a very high percentage of
epitaxial layers meet stringent production specifications.
ACKNOWLEDGMENTS
The authors gratefully acknowledge David Stanley, Gil Rykiel, Sam Ponczak, Ollie
Gildow, and Rich Siergiej for their assistance in reactor modifications, epitaxial growth, and
characterization. Development of this reactor was supported in part by the Department of the Air
Force under contracts F33615-92-C-5912 and F33615-95-5427 (Tom Kensky, contact monitor
and Laura Rea, program direction leader).
164
REFERENCES
1. S. Sriram, G. Augustine, A. A. Burk, Jr., R.C. Glass, H. M. Hobgood, P. A. Orphanos, L.V.
Rowland, T. J. Smith, C. D. Brandt, M. C. Driver, and R. H. Hopkins, IEEE Electron Device
Lett., EDL-17, p. 369 (1996).
2. R. R. Siergiej, S. Sriram, R. C. Clarke, A. K. Agarwal, C. D. Brandt, A. A. Burk, Jr., T. J.
Smith, A Morse, and P. A. Orphanos, Tech. Digest Int. Conf. SiC and Rel. Maf95, (Kyoto,
Japan 1995), p. 321.
3. A. A. Burk, Jr., M. J. O'Loughlin, R. R. Siergiej, A. K. Agarwal, S. Sriram, R. C. Clarke, M.
F. MacMillan, V. Balakrishna, and C. D. Brandt, J. Solid State Electronics, accepted for
publication.
4. Aixtron Inc. Kackerstr. 15-17, D-52072 Aachen, Germany.
5. P. M. Frijlink, J. Crystal Growth, 93, p. 207 (1988).
6. A. A. Burk, Jr., M. J. O'Loughlin, and H. D. Nordby, Jr., J. Crystal Growth, accepted for
publication.
7. A. A. Burk, Jr., M. J. O'Loughlin, and S. S. Mani, in Silicon Carbide, Ill-Nitrides, and
Related Materials, , edited by G. Pensl, H. Morkoc, B. Monemar, and E. Janzen (Materials
Science Forum, 264-268, Trans Tech Publications, Switzerland 1998), p. 83-88.
8. Calcarb, Inc., Rancocas, New Jersey, USA.
165
CHARACTERIZATION OF THICK 4H-SiC HOT-WALL CVD LAYERS
M.J. Paisley*, K.G. Irvine, O. Kordina, R. Singh, J.W. Palmour, and C.H. Carter, Jr.
Cree Research, Inc., 4600 Silicon Drive, Durham, NC 27703-8475, USA
*Mike_Paisley@Cree.com
ABSTRACT
Epitaxial 4H-SiC layers suitable for high power devices have been grown in a hot-wall
chemical-vapor deposition (CVD) system. These layers were subsequently characterized for
many parameters important in device development and production. The uniformity of both
thickness and doping will be presented.
Doping trends vs. temperature and growth rate will be shown for thep-type dopant used. The
«-type dopant drops in concentration with increasing temperature or increasing growth rate. In
contrast, the/?-type dopant increases in concentration with decreasing temperature or increasing
growth rate. A simple descriptive model for this behavior will be presented.
The outcome from capacitance-voltage and SIMS measurements demonstrate that transitions
from n to n, orp top', and even ntop levels can be made quickly without adjustment to growth
conditions. The ability to produce sharp transitions without process changes avoids degrading the
resulting surface morphology or repeatability of the process. Avoiding process changes is
particularly important in growth of thick layers since surface roughness tends to increase with
layer thickness.
Device results from diodes producing two different blocking voltages in excess of 5 kV will
also be shown. The higher voltage diodes exhibited a breakdown behavior which was near the
theoretical limit for the epitaxial layer thickness and doping level grown.
INTRODUCTION
The properties of silicon carbide, such as wide bandgap, high breakdown electric field
strength, and high thermal conductivity are characteristics that make the material highly
desirable for high power devices. Recent device demonstrations and measurements of the
superior electrical and physical properties of silicon carbide (SiC) have shown that it is the
premier semiconductor material for fabrication of high power and power microwave electronic
devices. In addition, SiC devices can operate at elevated temperatures allowing them to be used
at either high ambient temperature or with reduced cooling requirements in nominal ambients. In
recent years substrate and CVD developments have made great progress, thereby allowing for
the development of SiC-based high power high frequency devices such as Metal-Semiconductor
Field Effect Transistors (MESFETs)[l]. Other high power devices such as diodes, MOSFETs
and GTOs have also been demonstrated [2,3,4]. While current levels of uniformity of layer
thickness and doping are sufficient for production use, the current wafer diameters are not. Thus
work needs to continue on layer uniformity for increasing wafer sizes as they become available
to be ready for production when it becomes commercially viable.
EXPERIMENTAL SETUP
The wafers were grown in a horizontal hot-wall CVD reactor. The key component of the
reactor is the graphite susceptor which is similar in construction to those described in Ref. 5
and 6. The susceptor is heated inductively and is designed to obtain good heating uniformity over
a large area. The temperature is measured by a pyrometer focused on a position at the leading
edge of the susceptor. The susceptor is also tapered to compensate for gas reactant depletion
which may be quite severe in hot-wall systems. The precursors, silane and propane, are diluted in
a high flow of purified hydrogen. Growth temperatures in the range of 1500-1700°C were used.
Nitrogen was used as the n-type dopant and trimethyl aluminum (TMA) was used as thep-type
dopant.
The thicknesses and thickness uniformities were measured by observing the cleaved edge of
the sample in a scanning electron microscope (SEM). The difference in doping between the
highly doped substrate and the epitaxial layer gives sufficient contrast for determining the
167
Mat. Res. Soc. Symp. Proc. Vol. 572 ° 1999 Materials Research Society
thickness. The doping and doping uniformities were measured by a mercury probe capacitancevoltage (C-V) profiler.
RESULTS AND DISCUSSION
Epitaxial Layer Uniformities
30
Epitaxial layers were deposited on prototype
75 mm diameter wafers, cleaved and then
- \
25measured along the flow direction for layer
:
j
I
I
!
I
I ' !
thickness by cross-section scanning electron
microscopy (SEM). The results are shown in
20 Figure 1. The uniformity is calculated as the
standard deviation divided by the mean. The
Average: 27.2 urn
$ 15 -_
"epi-crown" (the raised ridge along the wafer
Uniformity: 1.21%
circumference) was not included in the
ß 10calculations.
'Additional wafers were grown and
5characterized by C-V measurements to determine
the doping uniformity. For a doping level of
7.5* 1015 cm"3 «-type the resulting uniformity
— i i i i | i i i i j i i i i | i i i i j i i i i j i i i i j i i i i | i i'
was 7%, also calculated as standard deviation
0
10 20 30 40 50 60 70
divided by the mean. Similar growth runs on
Distance from upstream edge (mm)
75 mm wafers for/?-type layers resulted in a
uniformity of 9%. The results for various
Figure 1. Epi thickness of 75 mm wafer.
diameter wafers are shown in Table 1.
Achieving the desired doping level is also very important to the final device performance.
However, reaching the desired levels of around 1 x 10 cm and below for thick layers of SiC is
often difficult. Nearly all reactor parameters can affect the doping at these levels to easily shift
the results outside allowable limits. Additionally, C-V measurements in this range are less
reliable when performed with a mercury probe profiler. We have collected data from over
80 reactor runs with different dopant types and
doping. All runs were for low doped thick layers Table 1. Uniformities of epitaxial layers
on various wafer sizes.
(>20 urn). The results show that if we accept
levels of within ±20% of the target doping, the
«-type
Thickness
p-type
Wafer
reactor yield is 93%. It should be noted that this
7%
1.2%
9%
75 mm
high yield applies only to standard product
1.6%
n/a
<1%
50 mm
offerings which are well characterized. Non2%
2%
<1%
35 mm
standard combinations of thickness and doping
levels will significantly impact the resulting
reactor yields.
Dopant Incorporation Model
The incorporation behavior of various dopants in SiC has been studied for some time. The
behavior of nitrogen incorporation in SiC has been established as decreasing with both
increasing growth temperature and increasing growth rate [7]. This has been our experience as
well, though it does appear that susceptor geometry can affect the magnitude of these changes, at
least in our laboratory. It turns out that the incorporation behavior of Al is remarkably different
where it still decreases with increasing temperature, but also increases with increasing growth
rate. Figure 2 shows the Al concentration measured by C-V as a function of temperature offset
from an arbitrary value. These temperatures were maintained for the duration of a single
deposition run. Temperature offset is used because the actual wafer surface temperature is not
known as the control temperature is measured at some distance from the wafer. Figure 3 shows
the growth rate of the layers shown in Figure 2, which was also held constant during each
individual run. As can be seen in the figure, the growth rate drops at higher temperatures due to
increased etching by the hydrogen carrier gas.
168
Figure 2 shows that the Na-Nd concentration drops rapidy with increasing deposition
temperature as was also observed by Kimoto [7]. Figure 3 shows the growth rate dependence for
the temperature range. It is reasonable to assume that the sticking coefficient of the Al will drop
as the wafer temperature is increased. At a constant growth temperature offset of 20°C, if the
growth rate is increased by 50% (from point to T point in both figures), the doping level
increases by 38%. But if the growth rate is increased again by 60%, then the doping level
increases by an additional 230%. At the highest temperature offset of 120°C, an increase in
growth rate of 83% (A point), results in a doping increase of 100%. So as the deposition
temperature is changed by 120°C while the deposition rate is held constant within 8% and the
TMA flow is held constant, the resulting Al concentration drops by a factor of 250. The
stoichiometry of the input precursors was kept the same though it is likely to assume that the
effective stoichiometry changes when the temperature is varied over this range. While we think
that stoichiometry may play some minor role in the observed behavior [8], it can not explain a
change of this magnitude.
8-
4-
I
2-
' | i i i i i i i | i i i i i i i | i i i i i i i | '
0
40
80
Temperature oflset (C)
'| > i i i i i i | i i i i i i i | i i i i i i i |'
120
0
40
80
Temperature offset (C)
120
Figure 2. Na-Nj concentration (from C-V) as Figure 3. Epitaxial growth rate as a function
a function of adjusted deposition
of adjusted deposition temperature,
temperature (at constant TMA flow).
The data points where the growth rate was increased indicates an initial proportional change
in doping levels with growth rate that becomes non-linear. This appears to us to indicate that the
/>-type doping is influenced by surface roughness or defects. Thus as the growth rate increases,
the surface roughens, and/or defect concentrations increase which provide additional sites for Al
incorporation. It may also explain the high incorporation rate of Al at lower deposition
temperatures where many more defects can act as bonding sites for Al. Nitrogen, in contrast,
decreases with increasing growth rate.
Doping Transitions
One component of device performance is the ability to make transitions from n to n, orp to
p~, and even n top levels. More abrupt transitions permit devices to be fabricated with layer
thicknesses closer to the ideal design thickness. Also, transitions from ntop orp to n levels will
result in potential recombination regions that can be minimized by the abruptness of the
transition. We chose a typical p~ layer structure that consisted of an n substrate followed by an
n buffer layer, then ap buffer layer and finally thep" region. A sample of this structure was
profiled using both C-V measurements and secondary ion mass spectrometry (SIMS). The SIMS
profile is shown below in Figure 4.
169
20
10'
10
Nitrogen
Aluminum
Boron
19
Subs.
n+Ü>uffer"
E
ü
|10
B
21
a
c
o
O
p+bjjffer
18
..player j
,17
,16
10
10
15
14
^^^4^
11111111111111111111111111
0
1
2
jwlfypf'
11111111111111111111111
3
4
Depth (urn)
5
11111111111111
6
I I I | llll
8
Figure 4. Secondary ion mass spectrometry (SIMS) depth profile of a multi-layer structure
indicating sharp transitions between adjacent regions.
It should be noted that the profiles observed with SIMS were also observed with C-V
profiling. These data are not shown since C-V profiles are subject to depletion effects and a
complete profile of the entire structure was not possible on a single structure. The SIMS data
itself is also subject to "layer-mixing" effects which can smear out transitions that occur at
greater depths. The Al profile shows a "spike" at the substrate interface which has been
previously observed by Burk and Rowland [9]. While removal of this "spike" is a matter of some
further research, doping levels are controlled such that no "buried interfaces" occur. The
transition from the n buffer to thep+ buffer shows that this change is possible within much less
than 0.5 urn. The Al profile also shows a "rounding up" behavior which is not present in layers
grown more recently, due to injection of larger quantities of trimethyl aluminum at the start of
such regions. Finally, the transition to the final p" level which falls three orders of magnitude
takes less than 3 urn, with the majority of the transition requiring only 0.2 um to fall to
3xl015 cm"3. The entire structure was deposited without changes in Si/C ratios or other
deposition conditions that might affect the quality of the layers. The surface morphology was
equivalent to single layers grown under these conditions.
High Voltage Device Behavior
Low doped, thick 4H-SiC epitaxial layers are essential for the realization of high voltage
devices. A simple high-power device structure that can use these epitaxial layers is a high
voltage P-i-N diode as shown in Figure 5. To ease the gradient of the electric field at the edge of
the device, an edge termination suitable for very high voltage must be used. A batch of wafers
using a 4H-SiC n layer with a thickness of 85 urn (also called the drift layer) and a later batch
with a layer thickness of approximately 50 urn were fabricated using junction termination
extension (JTE) as the edge termination technique. The termination region has a low dose
implanted/7-type ring surrounding the high voltage anode junction. While the edge termination
was not designed to fully exploit the capability of the epitaxial layer used, a high blocking
voltage in excess of 5.5 kV was achieved on one of the diodes of the first batch. With
170
improvements in the processing, a blocking voltage of 5.9 kV was achieved on one diode in the
second batch. The reverse current-voltage characteristics are shown in Figure 6. The first diode
batch gave a varying leakage current that went from mid-10"7 A to mid-10"5 A. The marked
improvement in leakage behavior of the second batch resulted in leakage current of 10"8 A over
most of the voltage range and demonstrates the much improved processing. Modeling of the
ideal breakdown behavior for SiC at a given layer thickness and doping level showed that the
second diode with a 50 urn epitaxial layer (vs. 85 um for the first batch) closely approached the
maximum theoretical voltage.
:
Anode
—
JTE termination
.
V.
V
P-type"
p-type J
.
* *
5
:
P-i-N Diode
I
Area = 7.85E-5 cm1
5.9 cV, 50 um la yet
n+-type substrate
tU
.——"
-5.5 cV, 85 um la yer
n- epitaxial layer
85 microns, 1-7x14 doped
p
,.— '
j
0
1000
2000
3000
4000
5000
6000
Reverse Bias (V)
Figure 5. The structure ofthe5.5 kV P-i-N
diodes in Figure 6.
Figure 6. Reverse bias I- V characteristics of
diodes.
In addition to a high reverse blocking voltage, a diode must also have a low on-state voltage
drop. In the diodes of the first batch achieving 5.5 kV, the forward voltage drop at 100 A/cm
was 5.4 V. In the second batch of diodes, the contacts were poorly annealed and had a much
higher resistivity which resulted in a larger voltage drop. However, we have recently produced
diodes with a 50 um drift layer which have a forward voltage drop of only 6.5 V at a current
density of 1000 A/cm2. This layer had a blocking voltage of 5 kV.
The formulation presented in Ref. 10 was used along with the on-state voltage drop observed
in the low doped drift region in the device to estimate a carrier lifetime. The results indicated
that the carrier lifetime of the low doped epitaxial layers used in these diodes was to be greater
thanl us [11].
SUMMARY
Highly uniform epitaxial layers of 4H-SiC have been achieved by optimizing growth
conditions in a hot-wall CVD reactor. The thickness and doping uniformity for 3" diameter
wafers have been shown for the first time at <2% and <10%, respectively. The thickness and
doping uniformities for other wafer sizes were also shown with results at or below 2% in all
cases.
A simple model was shown for the incorporation of Al in 4H-SiC showing that increased
growth rate resulted in higher doping levels. It was proposed that this behavior was due to
increased surface defects and/or roughness. SIMS data was presented that showed sharp
transitions across different doping levels and even dopant types. These changes were possible
without changes in deposition conditions which maintains the resulting surface morphology.
High voltage P-i-N diodes were fabricated on low doped epitaxial layers of two different
thicknesses. The best forward voltage drop of 6.5 V at 1000 A/cm2 was achieved using an
epitaxial layer thickness of 50 um. Improved processing resulted in both a higher blocking
voltage of 5.9 kV and dramatically improved leakage current behavior. Modeling showed that
the higher voltage diode approached theoretical limits for breakdown behavior in SiC for its
thickness and doping level. From the on-state resistance the carrier lifetime in the thick, low
doped blocking layer was estimated to be greater than 1 us.
171
ACKNOWLEDGEMENTS
This work was partially supported by DARPA through Air Force Contract No. F33615-C5426 and by ONR through its MURI program Contract No. N00014-95-1-1302.
REFERENCES
1. S.Sriram, G.Augustine, A.A. Burk, Jr., R.C. Glass, H.M. Hobgood, P.A. Orphanos, L.B.
Rowland, R.R. Siergiej, T.J. Smith, CD. Brandt, M.C. Driver, and R.H. Hopkins, IEEE
Electron Device Lett. 17, 369, (1996).
2. O. Kordina, J.P. Bergman, A. Henry, E. Janzen, S. Savage, J. Andre, L.P. Ramberg, U.
Lindefelt, W. Hermansson, and K. Bergman, Appl. Phys. Lett. 67, 1561, (1995).
3. J.N. Shenoy, JA. Cooper, Jr., and M.R. Melloch, IEEE Electron Device Lett. 18, 93 (1997).
4. A.K. Agarwal, J.B. Casady, L.B. Rowland, S. Seshadri, W.F. Valek, and CD. Brandt,
Submitted to IEEE Electron Device Lett.
5
O Kordina, C. Hallin, A. Henry, J. P. Bergman, I. Ivanov, A. Ellison, N. T. Son, and E.
Janzen, Phys. Stat. Sol. B 202, p. 321 (1997).
6. O. Kordina, A. Henry, E. Janzen, and C.H. Carter, Jr., Silicon Carbide, Ill-Nitride and
Related Materials 2, p. 107 (1997).
7. T. Kimoto, A. Itoh, N. Inoue, O. Takemura, et al, Mater. Sei. Forum 264-8, 675 (1998).
8. D.J. Larkin, Phys. Stat. Sol. (b) 202, 305 (1997).
9. A.A. Burk, Jr., and L.B. Rowland, Appl. Phys. Lett. 68, 382 (1996).
10. B. J. Baliga, Power Semiconductor Devices (PWS Publishing, 1996).
11. Paul Chow, Rensselaer Polytechnic Institute, private communication.
172
HOMO-EPITAXIAL AND SELECTIVE AREA GROWTH OF 4H AND 6H SILICON
CARBDDE USING A RESISTIVELY HEATED VERTICAL REACTOR
Ebenezer Eshun*, Crawford Taylor*, M. G. Spencer*, Kevin Kornegay**, Ian
Ferguson***, Alex Gurray*** and Rick Stall***
*Howard University, Materials Science Research Center of Excellence, Washington, DC
20059
**Department of Electrical Engineering, Cornell University, Ithaca, NY 14853
***EMCORE Corporation, SOMMERSET, NJ 08873.
ABSTRACT
Silicon carbide technology is rapidly developing into a production process. This is due to
rapid progress in the development of high quality epitaxy and substrates. We report on the
development of a resistively heated vertical reactor and it's application to homo-epitaxy and
selective area growth. Epitaxial growth of 4H and 6H-SiC requires high temperatures (in excess
of 1500°C). In this work we investigate resistive heating which offers advantages in cost,
temperature uniformity and power efficiency of heating. However, resistive heating presents
major technological challenges. Due to the power efficiencies possible with resistive heating we
are able to obtain temperatures in excess of 1750°C. Using this system we have grown "state of
the art" 4H and 6H-SiC. At 1580°C our background doping is p-type at a level of 3-5x1015cm"3
as measured by capacitance techniques in agreement with earlier results presented by
investigators from Siemens Corp using a similar system. The background concentration
increases by about an order of magnitude at 1680°C. This system has also been used to perform
experiments with selective area growth of SiC using a graphite mask. This masking technology
allows for the growth of SiC in specific regions at elevated temperature in excess of 1600°C.
INTRODUCTION
Silicon Carbide (SiC) has several superior properties as compared to other semiconductor
materials. These properties include high thermal conductivity, inertness to chemical reactions,
hardness, high breakdown field, high saturated electron drift velocity. These superior properties,
makes SiC an excellent candidate for high power and high temperature electronic devices. This
and the availability of high quality epitaxy and SiC substrates have resulted in the rapid
development of the SiC technology. In the epitaxial growth of 6H and 4H-SiC, high
temperatures in excess of 1500°C are required. High voltage, bipolar devices require thick, low
doped epitaxial layers with long carrier life times, and are most conveniently grown by Chemical
Vapor Deposition (CVD)[1].
The reactor used in this study is a EMCORE Corporation system, equipped with a
vertical chamber and a rotating disk technology. Based on the experience of EMCORE in the
field of m-V epitaxy, our reactor is a combination of the rotating disk technology and the high
temperature needs of homoepitaxial SiC growth [2]. In conjunction with EMCORE Corporation,
we have been able to develop a usable resistive heating system as an alternative to RF heating.
The construction of the reactor is shown in figure 1.
173
Mat. Res. Soc. Symp. Proc. Vol. 572 c 1999 Materials Research Society
Hydrogen
Pyrometer
ete
IMA
i trogen
Siiane
PjTOpane
^*>
To pu&p
püm
T
Ar
8on in
Figurel: Diagram showing the construction of the above-mentioned resistively heated CVD
reactor. (1) Resistively heated filament (2) wafer carrier or susceptor and (3) graphite cup.
It consists of a double walled, water-cooled vertical chamber, made out of stainless steel.
It has three view ports one of which is used for pyrometric temperature measurement. It is also
equipped with a RHEED gun for in-sftu RHEED measurements to characterize the grown layers
immediately after growth. The susceptor is resistively heated with a graphite filament with
power consumption at 1700°C of 8kW.
The challenge in the development of the resistive heating system has been the decrease of
the filament lifetime due to hydrogen etching. Hydrogen is present in the reactor as a carrier gas
and decomposition product of the reactant gases (SfflU and C3H8 ). Hydrogen etching is a
function of the filament temperature and the partial pressure of hydrogen in the region of the
filament. Studies at EMCORE indicate that filament failure is likely after a 5% reduction in the
filament cross sectional area. In order to increase the filament lifetime, hydrogen should be
eliminated from the filament region and/or the filament temperature minimized. Our current
design involves the use of a thin (20mil thick) graphite susceptor which drops into an open cup
resulting in a 20mil to 40mil air gap between the filament and the susceptor. This configuration
is used because in minimizes the temperature offset between the filament and susceptor. The
base of the cup is purged with argon as shown in figure 1 to reduce the partial pressure of
hydrogen in the vicinity of the filament. If the hydrogen can be eliminated the filament lifetime
will be determined by carbon evaporation. For our geometry the predicted filament lifetime is
over 1000 hours at a filament temperature of 2000°C. Because we have not been able to
completely exclude hydrogen from the filament the current filament life is still determined by
hydrogen etching. We have been able to achieve filament life times of 15 to 20 hours by this
method.
In our bid to achieve higher filament lifetimes (>100 hours), we are developing a growth
process, which uses argon as the shroud, thus reducing the volume of hydrogen significantly.
We are also designing a new cup and susceptor, which will lock in place to reduce the flow of
hydrogen into the cup and allowing a higher flow of argon for purging.
174
EXPERIMENTS
Homoepitaxial Growth / Reactor Calibration
For the growth of a- SiC, we used (0001) SiC substrates with the polished growth surface
tilted at angles between 3° and 8° from the basal plane . This tilt angle ensures that "step-flow"
homoepitaxial growth occurs and that good morphology is obtained [3],[4],[5],[6]. In order to
grow high quality a- SiC epitaxial layers and also to calibrate our CVD reactor, we have been
working on optimization of the growth parameters including growth temperature, chamber
pressure, reactant gas flows, growth rates, carrier gas flows and rotation speed. In our system a
shroud flow of 21-30 slm is required to stabilize the reactor. Typical growth parameters are
shown in table I.
Table I: Typical growth parameters for the homoepitaxial growth of o> SiC using a resistively
heated CVD reactor.
Growth
Chamber
Temperature Pressure
C3Hg
Flows
SiH4
Flows
Shroud (H2)
Flows
Rotation
Speed
1500°C
to
1700°C
8.4sccm
to
100 seem
275 seem
to
550 seem
21 slm
to
30 slm
750 rpm
50 Torr
to
250 Torr
Propane (C3H8) and silane (S1H4) were used as the reactant gasses with hydrogen as the carrier
gas. The substrates were heated resistively as discussed previously. Initial calibration growths
were performed without in-situ doping of the epitaxial layers thus we were able to find out about
the levels of background impurities during the growth process. We have performed growth
experiments of a- SiC at 1580°C and elevated temperatures (1680CC). We calibrate our reactor
at 1420°C by observing the melting of silicon.
Selective Area Growth of a- SiC
In order to selectively grow a- SiC, we mask parts of the substrate using our high temperature
graphite mask[7]. The graphite mask after growth was removed by oxidizing it and then etching
it in an HF/HNO3 (1:2) etch solution. The growth of the selective epi is verified with a scanning
electron microscope (SEM).
RESULTS AND DISCUSSIONS
In the characterization of our epitaxial layers, we employ several methods to access the
quality of the grown layers. The surface morphology is assessed using an atomic force
microscope (AFM) and an optical microscope .Other characterization methods include two point
probe to check the breakdown voltage and capacitance-voltage (CV) measurements (by
175
depositing Al schottky diodes) using a doping profiler which gives a quick assessment of the
background dopant type, background doping levels and hence a doping profile. Figure 2 shows
the doping profiles for two 4H samples grown under the same conditions except temperature.
~ 10"
<?
E
E
£10
O
Q
a
... ««• • "••
• "• • :
:
1«
a
o
"»»• «N»
Q
Iio15
§ io15
2
a
m io14
Temperature 1680 °C
Temperature 1580 °C
110Q.
'' "''.'''':
;' Silc ratio :i.O
>?
SiirCratfoH.ti '
2
r
:
oi
0.5
Iio14 0
1
1.5
Depth((im)
0.1 0.2 0.3 0.4 0.5 0.6 0.7
Depth (^m)
(b)
(a)
Figure 2: Doping profiles for unintentionally doped p-type 4H SiC epitaxial layers at (a) 1580°C
and(b)1680°C
These background doping levels are due to the fact that we do not use SiC coated parts, hence
although they are reasonably low, the use of SiC coated parts will enable us to achieve even
lower backgrounds.
High temperature growth is advantageous because higher growth rates could be achieved
to satisfy the thick layer requirements of SiC power devices. Results at the elevated
temperatures show significant step bunching as shown in figure 3 and also higher background
doping levels which we believe comes from the graphite out gassing at the higher temperature.
Experiments with the selective area growth show promising results. We see nucleation on the
mask after growth including some etching as shown in figure 4. The SEM picture shows the step
from substrate to epi. The rough edges are due to SiC growing into the undercut in the graphite
mask made during fabrication.
Figure 3: AFM picture of 4H SiC epitaxial layer, grown at 1680°C. Significant step bunching is
observed.
176
Selective epi 9
Mask
(a)
Figure 4: Optical microscope picture of (a) selectively grown 4H SiC epitaxial
layer(magnification 162x) (b)graphite mask surface showing nucleation during growth
(magnification 1618x).
In order to determine the growth rates, we determine the layer thickness. We have
developed the first direct measure of SiC epi thickness measurement by measuring the height of
our selectively grown epi as shown in figure 5, this is a unique advantage of our selective epi
technique.
Ti^^mw;
Figure 5: SEM picture showing selectively grown 4H SiC and a direct measure of epi thickness.
Rough edge is due to growth of SiC into the undercut in the graphite mask.
The graphite mask is easily removed with a silicon etch, leaving the masked area clean without
degradation.
CONCLUSION
We have successfully developed a usable resistively heated vertical CVD reactor for the
growth of SiC. Filament lifetimes of 15 to 20 hours have been achieved. These lifetimes are due
to the fact that we have not been able to totally eliminate hydrogen from the vicinity of the
filament. Because we know the degradation mechanism, we have various plans to increase the
lifetime further. We have grown "state of the art" a- SiC epi layers with this reactor, with a
growth rate of 5nm/hr for Si/C ratio of 1.0. Higher growth rates can be achieved with higher
177
Si/C ratios, this is possible because we have a wide window for the Si/C ratios for high quality
epi growths. We have found the background doping to be p-type ~ 5xl015 cm"3 as reported in [2].
The high temperature capability of resistive heating has enabled us to perform growth
experiments at elevated temperatures (1680°C). Significant step bunching is observed with an
order of magnitude higher background doping relative to growths at 1580°C. Process
optimization is underway to obtain quality epi layers at fester growth rates, an advantage of high
temperature growth. We have successfully grown high quality selective epi (using a graphite
mask) comparable to that grown without masking. Using our selective epi technique, we have
developed the first direct measure of SiC epi thickness measurement, resulting in an accurate
determination of growth rates. Critical device areas like the channel of a power MOSFET can
therefore grown selectively, eliminating the ion implantation damage which reduces carrier
mobilities.
ACKNOWLEDGEMENTS
The authors acknowledge the support of the MURI on Manufacturable Power Switching
Devices, contract manager John Zolper.
REFERENCES
[1]
[2]
[3]
[4]
[5]
[6]
[7]
O. Kardina et al, in Growth and Characterization of SiC Power Device Material,
(Linköping Studies in Science and Technology. Dissertations No. 352 1994) pp. 47
R Rupp et al., in Journal of Crystal Growth, 146(1994) 37-41
N. Kuroda et al., in Extended Abstracts of the 19* Conference of Solid State Devices and
Materials, edited by S. Furukawa, (Tokyo, Japan 1987 ) pp. 227
T. Ueda et al, in J. Crystal Growth, 104, 695 (1990)
H. Matsunami et al., Mater. Res. Soc. Symp. Proc. 162,397 (1990)
H. Matsunami et al., in Amorphous and Crystalline Silicon Carbide, edited by G.L Harris
and C.Y.W Yang, (Springer Proceedings in Physics 1989, Vol.34) pp. 34-39
Chris Thomas et al., Annealing of Ion Implantation Damage in SiC using a Graphite
Mask, presented at the 1999 MRS Spring Meeting, San Francisco, CA, 1999
(unpublished).
178
Properties of 4H-SiC by Sublimation Close Space Technique
S.Nishino, K.Matsumoto, Y.Chen and Y.Nishio
Department of Electronics and Information Science
Faculty of Engineering and Design, Kyoto Institute of Technology
Matsugasaki, Sakyo-ku, Kyoto 606-8585, Japan
ABSTRACT
SiC is suitable for power devices but high quality SiC epitaxial layers having a high
breakdown voltage are needed and thick epilayer is indispensable. In this study, CST
method (Close Space Technique) was used to rapidly grow thick epitaxial layers. Source
material used was 3C-SiC polycrystalline plate of high purity while 4H-SiC(0001) crystals
inclined 8° off toward <1120> was used for the substrate. Quality of the epilayer was
influenced significantly by pressure during growth and polarity of the substrate. A p-type
conduction was obtained by changing the size of p-type source material. The carrier
concentration of epilayer decreased when a lower pressure was employed. Schottky diode was
also fabricated.
INTRODUCTION
SiC is suitable for high power devices because of its wide-bandgap and high thermal
conductivity. To make a SiC device having high breakdown voltage, thick epitaxial layers are
needed. In conventional CVD method ( in which silane and propane are normally used),
epitaxial growth rate is about 3 jum/h . From industrial point of view, rapidly grown thick
epilayers are required. Sublimation epitaxy ( CST : close space technique) has been
demonstrated to grow thick epilayers at higher growth rate [1-5].
A close space technique has two advantages compared to other growth methods. One
is that the growth apparatus has a simple configuration. The other is the ability to keep the
growth system pure since only two materials are needed, one for the source and the other for
the substrate. This growth method had been applied to grow various semiconductors such as
Ge [6]. This technique has also been applied to SiC [7,8]. It is basically the same as
conventional sublimation method [9]. A merit of this configuration is that it minimizes the
area of graphite wall in the crucible. Normally in the conventional sublimation method
unwanted carbon species comes from the wide area of the graphite wall , and then the quality
of epilayers degrades. In CST, unwanted C species are minimized and sublimed vapor is
transferred to the substrate in quasi-thermal equilibrium condition.
We have previously studied homoepitaxy of 4H-SiC by CST[5]. The surface
morphology was observed by optical microscope in the Nomarsky mode and AFM ( Atomic
Force Microscope). Crystallinity and purity of the epilayer were characterized by Raman
spectroscopy. Smooth layers without step-bunching were obtained at a pressure lower than 40
Torr. The surface morphology of the epilayers was mainly influenced by the off-angle of
the substrate and the ambient pressure.
EXPERIMENT
3C-SiC polycrystalline plates of semiconductor grade (MUHSIC:Admap Ltd.) were
used as SiC source material while 4H-SiC (Si-face, C-face) substrates of commercially
179
Mat. Res. Soc. Symp. Proc. Vol. 572 ® 1999 Materials Research Society
obtained wafers inclined 8° off (0001) towards <1120> were used as substrates. Size of the
substrate was 5 mm x 5 mm. In order to have an accurate spacer height between source and
substrate, a graphite spacer with a square opening size of 4 mm x 4 mm was used. The source
and substrate were set in a graphite crucible. A crucible with outer diameter of 34 mm, inner
diameter of 24 mm and 33 mm long including cap was used. The crucible was thermally
shielded by graphite foam. A vertical quartz tube cooled by an air-fan was used as the
reaction tube. The temperatures of top and bottom of the crucible were measured
simultaneously by two-color pyrometers(Chino IR-AQ). The experiments were carried out in
Ar atmosphere. The crucible in the reaction tube was heated by an rf generator at a frequency
of about 45kHz. Our typical growth conditions were as follows: growth temperature of source
material (Tg) 1900-2100°C, temperature gradient ( grad T ) 3.5°C/mm, growth pressure (P)
0.3-100Torr, distance (AX) between source and substrate 1.5 mm.
The 3C-SiC plate was rinsed in HF to remove native oxide from the surface. The
substrate was also cleaned with HF before setting. The crucible was baked at around 2000 °C
in Ar atmospheric pressure before setting the substrate and source plate.
RESULTS AND DISCUSSION
The growth rate of epitaxial layers (Vg) was investigated by varying the substrate
temperature with AX-1.5 mm, P~10 and 100 Torr, and grad T~ 3 °C/mm . The growth rate
increased exponentially with growth temperature. The activation energies calculated by using
an Arrhenius plot was 162 kcal/mol at 100 Torr and 149 kcal/mol at 10 Torr as shown Fig 1.
Those two values are in the range of published data [10, 11]. According to the literature,
growth rate is limited by two mechanisms , one is surface reaction limited and the other is
4.2
4.3
4.4
4.5
4.6
4.7
4
Growth Temperature
1/T [ 10 / K ]
Fig.l Temperature dependence of the growth rate
180
diffusion limited [7]. Growth rate at 100 Torr was limited by diffusion , whereas, it was
limited by surface reaction at pressure of 10 Torr. To verify those rate limited regime, growth
rate dependence on Ar pressure was measured. Growth rate dependence on pressure is shown
in Fig.2. When the pressure was low , the growth rate increased. At lower pressure, the source
was activated and the concentration of chemical species increase. Generally speaking, growth
rate (Vg) is determined by three factors: etching (Ve), diffusion (Vd) and crystallization (Vc)
velocities. Ve and Vc are related to surface reaction rate(Vs). Consequently, Vg is expressed
by the competing processes of Vs and Vd. In Fig.2, growth rate decreased at higher pressure
indicating that the growth rate is limited by diffusion process [7]. However, when the
pressure was low, growth rate increased indicating that growth rate is limited by surface
reaction. This rate limiting process might be related to the different value of activation
energy at different growth pressure as shown in Fig.l.
Surface morphology of the epilayer on (OOOl)Si-face grown at a pressure of 10 Tonwas smooth as observed by optical microscope. Even though surface scratches remained on
the substrate , smooth surface was obtained after the epi-growth. In case of (OOOl)C-face, the
• STEP
1
O NO STEP |
o-
i
I 0
I i
30
<
P
10
"*
0
0.5
1
1.5
2
2.5
3
3.5
Growth Pressure : l/P[l/Torr]
Fig. 2 Pressure dependence of the growth rate
situation was quite different. The surface of as-received C-face was often rough compared to
the (OOOl)Si-face. A lot of polishing scratches were observed on it. To create a smooth
surface of the substrate, two kinds of treatments were carried out. One method was to use
molten-KOH etching and the other was RIE etching using CF4+O2. When the as-received
substrate was used for the growth, the scratched pattern remained in the epilayer even
though 40 mm thick epilayer was grown as shown in Fig.3 (a). Morphology of epilayer on
molten-KOH treated substrate was smooth but micropipes appeared which originally existed
in the substrate as shown in Fig.3(b). Morphology of the epilayer on RIE treated substrate was
also smooth and step-bunching was not observed as shown in Fig.3(c). In these
experiments, surface pre-treatment was a key to obtaining smooth surface. In comparing
the surface morphology of Si-face and C-face, C-face was very sensitive to pretreatment.
However, it is confirmed that the C-face gives a smoother surface, if surface treatment is
carefully done.This difference might be related with surface energies of those faces.
181
100 irni
I
1
Fig.3 Surface morphoplogy of the epilayers depending various pretreatments.
(a) epilayer on as-received substrate, (b) epilayer on molten KOH etching
substrate and (c) epilayer on RIE treated substrate.
Normally, surface energy of Si-face is high compared with C-face[12].
The purity of the epitaxial layers was investigated by Raman spectroscopy. Free
carrier concentration of the epilayer was characterized by measuring the LO-phononplasmon-coupled mode in Raman spectroscopy [13]. LO-phonon peaks at around 970 cm1
shifted to lower wave numbers, when the substrate temperature was increased. This shift
indicates that carrier concentration also decreased by increasing the substrate temperature.
This is related to the site-competition mechanism [14]. As higher concentration of C species
was produced at high temperatures, nitrogen was not incorporated in high C ambient, as
reported in homoepitaxial CVD growth.
Photon Energy (eV)
3.2
3.15
3
5
CM
3800
3850
3900
3950
Wavelength [Ä1
Fig.4 Photoluminescence spectrum of epilayer prepared at 0.3 Torr.
182
We also investigated the purity of the epilayers by performing photoluminescence
measurement at UK. Several sharp peaks, P-series and Q-series due to recombination of
excitons bound to neutral nitrogen donors were observed. Peaks of Po and Qo are zerophonon lines. The Qo peak, which is due to excitons bound to neutral nitrogen donors
substituting C atom on the cubic sites, was observed in the epilayer grown at 0.3 Torr.
However, this peak was not observed in the epilayer prepared at a pressure greater than 40
Torr. Those observations indicate that purer samples were grown at lower pressure. Phonon
replica of free exciton emission is also observed at 2K as shown in Fig.4. Phonon replica of
free-exciton band is also observed. The relative intensity of I-series (free exciton) and Q-series
(bound exciton) suggests concentration of nitrogen donors in epilayers [15]. Peak intensity of
the ratio between I series and Q-series was 3xl0~2, from which the carrier concentration of
epilayer is estimated to be the order of 1017 cm"3 [15].
P-type epilayer was made using p-type source material. A small amount of p-type chip
wafer ( 2mm x 2mm) was placed on the source plate of SiC (undoped n-type). The growth
condition was the same as for the undoped case. Photoluminescence of the epilayer is shown
in Fig5.D-A pair emission peaks ( BO,BTO, LO„CTO, LO)were observed and indicate that Al
was incorporated in the epilayer in this growth method. When a small amount of p-type
platelet was used, D-A pair emission band shifted to much lower energy side compared to
when entire p-type source was used. This is the result of large separation between donor and
acceptor achieved by adding a small amount of p-type source. This suggests the a possibility
of controlling doping concentration by changing the size of the p-type source materials.
Photon Energy (eV)
3.2
3.1
3.0
2.9
2.8
platelet
s
380 390 400
410
420
430 440
450
460
Wavelength [ Ä ]
Fig.5 Photoluminescence spectra of p-type epilayers
prepared by using different size of source plates.
183
To verify the usefulness of CST for large size wafer, we used 4H-SiC substrate of 1
inch size. One inch diameter spacer was made and the periphery (1 mm) of the substrate was
held by the spacer. Growth conditions were the same as described previously, and 40
micron-thick epilayer was obtained. Surface morphology was smooth and the entire area was
the same polytype as the substrate. However, central part was thicker than the periphery due
to the temperature gradient. The temperature uniformity will be optimized by changing the
crucible shape.
Schottky diode was also made on the n-type epilayer. Ohmic contact was made by Al
evaporation and annealing at 400 °C. Au was evaporated as Schottky metal. I-V characteristic
showed breakdown voltage of 40 V, built-in voltage of 2.3 V and n-value of 1.6. Carrier
concentration estimated by C-V measurement was about 7xl016 cm3 .
CONCLUSION
We have obtained smooth 4H-SiC homoepitaxial layers at a high growth rate of 40 (xm/h.
Surface morphology depended on the polarity of the substrate. When careful pretreatment
was done for the C-face, smooth and step-bunch-free surface was obtained on C-face.
Excitons bound to neutral nitrogen donors were observed by photoluminescence measurement
at 2 K, indicating high quality epitaxial layers. A low carrier concentration of 5xl016 cm3
was obtained at low pressure. P-type epilayer was obtained by adding small amount of p-type
source . The CST is applicable for large size epitaxial layer growth also.
ACKNOWLEDGEMENT
The authors would like to thank to Dr. H. Harima and Dr. S. Nakashima at Osaka University
for comment on Raman data. The authors also thank to Dr. A. Henry and Dr. E. Janzen in
Linkoping University for LTPL measurement. This work was partially supported by a Grantin-Aid for Science Research No.09450011 from the Ministry of Education, Science and
Culture, Japan and Kansai Electric Power Company, Ion-Engineering Ltd, and NEDO /FED.
REFERENCES
1) T.Yoshida et al. Proceedings of ICSC M-N97, Trans Tech Publications (1997), 155.
2) A.K.Georgierva et al.Proceedings of ICSC M-N97, Trans Tech Publications (1997)147.
3) M.Syvajarvi et al. Proceedings of ICSC1-N97, Trans Tech Publications (1997)143.
4) S.Nishino et al. Mat.Res.Soc.Symp.Proc.vol.483,(1988)307.
5) S.Nishino et al. ECSCRM98, Abstract (1998, Montpellie, France), .
6) F.H.Nicoll, J.Electrochem.Soc.l 10,(1963)1165.
7) S.KXilov etal. Phys.Stat.Sol.(a)37(1976)pp.l43.
8) M.M.Anikin et al. Material Science and Engineering,Bl 1(1992)113.
9) Yu.M.Tairov et al.
J. Cryst. Growth 36 (1976), 147.
10) LK.Kroko, J.Electrochem.Soc.ll3(1966)801.
11) J.Drowart, G.de.Maria, J.Chem.Phys.41(1958)1015.
12) E.Pearson et al J.Crystal Growth, 70(1984)33.
13) H.Harima, S.Nakashima, and T.Uemura, J.Appl.Phys.78, (1995)1996.
14) DJ.Larkin et al. Appl.Phys.Lett. 65,(1994)1659.
15) A.Itoh "Control of electrical Properties of 4H-SiC Brown by Vapor Phase Epitaxy for Power
Electronic Aplication" Ph.D. Thesis, Kyoto University, Kyoto (1995).
184
EFFECT OF Ge ON SiC FILM MORPHOLOGY IN SiC/Si FILMS GROWN BY
MOCVD
W.L. Sarney1, L. Salamanca-Riba1, P. Zhou2, M.G. Spencer2, C. Taylor2, R.P. Sharma3, and K.A.
Jones4
'Dept. of Materials & Nuclear Engineering, University of Maryland, College Park, MD
2
Materials Science Research Center of Excellence, Howard University, Washington D.C.
3
Center for Superconductivity, University of Maryland, College Park, MD
4
U.S. Army Research Laboratory, Adelphi, MD
ABSTRACT
SiC/Si films generally contain stacking faults and amorphous regions near the
interface. High quality SiC/Si films are especially difficult to obtain since the
temperatures usually required to grow high quality SiC are above the Si melting point.
We added Ge in the form of GeH2 to the reactant gases to promote two-dimensional
CVD growth of SiC films on (111) Si substrates at 1000°C. The films grown with no
Ge are essentially amorphous with very small crystalline regions, whereas those films
grown with GeH2 flow rates of 10 and 15 seem are polycrystalline with the 3C structure.
Increasing the flow rate to 20 seem improves the crystallinity and induces growth of 6H
SiC over an initial 3C layer. This study presents the first observation of spontaneous
polytype transformation in SiC grown on Si by MOCVD.
INTRODUCTION
Growing high quality SiC films on Si is challenging due to large differences in their
thermal expansion coefficients and lattice constants. SiC grown on Si usually forms in the cubic
polytype, typically with high densities of stacking faults and dislocations. Films that have a large
lattice mismatch with their substrate grow by either the Volmer-Weber or the Stranski-Krastanov
mode, both of which lead to island formation.1 Growth of SiC on Si is limited by the Si species.
It is believed that the Si is first absorbed on the surface, and then it migrates to the proper site
and is subsequently carbonized on either terraces or at step edges.2 In an attempt to improve the
quality of SiC films grown on Si at 1000° C, we added Ge in the form of GeH2 during growth.
We find that adding Ge promotes two-dimensional growth, improves film quality, and at high
concentrations may induce a cubic-to-hexagonal polytype transformation.
EXPERIMENT
The SiC/Si films were grown in a commercial vertical rotating disk CVD reactor. All of
the samples were grown with a constant Si/C ratio under identical conditions with the exception
of the Ge concentration. SiH3, C3H8, GeH2, and H2 were reacted at 1000CC on Si (111)
substrates. After growth the samples were characterized ex-situ by AFM, X-ray, RBS, and high
resolution transmission electron microscopy (HRTEM). The TEM samples were prepared using
tripod polishing and ion milling, and were observed in a JEOL 4000 FX operated at 300 KV.
RESULTS
Transmission electron microscopy was used to obtain high resolution lattice images and
diffraction patterns from each sample. The first sample was grown with 0 seem Ge, but due to
185
Mat. Res. Soc. Symp. Proc. Vol. 572 ° 1999 Materials Research Society
the presence of GeH2 in the CVD reactor, it was unintentionally doped with a trace amount of
Ge. Figure 1 is the HRTEM image and diffraction pattern from this sample. The distinct spots in
the (1 TO) diffraction pattern correspond to the Si substrate. The very light, diffuse ring seen in
the diffraction pattern corresponds to the SiC film and indicates that the film is primarily
amorphous. The elongated spots over the diffuse ring, however, indicate that some crystallites
are formed and they have a preferred orientation with respect to the substrate. The SiC/Si
interface in the lattice image is visible and denoted by the arrowheads. The amorphous-like
contrast of the film in Fig. 1 could not be the result of ion milling since SiC is much harder than
Si and the Si lattice in the image is intact.
Fig. 1 HRTEM image and diffraction pattern from the sample grown with a trace amount of Ge.
Figure 2 shows that the crystallite size increases and the preferred orientation in the film
improves with increasing Gelt flow. The lattice image of the film with 10 seem Ge in Fig 2.
shows regions with lattice fringes for SiC on a background of amorphous-like SiC. The
elongated spots and rings in the (112) diffraction pattern shown as inset to Fig. 2 are less diffuse
and sharper than the spots in Fig. 1, This result indicates that the film is polycrystalline 3C SiC
with some preferred orientation of the crystallites with respect to the substrate. The SiC/Si
interface in the lattice image is very abrupt and obvious. The TEM lattice image of the sample
grown with 15 seem Ge showed slightly improved crystallinity, but otherwise was not
appreciably different from the sample with 10 seem.
186
Fig. 2 HRTEM image from the sample grown with 10 seem Ge. The diffraction pattern indexes
denoted with the subscript s refer to the Si substrate.
Adding 20 seem GeH2 to the reactor led to the multi-polytype structure shown in Figs. 3
and 4. The diffraction pattern (Fig. 5) shows the presence of the 3C and 6H polytypes, and
corresponds to the (110) zone axis for the cubic structure and the (0110) zone axis for the
hexagonal structure. The initial structure is 3C SiC, which ends abruptly after approximately 14
nm from the interface and is followed by a 19 nm transistion region consisting of mixed cubic
and hexagonal polytypes. The remaining 33 nm of film is predominately 6H SiC with some
cubic regions, as shown in Fig. 3. The hexagonal structure is evident by the zig-zag stacking
sequence seen in the high resolution image (upper part of Fig. 3) and the presence of the (0001)
spots in the diffraction pattern (Fig. 5).
Although the SiC/Si interface is abrupt in all of the samples, it became more visibly
obvious for increasing amounts of Ge. Figure 4 shows that the sample containing 20 seem Ge
has a white band at the interface while Figs. 1 and 2 do not. Although surfactants are intended to
float at the growth front without incorporating into the film, our RBS results indicate that a small
amount of Ge is present at the interface. The white band may represent an interfacial layer of
SiGe. Although all of the Ge does not float to the surface, it certainly improves the crystalline
quality of the film. Similarly, Hatayama, et al, observed Ge at the SiC/Si interface in samples
grown by dimethylgermane source molecular beam epitaxy.3 In the case of MBE films, however,
no transition to 6H SiC was observed.
187
Miefe;
^
sjc]
ISifntcrfüceH
Fig. 3 HRTEM image from the sample grown with 20 seem Ge. The image shows from bottom
to top the Si substrate, the 3C SiC region, the transition region of predominately 6H SiC,
and the remaining region of 6H SiC.
Fig. 4 Higher magnification HRTEM image from the region close to the film/substrate interface
of the sample grown with 20 seem Ge. The upper part of the figure is the transistion
region.
188
Fig. 5 Diffraction pattern from the sample grown with 20 seem Ge. The indexes denoted with
the subscript s refer to the Si substrate. The spots labeled with three and four indexes
represent the cubic and hexagonal polytype of SiC, respectively.
0
5
10
15
20
Ge flow (seem)
Fig. 6 X-ray FWHM from SiC films vs. GeH2 flow in the reactor.
189
Ex-situ AFM results indicate that the surface roughness of the film grown with a trace
amount of Ge is approximately 16 nm. The roughness decreases with increasing Ge content up to
15 seem and then increases again for the film grown with 20 seem Ge. The increase in roughness
probably results from the presence of some cubic SiC in the 6H SiC layer. The samples with 10
and 15 seem Ge contain larger amorphous regions, which would tend to be smoother than the
crystalline regions.
X-ray experiments show that the rocking curve FWHM narrows as a function of Ge
concentration, as shown in Fig. 6. The x-ray linewidth is reduced by almost a factor of 2,
indicating that the crystallite size increases with increasing Ge, in agreement with the TEM
results.
CONCLUSION
To our knowledge, there are no reports of SiC films grown on Si with a sudden abrupt
transition from the 3C to the 6H polytype. The TEM images show that the crystallinity of 3C SiC
films grown on Si(l 11) by CVD is improved by adding small amounts of Ge during film growth.
We have obtained fairly high-quality 6H-SiC films and improved the crystalline quality of 3C
SiC films grown on (111) Si at 1000° C by adding 20 seem Ge. The quality of the SiC films may
be improved further by optimizing the growth conditions, particularly the Ge concentration. This
growth method could have very important implications in the fabrication of substrates of SiC for
the growth of GaN.
ACKNOWLEDGEMENTS
This work was supported by MRCP Army Grant No. DAAL 019523530.
REFERENCES
1. E. Tournie, K. Ploog, Thin Solid Films 231,43 (1993).
2. T.Kimoto, H. Nishino, W-S.Yoo, H. Matsunami, J. Appl. Phys., 73 , 726 (1993).
3. T. Hatayama, N. Tanaka, T. Fuyuki, H. Matsunami, J. Elec. Mater., 26,160 (1997).
190
Properties of Heteroepitaxial 3C-SiC Layer on Si Using Si2(CH3)6 by CVD
Y. Chen, Y. Masuda, Y. Nishio, K. Matsumoto, and S. Nishino
Department of Electronics and Information Science,
Faculty of Engineering and Design, Kyoto Institute of Technology,
Matsugasaki, Sakyo-ku, Kyoto 606, JAPAN
ABSTRACT
Single crystal cubic silicon carbide ( 3C-SiC ) has been deposited on Si(100) by
atmospheric CVD at 1350°C using Si2(CH3)6. The 3C-SiC epilayers were characterized by
XRD, Raman scattering and photoluminescence (PL). The 3C-SiC distinct TO near 796 cm"1
and LO near 973 cm"1 were recorded by Raman measurement. The PL spectra of SiC films at
UK included the nitrogen-bound exciton (N-BE) lines, the "defect-related" W band near 2.15eV,
and 2.13eV peak corresponding to D-A pair recombination as well as the "divacancy-related" Dl
peak at 1.97eV. The thickness dependences of Raman and PL measurement were made and it
was observed that tensile stress and strain in films decrease with increasing film thickness.
Electrical properties of the films were measured by making schottky diodes and using Van der
Pauw method. Above 300K, the electron mobility changed as \m - T _1-45 ~ _1-56 and the
highest mobility was about 400 cn^vls"1 at room temperature. In 3C-SiC the scattering
processes are affected prominently by acoustic scattering in this temperature range.
INTRODUCTION
The heteroepitaxial growth of 3C-SiC on Si substrates by CVD method has been
investigated intensively. This is because commercially available 6H and 4H-SiC wafers still have
some problems such as small substrate size and defects like micro-pipes. Atmospheric pressure
CVD (APCVD) method has been widely used for the growth of 3C-SiC on Si substrate. It has
been reported that SiC films can be grown on Si substrate at a relatively low temperature in
APCVD system by pyrolyzing HMDS (hexamethyldisilane: Si2(CH3)6), a single-source
organosilane precursor which contains direct Si-C bonds. Most of the work on the growth of SiC
films using HMDS reported SiC film thickness of about 5^m on Si(lll) substratefl, 2].
However, because the thermal stress in the films on Si(lll) is larger than that of Si(100), a
number of microcracks exit easily in 3C-SiC films on Si(lll)[3]. It is clear that using a (100) Si
substrate instead of a (111) Si substrate will reduce thermal stress. A higher quality crystallinity
3C-SiC films can be obtained on Si(100).
In this work, we report the growth of high quality single crystalline 3C-SiC films on
Si(100) up to n.5\xm using Si2(CH3)6 + H2 gases by APCVD. The resulting film surfaces are
all mirror-like. The growth rate of 3C-SiC films was about 4.3nm/h. This growth rate is rather
high compared with SiH4 + C3H8 system. The 3C-SiC epilayers were characterized by XRD,
Raman scattering and PL. We successfully obtained high quality single crystal 3C-SiC films and
observed the complete line structure of PL at 1 IK. Thickness dependences of the Raman and PL
measurement were determined and stress-related peaks were obtained. Electrical properties of
191
Mat. Res. Soc. Symp. Proc. Vol. 572
e
1999 Materials Research Society
the films were also measured by making Schottky diode and using the Van der Pauw method.
EXPERIMENTAL
The CVD system used in this study was made from a 50mm diameter air-cooled
horizontal quartz tube connected to a rotary pump. A SiC-coated graphite susceptor, supported
by a quartz boat, was heated by RF-induction, operating at 450KHz and 5KW. For Raman and
PL measurements, n-type Si (100) substrate with a resistivity of 0.01 - 0.02 ß • cm were used,
while films grown on p-type Si(100) substrate was used for Hall measurement. The silicon
substrate, cut into 25 X15 mm pieces, was placed on the graphite susceptor. The temperature of
the substrate was measured by means of an optical pyrometer focused at a hole in the susceptor
through a window at the end of the tube. HMDS flow rate was controlled by bubbling H2 gas
through the liquid HMDS. Manual flow control valves were used to control the flow rate. The
overall growth process consisted of three steps, (a): Si substrate was etched at 1175°C in
HCl(63.1sccm) + H2(lslm). (b): Si substrate was carbonized for 3 minutes at 1350°C in
C3H8(1.0sccm) + H2(lslm). This procedure provides a so-called "buffer layer" for subsequent
CVD. This is important to obtain a good crystallinity of SiC films, (c): After carbonization,
HMDS gas was quickly introduced for growth at 1350°C. HMDS flow rate was 0.5 seem and H2
flow rate was 2.5 slm. Usually, for a low flow rate of HMDS, the final surface of the 3C-SiC
films was mirror-like with some haze, and hillocks in the films appeared and became denser with
increase of the HMDS flow rate. The 3C-SiC epilayers on Si(100) were characterized by X-ray
diffraction (XRD), Raman scattering (excitation by argon-ion laser: 5145A) and
photoluminescence (PL: excitation by He-Cd laser with a line at 3250 Ä ) at UK. Electrical
properties of the films were also measured by making Au - Schottky diode and Van der Pauw
method.
RESULTS AND DISCUSSION
The epilayer surface orientations of each sample were characterized by XRD, and were
the same as the substrate Si(100). Only one peak appeared at 26=41.5 degree corresponding to
the SiC(200) peak. The full width at half maximum(FWHM) of X-ray rocking curves changed
from 1.04 to 0.19 degree with increasing film thickness from 0.5 to 17.5^m. This indicates that
the crystallinity of the 3C-SiC epilayers improved with increasing film thickness.
Raman scattering measurements were performed at room temperature. For every sample,
both the 3C-SiC TO( T ) phonon near 796cm"! and the 3C-SiC LO( T) near 973cm"1 were
recorded. The intensity of TO phonon peak is less than that of the LO phonon peak. With
increase of film thickness, the TO( T) and LO( T) peaks become sharper and stronger, which is
characteristic of improved crystal quality[4].
Fig.l shows the Raman spectra of 9.1fim and 17.5nm 3C-SiC films on Si(100) and
9.1^ free standing films removed from the Si substrate. For 9.1(xm films, the line position of
the 3C-SiC LO( T ) phonon shifts from 973.2cm"1 to 974.6cm"1 after the Si substrate was
removed. This indicates that the existence of the tensile biaxial stress in CVD films on Si
substrate, probably due to the thermal coifficent mismatch. A tensile stress in the 3C-SiC films
192
on Si(100) shifts the LO( T) phonon to lower energy. The Raman shift between the 3C-SiC on
Si and free 3C-SiC films is ^2cm_1. This corresponds to a biaxial stress of 0.4 -1.0 GPa and an
inplane strain of about 0.1% - 0.2% [4]. After removal of Si by HF/HNO3 l-'l etch, the intensity
of the TO and LO phonon peaks increase dramatically, and the TO phonon peak becomes much
more intense relative to the LO phonon peak. Both effects are the result of the strain relief in the
film[5]. Additionally, the LO phonon peak of 17.5(xm films on Si shifts to a higher frequency
relative to that of 9.1\im films on Si. This means that the tensile stress in films decreases
gradually with increase of the film thickness.
974.6cm-1
973.2cm"1
9.1nm free films
17.5um SiC films on Si
9.1 Um SiC films on Si
_L
800
850
900
950
Raman shift (cm')
1000
Fig.l The Raman spectra of 3C-SiC film
Fig.2 shows the CVD 3C-SiC/Si PL spectra for SiC films( 3.1 to 17.5nm ) at UK. The
spectra have a nitrogen-bound exciton (N-BE) zero phonon line No, one-, and two-phonon lines
and D-A (donor-acceptor: N-Al: Bo) pair recombination peak as well as Dl line[6,7]. The Dl
center no-phonon line at 1.97 eV, which is caused by divacancy related center, appears and
becomes stronger with increasing thickness. This is similar to that reported by Powell et al. [6]. It
indicates that a large number of dislocations and other extended defects near interface region are
reduced, and recombination due to point defect complexes becomes more prominent gradually in
the thicker films.
A peak at 2.13eV is observed in all samples. The peak energy corresponds to D-A pair
recombination in 3C-SiC[7]. The peak was confirmed by dependences of spectra on the
excitation intensity and on the measurement temperature. The peak energy of D-A recombination
shifts to higher energy side with increasing excitation intensity or temperature.With increases of
the excitation intensity and temperature, the number of neutral donors decreases more quickly
than that of neutral acceptors, because the donor level (53.6meV) is shallower than the acceptor
level (269meV). The recombination ratio of close D-A pairs to distant D-A pairs increases, so
193
that the contribution of Coulomb energy between D-A pairs becomes stronger and the peak
energy shifts to the high energy side.
3
17.5nm
S
£
öiöpjn
3.1nm
I „L I
1.8
1.9
2
2.1
2.2
2.3
I
I
I
2.4
photon energy (eV)
Fig.2 PL spectra of the 3C-SiC films on Si(100) at 1 IK
A broad band at ~ 2.15eV(W band) is also observed when the SiC film thickness is
thinner than about 3.1^m, which is caused by the dislocation and extended defects near the
interface region and heavy deformation in these films [6]. The energy region of the W band
around 2.15eV overlaps the two phonon energy region of the N-BE. But it is seen that the two
phonon replica of N-BE is enhanced and the W band decreases at greater film thickness. This
fact shows that a stronger strain exists in the thinner films. The strain is relaxed with increase of
the film thickness, corresponding to the Raman result. The N-BE line structure is developed and
No and one phonon replicas becomes stronger in the thicker films. The photoluminescence
spectra of the thicker 3C-SiC films consist of much stronger N-BE lines and stronger Dl line.
The result shows that the density of dislocations and extended defects decreases and the pointdefect complexes become more prominent, (he crystalline perfection of CVD 3C-SiC films has
been improved with increasing film thickness [6].
Hall measurements were done by using the Van der Pauw method. SiC films(5x5 mm)
removed from the Si substrate were used. Four Al electrodes of 0.5mm were deposited on the
epilayers and ohmic contact was obtained. Hall measurement were performed from 80K to
500K. All the epilayers obtained showed n-type conduction. The carrier concentrations were 1.5
X1017 at 17.5nm and 3 X1017 cm-1 at 8.3nm.
In Fig. 3, the Hall mobilities of 8.3nm and 17.5(jm free films are shown as a function of
temperature. The Hall mobility obtained was about 400 cnr^vVl at room temperature. Above
194
300K, the mobility decreases and changes as HH - T -1-45 - -1-56. It has been suggested that in
3C-SiC the scattering processes are affected prominently by acoustic phonon scattering in this
temperature range[8].
1000
i
-I
i i |
1—I—I—I I I
-1.45 --1.56
m,-T'
>
u
8.3um
17.5um
A
•
K
i 100
^v
^V
1000
100
T(K)
Fig.3 Temperature dependencies of Hall mobilities of non-doped crystals.
The capacitance-voltage characteristics of schottky barrier contacts were measured at a
frequency of 1MHz. The voltage dependence of the Au-3C-SiC diode capacitance is shown in
Fig.4. As shown in Fig.4, the plots are almost linear. The carrier concentration calculated
decreases from 7.3 X lO^toS.QX 1017for 2.6Mm and 6.6pm thickness, respectively, which
is in good agreement with Hall measurement. The barrier height calculated from the C-V
measurement is about 0.9 - 1.35 V.
2.5 10'
-r-r-r-i ] i
i i i
| i
i
i
i
1 ' ' ' '
i
i
i
i
-2.6jun
3.16)im
4.3flm
X
-A- - 6.6\im
2 10'
; Nd=3.9xl017cm'3
|
i
i
i,M i
-_
'I8
r
a 1.5 10'
I
^S*/^
1 10'
*-0
•-
^^^to
^^%^_
^^
^<Sv
5 10*
^^^b>
rNd=7.3&10"cm"3
.
" ....
I
-3
....
1
,
•2
N.
. . .
L^
l . . . .
"
^^
7>r->.rKi . .'
-1
0
Voltage (V)
Fig.4 C-V characteristics of Au - 3C-SiC schottky barrier contact
195
CONCLUSION
To summarize, a high quality single crystal films of 3C-SiC have been obtained using
HMDS by APCVD. The results indicated that the defects and tensile stress in the films on Si
decrease and the crystalline perfection has been improved with increase of film thickness. The
highest mobility was 400 cmSV-V1 at room temperature. The electron mobility changed as HH
_ T -1-45 - -1.56 above 300K. Compared with the SiC polytypes (|XH ~ T -2-0 - -2.6), the mobility
of 3C-SiC is still fairly large at high temperature. It suggests that 3C-SiC is a promising material
for the devices operated at high temperature.
REFERENCES
[1] K. Takahashi, S. Nishino, J. Saraie, J. Electrochem. Soc. 139 (1992) 3565-3571.
[2] C. H. Wu, C. Jacob, X. J. Ning, et al., J. Cryst. Growth 158 (1996) 480-490.
[3]H. P. Liaw and R.F. Davis, J. Electrochem. Soc.:SOLID-STATE SCIEVCE AND TECHNOLOGY,
Vol. 131 (1984) 3014-3018
[4] Z. C. Feng, A. L. Mascarenhas, W. J. Choyke, et al., J. Appl. Phys. 64(6) (1988) 3176-3186.
[5] J. A. Freitas, S. G. Bishop, A. Addamiano, et al., Mat. Res. Soc. Symp. Proc.46. 581- 586.
[6] W. J. Choyke, Z. C. Feng, J. A. Powell, J. Appl. Phys. 64 (6) (1988) 3163-3175.
[7] W. J. Choyke, L. Patrick, Phys. Rev. B2 (1970) 4959.
[8] K. Sasaki, E. Sakuma, S. Misawa, S. Yoshida, S. Gonda, App. Phys. Lett. 45(1), (1984) 72-73.
196
CHARACTERIZATION OF P-TYPE BUFFER LAYERS FOR SiC MICROWAVE
DEVICE APPLICATIONS
A.O. KONSTANTINOV, S. KARLSSON, P.-Ä. NELSSON, A.-M. SAROUKHAN,
J.-O. SVEDBERG, N.NORDELL, C.I. HARRIS, J. ERIKSSON* and N. RORSMAN*.
Industrial Microelectronics Center, IMC, S-164 40 Kista, SWEDEN
'Chalmers University, Dept. of Microwave Technology, Göteborg, SWEDEN
ABSTRACT
Low-loped p-type silicon carbide buffer layers are grown by chemical vapor deposition on
conducting and semi-insulating substrates. Capacitance-voltage and electrical admittance
techniques are developed for accurate non-destructive characterization. The electrical admittance
techniques suggested are capable of measuring the resistivity in a very wide range, up to 7 orders
of magnitude. MESFET devices using thick buffer layers on conducting substrates are reported
with Ff&A GHz and Fmal=32 GHz.
INTRODUCTION
Low-doped epitaxial p-type SiC is commonly used as buffer-layer material in power microwave
transistors. Silicon carbide microwave MESFETs formed on conducting substrates employ lowdoped buffer layer to minimize conductive and capacitive losses [1,2]. MESFET devices formed
on semi-insulating substrates often use p-type buffer layers to minimize the influence of
substrate impurities and of short-channel effects [2,3]. In the present work we report on the
growth and electrical characterization of low-doped p-type layers. We also report on the
application of low-doped p-buffer layers in SiC microwave device technology.
EXPERIMENT
3xio18[
^^w
<"> \
Epitaxial growth has been performed by chemical vapor
deposition (CVD) using two types of CVD reactors: a
single-wafer horizontal reactor built by Epigress AB [4]
and a multiwafer vertical TurboDisk reactor supplied by
2 kHz
10 kHz
EMCORE, Inc. Both reactors are operated at a low
100 kHz
1 MHz
pressure of about 200-400 mBar. The silicon and carbon
0
0.4
0.8
1.2
precursors used are silane and propane.
depth (jim)
Trimethylaluminum (TMA) is used as Al dopant source.
Nickel Schottky barriers 0.4 mm in diameter were
Figure 1. Acceptor concentration profiles for a p-type
4H SiC layers grown on a conducting n-type substrate.
deposited onto the front layer side. A large-area metal
contact on the front side or metallized wafer backside
was used as an Ohmic contact. Multifrequency capacitance-voltage measurements and
measurements of the electrical admittance were performed using the HP 4284A Precision LCR
Meter either in a shielded probe station or in a cryostat.
The results of room-temperature C-V measurements were processed using the standard
technique to extract the doping profiles. The resulting profiles for a p-type layer grown on a
conducting n-type substrate are plotted in Fig. 1. It is seen from the plots that the data tend to
converge for low measurement frequencies. By contrast, 1 MHz measurements yield a very large
197
Mat. Res. Soc. Symp. Proc. Vol. 572 ° 1999 Materials Research Society
error for the layers grown on n-type wafer. Lower frequencies are therefore preferable for
accurate measurement of doping profiles.
The frequency dispersion of electrical admittance can provide information on the layer
conductivity. Plotted in Fig. 2 are the measurement results for the parallel capacitance Cp and the
normalized conductance G/2%F for the layer with the doping profile plotted in Fig. 1. The
measurements were performed at zero bias. Only about a half of actual experimental plots are
shown in the figure. The peaks of two clearly different types are seen on the plots. The first type
appears in the low-temperature range, close to the freezeout point of aluminum acceptors, Fig.
2a. The other type can be observed even at room temperature, as those seen in Fig. 2b.
12r-T
*„
^0*w.
^X
\ \ 200
\" a 24D.
160
"*>.-1
(a)
\
\
S
\
v
ioo\
9
1t(A
\
120
\
(b)
\280
• W
\
\„,
-_ ^
1&
160
200
240
280
10*
omega (Hz)
omega (Hz)
Figure 2. Parallel capacitance (dashed lines) and conductance (solid lines) spectra of a low-doped p-type layer on ntype 4H SiC substrate measured at 100-280K. Circles are the simulation results.
SIMULTATION RESULTS AND DISCUSSION
Schottky metal
A simulation has been performed in order to understand the
origin of admittance spectra and to extract material parameters
from the measurement data. According to the simulation results,
the conductance peak originates from the conduction losses in
the high-resistivity layer. Two clearly different regimes of AC
current transport are observed depending on the layer resistivity.
At low temperatures the layer resistivity is very high. The
conduction losses are determined by vertical current transport
through the layer, whereas the lateral spreading of AC current is
Figure 3 Current pathways for the
regimes corresponding to low- and
high-temperature admittance peaks.
negligible. The conductance peak appears close to the dielectric
relaxation frequency, as a result of the transition of the p-type
material from conductor to insulator. The conductivity CTcan be
determined from the peak position F'°eak using the relationship:
a = 2nF'lk££0(w0 + wMS + wPN)
f <1
.1
'\
where £ is the permittivity, wMs, wPN and w0 are the thickness of the Schottky-barrier and p-n
junction depletion regions and of the non-depleted layer portion respectively. The parallel
capacitance will increase from the low-temperature dielectric limit, eS l(w0 + wus + wPN ) to
approximately Vi of the Schottky-barrier capacitance, eS l(wMS +wm), where S is the contact
198
(1)
area. The factor of Vi originates from the series connection of the p-n junction and the Schottky
barrier, as it is shown in Fig. 3a.
The high-temperature conductance peak originates from the conduction losses due to
lateral current transport. At a high temperature the lateral transport effectively eliminates the
contribution of the p-n junction to the device impedance, as it is shown in Fig. 3b. The
capacitance increases approximately twofold as result, and a peak of conduction losses appears
in the transition region. The layer conductivity can be determined from the conductance peak
position F*ak using the results of the numerical calculation performed:
0=
2JiF;akee0
K
w0wMS
where rc is Schottky contact radius w0 the non-depleted layer thickness, wMS is the depletion
region width for the Schottky barrier and the p-n junction respectively. Numerical factor K
depends on the ratio of the p-n junction to the Schottky-barrier depletion layer thickness. The
simulation yields K= 1.29 for wPN I wMS = 1, K- =1.61 for wm lwMS = 1.25 and £=1.93 for
-/w„
(2)
= 1.5.
A good quantitative agreement is observed between
theoretical and experimental spectra for both capacitance and
conductance, as one can see it from Fig. 2. Using the data
plotted in Fig. 2 we have determined the temperature
dependence of the layer conductivity for the temperature
range from 100 to 300K. The results are plotted in Fig. 4. The
slope of the dependence corresponds to the apparent
activation energy of 185 meV, which is close to the
activation energy of aluminum acceptors in 4H SiC. The
admittance results are in agreement with the data obtained for
this sample using the SMS technique.
1
1000/r (K' )
Electrical admittance measurements appear to be a
Figure 4. The temperature dependence of
convenient characterization tool for low-doped p-type
conductivity extracted from the admittance
epitaxial layers of silicon carbide. No processing apart from
data using Eqs. (1)and(2)
Schottky barrier deposition is required to determine the major
electrical properties such as the dopant concentration and conductivity. The technique can
therefore be used as a tool for non-destructive characterization and process monitoring.
MATERIAL APPLICATION IN MESFET TECHNOLOGY
Low-doped p-type buffer layers were used for fabrication of microwave MESFET
devices. The layers were unintentionally doped in the mid-1015 cm"3 range, grown in the
multiwafer EMCORE reactor to a thickness of 15 um. The dominant acceptor is boron as it was
determined by electrical admittance and SIMS measurements. The active layer was formed by
nitrogen implantation, and the capping n+ contact layer was grown by CVD. The wafers were
processed to form recessed-gate microwave MESFETs. The devices had a split gate
configuration, which is shown in Fig. 4. Sintered nickel is used as an Ohmic contact to source
and drain regions, and Ti/Au metal stack was selected as Schottky-barrier contact.
The gate length is varied in the range 0.25-2 urn, the drift region length is 1.5-2.5 urn.
The threshold voltage appeared to be lower than expected, only about 2 Volts. The reason for
this is presumably related to incomplete nitrogen activation. The on-state currents are
199
correspondingly low, about 40 mA/mm for the devices with a short gate length. The MESFETs,
nevertheless, show a good microwave performance. The cut-off frequencies F, and /w are as
high as 8.4 and 32 GHz for the gate length of 0.25 um. Only a marginal difference is observed
between the 0.25 and 0.5 um-gate devices; the latter have F,=7.8 GHz and Fma=31 GHz. The
cut-off frequencies, however, decrease more rapidly with further increase of the gate length.
Summary of the frequency performance measured at a drain bias of 40 V is given in Table I. The
frequency performance achieved is comparable with the best results reported for conducting
substrates [5], in spite of the fact that the channel doping is yet below the optimal value. Further
performance improvement can be achieved through optimization of the channel implant dose.
Table I.
Dependence of the frequency
performance on gate length.
Channel width is 200 urn.
gate
length
(Mm)
0.25
0.5
0.75
1
Figure 5. Layout of Microwave
MESFETs
F,
(GHz)
Tmax
8.4
7.8
7.3
5.9
32.0
30.7
28.3
21.2
(GHz)
SUMMARY
In conclusion we have developed the techniques for electrical characterization of low-doped ptype buffer layers for SiC microwave device applications. A very wide range of layer resistivity
is available for characterization with the admittance techniques suggested, up to 7 orders of
magnitude. Implanted-channel MESFET devices with high operation frequencies have been
fabricated using thick p" buffer layers on conducting substrates.
ACKNOWLEDGEMENTS
The authors would like to thank M. Linnarsson for performing the SMS measurements and to
ABB Corporate Research for access to the EMCORE TurboDisk deposition system. The work
was supported by the EU TELSiC program and by the Swedish Board for Industrial and
Technical Development (NUTEK). The Swedish Nanometric Facilities at Chalmers University
were used for electron-beam lithography.
REFERENCES
1. J.W. Palmour, C.E. Weitzel, K.J. Nordquist and C.H. Carter, IOP Conf. Ser. 137,495 (1994).
2. S.T. Allen, R.A. Sadler, T.S. Alcorn, J.Sumakeris, R.C. Glass, C.H. Carter, and J.W. Palmour,
Materials Science Forum 264-268,953 (1998).
3. O. Noblanc, C. Arnodo, E. Chartier and C. Brylinski, Materials Science Forum 264-268, 949
(1998)
4. Epigress AB, Ideon Science Park, S-223 70 Lund, Sweden.
5. C.E. Weitzel, Materials Science Forum 264-268, 909 (1998).
200
OPTICAL CHARACTERIZATION OF SiC WAFERS
J.C. BURTON*, M. POPHRISTIC*, F.H. LONG*, I. FERGUSON**
*Chemistry Dept, Rutgers University, Piscataway, NJ 08854, fhlong@rutchem.rutgers.edu
**EMCORE Corp., Somerset, NJ 08873
ABSTRACT
Raman spectroscopy has been used to investigate wafers of both 4H-SiC and 6H-SiC.
The two-phonon Raman spectra from both 4H- and 6H-SiC have been measured and found to be
polytype dependent, consistent with changes in the vibrational density of states. We have
observed electronic Raman scattering from nitrogen defect levels in both 4H- and 6H-SiC at
room temperature. We have found that electronic Raman scattering from the nitrogen defect
levels is significantly enhanced with excitation by red or near IR laser light. These results
demonstrate that the laser wavelength is a key parameter in the characterization of SiC by Raman
scattering. These results suggest that Raman spectroscopy can be used as a noninvasive, in situ
diagnostic for SiC wafer production and substrate evaluation. We also present results on timeresolved photo luminescence spectra of n-type SiC wafers.
INTRODUCTION
Silicon carbide (SiC) has recently been the subject of renewed interest as an important
material for a wide variety of high-power and high-temperature electronic applications. SiC
exhibits a large number (250) of polytypes with different structural and physical properties. '
The polytypes have the same chemical composition but exhibit different crystallographic
structures and stacking sequences along the principal crystal axis. Several important polytypes
of SiC such as 4H- and 6H- have Cey crystallographic symmetry. In the a-direction 4H- and 6HSiC are almost identical (< 1 % change); however, the 4H- polytype consists of 4 units in the cdirection and the 6H- consists of 6 units. Different polytypes have different band gaps, electron
mobilities, and other physical properties; for example, 4H-SiC has attracted significant attention
due to its high electron mobility and excellent thermal properties. Recently high quality wafers
of both 4H- and 6H-SiC have been grown.2 Wafers of SiC are also a promising substrate for
nitride semiconductor growth due to their compatible lattice structure and similar thermal
expansion coefficients. In this paper we discuss second-order Raman scattering and resonance
enhancements of the electronic Raman scattering from SiC in the near IR.
EXPERIMENT
Raman spectra were recorded using both confocal Raman microscopy and a bulk Raman
spectrometer. The bulk Raman system consisted of a Coherent Model INNOVA 90 Ar/Kr laser,
a SPEX Model 1877E triple monochromator, and a CCD (charge-coupled-detector) cooled with
liquid nitrogen. A liquid nitrogen cryostat was used to take low temperature data. This system
has been described elsewhere. All confocal data was collected at room temperature using both a
Dilor LabRam system and a Renishaw Series 1000 Raman microscope. Micro-Raman spectra
were obtained with laser excitation at 785 nm (1.58 eV) and 633 nm (1.96 eV), which was
compared with data taken at 514 nm (2.41 eV), 568 nm (2.18 eV) and 647 nm (1.92 eV) with the
bulk Raman spectrometer. The spectral resolution is approximately 1 cm"1 for all spectra. The
samples were aligned such that the collection of scattered light was in the near back-scattered
201
Mat. Res. Soc. Symp. Proc. Vol. 572 e 1999 Materials Research Society
geometry perpendicular to the (0001) face of the sample. The polarization of the laser light, both
incident and collected, was unspecified. Confocal Raman spectra taken 5-10 microns below the
surface of the SiC wafer yielded results very similar to the surface of the SiC; therefore we
conclude that the effects we have observed are not surface specific.
4H- and 6H-SiC samples were examined which came from a variety of sources. Most of
the samples studied were n-type nitrogen doped, except for one semi-insulating and onep-type
(aluminum) doped sample of each polytype.
RESULTS
Overtone Spectra
Many groups have studied the vibrational spectroscopy and dynamics of 4H- and 6HSiC.4 We have examined the two-phonon Raman spectra of 4H- and 6H-SiC. The optical
branch of the two-phonon Raman spectra for 4H- and 6H-SiC are shown in figures 1 and 2.
Two-phonon Raman spectra are known to be measures of the
6H-SiC(rKloped)
4H-SiC(n-doped)
4H-SiC (semi-insulating)
1500
1600
1700
1800
Raman Shift (cm1)
>
1900
Figure 1: Optical branch of the second-order Raman spectra of 6H- and 4H-SiC
at room temperature.
vibrational density of states and therefore these are essential to improved lattice dynamics
models of SiC. The two-phonon Raman spectra of the wurtzitic forms of the SiC were found to
be far more complex than their 3C (cubic) counterpart. The two-phonon Raman spectra are not
simply twice the one-phonon spectra, but reflect the entire three dimensional band structure. The
two-phonon Raman spectra for 4H- and 6H-SiC are quite different from each other. A detailed
discussed of our peak assignments for the overtone spectra can be found elsewhere.
202
Electronic Raman Scattering
Figure 2 shows room temperature Raman spectra of a single nitrogen doped 4H-SiC
wafer (n= 5.5 x 1018 cm"3) which were taken at different laser excitation wavelengths: 514 nm
(2.41 eV), 633 nm (1.98 eV), and 785 nm (1.58 eV). There is a very significant change in the
Raman spectra upon changing the laser wavelength. When the laser excitation is in the red or
near IR, clear resonant enhancements are observed for peaks at approximately 400, 530, and 570
cm'1, labeled Na, Nb, and Nc. At 620 cm'1, a Fano resonance is clearly observed with 785 nm
(1.58 eV) excitation. The strongest peak at 530 cm"1 (Nb) is broad and asymmetric; this
asymmetry is possibly due to another peak near 500 cm'1. The peaks at 530 and 570 cm"1 are
consistent with previous measurements of «-type 4H-SiC at low temperature.6 At low
temperatures, a sharp peak at 57 cm"1 has also been measured.6 The three high frequency peaks
and the low frequency mode at 57 cm"1 make for a total of four peaks that can be attributed to
nitrogen donors in 4H-SiC. This is more than the number of inequivalent sites, two, for 4H-SiC.
X=785nm
X=633nm
X.= 514nm
300
400
NL
500
600
Raman Shift (cm"1)
700
Figure 2: Raman spectra from a single n-type 4H-SiC sample taken at room temperature with
different laser excitation wavelengths: 514 nm (2.41 eV), 633 nm (1.98 eV), 785 nm (1.58 eV). Note the
clear appearance of several additional peaks, labeled Na, Nb, and Nc, as the laser wavelength is tuned to
the near IR. We attribute these peaks to electronic Raman scattering from the nitrogen defect levels. The
spectra are normalized to the peak at 777 cm"1.
203
Figure 3 shows Raman spectra for «-type 4H-SiC at different doping concentrations,
taken at room temperature with 785 nm (1.58 eV) excitation. As the «-type nitrogen doping
concentration increases from semi-insulating to 7.1xl018 cm"3, we clearly see an increase in the
intensity of the peaks at approximately 400, 530 and 570 cm"1. The absence of peaks N„, Nb, and
Nc in the semi-insulating sample demonstrates that these peaks are associated with nitrogen
doping. Careful inspection of the peak near 530 cm"1 determines that the absolute peak position
shifts to smaller values of Raman shift as the nitrogen concentration is increased.
300
400
500
600
Raman Shift (cm"1)
Figure 3: Raman spectra for 4H-SiC taken at room temperature at 785 nm (1.58 eV) for different
n-type doping concentrations. The peaks under discussion, Na, Nb) and Nc, are clearly not observed in the
semi-insulating sample. The spectra are organized in order of doping concentration, with the highest at
the top and the lowest at the bottom: 7.1xl018 cm"3, 5.5xl018,2.6xl018,2.1xl018, and semi-insulating.
The spectra are normalized to the peak at 777 cm" .
200
300
400
500 600 700 800
Raman Shift (cm"1)
900 1000
Figure 4: Raman spectrum from 6H-SiC with «-type doping concentration of 2.1 x 1018 cm"3, taken at
room temperature with excitation at 785 nm (1.58 eV). Insert is an enlargement of the region between
350 and 700 cm"1. Note the appearance of peaks attributable to nitrogen labeled Nd, Ne, Nf, and Ng.
204
Figure 4 shows the Raman spectrum for a w-type nitrogen doped (2.1 x 1018 cm"3) 6H-SiC
sample taken at room temperature with 785 nm (1.58 eV) laser excitation. At least four
additional peaks can be observed, which are labeled in the insert as Nd, Ne, Nf, and Ng. These
peaks are not clearly visible in 6H-SiC at room temperature with 514 nm (2.41 eV) excitation
(data not shown); in fact the electronic Raman scattering peak at 510 cm"1 is masked by the
weakly scattering 6H-SiC phonons at 505 and 513 cm"1 ?' We note that, for comparable
nitrogen concentrations, it appears as if the electronic Raman scattering from nitrogen donor
levels in 6H-SiC is much stronger than the 4H-SiC polytype.
The peaks observed in our Raman experiments are quite consistent with the values for
electronic Raman scattering in n-type 6H-SiC established by Colwell and Klein.7 The peak
observed by Klein at 113 cm"1 in «-type 6H-SiC appears in our data as a tail in the Raman
spectrum, as shown in Fig 5. Fano interference effects are observed in the peak at 204 cm" due
to these tails, again similar to effects were observed by Klein.7 The Raman peaks at 486, 505,
635, and 642 cm"1 measured by Klein7 and Gorban6 at low temperature are quite consistent with
the asymmetric peaks we observe in 6H-SiC around 510 and 638 cm"1 at room temperature.7 As
pointed out by Gorban et al. the number of electronic Raman peaks is not equal to the number of
inequivalent carbon sites for 6H-SiC.6
Nitrogen doped 4H- and 6H-SiC both have a greenish metallic color. In contrast, semiinsulating SiC is clear. The color is more intense in the heavily doped samples. The green color
coincides with a broad electronic absorption in the near IR. This absorption band is the probable
origin of the resonance enhancements observed in the electronic Raman scattering from the
nitrogen doping levels. The near IR absorption seen in nitrogen doped SiC is most likely due to
the formation of deep defects. It is reasonable to assume that nitrogen doping of SiC is not
unrelated to the well-studied problem of nitrogen doping of diamond. Nitrogen forms a deep
level in diamond associated with lattice relaxation and an observed UV absorption inside the
gap.
We have also used time-resolved photoluminescence to examine n-type SiC wafers. In
Figure 5 the time-resolved spectra clearly show three peaks at zero time delay. These can be
attributed to nitrogen, aluminum and boron in the sample. Further work is underway in our
laboratory to understand these spectra.
450
500
550
600
650
700
X[nm]
Figure 5: Time-resolved photoluminescence from n-type SiC at three time delays in ps.
205
CONCLUSIONS
We have examined the electronic and vibrational spectroscopy of 4H- and 6H- SiC. We
have used both one- and two-phonon Raman spectroscopy to investigate the vibrational
dynamics of SiC. The Raman spectra of SiC were found to be polytype dependent. The twophonon Raman spectra are not simply twice the one-phonon Raman spectra, but reflect the full
three-dimensional vibrational band structure of the SiC. We have also observed electronic
Raman scattering from nitrogen donor levels in both 4H- and 6H-SiC. We have found that the
electronic Raman scattering is enhanced with red or near IR laser excitation. This resonance is
due to the near IR absorption, typical of nitrogen doped SiC, which has been attributed to deep
defects in the material. The number of peaks observed in the electronic Raman scattering from
nitrogen donor levels in both 4H- and 6H-SiC strongly suggests that the standard picture of
nitrogen sitting in carbon vacancies is incomplete.
ACKNOWLEDGEMENTS
FHL would like to thank the Rutgers Research Council for financial support.
Acknowledgement is also made to the donors of the Petroleum Research Fund, administered by
the ACS, for the partial support of this research. Work at EMCORE was supported in part by
ONR contract number N00014-97-C-0210 and monitored by Dr. Colin Wood.
REFERENCES
1
2
3
W. J. Choyke and G. Pensl, MRS Bulletin 22,25-29 (1997).
R. C. Glass, D. Henshall, V. F. Tsvetkov, and C. H. Carter, MRS Bulletin, 30-35 (1997).
J. Burton, L. Sun, M. Pophristic, S. Lukacs, F. H. Long, Z. C. Feng, I. T. Ferguson, J. Appl.
Phys. 84 6268-6274 (1999).
4 S. Nakashima and H. Harima, Physica Status Solidi A 162, 37-63 (1997).
5 J. Burton, L. Sun, F. H. Long, Z. C. Feng, I. T. Ferguson, Phys. Rev B 59 7282-7284 (1999).
6 I. S. Gorban, V. A. Gubanov, V. D. Kulakovskii, A. S. Skirda, and B. N. Shepef, Sov. Phys.
Semicond. 30,928-930 (1988).
7 P. J. Colwell and M. V. Klein, Phys. Rev. B 6,498-515 (1972).
206
GROWTH OF SIC THIN FILMS ON (100) AND (111) SILICON
BY PULSED LASER DEPOSITION COMBINED WITH
A VACUUM ANNEALING PROCESS
Jipo Huang*, Lianwei Wang*, Jun Wen**, Yuxia Wang**, Chenglu Lin*,
Carl-Mikael Zetterling***, and Mikael Östling*
*State Key Laboratory of Functional Materials for Informatics, Shanghai Institute of Metallurgy,
Chinese Academy of Sciences, Shanghai, 200050, P. R.. China
"Department of Material Science and Engineering, University of Science and Technology of
China, Hefei, 230026, P. R.. China
***Royal Institute of Technology (KTH), Department of Electronics, Electrum 229,
SE-16440, Kista-Stockholm, Sweden
Keywords: SiC thin films; Pulsed laser deposition; Annealing
Abstract
Crystalline 3C-SiC thin films were successfully grown on (100) and (111) Si substrates by
using ArF pulsed laser ablation from a SiC ceramic target combined with a vacuum annealing
process. X-ray diffraction (XRD) and Fourier transform infrared spectroscopy (FTIR) were
employed to study the effect of annealing on the structure of thin films deposited at 800°C. It was
demonstrated that vacuum annealing could transform the amorphous SiC films into crystalline
phase and that the crystallinity was strongly dependent on the annealing temperature. For the
samples deposited on (100) and (111) Si, the optimum annealing temperatures were 980 and
920°C, respectively. Scanning electron microscope (SEM) micrographs exhibited different
characteristic microstructure for the (100) and (111) Si cases, similar to that observed for the
carbonization layer initially formed in chemical vapor deposition of SiC films on Si. This also
showed the presence of the epitaxial relationship of 3C-SiC[100]//Si[100] and 3CSiC[l 11]//Si[l 11] in the direction of growth.
1. Introduction
Silicon carbide's excellent physical and electrical properties, such as wide band- gap, high
thermal conductivity, high breakdown electric field, high saturated electron drift velocity and
resistance to chemical attack, provides a promising material for high-temperature, high-power
and high-frequency electronic devices [1-2], as well as optoelectronic devices [3-4]. Cubic SiC
(3C-SiC) is attractive owing to high electron mobility and high saturation velocity. As a result,
there is an increasing interest in the growth of high quality 3C-SiC heteroepitaxial films on
silicon. Since Matsunami and coworkers pioneered the single crystalline growth of 3C-SiC films
on Si(100) by chemical vapor deposition(CVD)[5], CVD has become the most frequently used
method for growth of SiC thin films [6-9]. In a typical CVD process, reactants consisting of a
mixture of SiHi and C3H8 diluted in H2 are flowed over a silicon substrate heated to 1300°C or
above, to achieve epitaxial growth of SiC films. However, this method suffers from high
hydrogen content and crystalline lattice defects in SiC films as a result of the very high substrate
207
Mat. Res. Soc. Symp. Proc. Vol. 572 ® 1999 Materials Research Society
temperatures. Although epitaxial 3C-SiC films on Si have been reported at 751TC \.IU\, the
development of other lower temperature methods is desirable.
Recently it has been demonstrated that crystalline SiC films can be grown on Si substrates
by pulsed laser deposition (PLD), a novel and powerful thin film growth technique at relatively
low temperature. Rimai [11,12], Balloch [13], and Capano [14] have all reported preparation of
3C-SiC films on silicon. However, the former employed temperatures of 1000°C or higher, while
the crystallinity of the SiC films reported by the latter two was poor. In this study, we report
oriented SiC films on (100) and (111) Si substrates grown by pulsed laser deposition (PLD) at
800°C combined with a vacuum annealing process. The purpose is to investigate the effect of
annealing on PLD-deposited SiC films, as well as the possibility of fabricating well-crystallized
SiC thin films at relatively low temperature by using this method.
2. Experimental
The SiC thin films were deposited inside a stainless steel vacuum chamber evacuated by a
turbomolecular pump to a base pressure of 2xl0"7 Torr and a working pressure was maintained at
5~6xl0'6 Torr. Radiation from an ArF excimer laser was focused to a spot size of 0.1x0.4 cm2 on
a rotating sintered SiC target at an angle of 45°. The pulse repetition frequency was 3 Hz and a
pulse energy of 150-200 mJ yielded an energy density of 2-4 J/cm2 on the target. The plume of
atoms and ions emerged in a cone normal to the target and impinged on the substrate with the
area of 2x3 cm2. P-type (100) and (111) Si wafers (20~35ßcm, 350 um thick) substrates cleaned
with a standard silicon IC procedure were located about 3-5 cm away from and parallel to the
surface of targets. The substrate was resistively heated up to 800°C, which was monitored by a
thermocouple. The typical average growth rate of SiC films with the energy density of 2-4 J/cm
was about -0.05 nm per pulse. After deposition, the samples were taken out and cut into several
slices, and then annealed in a vacuum furnace with a pressure of lxlO"5 Torr and temperatures
ranging from 900- 1050°C for 30 min.
The crystal structure of the SiC films was characterized by X-ray diffraction (XRD) using
Cu Koc radiation. Fourier transform infrared spectroscopy (FTIR), X-ray photoelectron
spectroscopy (XPS) and scanning electron microscopy (SEM) were used to investigate
composition and surface morphology.
3. Results and discussion
The XRD analysis of the SiC films deposited on Si(100) and Si(lll) substrates at 800°C
for 30 min could not detect any characteristic SiC diffraction peaks, indicating that the films were
amorphous. The lack of crystallinity in the SiC films appeared to be related to the low growth
temperature. It was found that the intensity of the refraction peaks varied as a function of the
annealing temperature for both Si(100) and Si(lll) cases, suggesting that there was an optimum
annealing temperature in order to obtain well-crystallized films. The optimum temperatures in our
experiment were 980°C and 920°C, respectively. The lower optimum temperature for (11 l)Si
over (100) might be explained by the lower free surface energy for the (111) plane. Fig. 1 shows
the XRD patterns of the SiC films annealed in a vacuum for 30 min, with the annealing
temperature for Si(100) substrate being 980°C, and 920°C for the Si (111). For films on Si(100),
the lines corresponding to the 200 (20=41.5°) reflections are observed, whereas the 111 reflection
(20=35.6°) is absent. The reverse is true for films on Si(lll).
208
3C-SiC(100)
3
CO
3C-SiC(111)
CO
c
CD
30
35
40
45
50
55
60
2e (degree)
Fig. 1 XRD patterns of PLD deposited SiC films (a) on Si(100) annealed at 980°C
(b) on Si(l 11) annealed at 920°C.
The characteristics of infrared active modes in SiC were investigated by Fourier transform
infrared spectroscopy. In the FTIR transmittance spectra of SiC films annealed in a vacuum on
both (100) and (111) Si cases, the strong phonon absorption peak centered at 800 cm'1 is
observed, which corresponds to the characteristic phonon mode of the Si-C bond [10 ]. These
phonon modes clearly show that the films contain pure SiC phase.
Fig. 2(a) and 2(b) show SEM micrographs of the SiC films on Si(100) and Si(lll) annealed
in a vacuum for 30 min, respectively. The well-defined oriented arrangements, rectangular
morphology for the (100) case and triangular morphology for the (111) case, could be observed,
which was similar to that observed for the carbonization layer initially formed in chemical vapor
deposition of SiC film on Si and was viewed as the indicative of the epitaxial growth of cubic SiC
films and the presence of crystalline films [12,17]. Clearly, the epitaxial relationship was 3CSiC[100]//Si[100] and 3C-SiC [1 ll]//Si[l 11] in the direction of the growth. It is possible that
these patterns originate in the underlying substrates and were associated with the SiC/Si interface.
The pits on the corresponding substrates diffused into the upper films during the annealing, and
the characteristic geometric patterns came into being gradually.
Fig. 3(a) and 3(b) present the XPS spectra of Si2p and Cls of the SiC film on Si(100),
annealed at 980°C. As is evident from Fig. 3, the Si2p and Cls exhibit a single peak at 100.2 eV
and 283.3 eV, respectively, but no peaks at 99.2 eV (corresponding to pure Si) and 282.6 eV
(corresponding to graphite). This result also indicated the formation of crystalline SiC and the
measured binding energies of the Si2p and Cls states agreed well with previous reports [15,16].
The atomic ratio of C/Si using the ratio of peak area and the empirical sensitivity factors [16]
(0.87 for Si and 1.0 for C) was 1.16, i.e. nearly stoichiometric.
209
Fig. 2 SEM micrographs of the SiC films on (a) Si(100) annealed at 980°C and
(b) Si(lll) annealed at 920°C
3
to
CO
c
CD
95
100
105
Binding energy (eV)
(a)
210
110
3
co
CO
c
280
285
290
Binding energy (eV)
(b)
Fig. 3 XPS spectra of the SiC film deposited at 800°C on Si(100) and annealed in vacuum
at 980°C for 30 min (a) Si 2p (b) C Is
4. Conclusion
Pulsed excimer laser deposition combined with a vacuum annealing process has been
developed to prepare stoichiometric crystalline SiC thin films on Si(100) and Si(lll) substrates.
The amorphous SiC films deposited at 800°C could be transformed into crystalline phase after
being vacuum annealed for 30 min. Both for the samples on (100) and (lll)Si substrates, there
were an optimum annealing temperature, 980 and 920°C, respectively. Different characteristic
microstructure for (100) and (111) Si cases exhibited in SEM observation revealed the presence
of the epitaxial growth relationship of 3C-SiC[100]//Si[100] and 3C-SiC[l 11]//Si[l 11].
Acknowledgments
This work is supported by National Natural Science Foundation of China under Grant No.
69776003 and the Shanghai Youth Foundation under Grant No. 97QD14034.
211
References
1. G. Müller, G. Krötz, and E. Niemann, Sensors and Actuators A 43, 259 (1994).
2. D. M. Brown, E. Downey, M. Grezzo, J. Kretchmer, V. Krishnamurti, W. Hennessy and G.
Michon, Solid State Electronics 59,1531 (1996).
3. J.W. Palmour, J. A. Edmond, H. S. Kong and C. H. Carter Jr., Physica B 185, 461 (1993).
4. S. Sheng, M. G. Spencer, X. Tang, P. Zhou, K. Wongchotigul, C. Taylor and G. L. Harris,
Mat. Sei. Eng. B 46, 147 (1997).
5. H. Matsunami, S. Nishino, and H. Ono, IEEE. Trans. Electron Devices 28, 1235 (1981).
6. S. Nishino, H. Suhara, H. Ono, and H. Matsunami, J. Appl. Phys. 61, 4886 (1987).
7. Y. Gao, J. H. Edgar, J. Chaudhuri, S. N. Cheema, M. V. Sidorov, D. N. Braski, J. Crystal
Growth 191, 439 (1998).
8. W. Wesch, Nucl. Instr. and Meth. in Phys. Res. B 116, 305 (1996).
9. J. A. Cooper, Jr., Mat. Sei. Eng. B 44, 387 (1997).
10.1. Golecki, F. Reidinger and J. Marti, Appl. Phys. Lett., 60, 1703 (1992).
ILL. Rimai, R. Ager, W.H. Weber and J. Hangas, Appl. Phys. Lett., 59, 2266 (1991).
12. L. Rimai, R. Ager, W.H. Weber, J. Hangas, A. Samman and W. Zhu, J. Appl. Phys. 77, 6601
(1995).
13. M. Balloch, R. J. Tench, W. J. Sielhause, M. J. Allen, A. L. Conor and D. R. Olander, Appl.
Phys. Lett., 57,1540 (1990).
14. M. A. Capano, S. D. Walck and P. T. Murray, Appl. Phys. Lett. 64, 3413 (1994).
15. Ming. Y. Chen and P. Terrence. Murray, J. Mat. Sei. 25, 4929 (1990).
16. A. Watanabe, M. Mukaida, T. Tsunoda and Y. Imai, Thin Solid Films 300, 95 (1997).
17. H.J. Kim and R.F. Davis, J. Electrochem. Soc. 134, 2269 (1987).
212
ON THE ROLE OF FOREIGN ATOMS IN THE OPTIMIZATION OF 3C-SiC/Si
HETEROINTERFACES
P. MASPJ *, N. MOREAUD *, M. AVEROUS *, Th. STAUDEN **, T. WÖHNER **, J.
PEZOLDT **
*Groupe d'Etude des Semiconducteurs, CNRS, UMR 5650, Universite Montpellier 2, cc074,
Place E. Bataillon, 34095 Montpellier cedex, France, masritaiinti,univ-montp2.fr
**TU Ilmenau, Institut für Festkörperelektronik, PSF 100565, 98684 Ilmenau, Germany
ABSTRACT
3C-SiC/Si structures with Ge incorporation are elaborated by solid source molecular
beam epitaxy (SSMBE). A comparison of the flatness of the SiC-surface and the interface
between SiC and Si by comparing the deposition with and without Ge is made. The results are
analyzed within the framework of a theoretical approach based on the theory of elasticity.
INTRODUCTON
Silicon carbide has garnered a particular attention because it has been guessed as a
promising material for high-temperature and high-power electronic devices. Sensors based on
SiC are indeed aimed because of the high thermal, chemical and mechanical stability of this
material. The elaboration of high quality epilayers, which can be used as an efficient active part
of a device, is strongly dependent on substrate surface quality. Among others, two approaches
of SiC film epitaxial growth are currently considered. The investigation of the state of art of
SiC wafer elaboration shows that wafer surface crystalline quality suffers from the existence of
extended defects as dislocations and micropipes. This really represents a serious disadvantage
of homoepitaxial growth as these structural defects can have dramatic consequences on device
performance, especially because of their replication in the grown SiC active layer.
Another alternative is afforded by heteroepitaxial growth of SiC on Si substrates. The
reasons for using Si is that large area of Si wafers, cheaper than SiC wafers, can be elaborated
with a high surface crystalline quality. However, when using this solution, an important problem, related to the lowering of residual stress, must be solved. This methodology presents indeed a disadvantage due to the large mismatches of lattice parameters and thermal expansion
coefficient respectively equal to 20 % and 8 %. Because of these mismatches and beyond layer
critical thickness, extended defects, like for instance dislocations, can be favored from the
formation energy point of view. Moreover, because of lattice parameter differences separating
SiC and Si, stresses can not be fully relaxed by the dislocation network introduced at the SiC/Si
interface during the carbonization of Si-substrate surfaces. As extended defects play a vital role
in the quality of the final product, it is then worth seeking for a methodology which enables to
smooth out their differences.
Among the different techniques which can smooth out the effect of large lattice mismatch, the buffer layer technique has been widely used to assist heteroepitaxial growth. The
carbonization of clean Si surface using hydrocarbon radicals is generally performed at the early
stage of the growth process. Gas source molecular beam epitaxy (GSMBE) experiments [1]
have shown that when one uses (CHs^Gefy (DMGe) as carbon source, the morphology of the
carbonized surface layer improves when compared with the one without Ge (carbon is then
provided by hydrocarbon radicals from cracked -C3H8). The thickness of the smooth carbonized layer is ~ 40 Ä with a Ge concentration showing a saturation around 0.4 % , and the pur213
Mat. Res. Soc. Symp. Proc. Vol. 572 ° 1999 Materials Research Society
pose of using DMGe is to introduce a large size atom at the heterointerface. Similar approaches
were also carried out by [2, 3].
In this work, we investigate the effect of the incorporation of Ge atoms into the 3CSiC/Si structure grown by SSMBE and we discuss the results within the framework of a theoretical approach based on the theory of elastic. This latter approach, which we have recently
developed [4], as proved itself to be very useful in many aspects of the physics of epitaxy.
EXPERIMENT
The SiC layers were deposited on on-axis (111) Si substrate by using solid source molecular beam epitaxy. For the deposition of the 300 nm thick SiC layers on Si, we used an
UMS 500 Balzers MBE system similar to the technique reported in [3]. The deposition procedure was similar to the method reported in [5] and consists of the following process steps: (1)
silicon wafer cleaning, described in detail in [6], (2) 0-1 ML Ge deposition on the 7x7 reconstructed Si surface at 325 °C by electron beam evaporation, (3) 6 ML carbon deposition on the
Si substrate, (4) heating up the Si wafer to 600 °C for 3 minutes, (5) gradually increase of the
substrate temperature in periods of 2 minutes by 50 °C up to 1050 °C, (6) the SiC deposition
started at 950 °C with a growth rate of lnm/min. The final epitaxial temperature was 1050°CThe deposition process was monitored by in-situ RHEED and in-situ spectroscopic ellipsometry. Ex-situ, the films were investigated by atomic force microscopy (AFM). The interface
width was determined with spectroscopic ellipsometry by using a three layer model consisting
of substrate-interface-silicon carbide-surface.
EXPERIMENTAL RESULTS
For a good substitutional incorporation of Ge into SiC during the carbonization process,
it is necessary to prevent the formation of Ge bonds in the GexSii.x surface alloy. For this reason the (7x7)-Ge reconstruction was chosen. The (7x7)-Ge on (11 l)Si surface reconstruction is
stabilized for Ge coverages below 1-1.5 ML [7] or for a GexSi,.x alloy with up to 0.24 according to [8]. For this structure, the Ge atoms randomly substitutes the sites of Si atoms [9]. Furthermore the Si dangling bonds are replaced in the (7x7)-Ge by those of Ge and that no Ge-Ge
bonds are formed [10]. During the Ge deposition the (7x7)-Ge reconstruction remains unchanged and no three dimensional islands were observed. This assumption was supported by
our experiments where a deterioration of the SiC nucleation behavior for Ge coverages above 2
ML was observed.
The deposition of carbon onto the silicon surface is an effective way to introduce large
carbon concentrations into the near surface region [11]. For this reason 6 ML carbon were deposited onto the (7x7)-Ge (11 l)Si surface at 325°C. To relax the so formed compound system,
it was annealed at the deposition temperature for several minutes. The following step wise annealing procedure had the task to form a thin initial SiC layer and to stimulate the Ge incorporation into the SiC lattice. During the subsequent epitaxy, no significant differences were found
in the RHEED pattern for the different Ge coverages. All the 3C-SiC layers grown on these
carbonized pseudo-substrates were single domain and show only two dimensional diffraction
with a strong (3x3)-Si reconstruction without transmission features. For that reason, the in-situ
spectroscopic ellipsometry was used to achieve information on the interface and ex-situ AFM
investigations were used to characterize the morphology.
The experimental results of terrace width/grain diameter, surface roughness and interface width obtained after the growth of 300 nm SiC on Si are given in Table I for several Ge
214
coverages ranging from 0 up to 1 ML, where ML represents the number of monolayers with
respect to the (111) Si surface.
Table I Variation of geometric surface and interface features in function of Ge coverages
Ge coverage Terrace width/grain diameter
Surface roughness, rms Interface width
(ML)
(nm)
(nm)
(nm)
1.2
1.85
0
133
0.5
1.73
145
0.25
1.1
1.48
0.5
192
6.1
2.19
169
1
The interface width given in Table I was determined after the growth of a 100 nm thick
layer during the deposition process. This was done to have information about the near starting
conditions.
As can be seen, Ge incorporation into the Si surface leads to an improvement of the
flatness of the SiC surface and to an increase of the grain size during conversion process. This
might be an indication that the Ge incorporation into the Si leads to a decrease of the sum of
the surface energy of the SiC nuclei and the interface energy. Additionally, the Ge incorporation improves the flatness of the SiC-Si interface. From the obtained results, it can be concluded that under our experimental conditions the largest positive effect can be expected for Ge
coverages below 0.5 ML.
THEORY
The lattice misfit between two materials A (substrate Si) and B (overlayer SiCGe mixed
layer or 3C-SiC) is related to the difference between their lattice parameters A= | aA-aB. Furthermore, we assume that the corresponding periodicity of the interface unit cell must match
the A and B lattices by a vernier process. If aA< aB, the mathematical expression of this
matching is [12] :
(ni+l)aA=niaB (l.a), with ni=aA/(aB-aA) (l.b)
Eq. (l.b) holds when the B lattice is under compression and the A lattice under extension. It
states that after ni jumps on the B lattice and (ni+1) jumps on the A lattice, we may find in
coincidence two interface sites belonging respectively to A and B. If the material A is under
compression and B is under extension (aA>aB), the geometric approach gives:
ni=aB/(aA-aB)
(2)
In ref. [13], the authors have developed the idea of material phases stabilized by epitaxial
strains. As the phase determining procedure is the minimization of lattice total energy as a
function of substrate lattice parameter after growth-axis relaxation, one must emphasize the
role of interface conditions. This means that now we have to deal with in-heterostructure material physics to take account of host material interactions. For cubic materials, the following
effective elastic constants f(Cij), for longitudinal phonon modes along directions [100], [110]
and [111], are relevant to the relaxation energy [13]:
f(Cij)=C, i (a), f(Cij)=C, ,+C12+2C44 (b), f(Cy)=C, ,+20,2+4044 (c)
(3)
As far as this in-heterostructure material physics is aimed, we must seek for relationships analogous to Eqs. 1 and 2, but which can (i) unify existing approaches on epitaxially constrained interfaces and (ii) provide way to interface optimization. In what follows, we present
the methodolgy which enables to answer questions (i) and (ii).
215
To tackle these problems, a good starting point is to consider, for cubic crystals, the
equations of the elasticity theory [13] which relate strain to lattice dynamics features:
^ = £u^+Cjde^+des)+Cii_faes_+de3;
(4)
dt2
p dx
p \ dx
8x ) p\dy
dz
where u is the x component of the displacement, p is the density, the Cy are the elastic constants, and the eCTCT' are the strain components (a,a'=x,y,z). Similar equations for y and z directions can be deduced from Eq. 4. The left hand side of this equation is proportional to co2, the
square of the angular frequency w. Because of this feature, these equations may be considered
as the signatures of a strain-dynamics correlation via the S=f(Cij)/p factor.
In the case of periodic lattices, we can carry out a Bloch waves analysis of o (g), where
g represents the phonon wave vector in the growth plane, say g (gx, gy), where the gx and gy
belong to the first two dimensional Brillouin zone (2DBZ) associated with the interface reciprocal lattice. For each interface configuration, i.e. for a specific growth plane, the dynamics
equations involve effective elastic constants f(Qj) which correspond to elastic waves propagating along the principal symmetry directions in cubic crystals. The expressions of f(Cjj) for
the longitudinal and transverse modes in cubic crystals are given in ref. [14].
In what follows, we consider the case of a system constituted of two host materials A
(B) representing the substrate (overlayer) with a lattice parameter aA (aB) and an S-factor equal
to SA (SB). The extension of the 2DBZ associated with the surface/interface lattice is scaled by
the wavevector components gx~7i/ma and gy~7t/na where a is the lattice parameter, and m and n
are integers which scale the corresponding extension of the unit cell associated with a periodic
lattice network (mxn). Then we have:
co2ocSG"'(a) where G=l/gxgy<x(mxn)a2 (b)
(5)
If we write down such a relationship for each host material A and B, we must introduce
the following quantities:
SA,B=[f(Cij)/p]A,B(a); GAoc(mxn)a2A(b) andGBoc(pxq)a2B (c).
(6)
GA and GB represent the geometric factors associated with A and B, respectively. If we match
up strain gradient components, we obtain the relationship :
GA/SA=GB/SB.
