Electrochemistry of Silicon Instrumentation
Electrochemistry of Silicon Instrumentation
Electrochemistry of Silicon Instrumentation
Electrochemistry of Silicon
Electrochemistry of Silicon: Instrumentation, Science, Materials and Applications.
Volker Lehmann
Copyright 2002 Wiley-VCH Verlag GmbH
ISBNs: 3-527-29321-3 (Hardcover); 3-527-60027-2 (Electronic)
Cover illustrations
Upper left: Electrolytic double cell for diffusion length mapping of 200 mm silicon wafers using
the ELYMAT technique, as discussed in Section 10.3. After [21].
Upper right: Electroluminescence from a micro PS film anodized in an O-ring cell viewed from the
top (10% acetic acid, 10 mA cm
2
, 2.6 cm
2
active area). Note that the luminescence appears or-
ange in the center line, where the film has been formed under high current density (in 1:1 ethanoic
HF at 200 mA cm
2
), while it appears red for low formation current density (10 mA cm
2
). After
[Le3].
Lower left: Free-standing porous silicon samples mounted on top of a 20 lm thick bulk silicon
grid (with grid bars of 7 lm width) and illuminated with white light from the back. Upper left
square: 70 lm micro PS of 69% porosity (50 min at 30 mA cm
2
in 1:1 ethanoic HF, 1 X cm p-
type), upper right square: 32 lm meso PS of 39% porosity (16 min at 30 mA cm
2
in 1:1 etha-
noic HF, 0.03 X cm p-type), lower left square: 69 lm macro PS of 72% porosity (1.85 lm diame-
ter pores in 2.3 lm trigonal pitch parallel to the light beam) and lower right square 7 lm bulk
silicon. Note that porosity and thickness of all porous samples has been selected to correspond
to 20 lm thick bulk silicon. After [Le27].
Lower right: First macroporous silicon-based chip capacitor (100 nF, 10 V) on a match for size
comparison.
Electrochemistry of Silicon: Instrumentation, Science, Materials and Applications.
Volker Lehmann
Copyright 2002 Wiley-VCH Verlag GmbH
ISBNs: 3-527-29321-3 (Hardcover); 3-527-60027-2 (Electronic)
Volker Lehmann
Electrochemistry of Silicon
Instrumentation, Science,
Materials and Applications
Electrochemistry of Silicon: Instrumentation, Science, Materials and Applications.
Volker Lehmann
Copyright 2002 Wiley-VCH Verlag GmbH
ISBNs: 3-527-29321-3 (Hardcover); 3-527-60027-2 (Electronic)
Author
Dr. Volker Lehmann
Infineon Technologies AG
Corporate Research
Otto-Hahn-Ring 6
81739 Mnchen
Germany
Library of Congress Card No. applied for
British Library Cataloguing-in-Publication Data:
A catalogue record for this book is available from
the British Library.
Die Deutsche Bibliothek
CIP-Cataloguing-in-Publication Data
A catalogue record for this book is available from
Die Deutsche Bibliothek
WILEY-VCH Verlag GmbH
D-69469 Weinheim, 2002
All rights reserved (including those of translation
in other languages). No part of this book may be
reproduced in any form by photoprinting, mi-
crofilm, or any other means nor transmitted or
translated into machine language without written
permission from the publishers. Registered
names, trademarks, etc. used in this book, even
when not specifically marked as such, are not to
be considered unprotected by law.
printed in the Federal Republic of Germany
printed on acid-free paper
Composition K+V Fotosatz GmbH, Beerfelden
Printing Strauss Offsetdruck GmbH,
Mrlenbach
Bookbinding Grobuchbinderei J. Schffer
GmbH & Co. KG, Grnstadt
ISBN 3-527-29321-3
n This book was carefully produced. Nevertheless,
authors, editors and publisher do not warrant
the information contained therein to be free of
errors. Readers are advised to keep in mind that
statements, data, illustrations, procedural details
or other items may inadvertently be inaccurate.
Electrochemistry of Silicon: Instrumentation, Science, Materials and Applications.
Volker Lehmann
Copyright 2002 Wiley-VCH Verlag GmbH
ISBNs: 3-527-29321-3 (Hardcover); 3-527-60027-2 (Electronic)
Dedicated to Hadley and other colleagues,
with thanks for good advice
Electrochemistry of Silicon: Instrumentation, Science, Materials and Applications.
Volker Lehmann
Copyright 2002 Wiley-VCH Verlag GmbH
ISBNs: 3-527-29321-3 (Hardcover); 3-527-60027-2 (Electronic)
Silicon has been and will most probably continue to be the dominant material in
semiconductor technology. Although the defect-free silicon single crystal is one of
the best understood systems in materials science, its electrochemistry to many
people is still a matter of alchemy. This view is partly a result of the interdisciplin-
ary aspects of the topic: Physics meets chemistry at the silicon-electrolyte inter-
face.