(7)
If isotropic geometric conditions are assumed for the MIS (m=n and p=q) and if we apply a
vernier procedure to both A and B lattices, we end up with the equation :
ns=aA(aB/VS'-aA)"1
(8)
with S=SB/SA and aB>aA (A is under extension and B under compression). Eq. 8 means that
after ns jumps on lattice B and (ns+1) jumps on lattice A, both lattices are in coincidence. If the
material A is under compression and B is under extension (S=SB/SA, aA>aB), we obtain :
ns=aB/(aA-aB)
(9)
Eqs. 8 and 9 are analogous to Eqs. 1, but with in addition a lattice parameter renormalizing
factor (VS). If S=l, one can see that ns=n,: the condition S=l implies that we have a matching
of the elastic constant-density ratios. It is only in this case that the density of dislocations predicted by the S-theory is equal to that given by Eqs. 1. If the heterosystem is formed out of
three host materials A (substrate), B (buffer layer) and C (overlayer), we have to consider two
interfaces, namely B/A (interface 1; S=Si) and C/B (interface 2 ; S= S2). The basic equation of
the interface optimization theory with respect to the S factors is the continuity of Sj throughout the heterosystem [4] :
S,=S2
(10)
216
APPLICATION TO THE 3C-SiC/Si INTERFACE
We have shown that during the Ge
0.14
deposition, the (7x7)-Ge surface superstructure remains unchanged, corresponding to a stable interface symmetry. We
0.12 can analyse this effect within the framework of our theory. To carry out this
0.10
analysis, we use Eqs. 7 and 8 to calculate
the interface superstructure (pxq) after film
0.08
deposition, by assuming that the bare (111) S
o
Si surface is (7x7) reconstructed. We are u
0.06
then able to show that, in terms of Ge and
C composition, there exist alloyed Sii_y_
0.04
xCyGex phases which can leave unchanged
the (7x7) reconstruction. Moreover, we
find that Ge plays a compensating role in
0.02
respect to C: when the C composition of
the alloy increases, we must increase Ge
0.00
0.04 0.05
0.01
0.02 0.03
0.00
incorporation in order to stabilize the surC
COMPOSITION
face structure. These results, depicted on
Fig. 1, show a linear variation of the Ge
Fig.l Variation of Ge composition vs y
composition (x) in function of C composition (y).
Considering the use of buffer layer technique for 3C-SiC/Si heteroepitaxy, we
concentrate on two approaches in SSMBE experiments. The first one is based on the
elaboration of Sii_yCy alloy layers deposited on
Si substrates. The second one concerns the
1.30
elaboration of Sii-y.xCyGex as buffer layers
aiming to improve the 3C-SiC/Si interface
3C-SiC/Sio 948_x C0 052 Gex
quality. We have applied our theory to the
Isystem SC-SiC/SiL^CyGex/Si (C/B/A) where
Sii-y.xCyGex is introduced as a buffer layer. In
this case, interfaces 1 (Si) and 2 (S2) correspond
to Si,.y.xCyGex/Si and 3C-SiC/Sii.y.xCyGex,
S.1.25
respectively. The results shown on Fig. 2
demonstrate that the continuity condition on Sj
can be fulfilled for Ge composition as low as
0.4 % in agreement with experimental results
Experiment^
[1] while C composition corresponds to y= 5 %.
Composition [Ref. 1]
This latter value agrees with that of ref. [15].
One must bear in mind that the C composition of
1.2(
the buffer layer must remain low (5 % is a
1.0
0.2
0.4
0.6 0.8
reasonable value) because of the distortions inGe Composition (x %)
troduced by C in the Si lattice. Then, a gradual
increase
of C composition must be attempted in
Fig. 2 Variation of Sj vs x
order to meet the conditions of a pseudomorphic
growth of 3C-SiC layer [7]. The incorporation of Ge atoms into the SiC heterosystem induces
new properties which, in our approach, depend on lattice parameter, elastic constants and
217
density. This enables to achieve a better matching of strain gradient factors (Sj) associated with
the host materials involved in the heterostructure. The agreement between experimental and
theoretical results demonstrate the relevance of Sj for heteroepitaxy optimization needs.
CONCLUSION
By using a theoretical approach based on the elasticity theory, namely the S-correlated
theory of interface optimization and its continuity condition, we have determined the composition of a buffer layer formed out of Si, C and Ge species which is meant to improve the 3CSiC/Si interface quality. The calculated composition, i.e., Sio.944C0052Geo.004 is in agreement
with the experimental results [1]. Ge is probably not the only host atom which can improve the
3C-SiC/Si heterointerface and several alternatives must be investigated. In our approach, the
host layer thickness is not introduced as an explicit parameter. If we consider the parameters
relevant to the S-theory, we can perform the following analysis. The density (p) is a material
property and it should not depend on layer thickness. Lattice parameters depend on the considered host material phase. This means that the corresponding layer thickness must be at least
that for which bulk structural features prevail. Eventually, elastic constants can be different for
very thin and very thick layers, although in this work we have used bulk values.
The theoretical approach could be verified by experimental preparation of samples with different Ge content incorporated into the Si-SiC interface. Ge coverages of approximately 0.5 ML
did influence the properties of the epitaxial layers positively. Investigations by AFM confirmed
a decrease in surface roughness and an increase in grain size. The fit of in situ ellipsometric
spectra using an optical model consisting of four layers (surface roughness, 3C-SiC layer, interface width and Si substrate) determines a reduction in interface width.
REFERENCES
1. T. Hatayama, N. Tanaka, T. Fuyuki, and H. Matsunami, J. Electronic Mater. 26, p. 160
(1997).
2. K. Zekentes and K. Tsagaraki, Mater. Sei. Eng. B56, (1999), in press
3. S. Mitchel, M. G. Spencer and K. Wongchotigul, Mater. Sei. Forum 264-268, 231 (1998).
4. P. Masri, Phys. Rev. B52, 16627 (1995).
5. A. Fissel, K. Pfenninghaus, U. Kaiser, J. Kräußlich, H. Hobert, B. Schröter and W. Richter,
Mater. Sei. Forum 264-268, 255 (1998).
6. J. Pezoldt, V. Cimalla, Th. Stauden, G. Ecke, G. Eichhorn, F. Scharmann, D. Schipanski,
Diam. Rel. Mater. 6, 1311 (1997).
7 T. Ichikawa, S. Ino, Surf. Sei., 136, 267 (1984)
8 P. Martensson, W.-X. Ni, and G.V. Hansson, Phys. Rev. B36, 5974 (1987).
9 B. Zhang, J.E. Northrup,, M.L. Cohen, Surf. Sei. 145, L465 (1984).
10 S. Hasegawa, H.Iwasaki,S.-T. Li, S. Nakamura, Phys. Rev. B32, 6949 (1985).
11 S. Ruvimov, E. Bugiel, HJ. Osten, J. Appl. Phys. 78, 2323 (1995).
12. See, for example, J. W. Matthews, in Epitaxial growth, Part B, Academic Press, New York,
505(1975).
13. Sverre Froyen, Su-Huai Wei, Alex Zunger, Phys. Rev. B38, 10124 (1988).
14. C. Kittel, in Introduction to Solid State Physics, 3rd edn., Wiley, New York, ((1968) p. 119
Eq.(31).
15. K. Eberl, S. S. Iyer, J. C. Tsang, M. S. Goorsky and F. K. Legoues, J. Vac. Sei. Technol.
BIO, 934 (1992).
218
3C-SiC BUFFER LAYERS CONVERTED FROM Si AT A LOW TEMPERATURE
H. M. Liaw*, S. Q. Hong*, P. Fejes*, D. Werho*, H. Tompkins*, S. Zollner*, S. R.Wilson*,
K.J. Linthicum**, and R. F. Davis**
»Motorola Inc., Semiconductor Products Sector, 2100 E. Elliot Road, Tempe, AZ 85284,
rwd720@email.mot.com
**Department of Materials Science and Engineering, North Carolina State University,
Raleigh, NC 27695
ABSTRACT
We have obtained single-crystal 3C-SiC films via conversion of the surface region of Si (111)
and (100) wafers at 970 °C by reaction with C2H4 in an MBE reactor. The major defects in the
films were clusters, voids, and misfit dislocations. Investigation by high resolution TEM images
showed low lattice strains in the epitaxial layer due to the formation of 1 misfit dislocation for
every 4 to 5 regular SiC planes that are bonded to Si at the interface. The clusters and voids often
occurred in pairs. A model for forming the void-cluster pairs is suggested.
INTRODUCTION
Heteroepitaxial growth of 3C-SiC on Si has been difficult due to the large mismatch in lattice
parameters. The difficulty has been partially overcome by use of a two-step growth process. The
process involves the growth of a buffer layer followed by the growth of second epitaxial layer.
The buffer layer growth is considered to be a critical step for the successful single-crystal growth
of 3C-SiC on Si. The first and most common method for growing a buffer layer has been via the
reaction of the Si substrate with propane as initially described by Nishino et al. [1, 2]. Propane
was introduced into an epitaxial reactor, and the Si substrate was heated rapidly from room
temperature to approximately 1360 °C and held at that temperature for a short time (< 5 minutes).
Carbon was deposited on the Si substrate from the decomposition of propane. The Si top surface
was converted to SiC by the reaction with the deposited C at elevated temperatures. The thin SiC
films achieved by this process were approximately 20 nm thick and were reported as
polycrystalline 3C-SiC. A single-crystal SiC buffer layer is preferable for achieving higher crystal
perfection in the second epitaxial layer.
The growth of a single-crystal buffer layer has been pursued by Cheng et al. [3] using a
lower temperature (in range of 1000 °C to 1170 °C) and a shorter carbonization time (in the range
of 5 to 30 seconds). They observed that the films started to grow as thin circular islands. Lateral
growth of the islands led them to coalesce and finally became a continuous film if the proper
process parameters were applied. A continuous SiC film (approximately 18 nm thick) could only
be obtained under a narrow process window that was at 1070 °C for 15 seconds. The converted
films were single crystals but highly defective. High concentrations of defects were observed at
the islands centers where nucleation initiated. The defect density decreased away from the island
centers. However, defects such as stacking faults were still present. In addition, the films were
embedded with high densities of etched pits which were created during the Si-to-SiC conversion
process. The SiC films in the vicinity of the pits were misoriented and contained numerous
stacking faults.
219
Mat. Res. Soc. Symp. Proc. Vol. 572 ° 1999 Materials Research Society
Further decreases in carbonization temperatures were carried out by Yoshinobo et al. [4]
with MBE. They used a temperature range from 720 °C to 830 °C with C2H2 as the carbon
source. Single crystal 3C-SiC layers were obtained only in a narrow range of substrate
temperatures near 790 °C. They used RHEED to confirm the growth of a single-crystal film. No
further characterization of the defects in the films was conducted.
In this study we carried out the carbonization using gas-phase MBE similar to that of
Yoshinobo et al. [4]. The single-crystal films were grown at 970 CC using C2H4. This paper
reports the characterization of these films deposited under this condition. The mechanisms for
low temperature growth of the converted SiC films will be proposed based on the
characterization data.
EXPERIMENTAL
The conversions of Si (111) and (100) surfaces to SiC were achieved simultaneously using a
custom MBE reactor designed, constructed and commissioned at NCSU for the growth of SiC
thin films [5]. The size of each substrate was approximately 1cm by 1cm. They were (111) and
(100) orientations with 3 degrees tilted toward the <110> flat. The native oxides on the Si
substrates were removed by dipping the substrates in a buffered HF solution prior to loading into
the reactor. After evacuation of the MBE chamber, the 1.8 seem flow of C2Et was introduced at
room temperature. The temperature was ramped to 820 °C with a ramp rate of 6 °C/min and then
ramped from 820 °C to 970 °C with a ramp rate of 3 °C/min. At 970 °C the substrates continued
to be exposed to the 1.8 seem flow of C2H4 for 60 minutes. This completed the conversion
process.
Scanning electron microscopy (SEM) was used to evaluate the surface defects introduced by
the carbonization. High resolution transmission electron microscopy (HREM) was used to
characterize the microscopic defects. The film thickness was evaluated by spectroscopic
ellipsometry. The film composition and carbon bondings were evaluated by X-ray photoelectron
spectroscopy (XPS).
RESULTS
1. Film defects
The converted SiC films were continuous across the entire surface of each substrate.
Investigation via SEM showed the presence of scattered clusters in the films. The clusters varied
in size. The largest clusters were as large as several micrometers, although the largest size shown
in Figure 1 is only hundreds of nm. The density of the clusters was estimated to be in the range
of 4 xl04/cm2to 10 xl04/cm2. The clusters contained voids between the film and the Si (111)
substrate, as shown in Figure 2. Each cluster was the aggregate of several micro-clusters. The
micro-clusters occurred preferentially along the edges of the voids. The edges of small voids were
bounded by (110) crystallographic planes and indicate a preferentially etched pattern. The large
voids became more rounded and were not crystallographically oriented.
Figure 3 is a plan view TEM micrograph showing a large cluster in the converted SiC film
deposited on a Si (100) substrate. The gray white contrast regions are the voids underneath the
film. The diameter of this cluster is approximately 5 |im. The voids along the edges of the cluster
are significantly larger than those in the internal region of the cluster. It also shows that the edges
of small voids are parallel to the {110} planes.
220
Figure 2. A cross-sectional SEM
micrograph of a cluster containing voids
and micro-clusters.
Figure 1. A SEM plan view of the
converted SiC film on Si (111)
showing presence of clusters
An electron diffraction pattern taken from the cluster is shown in Figure 4. The diffraction
spots from the SiC film as well as from the Si substrate are observed. The major crystal planes
are assigned and shown in Figure 4. It shows that the crystal planes of the SiC film align with the
same Miller indices as the Si substrate. This confirms the epitaxial relationship and the
monocrystalline character of the SiC film. Extra (or satellite) diffraction spots from the SiC
planes of the same Miller indices are also present in the pattern. The satellite spots result from
double or multiple diffractions but not from twinning in the film. The SiC diffraction spots are
distorted from circular to elliptical shapes. This suggests that there are slight mis-orientations in
some parts of the film. It is likely that the misorientations reside at or near micro-clusters.
High resolution electron micrographs (HREM) did not reveal polycrystalline regions and
twinning in the converted SiC film. However, high densities of misfit dislocations were observed.
Figure 5 shows the lattice image of the SiC film and the SiC/Si interface. The arrows in Figure 5
show the points of misfit dislocation initiation. These points are only several atomic spacings
above the SiC/Si interface. This micrograph also shows that there is approximately one misfit
dislocation (an extra SiC plane non-bonded to the Si substrate) for every 4 to 5 regular SiC planes
that are bonded to Si at the interface.
Figure 5 also reveals that the boundary (or interface) between the Si substrate and the
converted SiC is diffuse and wavy. The diffuse boundary may result from a tilt of the specimen
Figure 4. Electron diffraction pattern taken
from the cluster shown in Figure 3.
Fig. 3 A plane viewed TEM shows a
cluster in the converted SiC (100) surface.
221
surface away from perpendicular to the electron beam. Examination of several interface regions of
the TEM specimen found few interfaces sharper than that shown in Figure 5. It suggests that
this is the most common character of these interfaces. The width of undulation at the interface is
in the range of several nm.
2. Lattice strains of the film and Si substrate
Lattice strains of the SiC films and the Si near the interface were calculated by measuring the
change in lattice parameters from their theoretical values. The inter-planar distance of a given
family of crystallographic planes was obtained by Fourier transform of the lattice image. The
lattice parameters of the Si substrate further away from the interface were assumed to be the
same as the bulk Si values and were used for calibration of the measurement accuracy taken at the
vicinity of the interface. The measured results together with the calculated strains are listed in
Tables 1 for a converted SiC film. The results show that the lattice strains of the SiC film are less
than 1.2 %. We also evaluated the lattice strains of substrate in the vicinity of the interface and
found it to be less than 0.5%. It suggests that stresses from a large lattice mismatch are totally
alleviated by the misfit dislocation formation in the film. This agrees with the misfit dislocation
density evaluated from HREM micrographs. In other words, approximately 20 % of the film
lattices are dislocated in the misfit manner. This accounts for the disappearance of film strain
supposedly caused by the 20% lattice mismatch between the film and substrate.
3. Film thickness and uniformity
The film thickness was measured using a spectroscopic ellipsometer. The measured optical
spectra fitted fairly well with a layer model that assumed 25% voids in the films. The optical
constants of the SiC films used in the model were assumed to be the same as those of bulk SiC.
The thickness measured from the model was 48.8 A and 48.1Ä for the (100) and (111) films,
respectively.
Figure 5. A cross-sectional HREM
micrograph showing the initiation points
of misfit dislocation generations.
45
40
-<100>SiC
-<111>SiC
35
30
25
20
15
10
5
290
285
280
Binding Energy (eV)
Figure 6. XPS spectra of the converted
SiC films showing the bonded C peak at
282.5 eV and the free C peak at 284.2 eV.
222
Table 1. Measured lattice parameters of crystallographic planes and calculated lattice strains.
SiCon
(100)Si
1.53 A
film (220), vertical to
Atomic plane
BulkSiC
Lattice strain
1.54 A
-0.0064935
SiCon
Si(lll)
Lattice strain
thp oihctratp
film (111), +slant
film (111),-or zero
slant
2.55 A
2.55 A
2.52 A
2.52 A
0.01190476
0.01190476
2.54 A
2.49 A
0.0079365
-0.011905
film (200)
2.15 A
2.18 A
-0.0137615
2.19 A
0.0045872
The thickness of the converted SiC measured by XTEM showed a large variation with
location. The thickness varied in the range of 30 to 60 A for the regions of films containing no
voids. The thickness of the film located above the voids is considerable higher. This was
determined by obtaining electron diffraction patterns from a plan view TEM sample. The
diffraction pattern was visible when it was taken from a cluster. It was not visible when the
pattern was taken from a smooth area because it was too thin. The cross-sectional SEM
micrographs showed that the film was extremely rough at cluster regions. The film thickness in
these regions was estimated to be from lto 6 times ofthat at the smooth areas.
4. Film composition
The film composition was analyzed by XPS. A high resolution spectrum taken using C-ls
was analyzed for free carbon vs. bonded carbon. Figure 6 shows the bonded C peak at 282.5 eV
and the free C peak at 284.2 eV. A comparison of peak intensities suggests that about half of the
C deposited is not yet converted to SiC. It also shows that the degree of the conversion is
approximately the same for the SiC (111) and SiC (100). The Si 2p peak at 100.5 eV confirms
little difference in the amount of conversion from (111) and (100) Si (not shown).
DISCUSSION
Single-crystal 3C-SiC(l 11) and 3C-SiC(100) films have been obtained via conversion of the
near-surface region of Si (111) and Si (100) at 970 °C by reaction with C2H4 in a MBE reactor.
The major defects in the films were clusters, voids, and misfit dislocations. The clusters and
voids often occurred in pairs. The cluster-void pairs were also observed in 3C-SiC epitaxial layers
formed by carbonization at higher temperatures [3] as well as by direct gas-phase MBE growth
without any carbonization [6].
A model for formation of the cluster-void pairs has been proposed by Schmitt et. al. [6]. In
this model the voids (silicon etch pits) were assumed to be formed first by Si evaporation. The
clusters (referred to as islands in Ref. 6) subsequently nucleated and grew at corners of the voids.
It was assumed in this model that most voids were unsealed and that the clusters were made of
polycrystalline SiC.
Examination of our films showed no unsealed clusters and no polycrystalline SiC. Therefore,
a more suitable model for forming the void-cluster pairs in our films is suggested as follows: (1)
deposition of a thin carbon layer on the substrate surface, (2) conversion of the carbon in contact
with Si into SiC, (3) diffusion of Si underneath the Si/SiC interface into the free carbon and the
223
formation of SiC, and (4) continued deposition of free carbon on the top carbon layer and
formation of voids by outdiffusion of Si from the void surfaces.
Cheng et al. [3] observed that their converted SiC films were grown by the coalescence of
SiC islands. There is no evidence in our work that the films were formed by this mechanism. It
suggests that a 2-dimensional growth mechanism is dominant over a 3-dimensionaI growth
mechanism at 970 °C.
The undulations of the SiC/Si interface were likely caused by the inter-diffusion of C and Si
across the interface. The large thickness variation in the films resulted from locally enhanced
diffusion of Si. The Si can out-diffuse faster along the void surfaces than through the bulk of the
films.
It is likely that the deposited carbon layers before the conversion were amorphous. During
the conversion process, only the carbon atoms landing on top of Si atoms formed epitaxial SiC on
Si. The remaining un-aligned carbon atoms, landed between the Si lattice planes, were also
converted by inter-diffusions and formed extra planes or misfit dislocations. The misfit
dislocations had released the lattice strain in the films.
The XPS analyses showed that the concentration of the unreacted C is fairly high. The nonbonded C can provide a driving force for the outdiffusion of Si. The excessive out-diffusion of Si
can lead to the void formation and void growth.
The concentrations of misfit dislocations and the degree of lattice distortion in the converted
SiC are difficult to improve, since the growth is not pseudomorphic. However, the voids and
clusters may be reduced by optimizing the carbon layer thickness such that it is sufficiently thick
to encapsulate the Si surface and prevent evaporation yet sufficiently thin to avoid excessive Si
and C inter-diffusion. It is desirable to have a process condition so that the conversion does not
start before complete carbon coverage of the substrate surface. Moreover, carbon deposition
should not be continued after the conversion.
ACKNOWLEDGMENTS
The authors acknowledge the support of the Semiconductor Research Corporation and the
Motorola Inc. R. Davis was supported in part by Kobe Steel, Ltd. University Professorship.
REFERENCES
1. S. Nishino, J. A. Powell, and H. A. Will, Appl. Phys. Lett., 42, p.460 (1982).
2. S. Nishino, in " Properties of SiC", p. 204, EMIS, Dataview Series, an Inspec Publication
(1995).
3. T. T. Cheng, P. Pirouz, and J. A. Powell, Mat. Res. Soc. Symp. Proc. Vol. 148, p. 229
(1989)
4. T. Yoshinobu, T. Fuyuki and H. Matsunami, Jap. J. Appl. Phys. 30, p.L1086 (1991).
5. L. B. Lowlane, F. Tanaka, R. S. Kern and R. F. Davis, J. Mats. Res. 8, p.2310 (1993).
6. J. Schmitt, T. Troffer, K. Christiansen, R. Helbig, G. Pensl, and H. P. Strunk, Materials
Science Forum vol.264-268, pt.l p.247-50 Conference Title: Silicon Carbide, m-Nitrides
and Related Materials. 7th International Conference
224
TIME RESOLVED PHOTOLUMINESCENCE OF CUBIC MG
DOPED GAN
R. Seitz", C. Gaspar", T. Monteiro*, E. Pereira*, B. Schoettker**, T. Frey"*, DJ. As", D.
Schikora"", K. Lischka
*University of Aveiro, Department of Physics, Aveiro, PORTUGAL, seitz@fis.ua.pt
** University of Paderborn, Fachbereich Physik, Paderborn, GERMANY
Abstract
Mg doped cubic GaN layers were studied by steady state and time resolved
photoluminescence. The blue emission due to Mg doping can be decomposed in three
bands. The decay curves and the spectral shift with time delays indicates donor-acceptor
pair behaviour. This can be confirmed by excitation density dependent measurements.
Furthermore temperature dependent analysis shows that the three emissions have one
impurity in common. We propose that this is an acceptor level related to the Mg
incorporation and the three deep donor levels are due to compensation effects.
Introduction
GaN is found to exist in a thermodynamically stable hexagonal and in a
metastable cubic phase [1]. Substrates for the cubic material are much easier available,
cleavage is less complicated and due to the higher symmetry superior electronic
properties are expected. In order to achieve optoelectronic devices based on p-type cubic
GaN it is important to study its fundamental properties. As in hexagonal samples, the
basic properties like recombination channels of Mg doped cubic GaN samples are still
not completely understood. Lightly and moderately doped samples show besides the
excitonic transitions at 3.27eV, the donor-acceptor pair DAP transition at 3.15eV and a
Mg related DAP recombination at 3.04eV [2]. For heavily doped samples ([Mg]>1018cm"
*) deep Mg related transitions are observed. In this work these deep blue emissions are
studied by steady state (SS) and time resolved (TR) photoluminescence.
Experiment
The cubic GaN layers were grown by plasma-assisted molecular-beam epitaxy on
semi-insulating (100) GaAs substrates at a substrate temperature of 720°C. Details of the
growth procedure were reported elsewhere [3]. For Mg doping, a commercially available
effusion cell with an orifice of 3mm was used. The thickness of the GaN layers varies
between 700 and 900nm. Mg concentration and depth distribution was determined by
secondary ion mass spectroscopy. In this work samples with Mg concentration from
2.3*I018 to 5*10l8cm"~ were analysed. Results are compared with an non-intentionally
doped (nid) sample.
For steady state (SS) photoluminescence PL measurements we used the 325nm
line of a HeCd laser. Excitation intensity was varied by neutral density (ND) filters. Time
resolved (TR) measurements were carried out with a pulsed Xe lamp as an excitation
source and a boxcar system for detection. The emission was dispersed and detected in
both cases by a monochromator (lm, 1200/mm) and a photomultiplier, respectively. The
samples were cooled down to 10K by a closed-cycle He cryostat. The temperature could
225
Mat. Res. Soc. Symp. Proc. Vol. 572 ° 1999 Materials Research Society
he varied from 10K to room temperature by a heating resistance. The temperature was
measured by a thermocouple.
Results and discussion
Figure 1 shows SSPL spectra of a heavily Mg doped and a nid sample. In the md
layer there is a strong emission at 3.27eV and 3.15eV ascribed to excitonic and pair
emission, respectively. In the heavily Mg doped samples the PL spectra are dominated by
a broad blue band composed of three emissions. The 3.27eV and 3.15eV emissions are
hardly detectable in these layers.
Energy /eV
Fig. 1: Low temperature (1 OK) SSPL spectra of Mg doped and undoped GaN
sample
The exciton recombination at 3.27eV and the DAP emission at 3.15eV are
quenched and recombination takes place at lower energies. Doping cubic GaN layers
with Mg introduces alternative recombination channels.
The comparison of SS and TR spectra shows a shift to lower energies in TR
spectra (figure 2). Furthermore the low energy side of the emission is less quenched than
the high energy side. This is an indication for an energy dependent transition probability.
For a quantitative analysis the SS and TR spectra for various time delays were fitted to
three Gaussian curves. In the SSPL spectrum the three Gaussians are peaked at 2.968eV
(FWHM: 173meV), 2.789eV (FWHM: 239meV) and 2.512eV (FWHM: 318meV). The
energetic distance between the three bands were maintained when fitting the spectra. The
three bands shift to lower energies with increasing time delay (TD). The maximal shift
(79meV) is reached at a TD of 10ms. Further increase of the TD does not result in a
further shift. The shift to lower energies is accompanied by a very small narrowing of the
226
bands. The energetic distance between the three bands could be maintained which means
that all three bands show the same shift.
Decay curves were determined at various energies of the spectrum (figure 3).
They all follow a power law /(/) ~ f" rather than exponential decays. The exponent a is
increasing from 1.00 to 1.43 with increasing energy position. This is a sign for slower
emission at the low energy side. This behaviour is typical for distant DAP recombination.
^>
516.5nm (2.400eV)
a=1.00
_,
Q-
in.
450.8nm (2.750eV)
a=1.16
410.Onm (3.023eV)
a=1.43
"~~*VA. ..
"*&£*!.
Time / ms
Energy / eV
Fig. 3: Decay curves at various energies.
Fig. 2: SS and TR PL spectra of a heavily doped
GaN:Mg sample
The recombination energy of a DAP is [4]:
hv=£g-(£,A+£D)+e/47re0ei?
(1)
where ED and £A are the donor and acceptor ionisation energies and R is the
distance between the donor and the acceptor. The recombination energy decreases with
the distance between the impurities and so does the overlap of their wavefunctions.
Therefore the recombination probability is lower for more distant pairs and thus the
decay time is longer. Calculating the exact decay curves is rather complicated but it has
been shown [5] that the decay follows approximately a power law with exponents
between 1 and 2.
In figure 4 the spectra are shown for 3 different excitation intensities. With higher
excitation intensity the whole band shifts to higher energies and the high energy side
increases in intensity. The shift rate amounts approximately to 7meV per decade of
excitation intensity. The increase of emission intensity with excitation density follows a
power law:
227
L
(2)
with h showing values of 1.0 (2.95eV band), 0.9 (2.75eV band and 0.8 (2.5eV
band), typical for recombinations at impurities (DAP or deep defects) [6]. Excitonic
recombinations would behave superlinearily (b>\). The shift to higher energies of the
peak positions with increasing excitation intensity can also be explained with
recombinations at DAPs. With increasing excitation intensity, the very distant (low
energy) pairs get saturated and the recombination takes place at closer (high energy)
pairs.
2.95eVband: b=1.0
2.75eV band: b=0.9
2.5eV band: b=0.B
Excitation Intensity / a.u.
Energy / eV
Fiji. 4: PL spectra for various excitation densities
Fig. 5: Emission Intensity as a function of excitation
intensity for the three bands
Temperature dependent spectra show that there is hardly any change in lineshape
of the emission (figur 6). There is a shift of all three bands to lower energies with
increasing temperature. The shift follows exactly the temperature dependence of the band
gap of cubic GaN [7]. The energetic distance between the bands is maintained. The shift
is 24meV between 10K and 165K. Above 165K the emission is very poor and a
quantitive assessment is not possible anymore. DAP emissions are expected to behave
like this because both donor and acceptor levels maintain their energetic distance to the
corresponding band.
Usually the temperature dependence of DAP show exponential dependency of the
following form:
/(7>I(/[l+C,*exp(-£ai/kB7)+C2*exp(-£a2/kB7)]
228
(3)
C| and C2 are two temperature independent constants and kB is the Boltzmann constant.
2.5eV band
(
2.1
2.2
2.3
2.1
2.5
2.S
2.7
2.9
10
11
12
0.00
13
Fig. 6: SSPL spectra for various temperatures
0.02
0.04
0.06
0.08
0.10
0.1
r1 / K'1
Energy / eV
Fig. 7: PL intensity of the three bands as a function of
All three bands show a similar behaviour with Eai ~50meV for the 3 bands andEa2
varying between 4 and 6meV (figure 7).We could propose a recombination model with
one common impurity for the 3 transitions. Mg on the Ga site is supposed to introduce an
acceptor level in the band gap which however might be compensated by three donor
levels. Such a compensation mechanism was already found in hexagonal GaN [10].
Evidence for additional 3 donor levels is given by As et al. [11]. However since the
activation energy of 50meV cannot be directly related to any energy level of impurities
mentioned in literature it is very likely that a more complicated quenching mechanism is
present as proposed by Brandt [9]. No further information about the energy levels can be
obtained from the data as the bands are broad and structureless and probably a
superposition of unresolved single pair emissions and vibronic coupling.
229
Conclusions
We studied cubic GaN samples heavily doped with Mg. The results of SSPL and
TRPL measurements depending on excitation density and temperature are all consistent
with three DAP emissions. Since these emissions appear only in Mg doped samples we
conclude that Mg doping introduces levels in the band gap which result in DAP
recombination. At the moment we cannot determine the impurity levels with certainty.
References
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
J W Orton, Semicond. Sei. Techonol. 10, 101 (1995).
D.J. As, T. Simonsmeier, B. Schöttker, T. Frey, D. Schikora, W. Kriegseis, W.
Burkhardt, B.K. Meyer, Appl. Phys. Lett. 73, 1835 (1998)
D. Schikora, M. Hankeln, D.J. As, K. Lischka, T. Litz, A. Waag, T. Buhrow, and
F. Henneberger, Phys. Rev. B 54, R8381 (1996)
D. Thomas, J. Hopfield, W. Augustyniak, Phys. Rev. 140 (1 A), A202 (1965)
P. Dean, Progress in Solid State Chem. 8, 1 (1973)
T. Schmidt, K. Lischka, W. Zulehner, Phys. Rev. B 45, 8989 (1992)
G. Ramfrez-Flores, H. Navarro-Contreras, A. Lastras-Martfnez, R.C. Powell, and
J.E. Greene, Phys. Rev. B 50, 8433 (1994)
U. Kaufmann, M. Kunzer, H. Obloh, M. Maier, Ch. Manz, A. Ramakrishnan, B.
Santic, Phys. Rev. B 59, 5561 (1999)
O. Brandt, Group III Nitride Semiconductor Compounds - Physics and
Application, 11, 417-459, edited by B. Gil, Oxford Science Publications,
Clarendon Press, Oxford (1998)
L. Eckey, U. Gfug, J. Hoist, A. Hoffmann, B. Schineller, K. Heime, M. Heuken,
O. Schön, R. Beccard, J. Crys. Growth 189/190, 523 (1998)
D. J. As, Phys. Stat. Sol. (b) 210, 445 (1998)
Acknowledgements
One of the authors (R. Seitz) gratefully acknowledges financial support by the Fundacäo
Para A Ciencia E A Tecnologia. D.J. As , T. Frey, D. Schikora, and K. Lischka
acknowledges SIMS measurements by W. Kriegseis, W. Burkhardt and B.K. Meyer and
financial support by Deutsche Forschungsgemeinschaft.
For further information, contact:
Roland Seitz
University of Aveiro, Department of Physics, 3800 Aveiro, Portugal
Fax: +35134424965
Tel: +35134 370824
e-mail: seitz@fis.ua.pt
230
DIELECTRIC FUNCTION OF A1N GROWN ON Si (111) BY MBE
Stefan Zollner *, Atul Konkar *, R.B. Gregory *, S.R. Wilson *, S.A. Nikishin **, H. Temkin **
*Motorola Semiconductor Products Sector, Embedded Systems Technology Laboratories, MD
M360,2200 West Broadway Road, Mesa, AZ 85202
**Texas Tech University, Dept. of Electrical Engineering, Box 43102, Lubbock, TX 79409
ABSTRACT
We measured the ellipsometric response from 0.7-5.4 eV of c-axis oriented A1N on Si (111)
grown by molecular beam epitaxy. We determine the film thicknesses and find that for our A1N
the refractive index is about 5-10% lower than in bulk A1N single crystals. Most likely, this
discrepancy is due to a low film density (compared to bulk A1N), based on measurements using
Rutherford backscattering. The films were also characterized using atomic force microscopy and
x-ray diffraction to study the growth morphology. We find that A1N can be grown on Si (111)
without buffer layers resulting in truely two-dimensional growth, low surface roughness, and
relatively narrow x-ray peak widths.
INTRODUCTION
A1N, like all group-in nitrides, has recently received much attention because of its potential
applications in high-power and high-frequency electronics, optoelectronics, electronic packaging,
and deep-UV lithography. The resulting flurry of activities in this area, however, does not mean
that A1N is a new material. Publications on the physical properties of A1N go back more than 35
years [1]. Much of the early results are based on pioneering work by Slack, who synthesized
high-purity A1N single-crystals and studied their properties, particularly the high thermal conductivity [2,3]. Recent interest in A1N has shifted to thin films prepared on different substrates
(Si, SiC, sapphire) using a variety of techniques, such as molecular beam epitaxy (MBE), chemical vapor deposition [4], and RF sputtering [5].
The optical properties of A1N, i.e., the refractive index n and extinction coefficient k, have
recently been reviewed by Loughin and French [3]. Below the band gap of about 6.1 eV, the
material is essentially transparent, allowing the determination of n using bulk crystals. Such measurements are usually carried out employing the minimum-deviation prism method with
extremely high accuracy, not achieveable on thin films. Bulk measurements are not affected
much by surfaces and interfaces, although the chemical composition is important. It is generally
assumed that the crystals grown by Slack contain a very little oxygen or other impurities [3].
Therefore, the purpose of our work is not to determine the refractive index of A1N, but rather
to compare the refractive index of A1N thin films grown by MBE with that of bulk crystals and to
draw conclusions about the properties of the film. Our work is similar to that of Jones et al. [5],
except that their films were prepared by RF sputtering. Our A1N films were grown using MBE on
low-resistivity Si (111), cut 3° off towards (110), at 860°C with ammonia as the nitrogen source,
therefore they might contain some hydrogen or silicon, which tend to reduce n. Because of the
high affinity of A1N for oxygen, contamination with oxygen is a possibility [4], which would also
reduce the value of n. Infrared transmission measurements of A1N on high-resistivity two-side
polished Si are planned to determine the impurity content. Two films were studied here, one
displaying predominantly 2-D growth (#1, nominal thickness 1500 Ä), the other mixed 2-D and
3-D growth (#2, nominal thickness 2000 A).
231
Mat. Res. Soc. Symp. Proc. Vol. 572 ° 1999 Materials Research Society
A1N FILM PROPERTIES BY X-RAY DIFFRACTION, RUTHERFORD
BACKSCATTERING, AND ATOMIC FORCE MICROSCOPY
Since the A1N films are grown on Si (111), the crystal structure template defined by the substrate encourages growth of wurtzite A1N with the hexagonal axis along the growth direction.
This is confirmed by a 9-29 x-ray diffraction scan using a D-MAXB single-axis goniometer with
a Rigaku RU-200BH rotating-anode x-ray generator, see Fig. 2 (a). Because of the substrate
miscut, which cannot be compensated by our experimental arrangement, the Si (111) peak is
broad and weak. The A1N (0002) peak for film #1 is found at 29=36.05° corresponding to a
lattice constant c=4.98 Ä, in agreement with the literature for bulk A1N [1]. Therefore, the strain
in the film due to the lattice mismatch between A1N and Si is completely relaxed by misfit dislocations. The width of the 29 peak is 0.3-0.4°, slightly larger than the instrumental resolution of
0.2°. A rocking curve with the detector fixed at 29=36.05° shows a peak at 16.2°, indicating a tilt
between the c-axis of the film and the surface normal because of the substrate miscut. The
magnitude of the tilt cannot easily be determined with our goniometer. The width of the A1N
(0002) rocking curve peak, see Fig. 2 (b), is slightly less than 1°, giving an estimate of the mosaic
spread due to dislocations. The width of the A1N (0004) rocking curve peak is 0.86° (not shown).
Film #2 shows similar x-ray diffraction results.
Atomic force microscopy (AFM) analysis of the A1N films were carried out on a Dimension
3000 instrument in tapping mode. The surface roughness of the two A1N films was substantially
different. Whereas the film #2 with mixed 2-D and 3-D growth mode had a surface roughness of
about 142 Ä, film #1 with 2-D growth mode was much smoother and had surface roughness of
about 10 Ä. Shown in Fig. 1 is a perspective view AFM image of the A1N film #1 evidencing the
high growth quality. The film exhibits a clear terrace and step structure that is typical of films
with layer-by-layer 2-D growth on tilted substrates. The angle subtended by the terraces with the
average film surface is about 2°. Assuming that the terraces are the A1N (0001) planes the miscut
angle for A1N would be 2° which compares favorably with the intended miscut of the Si
substrate of 3°. The average step height measured by AFM is greater than 30 Ä. Since the
thickness of 1 ML of A1N is 2.49 Ä, the AFM results show that there is significant step bunching
during A1N growth resulting in steps that are greater than 10 MLs high.
Rutherford backscattering (RBS) finds that the density of the A1N film #1 is 3.0 g/cm3, about
10% lower than the published density of 3.26 g/cm3 [1]. For film #2, the density is 2.9 g/cm3, i.e.,
somewhat lower. Most likely, the differences in density are due to hydrogen (possibly saturating
dangling bonds at structural defects) or other impurities (such as oxygen) propagating along
dislocations or other growth defects (particularly in film #2). There is also a very small RBS
yield due to a contaminant with a mass similar to that of Ga (less than 0.1% by composition),
which is not expected to have a measureable influence on the optical properties.
ELLBPSOMETRY MEASUREMENTS
Ellipsometry measurements were carried out using a Woollam variable-angle ellipsometer
with compensator in the 0.74 to 5.4 eV spectral range at three angles of incidence <|>=65°, 70°,
and 75°. We obtained the ellipsometric angles \|/ and A, which are defined by rp/rs=tan\|/exp(iA),
where rp and rs are the complex Fresnel reflection coefficients (amplitude and phase) for p- and
s-polarized light. The finite bandwidth of our monochromator (about 1.5 nm) leads to a
depolarization with an amplitude of no more than 6%. There is no additional depolarization due
to thickness nonuniformities, which is also evident by the uniform coloration of the films across
the wafer surface. The ellipsometric angles for film #1 are shown in Fig. 3 (dashed lines).
232
Figure 1: Perspective view atomic force microscopy image of A1N film #1 with 2-D growth by
the terrace and step structure of the film.
We describe our data using a four-layer (ambient/surface/AIN film/substrate) model. An
interface layer between the A1N film and the substrate does not improve the agreement between
the data and the model. (In contrast to the work described in [4], no buffer was grown between
the substrate and the A1N film.) We first ignore the birefringence of A1N and describe n with a
Cauchy equation n=A+B/X+C/X2, where X is the wavelength. (Similar results are obtained with a
Sellmeier equation or using a Lorentz oscillator.) The extinction coefficient k of the film is
assumed to be given by an exponential Urbach tail. The absorption in the film is very small and
might be located at the A1N/Si interface. The surface layer is modeled within the Bruggeman
effective medium theory as a 50/50 mix of A1N and voids. (An oxide with properties similar to
those of Si02 or A1203 gives a similar description.) Our model thus contains five parameters for
the optical constants of the film plus the thicknesses of the film and the surface layer. These eight
parameters are varied using the Marquardt-Levenberg algorithm to minimize the least-square
deviation between the data and the model. The calculated ellipsometric angles, shown by the
solid lines in Fig. 3, are in good agreement with the measured data. This agreement can be
improved slightly by taking into account the small difference An between the ordinary and
extraordinary refractive index of wurtzite A1N, also described with a Cauchy model. The
anisotropy effects are small, on the order of 2% of n, see Ref. [3].
233
10°
=—I—1
1
1
1
1
i
1
_
: (a)
(0
o
o
o
g, 10-1
W
*-»
r
Si (111)
3° off
CM
O
o2- -
z-
- y\
C
zs
o
Ü
10"!
<:
\m
1111111
20
40
30
2 theta
Figure 2: X-ray diffraction results of A1N on Si (111), film #1. (a) 9-29 scan, (b) Rocking curve
of the A1N (0002) peak. Data for film #2 are essentially the same.
For film #1 (see data in Fig. 3), we obtain thicknesses of the surface and A1N layers equal to
17 A and 1500 A, respectively, and a refractive index n=2.05 at 2 eV for A1N with a probable
error on the order of 2%. A1N film #2 (data not shown) is found to have a thickness of 2026 A, a
surface layer thickness of 124 A, and «=1.98 at 2 eV, slightly smaller than for #1, which we
would expect due to the slightly smaller density found from RBS. The rms surface roughness
found by AFM is 10 A for film #1 and 142 A for film #2, in reasonable agreement with the
ellipsometry results.
DISCUSSION OF OPTICAL CONSTANTS
In Fig. 4, the ordinary refractive index determined from the Cauchy parameters and the
extinction coefficient from the Urbach tail of the A1N film #1 are given by solid and the dashdouble dotted lines, respectively. By fixing the thicknesses to their values determined in the
model and solving the ellipsometric angles for n at each wavelength without any assumption
about the dispersion model, we obtain n given by the dotted line, which is basically the same as
in the Cauchy model. This confirms that the Cauchy description for n is adequate in our spectral
range. The figure also shows the extraordinary refractive index found by the model (dashed line).
We compare our results with the data of Ref. [3] for bulk A1N and of Refs. [6-7] for thin films,
given by the symbols. The refractive index for the A1N film #2 is 3% lower (not shown).
DISCUSSION AND SUMMARY
It was shown that the refractive index of our A1N films grown on Si (111) by MBE is about
5-10% lower than that of bulk A1N single crystals [3]. Since RBS finds a density of our films that
is about 10% lower than the density of bulk A1N [1], the difference in density is the most likely
culprit for the low n. Impurities in the film (oxygen, hydrogen, or silicon) are also a possible
234
Figure 3: Ellipsometric angles for
AIN on Si (111) grown by MBE
(film #1). The dashed lines show the
experimental data at three angles of
incidence. The solid lines are the
results of a calculation, based on (i) a
Cauchy model with an Urbach tail
for absorption to describe the AIN
layer, (ii) a thin surface layer due to
surface roughness or a native oxide,
(iii) optical anisotropy due to the
wurtzite crystal structure of GaN, (iv)
finite bandwidth of the monochromator(1.5nm).
1500 A AIN on Si (111), Cauchy fit
2.0
3.0
4.0
Photon Energy (eV)
1500 A AIN on Si (111), Cauchy fit
2.0
3.0
4.0
Photon Energy (eV)
2.5
! 0.020
~i—i—i—i—|—i—i—i—i—|—i—i—r
ordinary index
extraordinary index
Ref. [6]
direct inversion
Ref. [7]
A
Ref. [3]
2.4
2.3
0.015
- 0.010
c 2.2
2.1
r
- 0.005
2.0 1.9
extinction-'coefficient k
_L
2
1=.
3
_L
4
I
I
I
E(eV)
235
I
I
I
I
I
0.000
Figure 4: Ordinary
(solid) and extraordinary (dashed) refractive index n and
extinction
coefficient k (double dotdashed) of AIN film
#1 based on the data
and model calculation from Fig. 3. The
dotted line shows n
calculated at each
wavelength from the
ellipsometric angles,
without assuming a
specific model for
the dispersion of
AIN. Data from the
literature are given
for comparison.
explanation, although their presence has not been demonstrated. The different refractive index
between two films is also due to density differences, as shown by RBS.
REFERENCES
1. O. Madelung, Semiconductors - Basic Data, 2nd ed. (Springer, Berlin, 1996), p. 69.
2. G.A. Slack, J. Phys. Chem. Solids 34, 321 (1973).
3. S. Loughin and R.H. French, in Handbook of Optical Properties of Solids, edited by E.D.
Palik, (Academic, Orlando, 1998), p. 373, and references therein.
4. S.Q. Hong, H.M. Liaw, K. Linthicum, R.F. Davis, P. Fejes, S. Zollner, M. Kottke, S.R.
Wilson, this volume.
5. D.J. Jones, R.H. French, H. Müllejans, S. Loughin, A.D. Dorneich, P.F. Carcia, (in print).
6. O. Ambacher, M. Arzberger, D. Brunner, H. Angerer, F. Freudenberg, N. Esser, T. Wethkamp, K. Wilmers, W. Richter, M. Stutzmann, MRS Internet Journal of Nitride Semiconductor Research 2, Article 22 (1997).
7. N.V. Edwards, M.D. Bremser, T.W. Weeks, R.S. Kern, R.F. Davis, and D.E. Aspnes, Appl.
Phys. Lett. 69, 2065 (1996).
236
The comparative studies of chemical vapor deposition grown epitaxial
layers and of sublimation sandwich method grown 4H-SiC samples
A. O. Evwaravea. S. R. Smith6, and W. C. Mitchel
Materials Directorate, MLPO, Air Force Research Laboratory, Wright-Patterson Air Force
Base, Ohio, 45422-7077. U.S. A
'University of Dayton, Physics Department, 300 College Park, Dayton, Ohio 45469-2314
"University of Dayton Research Institute, 300 College Park, Dayton, Ohio 45469-0167
ABSTRACT
Thermal admittance spectroscopy was used to characterize the shallow dopants in chemical
vapor deposition ( CVD) grown thin films and in sublimation sandwich method ( SSM )
grown 4H-SiC layers. The values of the activation energy levels of Ec - 0.054 eV for
Nitrogen at the hexagonal site and of Ec - 0.10 eV for Nitrogen at the cubic site were indices
of comparison. The net carrier concentrations ( N0 - Nv ) of the films were determined by
capacitance-voltage measurements. The net carrier concentrations for the SSM films ranged
from 2 x 10" to 7 x 10'7 cm'3. The two Nitrogen levels were observed in the CVD films.
Hopping conduction with an activation energy of Ec- 0.0058 eV was observed in one SSM
sample having ND - Nv = 7 x 10'7 cm'3.
INTRODUCTION
It is well known that silicon carbide sublimates when heated above 1800°C producing
Si, SiC,, and Si2C vapors. Sublimation of silicon carbide is the corner stone in the growth of
bulk silicon carbide single crystals. In the modified Lely technique or the physical transport
( PVT ) method, these vapors are physically transported through the growth chamber onto a
cooler seed where they condense to form single crystals. The sublimation process has also
been successfully applied to the growth of thin epitaxial layers. In the so called " sandwich"
method ( SSM ), the source ( polycrystalline ) is separated from the substrate by a small gap
of about 1 mm. The source and the substrate are heated so that a temperature gradient of about
10° C exists. It is claimed that the conditions are near equilibrium conditions which allow the
growth to be carried out in wider ranges of temperature and pressure. Sublimation etching can
also be carried out before thin film deposition. This is accomplished under reverse
temperature gradient—this time the substrate is at a higher temperature. However, this leaves
the surface deficient of silicon and is carbon rich. Epitaxial layers of 1- 20 urn of 6H- and 4HSiC samples with residual impurity concentration of less than 10'6 cm'3 and density of
dislocations less than 100 cm'2 have been reported1'2. Chemical vapor deposition ( CVD ) is
the more general method of growing thin films on substrates.
Nitrogen, which is incorporated into the crystalline lattice during growth, substitutes
for carbon atoms at the hexagonal ( h ) and cubic ( k) sites in 4H- SiC. Nitrogen at the two
inequivalent sites leads to two nitrogen activation energy levels. These have been determined
to at Ec - 0.054 eV and at Ec- 0.10 eV for hexagonal and cubic sites respectively3,4.
In this paper, we report the results of comparative studies of CVD and SSM grown
samples. The activation energy levels of nitrogen levels in 4H-SiC are the main indices of
comparison.
237
Mat. Res. Soc. Symp. Proc. Vol. 572 ® 1999 Materials Research Society
EXPERIMENTAL DETAILS
The six samples used in this study are listed, with their net doping concentrations and
their growth methods, in Table I.
Table I. List of samples with their net doping concentration and growth methods.
Growth method
Sample #
ND- NA(x 10'7cm'3)
5500
5635
5722
5929
5930
5931
CVD
CVD
SSM
SSM
SSM
SSM
0.0015
0.0015
2.0
4.0
7.0
4.0
Samples 5500 and 5635 were grown by chemical vapor deposition and obtained from Cree
Research Inc. and from Northrop-Grumman Science and Technology center respectively. The
other four samples were SSM grown 4H-SiC and obtained from the Institute of
Semiconductors, Kiev, Ukraine. After the usual cleaning in RCA solutions, Ni was sputtered
onto the back side of the samples and annealed at 900° C to form the ohmic contact. Circular
Schottky diodes were then fabricated on the thin films as described elsewhere5. Capacitancevoltage ( C-V) measurements were made with a 1 MHz capacitance meter. The net doping
concentrations listed in Table I were determined from the C-V data. The thermal admittance
spectroscopy ( TAS ) is described in detail in reference 6. In this experiment, the conductance
and the capacitance are measured as functions of frequency and temperature. TAS
measurements were made at frequencies ranging from 100 Hz to 1 MHz, and a temperature
range of 5 K to 300 K.
RESULTS AND DISCUSSION
The thermal admittance spectrum
of sample 5635 is shown in figure 1. The
two conductance maxima or the two
inflections in the capacitance curve
correspond to emissions from nitrogen at
the hexagonal and cubic sites
respectively. The activation energies
were obtained from the plots of ln(e„/T2)
versus 1/kT. The activation energies so
determined are Er-0.054 eV and
40
SO
80
TEMPERATURE(K]
Figure 1. The two conductance peaks are due to
excitation from N at hexagonal and cubic sites.
238
Figure 2 shows TAS of sample
5929. The device was first connected to
the grid ( planar) and TAS data
taken at 20 kHz. However, an identical spectrum was obtained when the device was
connected to the ground ( front to back ) thereby involving the substrate in the measurements.
N
X
(3
Figure 3 is the plot
of In (e/T2) versus 1/kT;
the slope of the straight line
through the data points
gives the activation energy
of thedefect. The value is
Ec-0.073 eV. The emission
of nitrogen from the
hexagonal site was not
observed, due to the
position of the Fermi level
Figure 4 shows the
TAS of sample 5930
( ND-NA = 7 x 10" cm'3)
100
150
taken at two frequencies
TEMPERATURE (K )
( 1 MHz and 80 kHz ). The
conductance maximum
Figure 2. Thermal admittance spectroscopy of sample 5929.
occurs at much lower
Only one conductance peak is observed.
temperatures than those in
figures 1 and 2. The Arrhenius plot of the TAS data yielded an activation energy of
Ec- 0.0058 eV. This is clearly
an activation energy for
hopping conduction.7
Normally hopping
conduction in silicon carbide
does not commence til the
background doping is in the
range of 4 x 10'8 cm'3. As
expected emission from N
into the conduction band was
not observed.
The devices of sample 5931
were first connected to the
grid so the measurements
were planar, not involving
the substrate. Figure 5 shows
the TAS of sample 5931 at
the measuring frequency of
Figure 3. Arrhenius plot of TAS for sample 5929. The slope
50 kHz. The activation
of the line fitted to the data yield the thermal activation
energy determined from the
energy.
plot of ln(e„/T2) versus 1/kT
is Ec -0.055 eV. This is the N level at the hexagonal site. However, when the devices were
connected to the ground to involve the substrate in the measurements, an additional peak
appeared in the spectrum. This is shown in figure 6. The additional conductance peak at 130
K must be due to the interface between the SSM film and the substrate. A SIMS analysis of a
239
a silicon carbide specimen annealed at 1900° C showed that the surface was carbon rich. So
the secondary peak in the spectrum may due a carbon complex at the interface. The devices in
the SSM film in sample
5931 were stripped off and
after thorough cleaning, a
new set of diodes were
fabricated on the substrate.
The TAS obtained was
identical to that in figure 5.
x
35
It was not possible to
deplete the substrate to the
3
interface in order to
observe the second
conductance maximum of
figure 6.
The TAS of sample 5722 (
TEMPERATURE ( K )
3
Figure 4. The TAS of sample 5yJ0 at two different frequencies ND -Nv = 2 x 10" cm" ) is
identical to that of sample
of 20 kHz and 1 MHz. The activation energy for hopping
5635 shown in figure 1.
conduction is 58 meV.
However, the activation
0.080 eV respectively
energies were found to be
40
60
80
100
120
140
TEMPERATURE ( K)
Figure 5. TAS of sample 5931; the devices were connected to the grid
without involving the substrate in the measurements. The activation energy
240
250
TEMPERATURE ( K)
Figure 6. The TAS of sample 5931 when the device was wired to ground through the
substance. The second peak is defect at the interface between the SSM film and the
substrate.
CONCLUSION
Thermal admittance spectroscopy has been used to determine the nitrogen activation
energy in two CVD grown samples and four SSM grown samples. The two N levels were
observed in samples 5500 ( CVD ), 5635 ( CVD ) and 5722 ( SSM ). Two SSM samples
5929 and 5931, even though they have the same net doping concentration ( 4 x 1017 cm'3)
behaved quite differently. Nitrogen at the hexagonal site ( Ec- 0.054 eV ) was observed in
sample 5931 while a deeper level Ec- 0.073 eV was observed in sample 5929. The degree of
compensation is the difference between these samples. The fact that hopping conduction took
place in sample 5930 is another indication that the SSM samples are compensated. The SSM
samples behaved more like bulk silicon carbide than the traditional epitaxial layers grown by
CVD.
ACKNOWLEDGEMENTS
We would like to acknowledge the technical contributions of Mr. Gerry Landis, Mr Robert
Bertke, and Mr. Robert Leese for the preparation of specimens.
241
REFERENCES
1. M.M. Anikin, V. A. Dmitriev, and V. E. Chelnokov,176 Meeting of Electrochem. Society
(Hollywood FL)
2. P.A. Ivanov and V. E. Chelnokov, Semicond. Sei. Technol. 7, 863, ( 1992 )
3. A. O. Evwaraye, S. R. Smith, and W. C. Mitchel,, J. Appl. Phys. 79,7726, ( 1996 )
4. W. Götz, A. Schoner, G. Pensl, W. Suttrop, W. J. Choyke, R. Stein, and S Leibenzeder, J.
Appl. Phys. 73, 3332 ( 1993 )
5. A. O. Evwaraye, S. R. Smith, M. Skowronski, and W. C. Mitchel, J. Appl. Phys. 75, 3472
( 1994)
6. D. L. Losee, J. Appl. Phys. 46, 2204 ( 1975) see also A. 0. Evwaraye, S. R. Smith, and
W. C. Mitchel, J. Appl. Phys. 75, 3472 (1994 )
7. W. C. Mitchel, A. O. Evwaraye, S. R. Smith, and M. D. Roth, J. Electron. Mater. 26,113 (
1997)
242
Part III
SiC Bulk Growth and
Characterization
IMPURITY EFFECTS IN THE GROWTH OF 4H-SiC CRYSTALS BY PHYSICAL
VAPOR TRANSPORT
V. BALAKRISHNA, G. AUGUSTINE, and R. H. HOPKINS
Northrop Grumman Science & Technology Center, 1350 Beulah Road, Pittsburgh, PA 15235
ABSTRACT
SiC is an important wide bandgap semiconductor material for high temperature and high
power electronic device applications. Purity improvements in the growth environment has
resulted in a two-fold benefit during growth: (a) minimized inconsistencies in the background
doping resulting in high resistivity (>5000 ohm-cm) wafer yield increase from 10-15% to 7085%, and (b) decrease in micropipe formation. Growth parameters play an important role in
determining the perfection and properties of the SiC crystals, and are extremely critical in the
growth of large diameter crystals. Several aspects of growth are vital in obtaining highly perfect,
large diameter crystals, such as: (i) optimized furnace design, (ii) high purity growth
environment, and (iii) carefully controlled growth conditions. Although significant reduction in
micropipe density has been achieved by improvements in the growth process, more stringent
device requirements mandate further reduction in the defect density. In-depth understanding of
the mechanisms of micropipe formation is essential in order to devise approaches to eliminate
them. Experiments have been performed to understand the role of growth conditions and
ambient purity on crystal perfection by intentionally introducing arrays of impurity sites on one
half of the growth surface. Results clearly suggest that presence of impurities or second phase
inclusions during start or during growth can result in the nucleation of micropipes. Insights
obtained from these studies were instrumental in the growth of ultra-low micropipe density (less
than 2 micropipes cm"2) in 1.5 inch diameter boules.
INTRODUCTION
To effectively compete as a viable semiconductor material suitable for commercial and
military technologies, SiC crystal perfection and diameter of SiC substrates must approach at
least GaAs-like specifications. Silicon Carbide (SiC) crystals are of special interest compared to
other contemporary high frequency semiconductors, such as Si and GaAs, because of its unique
properties for high power density microwave generation, for high power switching, and for high
temperature and radiation tolerant operation. As shown in Table I, SiC exhibits a higher thermal
conductivity (k,.), a higher critical breakdown field (EJ, and a saturated velocity (VM) equal to
Table I — Comparison of fundamental properties of Si, GaAs, and SiC
for device applications
E„(eV)
E,,(MV/cm)
V„,(10'cm/s)
K/WcmK)
£
Si
1.12
0.6
1.0
1.5
11.8
GaAs
6H-SiC
4H-SiC
1.42
3.02
3.26
0.6
3.2
3.0
2.0
2.0
2.0
0.5
3.0*
3.0*
11.8
10.0
9.7
245
Mat. Res. Soc. Symp. Proc. Vol. 572 ®1999 Materials Research Society
GaAs at high fields desirable for high power devices. As a consequence, solid-state SiC
electronics are having a pervasive impact on advanced DOD systems, as well as on commercial
applications, ranging from passively cooled high power and high voltage switches, satellite
communications, and cellular phones, to nuclear instrumentation, where high temperature and
radiation tolerant SiC properties offer major advantages over silicon-based circuits. However,
the success of the various commercial and military applications depend on the availability of
large diameter, low defect, high purity SiC-substrates at affordable prices. Contemporary SiC
crystals are limited in the purity, and have increasing residual stresses with an increase in
diameter. In addition, the physical vapor transport process is relatively an inefficient growth
process resulting in low yield and high cost of the wafers. Increase in wafer diameter (up to 4")
and reduction in microstructural defect density is necessary for widespread use of this unique
semiconductor material.
PVT SiC CRYSTAL GROWTH
Bulk SiC crystals are mainly grown by the physical vapor transport technique (or the
modified-Lely approach) [1-4]. Growth is achieved by the sublimation of the SiC source onto a
SiC seed. A schematic of the PVT growth technique is shown in Figure 1. By appropriately
controlling the temperature gradient between the source and the seed, the gas pressure inside the
growth chamber, and the orientation and quality of the seed crystal, the growth rate and the
polytype of the of the SiC crystal is defined. While, the baseline process at Northrop Grumman is
the growth of 35 mm diameter SiC boules, with less than 100 micropipes cm"2, single polytype
crystals, we have previously demonstrated two- and three-inch diameter growth. The current
focus is on the growth of 50 mm diameter (2 inch) whose properties are similar to the baseline
crystals, and on the development of three inch diameter crystals. Emphasis is to increase wafer
yield by accurate control over their
electrical properties and minimize defect
Q
density, and towards this several aspects
are being investigated which include:
(1) SiC Purity: The limitation to SiC
lm„-purity arises from the lack of
commercially available high purity
SiC sources. As clearly evident in
Table II, glow discharge mass
spectroscopy (GDMS) and calibrated
SIMS analysis indicates that most
sources are high in metallics,
transitional metals, or in nitrogen.
Most of these impurities are
electrically active and result in a high
degree of variability in the electrical
properties of the crystal [5-6]. For
example, boron and aluminum are
common shallow acceptor impurities
in SiC, while nitrogen and vanadium
are shallow and deep donors
respectively. Hence, depending on
Figure 1 - Schematic of conventional
the type of dominant impurity, the
PVT SiC furnace
growth could result in p-type or n-
o
o
o
o
o
246
1
1
type doping respectively. In addition, this work has clearly shown that presence of
impurities can also result in the nucleation of micropipes, and significantly affect device
yield.
(2) Microstructural defects: The presence of microstructural defects such as micropipes (which
are almost unique to SiC), edge boundaries and cracks, and high dislocation densities, has
limited the rapid commercialization of SiC technology. Micropipes are micron or submicron
sized , tube-like voids present in PVT-grown SiC, which degrade epitaxial layer morphology
and are detrimental to device performance and yield. Significant strides in the reduction of
microstructural defects (to <30 micropipes cm"2) and increase in wafer diameter (up to 2"
diameter) has resulted in the fabrication of practical SiC devices such as MESFETs [7], static
induction transistors (SITs) [8], and MOSFETs [9], opening the door for the economic
exploitation of this semiconductor. Currently, the wafers fabricated in our laboratory are low
in micropipe densities (ranging on average from <30 to about 300 micropipes cm"2, with a
best effort of <2 cm'2), and the dislocation densities are in the range of 10' to 105 cm'2.
However, to benefit power device fabrication (for example, devices with 25mm footprint
size or greater), and to provide a viable alternative for direct fabrication of devices on the
substrate, further reduction in micropipe and dislocation densities are needed.
There are several factors or combination of factors that can result in the formation of
micropipes [10]. For example, initial growth conditions favoring inhomogeneous nucleation and
growth of the crystals, constitutional supercooling, poor seeding technique, impurities in the
growth ambient, and lack of process control can result in the formation of micropipes or other
microstructural defects such as second phase inclusions, boundaries, and dislocations.
SiC PURITY
The focus of this study was (1) to improve the background crystal purity by raising the
source purity and the purity of the growth environment and thereby improve crystal yield, and
(2) optimize growth conditions to fabricate large diameter SiC crystals, while simultaneously
striving to reduce the defect density. Towards improving the crystal purity, effort was focussed
on evaluating and improving the source purity, and purity of the growth ambient
Source Purity
Various commercially available and in-house sources were analyzed to define the
Table II- GDMS analysis of SiC source used for PVT crystal growths.
Impurities in ppm by weight, nitrogen content measured by calibrated
SIMS.
Source
N
B
Al
V
Ti
Reacted Powder-1
-
5
10
130
80
3.9
19
65
42
8
2
2
5
8
2
3
8
1.5
0.1
0.001
0.013
Reacted Powder-2
CVD Source-1
CVD Source-2
CVD Source-3
l.OxlO
18
247
concentrations of various (known) electrically active impurities. Subsequently, N+ and semiinsulating 4H-SiC crystals grown from these sources were also analyzed to obtain insights into
the segregation behavior of the impurities. Earlier work clearly reveals the detrimental role of
boron and nitrogen in the growth of high resistivity 6H-SiC crystals [6], this study considers the
growth of N+ and semi-insulating 4H-SiC crystals, and their potential role in the nucleation of
micropipes. Glow discharge mass spectroscopy (GDMS) and secondary ion mass spectroscopy
(SIMS) measurements conducted on a variety of starting source materials, see Table II, reveals
the presence of acceptor impurities such as boron and aluminum in addition to a host of other
metallics such as titanium and vanadium. However, CVD sources (as seen in Table II) have a
significantly lower impurity concentrations and hence more suitable for SiC crystal growth. In
addition to the impurity incorporation from the starting source material, the carbon components
inside the growth chamber are a source of high contamination as shown in Table III.
Fortunately, as is evident, stringent high temperature halogen cleaning procedures can reduce the
impurity concentrations by over an order of magnitude.
Crystal Purity
High power switching devices, such as thyristors, operating at high temperatures require
high purity SiC substrates, while, MESFET devices operating at microwave frequencies require a
highly resistive (or semi-insulating) substrate. Sources with little or no boron, nitrogen, or
aluminum is critical to consistently grow highly resistive substrates; however, Tables II and III
clearly indicate the presence of these impurities in the source and in the carbon components used
in the growth chamber. It is hypothesized that similar to 6H-SiC crystals, nitrogen is a shallow
donor in 4H-SiC and hence compensates any residual acceptors that might be present in the
crystal such as boron and aluminum. Consequently, using ultra-pure CVD sources of SiC with
low background impurities in a high purity growth environment, crystals were successfully
grown with greater than 5000 ohm-cm, p-type resistivity.
The ability to have a consistently pure starting material and a high purity growth
environment has significant impact in the growth of low defect, semi-insulating 4H-SiC crystals.
Using a baseline approach similar to that used in the development of 6H-SiC crystals, semiinsulating completely 4H-SiC crystals with room temperature resistivities in the 1015 ohm-cm
range have been grown. To consistently grow semi-insulating 4H-SiC, vanadium is used to
compensate for any additional acceptor level impurities such as boron present in the growth
ambient either in the source or in the carbon components. Since vanadium, as an impurity, forms
a deep level close to the mid bandgap of 4H-SiC similar to 6H-SiC, it compensates for any
Table III - High temperature purification of carbon parts significantly
reduces impurity concentrations
B
AL
V
Ti
Fe
Ni
Zn
S
Untreated Carbon 1
1.7
<2
6.1
20
80
19
35
160
Treated Carbon 1
0.3
<2
0.07
6
<2
8
5
120
Untreated Carbon 2
7.6
10
380
21
450
580
87
8900
Treated Carbon 1
3
<2
0.1
18
<2
10
9
11
248
additional acceptor impurities that might be present.
Figure 2~ Variation of resistivity along length of crystal boule in vanadium
doped crystal; (a) non optimized growth, and (b) optimized growth in high
purity environment
The variability in the resistivity of the grown crystals is attributed to the variability in the
contamination levels of the various impurities, and has been one of the main aspects for poor
yield of semi-insulating wafers. As shown in Figure 2, the yield of semi-insulating wafers
increased from about 25%, as shown in Figure 2(a), to about 85%, as seen in Figure 2(b) due to
improvements in source and growth environment purity, and due to the improved growth
conditions.
MICROSTRUCTURAL DEFECTS
To understand the origin and mechanisms of micropipe formation various aspects of
nucleation and growth have been carefully evaluated. Significant insights have been obtained in
understanding the mechanisms of micropipe formation; recent results strongly suggest the
dominant role of impurities, initial crystal nucleation, and non-stoichiometric growth conditions
as the potential causes for micropipe formation. Result from the crystal nucleation experiments
and effect of impurities on micropipe formation are presented.
Nucleation Effect
It is clear that initial nucleation conditions are critical in the growth of highly perfect
crystals; as shown in Figure 3, a high quality seed is a necessary but not sufficient condition for
the growth of low defect boules. A transmission optical micrograph of a longitudinal section of
the boule clearly reveals the seed-crystal interface. Although the seed had a very low micropipe
density (approx. 20 micropipes cm"2), the ensuing crystal resulted in the formation of over 1000
micropipes cm'2. It is hypothesized that perturbations at the growth interface, droplet formation,
and/or presence of second phase inclusions can result in the formation of micropipes; dark
second phase inclusions are clearly seen in the micrograph.
249
Crystal
> 1000 cm-2
Seed
20 cnr2
Figure 3—Presence of droplets and/or second phase inclusions can result in
the formation of micropipes as clearly seen in the transmission optical
micrograph of a longitudinal section near the seed-crystal interface
Impurity Effect
Impurities or second phase inclusions can also result in the nucleation of micropipes. For
example, one factor limiting the yield of semi-insulating wafers is the presence of high
micropipe densities. It is rationalized that the increased density of micropipes is a result of
vanadium addition, which if not added in precise concentrations to compensate for the acceptor
impurities can result in the formation of precipitates or second phase inclusions resulting in the
nucleation of micropipes. Transmission optical micrograph (see Figure 4), verified by energy
dispersive x-ray analysis, clearly reveals the presence of vanadium suicide and vanadium carbide
precipitates formed in the crystals during growth, as shown in Figure 4(a). The second phase
inclusions act as nucleation centers for the micropipes as evident in Figure 4(b).
To verify the hypothesis that droplet formation or presence of second phase particles
during growth results in the formation of micropipes, controlled sites for nucleation of
micropipes were created by patterning a series of impurity 'dots' (1000 angstroms thick) on one
half of the seed face, see Figure 5. The dots varied in size from 2 to 50 ujn in diameter, so as to
20|im
Figure 4—Presence of vanadium suicide and vanadium carbide
precipitates/inclusions can cause the nucleation of micropipes; (a)
cluster of precipitates visible, and (b) nucleation of micropipes from
the inclusions clearly visible
250
encompass a wide range of nucleation sites. The transmission optical micrograph of the ensuing
crystal revealed the nucleation of a set of micropipes precisely at the location of the initial dots
or impurity locations. Although these results are not complete, it clearly demonstrates a strong
correlation between the presence of second phase inclusions on the growth surface and the
nucleation of micropipes.
Considering various other models for micropipe formation mechanisms, and our own
Ti dots patterned on / Wm W m, 9 W * ' SO^m
seed surface / ^
^
*
^
0'
/'
•
#
#
•
•
•
•
•
i
#
Figure 5—Micropipes emanating from each of the artificial nucleation site
created on the seed surface
experimental results, we have steadily reduced the micropipe density in the 4H-SiC crystals.
Large diameter Crystals
The current focus of our work is to develop 2" diameter completely 4H polytype wafers
as the baseline with less than 30 micropipes cm"2, in addition to demonstrating three-inch
2-0"diameter SiC crystal
1.S inch diameter SiC wafers
i
f
i
linn
l
|
I
l
i
i
|
I
Figure 6—Highly transparent 1.5" diameter wafers and 2" boules grown under
optimized growth conditions and in a high purity environment are representative of
the current progress
251
diameter growth. Simultaneously, finite element analysis modeling is providing insights to
minimize stresses in the crystals so as to reduce dislocation density and eliminate cracking.
Highly transparent 1.5" diameter wafers and 2" diameter boules, grown in our laboratories, are
clearly seen in Figure 6.
CONCLUSIONS
Large diameter (up to 50 mm diameter) 4H-SiC crystals grown by the PVT technique today
exhibit average micropipe densities near 100 cm"2, with the best being <2 cm"2. Focus is on the
development of highly perfect two- and three-inch diameter wafers. The results obtained thus far
can be rationalized as follows:
1. Residual impurities play a critical role in improving the yield of high resistivity SiC crystals.
Results suggest that sources of contamination are from the SiC source powder and from the
growth environment. Therefore, it is critical to perform high temperature halogen treatment
of the carbon components in the growth environment, while improving the purity of the SiC
source powder. Further improvements in purity are still necessary.
2. Improvements in purity of the growth environment and in the growth process has resulted in
yield improvements from 10-25% to 70-85% of high resistivity (> 5000 ohm-cm) 4H-SiC
crystal and is attributed to the improvements in the source and growth environment purity.
3. Impurities also aid in the nucleation of micropipes. Our studies suggest that the presence of
second phase inclusions, precipitates, or droplets during growth can result in the nucleation
of micropipes. Addition of vanadium as a deep donor for semi-insulating growth can result
in the formation of precipitates which can nucleate micropipes.
REFERENCES
1. Y. M. Tairov and V. F. Tsvetkov, Journal Cryst. Growth, 43, 209 (1978).
2. G. Ziegler, P. Lanig, D. Theis, and C. Weyrich, IEEE Transactions on Electron Devices, ED30, 277 (1983).
3. H. McD. Hobgood, D. L. Barrett, J. P. McHugh, R. C. Clarke, S. Sriram, A. A. Burk, J.
Greggi, C. D. Brandt, R. H. Hopkins, and W. J. Choyke, Journal of Crystal Growth, 137, 181
(1994).
4. V. Balakrishna. R. H. Hopkins, G. Augustine, G. T. Dunne, and R. N. Thomas, Inst. Phys.
Ser. No 60, DRIP-VII, Berlin, pp. 321-330 (1997)
5. J. R. Jenny, M. Skowronski, W. C. Mitchel, H. McD. Hobgood, R. C. Glass, G. Augustine,
and R. H. Hopkins, Appl. Phys. Lett., 68 (14), pl963 (1996).
6. R. C Glass,, G. Augustine, V. Balakrishna, H. McD. Hobgood, R. H. Hopkins, J. Jenny, M.
Skowronski, and W. J. Choyke, SiC and Related Materials, Inst. Of Phys. Ser. 142, p 37-40
(1995)
7. S. Sriram, G. Augustine, A. A. Burk Jr., R. C. Glass, H. McD. Hobgood, P. A. Orphanos, L.
B. Rowland, T. J. Smith, C. D. Brandt, M. C. Driver, and R. H. Hopkins, IEEE Electron
Device Letters, 17, 369 (1996).
8. R. C. Clarke, R. R. Siergiej, A. K. Agarwal, C. D. Brandt, A. A. Burk Jr., A. Morse, and P.
A. Orphanos, Proceedings IEEE/Cornell Conference on Advanced Concepts in High Speed
Semiconductor Devices and Circuits, 47 (1995).
9. J. W. Palmour, S. T. Allen, R. Singh, L. A. Lipkin, and D. G. Waltz, Technical Digest of the
International Conference on Silicon Carbide and Related Materials, Kyoto Japan, 813 (1995).
10. V. F. Tsvetkov, D. N. Henshall, M. F, Brady, R. C. Glass, and C. H. Carter, Jr., Mat. Res.
Soc. Sym. Proc. Vol. 512, p89 (1998)
252
CHARACTERIZATION OF VANADIUM-DOPED 4H-SiC USING OPTICAL
ADMITTANCE SPECTROSCOPY
S.R. Smith3. A.O. Evwarayeb, W.C. Mitchel, J.S. Solomon', and J. Goldstein, Materials
Directorate, MLPO, Air Force Research Laboratory, Wright-Patterson Air Force Base, Ohio,
45422-7077. USA
"University of Dayton Research Institute, 300 College Park, Dayton, Ohio 45469-0178.
"University of Dayton, Physics Department, 300 College Park, Dayton, Ohio 45469-2314
'University of Dayton Research Institute, 300 College Park, Dayton, Ohio 45469-0167
ABSTRACT
Vanadium is an important dopant in SiC because it gives rise to donor levels near the
middle of the bandgap which can be used to make the material semi-insulating, and semiinsulating material has many applications as a substrate material for high-power electronics.
However, conventional means of characterizing electronic levels in the bandgap of the material
require very high temperatures, in the neighborhood of 650-800 °C, in order to move the
Fermi level to midgap and cause ionization of the V donors. The technique of Optical
Admittance Spectroscopy permits the ionization of the midgap donors using light of the
appropriate energy, and thus avoids the need for high temperatures. Using this technique we
have examined several specimens of V-doped and high-resistivity 4H-SiC. We have identified
levels previously associated with V, and new levels we attribute to Ti. Pinning of the Fermi
level in some specimens was verified by high-temperature Hall effect measurements. SIMS
measurements were used to determine impurity concentrations. IR absorption measurements
were correlated with the Ti, V, and Cr concentrations determined by SIMS.
INTRODUCTION
High-power, high-temperature applications of semiconductors require a stable
platform or substrate. This substrate should be able to conduct heat very well and conduct
electricity very poorly. Semi-insulating 4H-SiC is such a material. SiC can be made semiinsulating by very close compensation of residual impurities, or by intentionally adding
elements that give rise to electrical levels near the middle of the bandgap in quantities sufficient
to pin the Fermi level there. One of the most popular elements for this purpose is vanadium.
However, it is almost impossible to add just V because there are usually impurities present in
the graphite parts of the growth chamber comprised of other transition metals (e.g. Ti, Cr, Zr).
If their concentration is small enough, they are of little consequence, but if their concentration
becomes large, it may be the impurities that affect the electrical characteristics of the
semiconductor as much as the intentional dopant. For the most part these impurities have been
assumed to be electrically insignificant, even though Ti has been determined to give rise to an
isoelectronic center in SiC, as well as forming Ti-N pairs.1'2
253
Mat. Res. Soc. Symp. Proc. Vol. 572 ° 1999 Materials Research Society
It is difficult to determine the position in the bandgap of energy levels that are near
midgap because of the extremely high temperatures that are required to ionize these elements
in conventional thermal characterization experiments. What is needed is a technique that does
not require high temperatures. One such technique is Optical Admittance Spectroscopy
(OAS). This technique utilizes the response of a Schottky diode to the excess charge created
when electrons are excited from either the valence band to an acceptor, or from a donor to the
conduction band. Other transitions are possible depending on the charge state of the impurity,
but the technique requires a free carrier to change the conductance of the Schottky diode. The
technique has been described in detail elsewhere.3,4
In this paper, we describe the results of using OAS measurements to characterize SiC.
The OAS spectra of several specimens of 4H-SiC were compared in the range from 200 nm
(6.2 eV) to 1300 nm (0.95 eV). The specimens were also characterized by SIMS
measurements to determine the concentration of transition metal impurities, principally Ti, V,
and Cr. High-temperature Hall effect measurements were used to determine the position of the
Fermi level. IR absorption data were correlated with the SIMS data to determine the effect of
varying Ti, V, and Cr concentrations on the amplitudes of peaks at 9682 and 6380 nm which
are normally attributed to vanadium.
An attempt was made to determine if there is a correlation between the magnitude of
vanadium optical absorption lines and the concentration of vanadium. Optical absorption
spectra were obtained using a Bomem DA-3 spectrometer fitted with an InSb detector. The
specimen was cooled to 10 K in a Cryo Industries continuous flow cryostat. The temperature
stability was better than ±0.03 K. It has been reported that the V absorption lines exhibit an
increase in absorption following exposure to intense broad-band illumination from a quartzhalogen source;5,6 therefore, a globar source was used to illuminate the specimens.
The V concentration, as well as the concentrations of Ti and Cr, were measured by
SIMS. A total of 16 different specimens were examined. Some of the specimens exhibited
very high resistivity and accumulated charge during the SIMS measurements, while others did
not. The charging of the specimen in the course of the measurement probably affected the
value of the concentration in a way for which we can not correct. 4H- and 6H specimens were
used, and some mixed polytype specimens as well, were measured.
RESULTS and DISCUSSION
Table I lists the specimens in ascending order of V concentration as determined by
SIMS. Also listed are the concentrations of Ti, Zr, and Cr. It is the purpose of this paper to
show that Ti, at least, has an electrical transition that is detectable using OAS.
254
Table I. Specimens used in this study sorted by V concentration. Concentration was
determined by SIMS.
Specimen
6147
5724
6129
5723
5633
5631
5713
5630
6130
N(V)
9.00E+14
5.00E+15
2.30E+16
3.00E+16
6.00E+16
4.30E+17
4.30E+17
ND
1.10E+17
N (Ti)
N(Zr)
N(Cr)
1.50E+16
1.50E+16
1.90E+16
6.10E+16
ND
6.00E+17
6.30E+17
7.00E+17
ND
ND
>lel7
3.90E+15
3.00E+16
ND
ND
ND
1.40E+17
ND
ND
>lel6
2.30E+17
>3.0el6
ND
ND
ND
3.80E+17
7.40E+17
ND = Not Detected
We determined where the Fermi level was pinned in the high-resistivity specimens by
high-temperature Hall effect measurements. This was critical in order to determine the nature
of the compensation. OAS can detect transitions from ionized acceptors in n-type material as
well as from neutral donors. Furthermore, in p-lypc material transitions from the valence band
to neutral acceptors, and ionized donors can be detected. Which center is giving rise to the
transition can not be determined unless the energy coincides with thermal energy differences,
or compensation rules out one or the other. Therefore it was important to know whether the
specimen was n- orp-type at high temperature, though semi-insulating at room temperature.
Figure 1 is an optical
10 T
1
admittance spectrum for
specimen 5723 which, from
Table I, it can be seen, had a
V concentration of 3.0 x
1016cm"3,andaTi
concentration of at least 6 x
1016cm'3. Also present in
non-negligible amounts were
Zr and Cr. The unusual
feature of this spectrum is the
large triplet peak in the
middle of the spectrum at 560
nm (2.21 eV). The peak at
1400
600
800
1000
800 nm (1.54 eV) is known
Wavelength (nm)
from previous experiments to
be attributable to the V
Figure 1. OAS spectrum of specimen 5723 containing 3 x
donor. The abrupt edge at
6
3
16
3
10' cm" V, and containing 6 x 10 cm" Ti.
850 nm corresponds to a
photon energy of 1.45 eV
255
which agrees with the hightemperature Hall effect
determination of the Fermi
level position. Another
specimen (5713) indicated a
Fermi level position of 1.1
eV; this was also verified by
the abrupt edge in the OAS
spectrum. This result is
shown in figure 2.
We were the first to
report the presence of a deep
level at 1.1 eV below the
conduction band in
1993.[rets. 3,4] At that time
2
3
4
5
6
7
we suggested that this peak
Energy (eV)
seen in the OAS spectrum
was related to a complex
Figure 2. Optical admittance spectrum of 4H-SiC in which the defect involving vanadium or
Fermi level was shown to be pinned at 1.1 eV below the
titanium. The defect had not
Conduction band.
been observed by any other
techniques. The specimens
cited above are the first
incidence of the 1.1 eV
defect being observed in SiC
by Hall effect measurements.
Very recent SIMS
measurements have revealed
the presence of oxygen in the
specimens of SiC which had a
Hall effect energy of 1.1 eV,
however, the carrier
concentration of the 1.1 eV
level did not scale with the
oxygen concentration, thus
ruling out oxygen or O
complexes as the source of
this level. From Table I it is
clear that we can rule out Cr
and Zr as sources for the
Energy (eV)
prominent peaks, as they
were not detected in
Figure 3. Comparison of OAS spectra of specimens 5633 and specimen 5713, but Ti was.
5724. Both are 4H-SiC considered to be semi-insulating, but
The fact that the pinning of
only 5724 had detectable Ti impurities.
the Fermi level is seen in the
'edge' of the OAS response
confirms the reality of the defect and its donor nature. However. both of these specimens
256
these specimens contain V and Ti according to SIMS measurements making identification with
either species difficult.
The OAS spectrum of specimen 5633 is shown in figure 3 compared to the spectrum of
another SI 4H-SiC specimen (5724). The most striking difference is the absence of a multicomponent peak at about 2.2 eV. Fitting the spectrum of specimen 5633 in the region of 2.2
eV results in a three-component peak having component energies of 2.09 eV, 2.20 eV, and
2.30 eV. Since we know from Hall effect data that the Fermi level was pinned at the midgap V
donor, the defect from which these peaks arise must have electrons residing on the center to be
excited to the conduction band by the incident light. Subtracting the excitation energy from
the bandgap energy of 3.026 eV yields an energy difference of Ev + 0.94 eV, Ev + 0.83 eV,
and Ev + 0.73 eV respectively. This means that the defect responsible for the triplet of peaks
is either a very deep donor with component energies approximately two eV below the
conduction band, or a compensated acceptor with component energies about 0.8 eV above the
valence band. Further scrutiny of the spectra in figure 3 reveals that another peak centered
near 2.9 eV is also inverted. Virtually all other features of these spectra are the same. As can
be seen from Table I, the only difference between these specimens is the presence of Ti in one,
and its absence in the other.
Dalibor, et ai, [ref. 1] have attributed deep acceptors with energies of EC - 0.117 eV
and EC - 0.160 eV to Ti in 4H-SiC. They assigned these levels to the ionized Ti acceptor
residing at the hexagonal and cubic sites, respectively. In the n-type specimens into which the
transition metal impurities were implanted, Dalibor could not have detected acceptors close to
the valence band via the DLTS measurements that they performed, nor would TAS
measurements have revealed acceptors close to the valence band. We suggest that there may
be Ti-related energy levels in the
SiC bandgap near the valence
band. Such a level has been seen
by us inp-type SiC in Optical
Admittance Spectroscopy
Absorption for peaks at:
experiments.7 We reported a
— — .9283 eV
deep level at Ev + 1.1 eV in
—• —.9284 eV
—*— .9681 eV
1994. At that time, we could
—T— .9694 eV
not assign this level to any
known impurity. However, it
appears that Ti, or a Ti-related
o
defect may have been
responsible for this line in the
OAS spectra. Further work with
__
SIMS and Hall effect will be
O.OOEtOOO 2.00E+017 4.00E+017 8.00E+017 a.OOE+017
ooEtoia USed to examine those early
Titanium concentration (cm"3]
specimens for Ti.
For the Cr ion, no
Figure 4. Correlation of Optical Absorption peak intensity correlation was seen with
with Ti concentration.
concentration for any of the
absorption lines for any of the specimens. For the V ion, a clear trend was observed in the four
specimens which had both sides polished and in which no charging was observed. The peaks
examined were those at 0.9200 eV, 0.9196 eV, 0.9460 eV, 0.9474 eV, and 0.9481 eV. The
trend is monotonically increasing to apparent saturation at the maximum V concentration. For
257
Ti, all 4H specimens exhibit a step-wise increase in absorption at a threshold concentration of
5 x 1017 cm"3. The maximum Ti concentration available was 5 x 1018 cm"3.
CONCLUSIONS
We have concluded that a peak in the OAS spectra of 4H-SiC, comprised of at least
three components, is related to the presence of Ti in the SiC lattice that produces an
electrically active site(s). Ti is present in most specimens that have been doped with transitionmetal species in order to produce semi-insulating material, however, in at least one case, Ti
was not added with the other impurities. In this specimen, the above mentioned peak was
absent. The source of the metal impurities was not determined, however, the specimens were
intentionally doped with V. Correlation of optical absorption peaks with V and Ti
concentration was observed in all specimens. The correlation of the V concentration was
monotonically increasing, whereas the correlation with the Ti concentration was step-wise with
a threshold at 5 x 1017 cm"3.
ACKNOWLEDGEMENTS
We would like to acknowledge the technical contributions of Mr. Gerry Landis, Mr.
Robert Bertke, and Mr. Robert Leese for the preparation of specimens. One of us, SRS, was
supported on Air Force Contract No. F615-95-C-5445.
REFERENCES
1. Thomas Dalibor, Gerhard Pensl, Nils Nordell, and Adolf Schöner, Phys. Rev. B 55, p 13
618 (1997)
2. T. Dalibor, G Pensl, H. Matsunami, T. Kimoto, W.J. Choyke, A. Schöner, and N. Nordell,
Phys. Stat. Sol. (a) 162,199 (1997)
3. A.O. Evwaraye, S.R. Smith, and W.C. Mitchel, Mater. Res. Soc. Proc. 325, 353, (1994)
4. A.O. Evwaraye, S.R. Smith, and W.C. Mitchel, J. Appl. Phys. 77,4477 (1995)
5. Jürgen Schneider and Karin Maier, Physica B 185,199 (1993)
6. J. Schneider, H.D. Müller, K. Maier, W. Wilkening, F. Fuchs, A. Dornen, S. Leibenzeder,
and R. Stein, Appl. Phys. Lett. 56,1184 (1990)
7. S.R. Smith, A.O. Evwaraye, and W.C. Mitchel, Phys. Stat. Sol. (a) 162, 227 (1997)
258
ONLINE MONITORING OF PVT SiC BULK CRYSTAL GROWTH
USING DIGITAL X-RAY IMAGING
P.J. WELLMANN, M. BICKERMANN, M. GRAU, D. HOFMANN, T.L. STRAUBINGER
AND A. WINNACKER
Materials Department VI, University of Erlangen, Martensstrasse 7, 91058 Erlangen,
GERMANY, Email: peter.wellmann@ww.uni-erlangen.de
ABSTRACT
An advanced method based on x-ray imaging is presented which allows us to visualize the
ongoing processes during physical vapor transport (PVT) growth of SiC. Using a high resolution
and high speed x-ray imaging detector based on image plates and digital recording we are able to
follow the SiC bulk single crystal growth as well as the evolution of the SiC powder source
inside the inductively heated graphite crucible on-line and quasi-continuously.
INTRODUCTION
SiC bulk crystals used to fabricate substrates for commercial device applications are presently
prepared by the physical vapor transport (PVT) method (so called modified Lely process) [1-4]
at high temperatures (T=2100°C ... 2400°C). This growth technique and the extreme thermal
conditions impose up to now intrinsic limitations to the visual control of SiC growth taking place
inside a graphite crucible. In order to improve the understanding of the PVT SiC growth process
and hence the crystal quality it would be highly advantageous to monitor the evolution of both,
the SiC crystal and the SiC source material during the whole process time. We have developed
an advanced method based on x-ray imaging which allowed us to visualize the ongoing
processes during PVT growth of SiC. Using a high resolution and high speed x-ray imaging
detector based on image plates and digital recording [5] [6] we were able to follow the crystal
growth inside the inductively heated graphite crucible quasi-continuously. Analyzing the
digitized x-ray images of the graphite crucible taken during the growth we have (i) monitored the
evolution of the SiC crystal and (ii) investigated the degradation of the SiC source powder (i.e.
graphitization) with increasing process time. In this paper we will introduce our new on-line xray imaging method of SiC growth and we will discuss its feasibility by showing several shots of
the high temperature SiC vapor growth process.
EXPERIMENT
6H and 4H SiC bulk single crystals have been prepared by the PVT technique in an
inductively heated graphite crucible at elevated temperatures of T = 2100CC ... 2300°C (figure
1). The temperature has been monitored at the seed end and at the source end of the crucible
using optical pyrometers. Argon has been used as carrier gas at a system pressure of p = 5mbar ...
40mbar. As a specialty of our setup, the induction coil can be moved electrically in vertical
direction giving rise to a wide range of geometrical configurations and the possibility of an instationary growth process with increasing growth time. As seeds we have used (0001) oriented
30mm ... 40mm SiC crystals grown by PVT in our laboratory. The micropipe density varied
between 200cm"2... 500cm"2. At defined times nitrogen was added for 10min to the argon carrier
gas in order to introduce doping striations revealing the shape of the growth interface [3]. The
source material (SiC powder) was synthesized from elemental Si and C. A typical growth run
can be divided in three parts: (i) Heating up is performed under an
259
Mat. Res. Soc. Symp. Proc. Vol. 572 ° 1999 Materials Research Society
PVT setup
x-ray
detector
Tbt
,::pyrörrieter:i:
Figure 1. PVT setup for the growth of SiC bulk crystals. The inductively heated graphite crucible (loaded with a
SiC seed crystal (top) and a SiC powder source (bottom)) is imbedded in a graphite like insulation inside a quartz
ampoule. Using a x-ray tube and an image plate based detector and scanner, shots of the ongoing processes inside
the graphite crucible were taken online during the growth process.
argon atmosphere (p = 800mbar). (ii) At the beginning of the growth (defined as Oh) the argon
pressure is lowered to 5mbar ... 40 mbar giving rise to a physical vapor transport of various SiC
species from the SiC source to the colder SiC seed, (iii) At the end of the growth run (typically
after 72h) the inductive heating is switched off.
In order to visualize the ongoing processes inside the graphite crucible during growth we have
applied a x-ray imaging system (x-ray tube, image plate detector and scanner) to the PVT setup
(figure 1). During x-ray exposure by the x-ray tube an image of the graphite crucible interior
(SiC seed/crystal and SiC source material) is transferred to the x-ray detector. While the x-ray
source is a conventional x-ray tube for medical care (y-energy = 60keV, I=7mA), a high speed,
high resolution and high dynamic range x-ray detector based on image plates and digital
recording was used.
An overview on the physics of image plate detectors and technical data of the image plate
scanner FASTSCAN can be found elsewhere [5][6]. In short: x-rays produce radiation defect
centers in x-ray storage phosphors (in our case a BaFBnEu based Fuji HR-V image plate) which
give rise to fluorescence under light excitation of proper wavelength (in our case X = 635nm).
The intensity of the so called "photo-stimulated luminescence" is proportional to the x-ray dose.
Thus, by scanning the image plate with a focused readout light beam one can extract the x-ray
image by monitoring the photo-stimulated luminescence point by point. The image plate is
recycled by exposure to intense light. Readout and recycling is done using the image plate
scanner FASTSCAN [6]. FASTSCAN has a maximum scanning size of 30cm x 20cm at a
resolution of up to 850dpi. The image readout time is 40sec for an image of 3500 x 5000 pixels.
In practice the resolution is limited by the image plate and is about 100u.m x lOOixm.
There are several advantages of the digital x-ray detector over conventional film techniques:
The fast digital image capture (t = 40sec) without time consuming developing stage allows an
260
Figure 2. x-ray images of the hot graphite crucible (T=2270°C) (a) with a fixed induction coil during x-ray exposure
and (b) with a moving induction coil during x-ray exposure.
online monitoring of the crystal growth process with the option of online process control. Due to
the large dynamic range (10* for image plates versus 10"... 10' for conventional x-ray films) and
high sensitivity the choice of a matching x-ray dose is uncritical. By computer processing of the
digitized image a large contrast of the graphite crucible interior can be obtained.
X-ray and y-ray techniques have been used in the past to study the solid-liquid interface
during crystal growth from the melt (see for example [7-9]). However, the experimental
challenge in the case of an inductively heated crucible as in our case is the large x-ray absorption
coefficient of the induction coil: The absorption of the x-ray beam is about 10" times larger in the
copper coil as compared to the graphite parts of the crucible, the SiC crystal and the SiC source.
Dark stripes are introduced on the x-ray detector by the induction coil (figure 2a). Although
digital image processing could partly uncover hidden parts of the digitized image of the crucible
interior, we have developed a hardware solution to get rid of the unwanted perturbation by the
copper coil. While moving the induction coil in vertical direction for at least 1/2 period (period is
defined as the distance between two turns of the coil, in our case 2cm) we expose the x-ray
detector continuously (typical time period = 3s). Hereby every part of the crucible interior will
be captured by the image plate once, resulting in a stripe free image. Instead of a continuous
exposure also a stepwise exposure (for example 16 steps) has been applied (figure 2b). After
finishing the exposure the coil will be moved back to the starting position. The whole process
takes about 10s to a few minutes. The temperature displays of the top and bottom pyrometers
indicate that the thermal conditions inside the graphite crucible do not change during coil
movement. Temperature changes of 1°C ... 2°C were observed after about 10 minutes if the coil
was not moved back to the starting position after x-ray illumination.
RESULTS
Figure 3 shows a series of x-ray transmission shots of the hot graphite crucible (T=2270°C) at
the beginning (Oh, figure 3a), 8h (figure 3b) and 22h after the beginning (figure 3c) of the
261
a) t = Oh
graphite
crucible
b) t = 8h
c) t = 22h
Figure 3. x-ray images of the hot graphite crucible (T=2270°C) (a) at the beginning of the growth process (Oh), (b)
8h and (c) 22h after the beginning of the growth process.
growth. One can clearly distinguish the SiC crystal, the SiC source material and parts of the
graphite crucible. The image resolution is about lOOiim.
At the beginning of the growth (Oh, figure 3a) the seed SiC crystal is visible as a dark stripe in
the x-ray image. The flat surface of the seed and the thickness of 1mm is well reproduced
indicating that the x-ray imaging system (x-ray source, detector and scanner) creates only minor
image distortions. After 8h and 22h of growth the SiC crystal has reached a length of 4.6mm and
8.4mm (figure 3b and 3c), respectively. The shape of the growth interface is well resolved in the
x-ray image. The growth interface has a convex shape in the central part (d<35mm); single
crystal growth of high quality and low micropipe density has been observed. In the outer part
(d>40mm) - at a diameter larger than the original seed crystal - a concave growth interface
occurs. In this area polycrystalline growth and the formation of various polytypes have been
observed. The SiC source material (figure 4) gives rise to a homogeneous contrast in the x-ray
image at the beginning of the growth. However, after 8h (figure 4b) and 22h (figure 4c) of
growth graphitization of the source material is observed in the vicinity of the hot graphite
crucible. Due to its low density the residual graphite powder of the SiC source material is almost
transparent for x-rays and gives rise to a light contrast in the x-ray image. The contrast of the
core part has become darker indicating a compression of the SiC source powder due to
sublimation in the hotter outer parts (close to the hot graphite crucible) and recrystallisation in
the colder central part.
262
a) t = Oh
b) t = 8h
graphite
crucible
c) t = 22h
Figure 4. x-ray images of the SiC source material inside the hot graphite crucible (T=2270°C) (a) at the beginning of
the growth process (Oh), (b) 8h and (c) 22h after the beginning of the growth process.
CONCLUSIONS
Digital x-ray imaging turns out to be a useful tool for a routine control of SiC bulk crystal
growth. The length of the SiC crystal and the status of the SiC source material can be monitored
online. The clear distinction between available and consumed (graphitization area) SiC source
material (figure 4) opens a way to optimize the yield (number of SiC wafers per crystal) for each
growth run. For a continuous process control the growth rate, the growth interface as well as the
consumption of the SiC source could be monitored online using automated digital image
processing. Through a feedback loop the whole growth could be supervised by a computer
controlling the inductive heating power (P^,), the coil position and/or the argon pressure,
respectively.
Digital x-ray imaging is as well a useful tool for understanding the physical vapor transport
process of SiC from a research and development point of view. The growth rate and the shape of
the growth interface can be studied without using demarcation doping striations by intermittently
introducing nitrogen gas during the growth [3]. The shape of the SiC crystal growth interface can
be determined with an accuracy better than lOOum (see figure 2 and 3). The study of the time
evolution of the SiC source material is another crucial point for understanding the thermal
conditions and underlying growth process. The graphitization which reflects the consumption of
the source material the process of SiC powder compression due to sublimation and
recrystallisation play an important role. Especially at the beginning of the growth (first 10 hours)
the SiC powder changes its properties giving rise to a change of growth velocity and change of
incorporated defects into the growing crystal. A detailed discussion of the time evolution of the
SiC source and its impact on the growth process will be given in a forthcoming paper.
In summary we have for the first time presented a digital x-ray imaging technique which
enable us to follow the SiC PVT process online during growth. We have introduced the
263
underlying measurement procedure and shown various x-ray images of the growing SiC crystal
and of the SiC source material inside the hot graphite crucible. Digital x-ray imaging is a
powerful tool to study the processes going on during PVT of SiC. In addition digital x-ray
imaging can be used for process control in the industrial growth of SiC.
ACKNOWLEDGMENTS
We would like to thank L. Kadinski and M. Selder (both Department of Fluid Dynamics,
University of Erlangen) for fruitful discussions in the field of numerical modeling and heat
transfer. This work has been supported by the Bayerische Forschungsstiftung (contract No.
176/96) and the Deutsche Forschungsgemeinschaft (contract No. Wi393/9).
REFERENCES
1. Y.M. Tairov and V.F. Tsvetkov, Investigation of growth processes of ingots of silicon
carbide single crystals, J.Cryst.Growth 43, 209, 1978.
2. G. Ziegler, P. Lanig, D. Theis and C. Weyerich, Single crystal growth of SiC substrate
material for blue light emitting diodes, IEEE Trans. Electron. Devices 30, 277, 1983.
3. R. Eckstein, D. Hofmann, Y. Makarov, St.G. Müller, G. Pensl, E. Schmitt and A. Winnacker,
Analysis of the sublimation growth process of silicon carbide bulk crystals, Mat. Res. Soc.
Symp. Proc. 423, 215-220, 1996.
4. D. Hofmann, R. Eckstein, M. Kölbl, Y. Makarov, St.G. Müller, E. Schmitt, A. Winnacker, R.
Rupp, R. Stein and J. Volkl, SiC-bulk growth by physical vapor transport and its global
modeling, J.Cryst.Growth 174, 669-674, 1997.
5. A. Winnacker, x-ray imaging with photostimulable storage phosphors and future trends,
Physica Medica IX 2-3, 95-101, 1993.
6. M. Thorns, H. Burzlaff, A. Kinne, J. Lange, H. von Seggern, R. Spengler and A. Winnacker,
An improved x-ray image plate detector for diffractometry, Mater.Sci.Forum 107 (1), 228231, 1995.
7. P.G. Barber, R.K. Crouch, A.L. Fripp, WJ. Debnam, R.F. Berry and R. Simchick, A
procedure to visualize the melt-solid interface in Bridgeman grown germanium and lead tin
telluride, J.Cryst.Growth 74, 228-230, 1986.
8. K. Kakimoto, M. Eguchi, H. Watanaba and T. Hibiya, In-situ observation of impurity
diffusion boundary layer in silicon Czrochalski growth, J.Cryst.Growth 99, 665-669, 1990.
9. T.A. Campbell and J.N. Koster, Visualization of liquid-solid interface morphologies in
gallium subject to natural convection, J.Cryst.Growth 140, 414-425, 1994.
264
POLYTYPE STABILITY AND DEFECT REDUCTION IN 4H-SiC
CRYSTALS GROWN VIA SUBLIMATION TECHNIQUE
Or\
|"\
Q
Ok
C*
Hfl
R.Yakimova ' , T.Iakimov , M.Syväjärvi , HJacobsson , P.Räback , A.Vehanen , EJanzen
"Dept of Physics and Measurement Technology, Linköping University, S-581 83 Linköping, Sweden
b
Okmetic AB, Box 255, 17824 Ekerö, Sweden
'Center for Scientific Computing, P.O. Box 405, FIN-02101 Espoo, Finland
d
Okmetic Ltd., PO Box 44, FIN-01301 Vantaa, Finland
ABSTRACT
Reproducible growth of 4H-SiC with good crystalline quality has been obtained in a temperature
interval around 2350°C and on 4H-SiC C-face seeds. It has been observed that morphological
instability may appear at the initial stage of growth, causing formation of defects. Experimental
evidence has been found that supersaturation and surface kinetics are responsible for the polytype
stability, while growth front and growth mode address defect reduction. An explanation of the
findings has been suggested. It has been shown that starting the growth with a relatively low
growth rate (= 100 iim/h) can be beneficial for the crystal quality.
INTRODUCTION
Silicon Carbide (SiC) is a material of expectation for high temperature power switching and high
frequency power generation. While SiC may offer an exclusive combination of physical and
electronic properties for many applications, the high temperature and chemical stability of this
material, as well as the variety of stacking sequence along the c-direction in the close-packed
structure of SiC, cause difficulties for growth of device quality crystals, especially of large single
crystals.
Among different crystallographic modifications of SiC, 4H polytype is the most interesting for
power device applications. However due to the low stacking fault energy it is difficult to restrict
syntaxy (parasitic polytype formation) during bulk crystal growth and thus to grow a single
polytype material. Another well known problem is the large number of structural defects such as
micropipes, mosaicity and dislocations. Moreover, these problems are interrelated to a large
extent; defects easily result in polytype disturbances [1], while polytype inclusions may lead to
defect formation [2]. However, little is known about the kinetics and thermodynamics of polytype
formation, growth stability, and also the mechanism that produces the periodic sequences. As
discussed in Reference [3] 3C-SiC may be the initial polytype that forms at virtually all growth
temperatures and thus acts as a necessary precursor for the phase transformation to other
polytypes. Several growth parameters, such as the growth temperature [4,5], supersaturation
[3,4], vapor phase stoichiometry and impurities [3,6] and polarity of seed surface [7] have been
discussed to influence the polytype stability.
Seeded sublimation growth has been the most successful method to date for growth of large 4HSiC boules that can be sliced into wafers. 4H wafers of 50 mm diameter are commercially
available and 35 mm wafers with 7 micropipes (0.7 cm"2) have been reported [8]. Generally, it is
more difficult to grow 4H polytype in comparison with 6H-SiC, considering the size of crystals
and the yield. On a large scale, the problem with micropipes, dislocation networks and stress in
the crystals still remains [9]. Although significant progress has been made in the polytype control
and defect reduction in SiC crystals, all mechanisms governing these processes have not been
completely understood.
This paper describes a study of the influence of the early stage of 4H-SiC crystal formation on the
polytype uniformity and defect occurrence in the subsequently grown crystals. An attempt has
been made to reveal the role of growth characteristics such as seed surface polarity,
supersaturation distribution and temperature for the morphological stability.
265
Mat. Res. Soc. Symp. Proc. Vol. 572
e
1999 Materials Research Society
EXPERIMENTAL
The growth experiments were performed with seeded sublimation technique in an Epigress SB50
sublimation system [10] with inductively heated graphite crucible and movable RF-coil. Rigid
graphite insulation was used for thermal shielding and the reactor outer walls were air-cooled. The
temperature was monitored both at the top and the bottom of the crucible with two-color
pyrometers. The growth was performed in the temperature range of 2300-2450°C measured at the
bottom of the crucible (Tb) and maintained constant during the growth run. The top temperature
was used only as a reference when different runs were compared. Growth took place at a reduced
Ar pressure ranging 5-30 mbar depending on the growth temperature. The SiC source powder
was purified and sintered before growth to prevent contamination of the seed crystal at the initial
stage of the growth process. As seeds we employed 4H polytype crystals produced via
sublimation technique, (0001) well oriented or misoriented to the [1120] direction. Growth was
performed on both Si- and C-terminated face. This variety of seeds was particularly necessary in
order to investigate the seed influence on the polytype uniformity and structural quality. When the
growth temperature was reached under nearly an atmospheric pressure the growth was initiated by
applying a controlled pressure reduction. In order to vary the starting supersaturation we used two
different pressure reduction schemes following an exponential decay of 200 sec and 15 min. The
supersaturation was changed also by changing the temperature gradient when keeping constant Tb.
We examined two groups of samples. First, we studied as-grown surfaces after 6 hours of
growth, referred to as early stage of growth. This allowed observation of the growth morphology
after nu'cleation had been completed and the crystal habit had been founded. The second group of
samples comprised wafers cut from boules and properly processed to permit structural and
polytypic uniformity evaluation. The grown material was investigated using an optical microscope
with Nomarski interference contrast and crossed polarizers, as well as by high resolution X-ray
diffraction (HRXRD). Etching in molten KOH at 500°C for different times was also performed.
The optical investigations are suitable to observe micropipes and domain boundaries on bare or
KOH etched surfaces, while HRXRD was used to assess domain misorientation and strain in the
crystal by recording 0) rocking curves and 29/(0 diffraction curves. X-ray diffraction
measurements were utilized to determine the polytype. Computer simulation was used to better
understand temperature and supersaturation distributions near the growth interface.
RESULTS
Growth results showing conditions for 4H
crystal polytype occurrence and stability are
summarized in Table I. Stable 4H growth with
good structural quality is achieved within a
narrow temperature range, 2350°C-2375°C and
only on C-face. The pressure of the process gas
is 5 mbar but we did not observe an effect of the
pressure and pressure reduction rate on the
polytype formation. Similarly the type of source
material did not affect the polytype. The growth
rate is about 100 um/h in the beginning and it
reaches 0.5 mm/h approximately after 1 mm of
growth. Depending on crucible geometry the
crystal shape is either cylindrical or slightly
conical while the growth front is nearly flat or
slightly convex. The largest diameter of the
crystals is 38 mm. The thickness of the
polycrystal ring around the monocrystal area
does not exceed 5 mm.
266
Fig. 1. A typical pattern on as-grown
surface of 4H SiC, C-face.
Table I. 4H polytype occurrence at differenl temperatures and 4H seed orientations.
Growth
Surface Seed surface Grown crystal
temperature orientation
polarity
polytype
Remarks
[°C]
Si-face
6H, 15R inclusions J spiral growth competition
on-axis {
C-face
4H (100%)
2350-2375
i single spiral with preferred
C-face
4H(100%)
off-axis |
Si-face
6H, 15R inclusions J lateral growth in [1120]
<2350
low crystal quality
>2375
polytype conversion 4H =» 6H
Fig. 2 (a) Non-uniform growth morphology manifesting growth instability; (b) stable growth
morphology with one growth center.
Morphologies of as-grown surfaces on C-terminated faces of 4H SiC seeds at early stage of
growth showed patterns (Fig. 1), which have been previously reported [11]. Typically, C-face
exhibits a small pseudo-flat area (mesa) with six ridges emerging in the six equivalent directions
<1120>. Often polygonized spiral can_pe distinguished on the mesa. Microsteps are observed in
between the ridges, running along < 1100> directions and thus the pattern reflects the six-fold
crystal symmetry. Morphological stability at early stage of growth is an important characteristique,
which was found to affect 4H polytype stability and defect formation. Fig. 2 displays uneven
growth morphology (2a) and regular morphology (2b) in case of two different supersaturations
over the growing surface when the other process parameters are the same. In the first case one can
observe several misoriented growth centers with an enhanced growth at the edges. After several
millimeters of growth the growth front acquired a concave shape, resulting in formation of domain
boundaries, micropipes and dislocations. At some defective areas 15R inclusions were found.
Fig. 3 depicts calculated temperature at the axis and inner crucible walls at different coil positions,
provided Tb is fixed. From this and our temperature measurements it was estimated that
temperature gradient can be tuned from 4-5cC/cm to 15-20 °C/cm. Large values of the temperature
gradient resulted in morphological instabilities (Fig. 2a). To avoid this we used a low temperature
gradient (4-5°C/cm) corresponding to a growth rate of approximately 100 iim/h. When growth
conditions are properly selected there is only one growth promoting center (Fig. 2b) from which
steps spread out over the whole growing surface in the course of the crystal growth. We examined
the as-grown surface of the top wafer of a 4H SiC boule. An optical micrograph taken under
crossed polarizers is shown in Fig. 4. There is strain-associated contrast located at the growth
center. A wafer from the region just below the top surface was polished and subjected to structural
investigations. Fig. 5 represents images taken in reflection light (5a) and crossed polarizers (5b)
modes from 1 cm2 area of the wafer after KOH etching. It was possible to trace the growth
267
2600
0
+20
z(mm)
Fig. 3. Calculated temperature at the axis and
inner crucible walls (dashed line) at four
different RF-coil positions when the bottom
temperature is constant.
Fig. 4. Optical micrograph under crossed
polarizers of the as-grown surface of a 4Hboule.
promoting center and observe rows of dislocations along the six symmetrical ridges. Strains are
still seen, however no micropipes are visible exactly in the spiral center. The two large hexagonal
etch pits outside the spiral center do not appear to be micropipes. The sample was "mapped" by
utilizing co rocking curves. Fig. 6 gives two representative co-scans from two areas of the sample,
A and B, with a spot size of 2x14 mm. The peaks over a large part (A) of the sample (not
illustrated) are sharp and symmetric with high intensity and FWHM value of 14". On the area with
defects (B), corresponding to Fig. 5, the rocking curves show peak broadening, asymmetry and
lower intensity. The 26/co-scan, taken with a large spot of 10x14 mm, shows a sharp peak with
FWHM of 17" and an intensity of 40 000 counts per second.
DISCUSSION
Polytype stability
From our results it follows that the occurrence and stability of 4H polytype depend on the
temperature and the type of seed. Similar findings have been reported in Ref. [4,5] and [7],
respectively. The temperature range of 4H stability is different in different studies. Commonly the
temperatures are lower than for the growth of 6H-SiC. The upper limit in our experiments is
2375°C above which 4H polytype tends to convert to 6H, while below 2350°C the crystal quality
is the limiting factor (Table I). The authors of Reference [12] suggest that SiC polytypes are
kinetically determined metastable phases rather than true thermodynamic phases and therefore a
well defined temperature stability range for each polytype can not be expected. Note that in our
experiments the pressure is not effective in determining the polytype. As to the effect of the face
polarity of the seed, one can speculate that the low surface free energy of the C-face facilitates the
nucleation of 4H polytype which otherwise is more difficult to form due to the higher formation
enthalpy than that of 6H-SiC. Consequently, other polytype inclusions can occur when growing
on the Si-face. 4H polytype maintains stable growth in two modes depending on the seed
orientation. On on-axis seeds the growth takes place via spiral competition ending with a dominant
center in the middle of the crystal. The growth proceeds with layer-spiral mechanism. On off-axis
seeds the growth center is at the seed edge and the lateral step growth is more pronounced in the
tilt direction. Both situations allow growth of single polytype material.
268
Fig. 5. Optical micrographs taken from an etched (0001) 4H sample with defects under
(a) reflection light and (b) crossed polarizers; random lines are polishing scratches.
Morphological stability
Non-uniform growth illustrated in Fig. 2a can be discussed in terms of a morphological instability
due to an enhanced nucleation at the edges of the seed. This is known to appear in case of
polyhedral crystal growth when the supersaturation exceeds some critical value [13]. The solution
of the diffusion equation yields a non-uniform supersaturation over the growing face, being
largest at the corners and smallest at the centers of faces [14]. Therefore growth in diffusion
limited sublimation growth is favored at the seed edges. Two consequences may be expected then:
(i) many growth centers with different orientations, respectively domains forming grain
boundaries with defects [2] and (ii) non-uniform growth rate leading to non-uniform material
properties, e.g. inclusions. On the other hand, surface kinetics may smooth the growing surface
by adjusting the density of growth steps. This can happen if the surface kinetics is fast enough
and supersaturation is not too high. We have shown that by decreasing the supersaturation (low
growth rate) and increasing the seed temperature,
growth can start with a stable morphology (Fig.
2b). The described instabilities are more
(0004)
*
pronounced in 4H growth due to a lower surface
6 - reflection
I ^^^ area A
diffusion coefficient and higher supersaturation
"©
over the growing surface in comparison with 6H
S
*
growth. To confirm that growth favored at the
,_,
edges of the seed is not due to an incorrect
S. 4
u
crucible design, we performed growth on two
adjacent seeds. The results have clearly indicated
>. 3
1*
that this is not the case. These experiments are
GA
areaB
jt \
described in Reference [14]. There could be
0 2
+rf
several other solutions to the discussed problem.
a
1-1
One is to use a large area seed. Our tests indicated
1
y,
that growth can proceed in an uniform way once
/-/
0
the tendency of many center growth is
i
i
i
i
i
suppressed. Similarly, morphologically stable
0.02
-0.02
0
growth can be achieved if diffusion mass
Aco[degrees]
transport is replaced by Stefan flow mass
transfer. Finally, we simulated a particular
Fig. 6. Two representative co-scans from
temperature profile over the seed, which may lead
area B shown in Fig. 5 and a good quality
to an even growth rate at each point of the seed
area A.
[15].
1
• 1
269
Defects
The probability of defect formation during growth increases with increasing the number of growth
centers. This motivates the importance of starting boule growth with only one growth center.
According to our results a large number of dislocations penetrating the growing surface may be
formed in the vicinity of the growth promoting center. These are not dislocations associated with
the spiral growth mechanism since we do not observe spirals emerging from them. They are found
at the ridges spreading in six equivalent <1120> directions, Fig. 5a. We suggest an explanation
of the defect origin based on the anisotropy of the step distribution from the growth center e.g. the
step velocity is different in the <1120> and <1100> directions. Furthermore, because of six-fold
symmetry of (0001) SiC face the steps undergo transitions at the <1120> directions to the
adjacent facet. This may lead to strain accumulation, which later relaxes in dislocation formation.
The shape of the rocking curves indicates that in the defective area there are domains that are
slightly misoriented only but do not form a mosaic structure. In addition, by optical examination
of the etched sample we did not observe imperfect domain boundaries. It is worth noting that the
average micropipe density in this sample is 170 cm'2 which corresponds to the micropipe density
in the seed crystal. It seems that our growth regimes do not provide conditions for creating new
micropipes by opening hollow cores of giant screw dislocations. The average density of randomly
distributed dislocations is about 15 000 cm"2.
CONCLUSION
4H polytype crystals have been grown having a single polytype structure and micropipe density
comparable with the seed. By controlling morphological stability, growth defects, such as
micropipes and mosaicity, can be reduced. Concerning 4H polytype growth stability, surface
kinetics plays an important role most probably due to the lower surface diffusion coefficient and
lower equilibrium vapor pressure compared with 6H. We propose to start the growth with a
relatively low growth rate compared with the following boule growth, which limits defect
formation and ensures polytype uniformity.
ACKNOWLEDGMENT
The SSF SiCEP Program, NUTEK and Okmetic Ltd. are gratefully acknowledged for support.
REFERENCES
1. M. Syväjärvi, R. Yakimova, P-A. Glans, A. Henry, M.F. MacMillan, L.I. Johansson, and E. Janzen,
J. Cryst. Growth 198-199 1019-1025 (1999).
2. M. Tuominen, R. Yakimova, E. Prieur, A. Ellison, T. Tuomi, A. Vehanen and E. Janzen, Diamond
and Related Materials 6 1272-1275 (1997).
3. Yu.M. Tairov and V.F. Tsvetkov, Progr. Crystal Growth Characterization 4 111-162 (1982).
4. M. Kanaya, J. Takahashi, Y. Fujiwara, and A. Moritani, Appl. Phys. Lett. 58 56-58 (1991).
5. G. Augustine, H. McD. Hobgood, V. Balakrishna, G. Dunne, R.H. Hopkins, Phys. Stat. Sol. (b)
202 137-148 (1997).
6. Yu.A. Vodakov, E.N. Mokhov, A.D. Roenkov, M.M Anikin, Sov. Tech. Phys. Lett. 5 147-148
(1979).
7. R.A. Stein and P. Lanig, J. Cryst. Growth 131 71-74 (1993).
8. C.H. Carter, Jr., V.F. Tsvetkov, D. Henshall, O. Kordina, K. Irvine, R. Singh, S.T. Allen and J.W.
Palmour, 2nd ECSCRM'98, Sept. 2-4, 1998, Montpellier, France, Abstracts pp.1-2 (1998).
9. P.G. Neudeck, W. Huang, and M. Dudley, Mat. Res. Soc. Symp. Proc. 483 285-294 (1998)
10. Epigress product information bulletin, 98.11.04
H.A. Okamoto, N. Sugiyama, T. Tani and N. Kamiya, Mat. Sei. Forum 264-268 21-24 (1998).
12. S. Limpijumnong and W.R.L. Lambrecht, Phys. Rev. B 57 12017-12022 (1998).
13. T. Kuroda, T. Irisawa, A. Ookawa, J. Crystal Growth 42 41-46 (1977).
14. R. Yakimova, M. Syväjärvi, M. Tuominen, T. Iakimov, P. Räback, A. Vehanen, and E. Janzen,
Mater. Sei. Eng. B (1999) in press
15. P. Räback, Ph.D. Thesis, Helsinki University of Technology, Finland (1999).
270
Growth and Characterization of 2" 6H-SHicon Carbide
Erwin Schmitt, Robert Eckstein, Martin Kölbl, Amd-Dietrich Weber
SiCrystal AG, Heinrich-Hertz-Platz 2,92275 Eschenfelden, GERMANY
Email: e.schmitt@sicrystal.com
ABSTRACT
For the growth of 2" 6H-SiC a sublimation growth process was developed. By different
means of characterization crystal quality was evaluated. Higher defect densities, mainly in
the periphery of the crystals were found to be correlated to unfavourable process conditions.
Improvement of thermal boundary conditions lead to a decreased defect density and better
homogeneity over the wafer area.
INTRODUCTION
During the last years silicon carbide has attracted more and more attention in many fields of
semiconductor research. A rising number of companies choose silicon carbide as a substrate
for device processing. The production of the blue Light Emitting Diodes based on 6H-SiC
has already started. Mass production is first of all a question of yield. Thus the prerequisites
for silicon carbide are size and quality of the wafers. In the case of 6H silicon carbide the
minimum diameter desired is 2". It is inevitable, that wafers have to fullfill the quality
requirements not only in a selected, but on a large fraction of their area. Achieving
homogeneous material properties is one of the major tasks at the present stage of SiC crystal
growth.
CRYSTAL GROWTH AND CHARACTERIZATION
Growth of semiconductor grade silicon carbide by sublimation was first reported by Lely [1]
in 1955. Another breakthrough in bulk growth was marked by Tairov and Tsvetkov [2], who
used single crystalline material as seeds. During the following years the sublimation
technique for silicon carbide has been further improved [3, 4]. Our growth experiments for
6H-SiC are based on the principles of the above mentioned technique. Evaluation of material
properties was performed by different means of characterization, as described in the
following: Optical microscopy of grown surfaces and wafers was carried out. The utilization
of crossed polarizers enables stress birefringency images of the area under observation.
Micropipe densities (MPD) were determined by focussing through the complete sample
thickness. For a selected amount of samples also KOH etching (10 minutes at 600 °C) and
scanning electron microscope investigations were conducted. No additional micropipes
could be detected by these methods, but other defects like grain boundaries, cracks,
dislocations and inclusions could be revealed. To determine the composition of inclusions,
scanning auger spectroscopy was applied. From rocking curves (XRD-analysis) the full
width at half maximum (FWHM) value of the (0006) reflexion was determined. The
broadening of the FWHM-value by a superposition of indivual peaks could be interpreted as
existence of slightly misoriented grains [5]. Measuring the sheet resistance (induction of
eddy currents), which is a comfortable and non-distructive technique, allows the calculation
of the specific electrical resitivity. A combination of resistivity and Hall-measurements gives
the correlation between specific electrical resistance and net doping concentration ND-NA.
271
Mat. Res. Soc. Symp. Proc. Vol. 572 e 1999 Materials Research Society
RESULTS
Test-area
The micropipe density is an
important value to determine the
material quality of the grown
crystals. After upscaling of boule
diameter we found out, that there
was a strong variation of the MPD
over the 2"-wafer. In all cases the
MPD at the periphery was
significantly higher than in the center
of the wafer. Figure 1 illustrates, how Figure 1: Determination of average micropipe densities (MPD)
average micropipe densities were
+ +0 = 2";
B+G = 1,625";
o=l,4"
determined, taking into account
different diameters for the area
under observation. Applying this
method to our 2"-wafers we found,
that for a "virtual" diameter of up to
1,625" the average MPD was below
200 cm"2 (see Figure 2, series I). For
the whole wafer ("real" diameter of
2") the average MPD was up to 380
cm"2. These growth experiments of
series I also showed a poor
crystalline quality according to the
s
rocking curves of X-ray diffraction.
°i
Additionally stress birefringency
images of wafers cut from these
wafer diameter (inch)
boules were made, to reveal the
lateral variation of stress contrast.
Figure 2: Dependence of average micropipe density
One typical image is illustrated in
(MPD) on wafer area under observation
Fig. 3a. Areas with low stress
for 2" 6H-SiC wafers
contrast are restricted to a small
central region of the wafer. Most of the wafer area has a high contrast in the stress
birefringency. Both, micropipe-distribution and stress birefringency show, that the desired
aim of homogeneity of crystal quality over the whole wafer was not achieved so far.
n
Figure 3b: Stress birefringency image
ofa2"6H-SiC wafer with
a nearly contrast-free
central region.
Figure 3a: Stress birefringency image
of a 2" 6H-SiC wafer with a
high stress contrast over a
large area.
272
For the investigation of the influence of different process conditions on the crystal quality,
two comparable wafers from the same boule were selected. Therefore the influence of seed
quality can be neglected. These samples showed nearly identical micropipe densities and
micropipe distribution, stress birefringency images and FWHM-values. From these two
seeds two 6H-SiC crystals were grown under different thermal conditions. The results can be
seen in Figure 4. High temperature gradients (indicated by T) lead to a poorer crystalline
quality (Al), while more moderate thermal boundary conditions (indicated by 4-) resulted in
smaller FWHM-values (BY). By using wafers from the boules Al and Bl as seeds for the
subsequent growth-runs, the two experimental series A and B were carried out. In all
experiments the achieved crystalline quality could be correlated with the applied thermal
conditions. Based on these results a significant improvement of wafer quality became
possible. From Figure 2 (series II) one can see a reduction of average micropipe densities
below the level of 200 cm"", over the whole wafer area. A comparision between Figures 3a
and 3b shows an obvious stress reduction.
In addition inclusions in silicon carbide were examined with auger electron spectroscopy.
These inclusions could generate micropipes and are so far ascribed to graphite [7, 8].
SSÖPiow
Si-rich
Figure 5a: SEM-image of an
inclusion, exposed after
successive sputtering.
Figure 5b: AES-image for
Carbon-Mapping
Figure 5c: AES-image for
Silicon-Mapping
Fig. 5a shows inclusions, that were exposed on the surface of silicon carbide samples after
successive sputtering with a scanning auger microscope. Silicon- and carbon-sensitive augermappings of this surface showed that these inclusions have a high carbon and a low silicon
content, relative to the surrounding silicon carbide matrix (Fig 5b, 5c). Carbonization due to
unfavourable process conditions could be the origin of these inclusions. The reduction of the
inclusion content could contribute to a further lowering of MPD.
Besides crystal quality and its lateral distribution, the homogeneity of electrical properties of
wafers is also of great importance. The results of a typical specific electrical resistivity
mapping is shown in Fig. 6. The resistivity lies between 0.0382 ficm and 0.0496 ficm, with
an average value of 0.0451 ficm and a standard deviation of 0.0028 ficm. The average value
for the resitivity corresponds to a net doping
concentration ND-NA of 3,3*1018 cm"3 .
,,,-<"7" \~"~~----.
SUMMARY AND OUTLOOK
x
We have shown that the enlargement of crystal
diameter can lead to a decreased crystal quality,
especially in the periphery of the crystals. By
improving process conditions we succeeded in
growing boules with significantly reduced defect
densities and improved homogeneity. For further
investigations, concerning the connection between
process conditions and crystal quality, additional
quality criteria like dislocation density, grain
273
y.
Figure 6: Mapping of specific electrical
resistivity of a nitrogen-doped
2" 6H-SiC wafer.
boundaries, etc. have to be taken into account. The calculation of the temperature
distribution inside the growth reactor by means of numerical modelling will also contribute
to a better understanding of the interactions of the growth process.
ACKNOWLEDGEMENTS
This work is supported by the Bavarian Research Foundation under contract No. 176/96. The
authors wish to thank the partners of the above mentioned cooperation project for their
contributions. That are: Inst. of Materials Science 6, Dept. for Appl. Physics and Dept. of
Fluid Mechanics (all University of Erlangen-Nürnberg). Special thanks to Inst. of Materials
Science 4 (University of Erlangen-Nürnberg) for auger scanning microscopy.
REFERENCES
[1]
J. A. Lely, Ber. Deut. Keram. Ges., 32,229 (1955)
[2]
Yu.M. Tairov, V.F. Tsvetkov, J. Cryst. Growth 43,209 (1978)
[3]
G. Ziegler, P. Lanig, D. Theis, C. Weyrich, IEEE TRANSACTIONS ON ELECTRON
DEVICES; VOL. ED-30, No. 4,277 (1983)
[4]
D.L. Barret, J.P. McHugh, H.M. Hobgood, R.H. Hopkins, P.G. McMullin, R.C.
Clarke, J. Cryst. Growth 128,358 (1993)
[5]
M. Tuominen, R. Yakimova, R.C. Glass, T. Tuomi, E. Janzen, J. Cryst. Growth 144,
267 (1994)
[6]
S. Milita, R. Madar, J. Baruchel, A. Mazuelas, Materials Science Forum Vols. 264268,29 (1998)
[7]
T. Tsvetkov, R. Glass, D. Henshall, D. Asbury, C.H. Carter, Jr., Materials Science
Forum Vols. 264-268,3 (1998)
[8]
D. Hofmann, M. Bickermann, R. Eckstein, M. Kölbl, E. Schmitt, A. Weber, A.
Winnacker, J. Cryst. Growth (1999), in press
274
EXPERIMENTAL AND THEORETICAL ANALYSIS OF THE
HALL-MOBILITY IN N-TYPE BULK 6H- AND 4H-SIC
ST.G. MÜLLER, D. HOFMANN, A. WINNACKER
Materials Science Institute VI, University Erlangen-Nürnberg,
Martensstr. 7, D-91058 Erlangen Germany
ABSTRACT
The electrical properties of nitrogen doped n-type 6H- and 4H-SiC bulk crystals grown by
the Lely- or modified Lely-method have been characterized by Hall-measurements. The
doping densities were determined by a fit of the neutrality equation to the experimental
data, accounting for in-equivalent lattice sites and the temperature dependence of the effective density-of- states-mass extracted from recent results of ab-initio-calculations of the
6H- and 4H-SiC bandstructure [1]. The theoretical analysis of the Hall-mobility is based
on an extended form of the Rode-Nag iteration algorithm [2]. The calculation scheme considers all relevant elastic and inelastic scattering mechanisms, the anisotropy of the crystal
modifications and the possible effect of spatial inhomogeneities in the distribution of donors,
acceptors or in the related electron system. Within these concepts it is possible to achieve
a quantitative agreement between theoretical and experimental mobility data in 4H- and
6H-SiC over the whole temperature range of band conduction. New values for the acoustic
deformation potentials Eac [15.0±0.5 eV (6H), 14.8±0.5 eV (4H)] and the coupling constants
for intervalley phonon scattering Dint [2.3 ± 0.1 x 109 eV/cm (6H), 2.6 ± 0.1 x 109 eV/cm
(4H)] are given.
INTRODUCTION
For the understanding of the limiting factors of the electron mobility in SiC a detailed analysis
of the scattering mechanisms is essential. A principal problem for early scientific approaches
to this subject [3, 4, 5] was the lack of critical material parameters of SiC at that time.
Recent theoretical results of ab-initio-calculations of the 6H- and 4H-SiC bandstructure [1]
and experimental data for the effective electron masses [6, 7, 8] provide the basis for a quantitative, theoretical mobility-analysis of these polytypes. Until now a detailed calculation of
the electronic mobility of SiC without the relaxation time approximation was only performed
for the cubic modification (3C-SiC) [9]. Similar GO tliis woric the mobility-analysis of n-type
6H- and 4H-SiC bulk crystals grown by the Lely- or modified Lely-method presented m this
study is based an extended version of the Rode-Nag iteration method [2], but modified to
account for the specific conductivity-anisotropy [10] of these hexagonal polytypes.
EXPERIMENT
The investigated nitrogen doped n-type 6H- and 4H-SiC samples (typical dimensions:
5 x 5 X 0.5 mm3 with the orientation of the basal plane perpendicular to the cristallographic
c-axis) were cut from crystals grown by the Lely- (L) or modified Lely-method (ML). Further details of the SiC sublimation growth at the Materials Science Institute VI (Erlangen,
Germany) can be found in [11]. Ohmic contacts at the four corners of the samples were
prepared by vaporization of titanium, followed by an annealing step at 900°C for 10 mm
in vacuum. For the measurement of the electrical conductivity a and the Hall-constant RH
at an magnetic field B of 0.42 T the van der Pauw [121 method was used: The electron
density n = rH/(e\RH\) and the Hall-mobility jiH = a{B = Q)\RH\ were calculated with
the simplified assumption of the Hall-scattering factor rH = 1. Based on the number of
hexagonal and cubic lattice sites the donor concentration ratios ND{h) : ND(k) for 4H- and
6H-SiC are 1 : 1 and 1 : 2, respectively. Hereby the energetic difference of the two cubic
sites (kuk2) in 6H-SiC is too small to be seperated by Hall-measurement. The donor concentration ND = ND(h) + ND(k), the compensating acceptor concentration Ncomv and the
ionization energies (AE(h), A£(fc)) were determined by direct numerical fitting of the neutrality equation to the n(T) data. In the neutrality equation the temperature dependence
of the effective density of states mass [1] was considered and the valley-orbit splitting of the
275
Mat. Res. Soc. Symp. Proc. Vol. 572 ° 1999 Materials Research Society
nitrogen ground-state [13, 14, 15] was taken into account by modifying the spin degeneracy
factor accordingly [16]. The numerical fitting results for the samples of Fig.l and Fig.2 are
summarized in the following table:
polytype
6H
4H
sample
ML1
LI
L2
ML4
ML5
ND
2.3
3.2
7.8
1.5
3.8
Ncomp [cm'3}
2.9 x 1016
4.0 x 101'
1.4 x 1018
6.7 x 1016
1.1 x 1018
[cm-*]
x 1017
x 1018
x 1018
x 1017
x 1018
&E{h) [meV]
94
77
72
51
53
A£(fc) [meV]
119
112
101
109
98
The dependency of the activation energies from the doping concentrations can be understood
within the theory of the formation and broadening of a donor impurity band [17]. Due to the
comparatively high donor binding energies in SiC the mechanism of Hopping-conductivity
[17] becomes important at relatively high temperatures of 40-75K, depending on doping and
compensation. Connected to this is the high temperature dependence of fiH (Fig. 1 and Fig.2)
in this temperature regime [18], where the theoretical interpretation within the formalism of
the So/izmann-equation used in the following chapter is not valid.
From the Hall-analysis of a series of n-type bulk 4H- and 6H-SiC samples, typically an
increase of iVcomp was found with increasing iVD, despite the fact, that in the low temperature
photoluminescence of the ML-samples no donor-acceptor pair spectrum was visible in the
blue spectral region. This is to be expected for the presence of flat acceptors like Al [19].
The photoluminescence spectrum is dominated by a broad emmission at 1.8 eV (6H) and
2.1 eV (4H), which can be correlated to the presence of deep intrinsic acceptors [20] and the
possible mechanism of self-compensation [21].
THEORY
The presented, theoretical analysis of the Hall-mobility is based on the linearized Boltzmarmequation. In the underlying coordinate system the z-axis is by definition parallel to the
c-axis of the hexagonal unit cell of SiC, while the y-axis is parallel to one of primitive
basis-vectors ax of the hexagon and the x-axis perpendicular to the corresponding face. For
the Hall-geometry under investigation with the electrical current density j ± c-axis and
the magnetic field B - (0,0,5-) || c-axis the electrical field is without loss of generality
defined by £ = (£x,0,0). The results of bandstructure calculations show, that due to
the low symmetry (C2v) of the fc-vector in the M - L direction of the Brillouin-zone, the
tensor of the effective electron mass for 4H- und 6H-SiC has 3 independent components
[1]. In the parabolic approximation for 6H- and 4H-SiC there are respectively 6 and 3
equivalent conduction band minima with ellipsoidal surfaces of constant energy described
by the effective masses parallel to the principal axis (mi (|| M — T), m2 (l| M - A'), m3
[M - L)) have to be considered. Using the Herring-Vogt Transformation [2] the ellipsoidal
energy surface in fc-space can be transformed to a spherical surface in the new w-space with
the energy dispersion Ew = h2w2/2mIT (my. arbitary normalization factor). The definition
of effective field intensities finally leads to an equivalent description of the carrier transport
by the 5o/femann-equation as in the case of an isotropic effective electron mass. After
linearization the coupled equations for the calculation of the electronic relaxation functions
<l>x and <j>y (not to be confused up with relaxation times) are given by [2]
Lc4>x — Uz4>y
UJzrj>x + Lc<t>y
(1)
(2)
= 1
= 0
with u, = eßz/(m1m2)1/'2 and the collision operator
Lc 4>i =
(m1m2m3)1!2
Vc
3
(2*)
3/2
ITlff
r
, 1 - ME') „,
Jdw T^ME)T^
276
HE)--HE1)
(3)
(/0: Fermi-Dirac distribution, Twwr. transition probability, E = Ew, E' = Ewi). After
the back-transformation from w-space to fe-space, taking into account the configuration of
the equivalent conduction band minima, the conductivity tensor for 4H- and 6H-SiC for the
special configuration B || c-axis is:
/
&=
0-12
0
-al2 an
ffii
0
0
0
\
(4)
(733
with
=
n£ t(M
+
(M) ,
ai2 =
and
9
^£^_ (^ , ,33 = ^ (MB = 0)>
(5)
oo
{h) = -ZJdE4nEl/JdEf0Et
(6)
o
o
In this case the Hall-constant and Hall-mobility are connected to the components of the
electrical conductivity tensor by:
RH
= <T« [Bz (<r2n + alt)}'1 , HH = <TII(0) \RH\
(7)
Within the existing accuracy of the mobility measurements and the lack of accurate data for
electron-phonon coupling constants in 4H- and 6H-SiC the extensive calculations involved
in an exact evaluation of (3) cannot be justified. Therefore in this work the transition
probability Tw wi is approximated as an isotropic process with the effective density of states
mass rrii = (m1m2m3)1/'3, which is exact for randomising scattering. The anisotropy of the
conductivity is then primarily connected to the superposition of the different contributions
to the conductivity expressed in (5). All collision integrals in (3) for the elastic and inelastic
scattering processes [2, 22] were calculated by numerical integration. No approximations
(like e.g. the Brooks-Herring formula to describe impurity scattering) were used, as this
turned out to be critical for the accuracy of the results. It has also to be stressed, that due
to the relatively high LA- (6H: 76.5 meV, 4H: 76.6meV [23]) and LO- (6H: 119.98meV,
4H: 119.73 meV [7]) phonon energies involved in the inelastic electronic scattering processes
in SiC (inter-valley scattering: LA, non-polar and polar optical phonon scattering: LO)
theoretical results based on the relaxation time approximation [24] are strictly speaking not
valid. Therefore in this work the calculations are based the Rode-Nag iteration method [2],
which was extended to account for several inelastic processes at one time. It was shown by
Tsukioka [9], that for the explanation of mobility data of 3C-SiC in addition to the scattering
by isolated ionized impurities it was necessary to introduce the concept of dipole-scattering
[22] to the theoretical mobility analysis, accounting for inhomogeneities in tue impurity or
related electron configuration. Also for a consistent mobility-analysis of n-type bulk 4H- and
6H-SiC it turned out to be necessary to split the concentration of ionized impurities into
Ndipole
=
XdipoU
NX ~ Xdipok
NA
Kiütei = K + N2- 2Ndipole a n + 2NA(l - xdipole)
(8)
(9)
with the fraction xdipoie of the acceptors (density: NA) forming dipoles of a length diivoi<.
with donors (density: ND). Possible reasons for this may be attributed to the already mentioned self-compensation mechanisms or the preferencial formation of so called " 1-complexes"
in a impurity band [17]. The considered elastic scattering processes additionally included
acoustical deformation potential (Eac) scattering, while piezoelectric- and neutral impurityscattering were ineffective for the investigated samples in the temperature regime of band
277
conduction. Eac, xdipoh, ddipoie and Dint (coupling constant for intervalley-scattering)_ were
used as fitting parameters for the calculations. The coupling constant Dnpo of the negligible
non-polar optical phonon scattering (hu>io > tt^LA^ cannot be extracted by the fitting
process. Thus Dnpo = D{nt was defined. Within these concepts, for the first time a consistent, quantitative agreement between theory and experiment (Fig.l and Fig.2) could be
achieved within the entire range of band conduction. An increase of xdipoie is found with
increasing donor concentrations. The fitting results correspond well to the parameters found
by Tsukioka [9] for 3C-SiC:
polytype
3C [9]
6H
4H
Eac [eV\
14
15 ±0.5
14.8 ±0.5
Dint [eV/cm]
2.2 x 109
2.3 ±0.1 x 109
2.6 ±0.1 x 109
ddipoh [nm]
0.74
0.7 ±0.05
0.7 ±0.05
^dipole
100% (postulate)
variable
variable
The temperature dependence of the Hall-factor rfj(T) also resembles the principal dependence, recently measured for 4H-SiC samples [25]. By correcting from some examples the free
electron concentration n(T) with the theoretical results of rH{T) ^ 1 the following values
are found after a new fit of the neutrality equation:
polytype
6H
4H
sample
ML1
L2
ML4
ML5
ND
2.2
7.8
1.7
4.1
[cm-3]
x 1017
x 101S
x 1017
x 1018
IVcomp \CTTi
j
10
2.6
2.0
6.3
1.3
x
x
x
x
10
1018
1016
101S
AE{h) [meV]
96
61
53
47
AE(k) [meV]
135
98
116
97
The deviations from the previous fitting results with ru = 1 are within the limits given by
the accuracy of the underlying experimental data and the parameters used for one fitting
procedure. It is therefore important to note, that the given analysis justifies the use of
TH = 1 for the evaluation of SiC-Hall-data for practical purposes. Only by this fact it can be
understood, that it was possible to achieve a quantitative aggreement between the described
theory and experimental data at all. Otherwise a far more complex iteration procedure
between the mobility calculation and correcting the experimental data with the theoretical
rji(T) values would have been necessary.
CONCLUSIONS
The doping densities and ionization energies of n-type bulk 6H- and 4H-SiC samples were
extracted from Hall-measurements considering recent results of ab-initio bandstructure calculations [1]. The theoretical analysis of the Hall-mobility is based on an extended version
of the Rode-Nag iteration algorithm [2] taking into account all relevant elastic and inelastic
scattering mechanisms. In order to achieve a consistent, quantitative agreement to experimental data the concept of dipole scattering has to be introduced into the analysis to account
for possible spatial inhomogeneities in the impurity- or related electron-system. The given
fitting parameters for the acoustic deformation potentials and the inter-valley coupling constants are close to recently published values extracted from high-field transport Monte Carlo
simulations for 4H- and 6H-SiC [26]. The results also justify the use of rH = 1 for practical
purposes of Hall-data evaluation. By transfering the presented theoretical concept of dipolescattering for the interpretation of Hall-data of SiC epi-layers one has to keep in mind, that
the Hall measurements in this case may be strongly influenced by the conductive substrate
and as the growth conditions for epi-layers are typically quite different from SiC bulk growth,
this may particulary influence selfcompensation effects and the spatial distribution of donors
and acceptors.
ACKNOWLEDGEMENTS
The author wants to thank A. Schöner (Industrial Microelectronics Center, Kista, Sweden)
for providing Hall-data of low doped 4H-SiC samples. This work was financially supported
by the Bundesministerium für Bildung und Wissenschaft (BMBF) (FKZ: 03 M 2746)
278
I03r
104
s^— acoustic
•.
\
\ -<' 1i non-polar^ - ^ \ \ \ optical
ioionized imp. /
x
10'
Figure 1. Theoretical and experimental Hall-mobility in 6H-SiC: a) scattering mechanisms, b)
Hall-mobility \iH and Hall-factor rH for different samples {xdiVcie- ML1(0%), Ll(40%), L2(70%)]
104
ui I |
a) \
\\ \ .
acoustic —-N
£ itf-
\\/ \
ionized imp./
/ *i\
/ inter-valley ^
S
u
non-polaroptical
>
polaroptical
>
\
\
\ \
X
i
1C? 7
\
\
Symbol M-Lely
•
....
i
100
\
ML4
.
,
300
. . \. 1
1000
T [K]
T [K]
Figure 2. Theoretical and experimental Hall-mobility in 4H-SiC: a) scattering mechanisms, b)
Hall-mobility \iH and Hall-factor rH for different samples [xdipoie: ML4(0%), ML5(15%)]
279
REFERENCES
[1] G. Wellenhofer, U. Rössler, phys. stat. sol. (b) 202, 107 (1997)
[2] B.R. Na,g,Electron Transport in Compound Semiconductors, vol. 11 of Springer Series
in Solid-State Sciences (Springer-Verlag, Berlin, 1980)
[3] L. Pattrick, J. Appl. Phys. 38, 50 (1967)
[4] J.J. Daal, Philips Res. Rept. Suppl.3 70, 1 (1965)
[5] B.W. Wessels, H.C. Gatos, J. Phys. Chem. Solids 38, 345 (1977)
[6] N.T. Son, O. Kordina, A.O. Konstantinov, W.M. Chen, E. Sörman, B. Monemar, E.
Janzen, Appl. Phys. Lett. 65, 3209 (1994)
[7] H. Harmia, S. Nakashima, T. Uemura, J. Appl. Phys. 78, 1996 (1995)
[8] D. Volm, B.K. Meyer, D.M. Hofmann, W.M. Chen, N.T. Son, C. Persson, U. Lindefelt,
O. Kordina, E. Sörmann, A.O. Konstantinov, B. Monemar, Phys. Rev. B 53, 15409
(1996)
[9] K. Tsukioka, Inst. Phys. Conf. Ser. 142, 397 (1996)
[10] M. Schadt, G. Pensl, R.P. Devaty, W.J. Choyke, R. Stein, D. Stephani, Appl. Phys.
Lett. 65, 3120 (1994)
[11] St.G. Müller, R. Eckstein, W. Härtung, D. Hofmann, M. Kölbl, G. Pensl, E. Schmitt,
E.J. Schmitt, A.-D. Weber, A. Winnacker, Mat. Sei. For. 264-268, 33 (1998)
[12] L.J. van der Pauw, Philips Res. Rep. 13, 1 (1958)
[13] W. Suttrop, G. Pensl, W.J. Choyke, R. Stein, S. Leibenzeder, J. Appl. Phys. 72, 3708
(1992)
[14] W. Götz, A. Schöner, G. Pensl, W. Suttrop, W.J. Choyke, R. Stein, S. Leibenzeder 73,
3332 (1993)
[15] P.A. Colwell, M.V. Klein, Phys. Rev. B 6, 498 (1972)
[16] J.S. Blakemore, semiconductor statistics, (Pergamon Press, Oxford, 1962)
[17] B.I. Shklovskii, A.L. Efros, Electronic Properties of Doped Semiconducters, vol. 45 of
Springer Series in Solid-State Sciences (Springer Verlag, Berlin, 1984)
[18] B. Molnar, J. Mat. Res. 7, 2465 (1992)
[19] M. Ikeda, H. Matsunami, T. Tanaka, Phs. Rev. B 22, 2842 (1980)
[20] A. Wysmolek, P. Mroziriski, R. Dwiliriski, S. Vlaskina, M. Kamiiiska, Acta Physica
Polnica A 87, 437 (1995)
[21] S.I. Vlaskina, Y.P. Lee, V.E. Rodionov, M. Kamiiiska, Mat. Sei. For. 264-268, 577
(1998)
[22] R. Stratton, J. Phys. Chem. Solids 23, 1011 (1962)
[23] W.J. Choyke, R.P. Devaty, L.L. Clemen, M.F. MacMillan, M. Yoganathan, G. Pensl,
Inst. Phys. Conf. Ser. 142, 257 (1996)
[24] T. Kinoshita, K.M. Itoh, J. Muto, M. Schadt, G. Pensl, K. Takeda, Mat. Sei. For.
264-268, 295 (1998)
[25] G. Rutsch. R.P. Devaty, D.W. Langer, L.B. Rowland, W.J. Choycke, Mat. Sei. For.
264-268, 517 (1998)
[26] R. Mickevicius, J.H. Zhao, Mat. Sei. For. 264-268, 291 (1998)
280
MID INFRARED PHOTOCONDUCTIVITY SPECTRA OF DONOR
IMPURTTTES IN HEXAGONAL SILICON CARBIDE
R. J. LINVILLE, G. J. BROWN, W. C. MITCHEL, A. SAXLER AND R. PERRIN
Air Force Research Laboratory, Materials & Manufacturing Directorate (AFRL/MLPO)
3005 P ST, Wright-Patterson AFB, OH 4533-7707
ABSTRACT
Mid-infrared photoconductivity (PC) is a useful technique for identifying and investigating
donor and acceptor centers in many semiconductors. This is especially true when the PC results
are combined with other measurements such as Hall Effect and DLTS. We report on the first
Fourier Transform Infrared (FTIR) photoconductivity spectra for n-type 6H and 4H-SiC. The
samples studied had temperature dependent Hall activation energies around 45 meV and 85 meV
in the 4H samples, and a single activation energy of 106 meV in the 6H. For the 4H samples, the
PC spectra showed an increase in photoresponse between 40 and 47 meV, with another sharp
increase at 120 meV. In the 6H-SiC, the photoresponse also had a rapid increase at 120 meV, and
at 77 meV in one sample. The photoresponse spectra of the n-type 4H and 6H-SiC samples
were distinctly different in the mid-infrared.
INTRODUCTION
SiC has, in comparison with Si, superior properties regarding high-power, high-frequency and
high-temperature electronics. The material has high thermal conductivity, can withstand high
electric fields before breakdown and also high current densities. The wide bandgap results in a
low leakage current even at high temperatures. The bandgap of SiC depends on the polytype and
ranges from 2.4eV (3C-SiC, T=4.2K) to 3.3eV (2H-SiC, T=4.2K).' Nitrogen is the main donor
impurity in all the polytypes of SiC2 and dominates the electrical properties of n-type SiC since
it has shallow levels in the upper half of the forbidden gap. Nitrogen substitutes for carbon in the
SiC lattice.2 The activation energy of the nitrogen levels depends on the polytype and the
substitutional lattice site involved. In recent years, there have been several studies on identifying
the nitrogen levels in 6H and 4H silicon carbide.
One technique that has not been used previously to study these energy levels is mid-infrared
photoconductivity. The only reported photoconductivity spectra for SiC are for deep levels
observed at energies higher than 0.5 eV. Mid-infrared photoconductivity (0.025 to 0.5 eV) can be
a sensitive technique for identifying both defect levels and impurities in semiconductors. •
Typically the optical activation energies identified in photoconductivity spectra agree well with
the thermal activation energies measured by temperature dependent Hall effect for samples from
the same wafer. In this study we compare the results from Hall effect and Fourier transform
infrared spectral photoconductivity for several samples of 6H and 4H n-type SiC.
EXPERIMENTAL
The 4H-SiC and 6H-SiC samples have been grown by physical vapor transport. All of the
samples were found to be of the n-type conductivity by Hall effect measurements. The
281
Mat. Res. Soc. Symp. Proc. Vol. 572 e 1999 Materials Research Society
investigated samples were cut from (OOOl) wafers of SiC crystals which were either vicinal or onaxis. For the Hall effect measurements we used samples ~1 cm2 dimensions. The ohmic contacts
were fabricated in the four corners of the samples (Van der Pauw geometry) by evaporating
Ni/Au contacts (n-type samples) and by subsequent annealing in forming gas at 925° C for 5 min.
The resistivity and Hall constant at a magnetic field of 5-8 Kilogauss were measured in the
temperature range of 10-1000 K.
The photoresponse measurements were made using a Bio-Rad FTS-6000 Fourier Transform
spectrometer with a wavelength range from 2 to 50 microns. The samples was mounted on the
coldfingerof a closed cycle refrigerator allowing the sample temperature to be varied from 10 K
to 290 K. The sample was mounted in series with a load resistor and voltage biased in a range
from 10 to 1000 volts, depending upon signal-to-noise requirements. The spectral resolution was
varied between 4 cm"1 to 1 cm"1 with the higher resolution reserved for samples with initially high
signal-to-noise ratios. For the photoresponse studies, the samples were 8mm x 6mm x 0.5mm
dimensions with two Ni/Au strip contacts covering all but a 1 mm strip down the center of the
sample's top surface. Gold wires were attached to the contacts with indium solder.
6H-SiC RESULTS
Figure 1 shows the measured free-electron carrier concentration (n) versus inverse temperature
for a n-type 6H-SiC sample in the concentration range from 1010 to 1017 cm"3 as a function of the
reciprocal temperature. The free-electron concentration was determined from the experimental
Hall constant RA and given by:
n=\leRH.
(1)
The carrier concentration versus inverse temperature data are fit by the least squares method to
the multi-donor charge balance equation:
n
+ Na= y\
¥*
(2)
Here n is the carrier concentration, Ndi is the concentration of the ith donor level with activation
energy Eai and degeneracy factor gj, Na is the concentration of compensating acceptors, and the
density of states in the conduction band is:
Nc = 2Mc(27tm*kT/h2)3/2
(3)
where Mc is the number of conduction band minima and m* is the electron effective mass. For
6H-SiC Mc = 6 and m* = 0.588. For 4H-SiC, Mc = 3 and m* = 0.390.
The fitting results for both 6H and 4H-SiC samples are given in Table I. This table shows
donor concentration, Hall activation energies, as well as the standard accepted energy levels,
based on IR absorption data, for the nitrogen donor in both SiC polytypes. The fits for the
sample in Fig. 1 indicated only one donor level with an activation energy of 106.5 meV. The
other 6H sample was highly resistive and the Hall data indicated a poorly defined deep level.
282
-—»
101>
Eo
icK:
c
o
*—»
101£r
CO
kCD
O
6H-SiC
1014r
;
c
o
1CK
(D
1CP;
O
.
i—
11
10 F
10'>1C
0
2
4
6 8 10 12 14
1000/T(K"1)
FIG. 1. Measured electron carrier concentration versus reciprocal temperature as obtained form
Hall Effect measurement of n-type 6H-SiC.
35
e
30 r
XI
25
o
3"
O
(D
W
XI
im
o
n
3
10
(D
20
(0
c
o
a
w
0)
15
B)
10
c
3
o
a.
100
150
200
Energy (meV)
250
FIG. 2. Fourier Transform Infrared photoresponse of n-type 6H-SiC samples. Sample 4
(dashed line) is the same as shown in the Hall data. Sample 3 (solid Line) shows response from
the shallower nitrogen level of 6H-SiC
283
TABLE I. SiC Donor Activation Energies
Sample
Nd (atoms/cm3)
Eact (IR Abs.)
Eact(Hall)
Eact (PC)
(meV)
(meV)
(meV)
a
1(4H)
1.2x10"
52
46.8
47.1
7.7xl0'6
92a
87.0
120
2(4H)
1.9x10"
52a
44.6
40.9
1.8x10"
92a
83.8
120
3(6H)
81"
—
77
-138,142"
120
4(6H)
8.4x10"
81"
138,142"
106.5
120
a) W. Götz, A. Schoner, G. Pensl, W. Suttrop, W. J. Choyke, R. Stein and S. Leibenzeder,
J. Appl. Phys. 73, 3332 (1993).
b) W. Suttrop, G. Pensl, W. J. Choyke, R. Stein and S. Leibenzeder, J. Appl. Phys. 72, 3708 (1992).
Figure 2 shows two FTIR photoresponse spectra of n-type 6H-SiC material. The spectrum
for sample 4, in the wavelength range 50-300 meV, shows a rapid increase in the photoresponse
at 120 meV with no lower energy states observed, probably due to the high compensation level in
this sample. A rapid increase in the photoconductivity occurs when the photon energy is greater
than the energy required to optically excite a donor electron from its ground state to the
conduction band. The energies determined from the photoresponse onset are generally accurate
measures of the optical activation energy of the impurity or defect levels in the material. The 120
meV activation energy is between the value from the Hall fitting and the expected standard value.
The sample 3 spectrum shows two distinct photoresponse onsets at 77 and 120 meV.
The mobilities for each of the above are less than 10 cm2/Vs at low temperature. These low
mobilities required high electric fields to be applied between the contacts and a very narrow
optical area between the contacts. The 6H samples had an applied bias voltages from 45.1 volts
to 1000 volts. To obtain measurable photoresponse, sample 3 had an applied voltage of 45.1 at
9.0K and sample 4 had an applied voltage from 500 volts to 1000 volts at 9.0K.
4H-SiC RESULTS
Hall effect results for one of the 4H samples is shown in Fig. 3. The data for both samples
were fit to eq. 2 with the 4H parameters. Both samples had two donors levels near the accepted
values for the hexagonal and cubic sites of nitrogen and the results are given in Table I. The 4HSiC samples had lower concentration levels approximately 10" cm"3 and higher mobilities on the
order of 150 cm2/Vs at low temperature (-30 K). For both 4H-SiC crystals a maximum value of
1000 cm2/Vs at T=100 K was obtained.
The low temperature photoconductivity spectra for two n-type 4H SiC samples are
compared in Fig. 4. The two spectra are nearly identical in shape throughout the mid-infrared.
Both spectra show strong increases in the photoresponse around 40 meV and at 120 meV. The
two samples show two slightly different energies for the onset of the shallow nitrogen level, 40
and 47 meV. As previously reported in DLTS studies, the activation energy of nitrogen at the
284
"E
c
g
2
c
(D
Ü
C
o
O
(0
Ü
0 2 4 6 8 10 12 14 16 18 2022
1000/T(K-1)
FIG. 3. Measured electron carrier concentration versus inverse temperature as obtained from the
Hall measurement of n-type 4H-SiC sample 1.
1 20
1
i
'
'
47.1,1
B
S
e
o
a
i
'
'
'
'
i
'
'
'
'
i
'
'
120
'
'
i
'
'
'
200
4H-S1C
8 0
1 50
6 0
1 00
^
o
o
>1
n
T3
O
P
V
4 0
J=
2 0
o
o
'
1 00
s-
<U
'
h/*^i
35 0
B
B
PH
50
100
150 200
Energy (meV)
250
0
300
FIG. 4. Photoresponse spectra of n-type 4H-SiC. Sample 1 (solid line) is the same as shown in
the Hall data. Sample 2 (dashed line) has a higher donor sample concentration than sample 1.
285
hexagonal site decreases with donor concentration.5 The 120 meV level is higher in energy than
expected. As shown in Table I, the theoretical energy for nitrogen on a cubic site is 92 meV.
The other influence on the SiC photoresponse spectrum is the strong absorption bands due to
the excitation of lattice vibrational modes. This process generates no electrons and hence no
additional photoresponse signal.
Instead, these phonon modes compete with any
photoexcitation processes for the available photons at selected wavelengths and create drops in
the background photoresponse. The influence of the SiC lattice absorption is clearly seen in the
4H spectra in the region from 95 mev to 220 meV.
CONCLUSIONS
Another powerful optical technique for examining impurity states in SiC has been
demonstrated. The mid-infrared photoconductivity spectra revealed the optically active
impurity states in n-type 6H and 4H-SiC. These photoionized impurity transitions are most
likely related to the nitrogen states known to be common in these materials. The photoresponse
spectra from 6H and 4H-SiC are observed to be distinctly different in many respects. However,
all the spectra show very strong photoresponse onsets at 120 meV. In principle, the 6H and 4H
polytypes should have different activation energies for substitutional nitrogen on the cubic lattice
sites.
If only the 6H results are considered, the agreement between the energies determined from
photoresponse, 77 and 120 meV, are in good agreement with Hall results in the literature, 85 and
125 meV. When the 4H spectra are considered however, the assignment of the 120 meV onset to
an optical activation energy may be in doubt. In 4H the nitrogen energy for the cubic sites should
be closer to 100 meV. One factor maybe the strong optical phonons known to occur between
98.7 meV (E, TO) and 119.9 mev (A! LO). Further work will be done on thinned SiC samples to
reduce the intensity of these lattice absorptions to see if any photoresponse is revealed below the
120 meV limit.
The agreement for the activation energies of substitutional nitrogen on the hexagonal sites in
both polytypes determined from the photoresponse was excellent. The 4H-SiC spectra revealed
a shallow nitrogen level at 47 meV for Nd= 1.2E17 cm"3 and 41 meV for Nd=1.9E17 cm"3. This
shift with donor concentration has been observed by other researchers. There are sharper
features in the photoresponse spectra of 4H-SiC related to nitrogen donor excited states that will
be published at a later date.
REFERENCES
1. W. J. Choyke and G. Pensl, Physica B 47, 212 (1991).
2. H. H. Woodbury and G. W. Ludwig, Phys. Rev. 124, 1083 (1961).
3. W. C. Mitchel, G. J. Brown, L. S. Rea and S. Smith, J.Appl. Phys. 71, 246 (1992).
4. J. J. Rome, R. J. Spry, T. C. Chandler, G. J. Brown, R. J. Harris and B. C. Covington, Phys.
Rev. B25, 3615-3618 (1982).
5. A. O. Evwaraye, S. R. Smith and W. C. Mitchel, Mat. Res. Soc. Symp. Proc. On Defect and
Impurity Engineered Semiconductors II. 510,187 (1998).
286
Part IV
GaN Growth and
Characterization
The influence of the sapphire substrate on the
temperature dependence of the GaN bandgap
Joachim Krüger, Noad Shapiro, Sudhir Subramanya, Yihwan Kim, Henrik Siegle,
Piotr Perlin, and Eicke R. Weber
Department of Materials Science, University of California at Berkeley
and
Materials Science Division, Lawrence Berkeley National Laboratory,
Berkeley, California 94720, USA
William S. Wong and Timothy Sands
Department of Materials Science, University of California at Berkeley
Berkeley, California 94720, USA
Nathan W. Cheung
Department of Electrical Engineering and Computer Science, University of California,
Berkeley, California 94720, USA
Richard J. Molnar
Lincoln Laboratory, Massachusetts Institute of Technology,
Lexington, Massachusetts 02173, USA
ABSTRACT
This paper analyses the influence of the sapphire substrate on stress in GaN
epilayers in the temperature range between 4K and 600K. Removal of the substrate by
a laser assisted liftoff technique allows, for the first time, to distinguish between stress
and other material specific temperature dependencies. In contrast to the prevailing
assumption in the literature, that the difference in the thermal expansion coefficients is
the main cause for stress it is found that the substrate has a rather small influence in
the examined temperature range. The measured temperature dependence of stress is
in contradiction to the published values for the thermal expansion coefficients for
sapphire and GaN.
INTRODUCTION
In recent years, GaN and related compounds have attracted a lot of academic as
well as commercial interest. This is due to the potential applications for UV-based optoelectronic applications as well as high-temperature electronics [Kahl]. Very bright blue
and green InGaN single quantum well diodes light-emitting diodes have been developed
and commercialized [Nakl], and a laser diode consisting of 4 InGaN multi quantum
wells has been reported to have a room temperature cw-operational lifetime of more
than 10.000 hrs [Nak2].
Since large-scale GaN substrates are not available, epitaxial layers of GaN are
deposited for the most part on foreign substrate materials like sapphire and SiC. These
materials are known to result in stress in the GaN main layer which can reach values of
up to 1.2 GPa [Krul], either compressive or tensile. It is commonly argued that the
lattice mismatch between layer and substrate and the difference in thermal expansion
coefficient (TEC) are the main causes of stress in the GaN layer at room temperature.
Consequently, considerable efforts have been spent on the exploration of alternative
substrate materials. These have been either focused on matching the lattice constant
289
Mat. Res. Soc. Symp. Proc. Vol. 572 ® 1999 Materials Research Society
(Si, GaAs, ZnO, ÜAIO3) or matching the thermal expansion coefficient (Ge [Siel]). As
to date, however, best results for GaN are still achieved for growth on sapphire.
Though methodologies have been empirically developed to control stress to some extent,
very little fundamental understanding of the exact causes of stress has been gained.
Based upon the linear thermal expansion coefficients for GaN (a = 5.59 x 10'6/K),
for sapphire (cc= 7.5 x lO^/K), and for SiC (a = 4.2 x 10^/K), all values are measured at
room temperature [Lanl], GaN crystals grown on sapphire should be compressively
stressed, whereas crystals grown on SiC should be under tensile stress. This statement
assumes that stress caused by the mismatch of the lattice constants is completely
relaxed. For most reported characterization studies, this assumption certainly holds,
though it has been recently shown that an adequate buffer layer design allows to grow
GaN in tension on sapphire [Krul] and compressively stressed GaN on SiC [Edwl]. Up
to this point, however, no publication is able to quantitatively explain the influence of
the substrate on the amount of stress in the GaN main layer. Due to experimental
constraints, it is unfortunately difficult to determine stress in-situ during growth and
subsequent cooling down. Therefore, in most cases, post-growth attempts to monitor
the temperature dependence of stress
are limited to the range between 4 K and
- : ' ' \ll rv';Yj{^;4^;^^fe%ös:^^i 800 K. Extrapolation of stress data taken
by Raman spectroscopy between 300 K
and 750 K [Agel] reveal a discrepancy
between the expected absolute amount of
mt/UBäj^r. - ^Ä^V
stress and actual measured stress values
for MOCVD and MBE grown samples.
fcfc»j^^.-i*-,^-.l*:f.:i*tif.
This effect was explained by the
tajHy|
assumption of onset of plastic flow
releasing stress during cool-down to an
unknown amount, [Kiel].
On the
contrary, S.Hearne et.al. [Heal] have
recently shown by in-situ stress
n
measurements that GaN tends to grow
Figure 1: SEM image of a free-standing GaN under tensiie stress and that plastic
membrane lifted offfrom sapphire.
relaxation during cool-down is negligible.
Stress can be most conveniently determined by photoluminescence at cryogenic
temperatures. It is known to shift the PL transition of the donor bound exciton (DX)
spectrum by 27meV/GPa with the stress-free position located at 3.467 eV, [Kiel].
Besides being subject to built-in stress at a given temperature, near band gap transitions
are also temperature dependent, as they follow the temperature dependence of the
bandgap. For the last 25 years, numerous publications have attempted to determine and
explain the bandgap temperature dependence of hetero-epitaxially grown GaN. The
huge differences in the observed dependencies have been ascribed to various material
and growth parameters, such as the growth method, post-growth cooling down rate,
sample thickness, to name a few. None of these publications, however, is in the position
to sort out one particular parameter and to describe or even explain its influence on the
GaN bandgap temperature dependence.
This study now takes an entirely new approach to the described problems.
Removal of the sapphire substrate via a laser-assisted liftoff technique [Wonl] provides
a reference sample which allows studying exclusively the temperature dependence of
stress. For the first time, it is therefore possible to distinguish between intrinsic and
material specific effects of the band gap temperature dependence on the one hand and
stress effects caused by the thermal mismatch between sapphire substrate and GaN
layer on the other hand. To prove the general applicability of the described findings,
HVPE and MOCVD grown samples are analyzed.
290
EXPERIMENTAL
AU GaN samples used in this study were grown on sapphire. The samples grown
by Metal-Organic Chemical Vapor Deposition (MOCVD) were supplied by Cree
Research and represent commercially
stress free
available state-of-the-art material.
The n-type carrier density was
specified as in the low lO^cm"3; the
sample thickness was 2 urn.
The
lifted off
sapphire
Hydride Vapor Phase Epitaxy (HVPE)
grown samples were about 25 |im
thick. Growth details are described in
[Moll].
For each set of samples, a
reference sample was produced by
removing its sapphire substrate by a
recently developed laser-assisted liftoff technique [Wonl]. This method is
3.450 3.455 3.460 3.465 3.470 3.475 3.480 3.485 3.490
known to preserve the optical and
Energy [eV]
structural quality of the GaN film, see
Figure 2: Low-temperature PL spectra of a free- figure 2.
standing and a GaN film still attached to its
Photoluminescence was excited
sapphire substrate.
by a 50mW HeCd laser, diffracted by a
0.85m double grating monochromator
and detected by an UV-sensitive photo-multiplier. Special attention was spent to keep
the excitation density constant for all samples. Further, with neutral density filters the
excitation density was sufficiently reduced to exclude any warming effects. For
temperatures between 15 K and 320 K a closed cycle refrigerator was used,
temperatures between 300 K and 600 K were achieved by using a hot plate. Sample
mounting with vacuum grease warranted a stress-free sample fixture and sufficient
thermal contact with the copper holder.
RESULTS
Comparison between the lowtemperature PL spectra of the MOCVD
140
grown GaN layer still attached to its
substrate and the free-standing layer shows
a red shift of the donor bound exciton
transition, figure 2.
The energetic
difference indicates a stress release of 0.4
GPa, according to the calibration provided
by [Kiel]. In particular, it is found that the
stress gradient across the cross section of
the HVPE grown layer which can be as
much as 0.8 GPa, [Sie2], has been
completely relaxed, too.
The energetic
positions of the PL spectra excited from the
top and the former interface side are
identical.
The near band gap PL signal at
cryogenic temperatures is comprised of
three transitions: the donor bound exciton
DX = 3.466eV, the free exciton A FX(A) =
0 100 200 300 400 500 600
3.472eV, and the free exciton B FX(B) =
Temperature [K]
3.479eV; energetic positions are given for
Figure 3: Temperature dependence of free the free-standing sample. For the whole
temperature range between 4 K and 600 K
exciton line full width at half maximum.
291
the entire near band gap spectrum could be
3.50 i 111 111 1111 111 1111 i 11 11 11
perfectly simulated by combination of these
lines under the assumption of a Lorentzian
line type. This finding proves that the
?. 3.48
nature of the luminescing transition does
not change over the whole range. In
particular it should be noted that the free
3.46
excitonic transition is traceable for the
whole
temperature
range
and
no
on sapphire
contribution from band-to-band transitions
3.44
lifted off
has been noticed, in agreement with
[Herl]. For most samples, the donor bound
exciton has been visible for temperatures
3.42
up to 150 K. The free exciton B is clearly
visible up to 260 K. Due to the rapidly
increasing line width (fig.4) and small
3.40
energetic distance to the free exciton A, it
can not be resolved anymore for higher
temperatures. As expected, the energetic
i .
11
3.38
spacing between these transitions are
10
20
30 40
50
60
temperature-independent,
the
found
T2/(T + ß)
energetic differences are: (FX(A) - DX =
Figure 4: Varshni plot of the energetic position
5.5meV, FX(B) - FX(A) = 6.5 meV)
of the free exciton A for both the free-standing
and the GaN film on sapphire.
0 + YphT + exp[h /k T| - 1
r = rr
VLO
B
The temperature dependence of the line width can be broken down into three
major regimes, figure 3. At cryogenic temperatures, the line width is dominated by
interface roughness, the exciton-exciton interaction, and exciton scattering by
impurities, [Visl]. The intermediate region is given by the coupling strength of the
interaction between excitons and acoustical phonons. For temperatures above 240K,
the line width is entirely dominated by the exciton interaction with longitudinal optical
phonons. We find the following parameters: To = 13 meV, Yph = 31±2 ueV, and TLO =
500±25 neV. These values are in agreement with a previous publication [Visl].
Figure 4 shows the temperature dependence of the free exciton transition
energy FX(A) for both the layer on
Experimental
a
sapphire and its free-standing reference
P
Reference
[l(r*
eV/Kl
[K]
technique
sample. It should be noted, that the GaN
Pfaotohuninescettce of
band gap energy is by 27 meV [Buyl]
[Visl]
400
5.0
EX (A)
larger than the free exciton energy; the
"
FX(B)
450
5.2
difference accounts for the free exciton
binding energy. Since this paper is only
[Monl]
PL excitation
-996
-5.08
concerned about relative changes in the
Photoabsorption
"
772
9.39
band gap energy, the given values are not
PhotorcflectioD of
corrected for this energy.
[Shall
835.6
8.32
FX(A)
For most
semiconductors
the
"
FX(B)
1194
variation of the band gap with temperature
10.9
can be described by the semi-empirically
Absorption of
[Manl]
737.9
5.66
MOCVD grown GaN
found Varshni formula [Varl]:
Absorption of
MBE grown GaN
PLofFX(A)
MOCVD, on sapphire
PLofFX(A)
MOCVD, lifted off
11.56
1187.4
"
13.25
1539
this study
13.56
1570
"
E(T) = ECr=0)-£lr
For both the free-standing and the
layer still attached to its sapphire
substrate, we find the same parameters
Table 1: Comparison of Varshi parameters within the experimental error, table 1.
determined for various GaN materials.
This is very remarkable, since the values
292
reported in the literature scatter considerably; for instance, the parameter ß which is
proportional to the specific heat Debye temperature [Manl] has been found to vary
between ß = -996 K [Monl] and ß = 3690 K [Shal].
This clearly proves that the
parameters a and ß are strongly sample
dependent and comparison of results taken
from different authors on different
samples is less meaningful. We argue that
this is due to the differing amount of point
and extended defects which certainly
influences the Debye temperature. As the
results of this study prove, stress caused by
thermal mismatch with the substrate plays
only a minor role at this scale.
It is more illustrative to analyze the
temperature dependence of the difference
in the FX(A) position between the sample
on sapphire and its respective freestanding reference sample, figure 5. This
should follow the temperature dependence
of stress in the GaN main layer.
Apparently, with decreasing temperature
also the difference decreases indicating
100 200 300 400 500 600
that the GaN layer is less compressively
Temperature [K]
stressed. This finding is surprising and in
Figure 5: Difference in energetic position of the contradiction
to
the
prevailing
freeexcitonA between GaN film on sapphire and understanding in the literature. Since the
the free-standing reference sample.
TEC of sapphire is higher than the TEC of
GaN, the GaN main layer should be increasingly compressively stressed with
decreasing temperature - provided that the TECs are temperature-independent. Consequently, the energetic difference between the PL spectrum of the free-standing layer
and the layer still attached to the substrate should increase. Strikingly, the opposite
tendency is experimentally observed.
DISCUSSION
The unexpected temperature behavior of stress in GaN hetero-epitaxially grown
on sapphire is the most interesting aspect of this study. Our results clearly indicate that
the influence of the thermal mismatch between sapphire and GaN epilayer on the GaN
band gap temperature dependence between 4 and 600 K is much less than generally
assumed in the literature. Monemar et.al. [Mon2], for instance, have found that the PL
spectra of GaN grown on sapphire and on SiC differ in their energetic positions by more
than 30meV when cooled down to helium temperatures. If this difference was solely
due to stress, the two different substrates would result in a difference of thermal stress
of more than 1 GPa!
Assuming a (temperature-independent) Young modulus of E = 200 GPa [Kiel]
and a Poisson ratio of v = 0.27 [Kiel], one estimates for a change of temperature from
300K to 4 K the stress for GaN on sapphire to be a = -0.16 GPa. This difference should
result in a blue shift of the PL spectrum by 4.3 meV. This should be clearly visible but is
experimentally not observed.
Two possible explanations for the findings of this study should be discussed:
First, the variation of stress in GaN grown on sapphire in the respective temperature
range is not governed by the thermal mismatch between GaN and sapphire. The second
and more likely cause is given by the uncertainty of published thermal expansion
coefficients for both sapphire and GaN, [Lanl, Lesl]. The reported values differ by a
factor of two, even when only values taken at room temperature are compared. The
TEC temperature dependence is even less known so that one may speculate that an
293
unanticipated temperature behavior of these parameters might be responsible for the
here observed stress dependence.
On the other hand it should be noted that stress is usually determined by
photoluminescence at cryogenic temperatures, whereas strain is almost always assessed
by X-ray diffraction at room temperature [Kiel]. Our results prove that the difference
in measurement temperature does not translate into an inconsistency between the two
methods, the maximum deviation is about 0.1 GPa which falls into the range of
experimental error of most methods.
The results of this work, however, strongly suggest to measure the temperature
dependence of the a and c lattice parameter by X-ray diffraction. Preliminary studies
are already available [Heil, Lesl], but their results are not conclusive yet.
Measurements of a lifted-off GaN film will provide the best basis for a solid data
collection.
SUMMARY
This paper analyses, for the first time, the temperature dependence of stress in
MOCVD and HVPE grown GaN films on sapphire in respect to a free-standing
reference GaN film which has been lifted off from its substrate. The measured
temperature dependence of stress in the temperature range between 4 and 600 K is in
contradiction to the published values for the thermal expansion coefficients for sapphire
and GaN.
ACKNOWLEDGMENT
This work was supported by the Office of Energy Research, Office of Basic
Energy Sciences, Division of Materials Sciences of the U.S. Department of Energy
under Contract No. DE-AC03-76SF00098.
REFERENCES
[Agel] J.W.Ager, et.al; Mat.Res.Soc.Symp.Proc. 449, 775 (1997)
[Buyl] I.A.Buyanova, et.al; Appl.Phys.Lett. 69,1255 (1996)
[Codl] G.D.Cody; in Hydrogenated Amorphous Silicon; ed. By J.Pankove,
Semiconductors and Semimetals Vol. 21, Part b, Ch. 2, p. 11 (Academic, New
York 1984)
[Edwl] N.V.Edwards, et.al; Appl.Phys.Lett. 73,2808 (1998)
[Heil] H. Heinke, et.al; J. of Crystal Growth 189/190, 375 (1998)
[Herl] HHerr, V.Alex, and J.Weber; Mater. Res. Soc. Symp. Proc. 482, 719 (1998).
[Kiel] C. Kisielowski in Gallium Nitride II; Semiconductors and Semimetals Vol. 57,
Ch. 7, p. 275(Academic, New York 1999)
[Krul] J.Krüger et.al; to be published
[Lesl] M. Leszczynski, et.al; JAppl.Phys. 76,4909 (1994)
[Heal] S. Hearne, et.al; Appl.Phys.Lett. 74, 356 (1999)
[Lanl] Landolt-Bornstein: Numerical Data and Functional Relationships in Science
and Technology, Springer, Berlin, 1982, Vol. 17b.
[Manl] M.O.Manasreh; Phys.Rev.B. 53,16425 (1996)
[Moll] R.J. Molnar, et.al; J.Crystal Growth 178,147 (1997)
[Monl] B.Monemar; Phys.Rev.B. 10, 676 (1974)
[Mon2] B.Monemar, et.al; MRS Internet J. Nitride Semicond. Res. 1, 2(1996)
[Siel] HSiegle, et.al; MRS '99 Spring meeting; these proceedings
[Sie2] H. Siegle, et.al; Appl.Phys.Lett. 71,2490 (1997)
[Shal] W.Shan, et.al; Appl.Phys.Lett. 66, 985 (1995)
[Varl] YP.Varshni, Physica 34,149 (1967)
[Visl] A.K.Viswanath, et.al; JAppl.Phys. 84, 3848 (1998)
[Wonl] W. S. Wong, et.al; Appl.Phys.Lett. 72, 599 (1998)
294
EFFECT OF N/Ga FLUX RATIO IN GaN BUFFER LAYER GROWTH BY MBE ON (0001)
SAPPHIRE ON DEFECT FORMATION IN THE GaN MAIN LAYER
S. Ruvimov,* Z. Liliental-Weber, J. Washburn,
Lawrence Berkeley National Laboratory, MS 63-203, Berkeley, CA 94720,
* present address: Mitsubishi Silicon America, 1351 Tandem Ay., Salem, Oregon, 97303
Y. Kim, G. S. Sudhir, J. Krueger, and E. R. Weber
Department of Materials Science and Mineral Engineering, University of California at Berkeley, Berkeley,
California, 94720
Transmission electron microscopy was employed to study the effect of N/Ga flux ratio in the
growth of GaN buffer layers on the structure of GaN epitaxial layers grown by molecular-beamepitaxy (MBE) on sapphire. The dislocation density in GaN layers was found to increase from
lx1010to6x10° cm'2 with increase of the nitrogen flux from 5 to 35 seem during the growth of the
GaN buffer layer with otherwise the same growth conditions. All GaN layers were found to contain
inversion domain boundaries (IDBs) originated at the interface with sapphire and propagated up to
the layer surface. Formation of IDBs was often associated with specific defects at the interface with
the substrate. Dislocation generation and annihilation were shown to be mainly growth-related
processes and, hence, can be controlled by the growth conditions, especially during the first growth
stages. The decrease of electron Hall mobility and the simultaneous increase of the intensity of
"green" luminescence with increasing dislocation density suggest that dislocation-related deep
levels are created in the bandgap.
1. INTRODUCTION.
Epitaxial GaN is a promising material for electronic applications such as visible light-emitting
diodes (LEDs) [1-4], blue lasers [3,4] and metal-semiconductor field-effect transistors [5]. Device
quality GaN on sapphire has been successfully achieved by metal-organic vapor phase epitaxy using
low temperature (LT) GaN or A1N buffer layers [6]. The LT buffer layer provides a high density of
nuclei for growth of the main GaN layer at high temperature and promotes lateral growth of GaN
[6]. The growth and subsequent coalescence of GaN islands finally leads to quasi-two-dimensional
(2D) growth [7]. The structural quality of the GaN layer appears to depend on growth evolution.
Because dislocations in the GaN have a low mobility [8] there is a low probability for their
interaction and annihilation compared to other III-V materials such as GaAs. The defect density in
GaN layers can be significantly reduced by optimization of growth conditions, especially at initial
growth stages. Thus, the structure of a LT buffer may significantly effect the dislocation density in
the GaN epilayer.
The LT buffer layer was found to transform during the temperature ramp/anneal with an
increase of its average grain size and root-mean-square (rms) roughness [9-11]. The growth
evolution can be affected by a number of parameters including the thickness of the LT buffer, the
temperatures for LT and HT growth, growth rate, Ga/N flux ratio, gas ambient, etc [12-19].
However, the effect of the Ga/N flux ratio in the growth of the LT buffer layer has not yet been
studied in detail. Here we report the results of a TEM study of the effect of the N/Ga flux ratio
during growth of GaN buffer layers on defect formation in epitaxial GaN layers grown by MBE on
(0001) sapphire.
295
Mat. Res. Soc. Symp. Proc. Vol. 572 c 1999 Materials Research Society
2. EXPERIMENT.
GaN epitaxial layers were grown by molecular beam epitaxy (MBE) on (0001) sapphire
substrates using a Riber 1000 system. Growth details are described elsewhere [20]. After cleaning
and nitridation of the sapphire substrate (700 °C, 10 min), a LT GaN buffer layer was deposited
(500 °C, 5 min) followed by the growth of a 2 um-thick GaN epilayer (725 °C, 4 h). A set of 4 GaN
layers were grown under various nitrogen flow rates (5, 15, 25 and 35 seem) during the buffer
deposition with a fixed N flux rate of 35 seem for the main layer. The Ga cell temperature was kept
constant at 880 °C during the growth. TEM studies were carried out on Topcon 002B and JEOL
200CX microscopes operated at 200 kV and on an ARM microscope operated at 800 kV. Crosssectional specimens were prepared for TEM study by dimpling followed by ion milling.
Atomic force microscopy (AFM) was employed to determine the surface morphology of GaN
layers as well as LT buffer layers with and without annealing at high temperature. Electrical
properties of GaN layers were evaluated by Hall effect measurements at room temperature. X-ray
diffraction was performed on a Siemens D-5000 diffractrometer to determine the crystalline quality
of GaN layers using both symmetric and asymmetric Bragg reflection.
3. RESULTS AND DISCUSSION.
TEM indicated that all layers contained a high density of threading dislocations and inversion
domain boundaries (IDBs). Growth parameters and structural characteristics of the GaN layers are
shown in Table 1.
Table 1. Growth and structural parameters of GaN layers.
Sample
Nflux
(buffer),
seem
Buffer
Nflux
(epi-layer), roughness,
nm
seem
A
B
C
D
5
15
25
35
35
35
35
35
Dislocation
density,
cm"2
l^lO1"
1.1
1.1
o.s-io1"
1.3
1.0X10IU
5.6"10'u
_
FWHM of
symmetric
(0002)
reflection,
arcmin
2.3
3.5
7.0
8.5
FWHM of
asymmetric
(0002)
reflection,
arcmin
16
15
17
34
Figure 1 shows two plan view TEM images of the same area taken under different diffraction
conditions where either dislocations (a) or IBDs (b) were visualized. Dislocations were often lying
within IDBs which form closed domains of 15-30 nm in diameter [Fig. 1 (b)].
Threading dislocations were usually arranged in small angle boundaries with both tilt and
twist components dividing the epitaxial layer into columnar grains. Diffraction analysis shows that a
majority of the threading dislocations are edge dislocations with b=a/3<ll 20> lying in the c-plane.
Therefore, grains were mainly misoriented around the c-axis being almost perfectly oriented in the
c-plane. As a result, the full width at half maximum (FWHM) for a symmetric (0002) x-ray rocking
curve from the GaN layer is typically much smaller than that of an asymmetric x-ray rocking curve
(see Table 1).
Many dislocations and all IDBs originated at the interface with the sapphire and propagated up
to the layer surface. Some dislocations form half loops near the interface and don't propagate
further into the GaN layer.
296
Fig. 1, a-b. Plan-view TEM images of GaN layer (the same area) taken at different g.
The GaN/sapphire interface is almost atomically abrupt, but contains steps of about 1-2
monolayer height. Formation of IDBs was often associated with specific defects at the interface
with a substrate (Fig.2). Fig 2 (b) shows an IDB associated with an atomic step at the interface. The
white contrast at the interface might indicate the presence of a thin A1N layer between the GaN and
sapphire. A one to two monolayer thick A1N layer is expected to form during the nitridation of the
sapphire. This A1N layer may reduce the possible deterioration of sapphire and, hence, the interface
roughening. As a result, it leads to a reduction of threading dislocation density and dislocation
rearrangement in both GaN buffer and epi layer.
7nm
1 nm
Fig. 2. Cross-sectional TEM (a) and HREM (b) images of GaN epitaxial layer
A white contrast often appears under the IDBs (see e.g. Fig. 2.) suggesting that A1N might
locally change the polarity and give rise for the formation of IDBs. Therefore, IDBs are expected to
originate at the earliest growth stage, during the buffer layer deposition. IDBs were observed in the
GaN buffer layer before and after the high temperature ramp/anneal. Fig. 3 shows a typical HREM
image of a LT GaN buffer layer after the high temperature anneal to 725 °C. The buffer layer has a
hexagonal structure and contains a high density of IDBs originating at the interface. Before
annealing the buffer layer often contained a fraction of the cubic phase (not shown) that transforms
into the hexagonal structure during the high temperature ramp/anneal. LT GaN in Fig. 3 has
relatively rough surface and smaller grains compared to the AFM data in Table 1. The small grains
(< 50 nm in diameter) were not visible on AFM images because they were buried under a thin
amorphous-like layer (not shown). The nature of this layer is unknown.
297
Fig. 3. Cross-sectional HREM image of LT GaN buffer layer after thermal anneal at 725 °C.
Initial growth of the GaN layer proceeds in three dimensional (3D) fashion due to a high
mismatch in lattice parameters between GaN and sapphire. The GaN/sapphire interface contains a
high density of misfit dislocations which accommodate almost all the misfit between their
crystalline lattices at growth temperature during the first growth stages. The orientation relationship
between GaN and sapphire, (0001)GaN//(0001)Ai2O3, [liOO]GaN//[H 2OU1203, provides the best
match between their crystalline lattices, but mismatch still remains high. The orientation
relationship results in a 6 to 7 "magic" ratio between crystalline lattices of GaN and sapphire:
every 6 planes ([flOO] or [II2O]) of GaN fits to 7 planes ([11 20] or [1100], respectively) of
sapphire with an error of about 2 % (Fig. 4). The period of 1.7 nm for the contrast oscillations
observed at the GaN/sapphire interface fits well to this "magic" value. The atomic structure of this
GaN/sapphire interface suggests formation of a high density of small coherent hexagonal nucleai as
the first stage of the growth and, then, their coalesce into larger incoherent GaN islands with misfit
dislocations at the interface. Threading dislocations in GaN appear at points where the misfit
dislocation network is disturbed, for example, by the presence of atomic steps at the interface or at
merging points of adjacent islands.
oooiA1N
• \ • • • • /• *\ • •
V*«. *• • .*
» • • • VA
• **• • *
o# o# o# os o# o. o. o# o. o# o. o# o. o(
I
• "»°• • • ">• *• •*° • *• *• "*°»° «° • '
1_
b
iiooA1,
,, 0001;
1120„
•
• «•«
•
J*
• • • •• •• mm
«•v •* ^
m» K.\m im
•\ •
• • • • •V •
i
7
i
7
k
Fig. 4. Coincidence site lattice (a) for adjacent Ga (or Al)
layers (two-dimensional lattices of Ga (or Al) atoms) in
crystalline lattices of GaN (or A1N) and sapphire, and GaN
(AlN)/sapphire interface (b).
The dislocation density is high in the GaN buffer layer and then drastically decreases over 0.2
urn toward the GaN top surface (Fig.2). This dislocation distribution being typical for many GaN
samples indicates that most of threading dislocations originate and annihilate during the early stages
of growth. Because of low dislocation mobility at the growth temperature [8], the dislocation
distribution in the GaN layer is, to a large extent, frozen after the layer growth and, hence, reflects
the growth process itself. Dislocation generation and annihilation are mainly growth-related
processes and, hence, can be controlled by the growth conditions, especially during the first growth
stages.
298
Formation of dislocation half loops near the interface can be caused by lateral overgrowth of
GaN islands at high temperature. The growth conditions for various grains are locally different
because the islands differ in their initial size and strain distribution within each island. Therefore,
larger islands with lower strain will grow faster than others. As a result, these islands will laterally
overgrow the others leaving them buried near the interface.
Threading dislocations that accommodate the misfit
threading
forced by this lateral overgrowth to bend into basal
dislocations
bended
planes, interact with other dislocations and often
^dislocations
annihilate. Schematically it is shown in Fig. 5. This
process results in the formation of many half loops in
the buffer and in the first 0.2 urn of the GaN and in the
decrease of overall dislocation density in the growing
GaN layer.
Fig. 5. Lateral overgrowth of islands during the early growth
stages.
The efficiency of this process depends on both thickness and structure of the buffer layer, and
on the growth conditions. This suggests that the structure of the buffer layer may effect the defect
formation and annihilation in the GaN epilayer. The roughness of LT buffer layer is highest at N
flux of 35 seem (Table 1) while the average grain size of about 60 nm (measured by AFM) doesn't
change with the N flux. Accordingly the dislocation density in GaN layers also increases up to
6X10 °cm"2 for N flux of 35 seem during the growth of the GaN buffer layer. The increase of the
dislocation density from 1.0x10'° to 6.0xl010 cm"2 corresponds to a decrease in a grain size of the
main GaN layer from 100 nm to 40 nm. This suggests that the lateral overgrowth during the
deposition of the main GaN layer may be affected by a "micro-roughness" of the LT buffer layer.
a
' 8res
tensile
^^T-500
700
T
compressive
5
10
15
20
25
30
35
N2 flow rate of buffer layer growth (seem)
Fig. 6. (a) An increase of the residual stress at room temperature due to partial stress relaxation of the LT buffer
at high temperature, (b) Electron Hall mobility and threading dislocation density of GaN epitaxial layers as a function of
nitrogen flow rate during the buffer layer growth.
The changes in the dislocation density with N flux are similar to those for the FWHM of
asymmetric x-ray rocking curves (see Table 1). This agrees well with the result of diffraction
299
analysis showing that the majority of threading dislocations is of edge type. FWHM of a symmetric
(0002) x-ray rocking curve is more sensitive to misorientations in c-plane so that its increase with N
flux during the buffer layer growth may indicate on the increasing fraction of mixed dislocation in
the GaN layer. Residual compressive stress in the GaN layer decreases with N flux during the buffer
layer growth according to the absorption spectroscopy measurements. A shift in the absorption edge
has been observed in absorption spectra due to the residual stress in the GaN layer. An increase of
the residual stress at room temperature for the layer grown on a "Ga-rich" buffer can be associated
with partial stress relaxation of the LT buffer during the high temperature anneal. Schematically it is
shown in Fig. 6 (a).
The decrease of electron Hall mobility from 85 to 12 cm2/Vs [Fig. 6 (a)] and the simultaneous
increase of the intensity of "green" luminescence (not shown) with increasing dislocation density
suggests that dislocation-related deep levels are created in the band-gap.
In conclusion, a decrease of the dislocation density in GaN layers was observed for "Ga-rich"
LT GaN buffer layers when the nitrogen flux was below 25 seem during the growth of the GaN
buffer layer under otherwise the same growth conditions. All GaN layers were found to contain
IDBs originating at the interface with the sapphire and propagating up to the layer surface.
Dislocation generation and annihilation were shown to be mainly growth-related processes and,
hence, can be controlled by the growth conditions, especially during the early growth stages. The
decrease of electron Hall mobility and the simultaneous increase of the intensity of "green"
luminescence with increasing dislocation density suggest that dislocation-related deep levels were
created in the band-gap.
This study was supported by the Director, Office of Energy Research, U.S. Department of
Energy under Contract No. DE-AC03-76SF00098. The use of the facilities at National Center for
Electron Microscopy and assistance of W. Swider for sample preparation are appreciated.
1. H. Amano, M. Kito, X. Hiramatsu, and I. Akasaki, Jpn. J. Appl. Phys. 28, L2112 (1989)
2. S. Nakamura, T. Mukai, and M. Senoh, Jpn. J. Appl. Phys. 30, L1998 (1991)
3. S. Nakamura, M. Senoh, S. Nagahama, et al, Appl. Phys. 69,4056 (1996)
4. F.A. Ponce, D.P. Bour, Nature 386, 351 (1997)
5. S.N. Mohammad, A. Salvador, and H. Morkoc, Proc. IEEE 83, 1306 (1995)
6. H. Amano, N. Sawasaki, I. Akasaki, and Y. Toyoda, Appl. Phys. Lett. 48, 353 (1986)
7. H. Amano, I. Akasaki, K. Hiramatsu, N. Koide, and N. Sawasaki, Thin Solid Films 163,415 (1988)
8. L. Sugiura, J. Appl. Phys. 81, 1633 (1997)
9. A.E. Wickenden, D.K. Wickenden, and T.J. Kistenmacher, J. Appl. Phys. 75, 5367 (1994)
10. T. George, W.T. Pike, M.A. Khan, J.N. Kuznia, and P. Chang-Chien, J. Electron.Mat. 24,241 (1995)
11. X.H. Wu, D. Kapolnek, E.J. Tarsa, B. Heying, S. Keller, B.P. Keller, U. Mishra, S. P. DenBaars, and J.S.
Speck, AppI.Phys.Lett. 68, 1371 (1996)
12. S. Nakamura, Jpn. J. Appl. Phys. 30, L1705 (1991)
13. S. Ruvimov et a!., unpublished
14. Y. Kim et al Mat.Res. Soc. Symp. 449,227 (1997)
15. J. Han, T.-B. Ng, R.M. Biefeld, M.H. Crawford, and D.M. Follstaed, Appl. Phys. Lett. 71,3114 (1997)
16. D. Kapolnek, X.H. Wu, B. Heying, S. Keller, B.P. Keller, U.K. Mishra, S.P. DenBaars, and J.S. Speck,
Appl. Phys. Lett. 67, 1541 (1995)
17. E.H. Tarsa, B. Heying, X.H. Wu, P. Fini, S.P. DenBaars, J.S. Speck, J. Appl. Phys. 82, 5472 (1997)
18. P. Hacke, G. Feuillet, H. Okamura, and S. Yoshida, Appl. Phys. Lett. 69,2507 (1996)
19. Z. Yu, S.L. Buczkowski, N.C. Giles, T.H. Mayers, and M.R. Richards-Babb, Appl. Phys. Lett. 69, 2731
(1996)
20. Y. Kim et al, unpublished
300
ENHANCED OPTICAL EMISSION FROM GaN FILM GROWN ON
COMPOSITE INTERMEDIATE LAYERS
Xiong Zhang, Soo-Jin Chua, Peng Li, and Kok-Boon Chong
Center for Optoelectronics, Department of Electrical Engineering
National University of Singapore
10 Kent Ridge Crescent, Singapore 119260, elezx@nus.edu.sg
ABSTRACT
GaN films have been grown on silicon-(OOl) substrate with specially designed composite
intermediate layers consisting of an ultra-thin amorphous silicon layer and a GaN/AlxGai.xN
(x=0.2) multilayered buffer by metal-organic chemical vapor deposition and characterized by
photoluminescence and x-ray diffraction spectroscopy. It was found that the GaN films grown on
the composite intermediate layers gave comparable or slightly stronger optical emission than
those grown on sapphire substrate under identical reactor configuration. Moreover, the full width
at half maximum for the GaN band-edge-related emission is 40 meV at room temperature. This
fact indicates that, by using the proposed composite intermediate layers, the crystalline quality of
GaN-based nitride grown on a silicon substrate can be significantly improved.
INTRODUCTION
Recently there has been successful demonstration of commercially available blue and green
1
2
light-emitting diodes (LEDs) and long life-time laser diodes (LDs) fabricated from group-Ill
nitrides which are generally grown on sapphire substrate. However, besides the large difference
in lattice constant and thermal expansion coefficient between the group-Ill nitride and sapphire
substrate, sapphire is an insulating material and extremely rigid. Therefore, it is not easy to
fabricate a group-Ill nitride-based semiconductor device on a sapphire substrate. Silicon is one of
the proposed substrate materials to overcome this shortcoming because of its high crystal quality,
large area size, low manufacturing cost, and the potential application in integrated optoelectronic devices. However, due to the even larger differences in lattice constant and thermal
expansion coefficient between the group-Ill nitride and silicon as compared with sapphire, it is
really difficult to grow high quality epitaxial layer of group-Ill nitride on a silicon substrate.
Attempts have been made to grow group-Ill nitrides on silicon substrates in the past decade using
various kind of materials as the intermediate layer between group-Ill nitride and silicon substrate.
These include A1N, ' carbonized silicon, ' nitridized GaAs, oxidized AlAs, and y-A^Os.
In particular, by using A1N thin film as the intermediate layer, ultraviolet and violet-light
emitting diodes of group-Ill nitride have been fabricated on a silicon substrate by molecular beam
epitaxy (MBE) recently. However, the turn-on voltages as well as the brightness of these diodes
do not approach the performance levels of the corresponding devices grown on a sapphire
substrate by metal-organic vapor phase deposition (MOCVD). Therefore, both the constitution of
the intermediate layer and the growth method for it need to be further optimized in order to
enhance the crystallinity of the group-Ill nitrides. In this letter, we report the growth of high
quality GaN epitaxial layers on a silicon substrate by MOCVD using the specially-designed
composite intermediate layers (CILs).
301
Mat. Res. Soc. Symp. Proc. Vol. 572 ® 1999 Materials Research Society
EXPERIMENT
GaN epitaxial layers were grown on silicon-(OOl) substrates by low pressure (100+2 Torr)
MOCVD in an EMCORE D125 vertical reactor. Trimethylgallium (TMGa), Trimethylaluminium
(TMA1), and high purity ammonia (NH3) were used as Ga, Al, and N precursors, respectively,
and hydrogen-diluted silane ( S1H4 ) was employed for depositing thin amorphous silicon film.
As shown in Fig. 1, after a chemical cleaning process, the silicon-(OOl) substrate was heated to
1,030 °C under hydrogen ambient for 10 min in order to produce a clean, oxide-free surface. The
silicon substrate was then cooled down to 525 °C, and an ultra-thin (less than 5 nm) amorphous
silicon film was deposited onto the surface of the silicon-(OOl) substrate by flowing the
hydrogen-diluted silane to form a "soft" buffer. Subsequently, a three period GaN/AlxGai_xN
(x=0.2) multilayered buffer (MLB) was grown on the top of the formed ultra-thin amorphous
silicon film to constitute the CILs. The details for the growth of GaN/AlxGai.xN (x=0.2) MLB
has been described elsewhere.10' " Finally, the temperature was raised to 1,000 °C and a 1 umthick unintentionally doped GaN epitaxial layer was grown on the surface of the formed CILs
consisting of the ultra-thin amorphous silicon film and the MLB. No cap layer was deposited on
the top of GaN epitaxial layers. For the purpose of comparison, undoped GaN films have also
been grown over a conventional 25 nm-thick GaN single layer buffer on sapphire substrates.
After the growth, room-temperature photoluminescence (PL) and x-ray diffraction (XRD)
measurements were carried out in order to characterize the crystalline quality of the grown
undoped GaN epitaxial layers. The optical properties, more specifically, the intensity and
linewidth of the PL emission and XRD peaks of the undoped GaN samples grown on silicon
substrates using the newly-developed CILs were compared with those grown on sapphire
substrate using the conventional GaN single layer buffer.
Undoped GaN crystal
AlxGai.xN
5 nm
GaN 3 nm
AlxGai.xN 5 nm
GaN 3 nm
AlxGai.xN 5 nm
GaN/A^Ga^N (x=0.2)
multilayered buffer
Amorphous silicon film
GaN
3 nm
Silicon-(001) substrate
FIG. 1. Schematic sectional view showing an undoped GaN crystal grown over the composite
intermediate layers (CILs) consisting of an ultra-thin (less than 5 nm) amorphous silicon film and
a three period GaN/AlxGai.xN (x=0.2) multilayered buffer on a silicon-(OOl) substrate.
302
RESULTS AND DISCUSSION
As shown in Fig. 2, the PL intensity of the dominant emission peak around 3.4 eV which is
attributed to the band-edge-related transition in wurtzite GaN epitaxial layer grown over the CILs
on a silicon substrate (in solid line), is comparable or even slightly stronger than that for the
undoped GaN sample grown over a conventional GaN single layer buffer on a sapphire substrate
(in dotted line) under identical experimental configuration. The defect-related yellow-band
emission centered at 2.35 eV is much weaker as compared with the dominant GaN band-edge
emission. Although it is impossible for us to compare the PL intensity with the results reported
by other research groups of the GaN samples grown over different intermediate layers on silicon
3 9
substrates, " we can still evaluate some conclusions by focusing on the PL linewidth at room
temperature, which is known to be nearly independent on the measurement conditions, such as
the power density of the excitation laser beam. The full width at half maximum (FWHM) of the
dominant GaN band-edge-related emission peak for the GaN/Si(001) sample shown in Fig. 2, for
example, was measured to be approximately 40 meV. This value is 38 % narrower than the best
result achieved so far for GaN film grown on a silicon substrate, 65 meV which was recently
reported by Osinsky et al. , and quite comparable to that for the low-doped GaN film grown on a
12
sapphire substrate. This fact indicates that the crystal quality of GaN-based nitrides grown on
silicon substrate can be significantly improved by using the CILs technique.
GaN/Si(001)
GaN/Sapphire (0001)
j
c
-5c
0
T=300 K
QL
i
1.5
2.0
.
i
.
2.5
i
3.0
.
i
3.5
.
4.0
Photon Energy (eV)
FIG. 2. The 300 K PL spectra for undoped GaN film grown over the CILs on a silicon substrate
(solid line) and that grown over a conventional GaN buffer on a sapphire substrate (dotted line)
under identical experimental configuration. The full width at half maximum (FWHM) of the
band-edge-related emission peak at 3.4 eV is 40 meV which is 38 % narrower than the narrowest
value reported, 65 meV, for an undoped GaN film grown on a silicon substrate.
303
1
i
i
i
i
i
GaN/Si(100)
GaN (0002)
'c
3
-5'w
c
0)
Si(004)
I GaN (0004)
GaN (1011)
Q
X
,_J
20
30
|
...
i
i
i
40
50
60
.*JL-..J<..
70
i
80
26 (degrees)
FIG. 3. X-ray diffraction profile and its rocking curve of the (0002) reflection for an undoped
GaN film grown over the CILs on Si(001) substrate. The dominant diffraction peak at 34.6 arcdegrees with a FWHM of 40 arc-minutes is attributed to the (0002) diffraction of the wurtzite
GaN, and is much stronger than the signal from the silicon substrate.
On the other hand, the XRD measurement result is illustrated in Fig. 3, showing a x-ray
diffraction profile and its rocking curve of the (0002) reflection for an undoped GaN epitaxial
layer grown over the CILs on a Si(001) substrate. Compared with the diffraction peak from the
silicon substrate at 69.3 arc-degrees, Si(004), the dominant diffraction peak at 34.6 arc-degrees is
much more intense, and identified as the (0002) diffraction from the wurtzite GaN crystal. The
weak diffraction peak near 73 arc-degrees is attributed to the (0004) diffraction of the wurtzite
GaN, and no diffraction peak from the zinc-blende (cubic) GaN is observed. The FWHM for the
dominant GaN (0002) diffraction peak was measured to be as narrow as 11 arc-minutes which is
much better than the corresponding value (108 arc-minutes) recently reported by Wang et al.
who employed a thin Y-A1203 film as the intermediate layer. These facts reveal once again that
the crystal quality of GaN-based nitrides can be greatly improved by using the proposed CILs.
It is clear that as the temperature is raised from a relatively low temperature of 525 °C to a
high temperature of 1,000 °C, both the ultra-thin amorphous silicon film and the amorphous or
poly-crystalline GaN/AlxGai.xN MLB formed at 525 °C will completely or partially change to
mono-crystalline or poly-crystalline due to the recrystallization. In fact, they serve as the seed
crystals for the subsequent growth of the GaN-based nitrides. Compared with the conventional
low-temperature-grown A1N intermediate layer, the CILs in this study turned out to be able to
304
accommodate the strain arising from the lattice mismatch between the group-Ill nitrides and the
silicon substrate, and to form the seed crystals more effectively. In other words, because the
strain-accommodating and recrystallizing effects are of crucial importance in improving the
crystal quality of the group-Ill nitrides, and they seem to be more profound in the CILs than in
the conventional intermediate layers, the crystal quality of the group-Ill nitrides has been
significantly improved by utilizing the CILs, as confirmed by the strong and sharp GaN-related
PL and XRD signals described above.
Note that the optimal values in total layer thickness and solid composition for the CILs
apparently depend on the selection of the constituent materials in the CILS as well as the
subsequently grown group-Ill nitrides. At the present time, however, since the physical origin of
the CILs is unfortunately not very clear, it is truly difficult to theoretically determine or predict
the optimal layer thickness of CILs for a specific material combination. In other words, the
optimal value for a specific material combination can now only be determined by experiment.
However, the existence of the optimal layer thickness for the CILs can be interpreted
qualitatively as follows. Generally an intermediate layer grown at a low temperature provides
seed crystals which act as nucleation sites with low orientational fluctuation to promote the
lateral growth of the group-Ill nitrides. The CILs in this study consisting of an ultra-thin
amorphous silicon film and a MLB provide more seed crystals and interfaces than a conventional
single intermediate or buffer layer to accommodate the strain and to terminate the misfit
dislocations. However, if the CILs are too thin, they may neither provide sufficient amount of
seed crystals necessary for the subsequent growth of the group-Ill nitrides nor effectively
accommodate the strain and terminate the misfit dislocations. On the other hand, if the CILs are
too thick, they tend to bring about excessive amount of the seed crystals with high orientational
fluctuation. Therefore, there should be an optimal value for the total layer thickness of the CILs.
CONCLUSIONS
We have grown high quality GaN films on silicon substrates using a novel CILs technique
by low-pressure MOCVD. The CILs consist of an ultra-thin (<5 nm) amorphous silicon film and
a three-period GaN/AlxGa,.xN (x=0.2) superlattice-like MLB. Based on the detailed study of PL
and XRD spectroscopy, the GaN films grown over CILs on the Si(OOl) substrates were verified
to be nearly pure wurtzite crystals, which gave comparable or even slightly stronger PL emission
than those grown over a conventional GaN single layer buffer on a sapphire substrate under
identical reactor configuration. Moreover the FWHM for the GaN band-edge-related PL peak is
measured to be as narrow as 40 meV at room temperature, which is the narrowest value ever
reported. This fact indicates that by using the proposed CILs technique, the crystalline quality of
GaN-based nitride grown on a silicon substrate can be significantly improved.
REFERENCES
S. Nakamura, M. Senoh, N. Iwasa, and S. Nagahama, Jpn. J. Appl. Phys. 34, L797(1995).
S. Nakamura, M. Senoh, S. Nagahama, N. Iwasa, T. Yamada, T. Matsushita, H. Kiyoku, Y.
Sugimoto, T. Kozaki, H. Umemoto, M. Sano, and K. Chocho, Jpn. J. Appl. Phys 37
L309(1998).
S. Guha and N. Bojarczuk, Appl. Phys. Lett. Vol. 72, 415(1998).
A. Osinsky, S. Gangopadhyay, J. W. Yang, R. Gaska, D. Kuksenkov, H. Temkin, I. K.
Shmagin, Y. C. Chang, J. F. Muth, and R. M. Kolbas, Appl. Phys. Lett. 72, 551(1998).
305
A. J. Steckl, J. Devrajan, C. Tran, and R. A. Stall, Appl. Phys. Lett. 69, 2264(1996).
Y. Hiroyama and M. Tamura, Jpn. J. Appl. Phys. 37, 630(1998).
J. W. Yang, C. J. Sun, Q. Chen, M. Z. Anwar, M. A. Khan, S. A. Nikishin, G. A. Seryogin,
A. V. Osinsky, L. Chernyak, H. Temkin, C. Hu, and S. Mahajan, Appl. Phys. Lett 69
3566(1996).
8
N. P. Kobayashi, J. T. Kobayashi, P. D. Dapkus, W. J. Choi, A. E. Bond, X. Zhang, and
D. H. Rich, Appl. Phys. Lett. 71, 3569 (1997).
L. Wang, X. Liu, Y. Zan, J. Wang, D. Wang, D. Lu, and Z. Wang, Appl. Phys. Lett. 72,
109(1998).
X. Zhang, S. J. Chua, P. Li, K. B. Chong, and W. Wang, Appl. Phys. Lett.. 73
1772(1998).
X. Zhang and S. J. Chua, Singapore Patent, 9801054-9 (1998).
E. F. Schubert, I. D. Goepfert, W. Grieshaber, and J. M. Redwing, Appl. Phys. Lett 71
921(1997).
306
PENDEO-EPITAXIAL GROWTH OF GAN ON SIC AND SILICON SUBSTRATES VIA
METALORGANIC CHEMICAL VAPOR DEPOSITION
K.J. Linthicum*, T. Gehrke*, D. Thomson*, C.Ronning*, E.P. Carlson*, C.A. Zorman**,
M. Mehregany**, and R.F.Davis*
Department of Materials Science and Engineering, North Carolina State University, Raleigh,
NC 27965
»»Department of Electrical, Systems and Computer Engineering and Science, Case Western
Reserve University, Cleveland, OH 44106
ABSTRACT
Pendeo-epitaxial lateral growth (PE) of GaN epilayers on (0001) 6H-silicon carbide and
(111) Si substrates has been achieved. Growth on the latter substrate was accomplished through
the use of a 3C-SiC transition layer. The coalesced PE GaN epilayers were characterized using
scanning electron diffraction, x-ray diffraction and photoluminescence spectroscopy. The
regions of lateral growth exhibited ~0.2° crystallographic tilt relative to the seed layer. The GaN
seed and PE epilayers grown on the 3C-SiC/Si substrates exhibited comparable optical
characteristics to the GaN seed and PE grown on 6H-SiC substrates. The near band-edge
emission of the GaN/3C-SiC/Si seed was 3.450 eV (FWHM ~ 19 meV) and the GaN/6H-SiC
seed was 3.466 eV (FWHM ~ 4 meV).
INTRODUCTION
Prior to the advent of the lateral growth techniques, heteroepitaxial gallium nitride films
typically contained threading defect densities that exceeded 109/cm2. These defects seriously
compromised the properties of subsequently fabricated devices. New growth techniques leading
to low defect-density nitride films were required to achieve commercialization of viable
optoelectronic and microelectronic devices.
The advancement of lateral epitaxial overgrowth (LEO) as a technique capable of
producing GaN epilayers with defect densities reduced to 105/cm2 has recently been
demonstrated by several research groups.',2,3,4 The significance of lateral overgrowth was
immediate as Nichia Chemical reported in 1997 that use of LEO aided in increasing the lifetime
of their blue laser diode from a few hundred hours to over 10,000 hrs,5 and then subsequently
introduced the first commercially available GaN blue laser diode.6 However, to benefit from this
reduction in defects, the placement of devices incorporating LEO technology is confined to
regions on the final GaN device layer that are located over the masked regions and not over the
window regions of the GaN seed layer. Figure 1 shows a scanning electron microscopy (SEM)
cross-sectional image of GaN grown using the LEO technique.
The authors are currently investigating another lateral growth technique for GaN, namely,
pendeo-epitaxy,7'8'9,10,11 as a process route to resolve the aforementioned alignment problem.
Pendeo-epitaxy (from the Latin: pendeo—to hang, or to be suspended) incorporates mechanisms
of growth used by the conventional LEO process by using masks to prevent vertical propagation
of threading defects, and extends the phenomenon to employ the substrate as a pseudo-mask.
Pendeo-epitaxy (PE) differs from conventional LEO in that growth does not initiate through
307
Mat. Res. Soc. Sytnp. Proc. Vol. 572 ® 1999 Materials Research Society
open windows but begins on sidewalls etched into the GaN seed layer. As the lateral growth
from the sidewalls continues, vertical growth of GaN begins and results in the eventual lateral
overgrowth of the masked seed form. Pendeo-epitaxial growth ultimately results in coalescence
over and between seed forms, producing a continuous layer of GaN, as shown in Figure 2.
Figure 2. Cross-sectional SEM
micrograph of PE GaN on SiC.
Figure 1. Cross-sectional SEM
micrograph of LEO GaN on SiC.
Although both LEO and PE research have led to low defect-density GaN material on two-inch
SiC and sapphire substrates, the use of this substrate does not resolve the problem of achieving
low-cost, large area GaN films necessary for commercialization of microelectronic devices.
Meeting this objective has renewed interest in using Si(l 11) substrates as an alternative to SiC
and sapphire. Several recent demonstrations of lateral growth process routes achieving growth
of (OOOl)GaN on (11 l)Si have been reported.8'12'13 In this paper we report on the process steps
for and characterization of PE GaN epilayers grown on both (0001) 6H-SiC and (111) Si
substrates.
EXPERIMENTAL PROCEDURES
GaN Seed Layers
(0001) 6H-SJC on-axis substrates: The initial 500nm thick GaN seed layers were grown
on lOOnm thick high-temperature A1N buffer layers previously deposited on the substrates via
metalorganic vapor phase epitaxy (MOVPE), as detailed in Ref. 14.
(1 inSilicon substrates: 500nm thick (11 l)3C-SiC transition layers were initially grown
on very thin (lll)3C-SiC layers produced by conversion of Si substrate surfaces via reaction
with GjHg entrained in H2. Both the conversion step and SiC film deposition were achieved
using atmospheric pressure chemical vapor deposition (APCVD), as detailed in Ref. 15. The
500nm thick GaN seed layers were subsequently deposited on 100 ran thick A1N buffer layers in
the manner used for the 6H-SiC substrates as noted above.
Pendeo-epitaxial growth of GaN
A 100 nm silicon nitride growth mask was deposited on the seed layers via plasma
enhanced CVD. A 150 nm nickel etch mask was subsequently deposited using e-beam
308
evaporation. Patterning of the nickel mask layer was achieved using standard photolithography
techniques and Ar-plasma sputtering. The final, tailored microstructure consisting of silicon
nitride masked GaN seed forms was fabricated via inductively coupled plasma (ICP) etching of
portions of the silicon nitride growth mask, the GaN seed layer and the A1N buffer layer. The
seed-forms used for this study were raised rectangular stripes oriented along the <1100>
direction. Various seed form widths and separation distances were employed. Pendeo-epitaxial
growth of GaN was achieved within the temperature range of 1050-1100°C and a total pressure
of 45 Torr. The precursors (flow rates) of triethylgallium (26.1 umol/min) and NH3 (1500 seem)
were used in combination with a H2 diluent (3000 seem). Additional experimental details
regarding the pendeo-epitaxial growth of GaN and AlxGai.xN layers employing 6H-SiC
substrates are given in Refs. 7-11.
The morphology and defect microstructures have been investigated using scanning
electron microscopy (SEM) (JEOL 6400 FE), transmission electron microscopy (TEM)
(TOPCON 0002B, 200KV) and X-ray diffraction (XRD) (Philips X'Pert MRD X-ray
diffractometer). Optical characterization was performed using a He-Cd laser (X=325 run).
RESULTS AND DISCUSSION
GaN Seed Layers
The pendeo-epitaxial phenomenon is made possible by using growth and mask
mechanisms similar to conventional LEO techniques and by using the substrate itself as a
pseudo-mask, i.e. the GaN does not nucleate on the exposed SiC surface when higher growth
temperatures are employed to enhance lateral growth. The Ga- and N-containing species more
likely either diffuse along the surface or evaporate (rather than having sufficient time to form
GaN nuclei) from this substrate. Since the newly deposited laterally grown GaN is suspended
above the SiC substrate, there are no defects associated with the mismatches in lattice parameters
between the PE GaN and the SiC substrate.
Pendeo-epitaxial techniques can be applied in general to other substrates as long as they,
or a transition layer deposited on the substrate, act similarly as a/weuc/o-mask. As demonstrated
above, silicon carbide meets this requirement. Therefore, by using a transition layer of 3C-SiC,
silicon can be successfully used as a substrate for PE growth. In addition to acting as a pseudomask, the 3C-SiC performs two other key functions. Firstly, deposition on (11 l)Si results in the
growth of (11 l)3C-SiC. The atomic arrangement of the (111) plane is equivalent to the (0001)
plane of 6H-SiC; thus it facilitates the sequential deposition of a high temperature (0001) 2HA1N buffer layer of sufficient quality needed for the GaN seed layer. Secondly, since the A1N
buffer does not act as a. pseudo-mask, it must also be removed from the areas between the GaN
seed forms to prevent the undesired nucleation of 'defective' GaN in these areas. Without the
presence of an A1N buffer or other transition layer, the silicon substrate is exposed to the growth
environment and consequent effects resulting from the reaction of Si atoms with the Ga and N
species. Therefore the 3C-SiC transition layer acts as a reaction/diffusion barrier between the Si
substrate and the GaN.
Figure 3 shows a SEM micrograph of a 500 nm (111) 3C-SiC transition layer deposited
on an on-axis (lll)Si substrate. A typical high resolution DCXRD scan is shown in Figure 4.
Although the quality of the 3C transition layer is less than optimal, as indicated from the
relatively rough morphology visible in Figure 3, and typical XRD FWHM values of 0.5°, it is of
sufficient quality for the deposition of 2H-A1N and 2H-GaN seed layers.
309
The surface morphology of the GaN seed layers deposited on the (111) 3C-SiC transition
layers were comparable to GaN seed layers deposited on (0001) 6H-SiC substrates, and had very
smooth surfaces. Low temperature (14K) photoluminescence analysis of GaN seed layers grown
on both 6H-SiC and 3C-SiC/Si substrates are shown in Figure 5.
15
15.5
16
16.5
17
17.5
18
XRD Diffraction Angle, CO (degrees)
Figure 4. X-ray rocking curve of
the 111 diffraction of a 0.5 micron
thick 3C-SiC transition layer.
Figure 3. Plane-view SEM
of a 0.5 micron thick 3C-SiC
transition layer on (11 l)Si.
The near band-edge emission was 357.7 nm (3.466 eV, FWHM ~ 4meV) and 359.4 nm
(3.450 eV, FWHM ~ 19 meV), respectively, and has been attributed to an exciton bound to a
neutral donor (X-D°). These results show that the quality of the GaN seed layers deposited on
3C-SiC/Si substrates is approaching the optical quality of GaN grown on 6H-SiC substrates.
Pendeo-Epitaxial Growth:
Figure 2 shows an SEM micrograph of a PE grown GaN using a 6H-SiC substrate with a
1 micron thick GaN seed layer and a lOOnm thick silicon nitride mask. The seed microstructure
was comprised of 3 um wide posts and 1.5 urn wide trenches oriented in the <1100> direction.
Analysis of Figure 2 reveals coalescence between and above the seed structures resulted forming
a continuous epilayer of GaN.
.
450
300
500
3.416V
FWHM 109meV
400
500
600
Wavelength (nm)
Wavelength (nm)
Figure 6. Room temperature
PL spectrum of coalesced PE
GaN on a 3C-SiC/Si substrate.
Figure 5. Low temperature (14 K)
PL spectra of GaN seed layers grown
on 6H-SiC and 3C-SiC/Si substrates.
310
Figure 6 shows the room temperature PL spectrum for a coalesced GaN PE epilayer
grown on a 3C-SiC/Si substrate. A band-edge emission of 363.6 nm (3.41 eV) was observed and
indicates the PE GaN on silicon is under tensile stress.
Figures 7 and 8 show SEM micrographs of PE grown GaN using 3C-SiC/Si substrates.
The seed microstructures were 2 um wide posts and 3 urn wide trenches oriented in the <1100>
direction. The sample shown in Figure 7 employed a 200 nm silicon nitride mask on top of the
seed forms. Analysis of this sample revealed void formation and poor coalescence over the
'thick' seed mask. Void formation in the trench regions is also visible; however this does not
appear to effect the quality of the coalesced GaN above the trench. This is typical of PE GaN
grown under nonoptimized conditions. The sample in Figure 8 is a maskless PE GaN epilayer.
PE GaN grown using this seed microstructure configuration is the equivalent of LEO techniques,
such that vertical propagation of threading defects from the original GaN seed is not prevented.
Analysis of Figure 8 reveals the same void formation at the coalescence point in the trench
regions.
Although the cause of these voids is not yet fully understood, it is believed that
optimization of the initial seed microstructure, including post-trench ratio, seed thickness, etch
quality and mask thickness, result in their elimination. Preliminary results suggest that
minimization of the trench width reduces the void formation in the trench region. Also,
minimizing the seed mask thickness helps to eliminate void formation in and poor coalescence of
GaN grown over the mask. Research optimizing all of the process and fabrication parameters is
ongoing.
Figure 9 shows a typical X-ray rocking curve for a PE GaN sample grown on a 3C-SiC/Si
substrate. Similar to the case of GaN grown using LEO techniques, there is ~ 0.2° tilting of the
laterally grown GaN compared to that of the seed layer, in the < 1120> direction (i.e.
perpendicular to the seed form orientation). However, unlike the LEO technique, it is difficult to
determine if the PE material, the LEO material, or both are tilted since there is both pendeoepitaxial growth between and LEO growth above the original seed forms. Work is in progress to
make this determination.
Figure 7. 45° SEM view of PE GaN
grown on a 3C-SiC/Si substrate.
Figure 8. 45° SEM view of maskless
PE GaN grown on a 3C-SiC/Si substrate.
311
Along <1100>
^(1 to posts)
\FWHM« 1100 arose
WHMW-2124 arcs-c
«.5
17
17.5
CD (degrees)
Figure 9. DCXRD rocking curves of the 0002 diffraction of PE GaN grown on 3C-SiC/Si
substrates.
CONCLUSIONS
The pendeo-epitaxy process route has been developed as an alternative to the
conventional GaN LEO technique and as a means of confining all the vertically threading
defects, stemming from the GaN/AIN and AlN/SiC interfaces, within the seed forms. This
results in the growth of a more uniform low defect-density GaN epilayer. Incorporation of
silicon nitride masks and SiC pseudo-masks (either as the 6H-SiC substrate or a 3C-SiC
transition layer) combined with etched sidewalls of GaN seed forms has allowed the
achievement of PE GaN films with low dislocation densities over the entire GaN epilayer
surface. The quality of GaN seed layers grown on 3C-SiC/Si substrates was shown to be
comparable to GaN layers grown on 6H-SiC. Investigations regarding the optimization of the
PE growth technique, including determination of the ideal GaN seed thickness, ideal seed
microstructure geometry (e.g. post width, trench width, etc.) and ideal seed mask material, is
ongoing. Additionally, optimization of fabrication steps for the PE GaN seed forms (e.g.,
photolithography, mask alignment, ICP etching process, etc.) is underway.
ACKNOWLEDGEMENTS
The authors acknowledge Cree Research, Inc. and Motorola for the SiC and the silicon wafers
respectively. This work was supported by the Office of Naval Research under contracts
N00014-96-1-0765 (Colin Wood, monitor) and N00014-98-1-0654 (John Zolper, monitor).
REFERENCES
1
2
O. Nam, T. Zheleva, M. Bremser, and R. Davis, Appl. Phys. Lett., 71,2638 (1997).
A. Sakai, H. Sunakawa, and A. Usui, Appl. Phys. Lett., 73,481 (1998).
312
3
H. Marchand, X. Wu, J. Ibbetson, P. Fini, P. Kozodoy, S. Keller, J. Speck, S. Denbaars, and U. Mishra, Appl.
Phys. Lett., 73,747 (1998).
4
H. Zhong, M. Johnson, T. McNulty, J. Brown, J. Cook Jr., J. Schezina, Materials Internet Journal, Nitride
Semiconductor Research, 3,6, (1998).