So far, researchers interested in this topic have had to choose either mono-
graphs that deal with the electrochemistry of semiconductors in general or recent
editions that deal with special topics such as, for example, the luminescent prop-
erties of microporous silicon. The lack of a book that specializes on silicon but
which gives the whole spectrum of its electrochemical aspects was my motivation
to write the Electrochemistry of Silicon.
With this book I hope to address different groups in the scientific community.
For beginners in the field a comprehensive overview of the topic is given in ten
chapters, including a brief historical review and safety tips. The practitioner will
find inspiration for instrumentation as well as examples of applications ranging
from photonic crystals to biochips. For experts the book may serve as a quick
reference with more than 150 technical tables, diagrams and micrographs, as well
as ca. 1000 references cited for easy access to in-depth information.
I did my best to eliminate mistakes and unclear descriptions, but I suspect that
even writing is governed by the laws of thermodynamics. So, I welcome com-
ments from readers and will attempt to correct any mistakes that they find.
VII
Preface
Electrochemistry of Silicon: Instrumentation, Science, Materials and Applications.
Volker Lehmann
Copyright 2002 Wiley-VCH Verlag GmbH
ISBNs: 3-527-29321-3 (Hardcover); 3-527-60027-2 (Electronic)
Preface VII
1 Introduction, Safety and Instrumentation 1
1.1 Early Studies of the Electrochemistry of Silicon 1
1.2 Safety First 3
1.3 The Basic Properties of Silicon 5
1.4 Common Electrolytes 7
1.5 The Electrodes 11
1.6 Cell Designs 15
2 The Chemical Dissolution of Silicon 23
2.1 The Basics of Wet Processing of Silicon 23
2.2 Silicon Surface Conditions and Cleaning Procedures 24
2.3 Chemical Etching in Alkaline Solutions 27
2.4 Chemical Etching in Acidic Solutions 30
2.5 Defect and Junction Delineation 33
2.6 Selective Etching of Common Thin Film Materials 36
3 The Semiconductor-Electrolyte Junction 39
3.1 Basics of the Semiconductor-Electrolyte Contact 39
3.2 The IV Characteristics of Silicon Electrodes in Acidic Electrolytes 42
3.3 The IV Characteristics of Silicon Electrodes in Alkaline
Electrolytes 49
4 The Electrochemical Dissolution of Silicon 51
4.1 Electrochemical Reactions 51
4.2 The Dissolution Valence 57
4.3 The Characteristic Anodic Currents in HF 59
4.4 Reverse Currents, Electron and Hole Injection 63
4.5 Electrochemical Etch Stops 68
4.6 Photoelectrochemical Etching 72
IX
Contents
Electrochemistry of Silicon: Instrumentation, Science, Materials and Applications.
Volker Lehmann
Copyright 2002 Wiley-VCH Verlag GmbH
ISBNs: 3-527-29321-3 (Hardcover); 3-527-60027-2 (Electronic)
5 Anodic Oxidation 77
5.1 Silicon Oxidation Techniques 77
5.2 Native and Chemical Oxides 78
5.3 Anodic Oxide Formation and Ionic Transport 79
5.4 Oxide Morphology, Chemical Composition and Electrical
Properties 82
5.5 Electrochemical Oscillations 89
5.6 Electropolishing 94
6 Electrochemical Pore Formation 97
6.1 Basics of Pore Formation 97
6.2 Porous Silicon Formation Models 99
6.3 Pore Size Regimes and Pore Growth Rates 104
6.4 Porosity, Pore Density and Specific Surface Area 108
6.5 Mechanical Properties and Drying Methods 114
6.6 Chemical Composition and Ageing Effects 117
6.7 Electrical Properties of Porous Silicon 120
7 Microporous Silicon 127
7.1 Micropore Formation Mechanism 127
7.2 Morphology of Microporous Silicon 128
7.3 Absorption, Reflection and Nonlinear Optical Effects 133
7.4 Luminescence Properties 138
7.5 Quantum Confinement and Models of the Luminescence Process 150
7.6 Oxidized Porous Silicon 159
7.7 Related Materials 162
8 Mesoporous Silicon 167
8.1 Mesopore Formation Mechanisms 167
8.2 Mesopores in Highly Doped p-Type Silicon 171
8.3 Mesopores in Highly Doped n-Type Silicon 174