5
S. Nakamura, M.Senoh, S. Nagahama, N. Iwasa, T. Yamanda, T. Matsushita, H. Kiyoku, Y. Sugimoto, T. Kozaki,
H. Umemoto, M. Sano, and K. Chocho, Proc. of the 2nd Int. Conf. On Nitride Semicond., Tokushima, Japan,
October, 1997
6
MRS Internet Journal of Nitride Semiconductor Research, January 13,1999
7
K. J. Linthicum, T.Gehrke, D.B. Thomson, E.P. Carlson, P. Rajagopal, T. Smith, and R.F.Davis, (submitted to
Applied Physics Letters)
8
K. J. Linthicum, T. Gehrke, D.B. Thomson, K.M. Tracy, E.P. Carlson, T.P.Smith, T.S. Smith, T.S. Zheleva, C.A.
Zorman, M. Mehregany, and R.F.Davis, MRS Internet J. Nitride Semicond. Res 4SI, G4.9 (1999).
9
T. Gehrke, K. J. Linthicum, D.B.Thomson, P. Rajagopal, A. D. Batchelor and R. F. Davis, MRS Internet J. Nitride
Semicond. Res 4SI, G3.2 (1999).
10
D.B.Thomson, T. Gehrke, K. J. Linthicum, P. Rajagopal, P. Hartlieb, T. S. Zheleva and R. F. Davis, MRS Internet
J. Nitride Semicond. Res 4SI, G3.37 (1999).
11
T. S. Zheleva, D.B.Thomson, S. Smith, P. Rajagopal, K. J. Linthicum, T. Gehrke, and R. F. Davis, MRS Internet
J. Nitride Semicond. Res 4SI, G3.38 (1999).
12
RMarchland, N. Zang, L. Zhao, Y. Golan, S.J. Rosner, G. Girolami, P.T. Fini, J.P. Ibbetson, S. Keller,
S.Denbaars, J. Speck, and U.K. Mishra, MRS Internet J. Nitride Semicond. Res 4,2 (1999).
13
P. Kung, D. Walker, M. Hamilton, J. Diaz, and M. Razeghi, Appl. Phys. Lett., 74,570 (1998).
14
T. Weeks, M. Bremser, K. Ailey, E. Carlson, W. Perry, and R. Davis, Appl. Phys. Lett., 67,401 (1995).
15
C. Zorman, A. Fleischman, A. Dewa, M. Mehregany, C. Jacob, S. Nishino, and P. Pirouz, J. Appl. Phys., 78,5136
(1995).
313
MASKLESS LATERAL EPITAXIAL OVERGROWTH
OF GaN ON SAPPHIRE
P. FINI*. H. MARCHAND**, J.P. IBBETSON**, B. MORAN*, L. ZHAO*,
S.P. DENBAARS*, J.S. SPECK*, U.K. MSHRA**
Materials Dept, Univ. of California, Santa Barbara; Santa Barbara, CA 93106
**ECE Dept., Univ. of California, Santa Barbara; Santa Barbara, CA 93106
ABSTRACT
We demonstrate a technique of lateral epitaxial overgrowth (LEO) of GaN, termed
'maskless' LEO, in which no mask is deposited prior to LEO regrowth. Instead, a bulk (> 2 urn)
GaN layer on sapphire is selectively dry etched, leaving -5 urn-wide stripe mesas oriented in the
<10T0>GaN direction, with a 20 um period. These stripes serve as seeds for LEO GaN growth,
which proceeds from the tops of the stripes and expands laterally, resulting in a 'T\ or overhang,
morphology. As for LEO over an Si02 mask, significant defect reduction (from ~109 cm"2 to
-10 cm"2) is observed in cross-sectional transmission electron microscopy (TEM). Atomic force
microscopy of the top surface of the LEO GaN reveals that no threading dislocations with screw
component terminate at the surfaces of laterally overgrown regions. X-ray diffraction
measurements reveal that the wings exhibit a crystallographic tilt away from the seed regions in
an azimuth perpendicular to the stripe direction; the tilt angle (-0.4 - 0.5°) is relatively
independent of growth temperature and wing aspect ratio.
INTRODUCTION
The technique of lateral epitaxial overgrowth (LEO) involves regrowth on a selectively
masked epilayer, such that the regrown material emerges from unmasked areas and overgrows
the mask. Through blocking and/or redirection of threading dislocations, the overgrown material
has a lower residual defect density than the seed material. Significant extended defect density
reduction has been demonstrated in GaN films grown by metalorganic chemical vapor deposition
(MOCVD) and hydride vapor phase epitaxy (HVPE) on sapphire,1"8 SiC,9"11 and Si(lll).5
Reductions in threading dislocation density (TDD) have led to direct benefits in device
performance such as long-lifetime C-W blue lasers,12"14 low reverse-bias leakage current
LED's15'16 and p-n junctions,17 low gate leakage current AlxGai_xN/GaN FET's,18 and low dark
current, sharp cutoff ALGai.xN-based solar-blind photodetectors.19
Much of the recent work in the LEO of GaN has utilized dielectric masks such as SiÜ2 and
SiNx, patterned with periodic arrays of stripes or holes. The use of Si02 in particular presents
difficulties in controlling the background carrier concentration of the LEO GaN, since
outdiffusion of Si and O may occur at the elevated temperatures of regrowth (e.g., 1050°C).
Such autodoping of the LEO GaN limits its usefulness for devices which require an insulating
base layer, such as field effect transistors. In the effort to obtain semi-insulating LEO GaN, one
approach is to eliminate the mask altogether, while still benefitting from extended defect
reduction. The present technique, termed 'maskless' LEO, has the potential of achieving this by
regrowing on GaN stripes that are separated by the sapphire upon which the seed layer was
grown. It is expected that the chemical and physical stability of the sapphire as well as the
regrowth morphology will lead to LEO GaN with lower unintentional impurity incorporation.
Thus maskless LEO has potential advantages over related techniques that yield similar
morphologies but use a dielectric mask.20 In this paper we present the structural and
morphological characterization of maskless LEO GaN grown under various conditions.
315
Mat. Res. Soc. Symp. Proc. Vol. 572 ® 1999 Materials Research Society
EXPERIMENT
Bulk (>2um) GaN layers were grown on sapphire using a standard two-step process.21
These layers were coated with photoresist (PR) and patterned using standard UV
photolithography, such that 5 urn wide PR stripes were formed, with 20 urn period and oriented
in the <10 1 0>G3N direction. Samples were etched using CI2 reactive ion etching, resulting in all
GaN being removed except the material beneath the PR stripes. Etching in this manner resulted
in stripes with some etch damage on the upper surfaces, and occasionally areas between stripes
where material ('grass') remained on the sapphire. Examples of both of these features are shown
in Fig. 1. It has been found in the present study that the selectivity of the LEO growth is not
affected by the features in Fig. 1(b).
Figure 1. Scanning electron micrographs of as-processed GaN stripes, showing (a) etch
damage on upper corners of stripe; (b) 'grass', or post-etch residual material.
The LEO GaN was deposited at 76 Torr in a H2 carrier, with an NH3 flow fixed at 1.77
slpm. Two trimethylgallium (TMG) flows (52 and 104 umol/min) and three surface
temperatures (1015°C, 1060°C, and 1100°C) were studied. The approximate surface temperature
was calibrated using an optical pyrometer. Growth durations of 15 min and 30 min were used for
frMG = 104 umol/min and 52 umol/min, respectively, to supply equimolar amounts of Ga to both
series of samples. Uncoated samples were characterized in cross section by scanning electron
microscopy (SEM) using a JEOL 6300F field emission microscope operating at 15 kV.
Specimens for transmission electron microscopy (TEM) were prepared by wedge polishing
followed by standard Ar+ ion milling and images were recorded on a JEOL 2000FX microscope
operated at 200 kV. Atomic force microscopy (AFM) of LEO GaN surface morphology was
carried out on a Digital Instruments D3000 scanning probe microscope. X-ray diffraction was
performed using four-bounce Ge(220)-monochromated Cu-Ka radiation in a four-circle
diffractorneter operating in receiving slit mode, with a 1.0 mm slit on the detector arm.
Specifically, rocking curves of the GaN 0002 peak were measured, with the scattering plane
perpendicular to the stripe direction. In this orientation, the rotation (rocking) axis is parallel to
the stripe direction.
RESULTS
The maskless LEO stripe morphologies that result from 15 min of growth at a TMG flow of
104 umol/min (nominal V/III = 750) are shown in Fig. 2. At the lowest temperature, the LEO
GaN is bound by the (0001) basal plane and both inclined {11 2n}_(where n «2) and vertical
{11 20} sidewalls. As the growth temperature is increased, the {1120} sidewalls dominate and
the lateral growth rate increases, which has been observed previously on LEO over an SiÜ2
316
mask.7 At all three temperatures studied, the LEO GaN grows vertically and laterally from the
upper part of the 'seed' GaN stripe, and not from the base of the {11 2 0} seed sidewalls. This is
believed to be due to etch damage such as shown in Fig 1(a), which would cause the LEO GaN
to start growing on the top surfaces of the stripes, and then expand laterally. Subsequently,
downward expansion of the LEO 'wings' also occurs, the extent_of which increases with
temperature. This implies that the growth rate of not only the {1120} sidewalls but also the
(000 T) plane increases with temperature.
Figure 2. SEM micrographs of as-cleaved LEO stripe cross sections, grown for 15 min with
104 umol/min TMG at (a) 1015°C, (b) 1060°C, and (c) 1100°C. Bar length is 1 urn.
When the TMG flow is set at 52 umol/min (nominal V/III = 1500), the LEO stripe
morphologies shown in Fig. 3 result. The same temperature dependence of facet formation as
seen above is observed, with the inclined {11 2n} facets disappearing at higher temperatures.
However, since the V/III ratio is higher in this case, the extent of lateral growth is also higher, as
observed for LEO over an Si02 mask.22 At the highest temperature, the LEO wings have
expanded downward to such an extent that they are in contact with the sapphire substrate. This
morphology is facilitated by the enhanced growth rates of the (0001) and {112 0} facets, which
could be directly dependent on the NH3 partial pressure. If it assumed that the decomposition of
NH3 is incomplete in the temperature range studied and the active nitrogen species have a short
residence time on the growing facets, then the 'local' V/III ratio in the immediate vicinity of the
stripe increases with increasing temperature. The relation between the ideal surface atomic
structures of the facets and their stability is discussed in more detail elsewhere.22
Figure 3. SEM micrographs of as-cleaved LEO stripe cross sections, grown for 30 min with 52
umoVmin TMG at (a) 1015°C, (b) 1060°C, and (c) 1100°C. Bar length is 1 urn.
Although the LEO stripes shown in Figs. 2 and 3 have (initially) suspended wings, it has
also been observed in the present study that the wings may be contact with the sapphire from the
onset of regrowth. This growth mode affects the dislocation density in the wings, as discussed
below. We believe that variations in the degree of etch damage on the seed sidewalls may alter
the manner in which LEO GaN initially grows on the seed.
Cross-sectional TEM micrographs of the stripes in Figures 2(a) and (c) are shown in Fig. 4.
The bright-field image in Fig. 4(a) reveals that some dislocations with edge component have
been redirected into the wing, which is expected when {11 2n} facets are present. The weakbeam images in Figures 4(b) and (c) reveal that the dislocations with screw and edge component,
317
respectively, have significantly lower density in the LEO wings than the seed regions. However,
the small wing volume in these samples precludes accurate estimation of the TDD at this time.
Figure 4. TEM micrographs of LEO stripes grown with 104umol/min TMG: (a) bright-field
cross-section; (b),(c) weak-beam images with two different diffraction conditions.
When the LEO wings are in contact with the sapphire while expanding laterally, an overall
reduction in defect density to ~106 cm"2 is observed, as shown in Fig. 5(a). However, vertical
arrays of edge dislocations with line directions parallel to the stripe and Burgers vector b =
1/3<11 20> are visible at various locations in the wings. These vertical arrays, highlighted by
arrows, would be expected to give rise to crystallographic tilt, since they are essentially lowangle tilt boundaries. Additionally, in Fig. 5(b) mixed-character threading dislocations generated
at the interface between the wing and the sapphire are highlighted with arrows.
Figure 5. Bright field TEM micrographs of (a) seed region and LEO wing in contact with
sapphire (arrows highlight vertical edge dislocation arrays), (b) close-up of LEO
wing / sapphire interface, with TDs highlighted by arrows.
The threading dislocation density reduction observed in cross-sectional TEM is also evident
in atomic force microscopy (AFM) micrographs of the top surfaces of the stripes, an example of
which is shown in Fig. 6. The right end of the micrograph is the LEO GaN located directly
above the seed, whereas the remainder is the wing region. The absence of step terminations in
this area indicates that there are no threading dislocations with a screw component emerging at
the surface, since their intersection with a free surface necessitates termination of two steps or a
single paired step.4 Step pairing as well as preferential orientation step orientation (dictated by
the in-plane symmetry of the wurtzitic crystal structure) is seen in the wing regions, as they are
unconstrained by step terminations.
318
• *!»*«£.' .
!p*='
r
t
'
&$y«
f
<ww>
r
11
Figure 6. Montage of AFM images for a stripe in contact with sapphire, showing 'wing'
(overgrown) region on left, and seed region on right. The total z-range is 2 nm.
Note that vertical lines are scan artifacts.
The x-ray diffraction rocking curves of the GaN 0002 peak in Fig. 7(a) reveal that for the
f(TMG) = 104 umol/min series, the LEO wings have a crystallographic tilt with respect to the
seed regions. This tilt, -0.4-0.5° for all three growth temperatures, is only observed in an
azimuth perpendicular to the stripe direction; no tilt is discernable along the <101 0>GaN stripe
direction. The same is true for LEO GaN stripes grown at f(TMG) = 52 umol/min, which have
tilts in the same range, as shown in Fig. 7(b). The fact that tilt is independent of aspect ratio
(which depends on temperature and V/III ratio) for these suspended morphologies may indicate
that it emerges at the onset of LEO. In contrast, LEO wings which grow in contact with an Si02
mask exhibit tilts that increase progressively with extent of lateral overgrowth.23
25000
14000
12000
oo
h
(b)
T=1060''C
T= 1100°C
Sj 10000
O 20000
s
&
*
£ 8000
\
i
_£. 15000
"5
IB 6000
in //
g loooo
*"
V' ?
4000
2000 1
0
§
/*—i;
. . _/._./,•.....
-0.5
-
5000
v
0
A to (°)
IK,
\ -V_ :
-1.5
0.5
-0.5
0
AGO
0.5
1.5
(°)
Figure 7. X-ray diffraction rocking curves of LEO stripes grown with (a) 104 umol/min for
15 min and (b) 52 umol/min for 30 min.
CONCLUSIONS
We have demonstrated a novel GaN lateral overgrowth technique, termed maskless LEO, in
which selective regrowth is achieved without the need for mask deposition. For the range of
growth conditions studied, morphologies with suspended wing (LEO GaN) regions largely result.
It is believed that RIE etch damage on the seed GaN stripes is the cause of this 'T' morphology.
TDD reduction has been observed in both TEM and AFM of LEO wing regions. The wings
exhibit crystallographic tilts away from the seed region, which are relatively independent of wing
aspect ratio. The origin of this tilt is the subject of ongoing study, as it is important in
determining the nature of the merge front when two wings coalesce.
319
ACKNOWLEDGMENTS
The authors would like to acknowledge funding from the Office of Naval Research (ONR),
monitored by C. Wood, and the Air Force Office of Scientfic Research (AFOSR), monitored by
G. Witt. PF acknowledges support from a NDSEG Fellowship, sponsored in part by the Office
of Naval Research. HM acknowledges support from the NSERC (Canada) and a Raychem
Fellowship.
REFERENCES
1. A. Usui, H. Sunakawa, A. Sakai, and A. A. Yamaguchi, Jpn. J. Appl. Phys., Pt. 2 36 (7B), L899-902 (1997).
2. A. Sakai, H. Sunakawa, and A. Usui, Appl. Phys. Lett. 71 (16), 2259-61 (1997).
3. A. Kimura, C. Sasaoka, A. Sakai, and A. Usui, presented at MRS Nitride Semicond. Symp., Boston, MA, USA,
1997 (unpublished).
4. H. Marchand, J.P. Ibbetson, P. Fini, P. Kozodoy, S. Keller, J.S. Speck, S.P. DenBaars, and U.K. Mishra, MRS
Internet J. Nitride Semicond. Res. 3 (3) (1998).
5. H. Marchand, N. Zhang, L. Zhao, Y. Golan, P. Fini, J.P. Ibbetson, S. Keller, S.P. DenBaars, J.S. Speck, and U.K.
Mishra, presented at 25th Int. Symp. Comp. Semicond., Nara, Japan, 1998 (unpublished).
6. A. Sakai, H. Sunakawa, and A. Usui, Appl. Phys. Lett. 73 (4), 481-3 (1998).
7. H. Marchand, J.P. Ibbetson, P. Fini, S. Chichibu, S.J. Rosner, S. Keller, S.P. DenBaars, J.S. Speck, and U.K.
Mishra, presented at 25th Int. Symp. Comp. Semicond., Nara, Japan, 1998 (unpublished).
8. H. Marchand, J.P. Ibbetson, P. Fini, X.H. Wu, S. Keller, S.P. DenBaars, J.S. Speck, and U.K. Mishra, MRS
Internet J. Nitride Semicond. Res. G 4.5 (G4.5) (1998).
9. T. S. Zheleva, Nam Ok-Hyun, M. D. Bremser, and R. F. Davis, Appl. Phys. Lett. 71 (17), 2472-4 (1997).
10. Ok-Hyun Nam, M. D. Bremser, T. S. Zheleva, and R. F. Davis, Applied Physics Letters 71 (18), 2638-40
(1997).
11. J. A. Freitas, Jr., O.-H. Nam, T. S. Zheleva, and R. F. Davis, J. Cryst. Growth 189-190, 92-6 (1998).
12. S. Nakamura, M. Senoh, S. I. Nagahama, N. Iwasa, T. Yamada, T. Matsushita, H. Kiyoku, Y. Sugimoto, T.
Kozaki, H. Umemoto, M. Sano, and K. Chocho, J. Cryst. Growth 189/190, 820-5 (1998).
13. S. Nakamura, M. Senoh, S. Nagahama, N. Iwasa, T. Yamada, T. Matsushita, H. Kiyoku, Y. Sugimoto, T.
Kozaki, H. Uimemoto, M. Sang, and K. Chocho, Japn. J. Appl. Phys. Lett. 37 (3B), L309-12 (1998).
14. S. Nakamura, M. Senoh, S. I. Nagahama, N. Iwasa, T. Yamada, T. Matsushita, H. Kiyoku, Y. Sugimoto, T.
Kozaki, H. Umemoto, M. Sano, and K. Chocho, Appl. Phys. Lett. 72 (2), 211-13 (1998).
15. T. Mukai, K. Takekawa, and S. Nakamura, Jpn. J. Appl. Phys. 37 (7B), L839-41 (1998).
16. C. Sasaoka, H. Sunakawa, A. Kimura, M. Nido, A. Usui, and A. Sakai, J. Cryst. Growth 189/190, 61-6 (1998).
17. P. Kozodoy, J.P. Ibbetson, H. Marchand, P.T. Fini, S. Keller, J.S. Speck, S.P. DenBaars, and U.K. Mishra,
Appl. Phys. Lett. 73 (7), 975-7 (1998).
18. R. Vetury, H. Marchand, J.P. Ibbetson, P. Fini, S. Keller, J.S. Speck, S.P. Denbaars, and U.K. Mishra, presented
at 25th Int. Symp. on Compound Semicond., Nara, Japan, 1998 (unpublished).
19. G. Parish, S. Keller, P. Kozodoy, J.P. Ibbetson, H. Marchand, P. Fini, S.B. Fleischer, S.P. DenBaars, and U.K.
Mishra, presented at 1998 Conference on Optoelectronic and Microelectronic Materials And Devices,
University of Western Australia, Perth, Australia, 1998 (unpublished).
20. T. Gehrke, K.J. Linthicum, D.B. Thomson, P. Rajagopal, A.D. Batchelor, and R.F. Davis, MRS Internet J.
Nitride Semicond. Res. 4SI, G3.2 (1999).
21. X. H. Wu, P. Fini, S. Keller, E. J. Tarsa, B. Heying, U. K. Mishra, S. P. DenBaars, and J. S. Speck, Japn. J.
Appl Phys. 35 (12B), L1648 (1996).
22. H. Marchand, J.P. Ibbetson, P.T. Fini, S. Keller, S.P. DenBaars, J.S. Speck, and U.K. Mishra, J. Cryst. Growth
195,328-332(1998).
23. P. Fini, J.P. Ibbetson, H. Marchand, L. Zhao, S.P. DenBaars, and J.S. Speck, unpublished (1999).
320
REPRODUCIBILITY AND UNIFORMITY OF MOVPE PLANETARY
REACTORS® FOR THE GROWTH OF GaN BASED MATERIALS
M. Heuken*, H. Protzmann*, 0. Schoen*, M. Luenenbuerger*, H. Juergensen*,
M. Bremser**, E. Woelk**
*AIXTRON AG, Kackertstr. 15-17, D-52072 Aachen, Germany, mail@aixtron.com
**AIXTRON Inc., 1670 Barclay Blvd., Buffalo Grove, IL 60089 USA, woe@aixtron.com
Production scale MOVPE reactors such as the AIXTRON 2000HT Planetary Reactor® offer
unique possibilities to fabricate highly efficient GaN based devices at a low cost of ownership. The
scope of this investigation is to understand the dependence of wavelength, thickness and doping
uniformity on parameters such as total gas flow, temperature distribution in the reactor and purity
of the precursors. Wafer to wafer uniformity in the 7x2" wafer configuration as well as run to run
reproducibility will be discussed. We obtained a wafer to wafer standard deviation of 2.7% for the
sheet resistance of Si-doped GaN/InGaN/GaN double heterostructures. The wafer to wafer
standard deviation of the main PL emission wavelength at 412.3 nm is 1.8 nm. The run to run
reproducibility of the main emission wavelength is <3 nm. We obtained reproducible resistivities of
GaN:Mg layers of less than 1 Dem which corresponds to 5-10xl017cm"3. Statistical data of p-type
doping taking 20 runs into account gave an average hole concentration of 5.5xl017cm"3. Together
with the wafer to wafer thickness uniformity of <1% the most sensitive layer properties are well
controlled to allow a cost-effective mass production process. Structures such as SQW and MQW
structures were grown to understand the performance of a production system with respect to
interface properties.
1.
Introduction
The market potential of blue, green, amber, red and even white LEDs based on GaN
heterostructures will increase dramatically. Since the profit margin of a LED is small, cost saving
production processes with a low consumption of expensive precursors and a high device yield is
essential [1, 2]. Therefore we developed the AIXTRON multiwafer Planetary Reactor® for the
growth of GaN-based material. This paper discusses reproducibility and uniformity issues under
production aspects.
2.
Experimental results
Undoped GaN layers were grown at TD = 1160 - 1180°C using a total pressure between 100 - 500
mbar. A typical V/III ratio for undoped GaN is -1000. These parameters result in a GaN growth
rate of about 2 um/h and an efficiency of the Ga precursor of 20%. Higher growth rates of 3 urn/h
and Ga precursor efficiencies up to 25% were also achieved. The typical background carrier
concentration of undoped GaN is 1 x 1016 cm"3. Depending on the ammonia purity and on layer
thickness background carrier concentrations as low as 5 x 101S cm"3 and mobilities as high as 600
cm2/Vs were achieved. The thickness uniformity on one single wafer without edge exclusion
measured by white light interference can be tuned to -1%. The wafer to wafer thickness uniformity
321
Mat. Res. Soc. Symp. Proc. Vol. 572
e
1999 Materials Research Society
in one growth run is well below 1%. Silicon doping of GaN was achieved by using SH4. Fig. 1
shows the sheet resistivity maps of four GaN layers grown in one run. The average value of the
wafer is 76.4 fi/ö. The standard deviation on each wafer is -1%. The wafer to wafer standard
deviation is <3%. Lines of constant sheet resistance are concentric circles on the wafer. This
indicates a linear depletion of the gas phase across the wafer together with a very uniform
temperature profile. Single and multiple quantum well structures of InGaN/GaN were grown and
investigated. The interface switching sequence, the growth temperature, total pressure for the
InGaN growth and the carrier gas for InGaN play an important role. We used in-situ refiectometry
measurements to optimize the growth step at the interfaces.
av. value: 73.8 ohm/sq
std. dev. 0.61 %
av. value: 79.2 ohm/sq
std. dev. 1.19 %
av. value: 76.2 ohm/sq
std. dev. 0.68 %
av. value: 76.4 ohm/sq
std. dev. 1.10 %
wafer-to-wafer standard deviation:
2.7%
Fig. 1: Wafer-to-wafer uniformity of Si-doped GaN/InGaN/GaN-DHS
322
Switching the precursor, ramping the pressure to the appropriate elevated pressure for InGaN and
the transition from H2 to N2 carrier gas was done in that way that the reflectivity of the grown
layer did not drastically decrease and that additional high quality GaN (as indicated by oscillation
of the reflectometry) can be grown on top of the quantum well. This growth technology was
developed to enable growth of high-quality LED structures.[3, 4]. Due to GaN specific design of
the temperature control system, the low thermal mass of the heated parts and the RF heating
system, fast heating and cooling cycles are possible. Fig. 2 shows the normalized emission
intensity as a function of emission wavelength at room temperature for seven satellite positions of
an InGaN/GaN multiple quantum well.
Satellite No.
-]
300
400
1
1
500
1
1
T"
600
700
800
Emission wavelength [nm]
Fig. 2: Normalized emission intensity as function of emission
wavelength for seven satellite positions
The emission wavelength of all layers is at 460 nm as indicated. Photoluminescence spectra
measured with a higher resolution show that the wafer to wafer difference is less than ± 1 nm.
Sometimes Fabry-Perot oscillations are superimposed on the PL making a precise evaluation of the
peak wavelength more difficult. Fig. 3 shows a PL wavelength mapping of one single wafer. The
323
average wavelength is 462.7 nm and the standard deviation is 1.9 nm. This result is in good
agreement with the conclusions of the measurements shown in Fig. 1 and in Fig. 2. To achieve
these data, an extremely uniform temperature profile across the reactor is required. Similar
structures were grown emitting at lower wavelength e.g. 394 nm. The standard deviation was 1.2
nm across one two inch wafer. Wafer to wafer standard deviations of less than 2 nm were already
achieved for structures emitting at 412 nm. To achieve InGaN quantum well structures emitting at
longer wavelength, tuning of the process parameters was necessary. Emission wavelengths up to
520 nm were achieved.
Peak Wavelength Mapping
Bm Size is 5.0 nm.
(mm)
SO
160.0
10
20
40
Average Wavelength:
Standard Deviation:
SO
462.7 nm
1.9 nm
Fig. 3: InGaN MQW structure grown on AIX 2000HT
In addition high temperature lasing experiments were carried out using these InGaN/GaN MQWs
under photo-excitation. Laser action was achieved up to 585K. The peak wavelength and the
FWHM of the lasing spectrum at 300 K were 423.2 nm and 9 meV, respectively. The laser
oscillation appeared near one of the peaks of the stimulated emission with increasing excitation
intensity. Lasing thresholds at T = 78 K, 300 K, 585 K were 24 kW/cm2, 95 kW/cm2, 555
kW/cm2, respectively. A characteristic temperature of 164K was derived from the temperature
dependence of the lasing threshold. This is in agreement with the value of 162 K reported by
others[5, 6]. Further details will be presented elsewhere [7].
324
3.
Conclusion
Optimized growth conditions for a GaN-based structures containing InGaN/GaN MQWs require
the use of different carriers gases, growth temperature and growth pressure. Our specially
designed MOCVD systems were able to achieve the results discussed here by precise control of
thermal and flow conditions. The high group III precursor efficiency, low V/III ratio and excellent
uniformity obtained will enable the low-cost production of GaN-based layers for LED fabrication.
4.
References
1. Y. Park, B.J. Kim, J. W. Lee, O. H. Nam, C. Sone, H. Park, Eunsoon Oh, H. Shin, S.Chae, J.
Cho, Ig-Hyeon Kim, J. S. Khim, S. Cho, T. I. Kim, MIJJNSR, Vol. 4, Article No. 1.
2. O. Schoen, M. Schwambera, B. Schineller, D. Schmitz, M. Heuken, H. Juergensen
Journal of Crystal Growth 195 (1998) 297-303.
3. Y. Kawaguchi, M. Yamaguchi, N. Sawaki, K. Hiramatsu, W. Taki, N. Kuwano,
K. Oki, T. Zheleva and R. F. Davis, Record of the 16th Electronic Materials Symposium,
L. Minoo, July 9-11, 1997.
4. O. Schoen, B. Schineller, M. Heuken, R. Beccard
Journal of Crystal Growth 189/190 (1998) 335-339.
5. S. Bydnik, T. J. Shmidt, Y. H. Cho, G. H. Gainer and J. J. Song, S. Keller, U. K. Mishra
and S. P. DenBaars., J. Appl. Phys. 72, 13, (1998) p. 1623.
6. S. Bydnik, Y. H. Cho, T. J. Schmidt, J. Krasinski, J. J. Song, S. Keller, U. K. Mishra,
S. P. DenBaars, Mat. Res. Soc. Symp. Proc. Vol. 512 (1998).
7. I. P. Marko, E. V. Lutsenko, V. N. Pavlovskii and G. P. Yablonskii, O. Schoen,
H. Protzmann, M. Luenenbuerger, M. Heuken, B. Schineller and K. Heime,
to be presented at the ICNS Montpellier 1999.
325
SYNCHROTRON X-RAY TOPOGRAPHY
OVERGROWTH OF GaN ON SAPPHIRE
STUDIES
OF
EPITAXIAL
LATERAL
PATRICK J. MCNALLY*, T. TUOMP*, R. RANTAMÄKI**, K. JACOBS***, L.
CONSIDINE****, M. O'HARE *, D. LOWNEY*, A.N. DANILEWSKY*****
Microelectronics Research Laboratory, Dublin City University, Dublin 9, Ireland.
**Optoelectronics Laboratory, Helsinki University of Technology, 02015 TKK, Finland.
***Dept of Information Technology (INTEC), Univ. of Gent, B-9000, Belgium.
****Thomas Swan & Co. Ltd., Harston, Cambridge CB2 5NX, U.K.
*****D-79108 Freiburg, Germany.
ABSTRACT
Synchrotron white beam x-ray topography techniques, in section and large-area transmission
modes, have been applied to the evaluation of ELOG GaN on A1203. Using the openings in 100 nm
thick Si02 windows, a new GaN growth took place, which resulted in typical overgrowth thicknesses
of 6.8 urn. Measurements on the recorded Laue patterns indicate that the misorientation of GaN with
respect to the sapphire substrate (excluding a 30° rotation between them) varies considerably along
various crystalline directions, reaching a maximum of a -0.66° rotation of the (0001) plane about the
[01«1] axis. This is -3% smaller than the misorientation measured in the non-ELOG reference,
which reached a maximum of 0.68°. This misorientation varies measurably as the stripe or window
dimensions are changed. The quality of the ELOG epilayers is improved when compared to the nonELOG samples, though some local deviations from lattice coherence were observed. Long range and
large-scale (order of 100 urn long) strain structures were observed in all multi quantum well
epilayers.
INTRODUCTION
Epitaxial lateral overgrowth (ELOG) holds out the potential for significant reductions in
threading dislocation densities for mismatched hexagonal-GaN on sapphire epitaxy [1-3]. Using
openings in a relatively thick Si02 mask a new MOVPE GaN growth is carried out. After an initial
phase of vertical growth upward through the mask window, the growth then proceeds laterally over
the mask itself. It is thought that a significant reduction in threading dislocation densities can be
achieved via mask blocking of vertically propagating dislocations and via a redirection of the
propagation of some dislocations at the growth front [4-5]. Studies have shown that the ELOG
technique can result in dislocation densities almost three orders of magnitude lower than in the nonELOG case, wherein typical densities of ~1 x 1010 cm"2 are often observed [3]. GaN-based opto- and
electronic devices are expected to benefit from this reduction in dislocation density and this
technique has recently been applied to GaN blue laser production [6-7]. However, an understanding
of the processes active during the ELOG procedure, and their impact on strain and dislocation
generation, is still far from complete. In this study, synchrotron white beam x-ray topography, in
section and large-area transmission [8-9] modes, have been applied to the evaluation of ELOG GaN
on AI2O3.
327
Mat. Res. Soc. Symp. Proc. Vol. 572 ®1999 Materials Research Society
EXPERIMENTAL
The measurements were performed at the Hamburger Synchrotronstrahlungslabor at the
Deutsches Elektronen-Synchrotron (HASYLAB am DESY, beamline Fl), Hamburg, Germany,
utilising the continuous spectrum of synchrotron radiation from the DORIS III storage ring bending
magnet. The ring operated at a positron momentum of 4.45GeV/c and at typical currents of 80150mA. The Laue patterns of topographs were recorded either on Kodak High-Resolution SO-343 or
SO-181 Professional X-ray film having an emulsion grain size of about 0.05 um. Transmission
topographic techniques were employed in producing these topographs [8-9]. For transmission section
topography, the incoming beam is collimated into a narrow ribbon by a slit typically 15-20 um in
width. When a single crystal is immersed in a white X-ray beam a number of lattice planes (hkl)
select out of the continuous spectrum the proper wavelengths to be reflected according to Bragg's
Law. Due to the low divergence of the synchrotron radiation beam (nominally 0.06 mrad FWHM),
each spot of this particular Laue pattern is itself a high-resolution topograph Thus, a set of Laue case
section topograph images of sample cross-sections are produced and, provided that the Bragg angle is
not too small, the image gives detailed information about the energy flow within the crystal and
direct depth information on the defects present in a particular crystal slice. For large-area
transmission topographs the collimating slits are opened to provide an incoming beam whose
dimensions are ~ 3 mm x 4mm.
The lattice mismatch between hexagonal wurtzite a-phase GaN and the underlying
rhombohedral (hexagonal) a-Al203 can achieve values as high as 13.9% [10] and standard epitaxial
deposition of GaN on sapphire leads to very high threading dislocation (TD) densities. However, as
stated earlier, the implementation of ELOG leads to significant reductions in TD densities from -10
cm"2 to ~107 cm"2. The epitaxial lateral overgrowth of GaN was carried out in a vertical rotating disk
MOVPE (Metal Organic Vapor Phase Epitaxy) reactor manufactured by Thomas Swan & Co.,
operated at a pressure of 100 Torr. Trimethylgallium (TMG), trimethylindium (TMI) and ammonia
(NH3) were used as Ga, In and N precursors respectively. At first, a 0.85-1.2 um GaN epilayer was
grown on top of a low temperature thin (33-50 nm) GaN buffer layer on a c-plane sapphire substrate.
Growth conditions of these layers are described elsewhere [11]. Afterwards 100-150 nm thick Si02
stripes were deposited using plasma enhanced chemical vapor deposition (PECVD) followed by
conventional photolithography and dry etching. Stripe widths (L) of 2 and 3 am were applied,
whereas the window openings (W) varied between 3 and 5 um. These stripes were oriented in the
<lT00> direction in order to obtain large lateral growth rates. The resulting structure was then reentered into the reactor after a 5 sec dip into a HF:H20 solution (1:4) to remove surface oxide. The
actual ELOG growth took place at temperatures ranging from 1060°C to 1080°C, using H2 as the
carrier gas. The thickness of the ELOG layer varied between 4.8 and 6.8 urn. Typical flow rates for
TMG and NH3 were 23.7 umol/min and 5 1/min respectively. When Multi-Quantum Well (MQW)
structures were implemented they were grown in N2 at 865°C and contained 5 periods of 3 nm thick
In02Gao.8N wells and 6 nm thick GaN barriers. Flow rates in this case consisted of 3 umol/min for
TMG and 9.6 umol/min for TMI. A diagram of the ELOG scheme used in this study is shown in
Figure 1. The wafers were divided into four regions, each with a differing fill factor, i.e. the ratio of
W to the total mask period, W+L. This information is given in Table 1.
328
ELOG
GaN
GaN epilayer
Si02
Mask
(0001) Surface Orientated A120
Figure 1 - ELOG structure
Sample Region
Window Opening
. W(nm)
a
b
c
d
SiOj Stripe
Width L (um)
5
3
4
3
4
2
3
3
Table 1 - ELOG Mask Dimensions
GaN
buffer
FiU Factor
W/(W+L)
]
'i
0.625
0.571
0.667
0.500
RESULTS
Figure 2(a) shows the Laue pattern taken from a complete film of Region a of the ELOG
GaN on A1203 material. The sample to film distance (D) was 50 mm. The major flat of the ELOG
GaN wafer is oriented in the [1120] direction of the GaN, so that means that the stripes on the
sample run along the [1 1 00] direction, which is perpendicular to the flat. Concerning the non-ELOG
sample, the shortest straight edge has the [1100] orientation, the other one is oriented in the [1120]
direction. Please note that all the aforementioned directions are for the GaN. There exists a 30°
rotation between the crystallographic orientations of the GaN and the sapphire substrate, which
means that the major flat of the wafer is one of the <1 100> directions for the sapphire. The direction
normal to the surface is the [0001] direction, so the surface is the so-called c-plane, or the (0001)
plane.
The images of the substrate and epilayer are separated from each other in certain diffraction
directions, e.g. the 11 • 3 A1203 and the 01 • 1 GaN reflections. A small separation between certain
reflections of the substrate and epilayer is expected, since there exists a 30° rotation between the
substrate and epilayer. A magnified image of the highlighted reflections is given in Figure 2(b) and
the two stripe images are the section transmission topographs for the substrate (uppermost) and the
ELOG GaN (lower image). The expected separation of the two reflections, based solely on the
designed 30° mutual rotation, is indicated by 2A<]>. However, the measured separation, 2A9, is much
greater. Therefore it must be concluded that the observed misorientation between the substrate and
ELOG GaN is mainly due to lattice dilatation or rotation. In the case of strained layer GaN epitaxy,
the perpendicular component of the lattice parameter will change while the parallel components
remain unchanged. This results in a distortion of the relative positions of the lattice planes in the
epilayer with respect to the substrate. Thus, along directions where the diffraction vector (g) and
distortion vector running normal to the distortion (h) are mostly parallel (i.e. any region where g.h #
0) there will be an observed shift in the position of the diffracted image of the epilayer with respect to
the remainder of the crystal, i.e. the substrate.
329
200 microns
2A<(>.D
2A9.D
ti*s
—r
\
(a)
i
2(2A\|/).D /
mm
1
Figure 2 - Section
Topographs (ELOG)
—
/M
It was found that the A9max for non-ELOG GaN epilayers was -0.68°, while the A9max for the
ELOG GaN was -0.66°, representing a combination of rotation of the (0001) plane about a [01*1]
axis or lattice dilatation. Further experimentation will be required to determine which, if either,
mechanism is dominant. In particular SXRT may be sensitive to tilting of the c-planes in the ELOG
GaN with respect to the seed material. The quality of the sapphire substrate appears to be rather
good. However, one can see that the image of the epilayer is severely broadened and appears to be
almost as thick as the substrate itself. This effect is most likely due to the fact that the quality of the
epilayer is far from perfect and this manifests itself via local deviations from lattice coherence
throughout the epilayer. Various regions in the epilayer, each slightly misorientated with respect to
its neighbour, though still generally macroscopically aligned, will each contribute to a topographic
image. However, each of these regions will produce images at slightly different locations on the film.
If it is assumed that these deviations are symmetrically distributed around the nominal
crystallographic plane positions, the measured broadening of the epilayer image in the section
topograph is given by 2Ay. For the example shown here we find these local lattice misorientational
deviations to give values of A\|/=±0.07° = ±250 arcsec across an 8.5 mm length of epilayer.
Significant differences between this non-ELOG sample and the ELOG wafers are observed.
For the non-ELOG wafer the overall dilatational misorientation (A9max) reaches a larger
maximum of a 0.68° rotation of the (0001) plane about the [01*1] axis. This is >3% larger than
previously recorded for the ELOG wafers. The local variations from coherence within the epilayer
are much larger and values of Ay are as large as A\|/=±0.12° across an 8.5 mm length of epilayer.
This is principally due to the apparent growth of highly misorientated features within the epilayer
(arrows B and C in Figure 3). These features correspond to highly misorientated features on the
substrate. The epitaxial replication of defects in the substrate is not observed for the ELOG samples.
2(2Ay).D
Figure 3 - Non-ELOG Topograph
330
It was also observed that the misorientation varies as a function of the window/stripe
separation. This was especially apparent when comparing two regions on the ELOG mask with
greatly differing fill factors. An example of this is shown in Figure 4, which is a 01» 1 transmission
section topograph as before. The image is left in its original orientation on the recording film, i.e. the
substrate image would appear above the epilayer image.
2AC0.D
Region d
Figure 4 - Misorientation across two regions of the ELOG mask
In this case the beam scanned two regions simultaneously, i.e. Regions c and d as indicated
on the figure. In region d large spikes are observed - indicated by arrows A- (similar features are also
seen in Figure 2(b)) and could be caused by a build-up of strain/misorientation at the SiC>2 stripe
edges. Since the stripes in this region are close to the threshold of resolution for SXRT (ca. 1.7 |xm at
HASYLAB), it is difficult to discern individual windows. However, this argument is supported by
the fact that these features are not easily seen for Region c, wherein the stripe widths are reduced by
33%. There is a clear shift in the separation of the two regions of the epilayer with respect to the
substrate. Region d displays a slightly greater misorientation than Region c. This difference is
indicated on Figure 4 by the angle Aco, and measurements yield a value of Aoo=0.015°. Figures 2, 3
and 4 indicate that the epilayer signals from the seed GaN and the ELOG and non-ELOG GaN
epilayers cannot be easily distinguished. This is most likely due to the fact that the seed layer is much
thinner than subsequent layers and cannot be easily separated on the topographic images.
Nevertheless, the clearly distinguishable "kink" in the epilayer image of Figure 4 suggests that an, as
yet unconfirmed, strain/relaxation effect impacts on the observed relative misorientation.
Figure 5 is a 11 • 3 (substrate)/01 • 1 (epilayer) large-area transmission topograph (LAT) of a
sample upon which MOW structures have been fabricated. The LAT image covered two regions,
namely MQWs on ELOG GaN epilayer and MQWs on a non-ELOG GaN epilayer. Again, due to the
inherent misorientation of the sapphire substrate with respect to the GaN-based materials, the images
of the substrate and epilayers are conveniently separated. There is a slight overlap where the
substrate and epilayer images meet. Thus, the topographic images of the epilayers are unambiguously
not contaminated by substrate contributions. The topographic images of the sapphire substrate are
identical for both cases. However, clear differences emerge in the topographs of the epilayers. The
MQW structures fabricated on the ELOG side of the wafer contains strain features, which are not
observed on the non-ELOG side. These features manifest themselves as an extra "orange peel"
roughening of the image as seen on the bottom left hand side (LHS) of Figure 5. These features are
of the order of 100 (im in size and their origins are uncertain and were only observed for MQW
fabrication on ELOG material.
331
SUBSTRATE
NON-ELOG
Side
SUBSTRATE
ELOG-Side •
ELOG
MQWs
NON-ELOG
MQWs
Figure 5 - Large area transmission topograph of MQW structures
CONCLUSION
SXRT was used to monitor the improvement in the quality of ELOG GaN on sapphire when
compared to non-ELOG material. Some local deviations from lattice coherence were observed. In the
best ELOG epilayers, section topographic measurements yielded misorientational deviations of up to
±0.07° across an 8.5mm length of epilayer. Topographic measurements also revealed variations in
ELOG epilayer quality as stripe/window dimensions changed, though the non-ELOG layers were
invariably worse. The fabrication of MQW structures on ELOG GaN produces an "orange peel"
strain structure in the epilayers, whose dimensions are of the order of 100 p.m.
ACKNOWLEDGEMENTS
The support of T. Wroblewski at HASYLAB, Hamburg, Germany is greatly appreciated.
This project was supported by the TMR-Contract ERBFMGECT950059 of the European Union and
the Irish Forbairt International Collaboration Programme.
REFERENCES
1. K. Kato, S. Kitamura and N. Sawaki, J. Crystal Growth, 144, 133 (1994).
2. D. Kapolnek, S. Keller, R. Vetury, R.D. Underwood, P. Kozodoy, S.P. DenBaars and U.K.
Mishra, Appl. Phys Lett., 71, 1204-1206 (1997).
3. A. Usui, H. Sunakawa, A. Sakai and A.A. Yamaguchi, Jpn. J. Appl. Phys., 36, L899 (1997).
4. T.S. Zheleva, O.K. Nam, M.D. Bremser and R.F. Davies, Appl. Phys. Lett., 71, 2472-2474 (1997).
5. A. Sakai, H. Sunakawa and A. Usui, Appl. Phys. Lett., 71, 2259-2261 (1997).
6. S. Nakamura, M. Senoh, S. Nagahama, N. Iwasa, T. Yamada, T. Matsushita, H. Hiyoku, Y.
Sugimoto, T. Kozaki, H. Umemoto, M. Sano and Chocho, Appl. Phys. Lett., 72, 211-213 (1998).
7. S. Nakamura, M. Senoh, S. Nagahama, N. Iwasa, T. Yamada, T. Matsushita, H. Hiyoku, Y.
Sugimoto, T. Kozaki, H. Umemoto, M. Sano and Chocho, Jpn. J. Appl. Phys., 36, L1568 (1997).
8. M. Sauvage-Simkin in Synchrotron Radiation Research, eds. H. Winick and S. Doniach, pp. 179204, Plenum (1982).
9. T. Tuomi, K. Naukkarinen and P. Rabe, phys. stat. sol. (a), 25, 93-106 (1974).
10. O Ambacher, J. Phys. D: Appl. Phys, 31, 2653-2710 (1998).
11. L. Considine, E.J. Thrush, J.A. Crawley, K. Jacobs, W. Van der Stricht, I. Moerman, P.
Demeester, G.H. Park, S.J. Hwang, J.J. Song, J. Crystal Growth 195, 192-198 (1998).
332
CONDUCTING (Si-DOPED) ALUMINUM NITRIDE EPITAXIAL FILMS GROWN BY
MOLECULAR BEAM EPITAXY
J.G. Kim, Madhu Moorthy and R.M. Park,
Department of Materials Science and Engineering, University of Florida,
Gainesville, FL 32611
Jgkim@solid.ssd.ornl. gov
ABSTRACT
As a member of the III-V nitride semiconductor family, A1N, which has a direct energygap of 6.2eV, has received much attention as a promising material for many applications.
However, despite the promising attributes of A1N for various semiconductor devices, research
on A1N has been limited and n-type conducting A1N has not been reported. The objective of
this research was to understand the factors impacting the conductivity of A1N and to control the
conductivity of this material through intentional doping. Prior to the intentional doping study,
growth of undoped A1N epilayers was investigated. Through careful selection of substrate
preparation methods and growth parameters, relatively low-temperature molecular beam
epitaxial growth of A1N films was established which resulted in insulating material. Intentional
Si doping during epilayer growth was found to result in conducting films under specific growth
conditions. Above a growth temperature of 900°C, A1N films were insulating, however, below
a growth temperature of 900°C, the A1N films were conducting. The magnitude of the
conductivity and the growth temperature range over which conducting A1N films could be
grown were strongly influenced by the presence of a Ga flux during growth. For instance,
conducting, Si-doped, A1N films were grown at a growth temperature of 940°C in the presence
of a Ga flux while the films were insulating when grown in the absence of a Ga flux at this
particular growth temperature. Also, by appropriate selection of the growth parameters,
epilayers with n-type conductivity values as large as 0.2 ß"'cm"' for A1N and 17 Q"'cm"' for
Alo.75Gao.25N were grown in this work for the first time.
INTRODUCTION
Research on the III-V nitride materials system (A1N, GaN, InN, and their ternaries) has
been one of the hottest issues in current materials research. Ever since the successful
fabrication of the first highly efficient blue-light-emitting diodes [1] and blue-diode lasers [2],
enormous attention has been given to the III-V nitrides. The III-V nitride materials are not only
good candidates for optoelectronic devices, but also promising materials for high temperature,
high frequency, and high power electronic device applications [3]-[6].
Among the III-V nitrides, A1N has recently drawn attention due to its potential for use
in many device application areas. A1N displays high thermal conductivity, high temperature
stability and a large direct band-gap of 6.2eV and has attractive piezoelectric properties, which
render it suitable for surface acoustic wave device applications [7]. A1N (and high Al content
AlGaN) is also of particular interest due to its negative electron affinity [8], which can be
exploited for cold cathode applications in high power vacuum electronics and flat panel
displays.
Despite these promising attributes, however, progress in the development of AIN-based
devices has been limited due to difficulties encountered in doping the material. Attempts to
dope A1N have been made by several researchers [9]-[ll], however, all of these attempts have
resulted in the production of highly resistive material (p > 1000 ß-cm).
333
Mat. Res. Soc. Symp. Proc. Vol. 572 ° 1999 Materials Research Society
The work described in this paper was motivated by the need to develop methods of
producing low resistivity A1N and high Al-content AlGaN, aided by an understanding of the
mechanisms that have prevented such an achievement thus far. The objectives of this work
were to develop ways to obtain conducting intentionally doped n-type A1N and to understand
how it is possible to produce such material.
EXPERIMENT
The samples in this work were grown by molecular beam epitaxy employing the
previously reported rf plasma discharge, nitrogen free-radical source [12]-[14]. In order to
minimize hydrogen effect during intentional doping of silicon, N2 gas was introduced to the
free-radical source instead of NH3 gas. Also, standard effusion cells were used for gallium,
aluminum and silicon. First, sapphire (0001) surfaces were etched in H2S04 : H3PO4 (3:1)
solution at 160°C for 90 minutes, after which gives smooth (5um x 5um AFM RMS value of
1.0Ä) and featureless (free of polishing damage) surface. The etched sapphire samples were
further cleaned in the MBE growth chamber by heating up to 940°C for 60 minutes followed by
2-hour nitridation. The rf power used for nitridation was 250W and a nitrogen flux was 1x10"
torr.
Three types of buffer layers were attempted, namely, GaN, low temperature A1N
(Tsubstrate = 750°C), and high temperature A1N (Tsubstrate = 900°C). A lum-thick GaN epilayer
was grown at 750°C after each type of buffer layer was grown. From the room temperature
Hall-effect measurements of the GaN epilayers, the high temperature A1N buffer layer was
chosen to perform the growth and Si-doping of A1N epilayers since only the GaN layer grown
on high temperature A1N buffer was insulating. Furthermore, intentionally Si-doped GaN
layers grown on high temperature A1N buffer layer exhibits higher room temperature freeelectron mobility (-200 cmVs"1 at 1016 < n < 1017 cm"3) than the values of the layers grown
on the other buffer layers.
Three sets of 0.4~1.0um thick undoped and Si-doped A1N epilayers were prepared on
previously mentioned high temperature A1N buffer layers. First, a series of undoped and Sidoped A1N films was deposited at various substrate temperatures (850, 880, and 940°C) under a
fixed HI/V ratio to 2.7 x 10"3. The second set of samples was grown at the same series of
conditions with additional 0.1 um thick undoped GaN buffer layers. Finally, the conditions
used in second series of samples were used except an additional Ga flux of 7.0 x 10" Torr was
included during growth. In all cases, the Si effusion cell temperature was 1280°C, which
resulted in a Si doping concentration of mid 1019 cm"3 in GaN grown using a similar growth
rate (-0.1 um/hr). The samples were analyzed using several analytical techniques, namely,
reflection high electron energy diffraction (RHEED), Hall-effect measurement, secondary ion
mass spectrometry (SIMS), AFM, cross-sectional secondary electron microscopy (SEM), and
electron microprobe.
RESULTS AND DISCUSSION
Figure la ~ lc is the AFM images of sapphire (0001) surface change before and after
various etching times. As the etching progressed, polishing damage was observed (Figure lb)
which was removed after 90 minutes (Figure lc).
All undoped A1N and high Al content AlGaN epilayers were insulating. RHEED
pattern during growth showed that most of the samples were weakly (2x2) reconstructed. At a
lower growth temperature, it was more difficult to see or maintain the reconstruction during
growth. The first set of Si-doped samples, directly grown on high temperature A1N buffer
334
layer, was all insulating while the second set of Si-doped samples, grown on additional GaN
buffer layer, showed a temperature dependence of resistivity (Figure 2).
(c)
Figure 1. 5 x 5 urn AFM images of (a) as-received sapphire (0001) surface (RMS = 3.32A),
(b) after 30-minute etching (RMS = 2.34Ä) and
(c) after 90-minute etching (RMS = 1.15Ä) in H2S04 : H3P04 (3:1) solution at 160°C.
The figure indicates that above the growth temperature of 900°C, A1N films were
insulating, however, below the growth temperature of 900°C, the A1N films were conducting.
I—r—1—I—|—rB
1—i—|—rM—I—I—|—I—I
Essentially insulating materials
Hall system measurement limit (- 500 Ohm-cm)
2
io fceu
s
O
21 10' t-
10° T
: with GaN buffer layer
: without GaN buffer layer
10''
810
840
870
900
930
Growth Temperature (°C)
Figure 2. Resistivity data obtained from Si-doped A1N films grown
at various substrate temperatures with and without a GaN
buffer layer.
Even though n-type conduction was observed from Hall-effect measurement, the Hall voltage
was fluctuating greatly making it difficult to determine either carrier mobility or concentration.
Figure 3a - 3b is the SIMS result of the Si-doped samples. Due to the interference of N2 or CO
signal, isotope 29Si signal was used to compare the Si dose. The result indicates that regardless
of the growth temperature the amount of Si in the film was about the same. Therefore, it seems
that the compensation, not Si incorporation, is the reason for these resistivity changes.
335
II I I I I I I I I I M | I II I | I I I I | I I I Ij
ri'1 • ' • '
I I I I I I I I l'
200
400
600
800
200
1000 1200
'' '' 'I'
400
600
800
1000 1200
Sputter time (seconds)
Sputter time (seconds)
(b)
(a)
i i i i 11 i i i | i i i i | i i i 11 i i i i | i i i i
J ii i i i I i i i i l i
0
200
400
600
800
1000 1200
Sputter time (seconds)
(C)
Figure 3. SMS depth profile of a Si-doped A1N film grown at (a) 850°C, (b) 880°C
and (c) 940°C
Further reduction in resistivity was achieved from the samples grown with a fixed
additional Ga flux of 7.0 x 10"8 Torr (Figure 4). Fraction of the Ga flux was incorporated after
electron microprobe analysis for the samples grown at 880 and 850°C. However, no evidence
of Ga incorporation was detected on the sample grown at 940°C even after SIMS analysis. The
SIMS results of the samples grown at 940°C with Ga flux and without Ga flux were compared
336
A1N
|M '
vT.
HULL————(
Essentially insulating materials
\
Hall system measurement limit (~ 500 Ohm-cm)^
102
A1N
I 10'
S
A1N
%
tt
10°
A1N =
10-1
A1
0.93Ga0.07N
Ga
N
Aln7S 0 25
0.75 -U.Z3
:
'
io-
S10
•
840
.
I
i
]
A: wim a Ga flux
:
0: without a Ga flux
i ~
870
'
'
'
900
'
'
'
'
930
Growth Temperature (°C)
Figure 4.
Resistivity data obtained from AlxGa,.xN films grown
at various substrate temperatures with and without the
presence of a Ga flux. Ternary compositions were
determined by microprobe analysis.
107
T"
T
106 r.
=r- 41 A1N (with Ga flux)
\
'o
52
(fl
-i—i—i—=
41 AIN (without Ga flux)
105 -
1 10" r
O
.£
5
1Q3
r"
:
c
29 Si (with Ga flux)
'.
2
M io r-.
29 Si (without Ga flux)
10'
I
10°
0
200
400
600
Sputter time (seconds)
Figure 5. Si profiles obtained by SIMS analysis in Si-doped AIN
films grown with and without the presence of a Ga flux
at a substrate temperature of 940°C.
337
800
in Figure 5, and an order of magnitude increase in Si dose was observed from the sample with
Ga flux.
CONCLUSIONS
In this study, n-type conducting (conductivities as high as 17 Q~ -cm") AlxGai.xN (x >
0.75) epilayers were produced for the first time by Si-doping. The influence of growth
parameters, such as growth temperature, and the addition of a Ga flux, on the conductivity was
studied. Conductivity values increased with decreasing growth temperature and also with the
addition of a Ga flux.
ACKNOWLEDGMENTS
The author would like to thank Mr. Lynn Calhoun for his valuable assistance and
discussions regarding this research and Dr. Margaret Puga, who performed the SIMS
measurements.
REFERENCES
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
S. Nakamura, T. Mukai and M. Senoh, Appl. Phys. Lett. 64, 1687 (1994).
S. Nakamura, Proc. of the SPIE V2693, 43 (1996)
M.A. Khan, Q. Chen, M.S. Shur, B.T. Dermott, J.A. Higgins, J. Burm, W. Schaff and
L.F. Eastman, Electronic Letters 32, 357 (1996).
Y.F. Wu, B.P. Keller, S. Keller, D. Kapolnek, P. Kozodoy, S.P. Denbaars and U.K.
Mishra, Appl. Phys. Lett. 69, 1438 (1996).
S.C. Binari, J.M. Redwing, G. Keiner and W. Kruppa, Electronic Letters 33, 242
(1997).
Q. Chen, R. Gaska, M. Asif Khan, M.S. Shur, A. Ping, I. Adesida, J. Burm, W.J. Schaff
and L.F. Eastman, Electronic Letters 33, 637 (1997).
G.R. Kline and K.M. Lakin, Appl. Phys. Lett. 43, 750 (1983).
M.C. Benjamin, C. Wang, R.F. Davis and R. J. Nemanich, Appl. Phys. Lett. 64, 3288
(1994).
T.L. Chu, D.W. Ing and A.J. Noreika, Solid State Electron. 10, 1023 (1967).
X. Zhang, P. Kung, A. Saxler, D. Walker, T.C. Wang and M. Razeghi, Appl. Phys. Lett.
67, 1745 (1995).
M.D. Bremser, W.G. Perry, T. Zheleva, N.V. Edwards, O.H. Nam, N. Parikh, D.E.
Aspnes and R.F. Davis, MRS Internet Journal Vol. 1, Article 8 (1996).
H. Liu, A.C. Frenkel, J.G. Kim and R.M. Park, J. Appl. Phys. 74, 6124 (1993).
J.G. Kim, A.C. Frenkel, H. Liu and R.M. Park, Appl. Phys. Lett. 65, 91 (1994).
H. Liu, J.G. Kim, H. Ludwig and R.M. Park, Appl. Phys. Lett. 71, 347 (1997).
338
INVESTIGATION OF THE MORPHOLOGY OF A1N FILMS GROWN ON
SAPPHIRE BY MOCVD USING TRANSMISSION ELECTRON
MICROSCOPY
W.L. Sarney1, L. Salamanca-Riba1, P. Zhou2, S. Wilson2, M.G. Spencer2, and K.A. Jones3
'Dept. of Materials & Nuclear Engineering, University of Maryland, College Park, MD
Materials Science Research Center of Excellence, Howard University, Washington, D.C.
3
U.S. Army Research Laboratory, Adelphi, MD
2
ABSTRACT
To determine the effect of growth conditions on A1N film morphology, we investigated
several A1N films grown on sapphire by MOCVD with various V/III ratios. Transmission
electron microscopy was used to characterize the film's crystalline quality and defect
morphology. TEM results show that the resulting film morphology depends on the V/III ratio.
Films grown with NH3 flow rates below 170 seem have high crystalline quality. In contrast, we
observe columnar growth and secondary interfaces in films grown with NH3 flow rates at or
above 170 seem. The secondary interfaces are likely to be inversion domain boundaries (IDBs)
and may be associated with strain relaxation. We discuss the V/III ratio's effect on crystalline
quality, surface roughness, and IDB and columnar structure formation.
INTRODUCTION
Because of its large energy gap, A1N is a good candidate for the fabrication of high
temperature semiconductor devices and as the dielectric material in high power, high frequency
devices operating at high temperatures. However, the quality of A1N films still needs to improve
before this material can be used in practical applications. Due to the large lattice mismatch, A1N
films grown on sapphire tend to grow three dimensionally, leading to a high defect density and in
many instances, a columnar structure. In order to reduce the lattice mismatch from 35% to an
effective mismatch of 13%, A1N films grown on sapphire have a 30° in plane rotation with
respect to the substrate. Three dimensional growth negatively affects the optical and electrical
properties of A1N, therefore inhibiting its potential for high power, high frequency, and high
temperature applications. The goal of this experiment was to determine which growth parameters
result in the highest quality films. In a previous study, we have obtained the optimum growth
temperature range (1110°C - 1150°C) for high quality A1N films. In this study, we obtain the
optimum V/III ratio for this temperature range.
EXPERIMENT
Several A1N films were grown on (0001) sapphire in a low pressure MOCVD reactor. All
of the films were grown for 15 minutes and at a pressure of 10 Torr. Samples were grown at four
different temperatures ranging from 1110° C - 1160° C. The growth precursors,
trimethylaluminum (TMA), and ammonia, the hydrogen flow rate and the temperature were held
constant during the growth of each individual sample. Various NH3 flow rates, ranging from 130
- 190 seem, were used for different samples. Since the TMA flow rate was 15 seem for each
sample, an increase in the ammonia flow results in an increased V/III ratio.
339
Mat. Res. Soc. Symp. Proc. Vol. 572 ® 1999 Materials Research Society
Cross-sectional TEM samples were prepared using a tripod polishing and ion milling, and
were examined in a JEOL 4000FX TEM operated at 300 KV. The film's surface was examined
with atomic force microscopy (AFM) and initial crystallinity was assessed with in-situ RHEED
at the end of the growth.
RESULTS
Sample A is typical of the high quality A1N films we have grown and is shown in Figs. 12. Figure 1 shows that the film is single crystalline and has a relatively low density of threading
dislocations. High resolution images (Fig. 2) show that the interface in this sample is sharp and
that the film is of high crystalline quality. The film does not show the columnar structure that is
often observed in AIN/Sapphire films. The white and dark bands seen near the interface are due
to strain contrast, as typically observed in highly mismatched systems. Sample A was grown
with a low ammonia flow of 150 seem. AFM results show that sample A has a very smooth
surface.
Fig. 1 (0002) Dark field image (DF) of sample A
5SS
Fig. 2. High resolution image of sample A. The s subscript in the diffraction pattern indices
denote the substrate spots. The zone axes are (1210) s/(10l0) f.
340
Several of the A1N samples have an interface within the film, which lies 18 - 20 nm
above the AIN/Sapphire interface. These films were grown with NH3 flow rates at or exceeding
170 seem. One such film is shown in Figs. 3 and 4, and will be referred to as sample B in this
paper. The upper interface was not expected since the growth was carried out under constant
conditions and with no interruptions. The (0002) dark field image (DF) shown in Fig. 3 clearly
shows the second interface. High resolution images and diffraction patterns (Fig. 4) indicate that
the film consists of the 2H polytype on both sides of this interface. The interface may be an
inversion domain boundary, which is caused by Al-Al or N-N atomic bonding. IDBs have been
observed by TEM in A1N films in previous experiments1"2. The lower A1N layer is single
crystalline, as seen in Fig 4. The upper A1N layer is crystalline near the interface, but becomes
polycrystalline as the film thickness increases. The diffraction pattern, shown as inset to Fig 4,
was taken from the film and substrate using a large selected area aperture. The pattern shows the
superposition of the (0110) pattern for the sapphire substrate, the (1210) pattern from the
single crystalline A1N layer, and several spots in a ring pattern representing the polycrystalline
portion of the film. Fig 3. shows that the film, especially the upper layer, is highly defective. The
upper layer is columnar and contains many regions of Moire fringes, which arise due to the
film's polycrystallinity. AFM results show that sample B has a rougher surface than sample A,
and contains several small grains.
Figure 3. (0002) DF image of sample B
'-.
15-*-.
,
*.
. IK«
/<• •
EKI
*»* # ai «MiwwJliw.
Figure 4. High resolution image of sample B. The zone axes are (0110) s/(l 210)f.
341
Another example of a sample showing a second interface is shown in Figs. 5-6, and will
be referred to as sample C. This sample was grown with a high ammonia flow of 190 seem. The
second interface is not as sharp as seen in sample B, and appears to consist of several
dislocations and stacking faults. The film is crystalline 2H A1N on both sides of the interface
and throughout the entire film. The (0002) DF image in Fig. 5. shows that this film is columnar,
but does not contain Moire" fringes as seen in sample B. AFM results show that sample C has
small grains and is slightly rougher than sample B.
Figure 5. (0002) DF image of sample C
Figure 6. High resolution TEM image of sample C. The zone axes are (0110) J (1210) f.
Fig. 7 is a high resolution image of sample D, which contains a second and third
interface. This film was grown with an ammonia flow rate below 170 seem, but at a higher
temperature than all of the other films grown in this experiment. The secondary interface is
located 18 nm from the film/substrate interface, similar to samples B and C. High resolution
images show that the film below the interface, labeled region 1 in Fig. 7, consists of the 2H A1N
polytype. The region above the interface (region 2) shows a markedly different atomic
arrangement than the one seen in the film below the interface. This region may consist of 6H
342
A1N. The diffraction pattern (inset to Fig. 7), shows the superposition of the (0110) pattern for
the substrate, the (T2T0) pattern for the film, and several extra spots. Microdiffraction confirms
that these extra spots come from the region between the second and third interfaces. The film
above the third interface (region 3) consists of the 2H polytype. We do not currently know why
our growth conditions would lead to the film morphology seen in this sample. Further
investigation, including photoluminescence and x-ray analysis, are needed to identify the
structure seen in region 2. AFM images show that the surface of sample D consists of several
large grains and is approximately 3 times rougher than sample B or sample C.
Fig. 7. High resolution TEM image of sample D. The zone axes are (0110) s/(l 210)f.
DISCUSSION
Secondary interfaces, which we believe may be inversion domain boundaries, were seen
in several samples, such as those represented by samples B and C. The IDBs were consistently
located between 18 and 20 nm from the substrate. In a similar experiment with AIN/6H-SiC
films, we observe IDBs consistently located at 100 nm from the film/substrate interface. It is
possible that the location of the IDB is related to the degree of film/substrate lattice mismatch,
which is in turn related to the threading dislocation density. The DF images of samples
containing IDBs indicate that there is a higher density of threading dislocations below the IDB
than above it. Also, the majority of the dislocations are not continuous across the IDB. Since
343
dislocations cannot end inside a crystal, the threading dislocations must bend on the plane of the
IDB. Despite the reduction in dislocation density above the IDB, the film quality generally
degrades above the IDB as the film becomes columnar.
All of the higher quality films we have observed, such as sample A, were grown with
ammonia flow rates below 170 seem and at temperatures ranging from 1110° to 1150° C. These
films do not have the columnar structure, IDBs, or Moire fringes commonly observed in
AIN/Sapphire films, such as samples B and C.
Sample D, which was grown with an ammonia flow rate of 130 seem but at a higher
temperature of 1160° C showed regions of mixed polytype. We do not currently now why these
growth parameters would induce a polytype transformation. Further experiments are planned to
determine which MOCVD reactor conditions would cause the structure seen in sample D to
arise. Other than sample D, which was grown at the higher temperature than all of the other
samples in this study, we did not find a correlation between small temperature deviations and the
film morphology.
Films grown with ammonia flows above 170 seem and within the temperature range
1110°C - 1150°C such as samples B and C, contain a high density of defects and secondary
interfaces. Inversion domain boundaries are seen in all samples grown with ammonia flows at or
above 170 seem. We suggest that the excess ammonia may have caused N-N atomic bonding,
causing the secondary interface to arise.
CONCLUSION
We have found the optimum TMA/NH3 flow rate ratio range for A1N films grown on
sapphire by MOCVD. We find that samples grown within the temperature range 1110° - 1150°
C with NH3 flow rates between 130 and 170 seem and TMA rates of 15 seem are of high
crystalline quality, have a smooth surface, and do not have inversion domain boundaries or a
columnar structure. In contrast, samples grown with higher NH3 flow rates consist of a columnar
structure, have a rougher surface and contain inversion domain boundaries at 18-20 nm from the
film/substrate interface. These IDBs seem to be related to threading dislocations in the film.
Above the IDBs the film's quality is degraded and in some instances the film becomes
polycrystalline. Further work is necessary to understand the formation of IDBs and their
dependence on the growth conditions.
ACKNOWLEDGEMENTS
This work was supported by MRCP Army Grant No. DAAL 019523530.
REFERENCES
1. J.P. Michel, I. Masson, S. Choux, and A. George, Phys. Stat. Sol, 146, 97 (1994).
2. A. Westwood, M. Notis, J. Am. Ceram. Soc. 74,1226 (1991).
344
TEMPERATURE DEPENDENT MORPHOLOGY TRANSITION OF GaN FILMS
A.RA. ZAUNER, F.K. DE THEUE, P.R. HAGEMAN, W.J.P. VAN ENCKEVORT, J.J.
SCHERMER, AND P.K. LARSEN
Research Institute for Materials, University of Nijmegen, Toernooiveld, 6525 ED Nijmegen,
The Netherlands
ABSTRACT
The temperature dependence of the surface morphology of GaN epilayers was studied
with AFM. The layers were grown by low pressure MOCVD on (0001) sapphire substrates in
the temperature range of 980°C-1085°C. In this range the (0001) and {1101} faces completely
determine the morphology of 1.5 |im thick Ga-faced GaN films. For specimens grown at 20
mbar and temperatures below 1035°C the {1101} faces dominate the surface, which results in
matt-white layers. At higher growth temperatures the morphology is completely determined by
(0001) faces, which lead to smooth and transparent samples. For growth at 50 mbar, this
transition takes place between 1000°C and 1015°C. It is shown that the morphology of the films
can be described using a parameter OcaN, which is proportional to the relative growth rates of the
(0001) and the {1101} faces.
INTRODUCTION
Due to its material properties gallium nitride (GaN) has attracted enormous attention in
recent years, certainly after the success of GaN-based blue light emitting diodes. The lack of
lattice matched substrates for epitaxial growth of GaN films has led to the application of a
variety of substrates of which sapphire is most frequently used [1].
Despite the large mismatch in lattice constants between GaN and sapphire, device quality
layers can be obtained using metalorganic chemical vapour deposition (MOCVD) and a two step
growth procedure. On top of an initial buffer layer [2] a GaN film is deposited at relatively high
temperatures. In this paper the influence, of the deposition temperature on the surface
morphology of the layers is investigated using atomic force microscopy (AFM) and scanning
electron microscopy (SEM).
EXPERIMENTAL
The GaN layers were grown in a horizontal MOCVD reactor equipped with a SiC coated
graphite susceptor with a hydrogen gas driven rotating disc to obtain optimum uniformity during
the growth process. The growth temperatures calibrated for the centre of the disc are determined
by a thermocouple. The discharge of the hydrogen flow causes a small temperature decrease
towards the periphery of the disc.
Immediately before growth, the two-inch sapphire (0001) substrates were cleaned in
organic solvents, etched in a solution of HC1:HN03 = 3:1, and finally rinsed in de-ionised water.
All deposition runs started with nitridation of the substrate surface, carried out in a
nitrogen/ammonia (N2/NH3) gas stream at 1030°C, followed by deposition of a 20 nm thick
GaN buffer layer at 500°C. On this buffer layer, the GaN films were grown.
Using trimethylgallium (TMG) and NH3 as precursors, and hydrogen (H2) as carrier gas,
two series of GaN layers, one at 20 mbar and the other at 50 mbar total reactor pressure, were
grown at temperatures between 980CC and 1085°C. Growth was performed with a TMG flow of
63 umol/min and a NH3 flow of 2.5 standard litre per minute (slm), diluted with H2 to a total
345
Mat. Res. Soc. Symp. Proc. Vol. 572 ° 1999 Materials Research Society
flow of about 5 slm. All samples were grown for 1 hour which resulted in a GaN layer thickness
of about 1.5 |Xm. AFM and SEM were applied to study the surface morphology of the samples.
RESULTS AND DISCUSSION
For the naked eye, lower deposition temperatures result in matt-white samples, whereas
at higher deposition temperatures the layers look colourless and mirror-like. In figure 1 the
specimen appearance is given as function of growth temperature and pressure. It shows that the
transition from matt-white to mirror-smooth appearance occurs between 1000°C and 1015°C for
samples grown at 50 mbar, and at 1035CC for those grown at 20 mbar. To ensure that changes in
polarity are not the cause for the observed morphology transition, the polarity of the deposited
GaN layers is determined by anisotropic etching [3]. For all samples it is found that growth
occurred in the [0001] direction (Ga-face).
Figure 1. The appearance of about 1.5 |J.m thick
GaN layers versus growth temperature for two
different pressures
T 40CD
A
E
* matt-white
O transparent
A morphology transition
i—|—i—|—.—|—i—|—.—|—i
970
990
1010
1030
1050
1070
1090
temperature (°C)
One sample, grown at 20 mbar and 1035°C, forms an excellent specimen to study the
morphology transition, since it exhibits a mixture of both appearances. This sample, which will
be referred to as sample 'B', has a matt-white appearance at the edges, where the deposition
temperature has been slightly lower than at the central region, and is mirror-like and transparent
at the centre. To determine the temperature difference during growth along sample B, the centre
of a complete matt-white sample grown at 1030°C and 20 mbar and slightly thicker as sample B
is used as a reference. AFM examination of this sample shows a morphology comparable to that
on sample B at a position halfway between the periphery and centre. This indicates, taking the
thickness difference into account, that the growth temperature near the edges of sample B has
been 1027 ± 3 °C. For which can be concluded that a temperature gradient of only about 8°C
can cause a dramatic change in morphology.
In figure 2 an AFM image from the periphery of sample B is shown. The main
characteristic of the surface is formed by large and irregularly shaped pits of about 300-600 nm
wide, with crystallographically oriented side faces and with a density of about 1.2 x 109 cm'2.
The depth of the pits is at least 250 nm, however, the sides are too steep to give reliable depth
measurements using AFM. SEM observations showed that the angle between the pit walls and
the (0001) face is about 60°.
From the crystal structure of wurtzite-GaN and its lattice constants [4] the theoretical
crystal morphology of GaN can be determined using the 'connected net' theory [5,6]. It states
that {hklm} faces parallel to a network of atoms interconnected by strong bonds, within a slice
thickness dhkim, determine the morphology of a crystal. Three faces, {0001}, {1101} and
346
{1100}, are found to be parallel with such a connected net. The occurrence of these faces was
indeed reported in literature [7,8].
The intersection lines and the inclination of the previously mentioned pit walls with the
(0001) surface indicate that these side walls are in fact {1101} faces, for which an angle of
61.9° with respect to the (0001) face can be calculated. Since growth is performed on (0001)
sapphire substrates and no etching of the GaN surface is expected to occur, it can be concluded
that the pits are induced by the coalescence of islands bounded by {1101} facets. The same
conclusion was drawn in references [9,10] on the basis of surface topographs showing the
nucleation, lateral expansion, and coalescence of islands for GaN grown on c-plane sapphire by
MOCVD.
For the surface region located in the intermediate zone between the edge and centre of
sample B, an AFM image is given in figure 3. It shows the transition stage between the lateral
expansion of the islands and the dislocation induced step flow growth of the planar (0001)
surface. The density of pits is still about 1.2 xlO9 cm"2, but most of them are partially closed to
form about 7 nm wide grooves.
Figure 2. Height and deflection (inset)
AFM image of a surface area near the
periphery of sample B, grown at
1035°C and 20 mbar. Pits can be
recognised which are bounded by
{1101} faces.
Figure 3. Height and deflection (inset) AFM
image of a surface area between the periphery
and the centre of sample B. Pits and grooves
can be recognised.
At the centre of sample B (figure 4) the same kind of cavities is found as shown in
figures 2 and 3. However, their average sizes and density are considerably less, and are found to
be, respectively, 100 to 300 nm and about 4.0 x 108 cm"2, indicating that the coalescence of the
islands is almost complete.
From the results discussed above, it is clear that the deposition temperature largely
influences the surface morphology of the GaN layers. It can be argued that a lower density of
nucleation islands at the periphery induces the larger size of pits near the edge of the sample.
However, it is known that a lower temperature is expected to result in higher chemical driving
force for GaN growth, therefore, the nucleation rate will certainly not decrease [9]. To determine
the nucleation density along the samples, a specimen is grown for 20 minutes to a thickness of
0.46 um (instead of = 1.5 urn) under simular conditions.
347
On this thinner sample, indicated as sample T, the separate islands can be distinguished.
The island density proved to be constant along the entire sample, therefore the different
morphologies cannot be explained by a different nucleation density.
At lower temperatures, the growth velocity in the (1101) directions is low compared to
that in the [0001] direction, causing slow lateral growth of the islands, and the morphology of
the 1.5 Jim thick layers is dominated by {1101} faces. For increasing temperatures, growth
velocity in the (1101) direction increases compared to the growth velocity in the [0001]
direction and therefore, the slow growth in the [0001] direction dominates the surface
morphology.
Analogous to the CVD growth of diamond crystallites [11,12], a growth rate factor is
defined to describe the morphology of the Ga faced GaN films. The growth rate factor ocoaN
equals O[0001]/v[liol]) coscp, where v[000l] and v[]iol] are the growth rates in the [0001] and
[1101] direction, respectively, and (p (=61.9°) is the angle between the (0001) and {1101} faces.
For c<GaN = 0 no {1101} faces develop and a completely closed (0001) layer is formed, for OtaaN
> 1 only {1101} facets are present on the surface. The present work shows that besides a strong
dependence on the deposition temperature, ocoaN is also pressure dependent.
Assuming that pits and islands are hexagonal in shape and (0001) faces top the islands,
the values of otGaN can be determined (figure 5). From the average (0001) surface area and the
nucleation density of the islands, the smallest centre-edge distance x of the hexagonal island can
be calculated. Using this value of x and the layer thickness t, 0(GaN is given by
OoaN = t / (t + x tan <p).
Figure 4. Deflection AFM image of
the surface at the centre of sample B.
Figure 5. Top view (A) and cross section (B)
of hexagonal islands.
The minimum layer thickness tnjn necessary to obtain a smooth film depends on OfeaN
and the distance d between the nuclei, and is given by
daGaNtan(p
(1)
£* — Z 0Cr,aw
The surface area covered by the (0001) faces is estimated from AFM images along the
surface of sample B. The nucleation density of the islands could be directly determined from
sample T and was found to be 2.1 x 109 cm"2. As already stated, the island density proved to be
constant along sample T, it does not change as function of temperature and is assumed to be the
348
same for sample B. SEM cross section observations on sample B showed that the growth rate in
the [0001] direction was also constant, resulting in a constant layer thickness, t, of 1.48 + 0.03
u.m. For sample B a film with completely coalesced islands, i.e. the islands just touching each
other with their (0001) faces, is obtained for OoaN values < 0.87. In table I it can be seen that the
small temperature gradient across the substrate causes an increase of O^N values from 0.87 to
0.94. Since the layer thickness of sample B is constant, the change in OtaaN values is only caused
by a difference in v..[1101]
Table I. Growth rate factor OtaaN across sample B
—
2
distance from centre (10" m)
0.00
0.25
0.50
0.75
1.00
1.25
otGaN values
0.87
0.88
0.89
0.91
0.93
0.94
More detailed AFM measurements on the centre of sample B (see figure 6) show a high
density of interacting growth spirals, which emerge from dislocations with a screw component.
Theoretical calculations [13] revealed that the GaN surface during growth has a surface energy
comparable with a surface free of adsorbates. This implies high step and kink energies, which
predict strongly polygonized growth spirals [14]. However, all spirals are found to be circular.
This can be explained by assuming that crystal growth is limited by surface diffusion rather than
by the integration of growth units at the step sites. The spirals consist of monoatomic (2.5 + 0.3
Ä) or double steps (5.0 ± 0.4 Ä), favourably comparing with the dooo2 distance of 2.59 Ä. Steps
emerging from dislocations have in general a height of 2 dooo2, indicating a screw component
[0001].
Figure 6. Deflection AFM image of the surface in
the centre of sample B at high magnification.
Monoatomic (ss) and double (ds) steps are shown
emerging from hollow cores (he) at dislocations
with a screw component. Hollow cores related to
edge dislocations (ed) can be recognised.
On these GaN layers also a number of small holes is observed with a density of
1.5xl09cm"2. The holes often coincide with dislocations containing a screw component and have
a narrow diameter distribution; 48 + 9 nm. The diameter of these hollow cores is much wider
than can be expected from standard stress theory of dislocations [15]. The large diameter can
349
possible be explained by the precipitation of vacancies to form voids along the dislocation
lines[16].
CONCLUSIONS
The morphology of =1.5 um thick Ga-faced GaN epilayers, grown by MOCVD on (0001)
sapphire substrates, is strongly temperature dependent. For the investigated temperature range of
980-1085°C the growth morphology is determined by (0001) and {1101} faces. To describe the
observed morphology a growth rate factor OfcaN is introduced. At higher growth temperatures the
morphology is governed by growth in [0001] direction, whereas at lower growth temperatures, a
slower lateral growth in the (1101) directions determines the morphology for a longer time. On
the mirror-like surfaces single and double steps can be seen emerging from the hollow cores of
dislocations with a screw component.
ACKNOWLEDGEMENTS
This work has been financially supported by the Dutch Technology Foundation (STW) and the
Dutch Technology Foundation for Chemical Research (SON).
REFERENCES
[I] III-V Nitrides, edited by F.A. Ponce, TX>. Moustakas, I. Akasaki, and B.A. Monemar (Mater. Res.
Soc. Proc. 449, Pittsburgh, Pennsylvania, 1996)
[2] H. Amano, N. Sawaki, I. Akasaki, Y. Toyoda, Appl. Phys. Lett. 48, 353 (1986)
[3] J.L. Weyher, S. Müller, I. Grzegory, S. Porowski, J. Crystal Growth 182, 17 (1997)
[4] F.A. Pearton, in Optoelectronic Properties of Semiconductores and Superlattices, volume 2:, GaN
and Related Materials, edited by SJ. Pearton (Gordon and Breach Science Publishers, Amsterdam, 1997)
[5] P. Bennema, in Handbook of Crystal Growth, Volume 1, edited by D.TJ. Hurle (North-Holland,
Amsterdam, 1993) 477
[6] R.F.P. Grimbergen, H. Meekes, P. Bennema, C.S. Strom, L.J.P. Vogels, Acta Crystallogr. A 54, 491
(1998)
[7] T. Akasaka, Y. Kobayashi, S. Ando, N. Kaboyashi, Appl. Phys. Lett. 71, 2196 (1997)
[8] T. Kozawa, M. Suzuki, Y. Taga, Y. Gotoh, J. Ishikawa, J. Vac. Sei. Technol. B 16 (1998) 833
[9] X.H. Wu, P. Fini, S. Keller, E.J. Tarsa, B. Heying, U.K. Mishra, S.P. DenBaars, S.J. Speck, Jpn. J.
Appl. Phys. 35, L1648 (1996)
[10] P. Vennegues, B. Beaumont, S. Haffouz, M. Vaille, P. Gibart, J. Crystal Growth 187, 167 (1998)
[II] C. Wild, P. KoidI, W. Miiller-Sebert, R. Kohl, N. Herres, R. Locher, R. Samlenski, R. Brenn,
Diamond Relat. Mater. 2, 158 (1993)
[12] R.E. Clausing, L. Heatherly, L.L. Horton, E.D. Specht, G.M. Begun, Z.L. Wang, Diamond Relat.
Mater. 1,411(1992)
[13] J.E. Northrup, R. Di Felice, Phys. Rev. B 56 (1997) R4325
[14] WJ.P. van Enckevort, in Facets ofFourty Years of Crystal Growth, edited by WJ.P. van Enckevort,
H.L.M. Meekes, and J.W.M. van Kessel (University of Nijmegen, 1997, internal publication), 62
[15] F.C. Frank, Acta Crystallogr. 4 (1951) 327.
[16] P. Coulomb, J. Friedel, in Dislocations and Mechanical Properties of Crystals, edited by J.C. Fisher,
W.G. Johnston, R. Thomson and T. Vreeland (Wiley, New York, 1957) 555.
350
COMPARATIVE STUDY OF EMISSION FROM HIGHLY EXCITED
(In, Al) GaN THIN FILMS AND HETEROSTRUCTURES
B.D. Little*, S. Bidnyk*, T.J. Schmidt*, J.B. Lam*, Y.H. Kwon*, J.J. Song*,
S. Keller**, U.K. Mishra**, S.P. DenBaars**, W. Yang***
*Center for Laser and Photonics Research and Department of Physics, Oklahoma State
University, Stillwater, OK 74078
**Computer Engineering and Materials Departments, University of California, Santa Barbara,
CA 93106
***Honeywell Technology Center, Plymouth, MN 55441
ABSTRACT
The optical properties of (In, Al) GaN thin films and heterostructures have been
compared under the conditions of strong nanosecond excitation. The stimulated emission (SE)
threshold from AlGaN epilayers was found to increase with increasing Al content compared to
GaN, in contrast to InGaN epilayers, where an order of magnitude decrease is observed.
Optically pumped SE has been observed from AlGaN films with aluminum concentrations as
high as 26%. Room temperature SE at wavelengths as low as 327 nm has been achieved. In
contrast to the increase of SE threshold seen for AlGaN films, we found that AlGaN/GaN
heterostructures which utilize carrier confinement and optical waveguiding drastically enhance
the lasing characteristics. We demonstrate that AlGaN/GaN heterostructures are suitable for the
development of deep ultraviolet laser diodes.
INTRODUCTION
(Al, In) GaN epilayers and heterostructures have drawn much attention in recent years
due to the potential for blue/UV light emitters and detectors for use in high density data storage,
high temperature electronics, solar-blind detectors, atmospheric sensing, and medicine [1].
Recently, nearly a dozen research groups have demonstrated lasing in InGaN-based
heterostructures, with the lowest reported emission wavelength being 376 nm [2]. However, to
obtain shorter wavelength laser diodes, it is necessary to use AlGaN-based structures.
Preliminary studies have shown that the incorporation of Al into GaN increases the stimulated
emission (SE) threshold [3]. In this work, we demonstrate that the introduction of strong optical
and carrier confinement into AlGaN/GaN heterostructures can significantly reduce the SE
threshold. We also demonstrate SE in AlGaN epilayers with emission wavelengths as low as
327 nm at room temperature, illustrating that AlGaN is a suitable material for the development of
deep ultraviolet laser diodes.
EXPERIMENT
The GaN, InGaN, and AlGaN epilayers studied were grown on (0001) sapphire substrates
by MOCVD. The thickness of the AlxGai_xN layers were -0.8 nm, and had alloy concentrations
of x = 0.17 and 0.26. For the purpose of comparison, we used GaN and InGaN epilayers with
thicknesses ranging from 100 nm to 7.2 um. The sample growth parameters have been reported
elsewhere [4, 5]. We also studied an AlGaN/GaN separate confinement heterostructure (SCH)
grown by MBE on 6H-SiC. The active region of the SCH was a 70 Ä thick GaN layer, which
was sandwiched between a 600 A Alo.06Gao.94N cladding layer and a 2300 A Alo.11Gao.89N
351
Mat. Res. Soc. Symp. Proc. Vol. 572 • 1999 Materials Research Society
c
c
e
-8
<,
(0
c
£
325 330 335 340 345 350 355 360 365
305 310 315 320 325 330 335 340 345
Wavelength (nm)
Wavelength (nm)
Fig. 1. RT SE spectra at several pump intensities above and below the SE
threshold, lth, for AlGaN layers with alloy concentrations of (a) 17% and (b) 26%.
Low-power cw spontaneous emission is given by the dashed line.
waveguiding layer. The structure was deposited on top of a ~3 um GaN buffer layer.
The experimental setup for the study of SE consisted of nanosecond tunable dye lasers
pumped by either the doubled output of an injection seeded Nd: YAG laser (~7 ns pulsewidth and
10 Hz repetition rate at 532 nm) or the 308 nm XeCl line of an excimer laser (~8 ns pulsewidth
and 10 Hz repetition rate). In the case of Nd:YAG pumping, the deep orange output of the dye
laser was doubled to -310 nm using a nonlinear crystal. The excimer-pumped dye laser emitted
directly in the UV. All experiments were performed in the edge-emission geometry. The UV
laser excitation source was focused to a line on the sample surface using a cylindrical lens [6]. A
continuously variable ND filter was used to attenuate the laser power. The resultant emission
was collected from the edge of the sample and focused onto the slits of a 1 meter spectrometer
with a UV-enhanced gated CCD. The samples were mounted to the cold finger of a closed cycle
helium refrigerator that allowed continuous temperature tuning from 10 K to 300 K.
Photoluminescence (PL) was performed using the 244 nm line of an intracavity frequency
doubled cw Ar+ laser. PL was performed in a backscattering geometry to avoid distortion of the
spectra due to reabsorption effects. For PL excitation (PLE) spectroscopy, the quasimonochromatic light from a Xe lamp was dispersed by a 1/2 m spectrometer and used as the
excitation source. The signal was collected from the sample and focused on the slits of a 1 meter
double grating spectrometer coupled to a photomultiplier tube for detection.
RESULTS
The SE emission from AlGaN epilayers qualitatively resembles that from GaN epilayers.
However, the emission wavelength lies in the deep UV. The RT emission spectra for AlGaN
layers with alloy concentrations of 17% and 26% are shown in Figures 1 (a) and (b),
respectively, for excitation densities above and below the SE threshold intensity /,/,. The dashed
352
(b)
RT
x = 0.26
L|3.4- 3
n
_ i0.9 w \
Sh •
1
SP
/ 1
10
lexc (MW/cm2)
lexc (MW/cm2)
Fig. 2. Power dependence of the peaks marked in Fig. 1 at RT. The
spontaneous emission peaks, denoted by circles, show an approximately linear
increase with excitation intensity. SE exhibits a superlinear increase with
power, and is represented by squares.
line indicates low power cw PL results. As seen from the figure, SE for both samples emerges
out of the low energy wing of the spontaneous emission peak as the excitation density is
increased. We believe that the SE observed from these epilayers (330 nm at RT) is the shortest
ever reported in the literature for a semiconductor material. The spontaneous and stimulated
emission peaks are separated by 10.5 and 8.5 nm, respectively, for the 17% and 26% epilayers,
which is comparable to the spacing observed in GaN epilayers [7] and is considerably smaller
than that seen in InGaN epilayers [8]. The redshift of the spontaneous emission seen under
nanosecond excitation in Figs. 1 (a) and (b) compared to the cw spontaneous emission peaks is
due to a reabsorption of the emitted radiation as it travels along the excitation path in the edgeemission experiments.
Figure 2 illustrates the emission intensity of the spontaneous and stimulated emission
peaks as a function of excitation intensity Iexc for the AlGaN epilayers presented in Fig. 1. The
spontaneous emission is seen to increase roughly linearly across the entire range ofIexc. Above
Ifi,, we see the emergence of a new peak which exhibits a superlinear increase in emission
intensity with Iexc, clearly indicating the onset of SE [9]. The values of Ig, obtained from the
AlGaN epilayers are slightly larger than those obtained from high quality GaN epilayers, and
more than an order of magnitude greater than that of InGaN epilayers [3, 7, 9]. The high SE
threshold for AlGaN epilayers makes the development of laser diodes using this material rather
challenging.
Through this study, however, we show that AlGaN/GaN-based heterostructures can be
used to produce lasing with both short emission wavelengths and low lasing thresholds. The
SCH used in this study possesses a high degree of optical and carrier confinement. By pumping
the sample at 335 nm, we obtained lasing at 358 nm (362 nm) at 10 K (RT) with a lasing
threshold as low as 125 kW/cm2. Figure 3 (a) shows the RT emission spectra from the SCH at
several excitation intensities above and below /,/,. The shorter wavelength peak at 341 nm is due
to spontaneous emission from the x = 0.06 cladding region. As Iexc is increased, we see a narrow
353
"1
'
1
'
GaN
active
region
siope=2.4
b)
.1
J
/
/
p
]
P
- AUGa^N
|
cladding
_
slope=075 j^
I
_
:
f
slope=Q.S1
,.l
340
350
,
,
360
Pump Intensity (kVtfcm2)
Wavelength (nm)
Fig. 3. a) 10 K power dependence of the SCH at several pump intensities above and below
the lasing threshold. The peak at 341 nm is due to the cladding layer, b) Intensitydependent behavior of the peaks shown in a). An approximately linear increase is seen for
the cladding layer for all pump intensities and for the active region until the lasing threshold
is reached. At higher intensities, a superlinear increase is observed at 358 nm along with
spectral narrowing. The FWHM of the lasing peak is 3 A.
lasing peak (3 A FWHM) emerge at a wavelength of 358 nm. We did not observe any
broadening of this peak as the temperature was raised up to RT. This indicates that the FWHM
of the lasing peak is determined by the finesse of the cavity, which remains independent of
temperature. In our previous study, we showed that the cavity is formed by microcracks caused
by strain relaxation of AlGaN grown on SiC [10]. Fig. 3 (b) shows the detailed behavior of the
emitted intensity as a function of Iexc. As can be seen from the graph, the cladding layer emission
increases almost linearly with increasing Iexc. The emission from the active region behaves
linearly until Ig, is reached, after which a superlinear increase is observed. The RT SE threshold
value for this sample was determined to be 125 kW/cm2, representing a drastic reduction in
comparison to GaN and AlGaN epilayers. This low threshold value is made possible by the
carrier confinement and optical waveguiding properties of the structure. In fact the lasing
threshold is comparable with the best InGaN epilayers which makes it suitable for the
development of short wavelength LDs (Emission < 370 am at RT)- n mayt)e possible to further
reduce the threshold by optimizing the sample parameters such as active layer thickness as well
as the thickness and alloy concentrations of the cladding and waveguiding region.
CONCLUSION
We have systematically studied the stimulated emission properties of (In, Al) GaN thin
films and heterostructures. The stimulated emission threshold of AlGaN epilayers was found to
increase with increasing Al content compared to GaN, in contrast to InGaN epilayers, where an
order of magnitude decrease is observed. Room temperature stimulated emission was observed
at remarkably short wavelengths, demonstrating that AlGaN-based structures are a suitable
material for deep ultraviolet laser diodes. Furthermore, we achieved a substantial reduction in
the lasing threshold for AlGaN/GaN-based heterostructures. We showed that optical and carrier
confinement will play the key roles in the reduction of the lasing threshold in these structures.
354
ACKNOWLEDGEMENTS
The work at Oklahoma State University was funded by BMDO, DARPA, and ONR.
REFERENCES
1. S. Nakamura and G. Fasol, The Blue Laser Diode, (Springer, Berlin, 1997).
2. I. Akasaki, S. Sota, H. Sakai, T. Tanaka, M. Koike, and H. Amano, Electron. Lett. 32, 1105
(1996).
3. T. J. Schmidt, Yong-Hoon Cho, J. J. Song, and Wei Yang, Appl. Phys. Lett. 74,245 (1999).
4. S. Keller, A. C. Abare, M. S. Minsky, X. H. Wu, M. P. Mack, J. S. Speck, E. Hu, L. A.
Coldren, U. K. Mishra, and S. P. DenBaars, Mater. Sei. Forum 264-268,1157 (1998).
5. S. Bidnyk, T. J. Schmidt, G. H. Park, and J. J. Song, Appl. Phys. Lett. 71, 729 (1997).
6. X. H. Yang, T. J. Schmidt, W. Shan, J. J. Song, and B. Goldenberg, Appl. Phys. Lett. 66,1
(1995).
7. S. Bidnyk, T. J. Schmidt, B. D. Little, and J. J. Song, Appl. Phys. Lett. 74, 1 (1999).
8. Yong-Hoon Cho, T. J Schmidt, S. Bidnyk, J. J. Song, S. Keller, U. K. Mishra, and S. P.
DenBaars, Proc. MRS Fall G6.54,161, Boston (1998).
9. S. Bidnyk, T. J. Schmidt, Y. H. Cho, G. H. Gainer, J. J. Song, S. Keller, U. K. Mishra, and S.
P. DenBaars, Appl. Phys. Lett. 72,1623 (1998).
10. J. J. Song, A. J. Fischer, T. J. Schmidt, S. Bidnyk, and W. Shan, Nonlinear Optics 18 (2-1),
269(1997).
355
ATOMIC SCALE ANALYSIS OF InGaN MULTI-QUANTUM
WELLS
M. Benamara, Z.Liliental-Weber, W. Swider and J. Washburn,
E.O. Lawrence Berkeley National Laboratory, Berkeley CA 94720.
R. D. Dupuis, P. A. Grudowski and C. J. Eiting,
Microelectronics Research Center, University of Texas, Austin TX 78712.
J. W. Yang and M. A. Khan,
ECE Dept., University of South Carolina, Columbia, SC 29208.
Abstract
InGaN multiquantum wells grown by MOCVD on GaN have been investigated by
transmission electron microscopy techniques and numerical analysis of high resolution (HREM)
images. One objective of this research was to correlate the atomic structure and emission
mechanisms of InGaN quantum well. The studied layers contained 13% or 20% In. It was
shown that GaN/InGaN interfaces are rather rough and exhibit an oscillating contrast. Structural
defects were found on these interfaces. The relative c-lattice parameter variation in the well was
determined using numerical processing of HREM images. The lattice spacings appear to be
larger than that expected from Vegard's law suggesting the presence of a biaxial strain. Further
observations also revealed a redistribution of In within the well. Instead of a continous In-rich
layer, quantum dots were often observed along the well with a regular spacing. The formation
of these In-rich dots was not intented and their presence suggests either a periodic modulation
of strain along the well or In-rich cluster formation.
Introduction
Recent years have witnessed a strong interest in rH-nitride semiconductors and especially
InGaN alloys because of the wide spectral range covered by these materials. These counpounds
in epitaxial layers are known to be highly strained. This strain affects the structural and physical
properties of the material through several mechanisms. It was recently shown by P. Perlin et
al.[l] that strain induced piezoelectric fields are responsible for the emissions from InGaN
quantum-wells (QW). Another explanation was given by Chichibu et al.[2] where emissions
were suggested to arise from the recombination of excitons located at potential minima along the
QW. This was supported by Krüger et al.[3] and by Grudowski et al.[4] whose
photoluminescence measurements on InGaN QW showed a broadening of peaks. Thus, there is
still some controversy about the emissions mechanisms of InGaN multi-quantum wells
(MQW). Furthermore, the low miscibility of InN in GaN[5] and the large lattice mistmatch
between these two counpounds (11 %) affect indirectly the emissions through phase
separation[6], segregation or defect generation[7]. In this paper, we report on an atomic scale
study of the microstructure of InGaN MQW using High-Resolution Electron Microscopy
(HREM).
Experiment
Several InGaN/GaN MQW structures were grown by metal organic chemical vapor
deposition (MOCVD) on 1 (im thick GaN layer using conventional (0001) oriented sapphire
substrates. Details of the growth conditions have been published elsewhere[4]. The structure of
sample 1 consists of five periods of 2.5 nm Ino.13Gao.87N separated by 50 nm Ino.03Gao.97N
barriers. The structure of sample 2 consists of twenty periods of 2.5 nm Ino.2oGa0.8oN
357
Mat. Res. Soc. Symp. Proc. Vol. 572 ® 1999 Materials Research Society
separated by 50 nm GaN barriers. The TEM samples were prepared by standard procedure
using ion-milling at liquid nitrogen temperature and examined using both Topcon 002B and Jeol
ARM microscopes operating at 200 kV and 800 kV respectively. Before the observations, the
samples were mounted on a gold covered grid and then dipped in a KOH (50%) solution for 5
min. This etch step was aimed at removing damage created during the ion milling.
For the numerical processing, HREM images were first digitized using a Kodak camera.
An intensity profile for all bright spots ("blob') in the image is fitted using an appropriate twodimensional function so that the exact coordinates of its center is accurately determined with
sub-pixel resolution. Since each bright spot in the image corresponds to either an atomic column
or an electron pathway (depending on the imaging conditions), the displacement of the atomic
positions away from their bulk values can be determined for each of the unit cells along any
crystallographic direction. This allows to maping the strain distribution. Thickness and
composition have the same effect on the projected potential. Since the projected potential can be
extracted and mapped from the intensity distribution within each unit cell, chemical fluctuations
or thickness variations can be revealed. A detailed explanation of the procedure can be found in
the original papers[8-9].
Result
Fig. 1 is a conventional image of sample 1 (five QW) structure taken along the [1120]
direction. Although the In content in the wells is only 10% different from that in the barriers,
the stack of layers is clear. Layers appear homogeneous and are easy to distinguish. The dark
stripes correspond to the wells (Ino.13Gao.87N). A magnification of one the wells is shown in
fig. 2. The layers are pseudomorphic to the underlying GaN layer and no misfit dislocations are
present at the interfaces. Interfaces with barrier layers on either side are not sharp but rather are
rough. An average thickness of layers was deduced from several parts of the sample and was
equal to about 2.5 nm (5 basal planes). The striking feature of this image is that the layer does
not appear homogeneous; it exhibits an oscillating contrast, which was also observed
earlier[10].
fig 1: TEM image showing the general structure of the 5 QW sample.
358
In
0.03Ga0.97N
In0.13Gat).87N
Fig 2: HREM image of one quantum well. The layer exibits a non-uniform contrast and
interfaces between adjacent layers are rough.
Quantitative HREM was used to map strain profile across the well at different positions. The
accuracy of the value was first determined by performing the same measurements in bulk GaN
where lattice parameters are not supposed to vary. This accuracy depends on the pixel
resolution at which the HREM image is digitized. A higher number of pixels results in a better
accuracy. Since the unit cell in the image is not square but rectangular (oa), the c lattice
parameter corresponds to more pixels than the a parameter, and hence the determination of the c
parameter is more accurate. In our case, the accuracy on a and c parameters in the bulk GaN is
0.8% and 0.5%, respectively; it allows us to detect any variation of the c-lattice parameter
greater than 0.002 nm for average interatomic distance of 5.2 Ä. Values of a and c lattice
parameters perpendicular to the well at three different positions are represented in fig 3.
atomic layers
fig 3: Average strain profile across the Ino.03Gao.97N/Ino.13Gao.87N/Ino.03Gao.97N at three
different positions. The lateral resolution is 5.2 Ä.
These profiles were obtained by averaging the measured lattice parameter values in each column
along each atomic rows parallel to the quantum well. It was noticed that the a parameter keeps
359
the same value (3.19Ä), which is not surprising since the well is only 25Ä thick, but c exhibits
a spike at the center of the well, corresponding to an expansion of the lattice parameter. The
scale graduated in pixels gives the variation of the c-lattice parameter. The height and the width
of the spike varies from place to place along the well as can be seen in fig.3. Its average
amplitude is 2 %. The same measurement was conducted at the Ino.03Gao.97N/GaN interfaces
and the average spike height was found to be equal to 0.4%. Simple considerations implie that
the c-lattice parameter of Ino.13Gao.87N is equal to c=CGaN * (l+2%+0.4%) = 5.31 Ä ± 0.02Ä.
An indium mole fraction x in the well can be extracted if one assumes the layer is relaxed. In
such a case, the application of Vegard's law, i.e. c=CGaN(l-x)+cinN(x) with CGaN=519 A and
cinN=5.71 A, leads to an In content of about 23 % in the well for c=5.31Ä, that is far above the
intented 13 % In. But the a lattice parameter of a relaxed Ino.23Gao.77N is
ao.23=aGaN(l-x)+ainN(x)=3.27Ä, that is far above the measured value a=3.19Ä. In case the
Ino i3Gao 87N layer would be relaxed, the lattice parameters in the well would be co=5.26Ä and
not 5.31 A+0.02Ä, and ao=aGaN(l-x)+ainN(x)=3.23Ä and not 3.19Ä. The fact that (oco) and
(a<ao) simultaneously lets us think that the layer is under biaxial compression. This observation
is supported by the TEM experiments which showed that the layer is pseudomorphic. If we
suppose that the In content in the well is the one intented (13%), the biaxial strain £c can be
calculated and is equal to £c=(c-co)/co=0.009+/-0.004. The measured a and c lattice constants
are linked to the Poisson ratio v by ec/ea=((c-c0)/co)/((a-ao)/ao)=-2v/(l-v)=0.7±0.3. All the
above numerical values lead to v=0.24±0.08. This value is in a good agreement with the one
previously published [11].
Romano et al.[12] determined the value of the Poisson's ratio for GaN using X-ray
diffraction. It is equal to v=0.18±0.01. Using this value, the combinaison of the above
mentioned formulas gives the corresponding In content of 15.5+0.4 % in the well. This value is
close to the intented In concentration (13%).
KalAirf'
hiA«N
Fig. 4: (a) HREM image of Ino.03Gao.97N/Ino.13Gao.87N/Ino.03Gao.97N quantum well and
(b) the corresponding electron scattering potential map. The growth direction is from left to
right.
Fig 4.b is a map of the electron projected potential that was extracted from the pattern change
in the image fig. 4.a. Since the beam amplitudes of InN and GaN are similar when the crystal
thickness is less than 20 nm, images were taken in thicker regions of the samples in order to
360
extract chemical information from any pattern change. The scattering potential increases with
sample thickness and with the average atomic weight of the atoms in each column. We can
assume that the sample thickness does not change across the well because it is only 25 Ä thick.
Therefore, the scattering potential is directly related to any compositional change. Thus, the map
in fig. 4.b can be seen as the representation of the Indium concentration across the well. It
confirms the non-uniformity of Indium distribution within the well because variations of the
potential reflects Indium content fluctuations[13]. The map shows also that interfaces are not
sharp and the roughness is broader on the upper interface. As seen in fig 2, the succession of
bright and dark areas along the well in Fig 2 suggests that a redistribution of In has taken place.
This feature is even more apparent when the In content increases from 13% to 20%. Fig. 5 is
an HREM of sample 2. Quantum dot like structure with regular spacing is evident. The dots
have an average diameter of 30 Ä and their separation distance is about 40 Ä. They have either a
spheroidal or an elongated shape and their presence in certain regions implies that nucleation
and growth are not the same all over the specimen. This regular arrangement of these clusters
makes it unlikely that they are artifacts due to sample preparation. Only images of regions above
30 nm thick of samples were considered to avoid strain relaxation at edges of samples.
• tltt«<!>'.'"
• tint':
' .;i>i" '
; Jw«
Fig 5: HREM taken along [1120] of 20 QW sample. Note the regular arrangement of dots and
the presence of strained regions and defects at InxGai-xN/GaN. Arrows indicate interfaces.
The large contrast variation and abrupt modifications of pattern of adjacent unit cells in parts of
the image prove that the structure is highly strained. The large difference in lattice parameter
between wells and barriers could be responsible for the strained regions. Moreover, the periodic
modulation of the c-lattice parameter along the well is a maximum inside the clusters. Formation
of these cluters is likely to be related to an accumulation of In atoms.
In addition, a high density of dislocations and small dislocation loops in the (0001) plane
were observed at interfaces quantum wellftarrier. These loops appear at the InxGai-xN/GaN
interfaces and not at the GaN/InxGai-xN ones. These loops may be the result of agglomeration
of Indium atoms. Because of defect formation and modification of the pattern within the unit
cell, no chemical mapping can be obtained from such sample areas.
361
Summary
InGaN multiquantum wells grown by MOCVD on GaN layers have been investigated by
transmission electron microscopy techniques and numerical analysis of high resolution (HREM)
images. It was shown that GaN/InGaN interfaces are rather rough and exhibit an oscillating
contrast. Structural defects were found on the interfaces. Their density increases with the In
mole fraction in the wells. The relative c-lattice parameter variation in the wells was determined
using numerical processing of HREM images. This allowed us to determine strain in the wells
and to estimate the Poisson's ratio. We showed that quantitative TEM is a suitable tool for
determining independently the In composition in the well. Further observations also revealed a
redistribution of In within the well. Instead of a continous In-rich layer, quantum dots were
often observed along the well with a regular spacing. The formation of these In-rich dots was
not intented and their presence suggests either a periodic modulation of strain along the well or
In-rich cluster formation.
Acknowledgment
TEM work was supported by the Director, Office of Basic Science, Materials Science
Division, U.S. Department of Energy, under the Contract No. DE-AC03-76SF00098. The use
of the facility at the National Center for Electron Microscopy at Lawrence Berkeley National
Laboratory is greatly appreciated. The work at The University of Texas was partially supported
by DARPA under Grant MDA972-96-3-0014 monitored by R. Leheny and by ONR under
grant N00014-95-1-1302 monitored by J. Zolper.
References
[I] P. Perlin, C. Kisielowski, V. Iota, B. A. Weinstein, L. Mattos, N. Shapiro, J. Kruger,
E. R. Weber and J. Yang, Appl. Phys. Lett. 73 (19), (1998).
[2] S. Chichibu, T. Azuhata.T. Sota and S. Nakamura, Appl. Phys. Lett. 69,4188 (1996).
[3] J. Krueger, C. Kisielowski, R. Klockenbrink, Sudhir G. S., Y. Kim, M. Rubin and E.
R.Weber, MRS Symp. Proc. Vol. 468, 299 (1997).
[4] P. A. Grudowski, C. J. Eiting, J. Park, B. S. Shelton, D. J. H. Lambert and R. D.
Dupuis, Appl. Phys. Lett. 71 (11), (1997).
[5] I. Ho and G. B. Stringfellow, Appl. Phys. Lett. 69 2701, (1996).
[6] N. A. El-Masry, E. L. Piner, S. X. Liu and S. M. Bedair, Appl. Phys. Lett. 72 (1),
(1998).
[7] F. Ponce, D. Cherns, W. Goetz and R. S. Kern, MRS Symp. Proc. Vol. 482, 453
(1998).
[8] P. Schwander, C. Kisielowski, F. H. Baumann, Y. Kim and A. Ourmazd, Phys. Rev.
Lett. 71, (25), 4150 (1993).
[9] C. Kisielowski, O. Schmidt and J. Wang, MRS Symp. Proc. Vol. 482, 369 (1998).
[10] Y. Narukawa, Y. Kawakami, M. Funato, S. Fujita, S. Fujita and S. Nakamura, Appl.
Phys. Lett. 70 (8), 981 (1998).
[II] C. Kisielowski, J. Krüger, S. Ruvimov, T. Suski, J. W. Ager III, E. Kones,
Z. Liliental-Weber, M. Rubin, E. R. Weber, M. D. Bremser and R. F. Davis,
Phys. Rev. B, 54 (24), 17745 (1996).
[12] L. T. Romano, B. S. Krusor, M. D. Mc Cluskey, D. P. Bour and K. Nauka, Appl.
Phys. Lett. 73 (13), 1757 (1998).
[13] C. Kisielowski, Z. Liliental-Weber and S. Nakamura, Jpn. J. Appl. Phys. Vol. 36, 6932
(1997).
362
TEM STUDY OF Mg-DOPED BULK GaN CRYSTALS
Z. LILffiNTAL-WEBER, M. BENAMARA, S. RUVIMOV, J.H. MAZUR, J. WASHBURN,
I. GRZEGORY* and S. POROWSKI*
Materials Science Division, Lawrence Berkeley National Laboratory, Berkeley CA 94720,
62/203;
* High Pressure Institut, Unipress, Warsaw, Poland
ABSTRACT
Transmission electron microscopy was applied to cross-sectioned samples to study surface
morphology, sample polarity and defect distribution in bulk GaN samples doped with Mg.
These crystals were grown from a Ga melt under high hydrostatic pressure of Nitrogen. It was
shown that the types of defects and their distribution along the c-axis depends strongly on
sample polarity. Based on this finding the growth rate along the c-axis for the two polar
directions was compared and shown to be approximately ten times larger for Ga polarity than
for N-polarity. In the part of the crystals with Ga polarity pyramidal defects with a base
consisting of high energy stacking faults were found. The parts of the crystals grown with Npolarity were either defect free or contained regularly spaced stacking faults. Growth of these
regularly spaced cubic monolayers is polarity dependent; this structure was formed only for the
growth with N polarity and only for the crystals doped with Mg. Formation of this
superstructure is similar to the polytypoid structure formed in A1N crystals rich in oxygen. It is
also likely that oxygen can decorate the cubic monolayers and compensate Mg. This newly
observed structure may shed light on the difficulties of p-doping in GaN:Mg.
INTRODUCTION
Progress in GaN technology has been delayed for a long time because of difficulties in
obtaining p-doping. Only recently success of Amano et al [1] followed by Nakamura [2]
resulted in p-doping of GaN. GaN p-n junction blue light emitting diodes (LEDs) and lasers
have been obtained [3]. Despite this success Mg p-doping is still not fully understood.
Originally low-energy-electron-beam irradiation (LEEBI) [1] and recently thermal annealing
have been found to activate Mg and result in p-doping [2,3]. There are reports that Mg is not
distributed uniformly in the layer and has a high tendency to diffuse to the layer surface [4].
EXPERIMENTAL
In this paper structural studies of Mg doped bulk GaN crystals will be described. The GaN
crystals have been grown by the High Nitrogen Pressure Solution Method [5] from a solution of
liquid gallium containing 0.1-0.5 at.% of Mg [6]. Before the growth conditions are established,
363
Mat. Res. Soc. Symp. Proc. Vol. 572 ° 1999 Materials Research Society
the solution was homogenized at a temperature of 1100°C for several hours. Then the
temperature was increased up to the growth conditions (T = 1500 - 1600°C) and N2 pressure
about 15 kbars was applied. Typical time of growth was in the range of 100 -150 hours.
Crystals in the form of hexagonal platelets (similar to those for undoped crystals) reaching
dimensions up to 8 mm were found in the cooler part of the crucible. This shows that the Mg
impurity does not influence the relative growth rates to a significant degree and therefore the
form of the hexagonal platelets is similar to that for undoped crystals. However, it was
observed that the platelets with Mg are generally somewhat thicker suggesting that Mg impurity
does accelerate the growth in the c-direction. Observation with the Nomarsky microscope shows
that as in undoped samples one surface normal to the c-axis is more rough, with many surface
steps while the opposite surface is always mirror like. Convergent beam electron diffraction
(CBED) was applied to study polarity of these crystals [7]. Earlier studies of undoped GaN
crystals showed that the polarity of the crystals can also be recognized by using mechanochemical polishing with aqueous solution of KOH [8, 9]. Cross-section samples were prepared
from crystals directly removed from the crucible and also from those etched chemically. Electron
microscopy was applied in order to detect the presence of any structural defects, observation of
surface morphology and determination of crystal polarity.
^m
11
II
im
sun.
^^^MJHIBBfflBl
JL
R| II Ga m
II
1©
N
Fig. 1: Cross-section TEM micrograph of the rough side of an as-grown GaN:Mg platelet
sample. The inset shows the experimental and calculated CBED patterns together with a hardball model showing atom arrangement along the c-axis.
Fig. 1 shows a cross-section electron micrograph from a crystal directly removed from the
crucible. It can be seen that one side of the as grown crystal has a rough surface (saw-like) with
an amplitude of 10 nm. The distance between the spikes was about 7-10 nm. The opposite side
of this crystal did not show such undulation, the surface was practically atomically flat.
Convergent beam electron diffraction (CBED) was applied to study polarity of these crystals and
it turned out that the rough side of the crystal grew with N polarity (Fig. 1-inset). This is
opposite to the undoped bulk crystals, where the rough side of the crystal grew with Ga
polarity.
364
Mechano-chemical polishing of undoped crystals with an aqueous solution of KOH showed
that the chemically active side grew with N polarity and the Ga-side was chemically inert [9,10].
Similar polishing applied to the Mg doped crystals showed that the previously rough crystal side
became smooth while the mechanical damage remained on the opposite side. CBED studies
confirmed that the Mg doped crystals behave similarly to undoped crystals in respect to the
etching behavior, the crystal side with N polarity is chemically active.
Transmission electron microscopy (TEM) applied to undoped cross-section samples showed
that not only surface morphology for N- and Ga-polarity is different but the part of the crystal
on the N polar side remains practically defect free, while the part near the Ga polar side has high
density of stacking faults distributed randomly (Fig. 2a). All three possible types of stacking
faults have been found in these undoped crystals. Dislocation loops decorated by Ga precipitates
were often attached to high energy stacking faults converting them locally to low energy
stacking faults.
a)
undoped
G»
Ga
\M^?
N
SF7
Fig. 2: Schematic drawing of defect distribution in the undoped bulk GaN crystals (a) and in
the Mg doped crystals (b). The inset indicates crystal polarity. Stacking faults (SF) with attached
dislocation loops (DL) decorated by Ga precipitates (P) are marked in the undoped crystals and
stacking faults and pyramidal defects in the Mg doped crystals.
Dramatically different defect distribution was observed in the GaN samples doped by Mg. A
large difference in crystal structural quality has also been observed for different growth runs.
Based on their structural quality one can divide the crystals into three groups. The first group of
crystals was practically free of any structural defects. In those crystals small precipitates could
occasionally be observed on the Ga side.
The second group of crystals had a similar defect distribution as for undoped crystals, the side
of the crystals with N-polarity was free of defects, while the opposite side of the platelet had a
high density of defects (Fig. 3). It was noticed that the defect free area of the crystal was much
thinner than the defective part of the crystal, opposite to the undoped crystals. The thickness of
the defect free region along the c-axis direction was about 10% of the platelet thickness. It can
be concluded that the growth of the crystal along the c-axis in Ga polarity is much faster than the
growth with N-polarity.
The defects observed in Mg doped crystals were not simple stacking faults on c-planes, each
defect had a pyramidal shape with the base of the pyramid facing the Ga side of the platelet.
365
High resolution transmission electron microscopy revealed that high energy stacking faults are
formed on c-planes at the base of these pyramids and their sides are also stacking faults on
planes inclined about 45° to their base (Fig. 3). Since these bulk crystals contain oxygen
impurities (in the range of lO1^ cm_3), it is possible that these defects are "dome" shaped
inversion domains as observed earlier in A1N crystals rich in oxygen, where the base of such
pyramid forms a flat inversion domain and the sides are corrugated domains [10]. However,
details of these defects are still under investigation. The fact that the sides of the pyramid can not
be viewed edge-on in high resolution micrographs makes their interpretation more difficult.
In the third type of crystal the pyramidal type of defect was also present but the part of the
crystal grown with N-polarity had a superstructure (Fig. 2b). Every 10 nm a monolayer with an
enhanced contrast was observed (Fig. 4) creating an equidistant layer structure. High resolution
imaging reveals the presence of stacking faults e.g. a unit of cubic material inserted equidistantly
in the hexagonal material. This layer arrangement results in additional satellite spots dividing the
(0001) reciprocal space into 20 equal spaces and giving 0.52x20=10.4 nm distances between
the monolayers. The appearance of these monolayers is assumed to be related to Mg distribution
Fig. 3: TEM micrograph shows defect distribution in GaN:Mg crystals. Note that the upper
part grown with N polarity is defect free.
in the sample. SIMS measurements for these crystals show Mg concentration 2x10^0 cm~3 on
the sample surface and flat Mg distribution in the range 6xl0^cm"3 in the remaining part of the
crystal. No difference in Mg content was observed on the two sides of the crystal. Si and
oxygen impurities were also detected in the crystals.
Based on theoretical predictions [11] Mg should substitute of Ga sites, the occupancy of Mg
on Ga sites and N sites are expected to be different. Mg segregation is expected to be more
366
Based on theoretical predictions [11] Mg should substitute of Ga sites, the occupancy of Mg
on Ga sites and N sites are expected to be different. Mg segregation is expected to be more
likely on N-sites of the crystals. Predicted positions [12] for Mg occupancy on the N-side
would likely lead to the formation of stacking faults. These structures are reminiscent of the
polytype defects formed in A1N rich in oxygen [12,13]. In the A1N crystals with oxygen the
formation of an octahedral metal layer surrounded by a mixture of N and oxygen atoms were
expected. In our GaN crystals oxygen content is much lower (1018cm-3) but it cannot be
excluded that oxygen decorates the stacking faults. The presence of oxygen would explain the
fact that p-doping was not obtained in these crystals, oxygen most likely would be responsible
for the observed compensation.
Fig. 4: High resolution image of low-energy stacking faults arranged equidistantly on the side
of the GaN:Mg crystal which grew with N-polarity.
CONCLUSIONS
These studies showed that GaN bulk crystals grown from a Ga melt doped by Mg under
hydrostatic pressure of nitrogen differ from undoped crystals. First of all the surface
morphology of the as-grown crystals is reversed with respect to the crystal polarity in
comparison with the undoped crystals. The N-side of the platelet is rough and the Ga site is
smooth. The second important observation was that a much higher growth rate (about 10 times)
was observed for the Ga polarity surface than for the N polarity surface. The third observation
was related to defect distribution in these crystals. Some of them were almost structural defect
367
pyramidal defects with the base on c-planes and the sides of these defects inclined approximately
45° are observed. A high energy stacking fault was observed as the base of these pyramidal
defects. For some crystals regions grown with N polarity did not show any structural defects.
However, in some crystals stacking faults ( a cubic monolayer) equidistantly distributed were
observed. This arrangement of stacking faults most likely is related to Mg segregation on Nsites and is very likely a new type of polytype like those observed earlier in A1N crystals rich in
oxygen. Our crystals have much smaller oxygen concentration, but it is very likely that stacking
faults are heavily decorated by oxygen. This would explain why p-doping was not obtained in
these crystals since compensation by oxygen would take place [14].
ACKNOWLEDGMENT
This work was supported by the Director, Office of Basic Science, Materials Science
Division, U.S. Department of Energy, under the Contract No. DE-AC03-76SF00098. The use
of the facility at the National Center for Electron Microscopy at E.O. Lawrence Berkeley
National Laboratory is greatly appreciated.
REFERENCES:
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
H. Amano, M. Kito, K. Hiramatsu, and I. Akasaki, Inst. Phys. Conf. Ser. 106,
72(1989).
S. Nakamura, M. Senoh, and T. Mukai, Jpn. J. Appl. Phys. 31, L1708 (1991)
S. Nakamura, Paper Plenary 1, presented at the 24th International Symposium on
Compoud Semiconductors, San Diego CA, 8-11 September 1997.
M.E. Lin, G. Xue, L. Zhou, J.E. Greeen and H. Morkoc, Appl. Phys. Lett., 63, 932
(1993).
I. Grzegory, J. Jun, M. Bockowski, St. Krukowski, M. Wroblewski, B. Lucznik and
S.Porowski, J. Phys. Chem. Solids 56, 639 (1995).
S. Porowski, M. Bockowski, B .Lucznik, I. Grzegory, M. Wroblewski, H. Teisseyre,
M. Leszczynski, E. Litwin-Staszewska, T .Suski, P. Trautman, K. Pakula and J.M.
Baranowski, Acta Physica Polonica A vol., 92,958 (1997).
Z. Liliental-Weber, C. Kisielowski, S. Ruvimov, Y. Chen, J. Washburn, I. Grzegory,
M. Bockowski, J. Jun, and S. Porowski, J. Electr. Mat. vol. 25, 1545 (1996).
J.L. Weyher, S. Müller, I. Grzegory, and S. Porowski, J. Cryst. growth 182, 17
(1997).
Z. Liliental-Weber, M. Benamara, O. Richter, W. Swider, J. Washburn, I. Grzegory, S.
Porowski, J.W. Yang, and S. Nakamura, MRS Proc, vol. 512, 363 (1998).
A.D. Westwood, R. A. Youngman, M.R. McCartney, A.N. Cormack, and M. R. Notis,
J. Mater. Res. 10, 1287 (1995)
C. Bungaro, K. Rapcewicz, and J. Bernholc, Rapid Communications 1999 in print.
R. A. Youngman, A.D. Westwood, and M.R. McCartney, Mat. Res. Symp. Proc. 319
45 (1994)
Y.Yan, M. Terauchi and M. Tanaka, Phil. Mag. A 77,1027 (1998).
C. G. Van de Walle, J. Neugebauer and C. Stampfl, in GaN and Related
Semiconductors, edts. J.H. Edgar, S. Strite, I. Akasaki, H. Amano, and C. Wetzel,
Emis Datareviews Series No. 23, An Inspec Publ (1999) p. 275.
368
DEFORMATION-INDUCED DISLOCATIONS IN 4H-SiC AND GaN
M. H. HONG", A. V. SAMANT*, V. ORLOV*, B. FÄRBER*, C. KISIELOWSKI**,
P.PIROUZ*
Department of Materials Science and Engineering, Case Western Reserve University,
Cleveland, OH, 44106-7204.
Lawrence Berkeley Laboratory, Berkeley, CA.
ABSTRACT
Bulk single crystals of 4H-SiC have been deformed in compression in the temperature range
55O-130O°C, whereas a GaN thin film grown on a (0001) sapphire substrate was deformed by
Vickers indentation in the temperature range 25-800°C. The TEM observations of the deformed
crystals indicate that deformation-induced dislocations in 4H-SiC all lie on the (0001) basal plane
but depending on the deformation temperature, are one of two types. The dislocations induced by
deformation at temperatures above ~1100°C are complete, with a Burgers vector, b, of -(ll20)
but are all dissociated into two -(lOTo) partials bounding a ribbon of stacking fault. On the other
hand, the dislocations induced by deformation in the temperature range 550<T<~1100°C were
predominantly single leading partials each dragging a stacking fault behind them. From the width
of dissociated dislocations in the high-temperature deformed crystals, the stacking fault energy of
4H-SiC has been estimated to be 14.7+2.5 mj/m2. Vickers indentations of the [0001]-oriented
GaN film produced a dense array of dislocations along the three < 1120 > directions at all
temperatures. The dislocations were slightly curved with their curvature increasing as the
deformation temperature increased. Most of these dislocations were found to have a screw nature
with their b parallel to < 1120 >. Also, within the resolution of the weak-beam method, they were
not found to be dissociated. Tilting experiment show that these dislocations lie on the {1100}
prism plane rather than the easier (0001) glide plane.
INTRODUCTION
SiC, GaN, and alloys of the latter, are wide bandgap semiconductors showing considerable
promise for high-temperature and optoelectronic applications [1,2]. Recent advances in crystal
growth have resulted in the production of single-crystal, single polytype, SiC boules [3] and verynearly single crystalline GaN thin films [2]. In spite of their widespread interest, the deformation
behavior and microstructure of these materials have not been studied in sufficient detail.
The most common polytypes of SiC are 6H and 4H; these are now produced commercially in
boule form and are available as wafers. The stable ambient form of GaN has a 2H structure, and
thin films of the metastable 3C (cubic) polytype can also be grown by CVD on certain substrates.
All the polytypes of SiC and GaN are tetrahedrally coordinated and consist of two variants of a
basic tetrahedron: normal, T, and twinned, 7"; each of these variants can occupy three spatial
positions, T,, T2, and T3 (or T',, T'2, and T'3) (for details, see, e.g. [4, 5]). They predominantly
exhibit hexagonal symmetry, denoted by H in the Ramsdell notation [6], e.g. 2H (wurtzite), 4H,
6H or, in general, 2nH (where n is an integer), or rhombohedral symmetry, denoted by R in the
Ramsdell notation, e.g. 15R or, in general, (2n+l)R. The 3C (zincblende) polytype is a subset of
the rhombohedral polytypes and is unique in that it consists of only one tetrahedral variant (either
all normal, T, or all twinned, 7", tetrahedra) and additionally exhibits cubic (C) symmetry. Fig. 1
shows the <1120> projected structure of 4H-SiC and 2H-GaN with tetrahedral sequences
...T,T2T'IT'3... (periodicity of four) and ...T,T'3... (periodicity of two), respectively. The
structure of each polytype can also be considered in terms of stacking of widely-spaced double
planes aA, ßB and yC where a, ß, /represent basal planes of carbon or nitrogen (silicon or
gallium) atoms and A, B, and C represent interleaved parallel planes consisting of silicon or
gallium (carbon or nitrogen) atoms. Thus, the structure of 4H-SiC and 2H-GaN may also be
369
Mat. Res. Soc. Symp. Proc. Vol. 572 ° 1999 Materials Research Society
described in terms of the stacking sequences ...aAßBocAyC... and ...aAyC..., respectively.
Note that all SiC and GaN polytypes are polar along the c-axis: the [0001] direction is distinct from
the opposite [OOO1 ] direction.
_m\
(a)
—t—
—o
—
\\
i
(b) -j
'4
1
*fc
"W
-<
I
-o-
5
—• —
^• jj
^
\
'i
"ioo;
u
\S
-o—
3
12
c?—
3
1
100]
I
Lt
—c
—O—
^
<)
-<>— -o— —
—c
3
',
-^s
\
, &
\
—• — —a
[0001]
^\
L-t
-*- —o-
-^
[0001]
t,
-o—
_^_
-—
c
*
12
-O-
•—
—i
H
H
J
-O-
12
3
—»_
-o—
k
—• —( — a
12
4H-S1C (22)
...T,T2T,T3...
2H-GSN (11)
...aAßBaAyC...
...aAyC...
?—c
3
Fig. 1. The structure of (a) 4H-SiC and (b) 2H-GaN
The aim of this work is to investigate the deformation-induced microstructure of bulk 4H-SiC
single crystals and a 2H-GaN thin film as a function of temperature using transmission electron
microscopy (TEM). The dislocation structure is discussed based on the differences in the core
structure of partials as well as the slip planes on which the dislocations lie.
EXPERIMENTAL PROCEDURES
The bulk 4H-SiC single crystals used in this study were grown by the modified sublimation
technique [3]. Parallelepiped specimens with nominal dimensions 2x2x4 mm3 were oriented and
cut for single glide such that one pair of their lateral faces was parallel to {1100}, and the (0001)
basal plane made an angle of 45° with respect to the compression axis. The samples were
compressed at 550-1300 °C in ultra-high purity argon to a strain of -4-6%. On the other hand, the
GaN film was grown by molecular beam epitaxy (MBE) on a (0001) sapphire substrate and was
deformed by Vickers indentation in the temperature range 25-800°C. The indentation diagonals
were aligned along the <1120 > and (l 1OO) directions on the (0001) surface of the film. From the
deformed samples, 0.3 mm thick slices parallel to (0001) plane were sectioned with a diamond
wheel cutter. Subsequently, the slices were ground with emery paper to a thickness of -100 fim,
then dimpled to a thickness of -20 fim, and ion-milled to electron transparency at a voltage of 5 k V
at an angle of -15°. GaN thin films were prepared by back-side thinning with the foil surface
normal to the [0001] direction from the sapphire substrate side. The thin TEM foils were examined
in a Philips CM20 electron microscope operating at an accelerating voltage of 200 kV.
370
RESULTS AND DISCUSSION
The easy slip plane of hexagonal and rhombohedral polytypes is (0001) and the dislocations
have a perfect b=l/3<l 120> Burgers vector [7]. As in other tetrahedrally-coordinated structures,
these dislocations are dissociated into two partials with Burgers vectors b,=l/3<1010> and
b=l/3<0110>, where the subscripts / and t denote the leading and trailing partials, respectively.
The dislocations dissociate as follows:
1/3<1120> = l/3<10l0> + 1/3<0110>
In tetrahedrally coordinated compounds, because of the polarity along the [0001] axis, the core of
the perfect or partial basal dislocations consists of only one species, i.e. silicon (gallium) or carbon
(nitrogen), and, because the partial dislocations belong to the glide plane, they are denoted as Si(g)
[Ga(g)]orC(g)[N(g)][8].
4H-SiC
Fig. 2 shows typical bright-field (BF) micrographs of 4H-SiC deformed in compression.
Both micrographs in this figure were obtained using the g=1011 reflection near the [1012] zone
axis. Samant [9] has recently shown that plastic deformation of 4H-SiC above ~1100°C takes
place by the activation of the (0001)<1120> slip system with the uncorrelated motion of partial
dislocations. This is clearly shown in Fig. 2(a) where dissociated dislocations, consisting of a pair
of leading/trailing partials bounding a ribbon of intrinsic stacking fault, are observed. Standard
strain contrast experiments indicate that, as expected, the Burgers vectors of the partials are,
respectively, parallel to < 1100 > and <1010> directions.
or
«H^^H b
i*Ö.5ym
Fig. 2. TEM micrograph of 4H-SiC deformed in compression (a) at 1300°C and (b) at 700 °C
From the width of partial separation, the stacking fault energy of 4H-SiC has been estimated to be
14.7±2.5 tnJ/m2; this value is nearly five times larger than that (2.9±0.5mJ/m2) of 6H-SiC
obtained by the same techniques [10].
Fig, 2(b) shows a BF micorgraph of dislocations induced by deformation at 700 °C. In this
case, the microstructure is dominated by single leading partials without the associated trailing
partials; each partial drags a stacking fault. The Burgers vectors of the single leading partials are
all parallel to the <1100> directions, and LACBED experiments on three different segments have
shown that they have a silicon core (i.e. they are Si(g) partials) [11]. Thus, it appears that in the
4H-SiC single crystals, deformation proceeds by nucleation and glide of single leading Si(g)
partials at low-temperatures (<~1100°C), whereas it proceeds by the generation and glide of total,
although dissociated, dislocations at high temperatures (>~1100°C). It has already been argued
371
[12] that only the leading partials, with a silicon core, nucleate at low temperatures (<~1100°C),
and a higher temperature (>~1100°C) is required to nucleate the associated carbon-core trailing
partials whereby the glide of leading/trailing pairs (i.e., dissociated dislocations) will carry out the
plastic deformation: at a certain applied shear stress, the activation barrier for nucleation of trailing
partials is higher than that of the leading partials. Since the formation of the same partial
dislocation from the same source cannot occur more than once on the same glide plane, plastic
deformation of the crystal can take place to a very limited extent at low temperatures. On the other
hand, once thermal activation is sufficient to from the trailing partial, repeated dislocation
multiplication can take place from the same source on the same planes, and plastic deformation will
proceed by the glide of total dislocations. At these temperatures, the crystal will be ductile and
large strains are obtained [12].
2H-GaN
Fig. 3 shows a (a) plan-view and (b) cross-sectional micrograph of the as-grown GaN
specimen (before deformation). As shown by various authors (see, e.g., [13, 14]) there is a high
density of threading dislocations in such films: some are those connected to misfit dislocation
segments, and some generated to accommodate the tilt and twist misorientation of neighboring
domains in the granular structure of the film. The plan-view micrograph in Fig. 3(a), with the
inserted [0001] SADP, shows the typical cell structure of the film with the grain boundary
dislocations defining the cell boundaries. They are mostly aligned along the <1120> directions
that are the Peierls valleys in non-cubic tetrahedrally-coordinated crystals. The cross-sectional
micrograph in Fig.3(b) shows that the thickness of the GaN film deposited on the (0001) sapphire
substrate is approximately 2 fjm. The dislocations in this micrograph are mostly parallel to the caxis, i.e., to the growth direction. Standard g. b=0 invisibility criterion shows that most of the
dislocations are a-type edge dislocations with a Burgers vector parallel to the [1120] direction.
The SADP obtained from an aperture covering both the 2H-GaN and the sapphire substrate shows
the orientation relationship to be [ 11 2 0]GaN//[ 1100 ]sap, (1100 )GaN//( 112 0)sap.
Fig. 3. (a) plan-view and (b) cross-sectional micrograph of the as-grown 2H-Gai\
Fig. 4 shows micrographs taken under different reflections from the same region of a
specimen indented at 450 °C. From these, and micrographs taken under other reflections, the
Burgers vectors of dislocations were determined. Practically all the dislocations appear under
reflections of the type g= 1120 shown in Fig. 4(a). In this micrograph, the short segments are
projections of threading dislocations (lying along the c-axis) in the film, while the dislocations
denoted by symbols A and B, aligned along the <1120> directions, are newly-generated
(presumably by indentation) dislocations that are parallel to the (0001) plane of the film.
Dislocations denoted by A are in contrast when imaged with reflections of the type g= 1100 (Fig.
4(b)), while they are out of contrast when imaged with reflections of the type g= 1010 (Fig. 4(c)).
On the other hand, dislocations denoted by B are out of contrast when imaged with reflections of
372
the type g= 1100 (Fig. 4(b)) and are in contrast with reflections of the type g= 1010 (Fig. 4(c)).
These and other micrographs indicate that dislocations A and B have Burgers vectors parallel to
[1120] and [1210] directions, respectively. Since, the line directions of dislocations A and B are
parallel to their Burgers vectors, it is concluded that all the indentation-induced dislocations have a
perfect screw character.
Fig. 4. TEM micrograph of a 2H-GaN indented at 450 °C; imaged by reflection (a) g= 1120, (b)
g=1100 and (c)g= 1010
Since two different polytypes (2H and 3C) have often been observed in GaN, it would be
expected that this material has a relatively low stacking fault energy and, consequently, that, as in
4H-SiC, all the basal dislocations in the GaN film would be dissociated into two partials. Indeed,
recent weak-beam TEM of GaN powder deformed by pulverization shows that basal dislocations
in this material are dissociated into two partials with a width of 3-8 nm, corresponding to a
stacking fault energy of -20 mj/m2 [15]. However, this is not the case in the present experiments;
despite many attempts using the weak beam technique of TEM, no evidence of dissociation was
found for any of the dislocations in the deformed film that were parallel to the basal plane. An
example is shown in Fig. 5 which is a g/3g weak-beam micrograph (with g= 1120) from a GaN
film indented at 300 °C\ all the dislocations appear as single lines and not as pairs.
One possibility for the non-dissociation of basal 1/3<1120> screw dislocations in Fig. 4
could be that they actually lie on the {1100} prism planes. Unlike the basal plane, the stacking
fault energy on the prism planes is quite likely very high and the perfect dislocations, if dissociated
at all on these planes, would have a small dissociation width (less than the resolution of the weak
beam technique, -2.5 nm). In order to check for this, extensive tilting experiments were carried
out in the microscope and the changes in the shapes of individual kinked dislocations were noted
with tilting. Figure 6 shows a typical example observed with the incident beam (a) nearly normal
(tilted by -5°) to the basal plane and (b) after tilting the foil approximately 35° away from position
(a). Care was taken that the same region of the foil was imaged during the tilt. In (a), a superkink pair is arrowed on an otherwise straight dislocation. In Fig. (b), the same super-kink pair has
widened considerably after the foil has been tilted by -35° about the < 1120 > axis. This clearly
373
indicates that the dislocation is not lying on the basal plane (in which case, the super-kink pair
would have shrunk with the tilt) and the results are consistent with the fact that it lies on the prism
plane. This sort of experiment was performed on five? kinked dislocation segments lying parallel
to the basal plane and, in all cases, the dislocations were found to lie on one of the three {1100 }
prism planes.
Fig. 5. Weak-beam TEM micrograph of the 2H-GaN deformed at 300 °C using reflection
g= 1120 close to the [0001] zone axis
Two possibilities may account for these unexpected results. Under the complex stress field of
the indentation, the 1/3<1120> screw dislocations may have nucleated on (0001) basal planes,
where they would be dissociated; subsequently, they could have cross-slipped onto the prism
planes by the Friedel-Escaig mechanism. More likely, however, the deformation-induced
dislocations were probably nucleated on the prism planes from the sample surface because of large
shear stresses on these planes induced by the indentation.
CONCLUSIONS
The deformation-induced dislocations in 4H-SiC are all basal dislocations and are divided into
two types. For deformations performed above ~1100CC, the dislocations are predominantly
dissociated perfect dislocations, while in crystals deformed below ~1100°C, the dislocations are
predominantly single leading partials without their corresponding trailing partials. From the width
of the dissociated dislocations, the stacking fault energy of 4H-SiC has been estimated to be
14.7+2.5 mJ/m2.
In the GaN film, Vickers indentation produced a dense array of dislocations on the {1100 }
prism planes; these dislocations lie along the three < 1120 > Peierls valleys on these planes. Most
of these dislocations were found to have a screw character and, because of the high stacking fault
energy on the prism planes, they were not dissociated.
ACKNOWLEDGEMENTS
This work was supported by grant number FG02-93ER45496 from the Department of
Energy, and subcontract number 95-SPI-420757-CWRU from the Silicon Carbide consortium.
374
Thanks are due to Dr. Don Hobgood (previously of Northrop-Grumman) for providing a single
crystal of 4H-SiC.
m *
• » !_"
*VJ'u.ium»
Fig. 6. TEM micrograph of deformed 2H-GaN (a) before and (b) after tilting around the [ 1120]
axis
REFERENCES
1. W. J. Choyke, H. Matsunami and G. Pensl (edited), Silicon Carbide, A Review of
Fundamental Questions and Applications to Current Device Technology, (Akademie Verlag GmbH
Volume I and II, Berlin, 1997).
2. S. Nakamura and G. Fasol, The Blue Laser Diode, (Springer-Verlag, Berlin-Heidelberg,
1997).
3. Y. M. Tairov and V. F. Tsvetkov, J. Crystal Growth 43, 209-212 (1978).
4. P. Pirouz and J. W. Yang, Ultramicroscopy 51, 189-214 (1993).
5. P. Pirouz, Solid State Phenomena 56, 107-132 (1997).
6. L. S. Ramsdell, Am. Mineralogist 32, 64-82 (1947).
7. P. Pirouz, "Extended Defects in SiC and GaN Semiconductors", in Proceedings of the
International Conference on Silicon Carbide, Ill-Nitrides and Related Materials, edited by G.
Pensl, H. Morkoc, B. Monemar and E. Janzen (Trans Tech Publications Ltd. 264-268, Zurich,
Switzerland, 1998), pp. 399-408.
8. H. Alexander, P. Haasen, R. Labusch and W. Schröter, Foreword to J. Phys. (Paris) 40,
ColloqueC6(1979).
9. A. V. Samant, Effect of Test Temperature and Strain-Rate on the Critical Resolved Shear Stress
of Monocrystalline Alpha-SiC, Ph.D. Thesis, Case Western Reserve University, 1999.
10. M. H. Hong, A. V. Samant and P. Pirouz, Phil. Mag. A (1999). In press.
11. X. J. Ning and P. Pirouz, J. Mater. Res. 11, 884-894 (1996).
12. P. Pirouz, A. V. Samant, M. H. Hong, A. Moulin and L. P. Kubin, J. Mater. Res. (1999).
In press.
13. B. Heying, X. H. Wu, S. Keller, Y. Li, D. Kapolnek, B. P. Keller, S. P. Denbaars and J. S.
Speck, Appl. Phys. Lett. 68, 643-645 (1996).
14. X. J. Ning, F. R. Chien, P. Pirouz, J. W. Yang and M. Asif Khan, J. Mater. Res. 11, 580592 (1996).
15. S. Takeuchi and K. Suzuki, Phil. Mag. Lett. (1999). In press.
375
Ca DOPANT SITE WITHIN ION IMPLANTED GaN LATTICE
H. Kobayashi* and W. M. Gibson
Department of Physics, the University at Albany, SUNY, Albany, NY 12222, USA
ABSTRACT
We have investigated the Ca dopant site within the GaN lattice using ion channeling in
combination with Rutherford backscattering spectrometry (RBS), particle induced x ray emission
(PIXE) and nuclear reaction analysis (NRA). Metalorganic chemical vapor deposition (MOCVD)
grown GaN on c-plane sapphire substrates implanted with 40Ca at a dose of lxlO15 cm-2 with
post-implant annealing were investigated. The channeling results indicate that more than 80 % of
Ca are near Ga sites even in as-implanted samples, however, they are displaced by ~ 0.2 Ä from
the Ga sites and that the Ca goes to the exact Ga sites after annealing at 1100°C. We think that the
displaced Ca in the as-implanted samples are electrically compensated due to formation of complex
defects with donor like point defects, such as CaGa-v*N and/or Caoa-GaN, and that Caoa becomes
electrically active when these complex defects are broken and the point defects diffuse away with
annealing at 1100°C.
INTRODUCTION
GaN has attracted great interest not only for the fabrication of blue light emitting lasers but also
for high power and high temperature devices because of its outstanding thermal and chemical
stability [1, 2]. It has been reported that Ca implanted GaN becomes p-type from n-type with
post-implant annealing at -1100°C. The ionization level of Ca was estimated to be -170 mV,
which is as shallow as that of Mg [3]. Since Ca can be a p-type carrier without any coimplantation while Mg implantation requires P co-implantation to achieve p-type conductivity [4],
Ca is expected to be a desirable p-type dopant. Ion implant damage and removal by annealing
have been studied as a function of implantation dose and annealing temperature [5, 6]. However,
most of these works concerned only electrical and optical properties or crystalline quality of the
GaN itself, no work on structural information on impurity lattice location has been reported other
than Si implanted GaN [7].
We have studied directly the lattice location of Ca implanted GaN using ion channeling
combined with Rutherford backscattering spectrometry (RBS), particle induced x ray emission
(PBCE) and nuclear reaction analysis (NRA) for Ga, Ca and N detection, respectively. We
estimated the substitutionality of Ca in the GaN lattice from minimum channeling yield and studied
whether the Ca is in Ga or N sites by comparing channeling angular distributions of Ca, Ga and
N. We will also discuss displacements of Ca from lattice sites. In this paper, we will demonstrate
the lattice location of Ca. We will also discuss the mechanism of p-type conductivity of Ca
implanted GaN with post-implant annealing.
EXPERIMENTAL
2 (im-thick GaN layers were grown on c-plane sapphire substrates by metalorganic chemical
vapor deposition (MOCVD) at 990°C with a 25 nm-fhick GaN buffer layer grown at 520°C. 4°Ca
ions were implanted into the undoped GaN layer at a dose of lxlO15 cm"2 at an energy of 200 keV.
The implantation was performed at liquid nitrogen temperature. The maximum ^Ca concentration
* present address : Research Center Sony Corporation, 134 Goudo-cho, Hodogaya-ku, Yokohama, Japan 240
e-mail: hkobayas@src.sony.co.jp
377
Mat. Res. Soc. Symp. Proc. Vol. 572 ® 1999 Materials Research Society
and projected range were estimated to be 9xl019 cm"3 and 110 nm, respectively, by TRIM
calculations [8]. After ion implantation, the samples were annealed at 1100°C for 10 minutes with
ramp-up and ramp-down times of 10 minutes each. The annealing was performed in flowing dry
N2 gas ambient (500 cc/min) using a conventional tube furnace made of quartz capable of operation
up to 1200°C. The tube furnace was maintained at constant temperature, and ramp-up and rampdown rates were controlled by inserting or removing the samples. Before annealing, samples were
held at 300°C for 1 hour to eliminate water on the samples to avoid oxidation of the surfaces.
Ion beam analysis was performed using a 4 MV dynamitron accelerator at the University at
Albany, SUNY. A 3 MeV He beam was used to obtain channeling yields of Ga and Ca by RBS
and PRE, respectively, and a 3.2 MeV proton beam was used for N detection by NRA. Nuclear
resonant elastic scattering, 14N (p, p) 14N was used for N detection. A Si surface barrier detector
(50 mm2, FWHM = 14 keV) was placed at 171° from the beam direction for the RBS
measurements. Figure 1 shows typical RBS energy spectra. The Ga signal window was set as
shown in Fig. 1, to correspond to the Ca implanted depth region. A minimum channeling yield
(Xmin) of 1.4 % for the unimplanted GaN indicates the excellent crystalline quality of the GaN
film. The %min of as-implanted samples and the annealed sample at 1100°C were 4.4 % and 1.9
%, respectively. This shows that the implant damage is considerably reduced by post implant
annealing. However, the recovery is not perfect, suggesting that the optimum annealing
temperature may be higher than 1100°C. Alternatively there may be some residual strain or
dislocation defects in the implanted layer. Electrical measurements have shown that Ca dopant
activation can be achieved at 1100°C. Samples annealed at 1150°C showed poor crystalline
quality due to surface decomposition, so channeling angular scans were not done. To prevent
such surface decomposition, either high N2 pressure or an A1N encapsulation layer is necessary
[9].
To investigate the implanted Ca lattice location, PKE signals were measured by using a Si(Li)
detector with a 4 nm-thick prolene window (20 mm2, FWHM = 160 eV) that was placed at 135°.
Figure 2 shows typical PKE spectra. We placed 8 urn-thick aluminized mylar in front of the
PIXE detector in order to reduce the intensity of the Ga L signal. The Ga L signal was reduced by
a factor of 35 while more than 95% of the Ca K signal remained. This aluminized mylar was also
useful to shield light from GaN luminescence produced by the ion bombardment.
3 MeV He 10(lC
random
2.0
Energy (MeV)
Fig. 1
2.5
RBS spectra for random and <0001> aligned GaN of as-implanted
(lxlO15 cm"2), annealed (1100°C) and unimplanted samples.
378
2500
800
2000
» 1500
1
I 400
u
'^,'"N(p,p)"N
resonance
h I
i jHJy \
(• 0***\y
random
aligned <0001>_
U 1000
/+**>
I Ga
500
2.0
3.0
4.0
Energy (keV)
-H. /..... .1
0
^'•^ /^^*fc':"i-j..
1.0
5.0
2.0
6.0
Fig. 2 PRE spectra of the annealed sample.
Fig. 3
2.5
Energy (MeV)
NRA spectra of the annealed sample.
Thus we minimized the dead time of the PIXE detector. In off-line data analysis, we obtained the
Ca yield by curve fitting which took into account the Gaussian shaped Ca K signal and quadratic
bremstrulung background. The same detector as used for the RBS measurement was used for N
detection by NRA. The 14N (p, p) 14N reaction has a narrow resonance at 3.2 MeV with a
resonance width of 20 keV. The reaction cross section is 35 times larger than that of RBS at this
energy [10]. Figure 3 shows a typical NRA spectrum. A sharp peak due to the nuclear reaction
can be seen. Since the resonance width corresponds to a depth width of 300 nm in GaN, we
obtained the N signal only from the region where Ca was implanted, and could thus eliminate
dechanneling contributions from deeper layers. To obtain the N yield precisely, we measured
spectra at just below the resonance energy (off-resonance) to obtain the amount of continuum
background under the resonance signal and subtracted the background from on-resonance spectra.
The samples were mounted on a two-axis goniometer and channeling angular distributions for
Ca and Ga were obtained simultaneously. The channeling angular distributions of N were
obtained in the same way. A total beam dose of 5 ~ 10 uC for each RBS, PIXE spectrum and 1
uC for the NRA spectrum was accumulated in a single angular scan. The beam spot size was 1.5
mm x 1.5 mm with divergence less than 1 mrad (0.06°).
RESULTS AND DISCUSSION
Channeling angular scans were performed in <0001> and < lOT 1 > axial directions. Tilts
through <0001> and < 1011 > axes were made at angles of 10° and 26°, respectively, with
respect to the {1120} plane to eliminate contribution of planer channeling at large tilt angles.
Thus, we obtained good reproducibility of channeling distributions. Figure 4 shows the
channeling angular distributions of an as-implanted sample. Channeling yields were normalized to
the random yield which was obtained for each angular scan. The angular distributions of Ca have
large dips even in the as-implanted sample in both the <0001> and the < 1011 > directions as
shown in Fig. 4. Table I summarizes Xmin of the as-implanted and the annealed samples. The
fraction of impurity in substitutional site (fsub) can be obtained by the following equation [11].
J sub
e»h
1-^(impurity)
l-Zn^nost)
379
We obtained fsub of 0.86, 0.85 for the <0001>, < 1011 > , respectively. The fact that a large
amount of Ca is already substitutional even in the as-implanted sample, suggests that dynamic
annealing takes place effectively in GaN even for low temperature ion implantations [5, 12].
Figure 5 shows the channeling angular distribution of the annealed sample at 1100°C. The
calculated fsub's are 0.82,0.83 for the <0001>, < 1011 > , respectively, indicating that
substitutionality of Ca in the annealed sample is almost same as that in the as-implanted sample.
The half angle of channeling dip can be expressed by the following equation [11],
Z1Z2
V„
where Z\ is the atomic number of incident atom, Z2 is the average number of the constituent atoms
in the row and E is energy of incident atom. Since the < 1011 > atomic row of GaN consists of
pure Ga and pure N rows, the half angle of Ga and N in the < 1011 > are different. Therefore, it
is possible to determine whether the Ca is in Ga or N sites by comparing half angles of Ca, Ga
and N in the < 1011 > direction. Table II shows a summary of V|/i/2 values. The ym's were
obtained from curve fitting procedures using a three-dimensional spline function. The accuracy of
the half width measurement was estimated to be 0.06° based on reproducibility. We corrected the
\|/l/2's of N obtained with 3.2 MeV protons to correspond to a 3 MeV incident He beam using the
above equation so as that we can compare \|/i/2's. For the annealed sample, the i|f 1/2 of Ca in the
< lOll > (0.49°) is almost the same as that of Ga (0.51°) and different from that of N (0.19°).
Therefore, we conclude that most of the substitutional Ca is in the Ga site in the annealed sample.
1?
1
'
'
1
'
'
'
• • T ' ':
1
.«•
0.8
" {• '
^ '•
-.
:
hi
0.4
'..-.-
'
0.8
0.6
;
1 * '"'' ' '
1
"Hi
r~&^-»^*,:
:
-
:
—• —Ga :
......... N :
0.2 - (b)<1011>
0
1
-1
1
0.4 :
-Ga :
• N 0
Ca :
0.2 . (a)<0O0I>
n
........
" r~~?~^'K -«*.
:
*
1
' •
X v
0.6
1.2
0
-1
0
1
tilt angle (deg.)
0
tilt angle (deg.)
Ca
'
1
Fig. 4 Channeling angular distribution of the as-implanted sample along (a) the <0001> and
(b) the < 1011 > axial direction.
"-T '
' I'
'
1.2
—• "' '" 1 ' ' ' ' 1
I
?"
**"**"* "V)\
/
\\
>
8
•a
*
0.8
~
0.4
0.2 :
0
9
\
06
(a) <0001>
•
<
1
^~~~s
0.8
f 1/
\J
.
.* • *• **..
.
7-"-!
''
T
1
1
1
1
1
1
1
1
- '^Ci"-"'
1
1
I
1
1
D
':
i
;
1
—• —Ga
......... N
0 Ca
:
0.2
....!..
0
- (b)<10ll>
1
Tv
1
-1
tilt angle
igle'(deg.)
-
A
0.4
......... N ;
;
*s[ '** r • "' K
\ * ' if
0.6
—*— Ga
lilt
4"i>
....
1
0
tilt angle (deg.)
Fig. 5 Channeling angular distribution of the annealed sample at 1100°C along (a) the
<0001> and (b) the < 1011 > axial direction.
380
,-
Table I. Summary of minimum channeling yield, %min.
Element
annealed
as-implanted
<0001>
<ioii>
<0001>
<1011>
0.18
0.34
0.20
0.25
Ga
0.04
0.22
0.02
0.10
N
0.08
0.66
0.02
0.34
Ca
Table n. Summary of half angle of channeling dip, \|/i/2 (degree). Corresponding
values to 3 MeV He incident beam are shown in parentheses for N.
Element
annealed
as -implanted
<0001>
<ioii>
<0001>
<ioii>
Ca
0.47
0.31
0.68
0.49
Ga
0.61
0.50
0.67
0.51
N
0.49 (0.72)
0.10(0.15)
0.50 (0.73)
0.13(0.19)
On the other hand, for the as-hnplanted sample, the \|/i/2's of Ca are narrower than those of Ga
in both the <0001> and the < 1011 > directions. This suggests that Ca is slightly displaced from
the Ga sites in the as-implanted sample. The yia in the < 1011 > direction would also be
narrowed if some fraction of the Ca is in the N site. However, the \\f\/2 in the <0001> direction
would not be narrowed. Since_our data show that the yfm in the <0001> direction is also
narrowed similarly to the < 1011 > direction, we conclude that Ca atoms sit near the Ga sites but
are slightly displaced. The projected displacement of impurity atoms from the host atoms (rx) is
related to the ym's by the following equation [11].
^-.(impurity)
VI/2(host)
{InKCg/rj'+l]}"2
{ln[(Ca/p)2 + l]}"2
C is a constant of magnitude -1.73, a is a screening distance and p is thermal vibration
amplitude. For GaN, a is ~ 015 Ä, p is ~ 0.10 Ä [13]. We obtained rx ~ 0.17 + 0.05 Ä in the
<0001> and ~ 0.24 + 0.05 A in the < 1011 > directions, respectively, using the measured V|/i/2!s
of Ca and Ga for the as-implanted sample. This equation assumes an ordered lattice. We
measured i|f 1/2's of an unimplanted GaN sample and found that the \|/i/2's of the unimplanted
sample and that of the as-implanted sample are consistent within error (0.06°). Therefore,
contribution of the disorder in the as-implanted sample to the calculation is not serious.
From our results and electrical properties reported previously [3], in which Ca implanted GaN
becomes p-type from n-type at an annealing temperature of 1100°C, we suggest the mechanism of
electrical conductivity of Ca implanted GaN as follows:
(i) Ca is compensated electrically in the as-implanted GaN due to formation of complex defects
with implantation induced donor like point defects, such as nitrogen vacancy (VN) and/or
antisite Ga (Gau), resulting in a complex in which Ca atoms are slightly displaced from exact
Ga sites. GaN remains n-type due to residual donor like point defects.
381
(ii) After annealing at 1100°C, these complex defects are broken and donor like point defects are
annihilated or diffuse away. After annealing, substitutional Caca works effectively as an
acceptor, and GaN becomes p-type.
CONCLUSION
We have demonstrated the lattice location of Ca in ion implanted GaN with post-implant
annealing using ion channeling. More than 80 % of Ca are near the Ga sites even in the asimplanted sample, however, they are slightly displaced from the Ga sites. Ca goes to the exact Ga
sites after annealing at 1100°C. We think that Ca is electrically compensated due to formation of
complex defects with donor like point defects, such as Caoa-VN and/or CaQa-GaN in the asimplanted sample. Caoa becomes electrically active when these complex defects are broken and
the point defects diffuse away with annealing at 1100°C.
ACKNOWLEDGEMENTS
We would like to thank T. Asatsuma for the GaN deposition. This work was supported by
Research Center, Sony Corporation.
REFERENCES
[I] J. C. Zolper and R. J. Shul, MRS Bull. 22, 36 (1997)
[2] M. S. Shur and M. A. Khan MRS Bull. 22, 44 (1997)
[3] J. C. Zolper, R. G. Wilson, S. J. Pearton, and R. A. Stall, Appl. Phys. Lett. 68, 1945
(1996)
[4] S. J. Pearton, C. B. Vartuli, J. C. Zolper, C. Yuan, and R. A. Stall, Appl. Phys. Lett. 67,
1435 (1995)
[5] H. H. Tan, J. S. Williams, J. Zou, D. J. H. Cockayne, S. J. Pearton, and R. A. Stall,
Appl. Phys. Lett. 69, 2364 (1996)
[6] H. H. Tan, J. S. Williams, J. Zou, D. J. H. Cockayne, S. J. Pearton, J. C. Zolper, and R.
A. Stall, Appl. Phys. Lett. 72, 1190 (1998)
[7] H. Kobayashi and W. M. Gibson, Appl. Phys. Lett. 73, 1406 (1998)
[8] J. F. Ziegler, J. P. Biersack and U. Littmark, Stopping and Ranges of Ions in Matter
(Pergamon, New York, 1988), Vol. I
[9] X. A. Cao, C. R. Abernathy, R. K. Singh, S. J. Pearton, M. Fu, V. Sarvepalli, J. A.
Sekhar, J. C. Zolper, D. J. Rieger, J. Han, T. J. Drummond, R. J. Shul, and R. G.
Wilson, Appl. Phys. Lett. 73, 229 (1998)
[10] S. Bashkin, R. R. Carlson, and R. A. Douglas, Phys. Rev. 114, 1552 (1959)
[II] L. C. Feldman, J. W. Mayer, and S. T. Picraux, Materials Analysis by Ion Channeling
(Academic Press, New York, 1982)
[12] C. Liu, B. Mensching, K Volz, and B. Rauschenbach, Appl. Phys. Lett. 71, 2313 (1997)
[13] J. R. Tesmer and M. Nastasi, Handbook of Modern Ion Beam Materials Analysis
(Materials Research Society, Pittsburgh, 1995)
382
GROWTH AND CHARACTERIZATION OF InGaN/GaN
HETEROSTRUCTURES USING PLASMA-ASSISTED MOLECULAR BEAM
EPITAXY
K. H. Shim, S.E. Hong, K.H. Kim, M. C. Paek, and K.I. Cho
Wide Band Gap Semiconductor Team, Microelectronics Technology Research
Laboratory, Electronics and Telecommunications Research Institute
161 Kajung-Dong, Yusong-Gu, Taejon, Korea 305-350
Structural and optical properties of In,, 2Gao.8N/GaN heterostructures grown by
plasma-assisted molecular beam epitaxy have been investigated as a function of rf
plasma power. Indium incorporation resulted in the higher rf power level suppressing
3D island growth with reduced introduction of defects in In^GaouN in comparison with
GaN. Sharp morphology at interfaces and strong transitions in photoluminescence
reveal the optimum rf power around 400 W in our experimental set up for the growth of
In(UGa0.gN/GaN heterostructures. Our experimental observations suggest that the
presence of indium on surface modulates the rate of plasma stimulated desorption and
diffusion, and reduces the formation of damaged subsurface.
INTRODUCTION
Recently, lots of achievements have been accomplished in the field of GaNbased nitride semiconductors by remarkable breakthroughs in epitaxial growth
technology [1]. Plasma assisted molecular beam epitaxy (PAMBE) has demonstrated
useful results along with its advantageous features such as low temperature processes
and atomic layer growth. However, the development of GaN-based epilayers using
PAMBE has been hindered by uncertainty about the effect of the plasma parameters on
film quality and inadequate control of the energetic species in the nitrogen plasma.
Meanwhile, the application of very low energy ion beams of a few tens of eV have
focused on the modulation of surface reaction kinetics, strain relaxation, and island
formation [2,3]. Despite the relevant role of plasma parameters, relatively few articles
have focused on the study of plasma parameters in GaN epilayers using PAMBE [4,5].
A fundamental understanding of the interactions of energetic particles with the surface
of GaN-based epi layers would enable the tradeoff between damage production and
surface enhancement.
The objective of this study is to understand the effect of the rf power on the
383
Mat. Res. Soc. Symp. Proc. Vol. 572 e 1999 Materials Research Society
growth of GaN-based epilayers by exploring correlations between the structural/optical
properties and the rf plasma power. Our experimental results on In^GaouN/GaN
superlattice (SL) structures provide a framework for understanding the complicated role
of plasma parameters on the growth of GaN-based heterostructures.
EXPERIMENT
GaN-based heterostructures were grown using PAMBE equipped with an
inductively coupled rf plasma source (SVT Associates) [6]. After the conventional
solvent cleaning and etching processes, 20-nm-thick A1N buffer layers were grown on
sapphire (0001) at 500°C. Then GaN epilayers were grown at 720°C and followed by
the growth often periods of In^GaogN/GaNpS/SOÄ) at 670°C. The thickness of InGaN
was designed to be smaller than the critical thickness, 7.6 ran, estimated from a
Matthew and Blakeslee model. The growth chamber maintained 7xl0"5 torr for whole
growth process. Langmuir probe technique was employed to analyze the distribution of
energy and flux of nitrogen ions. The surface structure of epilayer was monitored by insitu reflection of high-energy diffraction (RHEED) and the quality of obtained epilayers
was characterized by atomic force microscopy (AFM), cross-sectional transmission
electron microscopy (XTEM), and low-temperature photoluminescence (PL).
-O-GsN
-• -InGaN (X,„=0.2)
et
a
3
—'0
600
RF Input Power (W)
Fig. 1. Growth rate and roughness data plotted as a function of rf input powers
for GaN, open circles, and InojGaosN, closed circles.
384
RESULTS AND DISCUSSION
Fig. 1 represents the changes in both growth rate and root mean square (rms)
roughness as a function of the input rf power of PAMBE. A typical, panoramic feature
of PAMBE is well identified from the data in the figure 1. The decrease in growth rate
at high rf power indicates that the flux of reactive nitrogen is sufficient at around 300 W.
The surface roughness shows a significant decrease at low rf power regime and
decreases monotonously at high rf power regime above 300 W. The changes in rms
roughness revealed the evolution of surface morphology, corresponding to in-situ
RHEED observation. However, Two distinctive features are noticeable in the InGaN
growth: its growth rate saturates to 145 ran/hr at high rf power and its roughness is
roughly two folds of GaN at low rf power around 200W. The origins causing such
differences are discussed below in detail.
A
A
f
^*0
A
i2-
InGaN
c \
*
/
±
1
rf Power
a: 200 W
b: 300 W
c:400W
d: 500 W
A
1
•
1»
1M
«40
t*6
4M
0
I
t>owt< (VI a*%)
/b\
GaN (3 464 eV)
/
'
/3\
z^ir-*—
1
....
1
400
450
Wavelength (nm)
Fig. 2. LT-PL spectra measured at 13 K from InGaN/GaN superlattice
samples grown with the rf input powers of 200, 300,400, and 500 W.
In an effort to investigate the effect of rf power on optical properties,
Irio^GaogN /GaN SLs were prepared using different rf powers of 200, 300, 400, and 500
W. Shown Fig. 2 represents PL spectra for the In^Ga^N/GaN samples, where the
inset is a plot of energy and intensity of InGaN peaks. The spectrum c reveals optical
properties much better than the others in terms of high emission efficiency and the peak
385
position, 2.8 eV, of InGaN quantum wells corresponding to the energy levels of
electrons and holes. The strong PL emission spectra were observed at around 300W for
GaN [6]. Sharp and strong XRD peaks were observed from highly crystalline GaN
epilayers grown with 300-400 W. The increased energy and flux of nitrogen ions led
to a preferred orientation structure and that the 2D planar growth enhanced at high rf
power. The optimum rf power looks increased by -100 W in InGaN.
Shown Figs 3 (a) and (b) are XTEM images taken from the sample grown with
400 W. Many angled-dislocations, noted as D, originate from threading dislocations at
interfaces and facets of pyramid shape. We observed that threading dislocations with a
Burger's vector of g=<ll-20>/3 exist as much as 10" cm"2 at near to the interface, and
that the dislocation density was tremendously decreased to ~7xl 08 cm"2 at 0.4-1 um.
The lattice constants of GaN were measured to be very close to that of bulk
GaN, c=5.183 A. The lattice constant of In^GaogN is under the compressive strain of
1.5%, although the original strain 2% between In^GaogN and GaN. According to TEM
observations, dislocations were not newly generated at the interface of InGaN and GaN.
This confirms that the relaxation in InGaN occurred partially in the limit of elastic
deformation.
Fig. 3. (a) XTEM [11-20] images of InGaN/GaN SLs grown on sapphire
substrates and (b) HR-XTEM image of the rectangle in (a). The micrographs were taken
with g=0002 along [11-20] zone axis of hexagonal GaN. HR-XTEM image in (b) shows
the sharp interface of InGaN/GaN with on dislocations newly generated due to the
relaxation of lattice mismatch stress. The inset in (a) is the TED pattern showing spots
from the sapphire substrate and the GaN epilayer.
386
Among various process regimes of ion bombardments, distinguished as a
function of ion energy and ion/atom flux ratio, the operational conditions of PAMBE is
located at a boundary dividing surface desorption and on-effect [7,8]. The previous data
involving energetic particles often reveal discrepancies, which are particularly subject to
the substrate materials, reactive particles, and plasma parameters affecting the energy
and flux of ions. The enhanced diffusion of indium atoms on InGaN surface looks
predominant at low rf power operation and resulted in very rough surface as shown in
Fig.l. Therefore, the change caused by indium incorporation is attributed to the high
surface diffusion and the large mass of indium.
The role of energetic particles in PAMBE can be understood by analyzing the
changes in growth mode, which appears at the initial stage of thin film growth. The 2Dto-3D transition is unavoidable for the growth of heterostructures with large lattice
mismatch between adjacent layers. The effects of ion-induced displacement on the
nucleation rate can be estimated by adding a term for the breakup of the critical size
nuclei. The forward reaction of nucleation which occurs by the direct impingement of
Ga and the diffusion of single adatoms to critical-sized clusters is rate limited by
thermal and plasma-stimulated desorption. Thus, the 3D nucleation depends mostly on
the energy and flux of energetic particles at a certain temperature.
The optical properties of epilayers would be more likely depending on the ion
energy. The displacement energy of bulk atoms is usually approximated as three times
of the cohesive energy. By employing critical energies for the displacement of atoms at
subsurface, the threshold energy of particles for defect formation at the subsurface are
approximated as 16.5, 15, and 17 eV for A1N, GaN and InN, respectively [8].
Considering the Boltzman distribution with a dispersion of ~5 eV, the average ion
energy less than -14 eV looks safe for the growth of InN films without collisioninduced defects. The rf powers ejecting ions with an average energy of 15 eV are 300
and 400 W for 1 and 2 seem nitrogen gas flow, respectively. Consequently, the strong
PL emission in Fig. 2 could be achieved from the InGaN/GaN grown with 400W. By
raising the rf power from 400 W to 500 W, the plasma source becomes to supply ions
with the energy -21 eV and the flux increased by 30 %. The catastrophic degradation of
the spectrum d in Fig. 2 presents that the plasma rf power must be controlled below 500
W.
The defects produced by ion bombardment may be annihilated at a higher
temperature. However, the substrate temperature must be controlled below a certain
limit to prevent surface decomposition. Nevertheless, our observations suggest that the
387
energetic particles ejecting from a plasma source need to be removed completely or
controlled properly for a specific application like the manipulation of surface
roughening accompanied by stress relaxation.
CONCLUSIONS
From correlations between rf plasma power and the evolution of structural and
optical properties of In,^Ga^N/GaN SLs, the incorporation of indium was observed to
reduce the formation of damages, and correspondingly lead to enhance band-to-band
optical transitions. It was important to control the rf power level below 400 W in order
to suppress surface roughening without introducing defects at the subsurface of
In^GaogN epilayers. The experimental results associated with indium incorporation
could be explained in terms of the inelastic collision of energetic particles, stimulating
kinetic reaction processes at surface as well as the creation of defects at subsurface.
ACKNOWLEDGMENTS
The support of the Ministry of Information and Communications of Korea is
gratefully acknowledged. This work was partially initiated during Ph.D. research at
University of Illinois at Urbana-Champaign. Authors thank J.B. Park at KMAC for his
TEM analyses and Dr. C.S. Lee at KRISS for HR-XRD measurements.
REFERENCES
1. S. Nakamura, M. Senoh, S. Nagahama, N. Iwasa, T. Yamada, T. Matsushita, H.
Kiyoku, Y. Sugimoto, T. Kozaki, H. Umemoto, M. Sano, and K. Chocho, Appl.
Phys. Lett. 72, 211(1998).
2. C.J. Tsai, H.A. Atwater, and T. Vreeland, Appl. Phys. Lett. 57, 2305 (1990).
3. S.A. Barnett, C.H. Choi, and R. Kaspi, Mat. Res. Soc. Symp. Proc. 201 (1991).
4. M.S.H. Leung, R. Klockenbrink, C. Kisielowski, H. Fujii, J. Kruger, Sudkier G.S.,
A. Anders, Z. Liliental-Weber, M. Rudin, and E.R. Weber, Mater. Res. Soc. Symp.
Proc. 449, 221 (1997).
5. H. Fujii, C. Kisielowski, J. Krueger, M.S. Leung, R. Klockenbrink, M. Rubin, and
E.R. Weber, Mat. Res.Soc. Symp. Proc. 449, 227 (1997).
6. K.H. Shim, PhD thesis, University of Illinois at Urbana-Champaign, 1997.
7. M.V.R. Murty, H.A. Atwater, A.J. Kellock, and J.E.E. Baglin, Appl. Phys. Lett. 62,
2566(1993).
8. J.M.E. Harper, J.J.Cuomo, and H.T.G.Hentzell, Appl. Phys. Lett. 43, 547 (1983).
388
PIEZOELECTRIC COEFFICIENTS OF ALUMINUM NITRIDE
AND GALLIUM NITRIDE
CM. Lueng*, H.L.W. Chan*, W.K. Fong**, C. Suiya**, C.L. Choy'
Department of Applied Physics and Materials Research Center,
The Hong Kong Polytechnic University, Hung Horn, Kowloon, Hong Kong.
"" Department of Electronic Engineering,
The Hong Kong Polytechnic University, Hung Horn, Kowloon, Hong Kong.
ABSTRACT
Aluminum nitride (A1N) and gallium nitride (GaN) thin films have potential uses in high
temperature, high frequency (e.g. microwave) acoustic devices. In this work, the piezoelectric
coefficients of wurtzite A1N and GaN/AIN composite film grown on silicon substrates by
molecular beam epitaxy were measured by a Mach-Zehnder type heterodyne interferometer. The
effects of the substrate on the measured coefficients are discussed.
INTRODUCTION
Aluminum nitride (A1N) and gallium nitride (GaN) are III-V nitrides and the reported
lattice parameters of A1N and GaN with wurtzite structure are: a = 3.11 A, c = 4.98 A for A1N
[1] and a = 3.189 A, c = 5.185 A for GaN [1]. A1N and GaN have wide direct bandgaps of 6.2 [2]
and 3.39 eV [3], repectively. Both of them have potential applications in devices working in high
temperature and hostile environments [4]. Many different growth techniques have been used to
prepare A1N and GaN films and molecular beam epitaxy (MBE) is one of the techniques that can
produce high-quality epitaxial A1N and GaN films. Due to its superior properties, research in the
physical properties and applications of A1N and GaN has attracted considerable interest [2].
However, to date there appears to be limited data on the piezoelectric coefficients of A1N as well
as GaN. These parameters are important since both of them have potential use in microactuators,
microwave acoustic and other microelectromechanical (MEM) devices [5].
EXPERIMENT
A1N film grown by MBE
A1N thin films were grown by using a SVTA BLT-N35 MBE system. The nitrogen
molecules in a nitrogen gas stream were broken into atomic nitrogen by a SVTA RF plasma
source operating at 13.56 MHz. Aluminum was evaporated by the K-cells. The atomic nitrogen
and aluminum react to form A1N on a (111) silicon substrate.
The silicon substrate was cleaned by a degreasing process and etched in buffered HF.
After the substrate had been transferred into the deposition chamber, the substrate surface was
thermally cleaned at 940 °C for 1 hour. The substrate temperature was then decreased to about
600 °C and A1N thin film started to grow. The final thickness of the A1N film was about 450nm.
Sample geometry
In this experiment, two samples were studied. One is a 450nm thick A1N thin film while
the other is a composite film, comprising a 140nm thick GaN film grown on a 30nm thick A1N
389
Mat. Res. Soc. Symp. Proc. Vol. 572 e 1999 Materials Research Society
buffer layer. Both of the films were grown on silicon substrates with (111) orientation. Figures 1
and 2 show the X-ray diffraction (XRD) patterns of the A1N and the composite film, respectively.
The peak at 35.9° corresponds to the (0002) reflection of wurtzite A1N [6]. Another peak at 34.6°
corresponds to the (0002) reflection of wurtzite GaN [7]. The XRD patterns confirmed that the
films have the wurtzite structure, and the c axis is oriented along the normal of the substrate.
Si(U,)
SKI 11)
D
S-
&
c
I
c
1 AN(0002)
c
o
o
!5
b
25
1
V .
1
30
35
Two Theta (deg)
•
co
o
&
b
40
GaN(0002)
-
ni
'
25
I
,J
/ AIN(0002)
\\
30
35
Two Theta (deg)
40
Figure 2: The XRD pattern of the composite film.
Figure 1: The XRD pattern of the A1N film.
Circular dots of aluminum of diameter 1mm were thermally evaporated at different
positions on the top surface of the films. These dots act as top electrodes as well as mirrors to
reflect the laser beam from the interferometer. The silicon substrate was attached to an aluminum
block (connected to ground) fixed on a translation stage. An a.c. electric field was applied across
the top electrode and the aluminum block. The thickness of the film changes due to the converse
piezoelectric effect and the small displacement was measured using a Mach-Zehnder type
heterodyne interferometer.
Measurement of rf?? using laser interferometry
Figure 3 shows the schematic diagram of the Mach-Zehnder type heterodyne
interferometer (SH-120 from B. M. Industries, France) used to measure the displacement. A
linearly polarized laser beam, L (frequency fc, wave number k=27i/X, X = 632.8 nm for a He-Ne
laser) is split into a reference beam, R, and a probe beam, P, by a beam splitter (BS). R
propagates through a Dove prism and a polarizing beam splitter (PBS) into a photodiode. The
frequency off is shifted by a frequency fB (70 MHz) in a Bragg cell, and then this beam (now
labeled S), is phase modulated by the surface displacement of the film sample, x = ucosftrfj) ,
where/„ is the vibration frequency and u is the displacement amplitude. For small u, only the
sideband at_/ä +fu is detected and its amplitude is
J,(47M/X)/J0(4TM/X)~2TM/X
= it/1007
(1)
where J„ and // are the Bessel function of the zeroth and the first order, respectively. The ratio of
amplitudes of the zeroth order (centerband) to the first order (sideband) of the Bessel function
gives the absolute displacement of the sample surface. The ratio, R '= Ji(4mt/X)/J0(4mi/X) in
dBm can be measured using a spectrum analyzer (HP3589). Let R=l(f720, the vibration
displacement is [8]
390
u=1007*R'
(2)
The rfjj' coefficient (strain/applied field) of the film samples can be calculated as :
d33'= (u/t)/(V/t) = u/V
(3)
where t is the thickness of the sample and Kis the voltage applied across the film sample. Kwas
measured using an oscilloscope (Fig. 3) with a 50 O. termination connected across the sample so
as to ensure that the change in sample impedance with frequency does not cause a change in the
voltage.
i
i
Dove Prisn
!
QuarterwGv
Plate
BS
Cell
s v
Sample
Focusing1
Lens
Analyzer
Photodetec tor
-^3-
Signal
Processor
He-Ne Laser
Heterodyne Interferoneter
spectrun Analyze
Figure 3: A Mach-Zehnder type heterodyne interferometer.
Sample mounting
When measuring vibration displacement using an interferometer, it is important to mount
the sample properly. If the sample is not rigidly mounted on the aluminum block, other modes of
vibration such as the bending mode may occur. The bending mode may have a vibration
amplitude an order of magnitude larger than that of the thickness mode, hence a large error will
result if the bending mode is also excited. One way to eliminate the bending effect is to reduce
the size of the electrode and to glue the substrate to a rigid holder [9]. To ensure that no bending
mode was present, the probe beam was scanned across the sample surface. As mentioned before,
aluminum dots of diameter 1mm were deposited at different positions on the top surface of the
sample. Measurements of the displacement amplitudes at these different positions gave
essentially the same results, indicating that no bending mode was excited. As the electrode
diameter was 1mm, the probe beam (about lOOum diameter) was also scanned across an Al dot
to test whether bending vibration was excited within the 1mm dot. Again, constant vibration
amplitudes are observed indicating that the dominant vibration is the thickness mode [10].
Substrate Clamping Effect
The constitutive equation for a piezoelectric material is [11]
391
St=sJJ+dklEk
ij = l,...,6
* = 1,...,3
(4)
where 5, is the strain, 7} is the stress, s0 is the compliance, du is the piezoelectric coefficient and
Ek is the applied electric field. When a thin piezoelectric film is grown on a thick substrate, the
film is rigidly attached to the substrate and thus cannot vibrate freely. Hence, the measured
coefficient cannot represent the true value for the film. However, it is possible to correct for the
clamping effect, and the corrective factor depends on the symmetry and the orientation of the
piezoelectric layer [12].
Since the film is clamped by a substrate, it cannot freely expand or contract in the
interface. The assumption that strain Si, S2 and S6 are equal to zero holds if the measurements are
carried out at frequencies below the resonance frequencies of the film. If the top surface of the
film is assumed to be free, T3, T4 and Ts axe. zero on the free surface. For crystals with 6mm
symmetry wurtzite structure [11], it is found that
T,=T,2
-d,.E
31^3
~~
(5)
F
From the constitutive equation, we obtain
S,=2s^T1+di3E3
(6)
The measured (apparent) piezoelectric coefficient is d33 -Sj/E;, which can be related to
the true coefficient d33 through Eq. (5) and (6):
„£
S^
= rf33-2(4^V)
(7)
£3
Therefore, the true d33 coefficient differs from the apparent coefficient by a corrective
factor (the second term).
RESULTS
Measurements were made at the center of a 1mm diameter aluminum electrode located
close to the center of the film. Figures 4 & 5 show the variation of the displacement with driving
voltage at 10 kHz. From the slope of the line, the d33 coefficients of the A1N and GaN/AIN
composite film are found to be 3.9 pm/V and 2.65 pm/V, respectively. The correlation
coefficients of the best fit straight lines for the A1N and the composite film are 0.99475 and
0.99651, respectively. The correction factor for A1N was calculated by using equation (7) and the
compliance and d3i value given in ref. [13] and [14], respectively. The correction factor for A1N
is 1.2 pm/V, hence the true piezoelectric coefficient d33 of A1N film is 5.1 pm/V. Figures 6 & 7
show the variation of the measured d33' with frequency. It is seen that the value is approximately
constant over the full range except at frequencies near 70 kHz for the A1N sample and 50 kHz for
the GaN/AIN composite sample, where bending mode vibrations may have been excited. As it is
difficult to estimate the substrate clamping effect in the GaN/AIN composite film, we are in a
392
process of growing GaN films directly onto Si substrates. The results on these GaN films will be
reported in the near future.
0
12
3
4
5
4
6
Applied Voltage [V]
Applied Voltage [V]
Figure 4: Variation of displacement with driving
Figure 5: Variation of displacement with driving voltage
at 10 kHz for the GaN/AIN composite sample.
voltage at 10 kHz for the A1N sample.
Frequency [kHz]
Frequency [kHz]
Figure 7: Variation of d33' with frequency for
the GaN/AIN composite film sample.
Figure 6: Variation of dsl' with frequency for
the A1N sample.
CONCLUSIONS
In summary, the apparent piezoelectric coefficients rfjj' of A1N and GaN/AIN composite
films grown on Si substrates were found by a laser interferometric method to be 3.9 pm/V and
2.65 pm/V, respectively. The ds3' value of A1N film is approximately the same as that ( 4.0
pm/V) reported for A1N films grown by chemical vapour deposition [15]. By using a correction
factor, the true rf» coefficient of A1N was found to be 5.1 pm/V and this value is similar to that (
5.0 pm/V) reported in ref. [14].
REFERENCES
[1] Michael S. Shur and M. Asif Khan, Mat. Res. Bull., 22 (2), 44 (1997).
[2] S. Strite and H. Morkoc, J. Vac. Sei. Technol, BIO, 1237 (1992).
[3] H. P. Maruska and J. J. Tietjen, Appl. Phys. Lett., 59, 327 (1969).
393
[4] A. D. Bykhovski, V. V. Kaminski, M. S. Shur, Q. C. Chen and M. A. Khan, Appl.
Phys. Lett., 68, 818(1996).
[5] S. N. Mohammad, Arnel A. Salvador and Hadis Morkoc, Proc. of the IEEE, 83,
1306 (1995).
[6] Michihiro Miyauchi, Yukari Ishikawa and Noriyoshi Shibata, Jpn. J. Appl. Phys.,
31,L1714(1992).
[7] K. Kubota, Y. Kobayashi, and K. Fujimoto, J. Appl. Phys., 66,2984 (1989).
[8] Z. Zhao, H. L. W. Chan and C. L. Choy, Ferroelectrics, 195, 35 (1997).
[9] S. Muensit and I. L. Guy, Appl. Phys. Lett. 72, 1896 (1998).
[10] A. L. Kholkin, Ch. Wütchrich, D. V. Taylor and N. Setter, Rev. Sei. Instrum., 67,
1935 (1996).
[11] B. A. Auld, Acoustic Fields and Waves in Solid Volume I, 2" ed. (Krieger Publishing
Company, Malabar, Florida, 1990). p. 271, 370, 381.
[12] D. Royer and V. Kmetik, Electronics Letters, 28,1828 (1992).
[13] Lauroe E. McNeil, J. Am. Ceram. Soc, 76, 1132 (1993).
[14] Tsubouchi,K., Sugai,K., and Mikoshiba,N., Proc. IEEE Ultrason. Symp., 375 (1981).
[15] Supasarote Muensit, PhD thesis, Macquarie University, 1998.
394
FAST AND SLOW UV-PHOTORESPONSE IN n-TYPE GaN
R. ROCHA1, S. KOYNOV1, P. BROGUEIRA1, R. SCHWARZ*, V. CHU2, M. TOPF3, D. MEISTER3,
and B.K. MEYER3
(1) Physics Department, Institute Superior Teenico, P-1096 Lisbon, Portugal
(2) Institute de Engenharia de Sistemas e Computadores, P-1000 Lisbon, Portugal
(3) I. Physics Department, University of Giessen, D-35392 Giessen, Germany
ABSTRACT
The photocurrent decay in n-type GaN films prepared by low-pressure chemical vapor deposition
(LPCVD) was measured in the ms-to-s time range using steady-state UV light and in the us time regime
using short high-power pulses from higher harmonics of a NdrYAG laser. A power law time dependence
is observed with exponents ranging from -0.1 to -0.3, which is an indication of a broad distribution of
trapping states inside the band-gap. Combining Hall effect results and the magnitude of the initial slope of
the photocurrent decay we estimate a mobility-lifetime product of 2.1X10-4 cm2/V for photogenerated
electrons at times below a few us. Slow transients might be a handicap for applications of GaN in UV
detectors.
1. INTRODUCTION
Applications of GaN-related materials in devices like UV-detectors or high-temperature
transistors might be handicapped by slow current transients, as seen, for example, in the persistent
photoconductivity effect (PPC) [1]. The PPC effect, which manifest itself as a prolonged non-exponential
conductivity decay after exposure to sub-bandgap light, was attributed to internal strain or to charging
and discharging of deep traps in AlGaN alloy films. Metastable changes in conductivity were observed up
to several thousand seconds [2].
The samples we used for this study were prepared by low-pressure chemical vapor deposition
without intentional doping. Hall measurements show that the carrier concentration is about 1018 cm*3 and
the samples are n-type. Therefore the measured photocurrents were well below the dark current levels.
Nevertheless, effects similar to those encountered in highly resistive AlGaN alloy films occur, like
frequency-dependent photoconductivity and non-exponential current decay after pulsed excitation. We
discuss whether both the slow (s) and the fast (us) photocurrent decay can be described by changes of
occupation of deep traps.
2. SLOW PHOTOCURRENT DECAY WITH CHOPPED LIGHT
The GaN samples were prepared by low-pressure chemical vapor deposition (LPCVD) from
GaCl3 and NH3 sources on AI2O3 substrates [3]. The deposition temperature and process pressure were
965 °C and 0.5 mbar, respectively. GaCl3 was transported in a 50 seem N2 carrier gas heated to 70 °C.
The NH3 flow was set at 100 seem. This resulted in a growth rate of about 3 A/s.
395
Mat. Res. Soc. Symp. Proc. Vol. 572 ® 1999 Materials Research Society
Photocurrent decays were measured with coplanar Cr contacts. The 1 mm gap was illuminated
either with chopped light from a 50 W Xe-lamp or with 5 ns pulses of the 532 run line of a frequencydoubled Nd:YAG laser, or the fourth harmonic line at 266 nm. The applied voltage was kept between 1
and 3 V where ohmic behavior of the contacts was assured. The relatively large dark currents were
compensated for electronically since the photocurrents were below the percent level.
^
d
250
" ' I
200
.
150
.
.
CD
J.
10
°
n-typeGaN
•
B
y00^
I
r
%
J
L
^k
-
^^
Sample T10
XeJamp
12.5 K:
I
^S*J
I.I.
0.00
0.02
Fig. 1:
Modulated photocurrent
with chopped UV lightfrom
a 50 WXe lamp. The initial
decay time constant is 2.1
1
\
0.04
t
i
0.06
0.08
0.10
Time(s)
10
8
'
"i
i
-
Sample T10
Xe-lamp
.
/
1
1
4
• '
n-type GaN
I....,- -1
TO
r
1
d
—'
j—1—,—,—
12
,
2 -
Fig. 2:
Frequency dependence of
the peak-to-peak value of
the photocurrent with
chopped UV light.
1
•
100
200
i
i
300
400
500
Frequency (Hz)
As shown in the 12.5 Hz case in Fig. 1 the photocurrent never reaches a constant level due to the
1:1 duty cycle. Therefore, this experiment is not expected to reveal the very large time constants
encountered in the PPC effect [2,4]. The initial decay time (response time) is about 2.1 ms in Fig. 1. Even
though no saturation is reached one can estimate the exponent in a power law ansatz to the onset of the
current decay to be about -0.1. We obtain the curve of Fig. 2 when the peak-to-peak photocurrent value is
plotted as a function of the chopping frequency.
The complementary approach to the frequency domain is described in the next paragraph.
However, the time domain studied below does not cover the times measured in this section. Also, we
have to keep in mind that in the pulsed mode the average light bias, and the concomitant background
carrier density, will be much lower. This could increase the photocurrent decay rate, i.e. reduce the
396
observed response time, if it were governed by capture of free carriers in available shallow and deep
traps.
3. FAST DECAY AFTER PULSED PHOTOEXCITATION
The initial photocurrents were of the order of the dark currents under pulsed excitation with 5 ns
pulses from the 532 nm and 266 nm line of the Nd:YAG laser system. The power densities where about
14 and 5 MW/cm2, respectively. The UV line lies slightly above the bandgap of 3.4 eV of GaN.
However, the results shown in Fig. 3 for the 266 nm line do not differ significantly from the obviously
strongly non-exponential time decay when the 532 nm line of the Nd:YAG laser or the 337 nm line of a
pulsed nitrogen laser were used [5].
GaNonAI203
fest = 1.S6nS
Sample F83-1
266 nm excitation
3
CD
Fig. 3:
Photocurrent decay
after pulsed laser
excitation using 5 ns
pulses at 266 nm
wavelength.
Ü
O 0.1
g
sz
= 21xia"crr?/V
2x10"!
4x10"5
6x10"5
8x10-;
1x10"4
Time [s]
Again, the photocurrent decay is strongly non-exponential with an initial decay time of 1.9 |xs and
a long tail with a time constant of 154 p.s. The fast response time might be limited by the RC constant of
the experimental set-up. The second time constant will depend on the time window chosen, and is,
therefore, somewhat arbitrary. This is a typical feature of dispersive transport where the carrier mobility
decreases with time due to trapped carriers which cannot contribute to band transport [6].
We can try to estimate a majority carrier p.x-product (in our case (ur)e for electrons) if we
combine the initial response time of Fig. 3 with the mobility value of 120 cm2/Vs determined by Hall
measurements. The result is (ur)e = 2.1xl(H cm2/V. This is the "effective" value at about 1 fis, it is
expected to increase with time.
397
4. DISCUSSION
The three different time constants extracted from the photocurrent decays in Fig. 1 and 3 are quite
consistent if we assume a time-dependent carrier drift mobility ftft), as found in a number of disordered
semiconductor materials [6]. This will lead to a power law decay in the photocurrent, Ip/,(t) ~ /_m, with an
exponent m between 0 and 1. From Fig. 1 we find m = 0.1 and from Fig. 3 the range is from 0.27 to 0.33
for different samples and for different excitation wavelengths [5]. This would be compatible with a broad
tail state distribution. Both, the presence of such defect states and the high dark conductivity of the films
might be related to the heteroepitaxial growth of GaN on a foreign substrate (sapphire). Defect states
could also be located at the surfaces and interfaces of crystallites that are seen in the AFM micrographs of
our samples.
Without any further knowledge about the nature of trap states we can give an energy scale for
those states if we transform the above time constants into trap energies using E^ = kT In(vr) , which is
applicable for thermal emission. Under the assumption that electron traps are responsible for the above
photocurrent transients we can sketch the location of the corresponding trap states as shown in Fig. 4,
where we have chosen the phonon frequency v to be 1012 s_1. The energy levels that correspond to 1.9 ps,
154 us, and 2.1 ms are 0.36, 0.47, and 0.54 eV, respectively. A high defect density near the conduction
band (CB) would be responsible for the position of the dark Fermi level EF.
CB
Fig. 4:
Sketch of the energy levels
that correspond to the
observedphotocurrent decay
times, assuming thermal
reemissionfrom electron
traps. The magnitudes cannot
be determinedfrom the
photocurrent decay alone
and are therefore arbitrary.
>.
ro
b.
o
c
LU
VB
'' Log (Density-of-states)
From the n-type nature of the films we know that the photcurrent is carried by electrons.
However, similarly to the case of electrons trapped near the conduction band as sketched in Fig. 4, we
could also imagine that the time constants stem from reemission of holes trapped near the valence band,
which would then recombine with free electrons or, with less probability, shallow-trapped electrons. The
distinction should come from complementary experiments like photocapacitance measurements [7].
It is tempting to apply the Einstein relation to calculate the electron diffusion length Le that would
correspond to the above mentioned value of (pt)e = 2.1xl0"4 cm2/V. We obtain Le = 23 urn with an
effective lifetime of 1.9 ps. This can be compared to results from the literature where an electron
diffusion length of 0.2 urn and a lifetime of 0.1 ns was found in p-type GaN doped by Mg. The result was
obtained from the analysis of electron beam induced current measurements [8]. There, however, electrons
are the minority carriers, and the transport properties are low due to doping induced defects. We will be
able to obtain an independent measure of the electron transport parameters in our case, once we have
measured the steady-state photocurrent and the generation rate in absolute numbers.
398
5. CONCLUSION
In summary, our experiments hint to the existence of a broad state distribution inside the band
gap of unintentionally doped n-type GaN samples which can explain the non-exponential photocurrent
decay after pulsed laser excitation. The exponent of a power law decay is -0.1 in the case of chopped
UV-light, where the average light bias is strong, and it is about -0.3 in the microsecond time regime
measured with 5 ns pulses using both subbandgap and above bandgap laser light. We have also estimated
the majority carrier diffusion length of electrons to be 23 jxm at a response time of 1.9 \is.
Acknowledgment: S.K. would like to acknowledge a NATO fellowship. The work at 1ST is supported by
the Fundacäo da Ciencia e Tecnologia FCT (project no. PRAXIS/PCEX/P7FIS/7/98). All groups are
members of the EC-funded network under the COPERNICUS programme (project no. ERB IC15 CT98
0819). R.S. and B.K.M. acknowledge financial support through the exchange program Accöes Integradas
of CRUP in Portugal and DAAD in Germany.
References
[1] M.D. McCluskey, N.M. Johnson, CG. Van de Walle, D.P. Bour, M. Kneissl, and W. Walukiewicz.,
Phys. Rev. Lett. 80,4008 (1998).
[2] M.T. Hirsch, J.A. Wölk, W. Walukiewicz, and E.E. Haller, Appl. Phys. Lett. 71,1098 (1997).
[3] M. Topf, S. Koynov, S. Fischer, I. Dirnstorfer, W. Kriegseis, W. Burkhardt, and B.K. Meyer, Mat.
Res. Soc. Symp. Proc. 449 (1997). S. Koynov, M. Topf, S. Fischer, B.K. Meyer, P. Radojkovic, E.
Hartmann, Z. Liliental-Weber, J. Appl. Phys. 82,1 (1997).
[4] C.H. Qiu and J.I. Pankove, Appl. Phys. Lett. 70,1983 (1997).
[5] R. Schwarz, R. Rocha, P. Brogueira, S. Koynov, V. Chu, M. Topf, D. Meister, and B.K. Meyer, to be
published.
[6] H. Antoniadis and E.A. Schiff, Phys. Rev. B46,9842 (1992).
[7] Gyu-Chul Yi and Bruce W. Wessels, Appl. Phys. Lett. 68,3769 (1996).
[8] Z.Z. Bandic, P.M. Bridger, EC. Piquette, and T.C McGill, Appl. Phys. Lett. 73, 3276 (1998).
399
EPITAXIAL GROWTH OF GaN THIN FILMS USING A HYBRID PULSED LASER
DEPOSITION SYSTEM
Philippe Merel *, Mohamed Chaker *, Henri Pepin* and Malek Tabbal **.
*INRS-Energie et Materiaux, 1650 boul. Lionel Boulet, Varennes, Quebec, Canada J3X 1S2.
merel @ inrs-ener.uquebec.ca
**Department of Physics, American University of Beirut, P.O. Box: 11-0236, Bliss Street,
Beirut, Lebanon.
ABSTRACT
A hybrid Pulsed Laser Deposition system was developed to perform epitaxial growth of
GaN on sapphire(OOOl). This system combines the laser ablation of a cooled Ga target with a
well-characterized atomic nitrogen source. Taking advantage of the flexibility of this unique
deposition system, high quality GaN thin films were deposited by optimizing both the laser
intensity and the nitrogen flux. To date, our best GaN films show a FWHM of the GaN(0002)
rocking curve peak equal to 480 arcsec. This result has been obtained at a laser intensity of / =
7xl07 W/cm2, a substrate temperature of 800°C and under Ga-rich growth conditions.
INTRODUCTION
Gallium nitride (GaN) and similar materials (A1N, InN) are the subject of intensive
research because of their numerous applications in the fields of UV-visible light
emitting/detecting devices and high power electronics [1]. The growth of GaN layers with a low
density of defects has proved to be a challenge due to the lack of a high quality lattice matched
substrate. Many conventional deposition techniques, like MOCVD [2] and MBE [3], have been
used to grow these layers with a reasonable amount of success when combined with specific
substrate preparation like the ELOG (Epitaxial Lateral Over Growth) process [4]. In this
relatively new field, it is essential to search for alternative deposition techniques which could
produce high quality thin films while offering a relatively simple process. Also, the comparison
of the layers obtained using different growth methods can help to understand better the
fundamental aspects of GaN heteroepitaxy.
One of the interesting features of Pulsed Laser Deposition (PLD) is that the ablated
species (ions and neutrals) are available at high kinetic energy (10-100 eV) in the ablation plume
[5]. Other advantages include its simplicity and its versatility. Also, PLD can be used with a
reactive background gas leading to the growth of new compounds.
This work presents preliminary results on the epitaxial growth of GaN thin films using a
hybrid Pulsed Laser Deposition (PLD) system combining the laser ablation of a cooled Ga target
with an atomic nitrogen source developed at INRS-Energie et Materiaux [6].
EXPERIMENT
The deposition system is presented in Fig. 1. The atomic nitrogen source is provided by a
high-frequency (HF) plasma generated by an electromagnetic surface wave excited by a Ro-box
[7] (applied field frequency/= 13.56 or 40.68 MHz) or a Surfatron [8] (f= 440 or 2450 MHz)
wave launcher. The total HF absorbed power, PA, is varied from 10 watts to 200 watts. The
discharge is generated in a 4.5 mm inner diameter fused silica tube, 25 cm long, through which
401
Mat. Res. Soc. Symp. Proc. Vol. 572 ® 1999 Materials Research Society
pure N2 flows. The total gas flow, Q, and the pressure in the source, ps, are in the range of 50 to
750 seem and 1 to 30 Torr, respectively, as measured by flow meters and a Baratron pressure
gauge.
The spatial afterglow region is next to the discharge, and it expands to fill a larger quartz
vessel (diameter 3 cm, length 18 cm), which is connected at the other extremity to the deposition
chamber.
In
our
discharge conditions,
the
afterglow
is
separated into two
pressure
readout and
different regions as one
HF
NO gas inlet
leaves the discharge.
power
late afterglow I
First comes 1) the early
^discharge
or "pink" afterglow,
which is characterized
by
vibrationally
excited N2 molecules
0.5-1.5 ram
rotating
nozzle
leading
to
selfsubstrate
neutral beam
differential
ionization
reactions
heater
of atomic
pumping
(Penning effect), this
nitrogen
deposition
gives way to 2) a late
chamber 4—
discharge tube ps= 1-30 Torr
p = 0.5 mtorr
or "Lewis-Rayleigh"
Figare 1. Schematic representation of the deposition system.
late afterglow where
the dominant reaction
is volume recombination of nitrogen atoms. Measurement of the N-atom density through NO
titration of the N-atoms is performed in the afterglow region where an Ar- 1.3% NO gas mixture
can be introduced through a lateral gas inlet. The discharge and afterglow optical emissions are
collected perpendicularly to the discharge tube. The final section of the nitrogen atomic source
consists of a 30 cm long quartz tube (inner diameter of 6 mm) which extends into the deposition
chamber trough a Swagelok seal. The tube is terminated by a nozzle. The pressure inside the
deposition chamber, fixed at 0.5 mtorr, is controlled by both ps and the nozzle diameter <h.
Inside the stainless steel deposition chamber, evacuated by a diffusion pump, the beam of
a KrF excimer laser (248 nm, 100 mJ, 30Hz) is focused by a lens on a rotating Ga target (cooled
to -30°C) located 4 cm away from the substrate. The lens is mounted on a translating stage so
that the laser beam continuously sweeps the target over a one inch diameter. The rotation of the
target combined with the laser beam motion ensures a uniform erosion pattern. A rotating
heating stage holds the 1 inch sapphire(OOOl) substrate. A tungsten filament is used as the
radiation source. In order to increase heat absorption and temperature uniformity, the backside of
the substrate is coated with a 0.4 u,m thick Mo layer deposited by magnetron sputtering. Prior to
growth, the nitrogen source is used for the nitridation of the sapphire substrate at high
temperature (7>950°C), thus favoring GaN nucleation.
After deposition, the crystalline quality of the GaN thin films is determined by triple-axis
X-ray diffraction. The composition of the GaN layers is determined by EDX and XPS.
402
RESULTS
Atomic Nitrogen Source
The flexibility of our atomic nitrogen source enables us to measure the atomic nitrogen
production as a function of various discharge conditions. For example, Fig. 2 presents the Natom density measured in the late afterglow region as a function of the absorbed power, PA, at
different wave frequencies/(ps=4 Torr). Whatever the frequency, [N] first increases rapidly with
PA and then saturates at a level that increases with /. This saturation value increases by
approximately a factor of two
when varying/from 13.56 MHz
to 2450 MHz. One should further
note from Fig. 2 that the absorbed
power required to attain saturation
decreases
with
increasing
frequency,
ranging
from
approximately 130 W at 13.56
MHz to 40 W at 2450 MHz. The
f=2450MHz
initial rapid growth of [N] with PA
D
440 MHz
*
40.68 MHz
(observed before saturation) can
o
13.56 MHz
be linked to an increase of the
electron density [6].
40
120
To gain insight into the
. (watts)
saturation of the N-atom density
Figure 2. Nitrogen atom production, measured in the late afterglow, as with the increase of PA, we turn to
a function of absorbed power at different applied field frequencies (p^ the gas temperature. We assume
4 torr).
that the rotational temperature Tr
of the nitrogen molecules in the discharge is in equilibrium with the gas temperature Tg. The
value of Tr is obtained from the optical emission of the first positive system N2 (B3ng)-(A3E+U).
We find that the observed saturation of the N-atom concentration occurs at a value of Tr near
500 K whatever the value of/. This suggests that, in the pressure range investigated, there is a
threshold gas temperature above which the production of nitrogen atoms saturates. This can be
understood by considering that the production of atomic nitrogen occurs through vibrational
excitation of the N2 molecules. Collisions between vibrationally excited molecules and high
speed N2 molecules destroy these excited states. The saturation level of the N-atom density is
finally determined from the balance between vibrational excitation (which increases with the
electron density) and vibrational energy loss (which increases with Tg).
GaN Layers
For GaN depositon, the plasma source of atomic nitrogen is operated at/= 440 MHz and
= 100 watts. During our first attempts of GaN thin film deposition, the optimal laser intensity
was determined to be equal to 7xl07 W/cm2. Above this value, gallium droplets are observed at
the surface of the thin film while working at a lower laser intensity greatly reduces the
deposition rate. Figure 3 presents a o>29 diffraction pattern of a 0.5 (xm thick sample deposited
at 7>=800 °C. The atomic nitrogen source is operated with discharge conditions (ps = 30 torr, Q
= 750 seem and <|> = 0.5 mm) leading to an atomic nitrogen production percentage of
PA
403
[N]/[N2]=0.4%. These parameters
lead to a deposition rate of 0.08
|xm/h. A well defined GaN(0002)
peak is observed. From this, the
growth orientation is determined to
be GaN(OOOl) II Sapphire(OOOl).
(8
The lattice parameter c of GaN is
found to be 5.183Ä compared to
'S
the tabulated value of 5.185Ä.
G
Sapphire (0006) •
M
GaN layers were also grown
under nitrogen discharge conditions
i
(ps = 4 torr, Q = 370 seem and $ =
..—i—.—i—,—i—i—i—
1.5 mm) corresponding to a
42
40
38
36
34
production
percentage
of
[N]/[N2]=1.7%. By taking into
26 (°)
Figure 3. co-26 scan of a 0.5um thick GaN layer showing the account the nitrogen production
GaN(0002) and sapphire(0006) diffraction peaks (I=7xl07 W/cm2, rate which has been determined 35
Ts=800 °C). The sample is positioned in order to maximize the cm away from the substrate and the
GaN(0002) peak intensity.
nitrogen atom recombination along
i
i—
i
GaN(0002)
Table I. Properties of GaN layers deposited at two atomic nitrogen source pressures. (1=7x10 W/cm ,
T.-snn °cx.
Ps
(torr)
30
4
N-atom
production
rNi/rN2i (%)
Thickness
(urn)
0.4
1.7
0.5
0.7
Growth
rate
N content
(%)
FWHM
GaN(0002)
(arcsec)
48
55
480
936
(|im/h)
0.08
0.25
this path, we estimate that the N-atom flux on the substrate increases by about one order of
magnitude when the source
1
'
1
— '
pressure is varied from 30 torr to 4
•
torr. Table I presents the growth
"
-p =4 torr
.
rates in these two cases. At ps = 4
1 V'1
torr ([N]/[N2]=1.7%), the growth
/ \'\
rate is about 3 times higher than at
3
\ '•
ps = 30 torr ([N]/[N2]=0.4%). XPS
*' /
,' /
\ -_
and EDX analysis of the deposited
•
\ *
,' /
samples also show a 55% nitrogen
s
content for ps = 4 torr compared to
•
\
48% at ft
= 30 torr. This
corresponds to N-rich and Ga-rich
growth conditions respectively.
.
Figure 4 shows rocking
0,4
0,2
0,0
-0,2
-0,4
curves of the GaN(0002) peak of
AG>(°)
the two samples. The FWHM of
Figure 4. GaN(0002) rocking curve of GaN layers; grown ^t two - ^ ^
^ ^ j We
atomic nitrogen source pressures: ps = 30 and 4 torr (1=7x10 w/cm ,
f
r
Ts=800 °C). The FWHM are 480 and 936 arcsec respectively.
see that the Sharpest peaK,
FWHM=480 arcsec, is obtained in
.-\^y
404
Ga-rich growth conditions (ps = 30 torr) compared to FWHM = 936 arcsec when using N-rich
growth conditions (ps = 4 torr). The fact that Ga-rich conditions leads to higher quality GaN
layers has been observed in Molecular Beam Epitaxy experiments [3]. A possible explanation of
these results is presented in [9]. In this theoretical work, it is shown that the mobility of the Ga
atoms on a GaN surface is greatly reduced under N rich conditions, a low mobility of Ga atoms
resulting in a high density of defects in the growing layer.
CONCLUSIONS
An atomic nitrogen source based on a HF plasma has been developed for use in a hybrid
PLD system for the epitaxial growth of GaN. The flexibility and ease of operation of this source
enabled us to monitor systematically the N-atom concentration [N] as a function of the discharge
wave frequency/under otherwise constant discharge conditions. Saturation of [N] as a function
of/and PA was observed when the temperature of the gas exceeds 500 K.
Coupling this nitrogen source with a PLD system, epitaxial growth of GaN films was
achieved. We demonstrated that layers grown under Ga-rich conditions exhibit better crystalline
quality than those obtained under N-rich conditions.
Work is currently in progress to further improve our understanding of the influence of
deposition parameters such as the Ga and N fluxes, the substrate temperature and the presence of
a GaN buffer layer on both the crystalline quality and the electrical characteristics of the PLDGaN thin films.
ACKNOWLEDGMENTS
The authors gratefully acknowledge the financial support of MICRONET and Nortel.
The help provided by Dr. James Webb of the National Research Council is also acknowledged.
REFERENCES
[1] H. Morkoc, S. Strite, G. B. Gao, M. E. Lin, B. Sverdlov and M. Burns, J. Appl. Phys. 76,
1363 (1994).
[2] S. Nakamura, Y. Harada and M. seno, Appl. Phys. Lett. 58, 2021 (1991).
[3]D. Schikora, M. Hanklen, D.J. As, K. Lischka, T. Litz, A. Woay, T. Buhrow and F.
Henneberger, Phys. Rev B54, 8381 (1996).
[4] D. Kapolnek, S. Keller, R. Vetury, R. Underwood, P. Kozodoy, S. DenBaars, U. Mishra,
Appl. Phys. Lett. 71, 1204 (1997).
[5] D. H. Lowndes, D. B. Geohegan, A. A. Puretzky, D. P. Norton and C. M. Rouleau, Science
273, 898 (1996).
[6] P. Merel, M. Tabbal, M. Chaker, M. Moisan, A. Ricard, Plasma Sources Sei. Technol. 7, 550
(1998).
[7] M. Moisan and Z. Zakrzewski, Rev. Sei. Instrum. 58, 1895 (1987).
[8] M. Moisan, C. Beaudry and P. Leprince, Phys. Lett. 50A, 125 (1974).
[9] T. Zywietz, J. Neugebauer and M. Scheffler, Appl. Phys. Lett. 73,488 (1998).
405
EPITAXIAL GROWTH OF ALN ON SI SUBSTRATES WITH
INTERMEDIATE 3C-SIC AS BUFFER LAYERS
S. Q. Hong, H. M. Liaw, K. Linthicum*, R. F. Davis*, P. Fejes, S. Zollner, M. Kottke,
and S. R. Wilson
Motorola Inc., Semiconductor Products Sector, 2100 E. Elliot Road, Tempe, ZA 85282
Department of Materials Science and Engineering, North Carolina State University, Raleigh,
NC 27695
ABSTRACT
Single crystalline A1N was successfully grown on a 3C-SiC coated Si (111) substrate by
organometallic vapor phase epitaxy. The 3C-SiC film was obtained via the conversion of the Si
near-surface region to SiC using gas-source molecular beam epitaxy. The quality of the A1N was
mainly controlled by that of the SiC. The effects of Si pits and SiC hillocks formed during the
conversion on subsequent A1N growth are discussed. Process optimization is suggested to
improve the SiC buffer layer for subsequent A1N deposition.
INTRODUCTION
Silicon carbide (SiC) and aluminum nitride (A1N) are candidate materials for a wide variety of
high-power, high-frequency, and high-temperature applications because of their unique materials
properties as well as their excellent physical and chemical stability [1]. However, advances in
both SiC and A1N technologies are hindered by the absence of large-area and high-quality single
crystals for device fabrication. Successful epitaxial growth of 3C-SiC single crystals on Si
substrates has been demonstrated since 1983 [2], despite the large mismatches in lattice constants
and thermal expansion coefficients between the film and the substrate. Comparatively,
heteroepitaxy of hexagonal SiC has received much less attention, though 4H- and 6H-SiC are
actually of greater interest for power devices. This is due to their higher band gaps relative to the
3C material (3.265 and 3.023 eV, respectively, vs. 2.390 eV) and their commercial availability
for on-going evaluation of device. It has been demonstrated [3] that single crystalline 6H-SiC can
be grown at 1375°C on 2H-A1N coated sapphire (0001). The lattice mismatch between the
carbide and the A1N layer is less than 1%. It therefore seems feasible to deposit hexagonal SiC
on Si substrates if single-crystal 2H-A1N can be grown as an interlayer. The focus of this paper
is the epitaxial growth and characterization of A1N single crystal films on Si( 111) substrates. The
A1N films were grown on thin 3C-SiC buffer layers previously formed by carbonization of the
Si(lll) substrates.
EXPERIMENTAL
Si (111) wafers oriented 3° off axis toward <110> direction were used as substrates. The
surface of each Si wafer was converted to a thin 3C-SiC(l 11) layer via gas-source molecular
beam epitaxy (MBE) prior to chemical vapor deposition (CVD) of an A1N film. Before
conversion, the wafers were dipped in HF, rinsed with deionized H20, and placed on a SiCcoated graphite susceptor. The MBE reactor, with a base pressure of 10"9 Torr, was equipped
with an in-situ reflection high-energy electron diffractometer (RHEED) for surface
crystallographic analysis. The substrate was ramped from room temperature to 970°C at 7-8°C
/min and kept at 970°C for 60 min. Ethylene (C2H4) was supplied from room temperature at a
flow rate of 1.8 seem. The SiC coated Si substrates were loaded in a CVD reactor.
Trimethylalumimum (TMA) and ammonia (NH3) were used for A1N deposition. The flow rates
for TMA, NH3, and additional H2 were 26 (imol/min, 1500 seem, and 3000 seem, respectively.
The A1N was grown at 1100°C for 90 min at a growth rate of 11 Ä/min. For comparison, an A1N
film was also directly deposited on Si(l 11) substrate under the same process conditions.
407
Mat. Res. Soc. Symp. Proc. Vol. 572 • 1999 Materials Research Society
Film thickness was measured by spectroscopic ellipsometry (SE). Crystallinity was
examined with RHEED, X-ray diffraction (XRD), and electron diffraction. Scanning electron
microscopy (SEM) and transmission electron microscopy (TEM) were employed to study surface
morphology and defect formation. Auger electron spectrometry (AES) was conducted using a
5.0 keV primary beam at 30° angle of incidence to determine chemical compositions and the
extent of C and O throughout the films. Sputtering during Auger depth profile analysis was
accomplished by using a beam of 1.0 keV Xe+ incident at 45°.
RESULTS AND DISCUSSION
Figure 1 is a RHEED pattern from the [110] azimuth of the converted SiC on the Si(l 11)
substrate. It shows a single-domain (3x2) reconstruction of the cubic SiC. The formation of
single crystalline 3C-SiC was confirmed by XRD as shown in Fig. 2. The two peaks at 2-theta
=35.5° and 75.5° are SiC (111) and (222), respectively; the other two peaks represent Si (111)
and (222). The Si (222) peak is actually Si (444) observed in second order. The d-spacing of the
SiC (111) peak is 2.52A corresponding to a cubic SiC lattice constant of 4.36Ä, therefore the
SiC lattice is fully relaxed.
20
30
40
50
60
70
2-Theta
Figure 1
RHEED pattern from the
[110] azimuth of the converted SiC on the
Si(lll) substrate showing a single-domain
(3x2) reconstruction of the cubic SiC.
Figure 2 XRD theta-2theta scan of the
single crystalline 3C-SiC film grown on
the Si(lll) substrate.
The surface of the converted SiC film contains many hillocks having an approximate diameter
of 0.2-0.5 |J.m as revealed by SEM and shown in Fig. 3. Under the hillocks, {111} faceted pits
appear in the Si substrate. It is known [4] that the pits are generated by Si outdiffusion from the
substrate which reacts with C atoms forming SiC near pit openings and on pit walls. The
formation of the pyramid-shaped pits is attributed to the low surface energy of the {111} Si
facets. The polycrystalline SiC hillocks near the pit openings are an indication of strain relaxation
[4]. The thickness of the converted layer determined by spectroscopic ellipsometry is
approximately 3-5 nm. The higher thickness near hillocks/pits shown in Fig. 3 suggests an
enhanced growth rate in the defective area. Further discussion regarding pit formation and its
effect on subsequent 2H-A1N growth is presented below.
Figure 4 is an electron diffraction pattern of a cross-sectional AIN/SiC/Si sample, which was
taken using the Si [0-11] zone axis, showing single crystals of A1N and the Si substrate. The
diffraction from very thin SiC layer is too weak to be revealed in the same picture. Most of the
diffraction spots are from hexagonal A1N and Si, with a small percentage resulting from doublediffraction. It is noted that the A1N spots are distorted indicating the existence of slightly
misaligned grains despite its overall single-crystal nature. However, as revealed by RHEED and
408
XRD, A1N deposited directly on a Si substrate is polycrystalline (not shown). This result
suggests the necessity of using SiC as a buffer layer for A1N growth under the process
conditions of this research.
Figure 3
Cross-sectional SEM micrograph of a SiC/Si(l 11) sample
showing the existence of hillocks and pits.
Figure 4
Electron diffraction pattern of a cross-sectional
AIN/SiC/Si sample, showing the single crystalline A1N and Si
substrate.
409
Figure 5 is the cross-sectional TEM micrograph of the same sample showing a hill-and-valley
morphology on the surface of the A1N film. This morphology is believed to be associated with
island growth and subsequent coalescence of the islands [5] and/or the formation of stacking
mismatch boundaries at the steps on the SiC surface, as observed by Tanaka et al. [6]. Slight
misalignment between adjacent islands as well as the formation of the mismatch boundaries can
account for the distorted diffraction spots described above. A high density of dislocations are
expected in the A1N but are not revealed due to the high contrast from the highly stressed film.
pINJ
- ..
SiC
1""
El
.
50 nrrv tj
HI.,.»,,,,..;,,.
'
Figure 5
Cross-sectional TEM
micrograph of a AIN/SiC/Si sample
showing a hill-and-valley morphology on
the surface of the A1N film.
Figure 6
High resolution TEM
micrograph providing a lattice image of
the A1N and the converted SiC layer.
A high resolution TEM micrograph in Fig. 6 provides a lattice image of the A1N and the
converted SiC layer. The AIN-SiC interface is undulated because the surface of the converted
SiC layer possesses a similar wave-like morphology. Therefore, the formation of the hill-andvalley morphology on the A1N surface hear is due to the island growth of both A1N and SiC. At
the SiC-Si interface a high density of misfit dislocations are observed and indicated by arrows.
The SiC film is completely relaxed as discussed in details in a separate paper [7].
An Auger depth profile from A1N into the Si substrate is presented in Fig. 7. The signal from
carbon on the surface drops to the detection limit at about 7 min sputter time. There is no
detectable C throughout the remaining A1N layer. It is believed that C is only on the A1N surface
without extending into the film. The time required to sputter away the adventitious C is modified
by surface roughness, especially when the incident sputtering beam is 45° with respect to wafer
normal. Oxygen was detected throughout the A1N with heavier concentrations on the sample
surface and at the AIN-SiC interface as would be expected. Incorporation of oxygen is not
unusual during Al deposition as Al is a good oxygen getter. The thin SiC layer is not well
resolved, but the Auger data suggests that there is C within the Si substrate. This could be
associated with a SiC layer covering the walls of pyramidal pits under SiC-Si interface. These
pits are a result of Si outdiffusion from the substrate during the SiC conversion at high
temperatures as discussed earlier.
The Si pits and SiC hillocks formed during carbonization affect the quality of A1N layers.
The polycrystalline hillocks near pit openings are a result of stress relaxation and are always
unfavorable for subsequent A1N epitaxy. The effects of the Si pits fall into two categories. In the
410
32
48
80
Sputter Time (min)
Figure 7 Auger depth profile of an AlN/SiC/Si( 111) sample showing
(a) adventitious C on the sample surface and (b) oxygen in the A1N with
heavier concentrations on the sample surface and at the AIN-SiC
interface.
early stage of carbonization, C atoms only partially cover the Si substrate. Si diffuses out from
any of uncovered substrate regions resulting in the formation of {111} faceted voids. If the SiC
nucleation density is low, the small pitted areas cannot be completely covered or sealed by the
coalescence of C or SiC nuclei before a large pit opening is created. The continued dissociation
of C2H4 results in the formation of SiC on pit walls. The subsequent A1N film then conforms to
the morphology of the SiC layer. At high SiC nucleation densities, the originally pitted area is
rapidly sealed resulting in the formation of SiC bridges over the voids generated by continued Si
outdiffusion along {111} planes [8]. The subsequent A1N film appears to have a lower defect
density in the latter case. In this research almost all the pits were found to be sealed [7].
The nucleation density is controllable by varying growth conditions. To increase the SiC
nucleation density a low temperature is necessary in the early stage of conversion. Since the
driving force for Si outdiffusion is the reduction in surface energy, a quick termination of the
high temperature process after the SiC conversion is complete will further prevent the pit
formation. Based on XPS results of the converted SiC surface more than 50% of the deposited
carbon remained unreacted before A1N deposition. Moreover, wafer cleaning is also important as
localized contamination, defects and remnant native oxide tend to enhance pit formation.
411
CONCLUSIONS
Monocrystalline films of A1N have been grown on Si (111) substrate using converted 3CSiC as a buffer layer. Direct deposition of A1N on the Si substrate resulted in polycrystalline
growth under the process conditions of this research. The quality of the A1N on the SiC is mainly
limited by that of the SiC. Process optimization is necessary to improve the SiC quality, and
especially to prevent Si pitting during SiC conversion.
ACKNOWLEDGMENTS
The authors would like to thank Semiconductor Research Corporation and Motorola for
providing research funding. R. F. Davis was supported in part by Cobe Steel Ltd. University
Professorship.
REFERENCES
[1] R. J. Trew, J. Yan, and P. M. Mock, Proc. IEEE 79-5, 598 (1991)
[2] S. Nishino, J.A. Powell and H.A. Will, Appl. Phys. Letters 42, 460 (1983).
[3] B.S. Sywe, Z.J. Yu, S. Burckhard, J. H. Edgar and J. Chaudhuri, J. Electrochem. Soc.
141, 510 (1994).
[4] J. Schmitt, T. Troffer, K. Christiansen, S. Christiansen, R. Helbig, G. Pensl, H.P. Strunk,
Materials Science Forum 264-268, pt.l, 247 (1998).
[5] R. F. Davis, S. Tanaka, L. B. Rowland, R. S. Kern, Z. Sitar, S.K. Ailey, C. Wang, J. of
Crystal Growth 164, 132 (1996).
[6] S. Tanaka, R.S. Kern, J. Bentley and R.F. Davis, to be submitted.
[7] H. M. Liaw, S.Q. Hong, P. Fejes, D. Werho, H. Tompkins, S. Zollner, S. R. Wilson, K.
Linthicum and R. F. Davis, this volume.
[8] J.P. Li and A.J. Steckl, J. Electrochem. Soc. 142, p634 (1995).
412
SIMS AND CL CHARACTERIZATION OF MANGANESE-DOPED ALUMINUM
NITRIDE FILMS
R.C. Tucceri*, CD. Bland*, M.L. Caldwell*, M.H. Ervin**, N.P. Magtoto***, CM. Spalding*,
M.A. Wood**, H.H. Richardson*
Department of Chemistry and Biochemistry, Ohio University, Athens, OH 45701,
richards@helios.phy.ohiou.edu
**Army Research Laboratory/Shady Grove Industrial Park, 8705 Grovemont Circle
Gathersburg, MD 20877-4117, mark_wood@mail.arl.mil
***Department of Chemistry, University of North Texas, Denton, TX, 76201, noelm@unt.edu
ABSTRACT
We have recently carried out MOCVD experiments that showed for the first time the
doping of A1N thin films with manganese. Films of A1N that were doped with less than 0.1% of
manganese showed emission bands at 427 nm, 488 nm and 600 nm in accordance with previous
published excitation and emission spectra of manganese incorporated in bulk A1N. A film with a
higher percentage of manganese (1.7%) grown on top of a pure A1N layer (underlayer) showed
weak emission at 601 nm. A portion of the underlayer not covered during growth of the
overlayer has a very strong emission band at 408 nm and a weaker band at 605 nm. SIMS and
EDX analyses of this film revealed both carbon and oxygen contamination as well as diffusion of
manganese into the A1N underlayer. The band at 408 nm is associated with direct gap emission
from oxygen contaminated A1N and the band at 605 nm is from manganese incorporated by
diffusion into the A1N underlayer.