8.4 Mesopore Formation and Spiking in Low-Doped n-Type Silicon 177
8.5 Etch Pit Formation by Avalanche Breakdown in Low-Doped
n-Type Silicon 180
9 Macroporous Silicon 183
9.1 Macropore Formation Mechanisms 183
9.2 Macropores in p-Type Silicon 187
9.3 The Phenomenology of Macropore Formation in n-Type Silicon 190
9.4 Calculating Macropore Growth and Mass Transport 198
9.5 Design Rules and Limits of Macropore Array Fabrication 202
Contents X
10 Applications 207
10.1 Overview 207
10.2 AC Properties of Silicon Electrodes and Carrier Concentration
Profiling 208
10.3 Diffusion Length and Defect Mapping 211
10.4 Sensors and Biochips 219
10.5 Passive and Active Optical Devices 225
10.6 Porous Silicon-Based Electronic Devices 232
10.7 Sacrificial Layer Applications 236
Appendices 243
Supplier References 249
References 251
Subject Index 273
Contents XI
absorption
chemical 220
cross-section 137
coefficient 136, 212
optical 133, 145, 212
acceptor compensation 52
accumulation 39, 44
acidic etching 30
activation energy 11, 29, 61
active state 97, 186
aging 29, 117
alkaline etching 27, 49, 53, 193
alternating current properties 126, 208
ambipolar diffusion 124
amorphous silicon 131, 164
anisotropic etching 27, 5054
annealing 88, 117
anodic oxide 7796, 101, 149
chemical composition 86
defects 86, 87
density 78, 85
dissolution 67
electrical properties 88
etchrate 69, 83, 92
formation mechanisms 52, 56, 79
growth rate 81
morphology 83, 92
porous 90
refractive index 86
anodic regimes 4449
anti-reflective coating 227
anti-scatter grid 239
applications 207241
atomic force microscopy 85
attenuated total reflection 20
Auger recombination 6, 136, 145, 156
autocatalytic 33, 163
avalanche breakdown 103, 180
backside photo current 212
band-structure 139, 144, 151, 229
bifluoride 55
biochips 223
Bragg-filter 130, 222, 226
breakdown electrical 88, 103, 168
breakdown passivity 101
Brunauer-Emmet-Teller method 112
buffered oxide etch 36
capacitance-voltage curve 209
capillary forces 115
carrier concentration profiling 208
cathodic regime 45, 51
cell designs 1522, 214
double 19, 214
electrolyte circulation 21
immersion 17
internal 72, 75
materials 15
o-ring 16, 18
windows 16
chemical
dissolution 2338, 53
oxide 78
polishing 31
reactions 5157
vapor deposition 234
chemomechanical polishing 24, 64, 96
cleaning 24, 57
cleaving 4, 14, 17
cold cathode 232
collimator 239
colloidal silica 24
concentration 7, 201
conduction band 3950, 128, 144
contact 14, 39, 120
contact angle 24
273
Subject Index
Electrochemistry of Silicon: Instrumentation, Science, Materials and Applications.
Volker Lehmann
Copyright 2002 Wiley-VCH Verlag GmbH
ISBNs: 3-527-29321-3 (Hardcover); 3-527-60027-2 (Electronic)
Coulomb-blockade 122
cracking 86, 115
current
characteristic 5963, 90
critical 60
efficiency 83
multiplication 45, 54, 66
pore tip 59, 195
reverse 6368
transient 64
current-voltage characteristic 4150
acidic 4249, 65, 80
alkaline 41, 49, 50
basics 39
hydrofluoric acid 4249, 6068, 178,
196
dangling bond 121
dark current 65, 217
Debye length 100
deep level transient spectroscopy 211
defect
delineation 33, 70, 217
etch 35, 178
mapping 211
density of states 121, 145, 150
deoxyribonucleic acid 113, 223
depletion 39, 44
deuterium 52, 120, 142
diffusion
coefficient 7, 107, 125, 200, 212
current 184
doping 114, 123, 234
length 6, 100, 191
length mapping 211
limited aggregation 100
dielectric 154, 234, 226
dislocation 1,34
dissolution
chemical 2338, 110
electrochemical 11, 5175
hierarchy 71
oxide 27, 36, 67
valence 54, 57, 94, 111, 189, 200
dissolved oxygen concentration 26, 54, 63
divalent dissolution 46, 54, 57
double-sided electrolytic contact 214
early studies 1
Eden model 100
effective medium 125, 134
electrical serial resistance 208, 236
electrochemical dissolution 5175
electrochemical photocapacitance
spectroscopy 211
electrode 1115, 98
counter 12
geometry 85
reference 12
rotating disk 11, 21, 59, 62
electroluminescence 93, 148, 230