INTRODUCTION
Group III nitrides have shown promise as being suitable materials for microelectronic and
optoelectronic applications [1]. A1N attracts considerable interest because it possesses a very
large band gap compared to the rest of the group III nitrides (6.2 eV) [2]. In addition, it has a
high heat conductivity, excellent chemical and thermal stability; it will decompose rather than
melt at 2400 °C [3]. Recently, rare earth doped III-V semiconductors have been the subject of
great interest because of their strong visible luminescence [4,5]. However, very little attention
has been paid to incorporation of transition metals into semiconductor hosts. Manganese has
been incorporated in powder samples of A1N [6-8], with the tetravalent manganese ion, Mn+4,
occupying tetrahedral sites. The manganese activated A1N exhibits a maximum in the emission
spectrum in the red region (-600 nm). Doping of A1N thin films with manganese has not been
previously demonstrated. We present here for the first time, evidence that will show that
manganese can be incorporated into a film of A1N grown on Si(100) obtained from a variety of
surface analytical tools such as IR microscopy, SEM imaging, SIMS, EDX, XRF, CL, and XRD.
EXPERIMENTAL
Aluminum nitride (A1N) films were grown by metal organic chemical vapor deposition
(MOCVD) in a high vacuum stainless steel reaction chamber. This system consists of a growth
chamber, pumping unit, and a gas inlet. An Alcatel corrosive resistant turbo pump evacuates the
chamber into the 10"5 Torr range. The source gases for the A1N were ammonia (NH3) and
trimethyl aluminum (TMA). Both gases were constantly maintained in the chamber at a ratio of
2.5:1, respectively. Silicon (Si) (100) substrates, V* in. by Vi in., were mounted onto a boron
413
Mat. Res. Soc. Symp. Proc. Vol. 572 ° 1999 Materials Research Society
nitride ceramic heater and held in place by a molybdenum metal holder. A nichrome
thermocouple was attached to the holder to determine the temperature during the deposition
period. Pressures in the chamber were monitored with an ion gauge. The Si(100) substrates
were flash annealed above 1200 °C to remove the surface oxide coating. The substrates were
then kept at a constant temperature around 860 °C [9]. A stoichiometric ratio of 2.5:1 NH3 to
TMA was used for growing A1N films . Manganese decacarbonyl, ([Mn(CO)5]2), the dopant, was
introduced into the flow reactor via a pulse valve. These films were characterized ex situ with IR
reflectance microscopy, SEM imaging, XRF, XRD, CL, and SIMS.
RESULTS
With the use of IR microscopy, film thickness was determined by measuring the fringe
spacing of the Fabry-Perot oscillations, while the chemical composition was analyzed by
monitoring the peak frequency and bandwidth of the LO mode (880 - 935 cm"1). The sharpness
of LO mode observed in most of the films is related to the degree of crystalinity in the film [10].
The relatively fast growth rates (3-5 um per hour) produced multiple nucleation centers that lead
to a very rough surface. The film thickness ranged from 2 |j.m up to 27 |im.
Many samples show incorporation of manganese during the growth of A1N films. These
films show light emission during electron bombardment (cathodoluminescence). The film
referred to as sample 0209 is 12.18-um thick, and its manganese content relative to aluminum is
less than 0.1%. The CL spectrum of this film collected at room temperature exhibits few
emission bands as shown in Figure 1. First amongst these bands is a slight shoulder that
corresponds to an emission at 23419 cm'1 (427 ran). It is followed by a green band at 20491 cm"1
(488 nm), and then a red band at 16667 cm"1 (600 nm). The red band possesses multiple phonon
states that have been described earlier by Karel and coworkers [7]. Figure 1 also includes the
results of the curve fitting procedure in order to help locate the position of the multiple emission
bands. Although these transitions have been shown to originate from Mn4+ ions in a tetrahedral
field [7], only the emission from the red band has been observed previously. The blue and green
transitions were observed only in the excitation spectrum.
11000
17000
23000
29000
Wavenumbers (cm"1)
35000
Figure 1. Deconvolved spectra and original spectrum for sample 0209.
414
A full characterization of sample 1020 was also carried out. The film had an overall
thickness of 6.0 urn as calculated from TR reflectance microscopy data. In this sample, a
manganese activated A1N layer (overlayer) was grown on top of a pure A1N layer. SEM was
used to clarify the morphology of the films. At 8000 times the magnification, the two film layers
showed different structure. Figure 2 (panel a) displays an SEM image of the overlayer that
exhibits a grain like appearance. Panel b is an_SEM image of a portion of the sample not covered
by the overlayer where a dendritic structure is observed. This portion of the film is referred to as
the underlayer. The SEM images indicate that the films are rough. This is not surprising because
the growth rates (on the order of 3-5 lira per hour) are fast enough to produce multiple nucleation
sites.
Figure 2. a) The overlayer of A1N (Mn incorporation) has a thickness of 6 um. b) The underlayer
containing only A1N has a thickness of 2.9 (xm.
The elemental composition of the film was examined with x-ray fluorescence.
Manganese was found in both layers but at varying relative percentages. The overlayer of A1N
showed 1.7% manganese relative to aluminum, while the underlayer showed 0.6% manganese
relative to aluminum. Relative percentages of the manganese were calculated from a standard
material of aluminum and manganese. A sensitivity factor was then used to relate the intensities
of the two elements. After finding that manganese is present in the film, image cathode
luminescence (CL) was performed to determine the different regions of the film where visible
light is emitted. Figure 3 shows the reversed CL image in which the dark band denotes the area
where light is emitted. Three regions in the CL image are observed. The dark band in the center
is the underlayer (0.6% Mn). Light areas to the left and right are the overlayer (1.7% Mn) and
Si(100), respectively.
Figure 4 shows the EDX traces of both layers in sample 1020. Both the underlayer and
overlayer show the presence of oxygen and carbon contaminations. The concentration of
manganese, however, is below the limit of detection. The room temperature CL spectra from the
overlayer and underlayer region are shown in figure 5. The overlayer region shows an emission
band at 601 nm consistent with literature values for manganese activated A1N [8]. The CL
spectrum from the underlayer region has an intense emission band at 408 mn and a weaker band
at 605 nm. The band at 605 nm is most likely from the manganese that was incorporated in the
A1N by diffusion from the overlayer region. Because the blue emission band is considerably
stronger than the band at 605 nm, it cannot be attributed to Mn emission. Rather it could possibly
415
be associated with contaminants like carbon or oxygen. Since emission bands due to carbondoped A1N have been observed at wavelengths much lower than 400 rnn [9], the band at 408 nm
can be identified with oxygen contamination. Oxygen is introduced either as residual
atmospheric gases (C02 and H20) or as CO from manganese decacarbonyl used as the doping
agent.
Figure 3. A cathode luminescence image of sample 1020 showing the overlayer (left white
region), the underlayer (dark region), and the Si (100) substrate (right white region).
Al
1.5
Figure 4. EDX spectra of sample 1020. The dark traces are spectra from the overlayer and the
light traces are from the underlayer. The films show the presence of oxygen and carbon
contaminants.
SIMS depth profiling was carried out in order to determine the elemental positions in
sample 1020. The film was sputtered in the overlayer region of the sample. An oxygen ion beam
at 7 kV was used to examine the surface. This made detection of oxygen contamination
extremely difficult. The aluminum dimer peak was chosen because the monomer peak was off
scale. In order to properly relate the amount of manganese in the profile to that of aluminum, the
manganese trace is scaled up by a factor of 34. The results of this analysis are shown in figure 6.
The depth profile reveals three layers. The top layer (labeled overlayer) in the depth profile is
rich in manganese. Carbon is also found in this layer. It appears to be highly localized at the
416
surface of the sample, and follows the same behavior as that of aluminum. This indicates that the
observed carbon contamination is associated with the incomplete combustion of TMA. The
depth profile of the next layer, labeled as underlayer, shows the presence of both Al and Mn. It is
evident from Fig. 6 that the degree of manganese incorporation in the underlayer is significantly
smaller than that of the overlayer. Since the underlayer is grown as pure A1N, it should not
contain any manganese. Hence, the presence of manganese in this layer has to be the result of the
diffusion of Mn from the overlayer. This lends support to the idea that the 605 nm emission band
observed in the CL spectrum taken for the underlayer is from Mn that diffused from the
overlayer. Finally, the third layer is identified as the Si(100) substrate. It is also interesting to
point out that the silicon layer contains small amount of Mn, which indicates that Mn is able to
diffuse down to a depth of about 7 um. The absence of sharp, clearly defined interfaces as
shown by the SIMS depth profile is indicative of a very rough sample.
408 n
UndarlByar
"5"
£L
605 nm
ta
-
/V^-M" urn
y
'"
"~1
Onrliytr
1
X7 J
"T
——1
500
.
^-—~>~—r
|
i
|
|
«00
700
800
i
1-j
900
Wavelength [nm)
Figure 5. CL spectra of sample 1020 from the overlayer and the underlayer (top spectrum).
1
500000 I
400000
2
Overlayer
3
4
Underlayer
7
6
8
9
10
Silicon (100)
Mn*
Si+
300000
200000
5
Al2+
,C+ A \
100000 \ /x 34
0
1000
2000
3000
4000 5000
Time (s)
6000
7000
8000
9000
Figure 6. SIMS profile of sample 1020. The graph shows incorporation of manganese into the
A1N film, and suggests a process where manganese diffuses from the overlayer to the underlayer.
417
CONCLUSIONS
The incorporation of a small amount of manganese (< 0.1%) in A1N gives rise to three
emission bands at 427 nm, 488 run and 600 nm in the CL spectrum. These emission
wavelengths correspond to previously published excitation and emission data for manganese
activated A1N. A film with a larger percentage of incorporated manganese (0.6%) exhibits
strong emission at 408 nm and less intense emission at 605 nm. The blue emission band is the
direct gap emission from oxygen-contaminated A1N, while the red emission band is from Mn
incorporated in the A1N film. Manganese incorporation at 1.7% has emission only at 601 nm.
EDX analysis demonstrates the presence of carbon and oxygen contamination. However, only
oxygen contamination gives rise to emission at 408 nm. SIMS measurements reveal that
manganese diffuses from the Mn-rich overlay er to the previously pure A1N underlayer.
ACKNOWLEDGEMENTS
This work is supported by a BMDO URISP grant N00014-96-1-0782 entitled "Growth,
Doping and Contacts for Wide Band Gap Semiconductors". The authors would like to thank V.
Jadwisienczak and HJ. Lozykowski at Ohio University for the CL spectra.
REFERENCES
1. J. L. Rouviere, M. Arlery, R. Niebuhr, K. H. Bachern, O. Briot, MRS Internet J. Nitride
Semicond. Res. 1, 33 (1996).
2. F. Hasegawa, T. Takahashi, K. Kubo, Y. Nannichi, Jpn. J. Appl. Phys. 26, 1555 (1987).
3. D. V. Tsvetkov, A. S. Zubrilov, V. I. Nikolaev, V. A. Soloviev, V. A. Dmitriev, MRS
Internet J. Nitride Semicond. Res. 1, 35 (1996).
4. A. J. Steckl, R. Birkhahn, Appl. Phys. Lett. 73,1700 (1998).
5. H. J. Lozykowski, W. M. Jadwisienczak, Appl.Phys. Lett. 74, 1129 (1999).
6. F. Karel, J. Pastrnäk, Czech. J. Phys B. 19, 79(1969).
7. F. Karel, J. Mares, Czech. J. Phys. B. 22, 847 (1972).
8. F. Karel, J. Pastrnäk, J. Hejduk, V. Losik, Phys. Stat. Sol. 15, 693 (1966).
9. A. D. Serra, N. P. Magtoto, D. C. Ingram, H. H. Richardson, Mat. Res. Soc. Symp. Proc.
482,179(1997).
10. U. Mazur, A. C. Cleary, J. Phys. Chem. 94, 189 (1990).
11. X. Tang, F. Hossain, K. Wongchotigul, M. G. Spencer, Appl. Phys. Lett. 72,1501 (1998).
418
PHOTOLUMINESCENCE BETWEEN 3.36 eV AND 3.41 eV
FROM GaN EPITAXIAL LAYERS
R. Seitz*, C. Gaspar*, T. Monteiro*, E. Pereira*, M.A. Poisson", B. Beaumont***
*Universidade de Aveiro, Departamento de Fisica, Aveiro, Portugal
**Thomson-CSF, Orsay, France
***CHREA-CNRS, Valbonne, France
1. INTRODUCTION
GaN, its alloys, QWs and MQWs have gained an important place among shortwavelength optical emitters and high temperature electronic devices [1,2]. The
performance of such devices is limited by the presence of native and impurity defects.
The understanding of the optical properties of the basic material allows us to improve its
quality and thus increase the performance of these materials.
In non intentionally doped (nid) hexagonal good quality GaN layers grown on
sapphire, 6H-SiC or Si, free exciton (FXC, FXB, FXA), donor bound exciton (DX),
acceptor bound exciton (AX) and donor-acceptor pair (DAP) transitions have been
reported by several authors [3, and references therein]. Besides these typical emissions,
emission lines in the range 3.3 - 3.44 eV have been observed in nid and intentionally
doped hexagonal GaN layers. However the nature of these recombinations is not
completely clarified. Some authors assigned them to a superposition of LO phonon
assisted transitions of DX and FX [3-8], excitons bound to neutral donors with deeper
donor levels [5], band to impurity transitions and/or free to bound emission involving
oxygen [9-11], DAP transitions [12-14], shallow bound excitons of cubic phases [15],
excitons bound to structural defects [16-20] and Zn related recombinations [21].
In this work we analyse the luminescence between 3.36 eV and 3.41 eV of nid
hexagonal GaN samples grown on sapphire. We found sample dependent emission lines
with no DAP behaviour. From the data we are able to identify different kinds of
recombination processes in the same spectral region.
2. EXPERIMENT
Hexagonal GaN layers of ca. 2 u.m where grown by MOCVD on (0001) sapphire
substrates. Steady state (SS) photoluminescence (PL) is excited by the 325 nm line of a
He-Cd laser. The luminescence signal was dispersed by a SPEX 1704 lm
monochromator and detected by a Hamamatsu photomultiplier. Excitation intensity was
varied by neutral density (ND) filters. Time resolved (TR) measurements were carried
out with a pulsed Xe lamp as an excitation source and a boxcar system for detection. The
samples were mounted on a cold finger of a closed-cycle He cryostat and the temperature
of the samples was varied between 12-300K.
419
Mat. Res. Soc. Symp. Proc. Vol. 572
e
1999 Materials Research Society
Samples are oxygen free and present a background concentration of approximately
3. RESULTS
Fig. I shows SSPL and TRPL of an oxygen free sample (sample A) where besides the
excitonic emissions at 3.47 eV, DAP recombination at 3.27 eV and a yellow band (YB)
recombination a set of lines can be observed in the range between 3.3 and 3.41 eV,
namely at 3.400 eV, 3.342 and 3.328 eV (Fig. 2).
Sample A
SSPL
TRPL: TD: 0.08ms
TRPL: I'D: lOmil
Energy / eV
Biergy/eV
Fig. I: SS and TRPL of sample A at 1 OK
Fig. 2: Temperature dependent PL spectra of
sample A
From time dependent analysis of luminescence we are able to assume that all high
energy side emissions are fast, shorter than microseconds as they do not appear in time
resolved spectra. The TRPL spectra show that only the 3.27 eV DAP and the YB
transitions have contributions with lifetimes above 10 (xs. With increasing temperature
the excitonic emissions shift to lower energies while the 3.400 eV emission shifts to
higher energies (Fig. 2) indicating that they have different origin although they show a
similar quenching. The 60 meV energy separation indicates that the 3.4 eV emission
cannot be a LO assisted phonon replica of the excitonic recombinations. Also the relative
intensity of the 3.4 eV band and excitonic emissions is position dependent.
There is no significant shift of the peak position of the 3.4 eV band with varying
excitation density while under low levels of photoexcitation density a high energy shift of
the excitonic emission is observed (Fig. 3).
420
o
GO
Excitation Intensity
Energy / eV
Fig. 3: Excitation intensity
dependence of PL of sample A
Fig. 4: Emission intensity for various
excitation intensities (sample A)
On the other hand a super-linear behaviour of the 3.400 eV band with increasing
excitation intensity can be observed (Fig. 4)
Some of the nid samples present only excitonic lines and their vibronic replicas.
However in some samples an overlap of several high energy emission lines (peaked
between 3.3 to 3.43 eV) with LO phonon assisted replicas of excitonic transitions are
observed as shown in Fig. 5 and 6 (sample B). Some of these lines show a smaller
quenching than the excitonic emissions. It is interesting to note that a line that appears at
3.402 eV (very close in energy to the 3.400 eV emission of sample A) shifts to lower
energies with increasing temperature (broken line in Fig. 5)
The line at 3.383 eV (sample B) shows a quite different evolution with temperature.
In Fig. 7 and 8 a plot of luminescence intensity versus temperature is shown for the 3.400
eV line and DX emission of sample A and the 3.373 eV, 3.383 eV, 3.402 eV and DX line
of sample B. The overall shape of the quenching curve of emissions of sample A and B
are generally described by equations (1) and (2):
I(T) = 1(0) [1 + Ci exp (-Eai/kBT) + C, exp (-Eai/kBT)]-'
I(T) = I(0)[l+Ckexp(-Eak/kBT)r1
(1)
(2)
where the weighting factors C# express the weights of nonradiative dissociation channels
and E# stands for the thermal activation energy of the quenching processes. For the
different lines these values are given in table 1.
421
;\
ky-"'"\
I00K /
\
/
-v^V
\
A I
3.2X
.1.32
3.3ft
3.*)
Energy / eV
.U2
.1.34
3.36
3.3K
3.40
3.42
3.44
3.4G
3.4K
3.50
Energy / eV
Fig. 5: Temperature dependent PL spectra of
sample B
Fig. 6: PL spectra of sample B for 3
different temperatures
Donor Bound Exciton
Excilonic Transition
v
3.383eV line
3.371eV tine
3.402eV line
Fig. 7: Temperature quenching of sample A
Fig. 8: Temperature quenching of sample B
Sample B
SampleA
Line
(eV)
3.46
(exc.)
3.400
c,
5.8
5.0
Table 1:
Line
c3
(eV)
(meV)
18.8
3.461
140
30
(DX)
5.0
3.402
17.4
3.383
68
233
3.5
592
3.373
Quenching parameters of the emission lines
c2
E.1
(meV)
5.2
Ea2
422
Ea3
(meV)
13.6
5.6
29
10
c4
340
Ea4
(meV)
44
4. Discussion
In the spectral region between 3.33 and 3.43 eV several sample dependent emission
lines can be observed. The origin of the emission lines is discussed in detail for two
samples.
4.1 Sample A
In this sample besides the characteristic 3.27 eV DAP emission no other lines
reveal donor acceptor pair behaviour. In fact the 3.400 eV line shows a shift to higher
energies with increasing temperature as recently reported in literature [11,12]. However
no shift of the peak position to higher energies with increasing excitation density and no
long lived luminescence (provided from distant pairs) was observed indicating that in our
samples the 3.400 eV line is not a DAP recombination. It has been reported that this
luminescence was attributed to oxygen related transitions [9] but as our nid samples are
oxygen free as assessed by SIMS, this luminescence cannot be related to oxygen.
In our spectra no shoulder is observed at the 3.400 eV luminescence and no
enhancement of the luminescence was observed when the sample was excited from the
backside as has been mentioned in a recent report [19] indicating that no structural
defects at the interface are involved in this recombination line.
While the exciton energy accompanies the band gap dependence on temperature
the 3.400 eV line shifts to higher energies. This clearly shows that although both DX and
the 3.400 eV line show similar temperature quenching they have different origins. A shift
to higher energies usually indicates a free to bound or a bound to free transition. A free to
bound transition occurs usually when at low temperature DAP emission is quenched due
to ionisation of the donor. In our case no evidence of DAP is found. So we may assume
that the 3.400 eV emission may be due to a bound to free transition. This would fit the
following peak energy law [11]: hv= Ex - Ed + nkT with n=l. This value accounts for
the degeneracy of the top of the valence band. The donor level would be located about
100 meV below the conduction band. The similar quenching may suggest that this donor
is populated from the DX transition.
4.2 Sample B
As in previous reports [3-8] some samples show only excitonic lines and their LO
phonon replicas 92 meV apart. However, in some samples an overlap of several emission
lines can be observed as in our sample B. In this sample the increase of the FX emission
with temperature is clearly observed. Furthermore • the quenching of the DX emission
gives an activation energy equal to the spectroscopic separation between the FX and DX
emissions (13.6 meV).
In this sample the 3.383 eV line is close to 92 meV below the free exciton. Previous
results [7,8] clearly indicate a very small phonon coupling of the free exciton but in
sample B the emission at 3.383 eV reaches the intensity of the free exciton emission at
I00K. Therefore we can exclude a LO phonon replica. DAP transitions have been
ascribed to the recombination mechanisms of a 3.383 eV line in Be doped samples
[15,16]. Also a line at 3.382 eV has been ascribed to a strong phonon replica of a Ix
423
hound exciton peaked at 3.4735 eV[22]. However in our undoped sample B the 3.383 eV
line does not change in peak position with temperature, photoexcitation density and delay
times indicating that this emission is not a DAP recombination. On the other hand no Ix
line was observed as in reference [22].
10 meV lower in energy (3.373 eV) another line is observed. This line shows a
different quenching process than the other lines and cannot be ascribed to a LO phonon
replica of the DX emission.
Although the 3.402 eV line appears very close in energy to the 3.400 eV emission of
sample A and presents a similar quenching process the line in sample B shows a low
energy shift with temperature while the line of sample A shifts to higher energies. So the
two lines cannot be attributed to the same defect.
5. CONCLUSIONS
Emissions in GaN between the excitonic region and the 3.27 eV DAP transitions are
very complex and sample dependent. We show that lines that appear in the same spectral
region, or very close in energies in different samples present some common behaviour
but also differ in fundamental characteristics so that they can be ascribed to different
transitions. Even lines that might be LO phonon replicas due to their peak position are
shown to be of different origin due to their different temperature behaviour.
These results show that different recombination channels are present due to donor
and/or acceptor levels which are introduced in these samples.
REFERENCES
MIS Nakamura, M Senoh, N Iwasa, S nagahama, Jpn. J. Appl. Phys. 34 (1995) L797
[2] S Nakamura, M Senoh, N Iwasa, S Nagahama, T Yamada, T Mukai, Jpn. J. Appl. Phys., 34 (1995)
LI332
131 Annamraju kasi Viswanath, Joo In Lee, Sungkyu Yu, Dongho Kim, Yoowho Choi, Chang-hee Hong, J.
Appl. Phys., 84 (1998) 3848
[41 R Dingle, D d Sell, S E Stokowski, M Ilegems, Phys. Rev. B, 4 (1971) 1211
|51 W Götz, L T Romano, B S Krusor, N M Johnson, R J Molnar, Appl. Phys. Lett., 69(1996) 242
|6] M Leroux, B Beaumont, N Grandjean, P Lorenzini, S Haffonz, P Vennegues, J Massies, P Gibart,
ICAM/E-MRS'97, Spring Meeting, Symposium L, L-III.I
|71 W Liu, M FLi, S J Xu, Kazuo Uchida, Koh Matsumoto, Semicond. Sei. Technol. 13 (1998) 769
|8| D Kovalev, B Averboukh, D Volm, B K Meyer, H Amano, I Akasaki, Phys. Rev. B, 54 (1996) 2518
|9] B C Chung, M Gershenzon, J. Appl. Phys. 72 (1992) 651
110] M Smith! G D Chen, J Y Lin, H X Jiang, A Salvador, B N Sverdlov, A Botchkarev, H Morkoc, Appl.
Phys. Lett., 66(1995)3474
1111 A V Andrianov, D E Lacklison, J W Orton, D J Dewship, S E Hooper, C T Foxon, Semicond. Sei.
Teehonol. II (1996)366
[121 F Calle, F J Sanchez, J M G Tijero, M A Sanchez-Garcia, E Calleja, R Beresford, Semicond. Sei.
Technol., 12(1997) 1396
[13]F J sanchez, F Calle, M A sanchez-Garcia, E Calleja, E Munoz, C H Molloy, D J Somerford, J J
Serrano, J M Blanco, Semicond. Sei. Technol., 13 (1998) 1130
[ 14]D J Dewsnip, A V Andrianov, I Harrison, J W Orton, D E Lacklison, G B Ren, S E Hooper, T S
Cheng, C T Foxon, Semicond. Sei. Technol. 13 (1998) 500
115] Stefan Strauf, Peter Michler, Jürgen Gutowski, Hartmut Selke, Udo Birkle, Sven Einfeldt, Detlef
Hommcl, Jour. Cryst. Growth 189/190 (1998) 682
[161 S Fischer, C. Wetzel, W. Walukiewicz, E E haller, Mat. Res. Soc. Symp. Proc. 395 (1996) 571
424
1171 L Eckcy, J Ch. Hoist, P Maxim, R Heitz, A Hoffmann, I Broser, B K Meyer, C Wetzel, E N Mokhov
P G Baranov. Appl. Phys. Lett. 68 (1996) 415
I IS] C Wetzel, S Fischer, J Krüger, E E Halier, R J Molnar, T D Moustakas, E N Mokhov, P G Baranov
Appl. Phys. Lett., 68 (1996) 2556
1191 S Fischer, G Steude, D M Hofmann, F Kurth, F Anders, M Topf, B K Meyer, F Bertram, M Schmidt, J
Christen, L Eckey, J Hoist, A Hoffmann, B Mensching, B Rauschenbach, Jour. Cryst Growth 189/190
(1998)556
1201 S T Kim, Y J Lee, S H chung, D C Moon, Semicon. Sei. Technol., 14 (1999) 156
121 ] H Y An, O H Cha, J H Kim, G M Yang, K Y Lim, E -K Suh, H J Lee, J. Appl. Phys., 85 (1999) 2888
1221 D G Chtchekine, G D Gilliland, Z C Feng, S J Chua, D J Wolford, S E Ralph, M J Schurman, I
Ferguson, MRS J. Nitride Semicond. Res. 4S1 G6.47 (1999)
ACKNOWLEDGMENTS
This work was partially supported by European community under contract BRPR-CT96034. One of the authors (R. Seitz) gratefully acknowledges to JNICT by a maintenance
grant BD/16284/98 (Praxis XXI Programm}.
425
DISORDER INDUCED IR ANOMALY IN HEXAGONAL AlGaN SHORT-PERIOD
SUPERLATTICES AND ALLOYS
A. M. MINTAIROV *, A. S. VLASOV *, J. L. MERZ *, D. KORAKAKIS **, T. D.
MOUSTAKAS **, A. O. OSINSKY ***, R. GASKA ***, M. B. SMIRNOV ****.
*EE Department, University of Notre Dame , Notre Dame, IN 46556
**ECE Department, University of Boston , Boston,MA, 02215
***APA Optics. 2950 N.E. 84th Lane, Blaine, MN, 55434
****Institute for Silicate Chemistry, Odoevskogo 24/2, 199155 St.Petersburg, Russia
ABSTRACT
We report an experimental (infrared reflectance spectroscopy) and theoretical study of
the polar optical phonons in hexagonal ternary nitride compounds: AlNm/GaNn (n=2-8, m=4, 8)
superlattices (SL) and spontaneously ordered AlxGa|.xN (x=0.08-0.55) alloys. In infrared (IR)
reflectivity spectra we revealed two modes having strong LO-TO splitting (20-150 cm"1), and
several modes, having a small (1-3 cm"1) LO-TO splitting. All modes have a very high
damping parameter >20 cm"1. The unusual observation is the negative value of the oscillator
strength for the weak IR mode at -690 cm"1, suggesting possible lattice instability, consistent
with high damping observed. We found from lattice dynamical calculations that weak IR active
modes correspond to modes localized at GaN-AIN interfaces. Our analysis has shown that an
anomalous mode is induced by the disorder effects and arises due to strong overlapping of the
LO-TO phonon branches of the bulk GaN and A1N. In SL samples the anomalous mode
corresponds to phonons localized on interface inhomogenities.
INTRODUCTION
The study of the ITI-V nitrides GaN, A1N and their alloys has attracted considerable
attention during the last several years due to their application in wide bandgap optoelectronics
and microelectronics. The phonon properties of wurtzite GaN and A1N are intensively studied
and well understood. Less is known about the phonon properties of ternary AlGaN. In the
present paper we report the observation of an anomalous polar mode in IR reflectivity spectra of
AIN/GaN short-period SLs and spontaneously ordered AlGaN alloys. This mode, observed at
-690 cm"1, has negative oscillator strength. To our knowledge this is the first observation of
such modes in crystals. Our analysis and lattice dynamical calculations show that this
anomalous mode induced by GaN-AIN interface inhomogenities and disorder effects.
EXPERIMENTAL METHOD
Short period superlattice films were fabricated using a switched atomic layer
metalorganic vapor deposition process as described in [1]. The films were deposited onto 1.6 |im
GaN buffer layer, which was deposited on the basal plane of sapphire substrate. The superlattice
(SL) consisted of a few hundred periods of (AlNmGaNn) unit cell made up of m monolayers of
A1N and n monolayers of GaN. In the present work we studied superlattices having m=4, 8 and
n=2, 4, 6, 8. The SL period have been determined by X-ray diffraction and transmission
electron spectroscopy measurements [1].
427
Mat. Res. Soc. Symp. Proc. Vol. 572
@
1999 Materials Research Society
AlxGai-xN alloys films having x=0.08,0.22, 0.45, 0.55 and thickness 0.7, 1.4, 0.55,0.45
urn, respectively, were grown by molecular beam epitaxy on the basal plane of sapphire
substrates [2]. The X-ray data indicate SL ordering of the films [2].
The crystal symmetry of AlGaN SLs with even m+n and the alloy layers, grown on
(0001) sapphire substrates, can be characterized by space group C3v' [3]. This group has only ir
active Ai and E vibrational representations. We will distinguish the GaN- and AIN-type
character of the phonon modes by figures 1 or 2. It should be noted that for alloys, the C3v
space group follows from the fact that substitution of the Ga(Al) atoms in Al(Ga) sublattice
(chemical disorder) eliminates improper translation (00'/2) of the C46v space group (bulk GaN
and A1N), which replaces a sixfold screw axis by a threefold rotation one.
Polarized room temperature IR reflectance spectra were taken at oblique incidence (55°)
with a Broker IFS-66V spectrometer. In the simulations the reflectance coefficient was expressed
in terms of dielectric functions e„ and eK of constituent layers. The functions eaa (a=x, z) were
AnFaj
calculated from the standard expression: e^ (fl>) =e00+Jl .—
5
, (1)
a>TOaj-<° -i<Wj
Where 4nF^ , dhoaj, Yj and e„ are mode oscillator strength, transversal frequency, damping,
and high frequency dielectric function of the crystal. The subscript a shows the polarization of
the mode. The oscillator strength was expressed via transversal and longitudinal phonon
frequencies according to [4]. Phonon parameters for the sapphire substrate and buffer GaN layer
entering (1), and the values oiAnF^ were taken from [5] and [6].
For AlGaN layers E(TO) and E(LO) phonon frequencies enter £„, while Aj(LO) and
Ai(TO) frequencies enter e^. Only E-a contributes to s-polarized reflection, while both e« and ea
contribute to p-polarization. Our calculations have shown that the p-polarized reflection
coefficient very weakly depends on £H. The only Ea parameter which affected IR reflectivity is
the highest frequency Ai(L02) phonon.
IR SPECTRA OF AIN/GaN SL AND ALLOYS
In Fig. 1 we present measured and calculated s-polarized (Fig.l,a-e) and p-polarized (f-j)
IR spectra of three SL and two alloy samples. In most of the samples we can distinguish two
main AlGaN restrahl bands - E(l) and E(2), corresponding to the main GaN- and AIN-type
modes with frequencies of TO/LO components -570/620 and -640/840 cm"1, respectively. In
the p-polarization of the SL samples the E(2) band contains a sharp minimum at 740 cm"1 due to
the Ai(LO) phonon of the buffer GaN layer. The minimum corresponding to the high frequency
edge of E(2) band (-850 cm"1) in s- and p-polarizations related to the E(x,y) and Aj(z)
components of L02 phonon, respectively, enabling us to measure its angular anisotropy.
In SL samples (Fig. 1, a,f and c,h) the fine structure of E(l) mode is revealed. It appears
due to an additional weak (LO-TO splitting -2 cm"1) band e(l) near 600 cm"'. We can also see
that the E(l) band in alloy samples has asymmetry of its shape which can be an evidence of the
e(l) mode contribution.
The most interesting feature in the reflectance spectra in Fig. 1 is the fine structure of the
E(2) band observed in all samples. It consists of a small but distinct peak e(la) at -690 cm"
followed by a very weak minimum e(2) at -710 cm"1 and arises from two weak modes with
small LO-TO splitting (1-2 cm"1).
The unusual feature is the negative value of the oscillator strength for the e(la) mode,
i.e. peak in the spectra instead of the dip. This is demonstrated in Fig. 2. We see that only a
428
e(1a)
V,
<9>
A,(L02)
tt>
f£
e(1a)
E(1) ,X
A,(L02)
e(1a)
* • '
A,(L02)
(i)
0)
400 500 600 700 800 900 1000 400 500 600 700 800 900 1000
Wavenumber, cm"
Fig. 1IR spectra (a-e - s-polarization, f-j - p-polarization) of AlNmGaNn SLs (nxm: 8x4 - a,f; 4x4
- b, g; 2x8 - c, h) and AlxGai_xN alloys (x: 0.22 - d, i; 0.45 - e, j)
negative value of the oscillator strength for the e(la) mode leads to the characteristic peak at 690
cm"1 observed in the experiment. The important consequences of this anomaly is the high
phonon damping (y^20 cm"1) in AlGaN. It is clear from Fig. 2 that for smaller damping
(y=W cm"1) the e(la) reflection peak has at maximum unphysical value greater than unity.
This corresponds to negative value of imaginary part of the dielectric function of the crystal
Imfe,), i.e. to the amplification of the electromagnetic waves and lattice instability. For ^=20
cm"1 the broadening of the peak gives the positive value of Imfe) at 690 cm"1 and the value of
the reflection coefficient less than unity.
In Fig. 3 we summarize the mode frequencies versus A1N content of all measured
samples. The frequencies of the main E(TOl), E (LOl), E(T02), as well as weak e(la) and e(2)
IR modes are close in SLs and alloys with the same A1N content.
429
As can be seen from Fig. 3 there is a the strong difference in the frequency of the
Ai(L02) phonon in SLs and alloys indicating its different angular dispersion. In SLs the
1.0
6(1a)
A
O
/ve(2)*
\r^
/
/
O
E(1)
O
Reflectance
E(2)
,/\£:::;s*^^
0.8
3
i
500
600
700
Wave number, cm"1
.
Fig. 2. Effect of anomalous e(la)
mode in IR spectra of AlGaN.
Measured p-spectra of Alo.55Gao.45N
alloy - solid curve; calculated curves
(y in cm"1): dash-dotted - Fe(iaj>0, y}
=10; dotted - Fe(ia)<0,95=10; dashed
Fe(lal<0, gj=20 .
.
800
difference between frequencies of Ai(L02) and E(L02) is positive and has value -50 cm"1. In
alloys it is negative and equals -10 cm"1. In SLs the angular dispersion of the L02 mode
frequency reflects the differences in the average macroscopic polarization field for the modes
950
950
900
850
800
750
- 700
650
600
5501.
0.0
Fig. 3. Frequencies
oftheLO/TO
components of the
polar phonons of
AlNra/GaN„ SLs (a)
and AlGaN alloys (b)
versus A1N content.
Only TO components
frequencies are
shown for e(l), e(2),
and e(la) modes
because of their small
LO-TO splitting value
(-1-5 cm"1).
550
0.5
m/(n+m)
1.0 0.0
0.5
AIN - content
propagating along and perpendicular to the layers [7]. This implies the existence of the GaNA1N interfaces and has a purely long-range origin. In alloys the anisotropy of the L02 phonon
frequency reflects the anisotropy of the short-range atomic forces. As can be seen from Fig.3,b
this anisotropy in alloys is the same as in bulk materials. These angular anisotropy effects for
L02 phonon are well reproduced in our microscopic lattice dynamical calculations (see below).
It can be seen from Fig. 3 that frequencies of the transversal modes E(TOl) and E(T02)
differ slightly from those of the corresponding bulk materials compared with longitudinal ones
- E(LOl) and E(L02). For the GaN-type LOl-phonon a strong decreasing of the frequency in
ternary AlGaN (down to 610 cm"1) is unusual, because of its small dispersion in the bulk (670740 cm"1) [8]. However, the decreasing of the LOl phonon frequency below the T02 frequency
is required the necessity of having a positive value of the oscillator strength for GaN-type
430
phonons (lattice stability). If such a decrease involves short-range interaction, it occurs via a
strong deformation of the Ga-N bond. If this happens, the local strains will act to prevent the
bond deformation what can activate the anomalous e(la) mode. This was confirmed by our
lattice dynamical calculations.
LATTICE DYNAMICS OF ALN/GaN SL
We used lattice dynamical calculations of AlNm/GaNn SL optical phonon frequencies and
IR activity. The model is based on the short-range interatomic potentials and rigid-ion Coulomb
interactions developed for bulk GaN and A1N [8]. The positions of the atoms in the unit cell are
calculated using the condition of the conservation of the bulk bond lengths and the value of the
lattice parameter c determined from X-ray data.
The eigenvectors of GaN-type phonons calculated for a 4x4 SL for two different charge
values of interface N atoms are presented in Fig. 4. In Fig. 4 we used the subscript I (for />1) in
our notations of modes, which is equal to a quantum number of the wavevector along the SL axis
(see below). In model A the charges of the N interface atoms were equal to bulk GaN (-1.14)
and A1N(-1.25) values for AlGa3N and Al3GaN interfaces, respectively. In model B they
were taken equal to the averaged value over the charge of the neighboring metallic atoms, i.e.
-1.17 and -1.21, which qualitatively accounts for chemical bonding effects.
548.8
546.6
••••
• 0"O"0"° ®*0*s*9
J^.Q.O-0
E(T01)
e»®»®«®-
• •
E(T012)
•
552.1
553.5
«Q0<
E(T013)
• P-C^O'O"®***9*®
E(L014)
553.5
.
•
E(T014)
552.1
582
E(L012)
554.6
«Q . .Q - Q. . O" »• ©• «• ©
E(T012)
E(L013)
554.7
585.1
••••
«P'O °"0 -s*®*9*®
•
E(T01)
E(L013:
•
•
e(lo1)
549
«DQ-Cf0 "*• *• *• *"«OoVo.
551.5
••
«PO-G*n »#»• »• »
e(to1)
551.1
'oo .0
• •
aCyQL.o..O..»»(j>#®#®
E(L012)
548.8
547.2
545.2
• • • •
«p-o-o-o-y*V-
jp-o-o-o-"/»*»*'
E(L01)
E(T013)
E(L01)
Fig. 4. Displacement patterns of the GaN-type E-modes in AlN4GaN4 SL calculated in model A
(left) and model B (right).
We can see from Fig. 4 that in model A the confined E(TO/) modes are well described by the
standing waves envelopes with wave vector q = j-l, where k is the number of monolayers, a
is the thickness of the monolayer and \<l<k .In this case the layer has one main IR active
mode having l=\. In model B there is an additional IR active mode localized near the interface.
In this case E(TO) mode is confined in k-lmonolayers and has k-1 quantized components.
431
Our observation of the weak e(l) and e(2) modes
is qualitatively consistent with model B. This is
demonstrated in Fig. 5 where we presented the
transverse and longitudinal response functions of
a 4x4 SL extracted from experimental IR spectra
and calculated (using % =3 cm'1) in models A and
B. The modes e(l) and e(2) localized at AlGa3N
and A^GaN interfaces are clearly revealed in
model B. The differences in the calculated and
observed frequencies indicates a deformation of
bonds, which can be expected due to the
presence of the anomalous mode.
From Fig.4 and 5 we see that in an ideal SL
no anomalous mode exists because there is a
strong decreasing of the E(LOl) phonon
frequency down to 580 cm"1 due to its
penetration in the A1N layer, which has purely
electrostatic origin. In the calculation where we
exclude this effect, putting the mass of A1N equal
1000, we obtain the E(LOl) phonon frequency of
900
600
700
500
670 cm"1 which is close to the frequency of the
Frequency, cm-1
e(la) mode. This shows that modes strongly
Fig. 5. Im (£„) and Imi-Ve^)
localized in GaN bonds must have anomalous IR
functions of AlN^GaNt SL: a behavior in AlGaN material. While in alloys
experimental; b - model B. Insert
such localization is natural to assume due to
shows results for model A
disorder, in SLs the localization of modes can
occur at interface inhomogeneities: fluctuation of
layer thickness or layer intermixing. Our observation of the interface e(2) mode in alloys is
consistent with their SL ordering. As at the GaN-AIN interface the reduction of the LOl phonon
frequency occurs via long-range interaction which does not involve bond deformation, we can
suppose that SL ordering effects is the intrinstic feature of AlGaN.
An interesting consequence of the observed IR anomaly in AlGaN is the possibility of
amplification of IR radiation if one can create conditions for decreasing phonon damping (see
Fig. 2).
REFERENCES
1. M. Asif Khan, J. N. Kuzina, D. T. Olson, T. George, and W. T. Pike, Appl. Phys. Lett., 63,
3471,(1993)
2. D. Korakakis, K. F. Ludwig, Jr., and T. D. Moustakas, Appl. Phys. Lett, 71, 72 (1997).
3. Yu. E. Kitaev, M. F. Limonov, P. Tronc, and G. N. Yushin, Phys. Rev. B, 57, (1998).
4. C. T. Kirk, Phys. Rev. B, 38, 1255, 1988
5. A. S. Barker, Jr.and M. Degems, Phys.Rev. B 7, 743, 1973
6. A. S. Barker, Jr., Phys. Rev. 132, 1474 (1963)
7. M. Cardona, Superlattices and Microstructures, 7, 180, 1990
8. V. Yu. Daviidov, Yu. V. Kitaev, I. N. Goncharuk, A. N. Smirnov, J. Graul, O. Seminchinova,
D. Uffmann, M. B. Smirnov, A. P. Mirgorodsky and R. A. Evarestov, Phys. Rev. B 58, 12899
1998.
432
NONDEGENERATE OPTICAL PUMP-PROBE SPECTROSCOPY OF HIGHLY EXCITED
GROUP IH NITRIDES
T.J. SCHMIDT,00 J.J. SONG,(a) S. KELLER,® U.K. MISHRA,0" S.P. DenBAARS,*' and WEI YANG(c)
<a) Center for Laser and Photonics Research and Department of Physics
Oklahoma State University, Stillwater, Oklahoma 74078
(w
Electrical and Computer Engineering and Materials Departments
University of California, Santa Barbara, California 93106
(c)
Honeywell Technology Center, 12001 State Highway 55, Plymouth, Minnesota 55441
ABSTRACT
We report the results of nondegenerate optical pump-probe absorption experiments performed on
GaN and InGaN thin films and quantum wells under the conditions of strong optical band to band
excitation. The evolution of the band edge in these materials was monitored as the number of photoexcited free carriers was increased beyond that required to achieve population inversion and observe
stimulated emission. The band edge of InGaN is shown to exhibit markedly different high excitation
behavior than that of GaN, explaining in part the reduction in stimulated emission threshold that typically
accompanies the incorporation of indium into GaN to form InGaN. A comparison of the band edge
absorption changes observed in pump-probe experiments to the gain spectra measured in variable-stripe
gain experiments is also given.
INTRODUCTION
GaN and its respective alloys (InGaN and AlGaN) have been attracting an ever increasing
amount of attention due to their physical hardness, inert nature, and large direct band gaps, making them
promising materials for UV-Blue-Green light emitting devices and detectors.1 Current technological
advances have made high brightness light-emitting diodes (LED's) and cw laser diodes based on these
materials a reality.2 Their nonlinear properties are now becoming a focus for many research groups.
Femtosecond four-wave-mixing (FWM) experiments have been used to study the dephasing times of the
A and B free excitons in GaN,3,4 and femtosecond pump-probe transient transmission experiments have
been used to study the ultrafast carrier dynamics in InGaN expitaxial films.3 Picosecond FWM
experiments have shown strong optical nonlinearities below6 and at the band edge of GaN,7 as have
nanosecond FWM experiments.8 However, much information is still unknown about the optical
phenomena exhibited by these materials at the high carrier densities at which practical devices operate.
Recently, nanosecond pump-probe transmission experiments have shown exciton saturation due to
resonant and below resonance optical excitation of the excitonic transitions of GaN.7 Nanosecond pumpprobe experiments with optical excitation above the band gap of GaN have also been reported910 and
have shown large values of induced transparency and induced absorption in the band gap region with
increasing optical excitation. The time evolution of these band edge changes have been studied on
femtosecond," picosecond,12 and nanosecond13 time scales. Similar experiments have been performed on
InGaN thin films and multiple quantum wells (MQWs).14'15 A better understanding of the optical
phenomena associated with high carrier concentrations in this material system is important, not only for
general physical insight, but also as an aid in designing practical devices. We present here a direct
comparison of the absorption properties of GaN and InGaN-based structures in their highly excited state.
The magnitude of the nonlinearities studied in this work suggests the possibility of new photonic devices
based on the group IH nitrides as optical switches.
EXPERIMENT
The GaN samples used in this work were nominally undoped single-crystal films grown by
metalorganic chemical vapor deposition (MOCVD) on (0001) oriented sapphire substrates. Thin A1N
buffer layers, approximately 50 nm thick, were deposited on the substrate at 775 °C before the growth of
the GaN epilayers. The GaN layers were then deposited at 1040 °C directly on the A1N buffer layers. A
433
Mat. Res. Soc. Symp. Proc. Vol. 572
e
1999 Materials Research Society
GaN layer thickness of 0.38 Jim was used for the data presented here. The Ino.i8Gao.82N layer used in this
study was grown by MOCVD at 800 °C on a 1.8 iim thick GaN layer deposited at 1060 °C on (0001)
oriented sapphire . The hvusGao^N layer was 0.1 nm thick and was capped by a 0.05 Jim GaN layer.
The InGaN/GaN MQW was grown by MOCVD on a 1.8 \im thick GaN buffer layer grown on a (0001)
oriented sapphire substrate. The active region was made up of 12 quantum wells consisting of 3 nm thick
InGaN wells and 4.5 nm thick Si doped (n ~ 2xl018 cm"3) GaN barriers. The average In composition in
the wells was - 18 %. The structure was capped by a 0.1 |xm thick Alo.07Gao.93N layer. A detailed
description of the growth conditions is given elsewhere.16 The average In composition was measured
using high-resolution x-ray diffraction and assuming Vegard's law. We note that the actual InN fraction
could be smaller due to systematic overestimation when assuming Vegard's law in this strained material
system.17
The nondegenerate optical pump-probe experiments were performed using frequency doubled
radiation from a nanosecond dye laser as a UV pumping source and broadband fluorescence from a dye
solution as the probe source. The experimental system for the GaN studies consisted of an amplified dye
laser pumped by the second harmonic of an injection seeded Nd:YAG laser operating at 10 Hz. The deep
red radiation from the dye laser was frequency doubled in a nonlinear crystal to produce the near UV
wavelengths used to synchronously pump the GaN layers above their band gaps and the dye solution.
The UV fluorescence from the dye solution was collected and focused onto the GaN layer coincidental
with the pump beam. The broadband transmitted probe was then collected and focused on the slits of a
1 meter spectrometer and spectrally analyzed using a UV enhanced, gated CCD. A pump wavelength of
337 nm (3.678 eV) was used for the GaN data presented in this report. For the InGaN epilayer and
MQW experiments, the third harmonic of the Nd:YAG laser (3.49 eV) was used in place of the dye laser
to synchronously pump the individual samples and the dye solution. The sample temperature was varied
between 10 K and room temperature through the use of a closed cycle refrigerator.
RESULTS
Figure 1(a) shows the 10 K absorption spectra near the band gap for a 0.38 |im GaN sample
subjected to several different pump power densities (L.xc)- We note here that the unpumped absorption
spectra agree very well with published cw absorption values for the same sample.18 Fig. 1(c) shows the
measured absorption changes with respect to the unpumped spectra for the pump densities given in
Fig. 1(a), where Aa = a(IeXC) - a(0). Induced transparency (Aa negative) in the excitonic region and
induced absorption (Aa positive) in the below-gap region are clearly seen with increasing pump density.
Fig. 1(b) shows room temperature (RT) absorption spectra for several different pump densities. We note
that at RT, the A and B free exciton features seen in Fig. 1(a) have broadened and merged into one and
the resulting induced transparency with increasing Ie„ is about one third that at 10 K. The below-gap
induced absorption is seen in Fig. 1(d) to be approximately half that at 10 K. The decrease in the free
exciton absorption with increasing L„ is attributed to many body effects, such as exciton screening by
free carriers, causing a diminution of their oscillator strength.19 Lattice heating has been proposed as the
origin of the below-gap induced absorption,8'11 but is not consistent with observations in other nitride
materials (see the following section). It's origin, therefore, is still not well understood. We note that the
observed induced absorption is spectrally located in the region in which stimulated emission (SE) is
observed and net optical gain is expected, indicating a complex relation exists between induced
absorption and gain in MOCVD-grown GaN thin films. We also note that these samples are optically
thick (aL - 4 for the pump wavelength and sample thickness, L = 0.38 Jim, used in this study), so the
pump intensity is appreciably diminished as it traverses the sample thickness. Therefore, the resulting
values of Aa presented here are lower limit values of the actual change with pump density.
Figures 2(a) and 2(b) show the absorption spectra of the InGaN reference layer near the
fundamental absorption edge at 10 K and RT, respectively. The oscillatory structure is a result of thin
film interference. With increasing excitation density of the pump pulse, the absorption coefficient in the
band tail region is seen to decrease significantly. This bleaching was observed to saturate for L.JC
exceeding ~ 2 MW/cm2 at 10 K and RT. Similar behavior is observed for the MQW sample (not shown),
434
Photon Energy (eV)
Photon Energy (eV)
3.40
3.45
3.50
3.35
3.55
3.40
3.45
3.50
RT
GaN/sapphire
(b)- 1.0
Vy""
'
I
4
1„ = 0
0.35 MW/cm2
0.66 MW/cm2
1.3 MW/cm2
2.7 MW/cm2
m
0.0
RT
(d)
1 .r-
.. //A
0.35 MW/cm2
0.66 MW/cm2
1.3 MW/cm*
2.7 MW/cm2
3.40
3.45
3.50
3.35
3.55
3.40
E
u
q o*o
.1 I
vf
3.45
3.50
Photon Energy (eV)
Photon Energy (eV)
Figure 1: (a) 10 K and (b) room temperature absorption spectra near the fundamental absorption edge as a
function of optical excitation density (I„c) for a 0.38 um thick GaN layer grown by MOCVD on (0001)
oriented sapphire, (c), (d) Differential absorption spectra as a function of Ie„, Aa = a(Ie„) - a(0), for the
spectra in (a) and (b), respectively. The A and B free exciton transitions are clearly seen in the unpumped 10 K
spectrum. Complete exciton saturation is seen for IeXC approaching 3MW/cm2 at both 10 K and room
temperature. Induced transparency in the excitonic region and induced absorption in the below-gap region are
clearly seen with increasing Icxc at both temperatures. The excitation wavelength was 337 nm (3.678 eV).
the only difference being a larger spectral region exhibiting absorption bleaching due to the larger band
tailing exhibited by the MQW sample. The differential absorption spectra are also shown in Figs. 2(c)
and 2(d) for clarity. We note that the induced transparency associated with the absorption bleaching is
quite large, exceeding 3 x 104 cm"1 at both 10 K and RT. The spectral region in which SE is observed is
also indicated in Figs. 2(c) and 2(d). Clear features in the induced absorption bleaching spectra are seen
to coincide with these spectral regions and are attributed to net optical amplification (gain) of the probe
pulse. Again, similar behavior was observed for the InGaN/GaN MQW sample. Induced absorption was
not observed for either InGaN based structure. This lack of induced absorption in the gain region of
InGaN-based structures explains in part the typical reduction in SE threshold that accompanies the
incorporation of indium into GaN to form InGaN.20
The modal gain spectra measured by the variable-stripe method of Shaklee and Leheny21 are
shown in Figs. 3(a) and 3(b) for the InGaN/GaN MQW and InGaN thin film, respectively. The spectra
were taken for excitation lengths less than 200 jlm to minimize re-absorption induced distortions in the
spectra.22,23 The excitation densities in Figs. 3(a) and 3(b) are given with respect to the SE threshold (la,)
measured for long (> 2 mm) excitation lengths. A clear blueshift in the gain peak with increasing optical
excitation is seen for the MQW. This blueshift was observed to stop for 1^ > 12 x In,. Further increases
in Iexc resulted only in an increase in the modal gain maximum. The large shift in the gain maximum of
435
Photon Energy (eV)
Photon Energy (eV)
2.95
1.0
3.00
3.05
2.90
3.10
10 K
RT
ln
ln
0.18Ga0.82N/SaPPhire
2.95
3.00
3.05
3.10
(b)- 1.2
oi»GaoseN/saPPnire
1.0
0.8
E
ü 0.6
"o
^O^
a
0.8 -T
0.6 ^
jHj:
0.2
/
"' '
/,.-/./jf
0.0
/.Hi
/v/
f
0.05 MW/cm*
2
0.25 MW/cm
1.0 MW/cm2
2.0 MW/cm2
— u-o
MW/cm
— 0.10MW/cm
ZI 0.5
1.0 MW/cm
2
2.0 MW/cma
V\
V\
'^/.-'
'... ''
3.05
0.2
/
/ / -•
*V//• /•/• •
>^ /V//
3.00
-
0.0
|tia>BC\\.---'
?.95
0.4
2
a
2.90
3.10
2.95
3.00
0.05 MW/cm2
0.10 MW/cm2
0.5 MW/cm2 -| -3
1.0 MW/cm2
2.0 MW/cm2
3.05
3.10
Photon Energy (eV)
Photon Energy (eV)
Figure 2: (a) 10 K (b) room temperature optical absorption spectra of an InGaN thin film near the
fundamental absorption edge as a function of above-gap optical excitation, (c), (d) Differential absorption
spectra, Aoc(I„c) = oc(I„c) - a(0), as a function of above-gap optical excitation for the spectra shown in (a)
and (b), respectively. The oscillatory structure in (a) and (b) is a result of thin film interference. The
excitation wavelength was 355 nm (3.492 eV). The SE energy is indicated for completeness.
the MQW to higher energy with increasing \K is consistent with band filling of localized states in the
InGaN active layers. Similar behavior was observed at room temperature. The blueshift in the gain
spectra of the InGaN thin film is seen to be considerably smaller than that of the MQW. It was also
observed to stop at considerably lower excitation densities. The modal gain spectra of both samples are
seen to correspond spectrally with the low energy tail of the band tail state absorption bleaching spectra,
as shown in Figs. 3(c) and 3(d).
Figures 3(c) and 3(d) show a comparison of the absorption bleaching spectra of the InGaN/GaN
MQW and the InGaN thin film for several excitation densities below and above the SE threshold. An
interesting difference between the two structures is the behavior of the absorption bleaching as the SE
threshold is exceeded. As the pump density is increased, the bleaching is observed to increase for both
samples. As the SE threshold is exceeded, though, the bleaching of the MQW tail states is seen to
decrease significantly with increasing excitation, while the bleaching of the epilayer's tail states
continues to increase with increasing excitation density. This is clearly seen in Figs. 3(c) and 3(d), where
the dotted and dashed lines show the bleaching spectra for excitation densities below the SE threshold
and the solid lines show the bleaching spectra for excitation densities above the SE threshold. The SE
thresholds for the MQW and epilayer were measured to be - 350 and 300 kW/cm2, respectively, for the
experimental conditions. The bleaching behavior of the MQW is attributed to the opening up of the
bottleneck in the carrier relaxation process due to the fast depopulation of carriers participating in the SE
process.24 For the InGaN thin film, the recombination lifetime was found to always be significantly
436
Photon Energy (eV)
Photon Energy (eV)
2.8
2.9
3.0
300 ' InGaN/GaNMQW
250
10K
xv
i'A
200
150
3.1
3.2
2.9
3.0
3.1
- —'—l—'—'—'—'—r
151»
10 l„
5.0 I»
-..-2.01a
i
'
'.
- 200
InGaN epllayer
10 K
150~
V
E
o
100"
"cö
O
50 n
0.51a
100
50
0
Mi
-50
<a>
*>~^/
3.5 lth
I
Jj
1-7 lth
j
(b) . - -50
-100
2.7
2.8
2.9
3.0
3.1
2.9
3.2
Photon Energy (eV)
3.0
3.1
Photon Energy (eV)
Figure 3: 10 K modal gain spectra of (a) an InGaN/GaN MQW and (b) an InGaN epilayer as a function of
above-gap optical excitation. The excitation densities, I„c, are given with respect to the SE thresholds, U,
measured for long (> 2 mm) excitation lengths. A clear blueshift in the gain maximum and gain/absorption
crossover point is seen with increasing optical excitation for the MQW sample. This trend is much less
obvious for the InGaN epilayer. (c), (d) 10 K nanosecond nondegenerate pump-probe differential absorption
spectra for the structures in (a) and (b), respectively, showing absorption bleaching (Aa negative) of band
tail states with increasing Ic«. Aa(I„c) = a(Iexc) - a(0) and I,, = 100 kW/cm2. The SE threshold for the
MQW and epilayer was found to be - 350 and 300 kW/cm2, respectively, for the experimental conditions of
(c) and (d). An excitation wavelength of 355 nm (3.492 eV) was used for the data shown in (a) through (d).
shorter than the pump pulse, leading to no observable change in the relaxation dynamics of the higher
energy band tail states as the SE threshold is exceeded.
CONCLUSIONS
In summary, nanosecond nondegenerate optical pump-probe absorption experiments have been
performed on GaN thin films, InGaN thin films, and InGaN/GaN MQWs grown by MOCVD on (0001)
oriented sapphire in order to further understand the optical phenomena associated with high carrier
densities in this material system. The evolution of the band edge was monitored as the number of free
carriers was increased by photo-excitation. Exciton saturation was observed in GaN with increasing
carrier concentration, with a resulting decrease in the absorption coefficient approaching
Aa = - 4 x 104 cm"1 at 10 K and - 2 x 104 cm"1 at RT. In addition, large below-gap induced absorption
exceeding Aa = 4 x 104 cm"1 at 10 K and 2 x 104 cm"1 at RT was observed in GaN as the pump density
was increased to over 3 MW/cm2 . The exciton saturation is explained by many body effects, such as
screening by excess free carriers. The origin of the below-gap induced absorption is still not well
understood, but is most likely a result of the high density of defects in MOCVD-grown GaN films. In
437
contrast, InGaN based structures were observed to only exhibit strong bleaching of band tail states with
increasing above-gap optical excitation. The induced transparency associated with the absorption
bleaching was found to exceed 3 x 104 cm'1 at both 10 K and RT. The magnitude of the bleaching in the
InGaN/GaN MQW was found to be significantly affected by the onset of SE, indicating the carriers
responsible for bleaching and SE share the same recombination channels in these structures. Optical gain
was observed on the low energy tail of the absorption bleaching spectra for both InGaN structures. The
large values of induced transparency/absorption studied in this work suggest the potential of new
optoelectronic applications, such as optical switching, for the group HI nitrides.
ACKNOWLEDGMENTS
The work at Oklahoma State University was funded by BMDO, DARPA, and ONR.
REFERENCES
1. S.N. Mohammad and H. Morkoc, Prog. Quant. Electr. 20, 361 (1996), and references therein.
2. S. Nakamura, M. Senoh, N. Iwasa, S. Nagahama, T. Yamada, and T. Mukai, Jpn. J. Appl. Phys., Part 1 34,
L1332 (1995). S. Nakamura, M. Senoh, S. Nagahama, N. Iwasa,, T. Yamada, T. Matsushita, Y. Sugimoto, and
H. Kiyoku, Appl. Phys. Lett. 70, 868 (1997). S. Nakamura, MRS Internet J. Nitride Semicond. Res. 4SI, Gl.l
(1999).
3. A.J. Fischer, W. Shan, G.H. Park, J.J. Song, D.S. Kim, D.S. Yee, R. Horning, and B. Goldenberg, Phys. Rev. B
56,1077 (1997).
4. R. Zimmermann, M. Hofmann, D. Weber, J. Mobius, A. Euteneuer, W.W. Ruhle, E.O. Gobel, B.K. Meyer, H.
Amano, and Akasaki, MRS Internet. J. Nitride Semicond. Res. 2, Article 24 (1997).
5. C.K. Sun, F. Vallee, S. Keller, J.E. Bowers, and S.P. DenBaars, Appl. Phys. Lett. 70,2004 (1997).
6. B. Taheri, J. Hays, J.I. Song, and B. Goldenberg, Appl. Phys. Lett. 68, 587 (1996).
7. H. Haag, P. Gilliot, D. Ohlmann, R. Levy, O. Briot, R.L. Aulombard, MRS Internet J. Nitride Semicond. Res.
2, Article 21 (1997).
8. H. Haag, P. Gilliot, R. Levy, B. Honerlage, O. Briot, S. Ruffenach-Clur, and R.L. Aulombard, Appl. Phys. Lett.
74, 1436 (1999).
9. T.J. Schmidt, I.J. Song, Y.C. Chang, R. Horning, and B. Goldenberg, Appl. Phys. Lett. 72, 1504 (1998).
10. T.J. Schmidt, Y.C. Chang, and J.J. Song, SPIE Conf. Proceedings Series 3419,61 (1998).
11. A.J. Fischer, B.D. Little, T.J. Schmidt, J.J. Song, R. Horning, and B. Goldenberg, SPIE Conf. Proceedings
Series 3624-25 (1999).
12. H. Haag, P. Gilliot, R. Levy, B. Honerlage, O. Briot, S. Ruffenach-Clur, and R.L. Aulombard, Phys. Rev. B 59,
2254 (1999).
13. Theodore J. Schmidt, Ph.D. thesis, Oklahoma State University, 1998.
14. T.J. Schmidt, Y.H. Cho, G.H. Gainer, J.J. Song, U.K. Mishra, and S.P. DenBaars, Appl. Phys. Lett. 73, 1892
(1998).
15. T.I. Schmidt, Y.H. Cho, S. Bidnyk, J.J. Song, S. Keller, U.K. Mishra, and S.P. DenBaars, SPIE Conf.
Proceedings Series 3625-7 (1999).
16. B.P. Keller, S. Keller, D. Kapolnek, W.N. Jiang, X.F. Wu, H. Masui, X.H. Wu, B. Heying, J.S. Speck, U.K.
Mishra, and S.P. DenBaars, J. Electron. Mater. 24,1707 (1995). S. Keller, A.C. Abare, M.S. Minski, X.H. Wu,
M.P. Mack, J.S. Speck, B. Hu, LA. Coldren, U.K. Mishra, and S.P. DenBaars, Materials Science Forum, 264268,1157(1998).
17. T. Takeuchi, H. Takeuchi, S. Sota, H. Sakai, H. Amano, and I. Akasaki, Jpn. J. Appl. Phys. 36, L177 (1997).
M.D. McCluskey, CG. Van deWalle, C.P. Master, L.T. Romano, and N.M. Johnson, Appl. Phys. Lett. 72, 2725
(1998).
18. A.I. Fischer, W. Shan, J.J. Song, Y.C. Chang, R. Horning, and B. Goldenberg, Appl. Phys. Lett. 71,1 (1997).
19. C. Klingshirn and H. Haug, Phys. Rep. 70, 315 (1981), and references therein.
20. T.J. Schmidt, Y.H. Cho, J.J. Song, and Wei Yang, Appl. Phys. Lett. 74, 245 (1999).
21. K.L. Shaklee and R.F. Leheny, Appl. Phys. Lett. 18,475 (1971).
22. T.I. Schmidt, S. Bidnyk, Y.H. Cho, A.I. Fischer, J.J. Song, S. Keller, U.K. Mishra, and S.P. DenBaars, Appl.
Phys. Lett. 73, 3689 (1998).
23. T.J. Schmidt, S. Bidnyk, Y.H. Cho, A.J. Fischer, J.J. Song, S. Keller, U.K. Mishra, and S.P. DenBaars, MRS
Internet J. Nitride Semicond. Res. 4SI, G6.54 (1999).
24. T. Breitkopf, H. Kalt, C. Klingshirn, and A. Reznitsky, J. Opt. Soc. Am. B. 13,1251 (1996).
438
STUDY OF NEAR-THRESHOLD GAIN MECHANISMS IN MOCVD-GROWN
GaN EPILAYERS AND InGaN/GaN HETEROSTRUCTURES
S. BIDNYK *, T. J. SCHMIDT *, B. D. LITTLE *, J. J. SONG *
Oklahoma State Univ., Center for Laser and Photonics Research and Dept. of Physics,
Stillwater, OK.
ABSTRACT
We report the results of an experimental study on near-threshold gain mechanisms in optically
pumped GaN epilayers and InGaN/GaN heterostructures at temperatures as high as 700 K. We
show that the dominant near-threshold gain mechanism in GaN epilayers is inelastic excitonexciton scattering for temperatures below ~ 150 K, characterized by band-filling phenomena and
a relatively low stimulated emission (SE) threshold. An analysis of both the temperature
dependence of the SE threshold and the relative shift between stimulated and band-edge related
emission indicates electron-hole plasma is the dominant gain mechanism for temperatures
exceeding 150 K. The dominant mechanism for SE in InGaN epilayers and InGaN/GaN multiple
quantum wells was found to be the recombination of carriers localized at potential fluctuations
resulting from nonuniform indium incorporation. The SE spectra from InGaN epilayers and
multiple quantum wells were comprised of extremely narrow emission lines and no spectral
broadening of the lines was observed as the temperature was raised from 10 K to over 550 K.
Based on the presented results, we suggest a method for significantly reducing the carrier
densities needed to achieve population inversion in GaN, allowing for the development of GaNactive-medium laser diodes.
INTRODUCTION
The first results on stimulated emission (SE) in GaN at low temperatures were reported in
the literature more than a quarter of a century ago.1 GaN-based structures have been shown to be
chemically stable and able to withstand high temperatures.2 Recently commercialized bright blue
light emitting and laser diodes are all based on InGaN/GaN heterostructures.3 Understanding the
gain mechanisms in this material is extremely important from both a fundamental physics and a
device optimization standpoint. There have been several studies performed explaining the origin
of SE in GaN epilayers and InGaN multiquantum wells (MQWs) at various temperatures.
However, the results reported in the literature are often contradictory.4"7 In this work we
systematically studied the behavior of SE in GaN epilayers and InGaN/GaN MQWs over a wide
temperature range. We demonstrate that at temperatures below 150 K excitonic-related gain
mechanisms dominate the near-threshold emission behavior in GaN epilayers. For temperatures
above 150 K we show that EHP recombination is the mechanism responsible for SE in GaN. We
explain the SE behavior for the InGaN/GaN MQWs in terms of localized states recombination,
thus showing that SE in InGaN/GaN MQWs and GaN epilayers are distinctly different.
EXPERIMENT
The GaN epilayers and InGaN/GaN MQW samples used in this study were grown by
metalorganic chemical vapor deposition (MOCVD). Sample structures and growth parameters
are described in Refs. 8 and 9. The samples were mounted on a copper heat sink attached to a
custom-built wide temperature range cryostat/heater system. The SE part of this study was
439
Mat. Res. Soc. Symp. Proc. Vol. 572 ®1999 Materials Research Society
performed in an edge emission geometry. In order to avoid distortion of the spontaneous
emission spectra due to re-absorption processes, the laser beam was focused on the sample
surface and spontaneous emission was collected from a direction near normal to the sample
surface.8
RESULTS AND DISCUSSIONS
GaN epilayers
To determine if the SE threshold density occurs above or below the Mott density (the
critical density beyond which no excitons can exist), we studied the temperature behavior of the
SE threshold in GaN epilayers grown on sapphire and SiC substrates over a wide temperature
range (20 to 700 K), as shown in Fig. 1. For temperatures above 200 K, the SE thresholds
roughly followed an exponential dependence: Ilh = I0 exp(T/T0), with T0 s 170 K (Ref. 2).
This exponential behavior of the SE threshold is qualitatively similar to that observed in other
material structures. However, as the temperature is reduced to below 200 K, a significant (faster
than exponential) reduction in the SE threshold was observed in GaN epilayers. This is
associated with a change in the gain mechanism indicating a drastic increase in the SE efficiency
at low temperatures. Recently, Fischer et al. convincingly demonstrated the presence of excitonic
resonances in GaN epilayers well above RT through optical absorption measurements.10 Excitons
in GaN epilayers cannot be easily ionized due to the relatively large exciton binding energy.
However, at near-SE-threshold (near-I,,,) pump densities the picture is not straightforward due to
screening of the Coulomb-interaction by the photo generated free carriers. In general, the
existence of excitons depends on the strength of the Coulomb-interaction which in turn depends
on the density and distribution of carriers among bound and unbound states. It has been predicted
theoretically that in a material system with a relatively large exciton binding energy, one could
expect inelastic ex-ex scattering to have the lowest SE threshold at low temperatures." Schmidt
et al. performed pump-probe experiments and confirmed the presence of excitons at pump
10 i . i . i
ex-ex
scattering
200
400
Temperature (K)
FIGURE 1. SE threshold as a function of temperature
for GaN thin films grown on SiC (open triangles) and
sapphire (filled circles).
20
30
40
50
60
-looon^ic1)
FIGURE 2. Energy difference (AE ) between spontaneous
and SE peaks as a function of temperature for GaN thin films
grown on SiC (open triangles) and sapphire (filled circles).
440
densities above the SE threshold at 10 K in reflection spectra.12 These observations support the
idea that excitons persist above the densities required to observe SE in GaN at cryogenic
temperatures.
The effects of excitons on the SE process can be better understood by studying the
temperature and power dependence of the SE peak position. We measured the energy position of
the SE peak at near-I,,, pump densities and the spontaneous emission peak position using low
power cw PL techniques in the temperature range of 20 to 700 K for samples grown on sapphire
and SiC substrates, as shown in the inset of Fig. 2. The position of the spontaneous and SE peaks
in the GaN epilayers is influenced by residual strain resulting from thermal-expansion mismatch
between the epilayers and the substrates. This difference in energy position for the two samples
is largest at low temperature and gradually decreases as the temperature is increased.
To avoid strain-related complications, we restricted ourselves to an analysis of the
relative energy separation between the spontaneous and SE peaks, AE = Espm - ESE, as
depicted in Fig. 2. As we approach low temperatures (T<150 K), AE asymptotically approaches
the exciton binding energy (Ex =21 meV) measured by photoreflectance.13 However, at
temperatures above 150 K, AE monotonically increases and reaches values as high as 200 meV
at 700 K. The behavior of the energy difference between the spontaneous and SE peaks at low
temperatures (<150 K) is well explained by inelastic ex-ex scattering. In the case of ex-ex
scattering the energy difference between the two peaks can be estimated from:14
AE = Espo„ - E?r = (E, ~ Ex) - (E, - 2EX - Er") = Ex + £f* •
h
(D
where Eg is the bandgap energy and E'k~ is the kinetic energy of the unbound electron-hole
pair created during the excitonic collision. At low excitation densities and low temperatures, one
can consider the bands to be empty. The unbound electron-hole pairs created during the process
have small kinetic energy (E'k~h «0). Thus, AE approaches Ex as T -» 0 K, as shown in Fig.
2.
For high temperatures (T>150 K), the energy difference between the spontaneous
and SE peaks gradually increases from -35 meV to a few hundred meV. Both the large energy
difference and the relatively high SE thresholds in this temperature range (Fig. 1.) point to EHP
recombination as the dominant gain mechanism. In EHP recombination, a large number of
excited carriers cause band-gap renormalization effects resulting in large values of AE. Under
such high excitation conditions, excitons are dissociated by many body interactions.4 As further
evidence to support EHP in this temperature range, we point out that excitons have not been
observed in GaN at highly elevated temperatures, even under extremely low excitation
conditions.10 We therefore conclude that EHP recombination is responsible for gain in GaN thin
films at these elevated temperatures. Since no significant change in the behavior of the SE
threshold and SE peak position was observed for temperatures between 150 K and 700 K (Fig. 1
and Fig. 2), we conclude EHP recombination to be the dominant gain mechanism for
temperatures exceeding 150 K.
At temperatures below 150 K, the effect of the kinetic energy E't'h on ex-ex scattering
recombination process could be observed in the excitation density dependence of AE . As the
excitation intensity or temperature is increased, the bottom of the bands become filled. Thus,
unbound electron-hole pairs created in the process of ex-ex collision must have higher energies,
and the kinetic energy E'{h can no longer be neglected. The inset in Fig. 3 shows the power
dependence of AE at three different temperatures near the point when the gain mechanism
experiences a transition from inelastic ex-ex scattering to EHP recombination. For temperatures
441
GaN
. T=100K
•
*/^
.
?E
• %S
40
Pump density {MW/cm1)
0.0 0.5 1.0 1.5
J
? 60
E
)
B
a
4» 50
I
<LU 40
. . . .
0.0
0.5
1.0
jff
T=200 K
.»»
»**T=150 K
>^
«/*"* T=100K
1.5
below 150 K, we observed a rapid increase of AE
at near-I,,, pump densities, as shown in the inset of
Fig. 3. This shift is most likely associated with
band-filling which causes increased values of Eek'h.
For temperatures above 150 K, this strong near-Ith
shift in AE is not observed. At high temperatures (T
>150 K) SE originates from EHP recombination and
the gradual increase in AE with increasing
excitation density is caused by band-gap
renormalization effects. For one photon pumping
and elliptical bands, the band-filling effect
associated with ex-ex scattering gives a calculated
line shift E'k'h proportional to (/ - Ilh) (see Refs.
14 and 15). The substitution of this expression into
Eq. (1) yields:
Power (MW/cm2)
FIGURE 3. Energy separation (&E ) between
1/3
AE = Er + a{l - I,hf
(2)
the spontaneous and SE peaks as a function of
excitation density at 100 K. The solid line
represents a theoretical fit of the experimental
A fit of the experimental data (open circles)
data (open circles) in Eq. (2). The inset shows
taken at 100 K into Eq. (2) is shown in Fig. 3 by a
the change in the behavior of AE at different
solid line. From the fit, we obtained the values of
temperatures.
exciton binding energy Ex= 28 meV and the SE
threshold I,h = 100 kW/cm2, which is in a reasonable agreement with experimental results and
supports the idea of ex-ex scattering being the dominant SE mechanism for temperatures below
150 K.
Since ex-ex scattering has a lower SE threshold than recombination from an EHP (Fig. 1),
it would be advantageous from a device standpoint if SE was dominated by excitonic effects at
RT and beyond. This could be achieved, for example, by introducing 2-D spatial confinement of
carriers. By tailoring the width of a GaN active layer sandwiched between AlGaN confinement
layers, one would expect a significant increase in the exciton binding energy. For increased
values of exciton binding energy, a strong reduction of the homogeneous broadening due to
reduced Fröhlich interactions is expected.16 This could potentially extend the ex-ex scattering
gain mechanism to RT. The SE threshold for such structures would be significantly reduced due
to carrier confinement and a shift in the dominant near-I,,, gain mechanism to that of ex-ex
scattering.
Stimulated emission in InGaN/GaN multiple quantum wells
In this section, the results of SE studies from InGaN/GaN MQWs are presented. The
InGaN/GaN MOW samples were pumped under the same experimental conditions described
above for the GaN epilayers. Figure 4 shows an emission spectrum from the MQW sample
pumped above the SE threshold at RT. The spontaneous emission peak appears to be broad with
a full width at half maximum (FWHM) of approximately 20 nm. A drastic spectral narrowing
occurs when we excite the sample above the SE threshold, which generates narrow SE peak(s)
with FWHM of less than 0.1 nm. Note that the FWHM of the SE peak in high-quality GaN
epilayers is about 2 nm at RT. Such narrow SE lines in InGaN/GaN MQWs with inhomogeneous
In incorporation can only be explained in terms of deeply localized states. Large localization
442
InGaN/GaN MQWs
RT
.Q
FWHM = 0.1 nm
InGaN/GaN MQW
420
430
440
450
Wavelength (nm)
FIGURE 4. Emission spectra from InGaN/GaN
MQW sample at RT. The sample is pumped 1.1
times the SE threshold of 55 kW/cm2.
100
1000
Pump Density (kW/cm2)
FIGURE S. Integrated intensity of InGaN/GaN MQW
emission as a function of pump density for different
temperatures. The slope change from 0.8-1.3 to 2.2-3.0
indicates the transition from spontaneous emission to SE.
introduces discrete atomic-like levels which are observed as narrow peaks in SE spectra. To
further support this idea, we performed a study of SE in the wide temperature range. No
noticeable broadening of the SE peak(s) was observed when the temperature was varied over a
range of several hundred degrees. The low temperature sensitivity of the FWHM is consistent
with that expected from strongly localized carriers. The details of this study are reported
elsewhere.9 The SE threshold for the InGaN MQWs is observed to be an order of magnitude
lower than for a GaN thin film. It is likely that such a low SE threshold in the MQW is due to the
strong localization of carriers at potential fluctuations in the InGaN active layers.
We also found that an increase in temperature leads to a decrease in PL intensity. This
indicates the onset of efficient losses and a decrease in quantum efficiency of the MQW. At high
temperatures, only a small fraction of carriers reach the conduction band minima, and most of
them recombine non-radiatively. The modal gain depends only on radiatively recombining
carriers. Therefore, the temperature increase efficiently decreases modal gain and leads to an
increase in the SE threshold. To evaluate the number of carriers that recombine radiatively, we
studied the integrated photoluminescence intensity as a function of excitation power for different
temperatures, as shown in Fig. 5. For the temperature range studied, we found that under low
excitation densities, the integrated intensity Iinteg from the sample increases almost linearly with
pump density Ip (i.e. IilltegOcIpr, where y=0.8-l.3), whereas at high excitation densities, this
dependence becomes superlinear (i.e. 1^*1/, where ß=2.2-3.0). The excitation pump power at
which the slope of Iinteg changes corresponds to the SE threshold at each temperature.
Interestingly, the slopes of Itoteg below and above the SE threshold do not significantly change
over the temperature range involved in this study. This indicates that recombination of deeply
localized carriers is the dominant SE mechanism in InGaN/GaN MQWs in the entire temperature
range studied.
5. CONCLUSIONS
In conclusion, we have studied the gain mechanisms in GaN epilayers and InGaN/GaN
MQWs in the temperature range of 20 to 700 K. We observed that for temperatures below 150 K
443
the dominant near-threshold gain mechanism in GaN epilayers is inelastic exciton-exciton
scattering, characterized by a low stimulated emission threshold. For temperatures exceeding 150
K, the dominant gain mechanism was shown to be electron-hole plasma recombination,
characterized by a relatively high SE threshold and a large separation between spontaneous and
stimulated emission peaks. The emission spectra for the InGaN-based structures were found to be
drastically different than that of GaN over the entire temperature range studied. SE spectra in
InGaN/GaN MQWs were comprised of extremely narrow lines and no broadening of the lines
was observed as the temperature was raised by several hundred degrees. The InGaN/GaN MQWs
exhibited an order of magnitude lower SE threshold than that of GaN epilayers. The SE behavior
for the InGaN/GaN MQWs is explained in terms of carrier localization at potential fluctuations
in the InGaN layers. The temperature sensitivity of the SE threshold of InGaN/GaN samples was
measured and compared with GaN epilayers. This study demonstrates that the SE mechanisms in
InGaN/GaN MQWs and GaN are distinctly different.
ACKNOLEDGEMENTS
We gratefully acknowledge the support of ONR, DARPA, AFSOR, and NSF.
REFERENCES
1.
R. Dingle, K. L. Shaklee, R. F. Leheny, and R. B. Zetterstrom, Appl. Phys. Lett. 19, 5
(1971)
2. S. Bidnyk, B. D. Little, T. J. Schmidt, Y. H. Cho, J. Krasinski, J. J. Song, B. Goldenberg, W.
Yang, W. G. Perry, M. D. Bremser, and R. F. Davis, J. Appl. Phys. 85,1792 (1999)
3. S. Nakamura and G. Fasol, The Blue Laser Diode, (Springer, Berlin, 1997)
4. H. Amano and I. Akasaki, Proc. Topical Workshop on III-VNitrides, 193, Nagoya, Japan
(1995)
5. I. M. Catalano, A. Cingolani, M. Ferrara, M. Lugarä and A. Minafra, Solid State Comm. 25,
349 (1978)
6. J. Hoist, L. Eckey, A. Hoffman, I. Broser, B. Schöttker, D. J. As, D. Schikora, and K.
Lischka, Appl. Phys. Lett. 72,1439 (1998)
7. T. J. Schmidt, S. Bidnyk, Yong-Hoon Cho, A. J. Fischer, J. J. Song, S. Keller, U. K. Mishra,
and S. P. DenBaars, Appl. Phys. Lett. 73, 3689 (1998), and references therein.
8. S. Bidnyk, T. J. Schmidt, B. D. Little, and J. J. Song, Appl. Phys. Lett. 74,1 (1999)
9. S. Bidnyk, T. J. Schmidt, Y. H. Cho, G. H. Gainer, J. J. Song, S. Keller, U. K. Mishra, and
S. P. DenBaars, Appl. Phys. Lett. 72,1623 (1998)
10. A. J. Fischer, W. Shan, J. J. Song, Y. C. Chang, R. Horning, and B. Goldenberg, Appl. Phys.
Lett. 71,1981 (1997)
11.1. Galbraith and S. W. Koch, J. Crystal Growth 159, 667 (1996)
12. T.J. Schmidt, J. J. Song, Y. C. Chang, R. Horning, and B. Goldenberg, Appl. Phys. Lett. 72
(1998)
13. W. Shan, B. D. Little, A. J. Fischer, J. J. Song, B. Goldenberg, W. G. Perry, M. D. Bremser,
and R. F. Davis, 16369 (1996)
14. R. Levy and J. B. Grun, Phys. Stat. Sol. (a) 22,11 (1974)
15. X. H. Yang, J. M. Hays, W. Shan, J. J. Song, and E. Cantwell, Appl. Phys. Lett. 62, 1071
(1992)
16. H. Jeon, J. Ding, A. V. Nurmikko, H. Luo, N. Samarth, and J. K. Furdyna, Appl. Phys. Let.
57,2413(1990)
444
ELECTRON TRANSPORT IN THE III-V NITRIDE ALLOYS
B. E. FOUTZ*, S. K. O'LEARY", M. S. SHUR***, and L. F. EASTMAN*
* School of Electrical Engineering, Cornell University, Ithaca, New York 14853
** Faculty of Engineering, University of Regina, Regina, Saskatchewan, Canada S4S 0A2
*** Department of Electrical, Computer, and Systems Engineering, Rensselaer Polytechnic
Institute, Troy, New York 12180-3590
ABSTRACT
We study electron transport in the alloys of aluminum nitride and gallium nitride and
alloys of indium nitride and gallium nitride. In particular, employing Monte Carlo simulations we determine the velocity-field characteristics associated with these alloys for various
alloy compositions. We also determine the dependence of the low-field mobility on the alloy
composition. We find that while the low-field mobility is a strong function of the alloy composition, the peak and saturation drift velocities exhibit a more mild dependence. Transient
electron transport is also considered. We find that the velocity overshoot characteristic is
a strong function of the alloy composition. The device implications of these results are discussed.
INTRODUCTION
The III-V nitride semiconductors, gallium nitride ( GaN ), aluminum nitride ( A1N ), and
indium nitride ( InN ), offer considerable potential for electronic and optoelectronic device
applications [1]. Many nitride based devices employ alloys of these materials. For example,
alloys of A1N and GaN are used in field-effect transistors and photodetectors [2] and alloys
of A1N and GaN and alloys of InN and GaN are used in lasers [3]. These device applications
have fueled considerable interest in the fundamental properties these alloys, and thus these
alloys have been the focus of considerable attention in recent years.
In order to analyze and improve the design of electronic devices fabricated with alloys of
the III-V nitrides, a thorough understanding of the electron transport within these materials
is necessary. While electron transport in bulk GaN [4, 5, 6], A1N [7, 8], and InN [9, 10], has
been extensively examined, the transport of electrons in the III-V nitride alloys has yet to be
the focus of much attention. In a recent paper, however, Albrecht et at. [11] employed Monte
Carlo simulations to investigate electron transport in alloys of A1N and GaN. It was found
that while the low-field mobility decreases dramatically with the Al content, the saturation
drift velocity is found to be relatively insensitive.
In this paper, we further study electron transport in the alloys of the III-V nitrides. In
particular, using Monte Carlo simulations we study electron transport in both alloys of A1N
and GaN ( A^Ga^N ) and alloys of InN and GaN ( In^Ga^N ). Steady-state and transient electron transport are considered in this analysis, and the device implications of these
results are considered.
SIMULATIONS
The approach adopted here is similar to that employed by Bhapkar and Shur for the
treatment of wurtzite GaN [6]. In particular, a three-valley model for the conduction band
445
Mat. Res. Soc. Symp. Proc. Vol. 572 ® 1999 Materials Research Society
is employed. Non-parabolicity is considered in all valleys, the non-parabolicity being treated
through the application of the Kane model. We assume that all donors are ionized, and that
the free electron concentration is equal to the dopant concentration. For each simulation,
the motion of one thousand electrons is examined. The scattering mechanisms considered
are (1) ionized impurity, (2) polar optical phonon, (3) piezoelectric, and (4) acoustic deformation potential. Intervalley and alloy scattering are also considered, alloy scattering being
incorporated using the expression of Ridley [12]. Electron degeneracy and electron screening
effects are also accounted for.
The material parameters corresponding to GaN, A1N, and InN, are tabulated in Table I
of Foutz et al. [13]. The alloy material parameters are determined through linear interpolation. The alloy scattering potential, A, is nominally set to 1 eV [14]. For the purposes of
our analysis, the GaN, A1N, and InN band structures of Lambrecht and Segall [15, 16, 17]
are adopted, the corresponding band structural parameters being tabulated in Tables II, III,
and IV of Foutz et al. [13]. Owing to the uncertainty in the band structures, we ascribe an
effective mass equal to the free electron mass to all of the upper conduction band valleys.
RESULTS
We first consider steady-state electron transport. In Figure 1, we plot the steady-state
velocity-field characteristic associated with A^Ga^N for various alloy compositions. For
all cases, the temperature is set to 300 K and the doping concentration is set to 1017 cm-3.
We see that the velocity-field characteristic is changed rather dramatically as the alloy composition is varied. In particular, it is seen that the peak drift velocity diminishes, the field
at which the peak in the velocity-field characteristic occurs increases, the low-field mobility
decreases, and the sharpness of the peak softens as the Al content is increased. While Albrecht et al. [11] found that the saturation drift velocity is insensitive to the alloy content,
we find that the alloyed materials considered in Figure 1, Al0.2Ga0.sN and Alo.4Gao.6N, exhibit lower saturation drift velocities than that exhibited by either pure A1N or pure GaN.
This is attributable to our considerably greater upper valley effective mass and alloy scattering potential selections [14]. The steady-state velocity-field characteristic associated with
In^Gai-zN, for various alloy compositions, is depicted in Figure 2.
The dependence of the low-field mobility on the alloy composition, corresponding to both
AUGa^N and In^Ga^N, is depicted in Figure 3, these results being extracted from our
low-field Monte Carlo simulations of electron transport. Once again, in all cases we set
the temperature to 300 K and the doping concentration to 1017 cm"3. As was observed by
Albrecht et al. [11], the low-field mobility is a strong function of the alloy composition. In
particular, small amounts of Al in A^Ga^N dramatically reduce the low-field mobility.
Similarly, it has been found that small amounts of Ga added to InN dramatically reduce the
corresponding low-field mobility; this cannot be seen from Figure 3, but can be inferred from
the fact that pure InN is found to have a low-field mobility of 2900 cm2/V-s. These results
suggest that while alloy scattering plays a critical role in determining the low-field electron
transport in these materials, it plays a less important role in determining high-field electron
transport. Thus, III-V nitride alloys might be more suitable for short channel devices where
the electric fields are higher and alloy scattering will not significantly degrade the device
performance.
Steady-state electron transport is the dominant transport mechanism for devices with
446
o
-•
• i-H
o
o
o
o
oio
- o
>
<
• Alo.oGai.oN
o Al0.2Ga0.8N > Alo.4Gao.0N
< Ali.oGao.oN
o
. ° > << "
1«
-s^««««^«^
8fc>,?• • • • • • • • »• ««• • • )• «,
'<<!<].
e
1 h o t>
eg
Q
& & P o p 9 9 9 9 9 9 9 9.
ot><
<
Oi
0
100
200
300 400 500 600 700
Electric Field ( kV/cm )
800
900 1000
Figure 1: The velocity-field characteristic associated with AlxGai_xN for various alloy compositions. For all cases, the temperature is set to 300 K, the doping concentration is set to 1017 cm-3,
and the alloy scattering potential is set to 1 eV.
1
•
o
t>
<
100
200
300 400 500 600 700
Electric Field ( kV/cm )
1
Ino.oGai.oN _
Ir10.2Gao.sN
Ino.4Gao.6N
Irii.oGao.oN -
800
900 1000
Figure 2: The velocity-field characteristic associated with In^Gai-^N for various alloy compositions.
For all cases, the temperature is set to 300 K, the doping concentration is set to 1017 cm-3, and
the alloy scattering potential is set to 1 eV.
447
1200
1000
1
1
1
1
1
1
1
1
1
o InxGai_2;N o
• AlzGai^N -
-
-
0
>
800
—
o
-
•
o
o
o
o
600
3
400
—
•
o
•
200-
_
•
•
1
0
0.0
1
1
0.2
1
1
0.4
•
•
1
1
0.6
•
•
1
1
J»
1.0
0.8
Alloy Content ( x )
Figure 3: The low-field mobility for various alloy compositions for A^Gai^N and In^Gai-xN, as
determined from our low-field Monte Carlo simulations of electron transport. For all cases, the
temperature is set to 300 K, the doping concentration is set to 1017 cm-3, and the alloy scattering
potential is set to 1 eV. For In^Ga^N, the low-field mobility at x=0.8, 0.9, and 1.0, is found to
be 1500, 2000, and 2900 cm2/V-s. Note that the data points coincide for the case of x=0.0.
0
10
20
30
40 50 60 70
Displacement ( nm )
80
90
100
Figure 4: The average electron velocity as a function of displacement for various alloy compositions
for AlxGai-zN. For all cases, the temperature is set to 300 K, the doping concentration is set to
1017 cm-3, and the alloy scattering potential is set to 1 eV. The applied electric field is 500 kV/cm
in all cases.
448
larger dimensions. For devices with smaller dimensions, however, transient electron transport must also be considered when evaluating device performance. In studying transient
electron transport in the III-V nitride alloys, we follow the approach of Foutz et al. [18] and
consider the response of electrons to the sudden application of a constant electric field. We
assume that the electrons are in equilibrium prior to the application of the electric field. In
Figure 4, we plot the average electron velocity of electrons in AlxGai_xN as a function of
displacement for various alloy compositions; we assume that the electrons are initially distributed about zero displacement. We have applied a constant electric field of 500 kV/cm,
and set the temperature to 300 K and the doping concentration to 1017 cm-3 in all cases [19].
We see that the transient response varies dramatically with the Al content. In particular,
while the peak average electron velocity of pure GaN overshoots, by a considerable margin,
its corresponding steady-state velocity, this overshoot is dramatically reduced when the Al
content is increased. This would suggest that electronic devices fabricated with the III-V
nitrides should not use alloying if one wishes transient electron transport to enhance the
resultant device performance.
CONCLUSIONS
In conclusion, we have studied steady-state and transient electron transport in the III-V
nitride alloys. We have found that while the low-field mobility is a strong function of the
alloy composition, the peak and saturation drift velocities exhibit a more mild dependence.
Velocity overshoot effects are found to substantially diminish with alloy composition. These
results suggest that III-V nitride alloyed devices are more suitable for short channel devices
where the electric fields are high and the degradation in device performance due to alloy
scattering is not significant. We have also found that III-V nitride based alloys should not
be used in devices which hope to exploit transient electron transport effects.
ACKNOWLEDGEMENTS
The authors wish to thank the Office of Naval Research for financial support under
their MURI program: Grant # N00014-96-1-1223; Project Monitor: J. C. Zolper. One of
the authors ( S. K. O. ) gratefully acknowledges support from the Natural Sciences and
Engineering Research Council of Canada.
REFERENCES
[1] S. N. Mohammad and H. Morkog, Prog. Quant. Electron. 20, 361 ( 1996 ).
[2] M. S. Shur and M. A. Khan, Mater. Res. Bull. 22 (2), 44 ( 1997 ).
[3] S. Nakamura, Mater. Res. Bull. 22 (2), 29 ( 1997 ).
[4] M. A. Littlejohn, J. R Häuser, and T. H. Glisson, Appl. Phys. Lett. 26, 625 ( 1975).
[5] M. Shur, B. Gelmont, and M. A. Khan, J. Electron. Mater. 25, 777 ( 1996 ).
449
[6] U. V. Bhapkar and M. S. Shur, J. Appl. Phys. 82, 1649 ( 1997 ).
[7] S. K. O'Leary, B. E. Foutz, M. S. Shur, U. V. Bhapkar, and L. F. Eastman, Solid
State Commun. 105, 621 ( 1998 ).
[8] J. D. Albrecht, R. P. Wang, P. P. Ruden, M. Farahmand, and K. F. Brennan, J. Appl.
Phys. 83, 1446 ( 1998 ).
[9] S. K. O'Leary, B. E. Foutz, M. S. Shur, U. V. Bhapkar, and L. F. Eastman, J. Appl.
Phys. 83, 826 ( 1998 ).
[10] E. Bellotti, B. K. Doshi, K. F. Brennan, J. D. Albrecht, and P. P. Ruden, J. Appl.
Phys. 85, 916 ( 1999 ).
[11] J. D. Albrecht, R. Wang, P. P. Ruden, M. Farahmand, E. Bellotti, and K. F. Brennan, Mater. Res. Symp. Proc. 482, 815 ( 1998 ).
[12] B. K. Ridley, Quantum Processes in Semiconductors ( Oxford, New York, 1982 ).
[13] B. E. Foutz, S. K. O'Leary, M. S. Shur, and L. F. Eastman, J. Appl. Phys. ( in
press ).
[14] As was pointed out by Albrecht et al. [11], the selection of a specific numerical value for
A may be the the subject of considerable debate. Albrecht et al. [11] chose A = 0.01 eV for
their nominal selection.
[15] W. R. L. Lambrecht and B. Segall, in Properties of Group III Nitrides, No. 11 EMIS
Datareviews Series, edited by J. H. Edgar (Inspec, London, 1994 ), pg. 141.
[16] W. R. L. Lambrecht and B. Segall, in Properties of Group HI Nitrides, No. 11 EMIS
Datareviews Series, edited by J. H. Edgar (Inspec, London, 1994 ), pg. 135.
[17] W. R. L. Lambrecht and B. Segall, in Properties of Group III Nitrides, No. 11 EMIS
Datareviews Series, edited by J. H. Edgar (Inspec, London, 1994 ), pg. 151.
[18] B. E. Foutz, L. F. Eastman, U. V. Bhapkar, and M. S. Shur, Appl. Phys.
70, 2849 ( 1997 ).
Lett.
[19] The analysis of Foutz et al. [13] demonstrates that velocity overshoot effects can occur over substantially enhanced distances when the electric field is lower. In particular, for
an electric field selection of 210 kV/cm, Figure 2a of Foutz et al. [13] demonstrates that
velocity overshoot effects in pure GaN are exhibited over distances in excess of 0.3 ^m. We
chose 500 kV/cm as A1N does not exhibit velocity overshoot until the electric field in excess
of 450 kV/cm, as was demonstrated by Foutz et al. [13].
450
HIGH-QUALITY GaN GROWN BY
MOLECULAR BEAM EPITAXY ON Ge(OOl)
H. SEGLE***, Y. KIM*-**, SUDHIR G. S. *'**, J. KRÜGER *'**, P. PERLIN**, J.
W. AGER m**, C. KISIELOWSKI***, E. R WEBER*
»Department of Materials Science and Mneral Engineering, UC Berkeley, Berkeley,
CA 94720
** Lawrence Berkeley National Laboratory, Materials Science Division, Berkeley,
CA 94720
*** Lawrence Berkeley National Laboratory, National Center for Electron Microscopy,
Berkeley, California 94720
ABSTRACT
We report on growth of GaN on Germanium as an alternative substrate material. The
GaN films were deposited on Ge(001) substrates by plasma-assisted molecular beam
epitaxy. Atomic force microscopy, x-ray diffraction, photoluminescence, and Raman
spectroscopy were used to characterize the structural and optical properties of the films.
We observed that the Ga/N ratio plays a crucial role in determining the phase purity and
crystal quality. Under N-rich conditions the films were phase-mixed, containing cubic and
hexagonal GaN, while in the Ga-rich regime they were purily hexagonal. The latter
samples show bandedge luminescence with linewidths as small as 31 meV at low
temperatures.
INTRODUCTION
The availability of suitable substrate materials is a key requirement for the epitaxial
growth of GaN thin films. The lack of GaN substrates of appropriate size and structural
perfection has motivated the search for alternative substrates for heteroepitaxial growth.
The most common substrates, sapphire, SiC and GaAs, exhibit large differences in lattice
constants and thermal expansion coefficients to GaN, resulting in large amounts of stress
and high defect concentrations in the epilayers. [1,2] Ge in contrast possess the same
thermal expansion coefficient as GaN. [3] Since most of the stress in heteroepitaxial GaN
layers arises from the ill-matched expansion coefficients, growth on Ge would lower the
amount of stress in the GaN epilayers and thus improve their structural, electrical, and
optical quality. Besides, Ge can easily be doped and therefore could serve as a
backside/bottom contact.
451
Mat. Res. Soc. Symp. Proc. Vol. 572 e 1999 Materials Research Society
We demonstrate that Ge provide an alternative substrate material for GaN growth.
The GaN layers were deposited on Ge(OOl) substrates by plasma-assisted molecular
beam epitaxy (MBE). Their structural and optical properties were characterized by
atomic force microscopy, x-ray diffraction, Raman and Photoluminescence spectroscopy.
EXPERIMENT
A Riber 1000 MBE system was used for the epitaxial growth. An effusion cell
provided elemental gallium by evaporation and activated nitrogen was produced by a
constricted glow discharge (CGD) plasma source with pure nitrogen gas (99.9995%).
Details of the plasma source are given elsewhere. [4]
Intrinsic Ge (001) wafers with miscuts of 6 degree toward <110> served as
substrates. Prior growth they were degreased by boiling in acetone and isopropyl alcohol
and were finally rinsed in deionized water to remove Ge oxides. The substrates were then
heated up to 680°C in ultra high vacuum for thermal desorption of surface contaminants.
In contrast to growth on sapphire, the Ge substrates were not nitridated. Subsequently, a
GaN buffer layer was deposited on the substrate at 600°C for 5 minutes. Finally, the
GaN main layer were grown for 2 hours at 680°C. The Ga/N flux ratio at the deposition
of the buffer as well as the main layer was varied by changing the nitrogen flow rate while
the Ga cell temperature was kept constant at 880°C. We found best results with 5 seem
N flow at a background pressure of 3.5 mTorr for the buffer layer growth and 35 seem for
the main layer growth. The resulting thickness of the GaN epilayer was about 1 urn.
The surface morphology was tested using contact-mode atomic force microscopy
(AFM). The full width at half maximum (FWHM) of the rocking curve and the a- and clattice parameters were measured with a Siemens D-5000 diffractometer containing a fourbounce Ge monochromator. Room-temperature micro-Raman spectra were recorded with
the 488 am line of an Ar2* laser in backscattering geometry. A notch-filter suppressed
elastically scattered light while the Raman signal was analyzed by a single 0.5 m
spectrometer equipped with a charged coupled device detector. To get information on the
layers' optical properties we performed low-temperature photoluminescence (PL)
spectroscopy. PL at 4 K was excited by a 50 mW HeCd laser, diffracted by a 0.85 m
double-grating monochromator and detected by a UV-sensitive photomultiplier.
Secondary ion mass spectroscopy was carried out on selected layers to measure their
compositions.
452
T=10K X =325nm
3.466 eV
I
3
xi
\—
-5-
31meV
>.
+^
'w
c
ID
—i— —i—
1
—i—
2.2
2.4
2.6
2.8
3.0
3.2
3.4
3.6
3.8
Energy (eV)
Fig. 1: Typical low-temperature PL spectrum qfGaN grown on Ge after excitation at
325 nm (3.81 eV). The near-bandgap luminescence exhibits a relatively narrow
linewidth of 31 meV andpeaks at the stress-free position of 3.466 eV.
RESULTS AND DISCUSSION
Figure 1 displays atypical PL spectrum of the GaN layers grown on Ge substrates
taken at low temperatures after excitation at 325 nm (3.81 eV). The spectrum is
dominated by the hexagonal near-bandgap, exchonic luminescence at 3.466 eV. No cubic
PL can be observed indicating high hexagonal phase purity. Defect-related broad bands,
as e.g. the yellow luminescence, do
not appear in the spectrum either.
Along with the narrow linewidth of
31 meV, this demonstrates the good
optical quality of the layers which is
superior to growth on silicon with
best linewidths of around 40 meV
[5] and 100 meV [6], respectively.
The PL line position at 3.466 eV
corresponds exactly to the stressfree value [1].
_. ,.„,.^
j-,-,
2
tr- \T/r^
As can be seen from Fig.
a 2
Fig. 2: AFMpicture ofa 2x2 fim area ofGaN/Ge.
which
shows
m
The root-mean-square roughness is 20.6 nm.
atomic-force
453
microscopy (AFM) image taken of a 2x2 urn2 area, the growth mode is three-dimensional,
typical for MBE-GaN because of the low growth temperature. We determined a rootmean-square (rms) roughness of 20.6 nm. This value is in the range we normally observe
from GaN deposited on sapphire, [7].
To get information about the structural properties of our layers we performed microRaman spectroscopy analysis. This technique allows us to determine the stress and the
free-carrier concentrations in the layers and gives us at least a qualitative impression of
the crystalline quality by measuring the linewidths of the Raman modes and by looking
for disorder-activated scattering. For details about the use of Raman spectroscopy as a
characterization tool in general, the reader is referred to [8],
In Fig. 3a we plot a typical Raman spectrum taken in z(..)z geometry of an 1 urn
thick layer grown under optimized Ga-rich conditions. The spectrum is dominated by the
hexagonal ^(high) mode, which is the only mode allowed in this scattering geometry
assuming the c-axis to be perpendicular to the surface. We found linewidths in the range
of 5 cm"1 comparable to the values known from hexagonal GaN deposited on sapphire,
[9]. The absence of the A^O) mode indicates a very high free-carrier concentration in
the layer exceeding lxl019cnf3 [10]. Beside point defects, which might be caused by
non-perfect growth conditions, we expected the incorporation of Ge atoms in-diffusing
from the substrate during the growth process to be the main cause for the high free carrier
concentration. Similar to Si, Ge is a donor in GaN. To verify our assumption we
performed SIMS measurements on selected GaN layers. The SIMS profiles confirmed
the indiffusion of Ge into the GaN epilayers.
The üi(high) Raman mode reacts sensitively on pressure [11]. Since it is a nonpolar
mode its frequency neither depends on the propagation direction relative to the c-axis, as
observed from the Ax and Ex modes forming the so-called quasi modes [12] nor does the
mode shifts due to interaction with free carriers. Therefore, its frequency is a direct
measure of stress in the layer. We found in all our layers the Ei to be located at the
stress-free frequency of 567 cm"1 [13] in agreement with our PL results presented in Fig.
1. Our x-ray measurements yield similar results. The lattice parameter has been
determined to 5.1748 A.
Similar to the growth of GaN on GaAs [14,15] we observed that the Ga/N flux ratio
plays a crucial role in determining the phase purity of the layers. When growing in the
Ga-rich regime we found purily hexagonal GaN layers, as displayed in the PL and Raman
spectra of Figs. 1 and 3. With N-rich conditions the layers become phase mixed,
containing both cubic 3C-GaN as well as hexagonal 2H-GaN. On the right side of Fig. 3, a
Raman spectrum of such a phase-mixed layer is shown. One can clearly see the cubic TO
mode beside the hexagonal E2 mode. Similar results were found in the PL spectra, in
454
567 cm "1
E2(high)
n
CO
w
c
0)
—
c
j
c
d>
a(0
o
CO
,.,..
5 cm"1
z(..)z
».„„=488 nm
^
i
i
i
i
i
550 600 650 700 750
550 600 650 700 750
Raman Shift (cm
1
]Raman Shift (cm
1
)
Fig. 3a,b: Typical room-temperature Raman spectrum taken in z(..)z geometry of an
I fim thick GaN/Ge layer grown under optimized Ga-rich conditions (left, Fig. 3a) and
ofa phase-mixed layer (right, Fig. 3b).
which 200 meV below the hexagonal near-bandgap luminescence the corresponding cubic
PL occurred. Together with the coexistence of both phases in the layers the crystal
quality deteriorates, indicated by broader Raman lines and the appearance of disorderactivated Raman scattering [16]. The absence of long-range order in disordered materials
yields to a breakdown of the q = Aft = 0 selection, where q is the wavevector of the
phonon and Aft the difference between incident and scattered photon wavevector [17].
Thus, the Raman spectrum reflects the vibrational density of states.
SUMMARY
Summarizing our results, we achieved growth of high-quality GaN on Ge(001)
substrates. Optimization of the growth conditions resulted in stress-free GaN layers, as
indicated by PL and Raman spectroscopy. We found a strong dependency of the
material's quality on the Ga/N flux rate. When grown under Ga-rich conditions, the layer
exhibit a comparatively high optical quality, as indicated by the near-bandedge
luminescence linewidth of 31 meV and the absence of defect-related bands. Under N-rich
growth conditions, the GaN layers are comprised of both the hexagonal and the cubic
phase. In all cases, the layers exhibit a rather large free electron density which has to be
455
attributed to the in-diffusion of Ge from the substrate. Also, the current crystalline
quality calls for further optimization of the growth conditions.
ACKNOWLEDGMENTS
The authors thank M. Straßburg for carrying out the SIMS measurements. One of the
authors (H.S.) acknowledges the support of a DAAD fellowship. This work was
supported by the Office of Energy Research, Office of Basic Energy Sciences, Division of
Materials Sciences of the U.S. Department of Energy under Contract No. DE-AC0376SF00098.
REFERENCES
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
C. Kisielowski, J. Krüger, S. Ruvimov, T. Suski, J. W. Ager m, E. Jones, Z. LilienthalWeber, M. Rubin, E. R. Weber, M. D. Bremser, R. F. Davis, Phys. Rev. B 54, 17745
(1996)
J. Krüger, N.Shapiro, S.Subramanya, Y.Kim, H.Siegle, P.Perlin, E,R.Weber, W.S. Wong,
T.Sands, N.W.Cheung, R. J. Molnar, this MRS meeting
Landolt-Börnstein Tables, edited by O. Madelung, M. Schulz, and H. Weiss (Springer,
Berlin 1982)
A. Anders, N. Newman, M. Rubin, M. Dickinson, E. Jones, P. Phatak, and A. Gassmann,
Rev. Sei. Instrum. 67, 905 (1996)
Y. Nakada, I. Aksenov, H. Okumura, Appl. Phys. Lett. 73, 827 (1998)
B. Yang, A. Trampert, O. Brandt, B. Jenichen, and K. H. Ploog, J. Appl. Phys. 83, 3800
(1998)
H. Fujii, C. Kisielowski, J. Krüger, R. Klockenbrink, M. S. H. Leung, Sudhir G. S., H. Sohn, M.
Rubin, andE. R Weber; Mat.Res.Soc.Symp.Proc. 449, 227 (1997)
Lieht Scattering in Solids I-VL edited by M. Cardona and G. Guntherodt, Topics Appl.
Phys. (Springer, Berlin, Heidelberg)
H. Siegle, L. Eckey, A. Hoffinann, C. Thomsen, B. K. Meyer, D. Schikora, M. Hankeln,
K. Lischka, Solid State Communications 96, 943 (1995)
H. Harima, T. Inoue, S. Nakashima, H. Okumura, Y. Ishida, S. Yoshida, H. Hamaguchi,
Paper presented on the ICNS'97, October 27-31, 1997, Tokushima, Japan
H. Siegle, A. R. Goni, C. Thomsen, C. Ulrich, K. Syassen, B. Schöttker, D. J. As, D.
Schikora, Materials Research Society, Symposium Proceedings Vol. 468, ed. C. R
Abernathy, H. Amano, J. C.Zolper, pp. 225 (1997)
L. Filippidis, H. Siegle, A. Hoffinann, C. Thomsen, K. Karch, and F. Bechstedt, phys.
stat. sol. (b) 198, 621 (1996)
H. Siegle, A. Hoffinann, L. Eckey, C. Thomsen, J. Christen, F. Bertram, B. Schmidt, K.
Hiramatsu, Appl. Phys. Lett. 71, 2490 (1997)
O. Brandt, H. Yang, B. Jenichen, Y. Suzuki, L. Däweritz, and K. H. Ploog, Phys. Rev. B
52,R2253(1995)
D. Schikora, M. Hankeln, D. J. As, K. Lischka, T. Litz, A. Waag, T. Buhrow, F.
Henneberger, Phys. Rev. B 54, R8381 (1996)
H. Siegle, G. Kaczmarczyk, L. Filippidis, P. Thurian, A. Hoffinann, C. Thomsen,
Zeitschrift für Physikalische Chemie 200, 187 (1997)
M. H. Brodsky, in Light Scattering in Solids I. edited by M. Cardona (Springer, Berlin,
1975), pp. 205
456
CARRIER RECOMBINATION DYNAMICS OF AlxGai.„N EPILAYERS
GROWN BY MOCVD
Yong-Hoon Cho*, G. H. Gainer*, J. B. Lam*, J. J. Song*, W. Yang**, and S. A. McPherson**
* Center for Laser and Photonics Research and Department of Physics
Oklahoma State University, Stillwater, OK 74078
»Honeywell Technology Center, 12001 State Highway 55, Plymouth, MN 55441
ABSTRACT
We present a comprehensive study of the optical characteristics of AlxGai.xN epilayers by
means of photoluminescence (PL), PL excitation, and time-resolved PL spectroscopy. All
AlxGai_xN epilayers were grown by metalorganic chemical vapor deposition and the Al mole
fraction (x) was varied from 0 to 0.6. We observed that (i) the full width at half maximum of the
PL emission, (ii) the energy difference between the PL emission peak energy and the PLE
absorption edge, and (iii) the effective lifetime increase with increasing x. These facts indicate
that degree of band-gap fluctuation due to a spatially inhomogeneous Al alloy content
distribution increases with increasing x. We observed anomalous temperature-induced emission
shift behavior for AlxGai_xN epilayers, specifically, an S-shaped (decrease-increase-decrease)
temperature dependence of the peak energy with increasing temperature. This anomalous
temperature-dependent emission behavior was enhanced as the Al mole fraction was increased.
Since the band-gap fluctuation in AlxGai-xN epilayers due to inhomogeneous spatial variations of
the Al content increases with increasing Al content, we believe that band-gap fluctuation causes
the PL peak energy to deviate from the typical temperature dependence of the energy gap
shrinkage. Therefore, the anomalous temperature-induced emission shift can be attributed to
energy tail states due to alloy potential inhomogeneities in the AlxGai.xN epilayers with large Al
content.
INTRODUCTION
Much interest has been focused on III-V nitride compound semiconductors and their
heterostructures due to their potential applications such as short-wavelength light emitting
devices [1,2] solar-blind ultraviolet detectors [3], and high power and high temperature devices
[4,5]. In particular, the ternary compound AlxGai-xN has the potential for use in light emitting
and detecting devices covering nearly the entire deep-ultraviolet (UV) region of the spectrum
(190 - 350 ran). The recombination mechanism for InGaN-based structures has been widely
investigated by several authors and detailed emission properties from localized states was
discussed for both spontaneous and stimulated emission of InGaN/GaN quantum structures [6,7].
However, the detailed spontaneous emission properties of the AlGaN-based structures have not
yet been investigated.
In this work, we report the results on optical properties of AlxGai_xN epilayers as a
function of Al content x (0 < x < 0.6) by means of photoluminescence (PL), PL excitation (PLE),
and time-resolved PL (TRPL) spectroscopy. By studying alloy epilayers, we can avoid (or
minimize) other ambiguous effects such as strain-induced piezoelectric polarization, quantum
confinement, layer thickness variations, and interface-related defects usually involved in the case
of quantum structures. We observed that the full width at half maximum (FWHM) of the PL
emission, the energy difference between the PL emission peak energy and the PLE absorption
457
Mat. Res. Soc. Symp. Proc. Vol. 572 @ 1999 Materials Research Society
edge, and the lifetime increase with increasing x from 0.17 to 0.60, indicating that the degree of
Al alloy potential fluctuations increases with increasing x. From the temperature-dependent PL
measurements, we observed that the PL emission from AlxGai.xN with high x did not follow the
typical temperature dependence of the energy gap shrinkage.
EXPERIMENT
The ALGai-xN epilayers used in this work were grown by low-pressure metalorganic
chemical vapor deposition (MOCVD) on (0001) oriented sapphire at a growth temperature of
1050 °C. Prior to AlxGai_xN growth, a thin ~ 5-nm-thick A1N buffer layer was deposited on the
sapphire at a temperature of 625 °C. Triethylgallium, triethylaluminum, and ammonia were used
as precursors in the AlxGa,.xN growth. The AlxGa,.xN layer thickness was about 1 urn. In order
to evaluate the Al alloy composition of the AlxGai.xN thin films, the samples were analyzed with
a high-resolution x-ray diffractometer using Cu Ka\ radiation. The angular distances between the
AlxGai_xN and A1N (or GaN) peaks were obtained by 0-29 scans. PL experiments were
performed using the 244 nm line of an intracavity doubled cw Ar+ laser as an excitation source.
PL and PLE experiments were also carried out using the quasimonochromatic light emission
from a Xe lamp dispersed by a lA m monochromator as the excitation source. Both the PL and
PLE experiments used a photomultiplier tube in conjunction with a 1 m double spectrometer as a
Wavelength (nm)
340
1
—i—r
300
320
1
—i—«
280
i—|—,—i—i—i— —i
260
1
1—
A^Ga^N epilayers
x = 0.17
x = 0.26
x = 0.33
x~0.6
'c
e
c
3.6
3.8
4.0
4.2
4.4
4.6
Photon Energy (eV)
Figure 1. 10K PL and PLE spectra for AlxGa,.xN epilayers with Al content x = 0.17, 0.26, 0.33, and
0.60. The PL spectra were measured using second-order diffraction through a monochromator and all
the spectra were normalized. Note that the Stokes shift (the energy difference between the PL emission
peak energy and the PLE absorption edge) increases with increasing x, indicating that the degree of
potential fluctuations increases with increasing x.
458
detector. TRPL measurements were carried out using a picosecond pulsed laser system
consisting of a cavity-dumped dye laser synchronously pumped by a frequency-doubled
modelocked Nd:YAG laser for sample excitation and a streak camera for detection. We observed
room temperature UV stimulated emission from optically pumped AlxGai.xN epilayers with Al
content as high as 26 %. Detailed stimulated emission properties were reported elsewhere [8].
RESULTS AND DISCUSSIONS
Figure 1 shows 10 K PL and PLE spectra for AlxGai_xN epilayers with Al content
x = 0.17, 0.26, 0.33, and 0.60 with PL peak energies of ~ 3.70, 3.92, 4.22, and 4.76 eV,
respectively. The PL spectra were measured using second-order diffraction through a
monochromator and all the spectra were normalized. The decrease in PLE signal above the PLE
peak position with increasing excitation energy is due to the decrease in the excitation intensity
of a Xe lamp source. Note that the FWHM of the PL emission and the Stokes shift (the energy
difference between the PL emission peak energy and the PLE absorption edge) monotonically
increase with varying x from 0 to 0.60, as shown in Fig. 1.
TRPL and time-integrated PL (TIPL) results measured at 10 K are shown in Fig. 2 for
GaN, Alo.17Gao.83N, and Alo.33Gao.67N epilayers. In the case of the GaN epilayer, the lifetime of
the free exciton (FX) is about 30 ps while that of the bound exciton (BX) is about 40 ps. In the
case of both the Alo.17Gao.83N, and Alo.33Gao.67N epilayers, in contrast, the measured lifetime
increases with decreasing emission energy, and hence, the peak energy of the emission shifts to
the low energy side as time proceeds. This behavior is most likely due to alloy potential
fluctuations. We observed that the FWHM is about 4, 17, and 48 meV and the overall lifetime is
about 30, 250, and 450 ps for the GaN, Alo.17Gao.g3N, and Alo.33Gao.67N epilayers, respectively.
These facts indicate that the alloy potential fluctuation of the Alo.33Gao.67N epilayer is larger than
that of the Alo.17Gao.83N epilayer. These facts obtained from Figs. 1 and 2 indicate that the degree
Wavelength (nm)
339
336
333
...'
i
»
1
296
1
'"
500
1
1
292
1
1
1
1
1
1
0.33^^>.67
c
800
400
ATJVT
600
300
400
200
200
100
n
3.48
3.50
r-r 1 . . r*Z-
3.66
3.69
3.72
• ]\
-
/
y
n
c
>
\
<
4.16 4.20 4.24 4.28
Photon Energy (eV)
Figure. 2. 10 K time-resolved and time-integrated PL results for (a) GaN, (b) Alo.17Gad.83N, and (c)
AI0.33Gao.67N epilayers. A PL FWHM of about 4, 17, and 48 meV and an overall lifetime of about 35,
230, and 450 ps were obtained for GaN, Alo.17Gao.g3N, and Alo.33Gao.67N epilayers, respectively.
459
of Al alloy potential fluctuations increases with increasing x.
Generally, the PL peak energy (En) follows the well-known temperature dependence of
the energy gap shrinkage: Eg(T) = Eg(0) -aT2/(ß+T), where Eg(T) is the band-gap transition
energy at temperature T, and a and ß are known as the Varshni thermal coefficients [9]. The
parameters a= 8.32 x 10"4 eV/K and ß = 835.6 K for the GaN r9v - r7c transition were
previously extracted by photoreflectance studies [10]. The temperature-dependent PL peak shift
for the GaN and AlxGai-xN layers with small x value (x < 0.1) was consistent with the estimated
energy decrease, while the PL emission from Al„Gai.xN with higher x value did not follow the
typical temperature dependence of the energy gap shrinkage. The anomalous behavior of photon
energy as a function of temperature in the Alo.17Gao.g3N and Alo.33Gao.67N epilayers is shown in
Fig. 3. For the Alo.17Gao.83N (Alo.33Gao.67N) epilayer, with increasing temperature up to 20 (90) K
(region I), an initial small decrease in EPL was observed, followed by an increase in EPL in the
temperature range of 20 - 70 K (90 - 150 K) (region II) and finally a decrease again in the
temperature above 70 (150) K (region III). In the case of the Alo.33Gao.67N epilayer, EPL more
clearly shows the S-shaped (redshift-blueshift-redshift) behavior with increasing temperature. As
345
Wavelength (nm)
340
335
300
Wavelength (nm)
295
290
c
e
«0.
e
a.
3.55
3.60
3.65
3.70
3.75
Photon Energy (eV)
4.10
4.15
4.20
4.25
Photon Energy (eV)
4.30
Figure 3. Evolution of the PL spectra for the (a) Alo.17Gao.83N and (b) Alo.33Gao.67N epilayers over a
temperature range from 10 to 300 K. The main emission peak of both samples (closed circles) shows
an S-shaped shift with increasing temperature. All spectra are normalized and shifted in the vertical
direction for clarity. For the Alo.nGao.83N (Alo.33Gao.67N) epilayer, with increasing temperature up to 20
(90) K, an initial small decrease in peak energy was observed, followed by an increase in energy in the
temperature range of 20 - 70 K (90 - 150 K) and finally a decrease again in the temperature above 70
(150) K.
460
the temperature increases from 10 to 90 K for the Alo.33Gao.67N epilayer, EPL redshifts 8.6 meV.
This value is similar to the expected band-gap shrinkage of- 7.2 meV for the GaN epilayer over
this temperature range [10]. For a further increase in temperature from 90 to 150 K, the PL peak
blueshifts 20.2 meV for the Alo.33Gao.67N epilayer. If we consider the temperature-induced bandgap shrinkage of ~ 11.7 meV expected for the case of GaN, the actual blueshift of the PL peak
with respect to the band-edge is about 31.9 meV over this temperature range. When the
temperature is further increased above 150 K, the peak position redshifts again. From the
observed redshift of 31.7 meV combined with the expected band-gap shrinkage of ~ 47 meV
between 150 and 300 K, we estimate an actual blueshift of the PL peak relative to the band-edge
to be about 15.3 meV in this temperature range.
Consequently, the PL emission in region I exhibits a characteristic luminescence behavior
from lower energy tail states caused by an inhomogeneous spatial band-gap distribution in the
alloy materials. In region II, the carriers populate higher-energy tail states. In region III, the
emission is^due to thermally populated carriers at higher-energy states and follows the typical
temperature dependence of the energy gap again. We note that the PL peak energy deviation
from the £g(0) - a T 2l(ß+ T) function and the temperature ranges of regions I and II increase
with increasing Al content of the AlxGai.xN epilayers. Since the band-gap fluctuation in AlxGaiXN epilayers due to alloy potential variations of the Al content increases with increasing Al
content, we believe that band-gap fluctuation causes the PL peak energy to deviate from the
typical temperature dependence of the energy gap shrinkage. Therefore, the temperature-induced
emission shift observed in this work for AlxGai_xN epilayers with large Al content is attributed to
energy tail states due to inhomogeneous alloy potential fluctuations. We note, however, that we
did not observe an actual "effective" redshift with respect to the fundamental energy gap in
temperature region I, in contrast to the case of InGaN/GaN multiple quantum wells (MQWs)
[11]. In the InGaN/GaN MQWs, the redshift energy was about five times larger than the
expected band-gap shrinkage over the temperature range I and was attributed to the carrier
relaxation process into the localized states in the InGaN/GaN MQWs [11]. Further detailed
studies are in progress to clarify the relationship between the observed anomalous temperatureinduced emission shift and temperature-dependent carrier dynamics in the AL.Gai.xN alloy.
CONCLUSIONS
We have investigated the optical characteristics of MOCVD-grown AlxGai.xN
(0 < x < 0.6) epilayers by means of PL, PLE, and TRPL spectroscopy. We observed that (i) the
FWHM of the PL emission, (ii) the Stokes shift between the AlxGai.xN PL peak energy and the
band-edge obtained from PLE spectra, and (iii) the effective lifetime increase with increasing x.
These facts indicate that the degree of band-gap fluctuation due to a spatially inhomogeneous Al
alloy content distribution increases with increasing x. We observed anomalous temperatureinduced emission shift behavior for AlxGai.xN epilayers, specifically, an S-shaped (decreaseincrease-decrease) temperature dependence of the peak energy with increasing temperature. Both
the PL peak energy deviation from the typical energy gap temperature dependence and the
temperature range in which the anomalous emission shift occurs increase with increasing Al
content of the AlxGai_xN epilayers. This is attributed to spatially inhomogeneous Al content
variations causing more band-gap fluctuation with increasing Al content. Therefore, the
anomalous temperature-induced emission shift is related to thermal population of energy tail
states due to alloy potential inhomogeneities in the AlxGai.xN epilayers.
461
ACKNOWLEDGMENTS
This work was supported by AFOSR, BMDO, ONR, and DARPA.
REFERENCES
1. S. Nakamura, M. Senoh, N. Iwasa, S. Nagahama, T. Yamada, and T. Mukai, Jpn. J. Appl.
Phys.34,L1332(1995).
2. S. Nakamura, M. Senoh, S. Nagahama, N. Iwasa, T. Yamada, T. Matsushita, Y. Sugimoto, and
H. Kiyoku, Appl. Phys. Lett. 69,4056 (1996).
3. B. W. Lim, Q. C. Chen, J. Y. Yang, and M. A. Khan, Appl. Phys. Lett. 68,3761 (1996).
4. Y. F. Wu, B. P. Keller, S. Keller, D. Kapolnek, P. Kozodoy, S. P. DenBaars, and U. K.
Mishra, Appl. Phys. Lett. 69,1438 (1996).
5. X. H. Yang, T. J. Schmidt, W. Shan, J. J. Song, and B. Goldenberg, Appl. Phys. Lett. 66, 1
(1995).
6. Y. H. Cho, G. H. Gainer, A. J. Fischer, J. J. Song, S. Keller, U. K. Mishra, and S. P. Denbaars,
Appl. Phys. Lett. 73,1370 (1998).
7. T. J. Schmidt, Y. H. Cho, G. H. Gainer, J. J. Song, S. Keller, U. K. Mishra, and S. P.
Denbaars, Appl. Phys. Lett. 73, 560 (1998).
8. T. J. Schmidt, Y. H. Cho, J. J. Song, and W. Yang, Appl. Phys. Lett. 74, 245 (1999).
9. Y. P. Varshni, Physica34,149 (1967).
10. W. Shan, T. J. Schmidt, X. H. Yang, S. J. Hwang, J. J. Song, and B. Goldenberg, Appl. Phys.
Lett. 66, 985 (1995).
11. Y. H. Cho, G. H. Gainer, A. J. Fischer, J. J. Song, S. Keller, U. K. Mishra, and S. P.
DenBaars, Appl. Phys. Lett. 73,1370 (1998).
462
COMPARATIVE STUDY OF GAN GROWTH PROCESS BY MOVPE
Jingxi Sun, J.M. Redwing*, T. F. Kuech
Department of Chemical Engineering, University of Wisconsin-Madison, Madison, WI53706
*Epitronics, Phoenix, AZ85027
ABSTRACT
A comparative study of two different MOVPE reactors used for GaN growth is presented.
Computational fluid dynamics (CFD) was used to determine common gas phase and fluid flow
behaviors within these reactors. This paper focuses on the common thermal fluid features of
these two MOVPE reactors with different geometries and operating pressures that can grow
device-quality GaN-based materials. Our study clearly shows that several growth conditions
must be achieved in order to grow high quality GaN materials. The high-temperature gas flow
zone must be limited to a very thin flow sheet above the susceptor, while the bulk gas phase
temperature must be very low to prevent extensive pre-deposition reactions. These conditions
lead to higher growth rates and improved material quality. A certain range of gas flow velocity
inside the high-temperature gas flow zone is also required in order to minimize the residence
time and improve the growth uniformity. These conditions can be achieved by the use of either a
novel reactor structure such as a two-flow approach or by specific flow conditions. The
quantitative ranges of flow velocities, gas phase temperature, and residence time required in
these reactors to achieve high quality material and uniform growth are given.
INTRODUCTION
Gallium nitride (GaN)-based materials have been used successfully for fabricating shortwavelength optoelectronic as well as high-temperature and high-power devices. Metalorganic
vapor phase epitaxy (MOVPE) is the primary technique to grow these thin films and related
devices. However, the growth of GaN-based materials has been problematic. High growth
temperatures, gas phase pre-deposition reactions, and subsequent gas flow complexity
complicate the MOVPE growth of GaN. These complicating factors have impeded the design
and scale-up of MOVPE reactors for GaN-based material growth. The current understanding of
GaN MOVPE growth is limited to the influence of specific reactor geometries and growth
conditions on the growth rate and uniformity. The understanding of the dominant reaction
pathways and their interaction with transport phenomena has been insufficient for design and
optimization of MOVPE reactors for GaN growth.
In order to investigate the growth mechanism, the MOVPE of GaN has been extensively studied
using various numerical modeling techniques [1]. Based on theoretical calculation and analysis
of experimental data, kinetic mechanisms of differing complexity, which include a number of
gas phase and surface reactions, have been proposed in these modeling studies. Typically, the
proposed mechanism is substantiated by showing the consistency with a very limited set of
experimental observations, usually growth rate data. Due to the complexity of the proposed
growth mechanism and very limited growth-based data available for comparison, it is very
difficult for these modeling investigations to provide general information which can be directly
applied to practical technology, such as the design, scale-up and optimization of reactors capable
of producing deposition rate and composition uniformity necessary for fabrication of various
GaN-based devices.
463
Mat. Res. Soc. Symp. Proc. Vol. 572 • 1999 Materials Research Society
In this study, we studied the MOVPE of GaN-based materials by employing an alternative
modeling strategy. Since trimethylgallium (TMG) is very dilute in the NH3: H2 mixture, the gas
phase reactions have almost no affect on the thermal fluid flow behaviors. However, the growth
chemistry pathways are significantly determined by the thermal flow conditions, and can lead to
deleterious pre-deposition reactions for GaN growth. Therefore, we performed, without any gas
and surface reactions involved, a comparative study of the thermal flow behavior within two
different reactors that can grow device quality GaN-based materials. We identified some
common features of thermal flow inside these two different reactors. Based on these thermal
flow features, we are further studying the growth chemistry in a step-wise fashion, with an
increasing chemical complexity in the gas phase at each step, in an effort to discover how these
observed common thermal flow features determine the growth pathways. Compared to a typical
growth chemistry study, this step-wise growth chemistry investigation provides significant
insight into the primary reactions in the GaN MOVPE processes. Based on combined modeling
and experimental effort, we aim at providing a general model of GaN growth. In this paper, we
will address our study on thermal flow behaviors inside GaN reactors. Unlike the typical III-V
reactor, the GaN reactor involves much higher growth temperature and more extensive predeposition gas phase reactions, therefore the primary focus of this investigation are the flow
velocity distribution, temperature profile and residence time.
REACTORS & EXPERIMENT
Schematic diagrams of the two reactors used in this study are shown in Fig.l and Fig.2. The first
reactor we studied is a working vertical reactor, the second one is the two-flow reactor invented
by Nakamura [2]. We have detailed operating conditions for the vertical reactor, while our
simulation study of the two-flow reactor is based on the incomplete
experimental data distributed in several publications and patents by
Nakamura
[2] [3] [4].
Therefore, we present
Nj+H
our detailed study on
the vertical reactor.
The simulation of the
Y¥al6r
IR RADIATION
THERMOMETE1
T
cooled
two-flow reactor is
s
used as a comparison
uscep r to verify the features
H2 +NH3 + TMG
present in the vertical
reactor.
For the
Figure 1. Diagram of
vertical
reactor,
the vertical reactor
trimethylgallium
(TMG), in a hydrogen VACUUM
J
EXHAUSTcarrier gas, is supplied through the inner tube while
ammonia and hydrogen are supplied in the outer
tube. The outer wall of the reactor is watercooled. The graphite susceptor is inductively Figure 2. Diagram of the two-flow reactor ([2])
heated. The inner diameter of the reactor is 85
mm and the outer diameters of the two inlet tubes are 6.4 mm and 25.4 mm respectively. The
opening of the inner tube is widened at the end above the susceptor, which is 70 mm in diameter.
This reactor is operated at a pressure of 100 Torr and a susceptor temperature of 1100°C. The
coolant and the inlet gas temperature are assumed to be at room temperature. The films were
2
464
grown on 2-inch c-plane sapphire. The two-flow reactor is operated at 760 torr and a susceptor
temperature of 1000°C. The growth reactants (TMG + NH3 + H2) are supplied from a horizontal
inlet, and a vertical pushing gas (H2 + N2) is added in order to confine the reactants to the
growing surface [2]
MODELING
The modeling study was performed using a multi-purpose computational fluid dynamics code,
which includes conjugate heat transfer and radiation [5,6]. The fundamental equations of
continuity, momentum, and energy balances and species conservation are used to describe the
system [7]. The reactor model is based on numerical solution of the nonlinear, coupled partial
differential equations representing the conservation of momentum, energy, total mass and
individual species using the finite element method (FEM). The FEM transport model solves for
flow and heat transfer (including conduction in the walls, convection, and radiation) for the
reactor configuration. The thermal diffusion component that drives high molecular species away
from hot regions, towards cold regions, has also been included because of the high growth
temperature.
RESULTS
The detailed results for the
vertical reactor are initially
presented, followed by brief
-x=0
results on the two-flow reactor
5tl20O
O
— -X=0.4
for comparison. Figure 3 and 4
31000
X=1.0
show the temperature and flow
a
\
profile inside the vertical reactor
1 «00
V\
L
under the standard operating
400
Vv,
conditions. From Fig. 3(a), we
Height above susceptor (cm)
can see that the hightemperature flow zone is
confined to a thin gas flow sheet
above the susceptor. The
(b)
(a)
temperature decrease from
growth temperature to - 400K
Figure 3. Temperature profiles for the vertical reactor, (a)
within a very short distance
Temperature contour, (b) Temperature distribution above
away from the growth surface,
susceptor at three radial position x = R/Rwafer.
resulting in a very high
temperature gradient above the
growth surface. The bulk of the gas phase temperature remains around the inlet temperature,
which is about room temperature. Figure 4 (a) show the corresponding streamlines. Typically,
there are recirculations above the susceptor due to the high input velocity of the inner TMG feed
tube [8]. These recirculations could lead to trapping of paniculate matter, which may
subsequently deposit onto the film. The removal of recirculations for this case is accomplished
by changing the geometry. TMG flow velocity is lowered with a much wider opening at the end.
This modification improves the uniformity and quality of the deposited film on the substrate.
The radial velocity becomes increasingly uniform when the flow approaches the growth surface,
which is also a necessary requirement for the deposited film uniformity. In order to make
comparison between the two investigated reactors, we quantify the above thermal fluid behavior.
465
From Fig. 3 (b) and
Fig. 4(b), we can see
that the thickness of the
high-temperature zone
is about 1-2 cm, while
the gas flow velocity
inside the high
temperature flow zone
is about 0.01-0.6 m/sec.
D
C
O 1.5
Ü
0)
Ö)
ii.o
>
Y1=1.5mm
Y2=12.3mm
\.
Y3=15.9mm
Y4=34.3mm
N
-
N.
- %-__-._--
©
.
i
i
The
simulation
Radial Distance (R/Rwa(er)
geometry for the twoflow reactor and a
(b)
(a)
typical
temperature
profile is shown in
Figure 5. By comparing Figure 4. Flow profile for the vertical reactor, (a): Streamline,
to the corresponding (b): Velocity radial distribution above susceptor.
temperature profiles in
the vertical reactor, similar behaviors can be identified. A very thin high-temperature flow sheet
is also formed above the susceptor. The thickness of the high-temperature flow sheet is about
1.0-1.5 cm. The gas temperature outside this flow sheet is below 400K. Similar to the vertical
reactor, a uniform flow velocity
N2 + H2
_L
distribution is observed above the growth
surface. The gas flow velocity above the
NH,+ TMG + H2
susceptor from our simulation is in the
range of the reported value [4], 0.02-0.5
m/sec.
We further investigate these common
thermal fluid behaviors by studying
thermal diffusion and residence time
within the vertical reactor. The high
temperature gradient above the growth
surface leads to high thermal diffusion
coefficients.
High thermal diffusion
effects can reduce the concentration of
higher molecular weight species near the
growth front. Figure 6 indicates the
impact of a change in thermal diffusion
in the TMG concentration profile above
the substrate. The TMG concentration on
Figure 5. Temperature contour for the twothe substrate surface is significantly
flow reactor.
reduced by the high thermal diffusion
effect in this reactor. Since the reactant residence time affects the extent of gas phase reactions,
and hence the quality of the deposited film, the residence time is a very useful parameter to study
the thermal fluid behavior for GaN MOVPE reactors. We studied the residence time by
integrating the trace of released particles from the wide opened end of the inner TMG tube until
the particle leave the susceptor edge. Figure 7 shows the particle trace and radial distribution of
the residence times. The molar ratio of TMG to ammonia at every particle releasing point is also
466
included. From these results, we
can see that the total residence time
for TMG within this reactor is less
than 0.3 second. The residence
time decreases at higher ratio of
TMG to NH3.
With thermal diffusion
Without thermal diffusion
i
i
f
DISCUSSION
f
-
The most commonly used MOVPE
precursors
for
GaN,
trimethylgallium
(TMG)
and
0.2
0.4
0.6
0.8
1.0
ammonia, react very rapidly to
Radial Distance (R/Rwafer)
form an adduct, even at room
temperature. These adducts and
subsequent
adduct-related
Figure 6. The effect of thermal diffusion on the radial
reactions lead to changes in the
molar distribution of TMG on the substrate
nature and concentration of
reactants. The gas phase depletion
of the actual growth precursors can result in poor growth uniformity, and most probably material
quality. In addition, these adduct-related reactions take place away from the heated susceptor.
Additional parameters, such as the design of reactor inlet mixing critically influences the
(58)
300
123456
Point
1
2
3
4
5
6
R(mm)
0
0.5
1.5
3.0
4.0
8.0
'ft
200
^ NT
100
n
Molar Ratio of NH, to TMG
A(6»r
(204) (373) ^
—*
\
1
'
0
2
4
6
8
Radial Distance (mm)
(a)
(b)
Figure 7. (a) Particle releasing trace, (b) Residence time distribution. < are the total
residence time, while 6 are the residence time inside high-temperature zone (T > 400 K).
epitaxial growth.
From our simulation results, we can identify some common thermal flow features that are
observed both in a working vertical reactor and the two-flow reactor. A thin high-temperature
flow sheet of ~ 1 - 2 cm thick is formed above the susceptor, while the bulk of the gas phase
467
temperature is below 400K. Within the high-temperature flow sheet, the flow velocity is very
uniform and in a certain range of 0.01-0.6 m/sec. Since these two reactors have totally different
geometries and are operated under different pressures (100 Torr vs. 760 Torr), we believe that
these common features might be correlated to growth of device quality GaN-based materials.
Prevention of the adduct-related gas phase reactions plays a critical role in the GaN growth.
Reduction in the gas phase temperature and minimization of residence time are very effective
ways to prevent these pre-deposition reactions. The bulk gas phase temperature inside these two
reactors is kept at a low temperature of 300 ~ 400K. This temperature profile will minimize the
pre-deposition reactions, or delay these reactions until the high-temperature flow zone. Since
the high-temperature flow zone is restricted to a thin flow sheet right above the susceptor, the
reactants react vigorously within this high-temperature zone with a very short residence time,
leading to film deposition on the substrate. There is a small residence time of the reactants where
the NH3 and TMG are well mixed within the high temperature flow zone. This small contact
time reduces the extent of the pre-deposition reactions. This specific temperature profile is
achieved by a high TMG flow from the inner input tube in the vertical reactor, and by the
pushing gas flow in the two-flow reactor.
CONCLUSIONS
A comparative study of GaN growth by MOVPE was performed to identify the common thermal
fluid behaviors for MOVPE reactors that are being used to grow device quality GaN-based
materials. The comparison of the temperature and flow profiles for two high-performance
reactors has revealed several common features for the thermal flow behaviors inside these
reactors, which can be served as general engineering guidelines for design and optimization of
MOVPE reactors for GaN growth.
ACKNOWLEDGEMENTS
This work is supported by the ONR-MURI on compliant substrates and the EPRI-DARPA
initiative on high power devices.
REFERENCES
1. H. Jurgensen, D. Schmitz, G. Strauch, E. Woelk, M. Dauelsberg, L. Kadinski, Y. N.
Markarov, MU-NSR, (USA), Vol. 1 Art. 26 (1996).
2. S. Nakamura, Jap. J. Appl. Phys. Vol. 30 (8), (1991) 1620
3. Nakamura, T. Mukai, and M. Senoh, Appl. Phys. Lett. Vol.64, (1994) pl687.
4. S. Nakamura, United States Patent, No. 5334227, (1994).
5. H. M. Manasevit, J. Crystal Growth, 13/14, (1972) p306.
6 . CFDACE theory manual, CFDRC incorporation, Version 5 (1998).
7. R. B. Bird, W. E. Stewart and E. N. Lightfoot, Transport Phenomenon, Wiley, New York
(1960).
8. S. A. Safvi, J. M. Redwing, M. A. Tisher, T. F. Kuech, J. of Electrochem. Soc. 144(5) (1997)
1789.
468
PartV
GaN Devices and Processing
AlGaN Microwave Power HFETs on Insulating SiC Substrates
Gerry Sullivan, Ed Gertner, Richard Pittman, Mary Chen, Richard Pierson, Aiden
Higgins; Rockwell Science Center, Thousand Oaks, CA: Qisheng Chen; APA Optics,
Blaine, MN: Jin-Wu Yang; University of South Carolina, Columbia, S.C.: R. Peter
Smith, Raul Perez, Abdur Khan; JPL/Caltech, Pasadena, CA: Joan Redwing, Brian
McDermott; Epitronics/ATMI, Danbury, CT: Karim Boutros, Spectrolab, Sylmar, CA.
ABSTRACT
AlGaN HFETs are attractive devices for high power microwave applications.
Sapphire, which is the typical substrate for AlGaN epitaxy, has a very poor thermal
conductivity, and the power performance of AlGaN HFETs fabricated on sapphire
substrates is severely limited due to this large thermal impedance. We report on HFETs
fabricated on high thermal conductivity electrically insulating SiC substrates. Excellent
RF power performance for large area devices is demonstrated. On-wafer CW
measurements at 10 GHz of a 320 micron wide FET resulted in an RF power density of
2.8 Watts/mm, and measurements of a 1280 micron wide FET resulted in a total power
of 2.3 Watts. On-wafer pulsed measurements, at 8 GHz, of a 1280 micron wide FET
resulted in a total power of 3.9 Watts. Design of a hybrid microwave amplifier will be
discussed.
INTRODUCTION
Wide band gap semiconductors, such as AlGaN and SiC, are being developed for
a variety of applications. These materials have properties that make them very attractive
for microwave power generation. In Table 1, some of the materials properties of SiC
and GaN are compared to Si and GaAs. As can be seen, the band gaps and breakdown
fields of GaN and SiC are similar, and are much larger than those of GaAs and Si. The
maximum electron velocity in GaN is predicted to be a bit larger than that in SiC. The
low field electron mobility in bulk samples of SiC and GaN is also similar. At an AlGaN
/ GaN heteroj unction, however, the electron mobility can be much higher. Room
temperature Hall mobilities in excess of 2000 cm2/Vs have been measured in AlGaN /
GaN heterojunction structures, with corresponding sheet charge densities of 5 xlO12 cm"2.
This much larger mobility is a distinct advantage for AlGaN / GaN HFETs, relative to
SiC devices.
In power transistors, efficient thermal management is critically important. Table
1 shows that the thermal conductivity of SiC is much higher that that of GaN. This listed
thermal conductivity for SiC (4.9 W/cm K) is the value for very high purity, lightly Ntype material. Currently, insulating SiC has a somewhat lower conductivity (3.3 W/cm K
[1]), but this value should improve as the SiC substrate quality improves. Sapphire,
which is the typical substrate for AlGaN epitaxy, has a thermal conductivity of only 0.2
W/cm K. This very poor thermal conductivity severely limits the performance of RF
power HFETs on sapphire substrates. Growth of AlGaN HFET structures onto
electrically insulating SiC substrates should combine the superior electrical properties of
the AlGaN / GaN heterostructure with the superior thermal properties of bulk SiC,
resulting in high performance microwave power transistors.
471
Mat. Res. Soc. Symp. Proc. Vol. 572 ® 1999 Materials Research Society
Table 1. Selected materials properties of SiC and GaN, compared to Si and GaAs. (aelectron mobility in bulk, lightly N-type GaN; b- electron mobility at an AlGaN / GaN
heterojunction)
Si
GaAs
4H-SiC
GaN
Bandgap (eV)
Breakdown field [xlO5 V/cm]
Electron mobility [cmVVs]
1.1
2
1400
1.4
4
8500
3.3
8
800
Maximum velocity [xlO7 cm/s]
(E=8 E5 V/cm)
Thermal conductivity [W/cm K]
0.5
0.7
1.0
3.4
8
900"
2000 b
1.4
1.5
0.5
4.9
1.3
Property/Material
EXPERIMENT
The nitrides display a large piezoelectric effect, relative to other III-V materials
[2]. For the (0001) orientation typically used for epitaxial growth, the piezoelectric effect
is a major source of charge at AlGaN / GaN heterojunctions. Figure 1 shows that the
electron charge density densities in undoped AlGaN / GaN heterostructures increase
linearly with increasing AlGaN alloy composition, and densities well in excess of 1 xlO13
cm"2 can be achieved. This sheet charge density is several times larger than is typically
seen in AlGaAs / GaAs heterostructures. Because the mobility of electrons in AlGaN /
GaN heterostructures are several times lower than the mobility of electrons in AlGaAs /
GaAs, the total current for a given gate width is very similar for the two materials
systems. The drain breakdown voltage for the nitride structure is many times larger than
for the arsenide structure, for similarly sized transistors.
The structures reported on in this paper were all grown by low pressure MOCVD
using the standard metalorganic precursors for the group III material, ammonia for the
column V material, and silane for the dopant. Adding intentional donors to the AlGaN
layer increases the electron sheet charge density, as expected. Figure 2 shows the
measured Hall mobility versus sheet charge density for a variety of structures grown on
SiC substrates, some of which contain intentional doping in the AlGaN. The structure
with a 300K sheet charge density of 4.9 xlO12 cm"2 and a mobility of 2150 cm2/Vs, has a
77K sheet charge density of 4.9 xlO12 cm'2 and a mobility in excess of 12,000 cm2/Vs.
This large low temperature mobility attests to the high quality of the heterointerface.
472
^ 2x1(f
"E
,0,
1
'' 'i'''' r '
MOCVD
,|,, i,
fk.
m
:
? ixid3
E>
n
I
-
/
5x1012
(0
A
!| 1 1
AMBE
j* 1-5x1 d3
"55
c
I
II,,,.,
I i i i t
..!....
I '—
t i.,,,
I >
i,,
I t
i
0
0.1
I ,i >i Ii Ii 1I M
i i.
0.2
1 .1 . . .. i —
0.3
0.4
Aluminum Concentration \X\
Figure 1. Two-dimensional electron sheet charge density at AlGaN / GaN
heterostructures, as a function of the alloy composition of the AlGaN [2].
2500
Sheet Charge Density [cm"2]
Figure 2. Room temperature Hall mobility versus sheet charge density for a variety of
AlGaN/GaN structures grown by MOCVD on SiC substrates at Rockwell.
473
The epitaxial layer structure of the devices reported on in this paper consisted of a
50 Angstrom undoped AlGaN cap, a 100 Angstrom AlGaN region doped with 5 xlO
donors/cm2, a 50 Angstrom undoped spacer layer, a one micron thick GaN layer, and a
1000 Angstrom thick A1N nucleation layer. The aluminum alloy concentration in the
AlGaN is 25%. This structure is grown on a 350 micron thick electrically insulating 4H
SiC substrate. The structure has a room temperature sheet charge density of about 1
xlO13 cm"2, and a Hall mobility of 1350 cm2/Vs.
HFETs were fabricated in this layer structure. Implantation was used for
isolation. Ohmic contacts consist of Ti-Al annealed at 950 °C. The optically defined PtAu gate is 0.6 microns long. A microphotograph of the layout of the 1280 micron wide
device reported on in this paper is shown in Figure 3. It consists of 16 gates, each of
which is 80 microns wide. Successive sources are interconnected using air bridges.
Drain
Ground (Source)
Gate
Ground (Source)
Figure 3. Microphotograph of a 1280 micron wide AlGaN HFET.
RESULTS
Shown in Figure 4 are the transistor characteristics for HFETs on sapphire and
SiC. The thermally induced collapse of the drain current at high power that occurs for
the device on sapphire is largely absent for the device on SiC. This demonstrated the
efficacy of the SiC at conducting the heat away from the HFET. The difference in the
absolute magnitude of the drain currents is not due to the different substrates, but is
related to differences in the layer structure during the two different growths. The HFET
on SiC has a current density of about 1 amp/mm.
474
Ü.Ü8
©
W
p
o
4k
O
o
~~~
.
o
Drain Current [A]
Cn
O
o
(
i
S
o.t
o-J
litt
1
1
1
1
15
Drain to Source Voltage [V]
20
Drain to Source Voltage [V]
Figure 4. Comparison of the transistor characteristics of an 80 micron wide AlGaN HFET
on a sapphire substrate (a) and a SiC substrate (b).
The transconductance versus gate voltage for the HFET on SiC shown in Figure
4b is shown in Figure 5. The peak transconductance is 240 mS/mm, and the pinch-off
voltage is about -4.2 Volts.
250
E
|
/
.*
VL
200
'V
V
i,
M.
£
150
»
%
3
•o
c
o
u
(0
I
100 —t
I-
__J
\—
•
•
•
50
-4-3-2-10
1
Gate to Source Voltage [V]
Figure 5. Transconductance versus gate voltage for the 80 micron wide HFET on a SiC
substrate shown in Figure 4b. The drain bias was 7 Volts, and the source to drain spacing
is 2 microns.
475
The 10 GHz gain compressions characteristics for the HFETs on SiC are shown in
Figure 6. These measurements were done on the wafer using controlled impedance
probes. The small signal gain is in excess of 10 dB. With increasing drain bias, the
output power saturates as expected. Increasing the drain bias voltage does little to
improve the maximum output power. This is believed to be due to the high thermal
impedance for a wafer resting on a metal chuck.
Input Power [dBm]
Figure 6. CW on-wafer 10 GHz gain compression curves for a 240 micron wide HFET on
a SiC substrate. The source to drain spacing is 3 microns.
On-wafer CW 10 GHz power testing was done on these HFETs on SiC substrates.
To reduce the thermal impedance, the SiC substrate was attached to the metal chuck
using thermal grease. A power density of 2.8 Watts/mm was measured in a 320 micron
wide FET. A total power of 2.3 Watts was measured in a 1280 micron wide FET.
Pulsed testing of devices is a way to measure the devices while they are still in
wafer form, and to estimate the performance of the devices once they have been properly
packaged with efficient thermal management. The pulsed measurements were done by
applying a 10 microsecond DC pulse to the drain with a 0.1% duty cycle, and a 20
microsecond 8 GHz pulse to the gate which symmetrically overlapped the drain bias
pulse. Output tuners were used to match the fundamental frequency. As shown in Figure
7 a 1280 micron wide device with 3 micron source-to-drain spacing produced a
maximum power of 3.9Watts (3W/mm) with 7.7 dB gain and 26% P.A.E. The drain bias
was 32Volts and the gate bias was -1.5 V. A somewhat lower drain bias (25 V) resulted
in a somewhat higher 30% P.A.E., producing 3.3 Watts (2.6W/mm) with 8.5 dB gain.
476
40
30
35
—• —Efficiency
—D— Power
25
£
2, 30-|
O
CO
20
<D
o
0.
>
Q.
a
25
15
10
3
o
15
5
at
Q.
m
g
a>'
o
•^
<
,.-*
10
-1—
10
-1—
15
-r—
20
l
25
30
Input Power (dBm)
Figure 7. Pulsed on-wafer 8 GHz gain compression curves for a 1280 micron wide
HFET on a SiC substrate. The source-to-drain spacing is 3 microns, Vds is 32V and Vgs is
-1.5V.
Large signal 10 GHz measurements (Load pull) were done on these transistors to
determine optimal RF impedances. As shown in Figure 8, the input impedance is quite
low, as expected for a large power transistor with a comparatively long optically-defined
gate. Reducing the gate length using electron beam lithography will move the input
impedance closer to 50 Ohms, making the input matching easier and more efficient. This
low input impedance will probably limit the size of practical power HFETs to a few
millimeters of gate width.
The input impedance has also been rotated clock-wise around the Smith chart,
relative to a similarly sized GaAs HFET, so that its impedance is almost entirely real.
This rotation for the AlGaN HFET is believed to be a reflection of the lack of throughsubstrate via holes. The ground connection for the AlGaN HFETs source, instead of
being a low inductance via, is a long front-side metal line. This extra impedance between
the source and ground is very undesirable, and degrades the RF performance of the
existing HFETs.
The optimal output impedance for the 1280 micron wide HFET is high, relative to
a similarly sized GaAs HFET, reflecting the large drain bias possible with the AlGaN
HFET. The optimal impedance for maximum power moved even closer to 50 Ohms with
increasing drain bias, as expected. The output impedance mismatch which degrades the
477
total power by one dB is a fairly large circle, making the output matching for these
transistors fairly simple.
Optimal input impedance
Optimal large signal
output impedance
1 dB down output
power contour
Figure 8. Large signal on-wafer impedance measurements of 1280 micron wide HFETs,
optimizing for maximum power. The drain bias is 25 Volts, and the source-to-drain
spacing is 3 microns.
Shown in Figure 9 is a 9 GHz hybrid amplifier designed using two of these 1280
micron wide HFETs. The passive matching components are fabricated on 250 micron
thick GaAs substrates using microstrip waveguide format. The input and output are
matched to 50 Ohms. As can be seen, the input matching requires considerably more
circuitry than does the output matching, reflecting the greater impedance mismatch on the
gates of the HFETs relative to the drains. Also, the total area required for the passive
matching components is much larger that that required for the active area of the
transistors. This relatively large area for the passive matching components at 10 GHz is a
strong argument against fabricating MMICs, rather than hybrids, given the current high
cost and small wafer size of electrically insulating SiC. At higher frequencies, the
performance degradation associated with a hybrid design will force MMIC or flip-chip
designs to be used, in spite of the costs.
478
=1
Q.
Figure 9. 9 GHz Hybrid amplifier using two transistors, each of which is 1280 microns
wide.
CONCLUSIONS
AlGaN / GaN HFETs are the device of choice for microwave power applications.
For these power applications, the improved thermal conductivity of electrically insulating
SiC substrates provides a huge advantage in performance, relative to HFETs on sapphire
substrates. At 10 GHz, hybrid amplifiers are the cost-effective architecture, with the
large area passive matching components being fabricated on inexpensive substrates.
ACKNOWLEDGEMENTS
We gratefully acknowledge BMDO (Dr. Kepi Wu) for supporting this work.
REFERENCES
1- S.T. Allen, W.L. Pribble, R.A. Sadler, T.S. Alcorn, Z. Ring and J.W. Palmour, Recent
Progress in SiC Microwave MESFETs, Spring MRS Conference, Talk Y1.2, San
Francisco, 1999.
2- E. T. Yu, X. Z. Dang, P. M. Asbeck, S. S. Lau and G. J. Sullivan,
"Spontaneous and piezoelectric polarization effects in III-V nitride heterostructures", to
appear in J. Vac. Sei. and Tech., June/July 1999.
479
Recessed gate GaN MESFETs fabricated by the photoelectrochemical
etching process
Won Sang Lee,* Yoon Ho Choi,* Ki Woong Chung,*
Moo Whan Shin,** and Dong Chan Moon***
* Device & Materials Lab, LG Corporate Institute of Technology, 16 Woomyeon-Dong,
Seocho-gu, Seoul, Korea 137-724
**Department of Inorganic Materials Eng., Myongji University,
38-2 Nam-Dong, Yongin-Si, Kyunggi-Do, Korea
Department of Electronic Materials Eng., Kwang Woon University,
447-1 Wolgye-Dong, Seoul, Korea
Abstract
A new photo-electrochemical etching method was developed and used to fabricate
GaN MESFETs. The etching process uses photoresist for masking illumination and the
etchant is KOH based. The etching rate with 1.0 mol% of KOH for n-GaN is as high
as 1600 A/min under the Hg illumination of 35 mW/cm . The MESFET saturates at
VDS = 4 V and pinches off at VGS = -3 V. The maximum drain current of the device
is 230 mA/mm at 300 K and the value is remained almost same for 500 K operation.
The characteristic frequencies, fT and fmax, are 6.35 GHz and 10.25 GHz, respectively.
Insensitivity of the device performance to temperature was attributed to the
defect-related high activation energy of dopants for ionization and band-bending at the
subgrain boundaries in GaN thin films.
1. INTRODUCTION
Wide bandgap semiconductors based upon the IE-Nitride system, which are mainly
developed for optical application, have many properties which are ideal for electronic
devices for high temperature, high frequency, high power, and radiation hard application.
A variety of high frequency electronic devices can be fabricated from GaN-based
semiconductors; these devices are predicted to offer superior DC and RF performance
compared to more conventional Si and GaAs devices from the view point of high
power. Theoretical calculations for GaN MESFET predict output power density near 5
W/mm, power-added efficiency higher than 50 %, and linear power gain about 20 dB
for an optimized device structure [1]. In addition, GaN can allow for the AlGaN/GaN
heterostrüctures which was demonstrated to produce two-dimensional electron gas (2
, DEG) and thus makes possible several novel devices that can operate at frequencies
beyond the capability of SiC devices. Also, the 2 DEG permits low resistances and
low noise performance not possible with SiC [2].
Despite the excellent electronic properties of GaN, the fabrication of GaN MESFETs
with novel designs have been hindered by the chemical inertness of this material. Thus,
the most successful etching process so far has been dry etching method including
reactive ion etching (RIE) [3],
electron cyclotron resonance (ECR) RIE [4], and
inductively coupled plasma (1CP) RIE [5]. However, these techniques are known to
result in ion-induced damage on the etch surface which is highly undesirable for the
high frequency and high power operation of devices. Wet chemical etching is a
desirable substitute for dry etching method by providing low damage on the surface of
the active region. There has been a report on the photoelectrochemical wet etching
process for GaN films using KOH solution under the illumination of Hg arc lamp [6].
However, the experiment was performed using Ti metal mask which is not normally
used in conventional wet etching process and is not suitable for the fabrication of
microelectronic devices. In this paper, we report on the photoelectrochemical etching
481
Mat. Res. Soc. Symp. Proc. Vol. 572 ®1999 Materials Research Society
process for GaN thin films using photoresist mask. For the first time, the etching
method is demonstrated to be well applicable to the fabrication of recessed gate GaN
MESFET. The IDS, gm, and Vm of the device at different operating temperatures and fT
and fmax characteristics are discussed as well.
2.EXPERIMENTS
There are many ways of etching the wide bandgap semiconductors to make
transistors. The dry processing was used mainly because no convenientchemical solution
for good ething characteristics is available. Once we can provide the chemical solution,
the photoelectrochemical etching process is preferred, in particular, by the need for a
recessed gate GaN MESFET. The etching mask in this experiments is a i-line
photoresist (THMR-iP1800, TOK), eliminating the hard baking process which degrades
the adhesion of photoresist to the surface and hence leads to easy lift-off process. The
sample for this experiment consists of 1.3 ßm thick undoped (3~4 x 10 /cm) GaN
buffer layer, 500 A thick Si doped (2xl017/cm) n-GaN active layer and 300 A thick
Si doped (2xl018/cm3) n+-GaN cap layer grown on c-plane sapphire substrate by
metal-organic chemical vapor deposition (MOCVD) method. GaN epi layer quality was
measured by PL, XRD, and Hall measurement of 300 K. Dry etching for device
isolation, i.e. mesa etching, used an ECR dry etcher. Etching rate was varied with RF
bias, microwave power, and processing chamber pressure. Prior to the gate recess
etching process, the source and the drain ohmic contact with a contact resistance of
4 x 10 ,0 cm2 was obtained by the deposition of Ti/Al (=300 A/2000 A) bilayer
followed by a rapid thermal annealing (RTA) at 700 °C for 10 seconds. The recess
wet etching rate was examined as functions of the mole percentage of the KOH solution
and the etching time.
After an
extraction of
optimized conditions,
the
photoelectrochemical etching method was directly employed for the fabrication of a
+
recessed gate GaN MESFET. An n cap layer with a thickness of 300 A was etched
out for the recessed gate structure. The schottky contact was formed by a Pt/Au (=400
A/4000 A) alloy and it showed a good blocking capability. The gate length and the
gate width of the device were 0.7 ^m and 100 /im, respectively. The spacing between
the gate and source or drain was 5 ß m.
The morphology of the etched surface was
characterized by Scanning Electron Microscopy (SEM) and Atomic Force Microscopy
(AFM). The dc performance of the device was characterized at
different device
operating temp. From the S-parameter measurement in a frequency range of 1 — 18 GHz
with a microwave network analyzer, HP 8510C, we measured the fr and fmax3. RESULTS AND DISCUSSION
Fig. 1 shows the photoluminescence and X-ray diffraction of n-GaN epi layer at
room temperature. FWHM of band edge emission, (0002) 6 scan, (1012) $ scan are 43
meV and 7.2 arcmin and 22.5 arcmin, respectively.
Carrier concentration and mobility has changed by Si mole fraction. When the Si
mole fraction varied range of 0.5 ~ 7 nmol/min, carrier concentration, mobility are
increased of 2xl017/cm3~5.6xl018/cm3 and decreased of 430 cm/V.sec~135 cm/V.sec,
respectively. Optimum condition of GaN dry etching is Ar:5, BC13:4, Cl2:3, Pressure:3,
RF:450 V, Mw:300 W, He:3 and its etching rate has 950 A/min.
2
o
Ohmic contact metal used Ti/Al system, contact resistance was 5x10" ßcm at 700 C
10 sec RTP annealing. Fig. 2 shows the photoelectrochemical etching depth as functions
of etching time and the concentration of KOH in the etchant solution. The etching rate
is linearly increased with etching time. The etching rate is shown to increase as the
KOH concentration is increased. The etching rates for the 0.5 mol % and 1.0 mol % of
KOH in the solution are about 460 A/min and 1600 A/min, respectively. Note that the
etching rate observed in this experiments is comparable with the etching rate achieved in
482
a typical RIE method. The illumination intensity of Hg arc lamp during the etching
process was 35 mW/cm2. Fig. 3 shows the morphology of the etched surface displaying
a well defined etched edge. The sample in this figure was etched (DC bias 2.5 V, UV
Power of 35 mW) to a depth of about 2000 A. This figure demonstrates that the
photoresist can be used as a suitable pattern mask in this etching process.
300K
D
£
Band edge emission i
^\J
-^
3
<
c
aj
'
"
1
300
c
200
c
100
O
|
\
3
a_
400
*
34 .B
Yellow emission
300K
.
^
<"«> « oa p«»*.
A\
J v
/
tr* n te&S*^
39.1
^"^*^^o
39.4
3S.7
36.0
4> (degree)
36.3
30
(O02) Br.00 paak
1.8x10*
1.0x10'
:
a
c
I:
S.OxlO"
£
0.0
1B.O
1.4 1.6 1.8 2.0 2.2 2.4 2.6 2.8 3.0 3.2 3.4 3.6 3.8
18.4
16.8
17.2
17.6
16.0
6 (degree)
Energy (eV)
Fig. 1 Photoluminescence and Xray diffraction of n-GaN epi layer at 300K
A KOH base 0.5 mol%
-460 A/min
KOH base 1.0 mol%
~ 1600 A/min
10
20
Etching time [min]
Fig. 2 Photoelectrochemical etching depth
483
Fig. 3 Etched surface by SEM
The surface roughness was examined by AFM and the result is shown in Fig. 4.
The rms roughness from the etched GaN surface is about 37 Ä. The rms roughness for
the pre-etched GaN surface is measured to be 3.2 A. The surface morphology has been
changed by the etching, but the value of etching-induced roughness is low enough to
fabricate a MESFET structure with a gate length of 0.7 ^m and a channel thickness of
0.05 //m.
0.5 ^m
0.5 fin
b)
a)
Fig. 4 Surface roughness by AFM
a) as-grown b) etched surface
The optimized etching conditions were directly applied to the fabrication of recessed
gate GaN MESFET and Fig. 5
shows the room temperature and high
temperature current-voltage characteristics of the device. The gate voltage step is - 1 V
and the saturation of the source-drain current occurs at Vds = 4 V and the pinch-off
voltage at Vgs = - 3 V. The maximum drain-source current of the device operating at
300 K is about 240 mA/mm and no significant change is observed at 500 K. The
484
current level is somewhat lower than the value expected from the device design,
reasons for which will be discussed later.
250
-Vg = 0V
200
£
£
150
-Vg = -1V
<
-Vg = -2V
50
-
-Vg = -3V
15
10
20
vDSM
a)
E
E
E
300
2S0
260240
220
200
180
160
14012010080
60
40
20
0
VDS = 8V,Vos = 0V
Temperature [°C]
b)
Fig. 5 Current-voltage characteristics of GaN MESFET
a) Ids vs. Vds with Vgs (Vgs varies from 0 V to -3 V by -1 V from the top curve,
b) Drain current change to the operating temperature
(Vds = 8 V, Vgs = 0 V, temperature from room temperature to 200 °C)
485
The measured RF performance is in Fig. 6. The cut-off frequency and maximum
oscillation frequency are 6.25 GHz and 10.25 GHz at Vgs = 0 V, Vds = 8 V,
respectively.
5.o !
^J-w
^ v.
>*•
>
r1
\^ 5a
Mmax I
^
H21
*I {
°\
5 .0
1 .D
30. 0
Frequency
GHz
Fig. 6 fr and fmax characteristics of GaN MESFET
Theoretically, the current is supposed to decrease with the operating temperatures due
to the reduction of the electron velocity and mobility to the temperatures. One
explanation for no change in the drain current level is a high activation energy of
dopants for ionization. It has been reported that the activation energy of dopants in
thin films with two dimensional defects is known to be higher than that without the
defects [6]. When the activation is high and the dopants are in a "freeze-out region"
[7], the number of the effective free carriers increases with temperature. Therefore, it is
possible that the insensitivity of the drain current with temperature be attributed to a
presumable freeze-out of dopants in the active region of device, otherwise, the drain
current should be decreased due to the lower mobility and electron velocity at high
temperatures. Another explanation for the insensitivity is trapping of carriers at the
subgrain boundaries. GaN thin films are known to possess high density of subgrain
boundaries that are responsible for the band-bending 8]. The driving force for electrons
to cross the transverse boundaries is larger when the device operates at higher
temperatures. Thus, the mobility of electrons at higher temperature could be higher than
that at lower temperature as is the observed abnormal behavior in the mobility-doping
relationship [9].
4. CONCLUSION
For the first time, the photoelectrochemical etching process using a photoresist mask was
employed for the fabrication of recessed gate GaN MESFETs. The maximum etching
rate with
1.0 mol % of KOH for n-GaN was 1600 Ä/min. The fabricated GaN
MESFET shows a current saturation with the maximum current of 240 mA/mm at room
temperature. Insensitivity of the change in the drain current to the temperature was
attributed to the defect-related high activation energy of dopants and band-bending. The
486
developed photoelectric wet etching process is likely to be used widely in the fabrication
of devices using GaN, the chemically inert semiconductor.
REFERENCES
1. R. J. Trew and M. W. Shin, Int. J. High Speed Electronics and Systems,6, 211
(1995).
2. M. S. Shur, A. Khan, B. Gelmont, R. J. Trew, and M. W. Shin, Ins. Phys.
Conf. Series No. 141: Chapter 4, 419 (1994).
3. M. E. Lin, Z. F. Fan, Z. Ma, L. H. Allen, and H. Morcoc, Appl. Phys. Lett. 64,
887 (1994).
4. R. J. Shul, G. B. Aclellan, S. A. Casalnuovo, D. J. Reiger, S. J. Pearton, C.
Constantine, C. Barrat, R. F. Karlicek. C. Tran, and M. Schurman, Appl. Phys.
Lett., 69, 1119 (1996).
5. J. R. Milehan, S. J. Pearton, C. R. Abernathy, J. D. Mackenzie, R. J. Shul, and
S. P. Kilcoyene, J. Vac. Sei. Technol., A.14, 836(1996).
6. M. W. Shin, R. J. Trew, and G. L. Bilbro, IEEE Electron Device Lett., 15, 292
(1994).
7. M. W. Shin, R. J. Trew, G. L. Bilbro, D. L. Dreifus, and A. J. Tessmer, J. Mater.
Sei., 6, 111 (1995).
8. F. A. Fonce, MRS Bulletin, 22, 51 (1997).
9. Y. Park, B. Kim, I. Kim, and E. Oh, Proc. '98 Korea-Japan Joint Workshop on
SWSODM, 54 (1998).
487
CURRENT-VOLTAGE CHARACTERISTICS OF
UNGATED AlGaN/GaN HETEROSTRUCTURES
J.D. Albrechtt, P.P. Rudenf, S.C. Binari*, K. Ikossi-Anastasiou*,
M.G. Ancona*, R.L. Henry*, D.D. Koleske*, and A.E. Wickenden*
'ECE Department, University of Minnesota, Minneapolis, MN 55455
^Electronics Science and Technology Div., Naval Research Laboratory, Washington, DC 20375
ABSTRACT
Results of a systematic study of the current vs. voltage characteristics of ungated AlGaN/
GaN heterostructures grown on sapphire substrates are presented. It is experimentally observed
that the saturation current nearly doubles as the source-to-drain channel lengths decrease from
11.8 to 1.7|im. The average electric field at which current saturation occurs is 10 to 30kV/cm, i.e.
much less than the electron velocity saturation field. The experimental data is interpreted in the
framework of a new model that takes into account the non-uniformity of the electron density in
the channel, electron velocity saturation, and thermal effects. The temperature dependent electron
transport characteristics of the model are based on Monte Carlo simulations of electron transport
in GaN. It is shown that appreciable contact resistance, which leads to partial channel depletion
near the source, and significant self-heating of the devices under high drain-to-source bias are the
main reasons for the observed current saturation. The effective ambient temperature in the channel of the devices is calculated from a two-dimensional thermal model of heat dissipation through
the sapphire substrate. Equilibrium channel carrier concentrations and low-field mobilities are
determined from Hall effect data. The ungated structures are demonstrated to provide much useful materials and process characterization data for the development of AlGaN/GaN heterostructure field effect transistors.
INTRODUCTION
AlGaN/GaN heterostructures have excellent potential for the realization of high-power,
high-frequency heterostructure field effect transistors (HFETs) [1,2,3]. The GaN channel
material is characterized by favorable electron transport parameters, such as high mobility and
high peak velocity. In addition, polarization charges at the hetero-interface can neutralize a
large electron charge density, thus allowing for high channel carrier concentrations. However,
most epitaxial Ill-nitride structures are currently grown on sapphire substrates. Sapphire has a
relatively low thermal conductivity and, consequently, the performance of AlGaN/GaN HFETs
that are heatsunk through the sapphire substrate has been shown to be limited by deleterious
thermal effects at high voltages and currents [4]. Furthermore, the fabrication technology for
these devices is still relatively immature. Hence, contact resistances are often large and can
seriously impair device performance [4].
489
Mat. Res. Soc. Symp. Proc. Vol. 572
e
1999 Materials Research Society
Ungated HFET structures are of interest as simple characterization structures used in,
e.g., transmission line method (TLM) determinations of the equilibrium sheet resistivity and
contact resistance, and as saturated resistors in some circuit application. Here, we present a
systematic theoretical and experimental examination of ungated AlGaN/GaN heterostructures
grown on sapphire substrates.
DEVICE STRUCTURE
The Ill-nitrides used in this study were
grown by MOCVD on sapphire substrates.
The material structure consisted of an A1N es
nucleation layer, followed by 3|^m of
undoped GaN and 300Ä of Si doped
Al03Gao7N. Ohmic contacts were formed
using alloyed Ti/Al/Ni/Au. Device isolation ei
was accomplished by implantation with
nitrogen. The contact resistances varied
2
4
6
8
10
o
Spacing (um)
considerably with annealing conditions. The
impact of the different contact resistance Figure 1: Measured low-field resistances vs.
values on the current vs. voltage device length for ungated AlGaN/GaN heterostruccharacteristics of the ungated structures will tures.
be examined in more detail below. We will first focus on a set of devices, all from the same
wafer, with relatively low contact resistance. The devices tested consist of 75nm wide ungated
channels with 1.7, 2.0, 3.6, 4.7, 6.8, and 11.8nm source and drain contact spacings. The
measured low-voltage resistances are plotted as a function of the contact spacing in Figure 1.
Simple TLM analysis indicates that the sheet resistivity for this heterostructure is 610A/Q and
the specific contact resistance is 2.0£imm.
DEVICE MODEL AND KEY PARAMETERS
The device model developed for this investigation is a charge-control/gradual-channel
approximation model that incorporates salient results of previous Monte Carlo electron transport
simulations for GaN as a function of the ambient temperature [5]. The measured low-field
TLM data shown in Figure 1, together with independent Hall measurements of the electron
mobility, provide information about the equilibrium channel electron concentration, ns0=l/
qjxLFpQ, where |XLF is the low-field electron mobility and q is the electron charge. Under
applied drain-to-source bias, VDS, the local electron concentration in the channel can be related
to the difference between the surface potential, <|>s(x), and the channel potential, Vch(x). This
relationship can be expressed as,
(1)
qns(x) = /-[<l)s(x)-Vch(x)] + qns
u
eff
where e is the material permittivity. The surface-to-channel distance, deff, is the thickness of the
AlGaN layer plus the effective thickness of the two-dimensional electron gas (-20Ä), which ts
490
determined from a self-consistent calculation of the quasi-two-dimensional subband structure as
discussed in ref. [4] and references therein. The channel coordinate, x, ranges from zero at the
source-end to LDs at the drain-end. The surface potential, referenced to the source potential, can
be assumed to vary linearly between the source and drain contacts: <J)s(x) = (VDS/LDS)'XThe low-field mobility, the saturation (peak) velocity, and the critical (peak) field for
velocity saturation, results of Monte Carlo simulations for electron transport in GaN have been
parameterized as functions of temperature, doping concentration, and compensation [5]. The
velocity vs. electric field curve for field strengths below the critical field, Fc, is well
approximated with the simple analytic expression used in ref. [4]. Expressing this velocity vs.
field relationship in terms of a field dependent mobility, we obtain,
T
^LF( )
H(T)
1+
ch
I vc(T)
for
Fc(T)Jrfx
for
ch
dx
<F„
(2)
dV_ch
dx >Fo
where vc is the electron drift velocity at Fc. The drain-to-source drift current can now be
calculated from,
d\ch
I = qns(x)ndx
with the boundary conditions for the channel potential given by
(3)
Vch(0) = IRC
(4)
andV"ch^DS= VVDS"
ch(LDS)1 DS-IRC;
(5)
where Rc is the contact resistance. Integration of (3) over the length of the channel yields after
some manipulation the following implicit relationship between I and VDs:
L
-VDS = q[ns(LDS)-ns(0)] +
u
eff
IL DS
^LFV;DS
In
qns(LDs)
(0)+
WFC
^ C-
DS
H>LFVDS
L
DS
V
MLF DS
-T)
(6)
rj
Because of the voltage drops across the source and drain contacts under bias, the electron
concentration is reduced below the equilibrium value at the source-end of the channel and
enhanced above the equilibrium value at the drain-end. The non-uniform electron concentration
implies a non-uniform longitudinal electric field. This model is similar in character to a model
developed in refs. [6 and 7] for GaAs FETs.
Lastly, as indicated in equation (2) above, the transport parameters depend on the lattice
temperature. The power dissipated in the channel raises the ambient temperature due to the nonzero thermal impedance of the structure. The devices are heatsunk through the sapphire
substrate. The thermal impedance characterizing the heat transfer from the channel to the
491
heatsink was determined by executing a separate, two-dimensional heat-flow model.
Obviously, a two-dimensional model of the thermal conductance is only a rough approximation
to the real, three-dimensional problem. In addition, it was found that the thermal impedance
depends on the layout of the contact pattern and on the effectiveness with which these contacts
are heatsunk by the probes. However, for the parameter range of interest, a satisfactory
approximation to the effective thermal impedance obtained from these calculations (taking into
account the temperature dependences of the thermal conductivities of sapphire and of the IIInitrides) is given by R^ = (23.2 - 0.26-(LDS - 2))-(l + (T - Theatsink)/T0) K-mm/W, with T0 = 725
- 5.8(LDS - 2). Here LDS is given in urn. The channel temperature, T, is related to the power
dissipated:
T
heatsink
+ R
(7)
thIVDS
withTheatsink = 300K.
The thermal effects in the current calculation as a function of applied voltage are
incorporated by solving the system of coupled equations given by (6) and (7) to obtain a selfconsistent solution. The contact resistances are treated as temperature independent in all of the
calculations presented.
RESULTS AND DISCUSSION
Figure 2 shows the measured and calculated current vs. voltage characteristics of the ungated AlGaN/GaN heterostructures for various channel lengths. The curves are shown as broken
lines where the calculated channel temperature exceeds 650K. The agreement between the measured and calculated currents is very good, in particular for the longer devices. To explore the or1200
1000
800
3
I
u
600
400
200
VDS(V)
Figure 2: Measured (points) and calculated (solid
lines) current-voltage characteristics. The channel
lengths are LDS = 1.7 (top curve), 2.0, 3.6,4.7,6.8,
and 11.8^m (lowest curve).
Figure 3: Comparison of the calculated currents
for the same devices as in Fig. 2 with (solid lines)
and without (broken lines) thermal effects taken
into account.
492
800
HJ
n 35
^^^^
^v.
>
Ä
^^»^^^-.
<o
3 25
^^^_^^i
InC3
s2
<u
15
"
'
.
'
___
£
s
0
.
"
1/2
Channel Position (x/LDS)
1
1
Figure 4: Local longitudinal electric field corresponding to the current-voltage characteristics
shown in Fig. 2 for VDs = 10.0 V. The top curve
corresponds to the shortest device, the bottom
curve to the longest.
10
Rc (ß-mm)
Figure 5: Calculated saturation currents as a
function of the contact resistance for a series of
ungated AIGaN/GaN heterostructures with LDS =
5um. Also shown are experimental results from
devices with different contact resistances due to
different annealing conditions.
igin of the observed current saturation, Figure 3 displays the currents calculated assuming perfect
heatsinking (R^ = 0), together with the calculated results of Figure 2. Clearly, the thermal effects
are critical in leading to the measured saturation currents that increase with decreasing channel
length. The electron velocity depends most strongly on the temperature at low fields. Electron velocity saturation alone leads to saturation currents that are independent of the channel length (approximately equal to 2300mA/mm in our model), although the voltage at which saturation is
reached increases with channel length. The conclusion that the carrier velocity is actually relatively low in the devices studied here is supported by a calculation of the longitudinal electric field in
the channel. This field can be obtained by integrating the current equation (3) numerically. Figure
4 displays the longitudinal electric fields for the whole set of devices versus the normalized channel
coordinate, x/Lps» for an applied bias of 10V. It is apparent that even in the short devices the electric field at the source-end barely reaches 45kV/cm, well below the field where the velocity peaks
[5]. Evident from Figure 4 is the partial carrier depletion near the source-end of the channel. This
effect is larger in devices with short channels because the contact resistances play a greater role.
In a separate calculation, the effect of diffusion implied by the non-uniform carrier density was examined and it was found to be very small for these ungated devices [8].
From the results obtained, it is to be expected that the saturation current will depend on the
value of the contact resistances. A larger source contact resistance implies a stronger tendency for
the channel to be pinched off near the source-end and, consequently, a larger longitudinal electric
field and carrier velocity in that region when current saturation is reached. To examine the effect
of the contact resistance, the saturation currents were calculated as a function of the specific contact
resistance for devices with LDs = 5|xm. The result is shown in Figure 5 Also displayed are experimental data obtained from devices on a single wafer (although a different wafer from the one con-
493
sidered above) that underwent different annealing procedures. The contact resistances varied over
a relatively wide range when annealing conditions were changed from 830C for 2sec (highest contact resistance) to 860C for lOsec (lowest contact resistance) [4]. Good qualitative agreement between the calculation and the experimental data is observed. However, for the case of high contact
resistance there are quantitative discrepancies which may be attributable to variations in the equilibrium channel carrier concentrations with process conditions. The calculations assumed no variation in carrier concentration, transport parameters, or thermal impedance. A field analysis similar
to the one shown in Figure 4 demonstrates that electron velocity saturation ensures current saturation in devices with very high contact resistance and that thermal effects are relatively less important.
In summary, calculated current-voltage characteristics of ungated AlGaN/GaN
heterostructures have been presented and found to be in good agreement with experimental
data. The calculations reveal that the electric field and the carrier concentration remain fairly
uniform and that thermal degradation of the electron velocity is the main cause of the observed
current saturation in devices with relatively low contact resistance. Devices with high contact
resistance on the other hand show more pronounced field and carrier density non-uniformity.
This work also indicates that further study of the gate-to-source and gate-to-drain regions in
AlGaN/GaN HFETs is needed to fully understand the observed device performance. A
numerical model similar to the one presented here, which includes diffusion, will be the subject
of future work.
ACKNOWLEDGMENT
The work at the University of Minnesota was supported in part by Hughes Research
Laboratories and by the National Science Foundation. The work at NRL was partially supported
by the Office of Naval Research.
REFERENCES
1. N.X. Nguyen, C. Nguyen, and D.E. Grider, Electronics Lett. 34, 811 (1998).
2. G.J. Sullivan, J.A. Higgins, M.Y. Chen, J.W. Yang, Q. Chen, R.L. Pierson, and B.T. McDer
mott, Electrics Lett. 34, 922 (1998).
3. Y.-F. Wu, B. P. Keller, P. Fini, S. Keller, T. J. Jenkins, L. T. Kehias, S. P. Denbaars, and
U. K. Mishra, IEEE Electron Device Lett. 19, no.2, 50, (1998).
4. P.P. Ruden, J.D. Albrecht, A. Sutandi, S.C. Binari, K. Dcossi-Anastasiou, M.G. Ancona, R.L.
Henry, D.D. Koleske, and A.E. Wickenden, MRS Internet J. Nitride Semicond. Res. 4SI,
G6.35 (1999).
5. J.D. Albrecht, R.P. Wang, P.P. Ruden, M. Farahmand, and K.F. Brennan, J. Appl. Phys., 83,
4777 (1998).
6. T. Harm, K. Takahashi, and Y. Shibata, IEEE Trans. Electron Devices ED-30, 1743 (1983).
7. J. Baek. M.S. Shur, K.W. Lee, and T. Vu, IEEE Trans. Electron Devices ED-32, 2426
(1985).
8. J.D. Albrecht, P.P. Ruden, to be published.
494
Hydrostatic and uniaxial stress dependence and photo induced effects on the channel
conductance of n-AlGaN/GaN heterostructures grown on sapphire substrates
A. K. Funga, C. Caia, P. P. Ruden", M. I. Nathan", M. Y. Chenb, B. T. McDermottb,
G. J. Sullivan", J. M. Van Hove0, K. Boutrosd, J. Redwingd, J. W. Yange, Q. Chene,
M. A. Khane, W. Schafff, M. Murphyf
"Department of Electrical and Computer Engineering, University of Minnesota, Minneapolis,
MN 55455, afung@ece.umn.edu, nathan@ece.umn.edu
"Rockwell International Science Center, Thousand Oaks, CA 91358
C
SVT Associates, Eden Prairie, MN 55344
d
Epitronics/ATMI, Phoenix, Arizona 85027
e
APA Optics, Blaine, Minnesota 55449
f
School of Electrical Engineering, Cornell University, Ithaca, NY 14853
Abstract
We measure the hydrostatic stress, uniaxial stress, and photo induced dependence of the
channel conductance of two-dimensional electron gas AlGaN/GaN heterostructures grown on
c-axis sapphire. The structures examined are grown by nitrogen-plasma molecular beam
epitaxy and metal organic chemical vapor deposition. Electrical conductance measurements
are made with four point probes on Hall bar samples. Both, hydrostatic stress and uniaxial
stress result in changes in the conductance. Moreover, these changes in conductance have long
settling times after the stress is applied and may be due to deep level defects, the energy levels
of which change with stress. Stress coefficients extracted from the samples are partially
attributed to deep level defects and to the piezoelectric effect resulting from different
piezoelectric coefficients of GaN and A1N. Photo induced changes of the two-dimensional
electron gas are also observed. We find that pulsed illumination produces long transient times
in the conductance. These transients are reduced by thermal heating in some samples.
However, they can still be present at 153°C.
Introduction
Stress experiments are useful in the characterization of semiconductors. They can be
used to determine piezoelectric constants1 and they can reveal the presence of deep defects
having energy levels that change with pressure.2 Recent efforts have utilized applied static
stress/strain to piezoelectrically modify the two-dimensional (2D) electron concentrations at
AlGaAs/GaAs and AlGaN/GaN 4 heterostructure interfaces. The changes in channel
conductance can be related to differences in the piezoelectric constants and to differences in the
stress/strain profiles of the materials that constitute the heterostructures. Pressure has also been
used to quantify the amount of energy shift of deep levels through the changes in conductance
and photoluminesce.5 By the application of piezoelectric effects, and through the
understanding of the behavior of defects on the properties of semiconductors, novel and
improved device structures can be implemented.6,7
In this study we report on hydrostatic and uniaxial stress effects in the 2D electron gas
channel conductance of AlGaN/GaN heterostructures. We find that the channel conductance
exhibits a slow (on the order of 102 to 103 seconds) change with the application of pressure.
This slow transient follows a very fast change that occurs on the time scale of the application or
removal of pressure (a few seconds). We also examine the photo response of the samples in
attempts to gain more information about the cause of these long pressure induced transients.
495
Mat. Res. Soc. Symp. Proc. Vol. 572 e 1999 Materials Research Society
We find that the photo responses also have long transients and that increasing the temperature
of the sample can reduce them. However, the long transients can still persist even at 153°C.
Experiment
The four different epitaxial layers used in this study are grown by four different labs
with either nitrogen-plasma source molecular beam epitaxy (MBE) or metal organic chemical
vapor deposition (MOCVD). Here we will designate the MBE wafers as 1(a) and 1(b) and the
MOCVD wafers as 2(a) and 2(b). Schematic layer structures of the different wafers are shown
in Fig 1 Nominal non-illuminated room temperature Hall mobility (cm2/Vs)/electron sheet
carrier concentration (cm"2) are 580/5.40xl012, 723/1.38xl013, 933/3.50xl012 and 1520
/1.08xl013 for wafers 1(a), 1(b), 2(a) and 2(b), respectively. Standard microfabrication
techniques are used to produce Hall bars for our experiment.
Hydrostatic pressure up to 0.9GPa is applied to the samples in this study in a Unipress
cell through a 1:1 mixture of hexane:pentane. Temperature in the cell is monitored with a typeT copper-constantan thermocouple. Temperature changes in the cell follow a damped response
to the ambient lab temperature and can vary by as much as +0.5°C.
Compressive uniaxial pressure up to 0.6GPa is applied to a rectangularly prepared
sample by placing one end of the sample flush against a fixed aluminum platform and by
applying pressure to the other end of the sample with a tungsten anvil. The rectangular sample
dimensions are typically 0.85mm x 0.43mm x 2.5mm with the long edge parallel to the ^ g
direction of applied stress. Details of the apparatus and method can be found elsewhere. '
Photoconductivity measurements are performed with various excitation sources, light
emitting diodes with wavelengths of 470 and 928nm, a halogen lamp, and a mercury-gasfluorescent white lamp. Electrical measurements are performed with a HP4156A dc
semiconductor analyzer. Conductance measurements are obtained using four point contacts on
Hall bars.
5nm
5nm
Alo.15Gao.85N
18
25nm Alo.15Gao.85N n = 2xl0 cm"
3
GaN
20nm Al0.2Ga0.sN
5nm
Alo.15Gao.85N
lum
GaN
lum
GaN
5nm
A1N
Sapphire Substrate
Sapphire Substrate
Wafer 1(a)
18
3
30nm Alo.15Gao.85N n = 2xl0 cm"
30nm Alo.15Gao.85N
lOnm Alo.15Gao.85N
lOOnmGaN
5\im
n = 2xl017 cm"3
1 urn GaN
GaN
Sapphire Substrate
40nm A1N
Sapphire Substrate
Wafer 1(b)
Wafer 2(b)
Wafer 2(a)
Figure 1: Schematic epitaxial layer structures of wafers used in this study. Wafer 1(a) and
1(b) are grown by MBE, wafer 2(a) and 2(b) are grown by MOCVD.
496
Results
The application of hydrostatic stress result in changes in conductance of 1.0%/GPa and
3.5%/GPa for wafers 1(a) and 2(a), respectively. Compressive uniaxial stress produce changes
in the conductance on wafer 1(a) of -0.25%/GPa and -0.63%/GPa for stress in the [1120] and
[1010] GaN directions, respectively. These values are determined from the steady state values
after the pressure induced transient on the conductance and are the result of at least two effects:
the piezoelectric effect and a change in the ionization of deep level defects. For the
piezoelectric effect, the presence of inversion domains9 can modify the measured change in
conductance since the sign of the piezoelectric polarization depends on the orientation of the
(hexagonal) growth axis. Figure 2 shows time dependence of the conductance of wafer 1(a) for
an applied compressive uniaxial stress. The time domain behavior of the conductance shows
the presence of at least two processes, which result in a fast initial change in conductance and a
long decay in conductance with applied stress. The piezoelectric polarization at each interface
of dissimilar materials and their neutralization by available free carriers from the contacts or the
semiconductor bulk occur essentially instantaneously as do stress induced energy level shifts of
deep level defects. Both processes occur in a much shorter time than the time step of each data
point and will follow the applied pressure without time delay. In Fig. 2, at the leading edge
where stress is applied, the loading process is done over a period of a few seconds so as not to
overload the sample. After the stress is applied it can be seen that there is a slow process that
affects the conductance of the sample. The measured conductance is determined by the number
of carriers and their mobility. Although the mobility of the electrons can change with the
change in electron density due to screening and other effects, slow trapping and detrapping of
carriers with pressure most likely cause the long time constants. These slow trapping times
may be due to energy barriers associated with certain types of defects. Also, band
discontinuities of the different materials in the heterostructure and band bending from space
charge may additionally impede and slow movement of free carriers in response to applied
0.05
^"-0.05°
—
i
—4
HI
Uniaxial Stress [1 0101
AVIOOO
2000
3000
4C
o -0.1
tS -0-15 " Applying Stress
3
T3
C
-0.2
o 0.2 Removing Stress
o
c 0.1 a>
at
c
re
sz 0.1 J»
O
Uniaxial Stress [1010]
1000
2000
3000
4C30
0.2
Time [s]
Figure 2: Time dependence of conductance for compressive uniaxial stress on wafer 1(a)
along the [1010] GaN direction. Application of stress goes from 0.38GPa to 0.64GPa. The
removal of stress goes from 0.64GPa to 0.26 GPa.
497
pressure. In an attempt to further understand the slow transient behavior of the conductance
with pressure we also examine the photoconductive response of the samples under various
illumination conditions.
Figure 3 shows the pulsed illumination response for all four wafers examined in this
study. All the samples show slow conductance time transients with illumination. This is not
surprising; persistent photoconductivity (PPC) has been widely reported for bulk GaN10,11 and
AlGaN/GaN heterostructures.12'13 In our samples, we find that the conductance also changes
with infrared illumination (928nm). Wafers 1(a), 1(b) and 2(b) show an increase in
conductance with illumination whereas wafer 2(a) shows a slight decrease in conductance. The
increase in conductance may be caused by photo ionization of donor atoms or deep-donor-like
defect levels. The decrease in conductance can be due to the same processes that have been
reported to cause optical quenching.14,15 For wafer 2(a) prior to illumination, the sample may
be in a state such that thermally or optically generated electrons from the valence band are
present in the conduction band with their corresponding valence band holes at least partly
trapped in gap states that have long lifetimes." Then, upon illumination, the trapped holes in
the gap states within the photon energy range of the infrared LED (1.34eV) are freed into the
valence band where they can recombine with conduction band electrons directly or via
recombination centers in the band gap, resulting in a decrease in conductance.1 ,15 To elucidate
the slow conductance transients resulting from illumination we examine the effect of increasing
the temperature on optical quenching and on the PPC of our samples.
Since heating reduces both optical quenching and PPC and thermal noise makes it
difficult to observe these effects otherwise. It is necessary to increase the difference between
the initial (without optical quenching) and final (with optical quenching) conductive states for
optical quenching and also for the initial (illuminated) and final (non-illuminated) conductive
states for PPC. To generate electron-hole pairs we use a mercury-vapor-fluorescent lamp for
all the samples examined. To optically quench (reduce) the conductance for wafer 2(a) while
the fluorescent lamp is on, we use a halogen lamp. With increasing temperatures we find that
for wafer 2(a) optical quenching is reduced for fixed photon fluxes. Furthermore, at increasing
temperatures the effect of the halogen illumination increases the conductance rather than
decreases it (Fig. 4(a)). In addition to reducing optical quenching, heating can decrease
the PPC transient times for wafer 2(a). The transient times at higher temperatures are shorter
12
10
a
u
c
«j
o
3
o
C
o
o
8 6 4 -
a
D)
c
.c«I
u
-2
Turn Blue LED On
Time [s]
Figure 3: Pulsed illumination response for all four wafers studied in
this experiment. Long conductance transients are observed.
Listing wafers from greatest to smallest change in conductance at
300 seconds are wafer 2(a), 2(b), 1(b) and 1(a).
498
than at 27°C, however they can still persist up to 153°C (see Fig. 4(b)). Heating also reduces
the amount of photo-induced changes in the conductance for wafer 1(a). The change in
conductance for this sample from the illuminated state to the non-illuminated state at different
temperatures and a fixed fluorescent illumination intensity are -0.9%, -0.35% and -0.25% for
27°C, 90°C and 153°C, respectively. This effect can be the result of thermal ionization of states
in the band gap, leaving gap states emptier, such that it becomes less likely for photons with
energy less than the band gap to generate free carriers. Transient times are not reported for
wafer 1(a) since they could not be accurately quantified due to thermal noise. We also note that
the thermal coefficients for the electrical conductance of wafers 1(a) and 2(a) are linearly
interpolated to be -0.24%/°C and -0.46%/°C, respectively, between 27°C and 153°C with
fluorescent illumination. This can be due to a decrease in mobility and possibly to pyroelectric
effects that produce fixed polarization charges at the 2D electron gas interface that repel
electrons from the conduction path.
Conclusion
We have measured the hydrostatic and uniaxial stress dependence and the photo
response of AlGaN/GaN heterostructures. Hydrostatic stress results in changes in conductance
of 1.0%/GPa and 3.5%/GPa for wafers grown by MBE and MOCVD, respectively. For the
MBE wafer, uniaxial stress coefficients for the change in conductance are -0.25%/GPa and
0)
u
c
200
(0
c
o
o
c
6( 0
o>
c
ra
£
Time
[s]
Figure 4: (a) Optical quenching as a funcion of temperature for wafer 2(a).
The sample is illuminated with a fluorescent lamp at all times. A halogen
lamp is used to quench the conductance when it is turned on between 50 and
100 seconds, (b) Persistent photoconductivity as a function of temperature
for wafer 2(a). The sample is illuminated with a fluorescent lamp prior to 0
seconds. The lamp is turned off at 100 seconds. PPC response is nearly
identical at 90C and 153C.
499
-0.63%/GPa for stress in the [1120] and [1010] GaN directions, respectively. These
hydrostatic and uniaxial stress coefficients are attributed to the piezoelectric effect and stress
induced changes in ionization of defect levels. We can not quantitatively separate these two
effects. The application of stress results in conductance transients on the order of 10 to 10
seconds. Because of this, traps in the band gap due to defects are suspected. Photoconductivity
measurements show that the conductance of the MBE sample increases with light having
energy less than the band gap and may be due to the photo ionization of donors or deep level
electron traps. For the MOCVD sample, photoconductivity measurements show optical
quenching. This may be the result of hole trap levels in the band gap. Both the MBE and
MOCVD samples show persistent photoconductivity with long transients. The transients are
reduced with an increase in temperature, however even at 153°C long transients can still be
observed.
Acknowledgments
We thank Brian Ishaug and Xiaobo Zhang for providing the optical spectra of the light
sources used in this experiment and John Albrecht for helpful discussions. This work was
supported in part by NSF-ECS9612539, NSF-DMR and ONR N/N00014-9525758.
References
1. G. Arlt and P. Quadflieg, Phys. Status Solidi 25, 323 (1968)
2. J. Mercy, C. Bousquet, J. Robert, A. Raymond, G. Gregoris, J. Beerens, J. Portal, and P.
Frijlink, Surf. Sei. 142, 298 (1984)
3. A. Fung, J. Albrecht, M. Nathan, and P. Ruden, J. Appl. Phys. 84, 3741 (1998)
4. R. Gaska, J. Yang, A. Bykhovski, M. Shur, V. Kaminski, and S. Soloviov, Appl. Phys. Lett.
72, 64 (1998)
5. S. Kim, I. Herman, J. Tuchman, K. Doverspike, L. Rowland, and D. Gaskill, Appl. Phys.
Lett. 67, 380 (1995)
6. O. Ambacher, J. Smart, J. Shealy, N. Weimann, K. Chu, M. Murphy, W. Schaff, L.
Eastman, R. Dimitrov, L. Wittmer, M. Stutzmann, W. Rieger, and J. Hilsenbeck, J. Appl.
Phys. 85, 3222 (1999)
7. W. Yang, T. Nohava, S. Krishnankutty, R. Torreano, S. McPherson, and H. Marsh, Appl.
Phys. Lett. 73, 978 (1998)
8. S. Lu, C. Meng, F. Wiiliamson, and M. Nathan, J. Appl. Phys. 69, 8241 (1991)
9. L. Romano, J. Northrup, M. O'Keefe, Appl. Phys. Lett. 69, 2394 (1996)
10. E. Munoz, E. Monroy, J. Garrido, I. Izpura, F. Sanchez, M. Sanchez-Garcia, B. Beaumount,
and P. Gibart, Appl. Phys. Lett. 71, 870 (1997)
11. C. Qiu, W. Melton, M. Leksono, J. Pankove, B. Keller, and S. DenBaars, Appl. Phys. Lett.
69,1282 (1996)
12. J. Li, J. Lin, H. Jiang, M. Khan, and Q. Chen, J. Vac. Sei. Tech. B 15, 1117 (1997)
13. X. Dang, C. Wang, E. Yu, K. Boutros, and J. Redwing, Appl. Phys. Lett. 72, 2745 (1998)
14. Z. Huang, D. Mott, P. Shu, R. Zhang, J. Chen, and D. Wickenden, J. Appl. Phys. 82, 2707
(1997)
15. M. Hirsch, O. Seifert, O. Kirfel, J. Parisi, J. Wölk, W. Walukiewicz, E. Haller, O.
Ambacher, M. Stutzmann, Mat. Res. Soc. Symp. Pro. 482, 531 (1998)
500
THE INFLUENCE OF SPONTANEOUS AND
PIEZOELECTRIC POLARIZATION ON NOVEL
ALGAN/GAN/INGAN DEVICE STRUCTURES
B. E. FOUTZ, M. J. MURPHY, 0. AMBACHER, V. TILAK, J. A. SMART, J. R. SHEALY,
W. J. SCHAFF, and L. F. EASTMAN
School of Electrical Engineering, Cornell University, Ithaca, New York 14853
ABSTRACT
The strong spontaneous polarization and piezoelectric effects in the wurtzite IIInitride semiconductors lead to new possibilities for device design. In typical heterojunction
field effect transistors these effects are used to create large electron concentrations at the
AlGaN/GaN interface. However, we examine several other possible device structures which
include heterojunctions of AlGaN, GaN, and InGaN. For example, we find the strong electric fields present in these structures allow us to create quantum wells greater than 1 eV
deep. Both Ga-faced and N-faced materials are explored. The two-dimensional electron gas
concentrations in these structures are found using a self-consistent 1-D Schrödinger-Poisson
solver modified to incorporate the effects of spontaneous and piezoelectric polarization. The
boundary conditions at the heterojunction interfaces and at the surface and substrate are discussed in detail. Electron concentrations are compared with those obtained experimentally
through capacitance-voltage and Hall effect measurements.
INTRODUCTION
The group Ill-nitrides possess a large spontaneous and piezoelectric polarization. The
presence of this strong polarization is supported by both theoretical calculations of its existence [1] and the large electron concentrations which result at the AlGaN/GaN heterojunctions in transistor structures. Simple models [2, 3, 4] have been used to calculate the
electron concentration at a single heterointerface and support the hypothesis that the 2D
electron gas found at the interface is induced by polarization effects. However, to determine
the conduction band profile and electron concentrations in more complex device structures
more sophisticated and generalized methods are needed. One such method is the use of a
self-consistent one-dimensional Schrödinger-Poisson solver. The solver used in this study,
CBAND (for conduction band solver), has been modified to incorporate the effects of spontaneous and piezoelectric polarization. Specifically, the polarization is included in the solver
through the use of thin layers of space charge at each heterointerface of the structure. In this
paper we describe the theory used to calculate the conduction band profile and electron sheet
charge in nitride based heterostructures and then compare these results with experimental
data. Finally, we examine CBAND results for two novel device structures.
THEORY
The polarization present in the group Ill-nitrides is due to the lack of inversion symmetry
along the c-axis of the wurtzite crystal structure. In relaxed material there exists a builtin or spontaneous polarization [1]. This polarization points toward the substrate for Gaface material and points toward the surface in N-face material. (For Ga-face material the
501
Mat. Res. Soc. Symp. Proc. Vol. 572 e 1999 Materials Research Society
PSP
(C/m*)
633 (C/m2)
C3i (C/m2)
Ci3 (GPa)
C33 (GPa)
a0 (A)
GaN
-0.029
0.73
-0.49
103
405
3.189
A1N
-0.081
1.46
-0.60
108
373
3.112
InN
-0.032
0.97
-0.57
92
224
3.54
Table I: The constants used to calculate the polarization in Ill-nitride layers. Psp is the spontaneous
polarization. 633 and e3i are piezoelectric constants. C13 and C33 are elastic deformation constants
and ao is the lattice constant.
positive direction is toward the surface by convention.) The polarization in the material can
be changed by placing it under strain. This change in polarization is commonly called the
piezoelectric polarization and is given by
a-a0 (_
, C)
aQ (e31_e33^)'
rpK — Z—
(1)
where a is the lattice constant under strain, and o0 is the lattice constant of the relaxed
material. The constants e31 and e33 are piezoelectric constants and C13 and C33 are elastic
deformation constants. The total polarization in a given layer is simply the sum of the
spontaneous and piezoelectric polarization, i.e., P = Psp + PPE- The constants used in
our calculation are from Bernardini et al. [1] and Wright [5] and are shown in Table I. For
alloys, the constants are linearly interpolated. At a heterojunction there is usually a change
in the polarization on each side. This abrupt change in polarization causes a bound sheet
charge. In general, the bound sheet charge is the polarization of the bottom layer minus the
polarization of the top layer, a = P (bottom) - P (top).
We now examine the particular situation present in AlGaN/GaN field effect transistors.
We assume the thin AlGaN layer is pseudomorphically lattice matched to the thick GaN
layer below it. Figure 1 shows the situation for both Ga-face and N-face material. In a
Ga-faced structure, a positive bound charge is created at the deeper interface which causes
the formation of a 2-D electron gas at the lower interface. In N-face material, the positive
bound charge is present at the upper interface and the 2-D electron gas will form there.
With the proper structure geometry, it may be possible to form a 2-D hole gas at the other
interface [6], however, this has yet to be reported experimentally.
With a theoretical understanding of polarization we now wish to predict the effect this
polarization will have on conduction band diagrams and electron concentrations for a variety of device structures. For this purpose we use a 1-D Schrödinger-Poisson solver. First
the Schrödinger's equation is solved to find the electron concentration and then Poisson's
equation is solved to determine the conduction band profile. The new conduction band
profile will modify the solution of the electron concentration, so the process is repeated until there is little change in the solution, i.e., the solution is self-consistent. Our computer
program which solves this self-consistent problem, called CBAND, is very similar to other
self-consistent solvers [7]. The program however, must be modified to incorporate the effects
of the spontaneous and piezoelectric polarization. This is accomplished by adding thin layers
of charge at the heterojunction interfaces equivalent to the bound sheet charge caused by the
502
N-face
Ga-face
1 PSP
-o
tensile
strain
GaN
1 PSP
2DEG
j PPE
AlGaN
+c
2DEG
+ PSP
GaN
Figure 1: The direction of polarization and the location of the 2DEG in Ga-face and N-face AlGaN
HFETs. In both cases, the AlGaN layer is under tensile strain leading to both a spontaneous and
piezoelectric component to the polarization. For Ga-face materal the direction of polarization causes
the formation of a 2DEG at the lower interface. In N-face material the direction of polarization is
reversed causing the 2DEG to form at the upper interface.
polarization. The charge in these layers is added to the space charge used to solve Poisson's
equation.
It is also necessary to specify boundary conditions for the conduction band at the surface,
heterojunction, and substrate. In our structures we assumed a Schottky barrier contact at
the surface, pinning the conduction band to a fixed value. For AlxGaN, we assume the barrier
is $B = 0.84 + 1.3a; eV[2]. At the heterointerface, we use the conduction band discontinuity,
AEC, given by Wei and Zunger [8]. It turns out, for reasonable values of $B and AEC,
there is little impact on the resulting electron concentrations. This is the case, since the
electron concentration is dominated by the polarization induced bound charge. This can be
seen, for example, from the analytical expression given by Asbeck, Eq.(2) of [3] (or Eq.(15)
of [2]). At the substrate the conduction band is simply set to one half of the band gap of
the material next to the substrate. It turns out that the choice of conduction band level
at the substrate also has very little impact on the 2-D electron gas concentrations since the
substrate is typically far away from the heterojunction.
RESULTS
In Figure 2a the calculated solution for the conduction band diagram and electron
concentration for a typical AlGaN/GaN FET is shown. We assume the material is Ga-face,
and do not include the top GaN layer that is shown in Figure 1. The AlGaN barrier layer
is 300 Ä thick and contains 30% aluminum. In this case, the bound sheet charge, CT, is
1.68 x 1013 cm-2. Therefore a 6 A layer containing 2.8 x 1020 cm-3 positive space charge
is included at the interface. The conduction band discontinuity at the interface is 0.378 eV
and the Schottky barrier at the surface is pinned at 1.23 eV. The calculation uses a 1 micron
GaN buffer and the conduction band is pinned to 1.7 eV at the substrate boundary. From
the curvature of the conduction band we find the maximum electric field, located at the
503
2.0
a
ns= 1.4x10 cm"
1.5
73
C
pa
10"
- 4.5
10"
C-
<
E, = 0.004 eV -'_ 1.5
E0 =-0.125 eV
\
0.0
0.0
/
-0.5
c
o
3.0
1.0
0.5 -
o
U
6.0
200
h
c
c
o
U
c
SlO
• 3
td
Ö
gio
c
0
^m
0
ö
ä
SlO
.«
-1.5
10'
EF = 0eV
400 600 800 1000
Distance (Ang.)
200
400 600 800
Distance (Ang.)
1000
Figure 2: The calculated conduction band diagram and electron concentration for a typical AlGaN/GaN HFET. The 2DEG sheet density and the location of the subbands with respect to the
Fermi level are shown in (a). Figure (b) shows a comparison of the electron concentration calculated
by CBAND along with an experimentally measured profile determined by the capacitance-voltage
technique.
heterojunction, is about 2 MV/cm. The total electron sheet concentration is 1.4 x 1013
cm-2 and the maximum electron density is 5.4 x 1019 cm-3 with 22% of the electron sheet
density in the AlGaN layer. The electron concentration profile calculated by CBAND can
be compared with an experimentally determined profile through the capacitance-voltage
technique. As shown in Figure 2b, there is good agreement between the profiles over three
orders of magnitude in carrier concentration.
Next we compare the electron concentration of grown structures with our calculations.
Figure 3 depicts the measured electron concentration as a function of the aluminum content
in the barrier. The boxed items show the data for undoped structures grown by MBE [9] and
the circles show data for undoped structures grown by MOCVD. To demonstrate that both
spontaneous and piezoelectric polarization is important to explain all the sheet charge found
in AlGaN/GaN devices, calculated results are shown for three situations. The three curves
show the calculated sheet density for a 200 Ä AlGaN barrier with only the effects of piezoelectric polarization included, only the effects of spontaneous polarization, and with both. Most
of the experimental data lies between the calculation for the spontaneous polarization only
and the calculation that includes both spontaneous and piezoelectric polarization. Since the
CBAND sheet densities can be considered an upper bound it is clear that both spontaneous
and piezoelectric polarizations are important in determining the total sheet charge.
It is also fairly straight forward to use CBAND to investigate the conduction band profiles
and electron concentrations in novel device structures. In Figure 4 we show the results of
calculations for two possible device layers. Figure 4a shows how an AlGaN buffer can be
used to increase the confinement of the 2D gas by providing a greater than 1 eV conduction
band wall on either side of the 2D gas. In this structure a five percent AlGaN buffer is used.
When the material is abruptly changed to GaN the change in polarization induces a strong
504
10'
Spontaneous and Piezoelectric
10
Piezoelectric Only
c
Q 10
!-»
u
u
Spontaneous Only
c/5
10'
0.0
0.1
0.2
0.3
0.4
0.5
Aluminum Concentration, x
0.6
Figure 3: A comparison of experimentally determined sheet charge as a function of aluminum
concentration of the AlGaN barrier with that calculated by CBAND. The calculated curves show
results only including the effects of piezoelectric polarization, spontaneous polarization only, and
both.
electric field which bends the conduction band down sharply. In this case a 300 Ä GaN layer
is used. Then in the top barrier layer, a 30% aluminum layer, the direction of the electric field
is reversed and the conduction band is lifted sharply upward. The electron concentration in
this structure is predicted to be 1.2 x 1013 cm-2 and may increase performance due to the
increased confinement of the electrons. This may be especially important during transistor
operation, when the electron energy can rise dramatically due to the large electric field along
the channel.
We can further increase the sheet density of devices by using an InGaN buffer. The
conduction band diagram is shown in Figure 4b. Using 20% indium in the buffer increases
the strain and piezoelectric polarization of the the AlGaN barrier enough to increase the
sheet charge to 3.5 x 1013 cm-2. Increasing the indium content of the buffer continues to
increase the 2-D gas sheet concentration, however, at some point strain relaxation will occur.
CONCLUSIONS
In this paper the spontaneous and piezoelectric properties of the group Ill-nitrides have
been outlined. These effects have been incoporated into a self-consistent 1-D SchrödingerPoisson solver which calculates conduction band diagrams and electron concentration profiles
in group Ill-nitride device structures. It has been shown that both the spontaneous and
piezoelectric polarization are important for determining sheet carrier concentration observed
in grown devices.
ACKNOWLEDGEMENTS
The authors wish to thank the Office of Naval Research for financial support under their
MURI program: Grant # N00014-96-1-1223; Project Monitor Dr. J. Zolper. Dr. Ambacher
505
2.0
2.0
a
ns= 1.2x10 cm"
>
a
o
3 C
c
o
ta
/E, = 0.046 eV 1 i~c
I/
E0 = -0.115eV
o
c
o
O
1
-1 o
<=
1.0
i
c
o
o
o 0.0
§
u
Al30GaN
-1.0
5?
GaN
'
i
20
i
13
—2
ns = 3.5x10 cm
15 B
>
o
c
o
10 w
1.0
m
E, =-0.065 eV
E0 = -0.313 eV
c
o
3
T)
Al05GaN
Al30GaN
-1.0
400 600 800 1000
Distance (Ang.)
5
bc
u
o
3
0.0
c
o
-W
200
r
b
In20GaN
§
U
c
-5 go
.3
-10
200
400 600 800
Distance (Ang.)
1000
Figure 4: Two novel device structures calculated by CBAND. Figure (a) shows the results for
a structure which uses an AlGaN buffer to increase the 2DEG confinement. A structure which
includes indium in the buffer to increase the strain and piezoelectric polarization in the AlGaN is
shown in (b).
would like to thanks the Alexander von Humboldt Stiftung for a Feodor Lynen fellowship.
REFERENCES
[1] F. Bernardini, V. Fiorentini, and D. Vanderbilt, Phys. Rev. B 56, R10024 (1997).
[2] O. Ambacher, J. Smart, J. R. Shealy, N. G. Weimann, K. Chu, M. Murphy,
W. J. Schaff, and L. F. Eastman, R. Dimitrov, L. Wittmer, M. Stutzmann, W. Reiger,
and J. Hilsenbeck, J. Appl. Phys. 85, 3222 (1999).
[3] P. M. Asbeck, E. T. Yu, S. S. Lau, G. J. Sullivan, J. Van Hove, and J. Redwing,
Elec. Lett. 33, 1230 (1997).
[4] A. D. Bykhovski, R. Gaska, M. S. Shur, Appl. Phys. Lett. 73, 3577 (1998).
[5] A. F. Wright, J. Appl. Phys. 82 2833 (1997).
[6] F. D. Sala, A. D. Carlo, P. Lugli, F. Bernardini, V. Fiorentini, R. Scholz, J.-M. Jancu,
Appl. Phys. Lett. 74, 2002 (1999).
[7] I-H. Tan, G. L. Snider, L. D. Chang, and E. L. Hu, J. Appl. Phys. 68 4071 (1990).
[8] S. Wei, A. Zunger, Appl. Phys. Lett. 69, 2719 (1996).
[9] M. J. Murphy, B. E. Foutz, K. Chu, H. Hu, W. Yeo, W. J. Schaff, O. Ambacher,
L. F. Eastman, T. J. Eustis, R. Dimitrov, M. Stutzmann, W. Rieger, MRS Internet
J. Nitride Semicond. Res. 4SI, G8.4 (1999).
506
PIEZOELECTRIC SCATTERING IN LARGE-BANDGAP SEMICONDUCTORS AND
LOW-DIMENSIONAL HETEROSTRUCTURES
B.K. RIDLEY*, N.A. ZAKHLENIUK, C.R. BENNETT, M. BABIKER, and D.R. ANDERSON
Department of Physics, University of Essex, Colchester, C04 3SQ, UK
* also Department of Electrical Engineering, Cornell University, USA
ABSTRACT
We develop a rigorous theory of piezoacoustic phonon limited electron transport in bulk
GaN and GaN-based heterostructures. Within the Boltzmann equation approach we derive a new
expression for the momentum relaxation rate and show that the Pauli principle restrictions are
comparable in importance to a screening effect at temperatures up to 150 K provided that the
electron density is large. This is of particular importance for electrons in GaN/AlN-based
quantum wells where very high electron densities initiated by the piezoelectric effect have
recently been reported. Variations of the piezoacoustic phonon limited electron mobility with the
lattice temperature and with the electron density for a zinc-blende and wurtzite GaN are
presented.
INTRODUCTION
Large-bandgap semiconductors, such as GaN and A1N, have important piezoelectric
properties due to the large values of their piezoelectric tensor components [1]. As a consequence
of this, the electron interaction with piezoacoustic phonons provides a significant contribution to
the transport coefficients of the electrons within a very wide temperature range. The importance
of piezoacoustic scattering is well known and has been studied both experimentally and
theoretically. Usually piezoacoustic scattering dominates at low temperatures, but this is not the
case in the context of large-bandgap semiconductors where even near the room temperature this
scattering can be comparable with polar optical and deformation acoustic phonon scattering
[2, 3]. This situation appears to be ripe for investigations on piezoelectric scattering in lowdimensional heterostructures which rigorously and consistently take into account specific
features of the nitride large-bandgap semiconductors.
In this paper we study the piezoacoustic mobility of two-dimensional (2D) electrons
confined in an infinite potential quantum-well (QW) formed from a GaN/AIN double
heterostructure. Both the zinc-blende (ZB) and wurtzite (WZ) crystal structures of GaN are
considered. The effect of the screening on the electron-piezoacoustic-phonon interaction is
incorporated via the Lindhard dielectric function. For comparison we also calculate the electron
mobility of the three-dimensional (3D) electron gas in bulk GaN.
507
Mat. Res. Soc. Symp. Proc. Vol. 572 ®1999 Materials Research Society
THEORY
In order to calculate the electron kinetic coefficients it is necessary to solve the Boltzman
kinetic equation for the distribution function F(k),
eE dF(k)
' h dk
YXwnF{1')[1 - F(*)] - wwF^1 - F& 4
(1)
k
where Wrr, is the probability of scattering between electron states with wavevectors k and k'
due to the phonon emission and absorption processes, E is an external electric field, and e is the
electronic charge. Because we will consider an electron gas with an arbitrary degree of
degeneracy, the Pauli principle restrictions (PPR) are explicitly taken into account in Eq. (1).
Expressing F(k) as a sum of a symmetric part F0(e) and an antisymmetric part F\(k), we
obtain a solution for the antisymmetric part in the form:
F1(^) = AeT(£)^£)(^.f);
m
(2)
de
where m* is an electron effective mass, F0(e) = [exp{(e - eF) I k0T0} +1]
is an equilibrium
Fermi distribution function, eF is the Fermi energy, T0 is the lattice temperature, k0 is the
Boltzmann constant, and r(e) is the momentum relaxation time which is given by the
expressions
15,-T 24^IC(«!W^»<V*^ -«>.
®(E,h(Dq) =
\Nq + 1X1-F0(e-hcoq)]
|
Nq[\-F0{e + haq)-\
<»
(4)
Here C(q) and Gep(q) are, respectively, the coupling constant and the form-factor for the
electron-phonon interaction, eel(q') is the static Lindhard dielectric function [4] which accounts
for the electronic screening of the electron-phonon interaction, coq = caq is the phonon
frequency at wavevector q, ca is the longitudinal (or = L) or transverse (a = T) sound velocity,
Nq =\exp(hmq /k0T0)-i\
is the phonon distribution function. The expressions obtained
above are general for 3D and 2D electron gases interacting with bulk deformation acoustic or
piezoacoustic phonons. In the 3D case k is a three-dimensional electron wavevector and q' = q,
while in the 2D case k is a two-dimensional electron wavevector and q' = q\\ with q^ the in-plane
component of the phonon wavevector q. The form-factor Gep(q) = l for a 3D gas and
|G«p07l)| = rc%_|d / 2)~2 sin2(^d / 2) / [n2 - {qLd I iff for 2D electrons occupying only
the first subband of the infinite potential QW, where q±_ is the component of q normal to the
508
QW and d is the QW width. A careful analysis of the expression for <&{e,h(Oq) in Eq.(4), which
after using expressions for F0(e) and Nq can be presented in an equivalent form as
^(e,ho)q) = (2Nq + l)cosh2[(e - eF) 12k0T0] I [cosh2[(e - eF) 12k0T0] + smh2(fi(oq 12k0T0)l
shows that despite the fact that the phonon energy hcoq is always smaller than the electron
energy e, which is of the order of eF, the phonon energy must be retained in the argument of the
functions F0(e±fiö)„) provided that the lattice temperature is low, k0T0<^Bm caeF. A
numerical estimation of the right-hand side of this inequality gives respectively 30 K for bulk
GaN with electron density no=10l8 Cm~3 and 140 K for 2D electrons of density no=3xl0^ cm2
(such high electron densities have recently been reported [5] in a GaN/AIN heterostructure and
were attributed to the doping effect of the intrinsic piezoelectric field). The term in question in
Eq. (3) takes into account the effect of PPR on the momentum relaxation rate and its presence
means that the relaxation rate of an individual electron depends on the electron density n0 (even
if the screening is ignored) through the Fermi energy. At high lattice temperatures, when
k0T0 >^%tnc\eF , we obtain ®(e,haq)~(2Nq +\)~2k0T0 Ihmq and r(e) depends on n0
only in the presence of the screening. This brings qualitatively new features into the temperature
and density dependences of the electron mobilities. It is important to note that the expression
obtained in Eq. (3) for the momentum relaxation rate is only an approximate solution within the
Boltzmann equation approach. It becomes exact only in the test particle approach [6, 7] in the
Bloch-Gruneisen regime. The details of its derivation will be reported elsewhere.
The coupling coefficient \C(q)\ = \h(eh) / 2pVcaq\Ha(e) for the piezoacoustic
phonons depends in general on the phonon wave polarization a and on the orientation of the
vector e = qlq with respect to the crystal axes [8] (here p is a material density, V is a sample
volume, and h is a component of the piezoelectric tensor). The explicit form of the function
Ha(e) depends on the crystal symmetry. For the ZB symmetry h = hl4,
HL(e) = (6qxqyqz/q3)2 and HT(e) = 4[q2xq2 / q4+q2xq2 / q4+q2q2 / q4]-HL(e) are the
longitudinal and transverse components, respectively. For the crystal with WZ symmetry
h = h33, HL(e) = (qz I q)2[\ - {hx I h^){q2x +q2)/q2]2 with the notation hx =h33-h31- 2h15,
and HT{e) = [{q2 + q2)I q2][hi5 I h33 + (hx I' h33)q2 I q>2]2. The usual approximation is to use
the angular average of Ha(e).This approach is justified in the case of bulk semiconductors, but
it requires a special investigation in the case of low-dimensional heterostructures. The physical
reason for this is that for 2D electrons there is no momentum conservation along the quantization
direction. As a consequence of this, the phonon wavevector components q^_ and q\\ play
principally different roles in the electron-phonon interaction. Using the bulk-like expression for
the coupling coefficient may lead to the wrong estimate of the corresponding interaction strength.
It is necessary first to express Ha(e) as a functions of q± and q\\ and then average it over the
509
orientations of the in-plane wavevector q\\ [9]. As a result, the coupling coefficient does not
depend on the orientation of q«, but it depends on the ratio of q± I q^ and also on the crystal
orientation of the QW planes. Here we will consider three different orientations of the QW for
the ZB structure, when the 2D planes are parallel to the (001), (111), and (110) planes,
respectively, and one specific orientation of the QW for the WZ structure: namely when the 2D
planes are normal to the [0001] axis which is the usual growth direction.
The low field electron mobilities are calculated using Eqs. (2) - (4) and we obtain
m knT,
ith£
2f8J2S-\°l P
Jx2Tp(*)F0(*Xl-F0(x)]<fe,
\9n.
(5)
where p=3 for 3D electrons and p=2 for 2D electrons, x = e I k0T0. The Fermi energy at a given
lattice temperature T0 and a given electron density n0 is defined by the equation
2n2K3n0l{2m*k0T0)V2=[xll2[exv(x-z0) + \Yldx for the 3D electron gas, where
o
z0 = £p I k0T0, and for the 2D electrons z0 = ln[exp(7i/i n0 I m k0T0)-1].
RESULTS AND DISCUSSION
First we evaluate the mobility of 2D electrons as a function of lattice temperature for a given
sheet density. This is shown in the Fig. 1 for ZB and WZ. For ZB the mobility is almost
independent of the orientation of the plane of the QW. The following material parameters were
used in our calculations: p = 6.lg/cm, c^ =6.56x10 cmls, Cj-=2.68xl0 cmls,
es =9.5 for both ZB and WZ structures, m* =0.21mo, hH =6.66xl07 VIcm for ZB, and
m* = 0.23mo, h33 =10.86xl07 VI cm, h31 =-3.91xl07 VI cm, hi5 = -3.57xl07 VI cm for
and the electron sheet density is n0 = 10
:>
6.
10s
\
Screened Without PPR
,. \^ Unscreened With PPR
^ -. X^Screened Equipartition
" .
Vm
107
«£ 106
«fe 10
"o 105
i"
4
10
1000
.
10s
10'
„ 1°8
y io7
cm
ZB
~" 13
n =10
0
i"
-^TT~~~---~
2
cm'
~""~
104
1000
10
T0(K)
100
-~.rr
WZ
n =10
13
0
2
cm"
100
10
To(K)
FIG. 1. Temperature dependence of the 2D mobility in WZ and ZB GaN/AIN
heterostructures using different models for the momentum relaxation rate.
510
—
As is seen in the Fig. 1 the PPR plays an important role at temperatures below 50K.
Screening is also important and it leads to an increase in the mobility in comparison with the
unscreened case. But at T0 less than 20K the effect of the PPR is much stronger than the
screening effect. For comparison we also show the calculated mobility using the equipartition
approximation for the phonon distribution function. This approximation can be used only at
higher temperatures to give the correct value of the mobility.