electrolyte 711
acidic 39, 4249, 52
alkaline 39, 49, 52, 81
circulation 21, 33
convection 52, 107, 200
organic 56, 187
electrolytic metal tracer 72, 214
electron injection 46, 49, 54, 6368, 91
electron spin resonance 160
electronic devices 232
electropolishing 56, 74, 93, 203, 221
ellipsometry 49,9 1
epitaxial layer transfer 239
epoxy 21, 28, 241
equivalent circuit 208
etchpit 34, 97, 180
etchrate
aluminum 37
defects 33
nitride 36
oxide 27, 36, 67, 88
porous silicon 106111
silicon 2438, 94
etchstop 50, 6872
exchange splitting 143, 155, 158
exciton 138159
Fabry-Perot filter 228
Ficks first law 200
field current 184
filters 72
interference 222, 226
short-pass 72, 225
first aid 4
flat-band potential 48, 201
fluorine-termination 54
front side photocurrent 212
full isolation by porous oxidized silicon
237
gallium-indium 14
galvanostat 12
galvanostatic characteristic 79, 82, 90
gas sensors 220
generation rate 213
Subject Index 274
Gouy-Chapman model 40
gravimetric measurements 31, 57, 85, 108
hardness 114
haze 35, 85, 218
Helmholtz layer 39, 89, 208
hexafluorosilicate 32, 36, 55, 118
high-frequency-resistometry 43, 209
hillocks 29, 94
hole, definition 40, 97
hole burning spectroscopy 158
hole injection 32, 49, 64, 163, 190
hydrocarbons 25, 87, 117, 119
hydrofluoric acid
properties 711
safety 3, 219
sensor 219
toxic effects 4
hydrogen
bubbles 29, 34, 107, 163, 204, 217
desorption 136
formation 28, 32, 51, 55, 91
overpotential 52
termination 25, 53, 78, 118, 142, 155
hydrophilic 26, 53, 78
hydrophobic 25, 53, 78, 117
illumination 65, 72, 128, 190
backside 191
current-voltage characteristic 4144, 68,
196
inhomogeneous 72, 215
sources 22, 72
image etching 74
indirect bandgap 138
induction period 33
infrared spectroscopy 20, 54, 84, 91, 118
initial state 98
injection level spectroscopy 215
instrumentation 123
interference filter 221
internal surface 104
inversion 39, 44
ionic transport 79
junction
delineation 33
electrolyte-semiconductor 3950
pn 36, 69, 75
kinetics 51, 59
lattice expansion 114
Levich equation 59
lifetime 145, 155, 212, 231
light emitting diode 72, 215, 230
lithography 22, 193, 220241
linear stability analyses 99
luminescent microstructures 165
luminescent silicon 138158, 230
macropores 183205
applications 188, 223239
arrays 192205
calculations 198
conductivity 121
degradation 203
design rules 202
formation mechanisms 97, 183187
growth rate 200
n-type substrates 190205
phenomenology 190
p-type substrates 187
through-pores 203, 223
masking 22, 33, 37, 108, 236
mass transport 79, 198, 204
mesopores 167181
applications 226, 238
conductivity 121
formation mechanisms 97, 167171
highly doped n-type substrates 174
highly doped p-type substrates 171
low doped n-type substrates 177181
morphology 171
metal insulator semiconductor 41, 46, 89,
120
metal plating 35, 51, 75, 217
microelectromechanical systems 23,
219241
micromachining 12, 27, 30, 222241
micropores 127166, 236
applications 220, 226, 230, 232, 236
conductivity 122
formation mechanisms 97, 127
morphology 128
microscopy
atomic force 84
optical 105, 178, 188
scanning electron 171181
scanning tunneling 95
transmission electron 34, 129, 170
minority carrier density 213
minority carrier lifetime 211
mobility 40, 122
molecular recombination model 157
Mott-Schottky relation 210
Subject Index 275
nanoporous 104
native oxide 26, 78, 113, 119
neutrality 186
nitridized porous silicon 161
non-linear optical effects 133
normal hydrogen electrode 13
nucleophilic 53
Ohmic contact 214
Onsager model 124
optical devices 225
optical microbench 21
open circuit potential 42, 46, 49, 209
orientation dependent etching 27, 5054
oscillations 48, 60, 80, 89
Ostwald ripening 192
overall reactions 51
oxide replica 178, 195
oxidized porous silicon 119, 155, 159, 232,
237
oxidizing agents 27, 31
oxygen evolution 83
passivation potential 49, 53, 69
passive state 28, 69, 97
passivity breakdown 100