10°
10'
'»
10
*S^
^;
^^ ^^-^^ /
«fe 10
• Üfe
W
^^J>
^==^^^
s
o10
s=
4
tf
„ =: = ' "
"
T = 20K
°
:
^^s^
^-^»^^ ~~
r^~Z
1000
10"
^x
_- '
^^^
~~~
T = 200K
Screened Without PPR
Unscreened With PPR
10"
n (cm'
s
4
0
1000
10"
10"
10"
n (cm"
10
FIG. 2. Mobility of 2D electrons in WZ and ZB GaN/AIN heterostructures as a
function of the electron density at two different temperatures: T=200K and
T=20K using different models for the momentum relaxation rate. The upper curve
in each figure for each pair corresponds to WZ GaN.
In Fig. 2 we display the calculated mobility of 2D electrons for WZ and ZB GaN as a
function of the electron density at To=20K and To=200K. At low temperatures the PPR is
important if the density is bigger than 1012 cm-2. At high temperatures the PPR as expected is
not important, but the effect of screening is still considerable.
T0 = 200K
Bulk
Screened With PPR Unscreened With PPR 10"
10"
n (cm3
FIG. 3. Mobility of 3D electrons in bulk GaN as a function of the electron
density. The upper curve for each pair corresponds to WZ GaN.
511
In Fig. 3 the same dependences are presented for 3D electrons. Because the Fermi energy
of a 3D electron gas of density no=1018 cm-3 is smaller than that of 2D electrons of density
no=1013 cnr2, the PPR for 3D electrons is important at sufficiently lower temperatures. For the
range of densities n0 shown in Fig. 3 these temperatures are below 20K and the PPR is not
important here. There is a remarkable difference between effect of the screening in 2D and 3D
cases. In the 2D case the screening becomes weaker when n0 increases while in the 3D case this
is vice versa. The reason is that the screening length in the 2D case does not depend on n0 while
in the 3D case it does depend [4]. At the same time the increase in n0 leads to an increase of the
Fermi energy and consequently to increase of the wavevectors of the phonons interacting with
electrons. It follows from the Lindhard function [4] that in the 2D case this results in a
weakening of the screening effect.
CONCLUSION
The main conclusion which can be drawn from the analysis above is that in GaN at
temperatures below 150K both the effects of screening and the PPR should be taken into account
in order to obtain the correct result for the piezoacoustic phonon limited mobility in 2D case if
the electron density is 10*3 cm" 2 or higher. This also true for the deformation acoustic phonon
scattering. This conclusion becomes important in light of recent data about high electron sheet
densities in GaN/AIN heterostructures because of internal piezoelectric field doping. Also, we
note that WZ material has a slightly higher mobility than ZB material and that the mobility is not
sensitive to the orientation of the plane for 2D electrons.
REFERENCES
[1] A. Bykhovski, B. Gelmont, M. Shur, and K. Khan, J. Appl. Phys. 77, p. 1616 (1995).
[2] R. Oberhuber, G. Zander, and P. Vogl, Appl. Phys. Letters 73, p. 818 (1998).
[3] N.A. Zakhleniuk, C.R. Bennett, B.K. Ridley, and M.Babiker, Appl. Phys. Letters 73, 2485
(1988).
[4] For a 3D electron gas the Lindhard dielectric function can be found in G.D. Mahan, ManyParticle Physics, Plenum Press, New York, 1990, p. 438, and for a 2D gas in T. Ando, A.B.
Fowler, and F. Stern, Rev. Mod. Phys. 54, p.450 (1982).
[5] R. Gaska, M.S. Shur, A.D. Bykhovski, A.O. Orlov, and GL. Snider, Appl. Phys. Lett. 74, p.
287(1999).
[6] PJ. Price, Solid State Commun. 51, p. 607 (1984).
[7] V.E. Gantmakher and Y.B. Levinson, Carrier Scattering in Metals and Semiconductors,
North-Holland, Amsterdam, 1987, Chapters 2 and 4.
[8] J.D. Zook, Phys. Rev. 136, p.869 (1964).
[9] V. Karpus, Semicond. 21, p.1180 (1987).
512
ACTIVATION CHARACTERISTICS OF DONOR AND ACCEPTOR
IMPLANTS IN GaN
X. A. Cao,* S. J. Pearton,* R. K. Singh,* R. G. Wilson,** J. A. Sekhar,*** J. C. Zolper,**** J.
Han,***** D. J. Rieger,***** R. J. Shul,***** H. J. Guo ****** S. J. Pennycook,****** and J.
M. Zavada*******
*Department of Materials Science and Engineering, University of Florida, Gainesville, FL
32611, USA
* »Consultant, Stevenson Ranch, CA 91381, USA
***Micropyretics Heaters International, Inc. Cincinnati, OH 45215, USA
****Office of Naval Research, Arlington, VA 22217, USA
*****Sandia National Laboratories, Albuquerque, NM 87185, USA
******Oak Ridge National Laboratory, Solid State Division, Oak Ridge, TN 37831,USA
*******European Research Office, USARDSG, London, England
ABSTRACT
The ionization levels of different donor and acceptor species implanted into GaN were
measured by temperature-dependent Hall data after high temperature (1400 °C) annealing. The
values obtained were 28 meV (Si), 48 meV (S), 50 meV (Te) for the donors, and 170 meV for
Mg acceptor. P-type conductivity was not achieved with either Be or C implantation. Basically
all of the implanted species show no distribution during activation annealing. For high implant
doses (5xl015 cm"2) a high concentration of extended defects remains after 1100 °C anneals, but
higher temperatures (1400 °C) produces a significant improvement in crystalline quality in the
implanted region.
INTRODUCTION
Ion implantation is an effective technology for selected-area doping or isolation of GaNbased devices.1"3 As reviewed previously by Zolper,1 implantation of donors at high dose
(>5xl0 cm"2) can be used to decrease source and drain access resistance in field effect
transistors (FETs), at lower doses to create channel regions for FETs, while sequential
implantation of both acceptors and donors may be used to fabricate p-n junctions. Two different
device structures have been demonstrated using the latter method, namely a junction field-effect
transistor2 and a planar, homojunction light-emitting diode (LED).4
At high implant doses (>5xl0!4 cm"2) it is clear that conventional rapid thermal annealing
(RTA) at 1100-1200 °C can activate the dopants but not remove the ion-induced structural
damage.5 At higher annealing temperatures (>1400 °C), the equilibrium N2 pressure over GaN is
>1000 bar,6 and only two methods have proven effective in preventing surface decomposition.
The first is use of high N2 pressures (15 kbar),7 and the second is deposition of A1N
encapsulation layers8 (at 1400 °C the equilibrium N2 pressure above A1N is only 10"8 bar). The
latter is clearly more convenient.
What is needed is a better understanding of the optimum implanted species for creation of nand p-type regions in GaN, based on the highest achievable electron or hole concentrations, the
residual damage and the redistribution during high temperature annealing. In this paper we report
on the high temperature activation characteristics of donor implant species (Si, S and Te) and
513
Mat. Res. Soc. Symp. Proc. Vol. 572 ® 1999 Materials Research Society
acceptor species (Mg and C). The redistribution of all these dopants plus Se and Be was
investigated by using secondary ion mass spectrometry (SIMS). Finally, the efficiency of high
temperature RTA for removing lattice damage in implanted GaN was examined by transmission
electron microscopy (TEM).
EXPERIMENTAL
Layers of GaN 2-3 urn thick were grown at -1040 °C on c-plane A1203 by atmospheric
pressure Metal Organic Chemical Vapor Deposition (MOCVD), using triethylgallium and
ammonia. From x-ray diffraction and photoluminescence measurements we know this material is
typical of the current state-of-the-art heteroepitaxial GaN.
The samples were implanted at 25 °C with 150 keV 24Mg+, 80 keV 9Be+, 80 keV 12C+, 200
keV 32S+, 300 keV 80Se+, or 600 keV 128Te+ ions at doses of 3-5x1014 cm"2, and with 150 keV
28 +
Si at a dose of 5xl015 cm'2. This puts the projected range, Rp, of the implanted species at least
1500 Ä into the GaN in all cases, avoiding effects due to near-surface point defect injection. The
samples were capped with -1000 A of reactively sputtered A1N, and annealed at temperatures of
900-1500 °C under a N2 ambient in the Zapper furnace described previously.9 The dwell time
at the peak temperature was -10 sees. After annealing, the A1N was selectively etched in
aqueous KOH at 80 °C.10 For measurement of the electrical properties, Hgln ohmic contacts
were alloyed to the corners of 3x3 mm2 sections, and Hall effect data was recorded at 25 °C in all
cases. The atomic distributions before and after annealing were measured by SIMS, and the data
quantified using the as-implanted sample as a standard. In addition, the Si-implanted samples
were also examined by plan-view TEM, since Si remains the standard implant species for GaN
n-type doping, and it has a representative mass number that allows comparison to damage
expected with S, Ca and Mg.
RESULTS AND DISCUSSION
Figure 1 shows an Arrhenius plot of sheet carrier concentration in Si+ implanted material. In
the A1N encapsulated samples activation occurs with an activation energy of -5.2 eV before
saturating at -1400 °C. We interpret this activation energy as the average required to move the
interstitial Si atom to a vacant substitutional site by short-range diffusion and to simultaneously
remove compensating point defects so that the Si is electrically active. Note that at 1500 °C the
sheet electron density decreases, and this was accompanied by a decrease in carrier mobility.
This increase in compensation is consistent with Si beginning to occupy both Ga sites (where it
is a donor), and N sites (where it is an acceptor). This is commonly observed with Si
implantation in other III-V materials.11 The peak n-type doping level we obtained is -5x1020
cm"3, and the corresponding activation efficiency is 90%. This very high doping level produces
extremely good specific contact resistances for W and WSix metallization, with values <10"6
Qcm2 after annealing in the range 600-900 °C. This demonstrates the efficiency of the
implantation approach for reducing contact resistances in GaN electronic devices.
There is also interest in the group VI donors, S, Se and Te, which do not have the potential
drawback of being amphoteric like Si in GaN. Figure 2 shows the Arrhenius plots of S+ and Te+
activation in GaN. The sheet carrier concentrations show activation energy of 3.2 eV and 1.5 eV
for the temperature range between 1000-1200 °C for S+ and Te+ respectively. These are
significantly lower than that for Si, but their physical origins should be basically the same. The
sheet electron density essentially saturates above 1200 °C for S implantation, with maximum
volume density of ~5xl018 cm"3. However the electron density does not saturate even at 1400 °C
514
for the Te case. It is likely that because of the much greater atomic weight of 128Te, even higher
annealing temperatures would be required to remove all its associated lattice damage, and that
the activation characteristics are still being dominated by this defect removal process. Even
though implanted Si+ at the same dose showed evidence of site-switching and self-compensation,
it still produces a higher peak doping level than the non-amphoteric donors S and Te. This
suggests the group VI donors do not have any advantage over Si for creation of n-type layers in
GaN. From temperature-dependent Hall measurements, we find the ionization levels of Si, S and
Te are 28 meV, 48 meV and 50 meV respectively, so that the donors are fully ionized at room
temperature. This was verified by elevated temperature Hall measurements (up to 150 °C), where
no additional increase in free carrier concentration was observed. In this data, we assumed a
compensation level of-25%, as is typical for GaN implanted material.6
1018r
£
2
10"
g
o
10"
o>
10"
Si in GaN
100keV5x1015cm"2
10 sec anneal
0.6
0.7
0.8
iooon-(i/K)
Fig. 1. Arrhenius plot of sheet electron concentration versus inverse anneal temperature for Si+
implanted GaN.
— 10«
S
\
^s
E
8
1012
0.55
4 -
10«
0.60
0.65
,
S*,Te+->GaN
5 x 1014 cm-2
\
\
EaS = 3.2eV
Ea.Te=1-5eV
1
0.70
1 ^1-1
0.75 0.80
1
0.85
0.90
iooon"(ic )
Fig. 2. Sheet carrier densities in S+ or Te+ implanted GaN as a function of
annealing temperature.
The effects of post-implant annealing temperature on the sheet carrier concentrations in Mg+
and C+ implanted GaN are shown in Figure 3. There are two important features of the data: first,
we did not achieve p-type conductivity with carbon, and second only -1% of the Mg produces a
hole at 25 °C. Carbon has been predicted previously to have a strong self-compensation effect,12
and it has been found to produce p-type conductivity only in metal organic molecular beam
515
epitaxy where its incorporation on a N-site is favorable.13 Based on an ionization level of-170
meV, the hole density in uncompensated Mg-doped GaN would be calculated to be -10% of the
Mg acceptor concentration when measured at 25 °C. In our case we see an order of magnitude
less holes than predicted. This should be related to the existing n-type carrier background in the
material and perhaps to residual lattice damage which is also n-type in GaN, At the highest
annealing temperature (1400 °C), the hole density falls, which could be due to Mg coming out of
the solution or to the creation of further compensating defects in the GaN.
•
—«" !"
i
i
•
i
'
^z—• ——
^
/
o
Ü
a
D
Mg*, C*-»GaN 5x1014cm-2
a Mg p-type
Mg n-type
• C n-type
4
"E
O
W
i
2x10"
900
•
1000
i.i
1100
1200
1300
1400
1500
Annealing Temperature (°C)
Fig. 3. Sheet carrier densities in Mg+ or C+ implanted GaN as a function of annealing
temperature.
Figure 4 shows the SIMS profiles of Si as-implanted and 1400 °C annealed samples. There is
little redistribution of the Si at 1400 °C, with DSi < 10"13 cm2 -s"1 at this temperature calculated
from the change in width at half-maximum. Similarly, we found that S, Se, Te, Mg and C are all
extremely slow diffusers when implanted into GaN, with Deff <2xl0"13 cm2-sec_1 at 1450 °C. The
extremely stable nature of dopants in GaN means that junction placement should be quite precise
and there will be fewer problems with lateral diffusion of the source/drain regions towards the
gate. This is promising for the fabrication of GaN-based power devices, which require creation
of doped well or source/drain regions by implantation.
,1, 102°
a
o
:r^
- - as-implanted
1400 °C, 10 s
\.
\-
2 io's
a
o
10'8
10'7
N..
'S-..J
Si* in GaN 150 keV 5x1015 cm'2
_i
I
0.2
i
I
i
0.4
I
0.6
i
i_
0.8
Depth (um)
Fig. 4. SIMS profiles of implanted Si in GaN (150 keV, 5xl015 cm"2) before and after annealing
at 1400 °C for 10 s.
516
In the particular case of implanted Be, there was an initial broadening of the profile at 900 °C
(figure 5), corresponding to an effective diffusivity of ~5xl0"13 cm2-sec_1 at this temperature.
However there was no subsequent redistribution at temperatures up to 1200 °C. It appears that in
GaN, the interstitial Be undergoes a type of transient-enhanced diffusion until these excess point
defects are removed by annealing, at which stage the Be is basically immobile.
0.5
1.0
Depth (pm)
Fig. 5. SIMS profiles of implanted Be in GaN (80 keV, 5xl014 cm"2) before and after annealing
at different temperatures for 10 s.
Figure 6 shows a plan view TEM from a Si-implanted sample after annealing at 1100 °C
(left) and 1400 °C (right) for 10 sees. This high dose implant (150 keV, 5xl015 cm"2) represents a
worst-case scenario in terms of damage removal. The sample still contains a high density of
extended defects (~1010 cm"2) after 1100 °C annealing. We ascribe these defects to the formation
of dislocation loops in the incompletely repaired lattice. By sharp contrast, annealing at 1400 °C
for 10 sees brings a substantial reduction in the implant-induced defects. The lower density of
defects (~109 cm"2) could be ascribed to the threading dislocations arising from lattice-mismatch
in the heteroepitaxy. This appears to correlate well with the fact that the highest electron mobility
and carrier density in these samples was observed for 1400 °C annealing. Clearly the ultra-high
temperature annealing is required to completely remove lattice damage in GaN implanted with
high doses. However it may not be needed in lower dose material (<5xl013 cm"2) where the
amount of damage created is correspondingly less.
SUMMARY AND CONCLUSIONS
The redistribution and activation of a wide variety of possible donor and acceptor species in
GaN was examined. S and Te are found to produce lower room-temperature n-type doping levels
than Si at the same dose, while only Mg was found to produce p-type doping (C and Be
implanted samples remained n-type). None of the implanted species showed measurable
diffusion at 1450 °C, but Be did display an apparent defect-assisted redistribution at lower
temperature (900 °C). For high implant dose (e.g. 5xl015 cm"2 for Si+), annealing at 1100 °C is
insufficient to remove the lattice damage, while 1400 °C produces much lower defect densities.
517
Fig. 6. TEM plan view from Si+ implanted GaN (5xl015 cm"2,150 keV) after 1100 °C
(left) and 1400 °C (right), 10 s annealing.
ACKNOWLEDGEMENTS
The work at UF is partially supported by grants from DARPA/EPRI (D. Radack and J.
Melcher), MDA 972-98-1-0006, and NSF (L. Hess), DMR-9732865. The work of R. G. Wilson
is partially supported by a grant from ARO. Sandia is a multiprogram laboratory operated by
Sandia Corporation, a Lockheed-Martin company, for the US Department of Energy under
contract No. DEAC04-94 AL 85000.
REFERENCES
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
J. C. Zolper, GaN and Related Materials, edited by S. J. Pearton, p. 371 (Gordon and
Breach, NY 1997).
J. C. Zolper and R. J. Shul, MRS Bulletin 22, 36 (1997).
S. C. Binari, L. B. Rowland, W. Kruppa, G. Keiner, K. Doverspike and D. K. Gaskill,
Electron. Lett. 30,1248 (1994).
H. P. Maruska (unpublished, 1997).
H. H. Tan, J. S. Williams, J. Zou, D. J. H. Cockayne, S. J. Pearton and R. A. Stall, Appl.
Phys. Lett. 69,2364(1996).
S. Porowski and I. Grzegory, GaN and Related Materials, edited by S. J. Pearton, p. 295
(Gordon and Breach, NY 1997).
J. C. Zolper, J. Han, S. B. Van Deusen, M. H. Crawford, R. M. Biefeld, J. Jun, T. Suski, J.
M. Baranowski and S. J. Pearton, Mat. Res. Soc. Symp. Proc. 482, 609 (1998).
J. C. Zolper, D. J. Reiger, A. G. Baca, S. J. Pearton, J. W. Lee and R. A. Stall, Appl. Phys.
Lett. 69, 538 (1996).
M. Fu, V. Sarvepalli, R. K. Singh, C. R. Abernathy, X. A. Cao, S. J. Pearton and J. A.
Sekhar, Mat. Res. Soc. Symp. Proc. 483, 345 (1998).
J. R. Mileham, S. J. Pearton, C. R. Abernathy, J. D. MacKenzie, R. J. Shul and S. P.
Kilcoyne, Appl. Phys. Lett. 67,1119 (1995).
S. J. Pearton, J. S. Williams, K. T. Short, S. T. Johnson, D. C. Jacobsen, J. M. Poate, J. M.
Gibson and D. O. Boerma, J. Appl. Phys. 65,1089 (1989).
P. Bogulawski, E. L. Briggs and J. Bernholc, Phys. Rev. B 51,17255 (1995).
C. R. Abernathy, J. D. MacKenzie, S. J. Pearton and W. S. Hobson, Appl. Phys. Lett. 66,
1969(1995).
518
TRANSMUTATION DOPING OF DI-NITRIDES
GALINA POPOVICI
Rockbit International Ine, 7601 Will Rogers Blvd, Fort Worth, TX 76140
gpopovici@rbi-gearhart.com, rrmihai@showme.missouri.edu
Abstract
Transmutation doping of (In, Ga, A1)N compounds by neutron irradiation is a
promising and totally unexplored field to date. It is much more effective than that of
Si due to large neutron capture cross section and abundance of In, Al and Ga isotopes
participating in reaction. This should make the irradiation possible in low-flux reactors
and result in smaller radiation damage. Annealing of the radiation damage seems
feasible.
Doping is a controlled introduction of impurity atoms in appropriate sites and states
in the crystal lattice. It is designed to change the electrical, optical and/or other properties
of semiconductors in a controllable manner. There are four methods of doping: during
growth, by ion implantation, by diffusion and by transmutation.
In the case of nitrides and other wide band gap m-V semiconductors, doping is
generally performed simultaneously with the growth. Noteworthy progress in the growth
and doping of high quality epitaxial Ill-nitride films by a variety of methods has recently
been achieved.1'2 It is well known that it is easy to obtain GaN and InN of n-type through
unintentional doping. The source of the unintentional doping is still controversial.1'2,3
Unintentionally doped IriN has high electron concentrations up to 10 - 10 cm".
Controlled n-type conductivity in GaN and diluted InGaN alloys is generally achieved by Si
doping during growth 3 or by ion implantation.4 Ways to successfully and reproducibly
dope A1N have not been found.1 Under many conditions of growth and doping A1N
remains an insulator with resistivity of 109 to 1012 ficm.1
Ion implantation can be a useful method of GaN doping. Since p- and n-type, and
also highly resistive layers, have been realized, the method can be used to fabricate of allimplanted devices.5' 6 However, much more research should be done to achieve good
reproducibility.
Transmutation doping is used industrially for obtaining n-type silicon. Samples of
pure Si are irradiated with neutrons in nuclear reactors. One of the isotopes of Si can
absorb thermal neutrons and undergo subsequent ß-decay. As a result, Si transmutes into
phosphorus. The material is then annealed for healing the radiation damage related to the
unavoidable presence of fast neutrons. It is possible to control precisely the concentration
of P atoms by controlling the neutron fluence. Transmutation of B into Li was studied for
diamond doping 7'8 and showed promising results.
Data on relevant transmutation reactions for A1N, GaN and InN are given in Table
L Nitrogen has a very small cross section of neutron absorption. Even if it absorbs a
neutron, it changes into another stable isotope of N. Transmutations of Al, Ga and In
occurs with the absorption of a neutron and subsequent ß-decay, except for a weak channel
of transition of In into Cd, when the neutron absorption is followed by the capture of a K-
519
Mat. Res. Soc. Symp. Proc. Vol. 572 • 1999 Materials Research Society
shell electron or by the emission of a positron. 9,1° Absorption of the neutron puts the
mother nucleus in an excited state, from which it decays with a certain half life, given as tia
in Table I. In all cases given in Table I, the daughter isotopes are stable. The last column
in Table I gives the concentration of the transmuted atoms Nt for one hour of neutron
irradiation at a thermal neutron flux of 1013 cm"2 s"1. a) It was calculated through the
relation:
Nt = NasFt
where N is the concentration of mother atoms, a is the isotopic abundance of those
atoms, S is the cross section for neutron capture (averaged over thermal neutron
spectrum), F is the thermal neutron flux, t is the irradiation time.
Table I. Data on transmutation by neutron capture of Si, Al, Ga and In "
Mother
Nucleus
(Z,m)
Si (14, 30)
Al (13, 27)
Abund
ance
3.1
100
Cross
Section
(barn)
0.11
0.23
Daughter
Nucleus
(Z,m)
P (15,31)
Si (14, 28)
Ga(31,69)
Ga(31,71)
In (49, 115)
In (49, 115)
In (49, 113)
60.1
39.9
95.7
95.7
4,3
1.8
0.15
72
42
3
Ge (32, 70)
Ge (32, 72)
Sn(50, 116)
Sn(50, 116)
Sn(50, 114)
In (49, 113)
4.3
0.16
Cd (48, 114)
(%)
tl/2
Energy
(MeV)
1.49
2.6 h
2.3
2.82
min
21 min 1.65
14.1 h 1.5-3.15
54 min 0.6-11.6
3.3
14 s
1.2
~kV
nun
49.5
~kV
days
N, in 1 h
(cm"3)
4.5xl012
4.0x1014
1.6xl015
l.OxlO14
lxlO16
0.94xl016
3.0xl013
0.16xl013
Z is the atomic number, m is the atomic mass, cross sections are in barns
(lbarn=10-24crn2).
Concentrations of atoms used in the calculation:
N(Si) = 0.37xl023cm"3
N(A1) = 0.48>dO23 cm"3
N(Ga) = 0.43:xlO23 cm"3
Win) = 0.65x 10 cm
As one can see from Table I, neutron capture for In, Ga, and Al is more effective
than that for Si by two orders of magnitude. Therefore, neutron fluences for obtaining the
same dopant concentration will be about two orders of magnitude smaller. This will result
in smaller radiation damage and will make the irradiation possible in low-flux reactors.
All three transmutated atoms Si, Ge and Sn are amphoteric impurities and can enter
both sublattices, m and V. But it is unlikely that these atoms will enter the nitrogen
* For comparison: Missouri University Research Reactor has a thermal neutron flux of
l.lxlO'WW
520
sublattice on account of the small radius of the nitrogen atom (see Table II). Most
probably Si, Ge and Sn will enter the III-sublattice. All three dopants will thus be donors.
Table II. Atomic radii of the elements.
Atom
Atomic radius (um)
Al
Ga
N
Si
Ge
Sn
0.143 0.122 0.070 0.117 0.122 0.140
The recoil energy of transmutated atoms will be in the range used for ion
implantation ( 50-250 keV). Though the energy of the nuclear reaction is of the order of
MeV, from the conservation of impulse it follows that most of energy is taken by the
electron.
Successful implantation doping of GaN layers has shown that annealing of radiation
damage can be effective.4'5'6'12 A1N has a high radiation resistance and can be annealed to
high temperatures due to the fact that A1N is a much more stable compound than GaN and
InN.1 To our knowledge there are no papers on implantation or irradiation of InN layers.
However, transmutation doping of InN may be efficient. In has a high efficiency of
transmutation reaction, three order of magnitude higher than that of Si. It should result in a
short time of irradiation and consequently in a weak radiation damage.
Transmutation doping of IH-N semiconductors is promising and to date totally
unexplored field. Doping of GaN by transmutation of Ga into Ge should be feasible. Since
doping of A1N by more conventional methods was unsuccessful, an attempt to obtain ntype A1N by transmutation doping would be interesting from both experimental and
theoretical points of view. InN has the highest efficiency of transmutation. This should
result in a weaker radiation damage. InN transmutation doping thus may also be achieved.
REFERENCES
1
G. Popovici, H. Morkoc, and S. N. Mohammad, Deposition and Properties of IllNitrides by Molecular Beam Epitaxy in" Group III nitride semiconductor compounds,
Physics and Applications" Edited by Bernard Gil, (Claredon, Oxford UK), p. 19-69,1998.
2
G. Popovici and H. Morkoc, Growth and Doping of, and Defects in Ill-Nitrides in"
"III-V Nitride Semiconductors and Optoelectronic Devices" v.2 " edit by S. Pearton, to
be published in 1999.
3
W. Kim, A. E. Botchkarev, A. Salvador, G. Popovici, H. Tang and H. Morkoc,. J. Appl.
Phys. 82, 219 (1997).
4
S. J. Pearton, C. B. Vartuli, J. C. Zolper, C. Yuan and R. A. Stall, Appl. Phys. Lett,
67,1435 (1995).
5
M. Rubin, N. Newman, I. C. Chen, T. C. Fu, and J. T. Ross, Appl. Phys. Letters, 64, 64
(1994).
521
6
C. J. C. Zolper, R. G. Wilson, S. J. Pearton, and R. A. Stall, Appl. Phys. Lett, 68,1945
(1996).
7
G. Popovici, A. A. Melnikov, V.S. Varichenko, S. Khasawinah, T. Sung, M. A. Prelas,
A. B. Denisenko, N. M. Penina, V.A. Martinovich, E. N. Drozdova, A .M. Zaitsev, and
W. R. Fahrner, Diamond Relat.Mater.5, 761 (1996).
8
S.A. Khasawinah, G. Popovici, J. Farmer, T. Sung, M. A. Prelas, J. Chamberlain, and H.
White, J. Mater. Research, 10,2523 (1995).
9
R.L. Heat, Table of the Isotopes, in CRC Handbook of Chemistry and Physics, Ed.
RC. Weast, CRC Press, West Palm Beach, 1978, pp.B-270 to B-354.
10
F.W.Walker, J.RParrington, F. Feiner, Nuclides and Isotopes, 14-th Edition, GN
Nuclear Energy, General Electric Company, 1989.
11
The data compiled by M. Popovici, Missouri University Research Reactor.
12
H. H. Tan, J. S. Williams, J. S. Zou, d. J. H. Cockayne, S. J. Pearton, R A. Stall, Appl.
Phys. Lett. 69, 2364 (1996).
522
High Barrier Height n-GaN Schottky diodes with a barrier height of 1.3 eV
by using sputtered copper metal
W.C. Lai*, M. Yokoyama*, C.Y. Chang", J.D. Guo"\ J.S. Tsang*", S.H. Chan"*
and S.M. Sze'
"National Cheng Kong University, Department of Electrical Engineering, Tainan, Taiwan,
R.O.C.
"National Chiao Tung University, Institute of Electronic, Hsinchu 30050, Taiwan, R.O.C.
'"National Nano Device Laboratories, Hsinchu 30050, Taiwan, R. O. C.
ABSTRACT
Copper Schottky diodes on n-type GaN grown by metal-organic chemical vapor
deposition were achieved and investigated. Ti/Al was used as the ohmic contact. The copper
metal is deposited by the Sputter system. The barrier height was determined to be as high as
=1.13eV by current-voltage (I-V) method and corrected to be <DB =1.35eV as considered
the ideality factor, n, with the value of 1.2. By the capacitance-voltage (C-V) method, the
OB
barrier height is determined to be OB =1.41eV. Both results indicate that the sputtered copper
metal is a high barrier height Schottky metal for n-type GaN.
INTRODUCTION
GaN-based materials have been intensively studied recently for the blue emission diodes.
Besides that, they have the potential in the applications of high temperature, high frequency
and high power electronic devices due to its wide band gap and chemical stability
characteristics. For some electronic devices like MESFET, it is necessary to fabricate a
Schottky contact with a high breakdown voltage and a low reverse leakage current, up to a
high reverse voltage. The metals with high work functions have been reported by many
researchers to be used as a Schottky contact for GaN1"". In this study, the Cu metal deposited
by the sputter system was used as a Schottky contact to fabricate a Cu/n-GaN Schottky diode.
The Schottky barrier height of the copper metal for n-GaN grown on the A1203 substrate was
investigated by using the I-V and C-V measurements.
523
Mat. Res. Soc. Symp. Proc. Vol. 572 ®1999 Materials Research Society
EXPERIMENT
The samples used for this study were grown by the low pressure metal-organic vapor
phase epitaxy (MOVPE) system. Trimethylgallium (TMGa) and NH3 were used as source
materials, and SiH4 diluted in H2 was used as the n-type dopant source. The sapphire substrate
was first heated at 1100°C for 20min in the stream of H2. A 25nm thick GaN layer was
deposited as the buffer layer at 525°C. Then, the substrate was heated up to 1050°C and a 2um
thick Si doped GaN layer was grown on the GaN buffer layer. The carrier concentration of the
Si doped GaN film was 5xl017cm"3 measured by the Hall measurement. The GaN samples
were first cleaned with organic solvents, followed by etching in a HC1:H20 solution, and then
were loaded in an E-gun evaporator equipped with the cryo pump. Ti (50nm) and Al (lOOnm)
were evaporated using the conventional lift-off technique. After lift-off process, the sample
was annealed at 550°C for 5min in nitrogen ambient in order to obtain ohmic characteristics.
Then, Cu (lOOnm) was deposited on the GaN film by sputtering and the pattern of the Cu
contact was formed by using the wet etching with a solution of 10% HN03. Prior to the
sputtering process, the sample was cleaned in foiling aqua regia for 5 min and dipped in
HC1:H20 for lmin. The area of the Schottky contact was 9.5xl0"5 cm2. The current-voltage (IV) and capacitance-voltage (C-V) characteristics were measured with HP4145 semiconductor
parameter analyzer and Keithly 590 C-V analyzer.
RESULT AND DISCUSSION
The typical I-V characteristics of the Cu/GaN Schottky diode measured at room
temperature are shown in Fig.l. An increase of the forward current was observed above the
applied voltage of 0.4V. The I-V characteristic function of the Schottky diode is described by
the following equations[ll]:
J=J0(exp(qV/nkT)-l)
J0=A**T2exp(-q(|)b„AT)
(1)
(2)
where J0 is the saturation current density, n the ideality factor, A** the effective Richardson
constant, k the Boltzmann's constant and <|>bn0 the measured barrier height. The value of <&,„,
can be deduced from the I-V measurement as the effective Richardson constant is decided.
The theoretical value of A** is 26 A/cm"2K"2 [1]. From the forward bias log I-V plot as shown
in Fig.l, the values of the Schottky barrier height and the ideality factor are 1.13eV and 1.2,
respectively.
To eliminate the effect of the series resistance of the diodes, the Norde's method was
used to calculate the Schottky barrier height[12]. From this method, the Schottky barrier
height of the Cu/GaN can be calculated by using a plot of the following function:
524
0.0
0.5
1.0
1.5
2.0
2.5
-15
3.0
-10
-5
Reverse voltage(V)
Forward voltage (V)
(b)
(a)
Fig. 1 The I-V characteristic plot of the Cu/GaN Schottky diode at room
temperature (a) forward biases (b) reverse biases.
V kT
2~~q~
I
)• •
AA**T2
..(3)
From the plot of the F(V) vs. V suggested in Fig.2, the minimum point F(VJof F(V) can be
find by setting dF/dV=0. And then the corresponding voltage Vm can be obtained. Therefore,
the Schottky barrier height can be calculated by the following equation[13]:
2-n kT
..(4)
n
q '
where o)B is Schottky barrier height, Vm is correspond voltage of the minimum point of F(V)
and n is the ideal factor. From Eq.(4), the calculated Schottky barrier high of the Cu/GaN is
h=nvm)-t 1
1
)Va
1.15eV.
Besides the I-V methods, the Schottky barrier height of Cu/GaN can also be measured by
using the C-V characteristic method. Fig.(3) shows the C-V characteristic curve and the 1/C2V plot as a function of DC bias changing from -5V to 0V in steps of 0.02V, with a small
signal of 100kHz. The capacitance increases gently with the increase of the applied voltage,
which is shown in Fig.3(a). That is because of the shrinkage of the depletion region of the
Cu/GaN Schottky contact. The built in potential of this Cu/GaN Schottky barrier is obtained
from the l/C2-voltage plot via Vbi=V„+kT/q, where V0 is the voltage intercept. V0 is
determined to be 1.37eV which is fitted and marked in Fig.3(b). Neglecting image force
lowering, the barrier height, Obn(C-V) is given by
®ta (C-V) = qVbi + 0VE,)
(5)
Where (Ec-Ef) is the energy difference between the conduction band and the Fermi level and
525
given by (Ec-Ef) = (kT/q)]a(NJN^). Nc is the effective state density in the conduction band and
Nd is the donor concentration of the film. The theoretical value of Nc is defined as
0.6
0.7
0.8
0.9
Forward voltage(V)
Fig.2 The norde plot of the Cu/GaN Schottky Diode
2(27im1*kT/h2)3'2. The typical value of the effective mass, m/, is equal to 0.22m,,. As
mentioned above, the doping concentration of GaN measured by the Hall measurement is
5xl0,7cm"3. Therefore, the values of (E^Ef) and the barrier height calculated from Eq.5 are
0.04 eV and 1.41eV, respectively.
The value of the barrier height measured from the C-V method is larger than that
measured by the I-V method. This is because the transport mechanism of these diodes is not
purely due to the thermionic emission. For these diodes, the barrier height of <Dbno is voltage or
electric field sensitive and Obn is not.10' The two barrier heights are related but are not
identical quantities. Because the ideality factor of the I-V characteristic is not 1 for the Cu/nGaN Schottky diode, the Schottky barrier height of the copper metal on n-GaN must be
correlated with a more fundamental barrier height, Obf, at zero electric field by the following
equation[14]:
«brnOtao-Cn-iykT/qInCN./NJ
(6)
where Nc is the effective conduction-band density of the states. When the junction is at zero
electric field, i.e., the flat band condition, there should be no tunnelling or image force
lowering. The corrected barrier height calculated from Eq.(6) is 1.35eV. The quantitative <Dbf
is, therefore, more appropriate value to compare with <3>bn. The corrected barrier height, 4>bf,
now can be considered as the barrier height of a Schottky diode with a transport mechanism
of thermionic emission and without any image force lowering effect. [10]
The results of the barrier height calculated from above methods are summarized in Table
I. The Schottky barrier height of the Cu/GaN diodes measured from I-V and C-V methods are
different from that predicted by the Schottky-Mott model, which indicates that the barrier
height is equal to the difference between the metal work function of Cu and the electron
526
affinity of GaN (x=4.2eV). Such a high Schottky barrier height of the Cu/GaN contact
obviously can not be explained by the Schottky-Mott model and it may result from the
Table I. Schottky barrier height of Cu/GaN diode
Ideality n
Cu/GaN
Schottky diode 1.2
<Db(Norde Plot)
rEq.(4)l
4UC-V)
rEq.(5)l
Obf[Eq.(6)]
rEq.(l)l
1.13 eV
1.15 eV
1.41 eV
1.35 eV
26
6.00E+015
24
5.00E+015
22
JS
20
"u.
1.8
I
4.00E+015
3.00E+015
1.6
2.00E+015
1.4
1.00E+015
U
0.00E+000
5
-4
-3
-2
-1
0
-5-4-3-2-1012
Applied voltage(V)
/applied vdtageM
(a)
(b)
Fig. 3 (a) C-V characteristics and (b) 1/C2 plot of Cu/GaN Schottky diode.
interface reaction or the Fermi level pinning effect which is also found in the GaAs-based or
InP-based materials. The details of the Schottky contact Cu on GaN, such as the reaction
between Cu and GaN, are currenty extensively studied and will be published later.
In conclusion, the Cu/GaN Schottky diode with a high barrier height has been achieved
and investigated. The Cu metal was sputter-deposited on n-GaN. The barrier height of the
Cu/GaN Schottky diodes measured by the I-V and C-V method are 1.15 and 1.41eV,
respectively. The contact mechanism of Cu/GaN can not be completely explained by the
Schottky -Mott model and the mechanism is suggested to be the surface fermi-level pinning
effect.
The authors wish to acknowledge the support a portion for this work by the National
Sciences Council (NSC-88-2215-E009-015) and Epistar Corporation .
Reference
1) P. Hacke, T. Detchprohm, K. Hiramatsu and N. Sawaki: Appl. Phys. Lett. 63 (1993)
2676.
2) S. C. Binari, H. B. Dietrich, G. Keiner, L. B. Rowland, K. Doverspike and D. K. Gaskill:
527
3)
Electron. Lett. 30 (1994) 909.
M. R. H. Khan, T. Detchprohm, P. Hacke, K. Hiramatasu and N. Sawaki: J. Phys. D 28
4)
(1995) 1169.
J. G. Guo, M. S. Feng, R. J. Guo, F. M. Pan and C. Y. Chang: Appl. Phys. Lett. 67 (1995)
5)
2657.
L. Wang, M. Y. Nathan, T. -H. Lim, M. A. Khan and Q. Chen: Appl. Phys. Lett. 68
6)
7)
8)
9)
(1996) 1267.
A. T. Ping, A. C. Scmitz, M. A. Khan and I. Adesida: Electron. Lett. 32 (1996) 68.
S. N. Mohammad, Z. F. Fan, A. E. Botchkarev, W. Kim, O. Aktas, A. Salvador and H.
Morkoc: Electron. Lett. 32 (1996) 598.
K. Suzue, S. N. Mohammad, Z. F. Fan, W. Kim, O. Akats, A. E. Botchkarev and H.
Morkoc: J. Appl. Phys. 80 (1996) 4467.
E. V. Kalinina, N. I. Kuznetsou, V. A. Dmitriev, K. G. Irvine, and C. H. Carter Jr.: J.
Electron. Mater. 25 (1996) 831.
10) Q. Z. Liu, L. S. Yu, S. S. Lau, J. M. Redwing, N. R. Perkins and T. F. Kuoch: Appl. Phys.
Lett. 70 (1997) 1275.
11) S. M. Sze: Physics of Semiconductor Devices (John Wiley & Sons, New York, 1981) 2nd
ed., Chap. 5.
12) H. Norde: J. Appl. Phys. 50 (1979) 5052.
13) D. K. Schroder: Semiconductor Material And Device Characterization (John Wiley &
Sons, New York, 1990), 153-154.
14) L. F. Wagner, R. W Young, and A. Sugerman: IEEE Electron. Device Lett. ED-4, (1983)
320.
528
HIB- Nitride Semiconductors for High Temperature Electronic Applications
X. BAI, D. M. HILL and M. E. KORDESCH,
Department of Physics and Astronomy, kordesch@helios.phy.ohiou.edu,
Ohio University, Athens OH 45701
ABSTRACT
Thin films of ScN and YN were grown on silicon, quartz and sapphire using metal
evaporation and an RF atomic nitrogen source. YN decomposes on contact with water vapor,
and only A1N capped films could be stabilized. ScN is stable in air and water, and thin films of
this material deposited at temperatures between 300 and 900 °C show a substrate-dependent film
texture. Typical growth rates were -0.1 nm/second with a 300W N discharge at about 0.1
mTorr Nitrogen pressure. Structural characterization by x-ray diffraction, infrared transmission
spectroscopy and Hall effect measurements on n-type ScN and the fabrication of p-n junctions
of n- type ScN with silicon are presented.
INTRODUCTION
The IIIB metals Scandium, Yttrium and Lanthanum are less commonly used for wide
bandgap nitride semiconductors compared to the ITIA metals Aluminum, Gallium and Indium.
The HIB metals have one d electron rather than one p electron in the outer shell. The IIIBnitrides have bandgaps in the 2-2.4 eV range, crystallize in the rock salt structure, and melt
above 2600 °C. The films are yellow to deep red in color [1]. Several groups have investigated
ScN, both for its mechanical properties in analogy to the well-known coating material TiN [2,3],
and both experimental and theoretical investigations of its electronic properties [4-6]. The cubic
ScN lattice is a good match for the IIIA nitrides; because of the high melting temperatures of
both Sc and ScN this material might ultimately be used to replace InN in IHA.-V semiconductor
alloys for high temperature applications.
We report the results of vacuum evaporation of scandium metal in an atomic N ambient
for producing thin films of ScN, and diode structures made from n-type ScN and p-type Si. YN
nitride films decomposed in humid air in a matter of hours to days.
EXPERIMENT
The vacuum system is based on a stainless steel 6-way cross with 150 mm outside
diameter copper sealed flanges. The vertical ports are used for sample insertion (top), the
evaporator feedthroughs (bottom). The four horizontal ports are used for the SVTA Inc. radio
frequency atomic nitrogen source, viewports, shutter, main chamber pump and quartz crystal
thin film thickness monitor. The system has a 100 1/sec turbo pump roughing system, and a 1500
1/sec cryopump for use during deposition, because the atomic nitrogen caused corrosion and oil
decomposition problems with our conventional pumping system. Base pressure in the unbaked
system was in the 10"7 Torr range in several hours.
The substrates were clamped to a pyrolytic boron nitride-coated graphite heater, the
substrate temperature was monitored with an optical pyrometer. A four-conductor (6mm
diameter copper conductors) feedthrough was used to support and make contact to two tungsten
529
Mat. Res. Soc. Symp. Proc. Vol. 572 ° 1999 Materials Research Society
boats (6mm wide tungsten foil 0.25mm thick with a central dimple) for evaporation of two
metals at one time. Typical deposition currents were in the 100-125 Ampere range. The
nitrogen source was operated on pure nitrogen at about 2 xlO~4Torr. Sc deposition rates were
controlled manually in the 0.05-0.3 nm/sec range.
Substrates were washed in alcohol and preheated in the chamber to degas the substrates
for about 1 hr. Small (2-3 mm diameter) chips of Sc or Y metal (99.99%) were placed in the
boats for evaporation. Hall effect measurements were made with a commercial Hall effect set-up
(Keithly Instruments).
A schematic of the experimental set-up is shown in the figure 1.
Chamber
Boats
atomic N
Substrate
EvaporotDn
Shutter J2I7
View from top
Evaporators
Figure 1. Schematic of experimental set-up.
RESULTS
Several dozens films were grown on various substrates: quartz microscope slides (Quartz
Scientific, Inc. Fairport Harbor OH) and silicon (100). Typically, a low metal evaporation rate
was used, because higher rates were very difficult to control, and resulted in Sc- rich ScN or
metal capped films. Films where the deposition resulted in a Sc metal overlayers were usually
discarded immediately, and no further ScN growth was attempted. ScN and YN "skin" or
"crust" formed on the metal in the evaporation boat. Well-formed bulk cubic crystals were
observed in SEM observation of the evaporation remnants. While still in the chamber, YN films
showed a deep red color, ScN films were yellow to greenish-yellow. Upon Removal from the
vacuum system, the YN films became transparent in a matter of hours. YN capped with A1N
was more stable, but decomposed from the edges of the film inward with the same eventual
outcome.
Sc metal films could be evaporated in this system, which showed only Sc metal lines in
x-ray diffraction, with no detectable Sc-oxide.
530
3
40.00
20.00
60.00
theta-2*theta
100.00
80.00
Figure 2: Typical 9-26 x-ray diffraction curves for ScN on various substrates. Nominal
deposition parameters: Curves (1) and (5), Silicon and quartz substrates, respectively. Curve (2)
240 nm ScN, deposited at 650 °C, 0.05 nm/sec, (3) 400 nm ScN, 700 °C 0.2 nm/sec, (4) 170 nm
ScN, 700 °C, 0.3 -0.4nm/sec. (6) 220 nm ScN on quartz, 650 °C, 0.03 nm/sec.
100
£
80
c
a
8000 7000 6000 5000 4000 3000 2000 1000
0
W avenum ber (cm -1)
Figure 3. Infrared Transmission of a ScN/quartz film. The LO phonon mode is observed at
1104 cm"1. The sinusoidal oscillations 2000-8000 cm"1 are interference fringes.
531
Infrared transmission measurements were made on these films, originally to determine
the thickness, however the refractive index of scandium nitride is not known to sufficient
accuracy for thickness measurements using interference fringes. The thermal conductivity of
ScN is reportedly in the range for A1N, and the HIA nitrides, which are quite high. The phonon
modes for these materials have recently been reported [7], and the ScN LO mode value observed
is consistent with the magnitude of the IUA-nitrides.
Several types of metal contacts were sputter-deposited onto ScN thin films for Hall
measurements and to form simple schottky-diode contacts. The Hall data are reported in Table 1
below. No successful schottky diodes were made.
Diode structures were formed by deposition of ScN on conductive p-type Si(100) (Boron
doped, 0.2-1.0 -cm, from Virginia Semiconductor). Indium metal contacts were pressed onto the
surface of the ScN film manually, the silicon contact was made by pressing In onto a freshly
scrapped area of the film where silicon was exposed. Gold contacts were also used. The I-V
curves for indium and Au contacted diodes are shown in figure 4.
-20-
15
10
MM»
•5
-10
-5
0
5
10
15
V(V)
V(V)
Figure 4: LEFT: 100 nm ScN on p-Si(100), gold contacts..
RIGHT: 130 nm ScN on p-type Si(100), indium contacts.
Hall effect measurements were made on several of the ScN films. Films that were
sufficiently conductive for this measurement were n type. It was suggested in ref. [4] that Mg
might be suitable for p-doping of ScN. Magnesium metal was co-evaporated in the second boat
at a rate corresponding to about 1% Mg. The Mg content was checked with x-ray fluorescence
and is in the 1% range. The Mg-doped ScN sample does show a positive Hall coeffcient. ScN
p-n junctions are planned.
The data are summarized in the table below.
532
Table 1. Hall Effect Data
Sample
(substrate)
Thickness
(nm)
Contact
980727A
(quartz *)
130
980728B
(quartz *)
990129
(quartz **)
before anneal
990129
(quartz **)
annealed
Ti
Resistivity Hall
(Q-cm)
Coefficient
(cm3/gs)
6.39E-2
-4.66E0
Carrier
Density
(cm"3)
1.34E18
180
Ti
6.03E-1
-3.57E0
1.75E18
500
Ni/Au
2.97E-2
2.47E1
2.50E17
500
Ni/Au
1.18E-1
2.53E1
2.47E17
Anneal condition: N2 , temperature: 447°C, duration: 5min.
DISCUSSION
ScN thin films grown on silicon and quartz show similar orientation as observed for
reactively sputtered ScN on MgO in ref.[2,3]. On silicon we have grown both (100) and (111)
textured films, on quartz, the (100) orientation predominates. The as-grown films with good
conductivity are n-type, suggesting oxygen as the n-dopant, in analogy to GaN. In this case, we
expect that the source may be water that desorbs from the system walls during the deposition. In
several cases, insulating films were obtained, although similar in color, the bandgap
measurements (in the 2.5-3 eV range) on these films (on quartz) indicates some Sc-oxide in the
film.
Initial attempts show that Mg may indeed be a p-type dopant for ScN.
CONCLUSIONS
ScN thin films were grown by thermal Sc evaporation in an atomic nitrogen environment.
Films were grown on silicon and quartz. Conductive n-ScN/p-Si(100) diodes were tested. YN
films were unstable in humid air, and were not investigated further. The results for ScN thin
films suggest that diodes and possibly other electronic devices made from ScN junctions may
survive operation at elevated temperatures. Although not investigated further in this study, the
transition from ScN to Sc-metal is very simple in this growth method, so that Sc contacts for
high temperature use should also be considered.
533
ACKNOWLEDGMENTS
We would like to thank John Dismukes for helpful discussions. This work supported by BMDO
through ONR Grants N00014-95-1-0298, and N00014-96-1-0782, -1183.
REFERENCES
1. Gmelin's Handbuch der anorganischen Chemie, 8 Aufl., Deutsche Chem. Gesell., R.J. Meyer,
ed., Leipzig-Berlin, Verlag Chemie, 1924-. Volume SE C2, Nitride, Vergleichende Angaben.,pg
146ff.
2. D. Gall, I. Petrov, N. Hellgren, L. Hultman J.E. Sundgren and J.E. Greene, J. Appl. Phys. 84,
6034 (1998).
3. D. Gall, I. Petrov, L.D. Madsen, J.E. Sundgren and J.E. Greene, J. Vac. Sei. Technol. A16,
2411 (1998).
4. J.P. Dismukes and T.D. Moustakas, Proceedings of the JU-V Nitrieds Materials and
Processing Symposium, The Electrochemical Society, Pennington NJ, 1996.
5. A.G. Petukhov, W.R.L. Lambrecht and B. Segall, Phys. Rev. B. 53, 4324 (1996).
6. K. Kunze and J.F. Harrison, J. Amer. Chem. Soc. 112, 3812 (1990).
7. H.Harima, T.Inoue, S. Nakashima, H. Okumura, Y. Ishida, S. Yoshida, T. Koizumi, H.Grille,
and F. Bechstedt, Appl. Phys. Lett. 74, 191 (1999).
534
PHOTO-ASSISTED RIE of GaN in BChfCyS2
N. Medelci, A. Tempez, I. Berishev, D. Starikov, and A Bensaoula
Nitride Materials and Devices Laboratory, SVEC-University of Houston, Houston, TX
ABSTRACT
Gallium nitride (GaN) has been under intense investigation due to its unique qualities
(wide band gap, chemical and temperature stability) for optoelectronic and high
temperature/high power applications. To this end, reactive ion etching (RIE) experiments
were performed on GaN thin films using BCyCtyAr. These resulted in etch rates of 1400
Ä/min at -400 V dc bias1. However, rough etched surfaces, nitrogen surface depletion and
high chlorine content were observed. In order to remedy these shortcomings, a photo-assisted
RIE process using a filtered Xe lamp beam was developed, resulting in higher etch rates but
again in nitrogen depleted surfaces2. Preliminary results on using nitrogen instead of argon in
the process chemistry show a big improvement in photo-asssisted etch rates (50%) and Ga/N
ratio (0.78 versus 1.25). In this paper, the effects of epilayer doping, dc bias, nitrogen flow
rate and photo-irradiation flux on GaN etch rates, surface morphology and composition are
presented. Finally, preliminary results on the use of a KrF excimer laser beam in the GaN
photo-assisted RIE process are presented.
INTRODUCTION
Wide band gap m-V nitrides are emerging as the materials of choice for high
temperature and high power electronics and blue-UV emitters and detectors. Synthesis of
gallium nitride (GaN) is being developed for the fabrication of light emitting diodes, laser
diodes and flat panel displays. These characteristics, resulting from their strong chemical
strength, become a drawback for their processing and thus for their industrial development.
Conventional reactive ion etching (RIE) using halogen-based chemistries achieved
relatively low etch rates3"6. High density plasma techniques such as electron cyclotron
resonance (ECR), inductively coupled plasma (ICP) and magnetron RIE lead to higher etch
rates7"13. However, these ion-assisted methods cannot avoid ion bombardment damage and
surface roughening at high RF powers. In addition, nitrogen depletion is usually associated
with high ion energies. Vertical sidewalk have been realized with chemically assisted ion
beam etching (CAIBE) using HC114. Alternative dry etching methods for low lattice damage
are low energy electron enhanced etching (LE4)15 and photo-assisted etching. The last
technique has already been demonstrated for GaAs and Si
. Photo-assisted etching of GaN
has also been tested using an ArF excimer laser in a HC1 ambient18. However, the preliminary
etch rate was low (optimized etch rates have not yet been reported).
Encouraging results for RIE of GaN have been reported using BCl3/Cl2/Ar, with etch
rates reaching up to 1,200 A/min at 200 W RF power1. Nevertheless, nitrogen depletion,
which increases with increasing dc self-bias, was observed after etching. Moreover, higher
etch rates are desirable. Therefore, a photo-assisted RIE process using the same chemistry was
developed, resulting in higher etch rates but still nitrogen depleted surfaces2. Substitution of
Ar with N2 in the gas chemistry was tested with the aim of reducing the surface nitrogen
depletion. Encouraging results for both etch rates and nitrogen surface content were obtained.
In this paper, photo-assisted RIE of GaN in BC13/C12/N2 using both a filtered Xe lamp and a
535
Mat. Res. Soc. Symp. Proc. Vol. 572 ® 1999 Materials Research Society
KrF excimer laser are evaluated and compared to standard RIE. Doping, nitrogen flow rate
and lamp optical power density effects on etch rates, surface morphology and composition are
investigated.
EXPERIMENTAL DETAILS
The gallium nitride films were grown on sapphire wafers at 750 °C by molecular beam
epitaxy19. The GaN samples were patterned using photoresist (Shipley 1813). For a given set
of etching conditions, two unassisted samples (one blank and one patterned) and two lightexposed ones were processed in the same run Etching was performed in an Oxford Plasma
Technology 80 up RIE reactor. The base pressure of the reactor was 2 x 10"6 Torr. A xenon
lamp with an emission spectrum ranging from 200 to 900 nm was used. The light beam
traveled through a glass window (UV filter), an IR filter, reflected on a mirror and impinged
at normal incidence onto the samples with a spot size of 2 cm in diameter through a quartz
window. The maximum light power density was 38 mW/cm2. A KrF excimer laser (248 nm)
was also used as a light source in the photo-assisted RIE process.
Gallium nitride thin films were etched in 10 seem BCI3/IO seem cyN2 at 30 mTorr.
The RF power was varied from 100 to 400 W, which corresponds to a dc self-bias range of 200 to -480 V. Etch rates were measured on the patterned samples using a Tencor 250-alpha
step profilometer. The GaN surface morphology and etch profiles were checked with a JEOL
JSM-5410 scanning electron microscope (SEM). XPS was employed to investigate any
resulting surface stoichiometry modification from etching. For this purpose, a Perkin-Elmer
PHI ESCA system was used. Mg Ka radiation was utilized as the source of excitation The
energy scale was calibrated from the C Is line at 285 eV from adventitious carbon on the
surface. The base pressure during the analysis was in the low 10"10 Torr range. The surface
composition was determined from integration of B Is, C Is, N Is, O 2s, Ga 2p3/2 and Cl 2p
peak area, using the sensitivity factors 0.171, 0.314, 0.499, 0.733, 2.751 and 0.954 for each
element, respectively. In the case of GaN grown on an insulating substrate, surface charging
build-up was compensated by a flow of low energy electrons. The detection angle was 45°
with respect to the sample surface.
RESULTS AND DISCUSSION
Etch Rate Results
GaN thin films were etched in BClj/Cl2/N2 at 100 to 400 W RF powers (-220 to -480
V dc self-biases). The standard RIE rate increased from 400 A/min at -220 V up to 1330
A/min at -470 V (Figure 1). The etch rate at a given dc self-bias under filtered Xe lamp light
irradiation was higher when compared to the standard RIE rate. The photo-assisted etch rate
also increased as a function of dc self-bias. The highest etch rate, 3240 A/min was reached at
-470 V dc self-bias. The highest standard RIE rate obtained at -470 V could be obtained at 250 V with the use of photo-radiation. The photo-assisted to standard rate ratios are in the 1.94.4 range with the highest photo-enhancement observed at 200 W RF power (-300 V dc selfbias).
The effects of nitrogen flow rate, Xe lamp beam power density and GaN doping level
on etch rates have also been investigated for both the standard and photo-assisted RIE
processes. Both n and p-type GaN etch rates decrease with increasing carrier concentration,
536
3600 1
3000
1
-
2400
S.
1800 t-
1
1
I
1
T
'"
A
1
i
^
c
1
• StandardRIE
—*— KrF Laser-Assisted RIE
-*• • •Xe
- LamffcAssisted RIE
J
600
l""'— 1
...
1200
.
'
1
250
200
.
1
.___.
1
300
1—
350
1
l
L.
i
450
400
500
-de Self-Bias (V)
Figure 1. Photo-assisted (both filtered Xe lamp and KrF excimer laser) and standard etch rates
and ratios of GaN as a function of dc self-bias voltage. The GaN films were etched in 10 seem
CW10 seem BCI3/IO seem N2 at 30 mTorr and 38 mW/cm2 optical power density.
1
4JW
1
1300
rHypeGBN
—• —StattenJRE
Rxto^ssfctedRE
T
r
'
1
—r
1
"
1
1300
ptypeäN
-• -SsrdsrdHE
-
1000
-.
s
•
. s
S»
>
a1
1
10
K30
-
n
r
2
'
«
3
•!
4
5
OatierOorcertaicnflo" ori8)
Carrier QiuBMiuifKfcnf)
Figure 2. Photo-assisted and standard etch rates of n and p-type GaN as a function of carrier
concentration. The GaN films were etched in 10 seem Cl2/10 seem BCI3/IO seem N2 at 30
mTorr and 38 mW/cm2 optical power density.
537
particularly under filtered Xe lamp-assisted RIE conditions, even though they are higher than
the standard ones (Figure 2). Optimum N2 flow rate was found to be in the 10-15 seem range.
As the Xe lamp beam power density increases, we notice a sharp increase in the etch rate at
30 mW/cm2.
Etching under the same conditions but using a KrF excimer laser beam (248 nm) has
also been performed. As shown in Figure 1, the laser-assisted etch rates (maximum: 1700
A/min at -470 V) and photo-assisted to standard rate ratios (maximum: 1.7 at -220 V) were
lower than those obtained with the filtered Xe lamp.
Surface Morphology and Composition
The GaN surface morphology as examined by SEM exhibits 'etch pits' at high dc bias
(-470 V). We notice however that, at all dc biases, the etched surfaces under illumination arc
smoother than the standard etched ones. Moreover, we observe that the photo-assisted etched
surface at -300 V is smoother than the as-grown surface.
Using XPS analysis, the surface composition change due to etching was investigated.
In the standard RIE process, the surface Ga/N versus dc self-bias does not show a defined
trend (Figure 3). At -300 V, the surface is more depleted in N than the starting GaN surface.
For -200V and -480 V, the surface Ga/N ratio is similar to that of the as-deposited surface
(1.55). However, after etching at -380V, the surface is even closer to stoichiometry than the
starting GaN. When the GaN sample is illuminated during the etch process with the filtered
Xe lamp beam, the surface Ga/N ratio decreases as the dc self-bias increases, reaching 1.37. at
3.5
Ä-
3.0
* Standard RIE
- -T- - • Rioto-Assisted RIE
'i
.. *'
z
o 9
8
2.
2.0
-2 I.
o
3
9
1.5 -
1.0
500
600
-dc Self-Bias (V)
Figure 3. Surface XPS Ga to N ratios and Cl atomic concentrations as a function of dc bias.
538
-480V When the sample is ffluminated with the KrF excimer laser beam, the variation of the
Ga/N ratio with the dc self-bias follows the fluctuations of the rat» obtained with standard
RIE
The standard RIE process always results in a surfece containing 2% or more of
chlorine (Figure 3). The lowest chlorine residual content is obtained at -220V dc self-bias and
the highest content at -300V. The photo-assisted etching process (for both KrF excimer laser
and filtered Xenon lamp) always leads to a surface as clean or cleaner than the standard RIE
process In the case of the filtered Xe lamp, the surface residual chlorine content decreases as
the dc self-bias increases. At -300V dc self-bias, the chlorine content is below 0.8%. When
using the KrF excimer laser, the photo-assisted process is only better man standard RIE for the
highest dc self-biases (-420 and -480V in our study).
It is difficult to explain how irradiation influences and enhances the etching process
without knowledge of the surface composition during etch and real time monitoring of the
etch products. It is believed that irradiation may excite the adsorbed species (reactants and
products) and thus enhance product desorption. This would also lead to higher coverage of the
material surface with reactive species. The availability of N radicals in the plasma and the
photo-excitation of the surface resident species might enhance surface diffusion and
accelerate the formation of more volatile end products. In addition, photo-excited surface Ga
atoms with unsatisfied bonds react more readily with N atoms. Such process should result in
higher surface turn-over and etch rates. Finally, photo-generated carriers take part m the
photochemistry, as shown by the decrease in etch rate for the p-type material as doping level
increases.
CONCLUSIONS
Photo-assisted reactive ion etching of GaN using both a filtered Xe lamp and a KrF
excimer laser in BCVCU/Nj was investigated. The standard and photo-assisted RIE etch rates
always increase with increasing dc self-bias. Higher etch rates are observed when the
materials are exposed to both a filtered Xe lamp and a KrF excimer laser beam, reaching 3240
and 1700 Ä/min at -470 V dc bias, respectively. Higher photo-assisted to standard etch ratios
were observed for the filtered Xe lamp case, with a maximum (4.4) observed at -300V dc
bias Use of illumination lowers surface chlorine atomic concentration. Etch rates decrease as
GaN doping level increases for both p and n-rype materials. Hence, the photo-assisted RIE
process of GaN in BCI3/CI2/N2 results in higher etch rates, and smoother and cleaner etched
surfaces when compared to the standard process performed using the same conditions.
ACKNOWLEDGMENTS
This work was supported by funds from a NASA cooperative agreement #NCC8-127
to SVEC a Texas Advanced Research Program Grant # 1-1-27764, and a Texas Advanced
Technology Program Grant # 1-1-32061. This material is also based upon work supported by
the U S Civilian Research and Development foundation under Award No. REI-247. This
work made use of TCSUH/MRSEC Shared facilities supported by the State of Texas through
the Texas Center for Superconductivity at the University of Houston and by the National
Science Foundation under Award Number DMR-9632667. The authors would also like to
thank Darren Tucker for helping with the RIE system and Susan Street for performing the
SEM analysis.
539
REFERENCES
1
N. Medelci, A. Tempez, E. Kim, N. Badi, D. Starikov, I. Berichev, and A. Bensaoula, Mat
Res. Soc. Symp. Proc. 512,285 (1998).
2
A. Tempez, N. Medelci, N. Badi, D. Starikov, I. Berishev, and A Bensaoula,
"Photoenhanced reactive ion etching of III-V nitrides in BCI/CI/Ar/N} plasmas", to be
published in J. Vac. Sei. and Technol. A (1999).
I. Adesida, A Mahajan, E. Andideh, M Asif Khan, D. T. Olsen, and J. N. Kuznia, Appl.
Phys. Lett. 63(20), 2777 (1993).
4
Heon Lee, David B. Oberman and James S. Harris Jr., Appl. Phys. Lett. 67(12), 1754 (1995).
5
M. E. Lin, Z. F. Fan, Z. Ma, L. H Allen, and H. Morkoc, Appl. Phys. Lett. 64(7), 887 (1994).
6
A. T. Ping, I. Adesida, M Asif Khan and J. N. Kuznia, Electronics Letters, Vol. 30, No. 22,
1895(1994).
7
C. B. Vartuli, S. J. Pearton, J. W. Lee, J. D. McKenzie, C.R. Abernathy, and R. J. Shul, J.
Vac. Sei. and Technol. A 15(3), 638 (1997).
8
G. F. McLane, T. Monahan, D. W. Eckart, S. J. Pearton, and C. R. Abernathy, J. Vac. Sei.
and Technol. A 14(3), 1046 (1996).
9
C. B. Vartuli, S. J. Pearton, C.R. Abernathy, and R J. Shul, A. J. Howard, S. P. Kilcoyne, J.
E. Parmeter and M. Hagerott-Crawford, J. Vac. Sei. and Technol. A 14(3), 1011 (1996).
10
G. F. McLane, L. Casas, S. J. Pearton, and C. R Abernathy, Appl. Phys. Lett. 66(24), 3328
(1995).
n
G. F. McLane, L. Casas, R T. Lareau, D. W. Eckart, C. B. Vartuli, S. J. Pearton, and C. R
Abernathy, J. Vac. and Sei. Technol. A 13(3), 724 (1995).
12
Hyun Cho, C. B. Vartuli, S. M Donovan, C.R Abernathy, S. J. Pearton, R J. Shul, C.
Constantine, J. Vac. Sei. and Technol. A 16(3), 1631 (1998).
13
R. J. Shul, C. G. Willinson, M. M. Bridges, J. Han, J. W. Lee, S. J. Pearton, C.R. Abernathy,
J. D. McKenzie, S. M Donovan, L. Zhang, and L. F. Lester, J. Vac. Sei. and Technol. A 16(3),
1621 (1998).
14
A. T. Ping, I. Adesida, M. Asif Khan, Appl. Phys. Lett. 67(9), 1250 (1995).
15
H. P. Gillis, D. A Choutov, K. P. Martin, M D. Bremser, and R. F. Davis, Journal of
Electronics Materials Vol. 26 No. 3 (1997).
16
F.A. Houle, Phys. Rev. B 19,10120 (1989).
17
S. Takatani, S. Yamamoto, H. Takawaza, and K. Mochiji, J. Vac. Sei. and Technol. B 13(6)
2340 (1995).
18
R.T. Leonard, S. M. Bedair, Appl. Phys. Lett. 68(6), 794 (1996).
19
1. Berishev, E. Kim, and A. Bensaoula, J. Vac. Sei. and Technol. A 16(5), 2791 (1998).
540
CORRELATION OF DRAIN CURRENT PULSED RESPONSE WITH MICROWAVE
POWER OUTPUT IN AlGaN/GaN HEMTs
S. C. Binari, K. Ikossi-Anastasiou, W. Kruppa, H. B. Dietrich, G. Keiner, R. L. Henry,
D. D. Koleske, and A. E. Wickenden
Naval Research Laboratory, Washington, D. C, 20375
ABSTRACT
The drain-current response to short (<lus) gate pulses has been measured for a series of
GaN HEMT wafers that have similar dc and small-signal characteristics. This response has been
found to correlate well with the measured microwave power output. For example, for devices
where the pulsed drain current is greater than 70% of the dc value, output power densities of up
to 2.3 W/mm are attained. This is in contrast with 0.5 W/mm measured for devices with low
pulse response (less than 20% of the dc value). These results, which can be explained by the
presence of traps in the device structure, provide a convenient test which is predictive of power
performance.
INTRODUCTION
GaN-based microwave power transistors have set the state-of-the art for output power
density [1] and have the potential to replace GaAs-based transistors for a number of high-power
applications. The rapid advances made in GaN-based devices stem from the electronic properties
of the GaN-based material system. However, due to the relative immaturity of the materials
growth and device processing, the measured microwave power output is frequently limited by
trapping effects [2]. Trapping effects have been shown to play a role in the operation of many
GaN-based FETs. In addition to the limitations they impose on microwave power output, traps
have been shown to result in drain-current transients subsequent to a gate or drain voltage pulse
[2], drain current collapse after the application of a high drain bias [3], and transconductance and
output resistance frequency dispersion [4]. One of the objectives of this work is to develop a test
or measurement technique that can serve as a useful predictor of device microwave power
performance. The earlier this test can be performed in the device processing sequence, the more
useful it will be.
During the development of GaAs field-effect transistors, significant attention was
directed toward the understanding and minimization of trapping effects. In the GaAs technology,
the drain current response to gate- and drain-voltage pulses was extensively utilized [5, 6] as a
means of investigating trapping effects. It was also shown that pulsed current-voltage
characteristics can serve as a indicator of microwave power performance [7,8]. In the work
presented here, the measured microwave power output in GaN-based HEMTs has been found to
correlate well with the drain-current pulsed response.
MATERIALS GROWTH AND DEVICE FABRICATION
The device cross section is shown in Fig. 1. All HEMTs were fabricated with 4 to 6 um
source-drain spacings and a nominal gate length of 1 um. The gate consists of two gate fingers
with a total gate width of 150 urn. The ohmic contacts were Ti/Al/Ni/Au and had a contact
resistance in the range of 1-2 ö-mm. Pt/Au was used for the gate metallization and the devices
were isolated with N implantation.
541
Mat. Res. Soc. Symp. Proc. Vol. 572 ® 1999 Materials Research Society
Implantation
damage
Ti/Al/Ni/Au
0.4
Pt/Au
I
n mm.
/
-1 V/step
,0.3-
300 Ä AlGaN
Q
E 0.2
3 um Sl GaN
-
II /
200ÄAIN
2-DEG
Sapphire
ja 0.1
Q
-
~V
0.0
10
Drain Vortage (V)
15
Fig. 2. Representative drain characteristics.
Fig. 1. Device cross section.
The epitaxial layers used in this work were grown in an inductively-heated, water-cooled,
vertical MOCVD reactor. Triethylaluminum, trimethylgallium, and NH3 were used as the
reactant sources, and Si2H_ was the Si dopant source. The AlGaN/GaN HEMT structures
consisted of a 3 urn thick undoped, high-resistivity GaN buffer layer grown on top of a thin (200
Ä) low-temperature nucleation layer. The buffer layer was employed to spatially remove the
active part of the device from the higher-defect-density material near the substrate interface. An
AlGaN layer with a total thickness of 300 Ä was grown on top of the GaN buffer. Five epitaxial
layer designs were used in this work. The salient features of these wafers, including the average
values measured across a wafer for sheet resistance, mobility, sheet carrier concentration, Imax
(the maximum drain current at a forward gate current of 0.1 mA/mm), and the threshold voltage,
V_„ are summarized in Table I. Wafers 1 and 5 were undoped, wafers 2-4 had Si-doped AlGaN
layers with -30 Ä undoped AlGaN spacer layers, and wafers 3 and 4 had a ~50 Ä undoped
AlGaN cap.
HEMT
Al
Wafer # fraction
1
0.3
2
0.3
3
0.3
4
0.3
5
0.4
Si2H6 flow
(seem)
0
0.26
0.26
0.52
0
Rsh
(__/D)
760
690
590
610
700
Mobility
(cm2/V-s)
800
830
845
870
725
Amax
vth
(mA/mm)
480
550
520
560
500
(V)
-4.5
-4.0
-4.5
-5.5
-4.5
n Sh
(1013 cm'2)
1.0
1.1
1.3
1.2
1.2
Table I. Materials and device electrical characteristics.
RESULTS
The current-voltage characteristics, drain current pulsed response, and the microwave
power output were measured for the HEMTs described above.
Drain characteristics
representative of the devices studied here are shown in Fig. 2. The negative slope at the larger
drain voltages are due to thermal effects. The transfer characteristics for a group of devices (that
542
<- ov
I—-- <- OmA
-4-2
0
2
Gate-Source Voltage (V)
Time (200 ns/div)
Fig. 4. Drain current response to a gate
voltage pulse. The gate voltage is
pulsed from a level less than Va, to 0 V.
Fig. 3. Transfer characteristics for VDS = 7
V. The maximum forward gate current
was 0.1 mA/mm. The line style used
represents the different output power levels
measured.
were subsequently measured for pulsed-response and power output) are shown in Fig. 3. The
maximum drain current level, Imax, is within the range of 0.5 to 0.65 A/mm.
An estimate of the maximum power that can be obtained from the transistor is AIAV/8,
where AI is the available current swing (= Iraax), and AV is the knee voltage subtracted from the
gate-drain breakdown voltage at pinch-off. Based on the dc values of current and voltage, the
estimated power output from this group of wafers should be comparable. However, as described
below, the power output level varies by more than a factor of 5.
The drain current response to a gate voltage pulse was also measured for this group of
devices. An example of this is shown in Fig. 4. For this case, the gate voltage is pulsed from a
level of -8V to 0V and VDs was IV. The pulse width was 0.6 |is and the duty cycle was 1%. The
drain current was measured with a current probe. The dc value of IDss (VGS = 0) is also shown in
the figure. This device exhibits a high pulse-to-dc current ratio, 90%. Other devices that were
measured exhibited a much lower ratio; as low as 5%.
The choice of the starting value of VGs has a significant effect on the observed drain
current pulse. In general, to see a significant difference between the pulsed and dc current level,
the starting VGs must be at the threshold voltage or lower. This variation is shown in Fig. 5. In
addition, values of VDs from 0.1 to 10V were investigated and it was determined that VDs
typically had a minimal effect on the current ratio. This behavior is shown in Fig. 5, where the
starting VGs is varied from -8 to 0V for two different values of VDsThe microwave power output for these devices was measured using on-wafer microwave
probes. The input and output tuning as well as the gate and drain bias were adjusted to maximize
the output power. The measurement frequency was 2 GHz, at which the small-signal gain was
between 10 and 13 dB. Output power levels as high as 2.3 W/mm and as low as 0.4 W/mm were
measured. The power output is plotted as a function of drain current pulse response in Fig. 6,
and a strong correlation is observed.
543
The results described above were obtained under normal room lighting conditions. The
effect of ultraviolet illumination on both the microwave and pulsed current measurement was
investigated. It was generally observed that ultraviolet illumination significantly increased the
power output and pulse response ratio for HEMTs where these values were low, but had a
minimal effect for devices where these values were high. For example, for devices that
performed poorly in normal lighting, the current ratio was found to increase from 0.12 to 0.65
and the microwave power output was found to increase by 4 dB with ultraviolet illumination.
For devices that performed well, the pulse ratio increased from 0.84 to 0.90 and the power output
increased by <0.5 dB with the application of ultraviolet illumination.
1.1
1.0
.2 °-
vK = iV-
,* *"'
9
to
DC
JS0.8
J3
-'0.7
0.6
0.5
-8
tiii
-7
-6
i
i
-5
-4 -3 -2
Starting Ves(V)
i
-1
0
0.0
0.2
0.4
0.6
0.8
Current Ratio (lpulse/ldc)
Fig. 6. Correlation of microwave power
output at 2 GHz with drain current response.
Fig. 5. Variation of pulse response with
starting gate-source voltage.
DISCUSSION & CONCLUSIONS
The measured results can be explained by the presence of traps in the device structure.
The pulse measurement reported here is related to gate lag measurements done originally on
GaAs MESFETs to assess their performance potential in digital circuits. In those cases, the gate
lag phenomenon was usually attributed to surface states in the access regions between the metal
contacts which acted as electron traps. Although in the present case, additional study is
necessary to establish with certainty that surface states are involved in the observed phenomena,
the lack of a gate recess or a systematic passivation procedure, makes this a likely possibility. A
plausible mechanism consists of electrons being trapped at the access region surface with a
negative gate voltage and causing a lifting of the quantum well below, thereby reducing its free
electron density. When the gate voltage switches to 0 V, thereby opening the channel under the
gate, the trapped electrons cannot respond quickly, and the 2 DEG density below increases only
slowly to its full value.
In the power output tests, devices with more surface states will have lower values of
output power, since the maximum drain current is limited by the steady-state trap occupancy of
the surface states which causes a reduction in the electron density in the quantum well. Although
544
further tests are necessary to fully characterize and then hopefully eliminate the mechanism
causing the limitation in output power, the observed correlation between gate-pulse response and
power performance yields a simple screening procedure which can be utilized early in the device
fabrication process.
ACKNOWLEDGMENTS
This work was supported by the Office of Naval Research.
REFERENCES
1. S. T. Sheppard, K. Doverspike, W. L. Pribble, S. T. Allen, J. W. Palmour, L. T. Kehias, and T.
J. Jenkins, to be published, IEEE EDL, Apr. 1999.
2. S. C. Binari, H. B. Dietrich, W. Kruppa, G. Keiner, N. S. Saks, A. Edwards, J. M. Redwing, A.
E. Wickenden, and D. D. Koleske, Proc. Inter. Conf. Nitride Semicond., pp.476-478, 1997.
3. S. C. Binari, W. Kruppa, H. B. Dietrich, G. Keiner, A. E. Wickenden, and J. A. Freitas, Jr.,
Solid-State Electron. 41, pp. 1549-1554, (1997).
4. W. Kruppa, S. C. Binari, and K. Doverspike, Electronics Lett. 31, pp. 1951-2, (1995).
5. M. Rocci, Physica 129B, pp. 119-138, (1985).
6. R. Yeats, D. C. D'Avanzo, K. Chan, N. Fernandez, T W. Taylor, and C. Vogel, IEEE IEDM
Digest, pp. 842-845,(1988).
7. A. Platzker, A. Palevsky, S. Nash, W. Struble, and Y. Tajima, IEEE MTT-S Digest, pp. 11371140,(1990).
8. J. C. Huang, G. Jackson, S. Shanfield, W. Hoke, P. Lyman, D. Atwood, P. Saledas, M.
Schindler, Y. Tajima, A. Platzker, D. Masse, and H. Statz, IEEE MTT-S Digest, pp. 713-716,
(1991).
545
PHOTOIONIZATION SPECTRA OF TRAPS RESPONSIBLE FOR CURRENT
COLLAPSE IN GaN MESFETS
P.B. KLEIN, J.A. FREITAS, Jr., AND S.C. BINARI
Naval Research Laboratory, Washington DC 20375-5347, klein@bloch.nrl.navy.mil
ABSTRACT
Current collapse in GaN MESFETS is believed to result from the trapping of carriers in
the high resistivity GaN layer, and can be reversed by the application of light. Light
photoionizes (or photoneutralizes) the carriers, releasing them from the traps and restoring all or
part of the original I-V characteristics of the device. In these investigations we have taken
advantage of this effect to characterize the traps responsible for current collapse in an n-channel
GaN MESFET. At fixed source-drain voltage, the incremental light-induced drain current,
above that measured in the dark, and normalized per incident photon, is measured as a function
of wavelength. The resulting photoionization spectrum reflects two absorption thresholds
corresponding to two distinct electron traps. Because of the nature of the measurement, these
traps can be identified as those responsible for current collapse in the device.
INTRODUCTION
Significant progress has been made over the last several years in understanding the
physical characteristics of the group-m nitride material system, as well as in developing
optoelectronic and electronic devices based on these materials. The material parameters of this
system promise the possibility of convenient sources and detectors in the UV and blue portions
of the spectrum, as well as electronic devices capable of operating at high power, high
temperature and in adverse environments. Indeed, blue-green LED's , as well as blue lasers
operating for several thousand hours are now commercially available, in spite of the fact that the
materials used to fabricate these devices contain relatively high concentrations of defects.
Similarly for electronic devices, FETs have been successfully produced in GaN as well as in
AlGaN/GaN HEMT structures. While the properties of these FETs continue to improve, traprelated phenomena still hinder the reproducible fabrication of high quality electronic
devices.
Two of the most commonplace trapping phenomena are persistent photoconductivity
(PPC) and "current collapse". PPC is the optical excitation of photoconductivity in a material
that exists for times long after the optical excitation source is removed. This process is often
associated with the presence of a metastable deep defect, such as the DX center in AlGaAs, or
can be induced by the transport of photoexcited carriers across macroscopic potential barriers
[1]. It has been noted with respect to the AlGaAs/GaAs system that while PPC poses no direct
problem for FETs, its presence does indicate the possibility of other transient device instabilities
associated with charge trapping, such as shifts in the threshold voltage [2]. "Current collapse"
refers to the trapping of charge at deep level centers in the structure, resulting in a dramatic
reduction in the current flowing through the device, and hence in its output power.
Current collapse is initiated by subjecting the device to a high electric field, such as that
experienced by ramping the drain voltage up to a large value. This is thought to inject hot
carriers from the channel into regions of the structure where they can be trapped by deep defects.
The resulting buildup of charge forms a depletion region in the channel which tends to pinch off
547
Mat. Res. Soc. Symp. Proc. Vol. 572 ° 1999 Materials Research Society
the channel and reduce the source-drain current. Khan et al. [3] studied collapse in an
AlGaN/GaN heterostructure insulated gate FET (HIGFET) and, following analogous work in
similar AlGaAs/GaAs structures [4], suggested that the trapping occurred in the AlGaN gate
insulator. Similarly, Binari et al. [5] observed this effect in GaN MESFETS. In that case, the
carrier trapping was assumed to occur in the semi-insulating GaN layer below the active n-type
channel layer. In both of these studies it was observed that the current collapse could be
reversed by the application of light, and that the effectiveness of the light illumination
diminished monotonically with increasing wavelength.
It is the purpose of this investigation to take advantage of the light sensitivity of the current
collapse phenomenon in order to begin to identify the traps responsible for this effect in GaN
MESFETs. While current interest is more focused on the more efficient HFET structures, it is
important to understand these processes in the simpler MESFET, where only the GaN material
contributes. With channel carriers trapped on deep level centers in the structure, the optical
reversal of current collapse should be viewed as a photoionization (or photoneutralization,
depending on the charge state of the trap) of the trapped carrier from the trap. Consequently, the
dependence of the light-induced increase in the drain current on the illumination wavelength
should reflect the photoionization spectrum of the trap. This spectrum, and its associated
photoionization threshold, can serve as an identifier for the trap involved in the current collapse
process.
EXPERIMENT
Details of the MESFET design and characterization are described in Ref. 5. The device
layout for the MESFET is shown in Fig. 1 and a cross-section is shown in Fig. 2. The FET was
fabricated with a source-drain spacing of 5(xm, a gate width of 150(J.m and a gate length of
1.5u,m. The active channel was grown on top of a thick, undoped semi-insulating buffer layer in
order to spatially remove the active part of the device from the higher-defect-density material
near the sapphire substrate. Hall measurements at 300K indicated a channel carrier
concentration of 2xl017cm"3 and a mobility of 410 cm2/V-sec. The current collapse
measurements were carried out by obtaining drain current characteristics of the device with light
illumination and in the dark. These were determined using an HP4145B semiconductor
parameter analyzer, which measures the drain characteristics with a single sweep of VDS for each
gate bias.
Pt/Au
Ti/AI
JZL
He implantation
damage
Ti/AI
200 nm n-GaN
3 \im SI GaN
20 nm AIN
Sapphire
Fig. 1. MESFET layout. Solid lines indicate
source, drain and gate metallizations. Dotted
lines indicate implant isolation regions.
Fig. 2. MESFET cross-section.
548
For reproducible measurements of current collapse, it was found necessary to initialize the
device before each measurement by proximity illumination with a blue GaN LED (lmW), which
emptied all or most of the traps. The measurements were carried out near zero gate bias by
setting the system to record two consecutive I-V curves with VGs at OV and -10mV, with a 5 sec
delay between scans. After the first I-V curve was recorded in the dark, an electronically
activated shutter was opened, allowing the device to be measured during the second scan under
optical illumination. Monochromatic light was obtained from a 75W Xe arc lamp (PTI model
A1010) and a Spex 1680B 0.22m. double monochromator with a 1200 gpm holographic grating,
with the spectrometer bandpass set to approximately 3.5nm. The light was collected with a
spherical mirror and focused onto the device with an output power of generally tens of p.W in an
image area of about 20mm2. I-V curves were obtained for light-on and light-off (dark)
conditions, and the fractional increase in drain current taken under illumination (at wavelength
X) over that taken in the dark, (Ix-WkVW , was measured at a predetermined fixed drain
voltage (taken at VDS=5V). This light-induced increase in IDS was normalized at each
wavelength by the incident photon flux © (photons/cm2/sec). The resultant quantity, defined
by R(X) = [(I^-Idar^/IdariJ/O^) is a measure of the number of traps emptied by photoionization
per incident photon. The spectral dependence of R(X) reflects the photoionization spectrum [6]
of the trap, thus allowing us to probe the traps responsible for the current collapse.
RESULTS
The experimentally determined spectrum R(hv), plotted as a function of incident photon
energy (hv = hc/X.), is shown as the filled circles in Fig. 3. Two broad absorptions are observed.
In addition, a rise in the drain current is observed near the GaN bandgap, and is assumed to
result from photoexcited carriers injected into the channel. As observed by Binari et al. [5], a
monotonic decrease in the effectiveness of light to reverse current collapse is observed with
increasing wavelength.
This optically-induced reversal appears to result from the
photoionization of two distinct trapping centers, labeled Trap 1 and Trap 2. It should be
emphasized again that the absorptions seen here represent optical transitions from traps that are
intimately connected with the current collapse phenomenon.
The dotted line in the figure represents a best fit of the data to the functional form for the
photoionization cross-section of a deep-level defect [7,8], where a vertical, "forbidden" optical
transition is assumed: o(hv) °= (hv-E(h)3/2/(hv)3, where E«, is the absorption threshold. A similar
form for the analogous "allowed" transition [8], which employs an additional fitting parameter,
does not significantly improve the fit. The failure of this fitting procedure appears to result from
the fact that the observed absorptions are particularly broad. Large optical linewidths associated
with deep centers in semiconductors often reveal the presence of strong coupling of the
electronic states of the deep center to the vibrational states of the lattice [6]. Such coupling is
usually associated with a significant lattice distortion at the defect site. As sketched in Fig. 4, at
300K, absorption between the trap and the conduction band can take place from vibrational
excited states as well as from the lowest trap state, thus resulting in a Gaussian broadening of
the absorption.
In such cases, photoionization data must be fitted by a convolution of the photoionization
cross-section with a Gaussian broadening function. This procedure has been successfully
549
i i i i l i i i i I i i i i i n i i i i i ii i i i i i i i.
10-13
10"14b
-10-15 E.
x:
a P
Reddy et al.
0 0 0 0 Hirsch et al.
' / '
'
'
'
'
1.5
'
i
i
i
i
i
i
i
i
i
i
'
'
2.0
2.5
3.0
Photon Energy (eV)
i
'
'
i
'
'
'
'
i
' -
3.5
Fig. 3. The filled circles represent the spectral dependence of the optical reversal of current
collapse, R(hv), that reflects the photoionization spectrum of the traps causing this effect. The
dotted line represents the best fit of the data to a standard deep level photoionization spectrum,
while the solid line is a fit employing the convolution of the photoionization spectrum with a
Gaussian broadening function. The open squares and diamonds represent data from recent
photoconductivity studies (individually scaled).
employed in studies of the DX center in AlGaAs [9] and the EL2 center in semi-insulating GaAs
[10], and has recently been applied to account for capacitance transient spectra of the deep "E2"
center in n-GaN [11]. Using the approach of Mooney et al. [9], the solid line in Fig. 3 represents
such a fit of the current collapse data, and accounts quite well for the spectral dependence of
R(hv). The dashed lines are the Trap 1 and Trap 2 components of this fit. For each absorption,
an amplitude, an energy threshold and a Gaussian broadening parameter are deduced. For Trap
1, Ai=1.5xl0"16, E,=1.8 eV, and O"i=0.26 eV. For Trap 2, A2=1.4xl0"14, E2=2.85 eV, and
O"2=0-10 eV. Note that these broadening parameters correspond to Gaussians with FWHM of
approximately 0.6eV and 0.25eV, respectively, which verifies the large breadth of the
absorptions. Unlike the unbroadened photoionization cross-section, there is significant
absorption at energies lower than the threshold energy. As can be seen from the diagram in Fig.
5, these two threshold energies correspond to two very deep traps. It is also important to note
that these optical threshold energies exceed the trap depths that might be determined from a
DLTS measurement of the same center, for example, by the lattice relaxation energy (or FranckCondon energy) dFc indicated in Fig. 5.
550
Electronic +
Vibrational
The data in Fig. 3 is also compared
*
to two recent investigations of
photoconductivity in GaN. Hirsch et al.
[12] studied the spectral dependence of
Energy
PPC in MOCVD grown, unintentionally
Broadening
doped n-GaN, and deduced the
photoionization spectrum (diamonds)
shown (scaled) in Fig. 3 from the time
dependence of photoconductivity buildup
in their samples. Similarly, Reddy et al.
[13] also measured PPC in undoped,
MBE-grown GaN, but in a slightly
different spectral region: their photocurrent results are shown (scaled) as the
squares in Fig. 3. That these data agree
with our current measurements for Trap
Qo
2 and Trap 1, respectively, suggests the
Lattice Distortion
possibility that we are observing the
same traps, but it is curious that each of
Fig. 4. Broadening of absorption lines from a
these studies reports a single trap. The
strongly lattice-coupled deep level defect.
signal-to-noise of the data of Hirsch et al.
would suggest that they would have been
unable to detect Trap 1, but it is not clear
whether Reddy et al. investigated the
energy range of Trap 2, or whether that trap simply does not exist in MBE-grown material. If
the latter were found to be true it would be a significant point, as the trap is a major contributor
to current collapse in GaN-based devices.
We have also studied the dependence of the light-induced reversal of current collapse on
the excitation intensity of the light source. It was found that at 400nm (3.1eV), the optical
response from Trap 2 is almost fully saturated (i.e. most of the traps are empty) at the highest
power used, corresponding to about lmW/cm2. For Trap 1, studied at 600nm (2.07eV), the
highest excitation power was found to be within an order of magnitude of saturation values.
Saturation at such low power levels suggests either a low trap density, a large photoionization
cross-section, or a small capture cross-section, although these parameters cannot yet be
unraveled by the present measurements. However, Hirsch et al. have observed very slow
photoconductivity buildups associated with the deep center that appears to be identical to Trap 2
in the current work. From these results they concluded that the capture cross-section for this
trap must be extremely small, in agreement with our current observations.
CONCLUSIONS
In these studies we have taken advantage of the optical reversibility of the current collapse
phenomenon in order to probe the deep traps responsible for this process in a GaN MESFET. In
addition to a near-bandedge effect believed due to the injection of photoexcited carriers into the
channel, two broad, below-gap absorptions were observed, which we have associated with the
photoionization of trapped electrons at two distinct deep centers in the semi-insulating GaN
551
layer. These absorptions were much too broad
to fit with the standard spectral dependence for
a deep-center photoionization cross-section.
Assuming strong coupling of the deep trap
with the lattice, the data was well accounted
for by a convolution of the photoionization
cross section and a Gaussian broadening
function. The resulting threshold energies
reflect two very deep traps associated with
current collapse, with absorption thresholds at
1.8 eV and 2.85 eV. These traps may have
been observed in recent PPC studies as well.
Energy(eV)
n
ACKNOWLEDGEMENTS
This work has been supported in part by
the Office of Naval Research.
Fig. 5. Threshold absorption energies of Trap
1 and Trap 2 relative to the GaN band edges, as
a function of local lattice distortion at the
defect site.
Lattice Distortion
REFERENCES
1.
2.
3.
4.
H.J. Quiesser and D.E. Theodorou, Phys. Rev. Lett. 43,401 (1979).
T.N. Theis and S.L. Wright, Appl. Phys. Lett. 48, 1374 (1986).
M.A. Khan, M.S. Shur, Q.C. Chen and J.N. Kuznia, Electron. Lett. 30, 2175 (1994).
R. Fisher, T.J. Drummond, W. Kopp, H. Morkoc, K. Lee and M. Shur, Electron. Lett.
19,789 (1983).
5. S.C. Binari, W. Kruppa, H.B. Dietrich, G. Keiner, A.E. Wickenden and J.A. Freitas, Jr.,
Solid State Electron. 41,1549 (1997).
6. B.K. Ridley, Quantum Processes in Semiconductors, 3rd ed. (Clarendon Press, Oxford,
1993).
7. G. Lucovsky, Solid State Commun. 3,299 (1965).
8. J.C. Inkson, J. Phys. C 14, 1093 (1981).
9. P.M. Mooney, G.A. Northrup, T.N. Morgan and H.G. Grimmeiss, Phys. Rev. B 37, 8298
(1988).
10. A. Chantre, G. Vincent and D. Bois, Phys. Rev. B 23, 5335 (1981).
11. P. Hacke, P. Ramvall, S. Tanaka, Y. Aoyagi, A. Kuramata, K. Horino and H. Munekata,
Appl. Phys. Lett. 74, 543 (1999).
12. Michele T. Hirsch, J.A. Wölk, W. Walukiewicz and E.E. Haller, Appl. Phys. Lett. 71, 1098
(1997).
13. C.V. Reddy, K. Balakrishnan, H. Okumura and S. Yoshida, Appl. Phys. Lett. 73, 244
(1998).
552
AUTHOR INDEX
Agarwal, A.K., 23
AgerIII,J.W.,451
Ahmed, S.I., 99
Albrecht, J.D., 489
Alcom, T.S., 15
Allen, S.T., 15
Alok, D., 63
Ambacher, O., 501
Ananthanarayanan, P., 81
Ancona, M.Q., 489
Anderson, D.R., 507
Arai, K., 105
Arnold, E„ 63
As, D.J., 225
Augustine, Q., 245
Avci, R., 99
Averous, M., 213
Ayoub, M.A., 123
Babiker, M., 507
Bai, X., 529
Balakrishna, V., 245
Baliga, B.J., 81
Beaumont, B., 419
Beling, CD., 117
Benamara, M., 357, 363
Bennett, C.R., 507
Bensaoula, A., 535
Berishev, I., 535
Bickermann, M., 259
Bidnyk, S„ 351, 439
Binari, S.C., 489, 541, 547
Bland, CD., 413
Boutros, K., 471, 495
Bremser, M., 321
Brogueira, P., 395
Brown, Q.J., 281
Burger, A, 87
Burk, Jr., A.A., 161
Burton, J.C, 201
Cai, C, 495
Caldweil, M.L., 413
Cao, X.A., 513
Capano, M.A., 3, 45
Carlson, E.P., 307
Carter, Jr., C.H., 167
Chaker, M., 401
Chan, H.L.W., 389
Chan, S.H., 523
Chang, C.Y., 523
Chatty, K., 63
Chen, M.Y., 471, 495
Chen, Q., 471, 495
Chen, X.D., 117
Chen, Y„ 179, 191
Cheung, D.W., 289
Chilukuri, R.K., 81
Cho, K.I., 383
Cho, Y-H„ 457
Choi, Y.H., 481
Chong, K-B., 301
Chow, T.P., 63
Choy, C.L., 389
Chu, V., 395
Chua, S-J„ 301
Chung, K.W., 481
Cimalla, V., 99
Cole, M.W., 93
Coleman, P.O., 129
Collins, W.E., 87
Considine, L., 327
Cooper, J.A., 3
Crenshaw, T., 87
Danilewsky, A.n., 327
Das, M.K., 3
Davis, R.F., 219, 307,407
Demaree, D.J., 93
DenBaars, S.P., 315, 351, 433
De Theije, F.K., 345
Dietrich, H.B., 541
Do, S.J., 141
Drube, W., 39
Dubaric, E., 69
Dupuis, R.D., 357
Eastman, L.F., 445, 501
Ecke, Q., 99
Eckstein, R., 271
Eickhoff, Th., 39
Eiting, C.J., 357
Eriksson, J., 197
Erler, F., Ill
Ervin, M.H., 413
Eshun, E., 173
Evwaraye, A.O., 237, 253
Farber, B., 369
Fejes, P., 219, 407
Ferguson, I., 173, 201
Fini, P., 315
Fong, W.K., 389
Fountzoulas, CO., 93
Foutz, B.E., 445, 501
Freitas, Jr., J.A., 547
Frey, T., 225
Fukuda, H., 105
Fung, A.K., 495
Fung, S„ 117
Gainer, Q.H., 457
Qaska, R., 427
Qaspar, C, 225, 419
Qehrke, T., 307
George, M.A., 123
553
Qertner, E., 471
Qibson, W.M., 377
Qlans, F-A., 39
Qoldstein, J., 253
Qong, M., 117
Qrau, M., 259
Gregory, R.B., 33, 231
Qrehk, T.M., 39
Qriffin, J., 45
Qrudowski, P.A., 357
Qrzegory, I., 363
Quo, H.J., 513
Quo, J.D., 523
Qurray, A., 173
Qutmann, R.J., 63
Qwilliam, R.M., 129
Hageman, F.R., 345
Hall, W.B., 23
Han, J., 513
Hanson, T„ 23
Harris, C.I., 197
Hecht, C, 149
Henkel, T., 117
Henry, R.L., 489, 541
Heuken, M., 321
Higgins, A., 471
Hill, D.M., 529
Hjelm, M., 69
Hofmann, D„ 259, 275
Holland, O.W., 33
Homewood, K.P., 129
Hong, M.H., 369
Hong, S.E., 383
Hong, S.Q., 219, 407
Hopkins, R.H., 245
Huang, J., 207
Hubbard, C.W., 93
Iakimov, T., 265
Ibbetson, J.P., 315
Ikossi-Anastasiou, K., 489, 541
Ha, D., 123
Irvine, K.G., 167
Jacobs, K„ 327
Jacobsson, H„ 265
Janzen, E., 265
Je, J.H., 141
Johansson, L.I., 39
Johnson, CM., 129, 135
Jones, K.A., 185,339
Juergensen, H., 321
Kackell, P., 69
Kang, S.C., 141
Karlsson, S., 197
Keller, S„ 351, 433
Keiner, Q., 541
Khan, A., 471
Khan, M.A., 357
Khan, M.A., 495
Khatri, S„ 23
Khlebnikov, I., 57
Kim, J.Q., 333
Kim, K.H., 383
Kim, Y„ 289, 295, 451
Kisielowski, C, 369, 451
Klein, P.B., 547
Knights, A.P., 129
Kobayashi, H„ 377
Kobayashi, N„ 117
Kölbl, M., 271
Koleske, D.D., 489, 541
Konkar, A., 231
Konstantinov, A.O., 197
Korakakis, D., 427
Kordesch, M.E., 529
Kordina, O., 167
Kornegay, K, 45, 173
Kottke, M., 407
Koynov, S., 395
Krüger, J., 289, 295, 451
Kruppa, W., 541
Kuech, T.F., 463
Kum, B.H., 141
Kwon, Y.H., 351
Lai, W.C., 523
Lam, J.B., 351, 457
Larkin, D.J., 123
Larsen, P.K., 345
Lee, W.S., 481
Li, P., 301
Li, Y„ 3
Liaw, H.M., 219, 407
Liliental-Weber, Z., 295, 357, 363
Lin, C, 207
Lindner, J.K.N., 111
Linthicum, K.J., 219, 307, 407
Linville, R.J., 281
Lischka, K., 225
Little, B.D., 351, 439
Long, F.H., 201
Lourenco, M.A., 129
Lowney, D., 327
Lu, W.J., 87
Luenenbuerger, M., 321
Lueng, CM., 389
Macfarlane, P.J., 51
MacMillan, M.F., 23
Madangarli, V., 57, 75
Magtoto, N.P., 413
Mani, S.S., 23
Marchand, H., 315
Masri, P., 213
Masuda, Y., 191
Matin, M., 3
Matsurnoto, K., 179, 191
Mazur, J.H., 363
McDermott, B., 471
McDermott, B.T., 495
McQlothlin, H.M., 3
554
McNally, P.J., 327
McPherson, S.A., 457
Medelci, N., 535
Mehregany, M., 307
Meister, D., 395
Melloch, M.R., 3
Merel P., 401
Merz, J.L., 427
Meyer, B.K., 395
Mintairov, A.M., 427
Mishra, U.K., 315, 351, 433
Mitchel, W.C., 237, 253, 281
Molnar, R.J., 289
Monteiro, T., 225, 419
Moon, D.C., 481
Moorthy, M„ 333
Moran, B„ 315
Moreaud, N., 213
Morisette, D.T., 3
Morrison, D.J., 129
Moustakas, T.D., 427
Müller, St.Q., 275
Murphy, M.J., 495, 501
Nagai, K., 105
Nagapudi, V., 81
llathan, M.I., 495
Nikishin, S.A., 231
riilsson, H-E., 69
Nilsson, P-A., 197
liishino, S., 179, 191
Nishio, Y., 179, 191
Mordby, Jr., H.D., 161
Nordell, N., 197
O'Hare, M., 327
O'Leary, S.K., 445
O'Loughlin, M.J., 161
O'Neill A.Q., 129, 135
Orlov, V., 369
Ortolland, S., 129
Osinsky, A.O., 427
Östling, M., 207
Paek, M.C., 383
Paisley, M.J., 167
Paimour, J.W., 15,167
Park, R.M., 333
Pearton, S.J., 513
Pennycook, S.J., 513
Pepin, H., 401
Pereira, E., 225, 419
Perez, R., 471
Perlin, P., 289, 451
Perrin, R., 281
Persson, C, 69
Petersson, C.S., 69
Pezoldt, J., 213
Pierson, R., 471
Pirouz, P., 369
Pittman, R., 471
Poisson, M.A., 419
Pophristic, M„ 201
Popovici, Q., 519
Porowski, S., 363
Pribble, W.L., 15
Protzmann, H., 321
Raback, P., 265
Rantamäki, R., 327
Redwing, J.M., 463, 471,495
Ren, P., 93
Rendakova, S„ 45
Richardson, H.H., 413
Ridley, B.K., 507
Rieger, D.J., 513
Ring, Z., 15
Rocha, R., 395
Rodrigues, R., 23
Romanus, H., 99, 111
Ronning, C, 307
Rorsman, N., 197
Rose, W.L., 45
Ruden, P.P., 489, 495
Rupp, R„ 149
Ruvimov, S., 295, 363
Ryu, S-H., 3
Sadler, R.A., 15
Sakamoto, K., 105
Salamanca-Riba, L., 185, 339
Samant, A.V., 369
Sands, T., 289
Sänger, P.A., 23
Sarney, W.L., 185,339
Saroukhan, A-M„ 197
Saxler, A., 281
Schaefer, J.A., 99
Schaff, W.J., 495, 501
Schermer, J.J., 345
Schikora, D., 225
Schmidt, T.J., 351, 433, 439
Schmitt, E., 271
Schoen, O., 321
Schoettker, B., 225
Schwarz, R., 395
Seitz, R., 225, 419
Sekhar, J.A., 513
Sekigawa, T., 105
Seshadri, S., 23
Shapiro, N„ 289
Sharma, R.P., 185
Shealy, J.R., 501
Sheppard, S.T., 15
Shi, D.T., 87
Shim, K.H., 383
Shin, M.W., 141,481
Shul, R.J., 513
Shur, M.S., 445
Siegle, H„ 289, 451
Singh, R., 167
Singh, R.K., 513
Smart, J.A., 501
Smirnov, M.B., 427
555
Smith, R.P., 471
Smith, S.R., 237, 253
Solomon, J.S., 253
Soloviev, S., 57, 75
Song, J.J., 351, 433, 439, 457
Spalding, CM., 413
Speck, J.S., 315
Spencer, M.G., 45, 173,185, 339
Spiess, L„ 99, 111
Spitz, J., 3
Stall, R., 173
Starikov, D., 535
Stauden, Th., 213
Stephan!, D„ 149
Straubinger, T.L., 259
Subramanya, S., 289
Sudarshan, T.S., 57, 75
Sudhir, Q.S., 295,451
Sullivan, Q.J., 471, 495
Sun, J., 463
Surya, C, 389
Svedberg, J-O., 197
Swider, W., 357
Syväjärvi, M., 265
Sze, S.M., 523
Tabbal, M., 401
Tanaka, Y., 117
Tanoue, H., 117
Taylor, C, 45, 173, 185
Temkin, H., 231
Tempez, A., 535
Thomas, C, 45
Thomas, D.K., 33
Thomson, D., 307
Tilak, V., 501
Tompkins, H„ 219
Topf, M„ 395
Tsang, J.S., 523
Tucceri, R.C., 413
Tuomi, T., 327
Van Enckevort, W.J.P., 345
Van Hove, J.M., 495
Vehanen, A., 265
Vlasov, A.S., 427
Wahab, Q., 39
Wang, L., 207
Wang, Y., 207
Washburn, J., 295,357, 363
Weber, E.R., 289, 295,451
Wellmann, P.J., 259
Wen, J., 207
Werho, D., 219
Wetteroth, T.A., 33
Wickenden, A.E., 489, 541
Wiedenhofer, A., 149
Wilson, R.Q., 513
Wilson, S., 339
Wilson, S.R., 33, 219, 231, 407
Winnacker, A., 259, 275
Woelk, E., 321
Wöhner, T., 213
Wong, W.S., 289
Wood, M.A., 413
Woodall, J.M., 3
Wright, M.Q., 129, 135
Yakimova, R., 265
Yang, J-W., 357, 471
Yang, J.W., 495
Yang, W., 351, 433, 457
Yokoyama, M., 523
Yoshida, S., 105
Zakhleniuk, H.A., 507
Zauner, A.R.A., 345
Zavada, J.M., 513
Zetterling, C-M., 207
Zhang, Q., 57, 75
Zhang, X., 301
Zhao, I., 315
Zhou, P., 185, 339
Zollner, S., 219, 231, 407
Zolper, J.C., 513
Zorman, CA., 307
Zvanut, M.E., 51
556
SUBJECT INDEX
acoustic deformation potential, 275
activation, 513
energy, 237
admittance spectroscopy, 197
AES, 99
ATM, 45, 87, 345
AlQaW, 457, 471
AlQaM/QaM, 495
heterostructures, 489
AIM, 231, 333, 339, 407
films, 413
alpha-factor, 345
aluminum nitride, 389
annealing, 207
atomic
force microscopy, 315
nitrogen source, 401
bandgap, 289
fluctuation, 457
beryllium, 117
bias-temperature-stress (BTS), 63
breakdown
field, 57
voltage, 75,141
buffer layer, 407
cathode luminescence, 413
channeling, 377
chemical
sensor, 123
vapor deposition, 237
composite, 301
computational fluid dynamics, 463
conductivity, 333
contact. 111
copper, 523
C-related centers, 51
crystal growth, 271
cubic(/)
hexagonal phase, 451
SiC, 191
current
collapse, 547
saturation, 489
CVD, 149, 167
doping, 117, 519, 535
uniformity, 161
edge termination, 75
electrical resistivity, 99
electron paramagnetic resonance (EPR), 51
elevated temperature, 123, 173
ellipsometry, 231
emission lines, 419
energy tail states, 457
epitaxial, 237
growth, 179, 197, 333, 401, 407
lateral overgrowth, 327
epitaxy, 149, 161, 219, 295
exciton scattering, 439
extended defects, 513
field
effect mobility, 63
plate, 81
forward current density, 75
4Hpolytype, 265
SiC, 105, 167, 179, 245
MOSFET, 63
FTIR, 281
gain mechanisms, 439
gallium nitride, 307, 315,327, 345, 389, 463
QaM, 225, 295, 301, 369, 377, 383, 395,
401, 419, 433, 451, 481, 501, 507, 535,
541, 547
A1N, InN, 519
HEMTs, 15
heteroepitaxial growth, 289
gate turn-off thyristor (QTO), 23
Qe substrates, 451
graphite mask, 45,173
growth, 245,339
Hall(-)
effect, 281
mobility, 275
heteroepitaxial growth, 191
heterostructure, 383
HFET, 471
high(-)
frequency power, 15
power switching, 93
temperature, 529
annealing, 513
voltage, 81
homoepitaxy, 173
hot-wall, 167
hybrid, 471
hydrogen
annealing, 105
implantation, 33
deep defects, 395
defect(s), 265
reduction, 315
deformation, 369
depo-conversion, 57
digital recording, 259
diode, 23
dislocations, 369
distribution, 513
DLTS, 129
dopant site, 377
557
characteristics, 457
phonons, 427
oxidation, 51,105,135
hydrostatic stress, 495
image plate, 259
implant damage, 33
implantation. 111, 117, 513
impurity, 245
infrared spectroscopy, 427
InQaM, 357,383
InQaN/QaM - SQW, MQW, 321
interface optimization, 213
interferometer, 389
intermediate, 301
inverter, 23
ion implantation, 45
Pauli principle restrictions, 507
Pd, 87
pendeo-epitaxy, 307
photo-assisted, 535
photoconductivity, 281,495
photoelectrochemical etching, 481
photoelectron spectroscopy, 39
photoionization, 547
photoluminescence, 201, 225, 289, 419
photoresponse, 395
physical vapor transport, 259
physics of epitaxy, 213
piezoelectric, 501
coefficient, 389
scattering, 507
P1XE, 377
planar edge termination, 81
planetary reactors, 321
plasma-assisted growth, 383
p-n junction diode, 81
polarization, 501
poly-Si, 105
polytype, 185
polytypism, 339
positron annihilation, 129
power, 471
p-type SiC, 75
pulsed laser deposition (PLD), 207, 401
PVT, 245
large-bandgap semiconductor, 507
Iasing, 351
lateral
epitaxial overgrowth, 315
growth, 307
lattice dynamics, 427
layers, 301
low-dimensional heterostructure, 507
macrosteps, 45
MBE, 295, 451
MESFET(s), 15, 481, 547
metal-SiC interface, 93
metallization, 111
Mg doped, 225
micropipe density, 271
microwave, 471, 541
devices, 197
power, 15
mid-infrared, 281
misfit dislocations, 219
Mn doped AIM, 413
mobile ions, 63
MOCVD, 185, 339, 345, 357
modeling, 135
molecular beam epitaxy, 213, 231, 333, 383
Monte Carlo, 69, 135
simulations, 489
morphology, 265,345
MOS, 3
MOSFET, 69
MOVPE, 321, 463
multi-quantum wells, 327
multiwafer, 149
quantum wells, 357,433, 439
Raman, 201, 451
RBS, 377
refractive index, 231
reproducibility, 161, 321
resistively heated, 173
resistivity mapping, 271
RIE, 535
sandwich method, 179
scandium, 529
scattering mechanisms, 275
Schottky
barrier
diode, 141
height, 523
diode(s), 3, 75, 129
selective area growth, 173
semi-insulating, 253
separate confinement heterostructure, 351
SiC, 369, 407
MOS, 57, 63
on Si, 191
SiCOI, 33
Si-doping, 333
suicides, 141
neutron irradiation, 519
Mi contacts, 93
nitride(s), 351, 433, 439, 529
compounds, 427
nitrogen doping, 201
nonlinear spectroscopy, 433
normal pressure CVD, 191
I-V, 123
optical
admittance (spectroscopy), 253
558
thick oxide, 57
thickness uniformity, 161
thin(-)film(s), 99, 207, 401
separation, 33
3C-SiC, 191
IH-nitrides, 519
time resolved, 225
transistor, 541
transmission electron microscopy, 315
traps, 547
tungsten
carbide, 99, 111
suicide. 111
2D electron gas, 501
Silicon, 301, 307
carbide (SiC), 3, 23, 39, 51, 69, 81, 87,
111, 117, 123, 129, 135,141, 149, 161,
185, 197, 201, 207, 219, 237, 253, 259,
265, 275, 281, 307
interface, 213
SIMS, 117, 253
characterization, 413
Si02,105
Si02/SiC interfaces, 39
6H(-)
and 4H-poIytype, 275
silicon carbide, 271
sputtering, 523
stimulated emission, 351, 433, 439
strain, 327
stress, 289
birefringency, 271
structure, 295
sublimation, 237, 265
method, 179
surface
compounds, 87
conversion, 219
mobility, 69
morphology, 45
surfactants, 185
switch, 23
ungated HFETs, 489
uniaxial stress, 495
unit current gain frequency, 69
UV
detector, 395
emitters, 351
vanadium, 253
wafer-to-wafer uniformity, 321
wide(-)bandgap, 3
semiconductor(s), 213, 541
XPS, 87, 123
x-ray
diffraction, 99
imaging, 259
topography, 327
TEM, 295, 357, 369
temperature-induced emission shift, 457
thermal
effects, 495
on electron transport, 489
expansion, 289
fluid, 463
stress, 93
559