persulphate 150
photo conductive decay 215
photoconductivity 124
photoelectrochemistry 72, 128, 140
photoluminescence 123, 138158
Auger recombination 145
decay 146
dependence on chemical environ-
ment 141
dependence on formation current 139
dependence on hydrostatic pres-
sure 141
dependence on illumination 140
dependence on magnetic field 141, 176
dependence on temperature 146, 160
excitation spectroscopy 142
green-blue band 147
infrared band 148
models 157
oxidized porous silicon 160
polarisation 140
quenching 123, 147
red band 139
silicon compounds 166
silicon microstructures 165
stability 161, 231
thermostimulation 145
photonic crystals 229
pirhanja-clean 26
pitch 109, 192, 199
Poisson ratio 115
polysilicon 13, 31, 37, 164, 232
pore
arrays 192205, 223239
bottleneck 192, 200, 203
branching 120, 189, 192
definition 97
degradation 174, 203
density 108113
dying 192
facetting 195
filling, liquid 123, 141, 154
filling, solid 189, 235, 238, 240
geometry 98, 167, 195
initiation,nucleation 98, 171, 187
orientation dependence 105, 176180, 197
position 99
spiking 178, 195
tip 168, 185, 195, 201
wall 99, 168, 189, 192
porous oxide 85, 90, 159
porous semiconductors 164
porous silicon 2, 46, 97205
aging effects 117
biocompatibility 223
carrier mobility 125
chemical composition 112, 113, 117
conductivity 121125
critical thickness 115
dielectric constant 125, 154
doping dependence 141
drying 109, 114
electrical properties 120
formation models 97, 99104
growth rates 17, 104, 108
mechanical properties 114
morphology 128, 171, 188, 196
orientation dependence 105, 170, 178,
180, 195
oxidation 119, 155, 159, 232, 237
photoconductivity 124
size regimes 104, 113
specific surface 108113
porosity 108, 113, 198
potential-charge curve 80
potentiostat 12
potentiostatic characteristic 41, 50, 60, 80
precipitates 215
probability analyses 100
pseudo-reference 13, 21, 195
Subject Index 276
pumps 21
punch-through 170
quantum
confinement 102, 127, 150161
dot, wire, well 151, 165
efficiency 65, 148, 165, 228, 231
recombination model 157
radiative recombination 138158
Raman spectroscopy 130, 141
rapid thermal annealing 77
reflection 133, 227
refractive index 126, 134, 227
related materials 162
roughness 25, 30, 84, 95, 107
sacreficial layer 236
safety 3, 219
saturated calomel electrode 13
scanning tunneling microscopy 54, 95
Schottky contact, junction 41, 46, 169, 215
selectivity 36, 68, 70
semiconductors-electrolyte junction 3950
sensors 219
silanol 26, 53, 78, 148
silicon
capacitor 233
fluorides 55, 118
nitride 36, 159, 234
on insulator 238
phase transitions 141
properties 5
surface conditions 24
wafer 13, 16, 72, 216
siloxene 157
size regimes 104
small angle x-ray scattering 133
spin 144
Stern model 40
surface
photovoltage spectroscopy 215
recombination 46, 94, 154, 191, 214
states 94, 123, 157, 208
tension 115
surfactant 11, 117, 188, 201
space charge capacitance 210
space charge region 6, 101, 168, 215
spiking 170,
stain film 31, 75, 162
Stokes shift 142, 153, 156, 166
stress 114, 131, 159
striations 107
surfactant 117
susceptibility 137
Tafel plot 46
tetravalent dissolution 32, 48, 56, 57
thermal
conductivity 115
desorption spectroscopy 87, 114, 120
expansion coefficient 114
oxidation 77
thermionic emission 185
thin films 36
through-pores 224
transient currents 42, 68, 80, 178
transistor 1, 43, 70
transmission 136, 226, 229
tunneling 81, 103, 167180
valence band 3950, 128, 144
Van der Waals forces 117
viscosity 11, 31, 96
voltammogram 59
waveguide 227, 230
wet processing 23
x-ray diffraction 131
x-ray absorption finestructure 133, 152
x-ray photoelectron spectroscopy 78
Youngs modulus 114
Zeeman splitting 141
Subject Index 277
1.1
Early Studies of the Electrochemistry of Silicon
This section briefly surveys the history of the electrochemistry of silicon. Electro-
chemistry is a much older science than the solid-state physics of semiconductors.
Batteries had already been built, by Volta in 1799, when Berzelius first prepared ele-
mentary silicon in 1823 by reducing SiF
4
with potassium. In 1854 Deville prepared
silicon by electrochemical methods. Faraday, who found the exchanged charge and
the deposited matter at electrodes to be proportional, also observed that the resistiv-
ity of certain materials decreased with increasing temperature. However, it took an-
other century for a deeper understanding of the semiconducting state to be devel-
oped, based on the pioneering work of Bethe, Bloch, Braun, Lark-Horovitz, Mott,
Pohl, Schottky, Wilson and many others between 1930 and 1940. Their results final-
ly led to the invention of the transistor by Bardeen, Brattain and Shockley in 1947. In
one of the first papers about the semiconductor-electrolyte junction, by Brattain and
Garrett [Br2], it was already realized that holes control anodic oxidation and that cur-
rent multiplication effects at illuminated n-type electrodes are caused by electron
injection during the electrochemical dissolution process. The first transistor, how-
ever, was made from germanium, because silicon single crystals were not grown un-
til 1950, by crucible pulling. Two years later the float-zone (FZ) method was invented
and the first silicon-based transistor was presented in 1954 by Teal. Since 1961, the
preparation of silicon has involved its transformation into silane, which is then pu-
rified by distillation and adsorption and finally retransformed to elemental silicon by
chemical vapor deposition (CVD). The availability of dislocation-free silicon single
crystals and the idea of an integrated silicon circuit, developed by Kilby in 1958,
were the beginnings of what today is known as the silicon age.
Silicon has long been the subject of numerous electrochemical investigations.
Early electrochemical studies on silicon dealt mainly with problems of anodic oxida-
tion, electropolishing and chemical etching. The first experiments attempting to
grow anodic oxides on silicon were performed by Guentherschulze and Betz as early
as 1937 [Gu1]. Schmidt and Michel carried out a more detailed study in 1957 [Sc1],
leading to a method of local anodic oxidation by the projection of light patterns onto
n-type silicon electrodes [Sc3]. At this time the first etchants for defect delineation
1
1
Introduction, Safety and Instrumentation
Electrochemistry of Silicon: Instrumentation, Science, Materials and Applications.
Volker Lehmann
Copyright 2002 Wiley-VCH Verlag GmbH
ISBNs: 3-527-29321-3 (Hardcover); 3-527-60027-2 (Electronic)
[Da1, Si1] or chemical polishing (CP) [La1] were also developed. Alkaline etching in
sodium hydroxide solution was studied by Seipt [Se1], who observed passivation un-
der anodic bias, and he interpreted this in terms of an insoluble oxide film. Seipt and
others [Bo1, Ef1, Hu1] reported that bias-dependent variations in the capacitance of a
silicon-electrolyte junction could be used to verify the existence of a depletion layer
in the electrode. Pleskov [Pl1] was the first to apply independent electrolyte contacts
to both sides of an n-type germanium disk. He reported hole injection when the elec-
trode reaction involved the cathodic reduction of an oxidizing agent and pointed out
that the amount of collected holes depends on the diffusion length of these holes as
well as on the thickness of the germanium disk. An extension of this method was
used by Harvey [Ha1] to measure the surface recombination velocity of the elec-
trode. A detailed study of the anodic dissolution mechanism of germanium and sil-
icon was carried out by Beck and Gerischer [Be1, Ge1].
The first report of porous silicon (PS) dates back to 1956. In this study, which
dealt mainly with electrolytic shaping of germanium, Uhlir found matte black,
brown or red deposits on anodized silicon samples and tentatively supposed them
to be a suboxide of silicon. He found that smooth etching occurred for higher cur-
rent densities and a dissolution valence of four while at lower current densities
the dissolution occurred under hydrogen evolution at a valence of about two
[Uh1]. Shortly after, Fuller and Ditzenberger reported similar films, which devel-
oped without any applied bias in HF/HNO
3
solutions [Fu3]. Anodically formed
films were studied in more detail by Turner and by Schmidt and Keiper [Tu1,
Sc2], while chemically formed films were investigated by Archer [Ar1]. Turner
found electropolishing to occur above a critical current density, which increased
with HF concentration and temperature, but decreased with viscosity. He inter-
preted this critical value to be a result of mass transfer in the electrolyte. Below
the critical current density he observed a thick film with an orange-red color and
a glassy appearance and speculated that it was a silicon subfluoride. Turner as-
sumed SiO
2
to be present during electropolishing, and observed oscillations of
cell current and potential for current densities above the critical value [Tu2]. In
1960, Gee [Ge2] observed anodic electroluminescence (EL) in different electrolytes
from stain films grown chemically or electrochemically on silicon electrodes. An
excellent review of these early studies is given by Turner [Tu2].
In 1965 Beckmann [Be2] investigated stain films on silicon by means of infrared
(IR) spectroscopy. He found the chemical composition of electrochemically formed
films to be between SiH and SiH
2
, and interpreted this as polymerized silicon
hybrids. In contrast to these findings, films grown chemically in mixtures of HF
and HNO
3
showed high amounts of oxygen. In 1966, Memming and Schwandt
[Me11] presented a dissolution mechanism for silicon electrodes in HF and pro-
posed the resulting films to be a result of redeposition of silicon from SiF
2
. Macro-
pores on n-type substrates and their dependence on crystal orientation were first re-
ported by Theunissen and co-workers in 1970 [Th1]. In that same year, the first stud-
ies on electrochemical etch-stop techniques [Di1] and photoelectrochemical etching
[Da2] of n-type silicon were published. In 1971, Watanabe and Sakai first reported
the porous nature of electrochemically formed films on silicon electrodes [Wa7].
1 Introduction, Safety and Instrumentation 2
The number of publications dealing with the electrochemistry of silicon and PS
has increased rapidly since 1971. The first model for pore formation in n-type sili-
con electrodes, based on a breakdown of the depletion layer, was proposed by
Theunissen in 1972 [Th2]. In 1988, it was shown that macropores could be etched
in arbitrary patterns using a pre-structured n-type silicon electrode [Le11]. Using
electron diffraction, Arita and Sunohara proved in 1977 [Ar2] that PS on silicon
electrodes, independent of their doping, is single crystalline with the same orien-
tation as the substrate. This allowed them to conclude that localized dissolution
generates pores in the electrode and the remaining substrate forms the PS. Bom-
chil et al. demonstrated in 1983 [Bo2] using gas absorption that the pore diame-
ters in PS may be as small as 2 nm.
The conversion of PS to SiO
2
by thermal oxidation was reported in 1971 by
Watanabe and Sakai [Wa7]. Arita in 1978 [Ar3] and Unagami in 1980 [Un1] per-
formed thermal oxidation experiments on PS, which a few years later led to an sil-
icon-on-insulator (SOI) technology based on oxidized PS [Ho1, Im1]. Another
approach to manufacturing SOI structures was developed in 1986 by Lin and co-
workers [Li1], by growing a Si molecular beam epitaxy (MBE) film on PS and sub-
sequent oxidation. However, a major drawback of PS-based SOI technologies is
the need for windows in the Si film to carry out the oxidation of the underlying
PS.
Pickering and co-workers observed visible photoluminescence (PL) from PS at
4.2 K in 1984 [Pi1], which they interpreted as due to a complex mixture of amor-
phous phases. The questions of why PS is transparent for visible light and why it
is photoluminescent remained unanswered until 199091 when a quantum size
effect was proposed as an explanation [Ca1, Le1]. Two years later PL was also
found for oxidized PS [Le15, It2]. These astonishing optical properties of PS in-
itiated vigorous research and resulted in more than a thousand publications, as
well as several books and reviews [Cu2, Th7].
1.2
Safety First
It is very important to first consider the safety aspects of electrochemical experi-
ments with silicon. The single most dangerous compound, which cannot be avoided
in the electrochemistry of silicon, is hydrofluoric acid (HF). HF in its anhydrous
form and in concentrated aqueous solutions is highly corrosive towards living tis-
sue. Inhalation, ingestion or skin contact with HF are all extremely hazardous.
For aqueous HF of high concentrations (> 10%) or for elevated temperatures
(above room temperature, RT), HF is in its most dangerous phase, the vapor
phase. The liquid solution generates considerable amounts of HF vapor. HF has a
stinging smell not unlike hydrochloric acid, and its smell is detectable at levels
above 0.04 ppm. The permissible exposure limits in industrial countries vary from
0.5 to 2 mg m
3
(0.622.50 ppm). At 30 ppm, HF is immediately dangerous to life
and health. An exposure to 50 ppm for 30 min can be lethal. It is self-evident that
1.2 Safety First 3
HF should only be handled under a hood with proper ventilation. However, even
if all safety regulations are obeyed the risk of accidents cannot be totally elimi-
nated. In case of large spills, the contaminated laboratory area should be evacu-
ated immediately because of the danger of inhalation.
If HF vapor is inhaled, a corticosteroid aerosol and inhalation of pure oxygen are
recommended as first aid, because they relieve inflammatory reactions such as pul-
monary edema or hypersecretion of mucus in bronchial tubes and help to prevent
bronchospasm. A b-mimetic aerosol can be given to control apparent bronchospasm.
For concentrations below 10%, the evaporation of HF is reduced and direct con-
tact with the liquid becomes the greatest risk. If HF is swallowed, it is advisable
to drink lots of water, if possible with activated carbon added, in order to dilute
the acid. Small amounts in the eye can cause intense irritation of the eyelids and
slow ulceration of the conjunctivae. Large amounts in the eye cause immediate
blindness. As first aid treatment the eyes should be irrigated immediately and co-
piously with clean water for a minimum of 15 min. Immediate medical care is
mandatory after all the accidents mentioned above, even if no symptoms are ap-
parent, because respiratory problems or other symptoms of poisoning can be de-
layed for hours after the incident has occurred.
In contact with skin, HF causes burns that show a progressive necrosis, often
resulting in permanent tissue loss. The dissociation of HF yields H
+
ions, which
exhaust the buffering capacity of the tissue, and F
, PF
6
, CF
3
SO
3
, AsF
6
and SbF
6
and
various hydrofluoric species such as HF
2
and (HF)
2
according to the reactions:
HF = H
with K
1
=([H
+
][F
])/[HF] (1.13)
HF F
= HF
2
with K
2
=[HF
2
]/([HF][F
] (1.14)
2HF = (HF)
2
with K
3
=[(HF)
2
]/([HF][HF]) (1.15)
These equilibrium constants vary with molarity of the HF solution. Measured
values corrected for zero ionic strength at 258C are K
1
=6.7110
4
mol l
1
,
K
2
=3.86 l mol
1
, and K
3
=2.7 l mol
1
[Br10, Iu1, Wa11], implying a dissociation of
only a few percent. This unusual behavior is still controversial and has been at-
tributed to the greater strength of the HF bond compared to the other hydrogen
halides [Pa1], to the presence of the dimer (HF)
2
[Wa1], or to polymers that may
1.4 Common Electrolytes 9
form ring structures [Hy1]. Spectroscopic measurements indicate the formation of
a hydrogen-bonded ion-pair H
3
O
+
F
, (HF)
3
F
and (HF)
4
F
and F
are shown
as a function of c
HF
in Fig. 1.2. For unbuffered HF of concentrations above
0.25 mol l
1
the composition is roughly constant, as follows: 90% HF and (HF)
2
,
4% HF
2
and 2% F
. The pH=log([H
+
]) of the solution can be calculated for a
known HF concentration [c
HF
] =[HF] + 2[(HF)
2
] + 2[HF
2
] + [F
[ = [HF
2
[ [F
[ (1:16)
Note that the exact concentrations of the species H
+
, HF, HF
2
and F
in dilute
HF solutions of a certain molarity can be obtained from K
1
and K
2
only with
some function to represent the activity coefficients [Ha13]. Equilibrium constants,
obtained from measurements of pH, differ slightly from the values given above
(K
1
=7.710
4
mol l
1
, K
2
=5.59 l mol
1
[Se2]). The pH for different HF concentra-
tions, as given in the table shown in the inner back cover of this book, has been
calculated using the latter constants. Note that the dissociation of HF is further re-
duced by addition of ethanol to the solution [Ga3].
For the case of SiO
2
etching, HF, (HF)
2
and HF
2
ion becomes negligible, which leaves HF and its polymers to be the active species
[Ve3]. Because for high current densities the electrochemical dissolution of silicon
occurs via a thin anodic oxide layer it can be concluded that, at least for this re-
gime, the same species are active. This is supported by the observation that F
is
1 Introduction, Safety and Instrumentation 10
Fig. 1.2 The calculated fraction of each component in an aqueous HF solution as a function of
pH for a fixed total fluoride concentration of 7.5 mol l
1
. Redrawn from [Ve1].
inactive in the electrochemical dissolution kinetics [Se2]. It is found that HF
2
is
also inactive for the pH range investigated which again leaves HF and its poly-
mers to be the active species in the electrochemical dissolution reaction of silicon
[Se2].
The diffusion coefficient D
HF
of the HF molecule has been determined from
the etch rate on oxide films to be between 210
6
and 210
5
cm
2
s
1
[Mo5]. A
conductimetric technique gave D
HF
=1.6810
5
(0.02) cm
2
s
1
for concentrations
of aqueous HF in the order of 0.2% at 258C [No1]. In ethanoic solutions values in
the order of 2.210
5
cm
2
s
1
have been reported [Me14]. Due to ionization of
neutral HF molecules to F
(1.47310
5
cm
2
s
1
), HF
2
(2.3510
5
cm
2
s
1
) and
highly mobile H
+
(9.31510
5
cm
2
s
1
), the measured average diffusion coefficient
increases rapidly for HF concentrations below 0.02%. The water molecule, by
comparison, has a diffusion coefficient of 210
5
cm
2
s
1
at RT. The viscosity of
solutions is dependent on temperature, which produces a temperature depen-
dence of D
HF
. From measurements of viscosity versus temperature, activation en-
ergies of 0.16 and 0.12 eV have been calculated for diffusion-controlled reactions
in water and ethanol, respectively. These results are supported by rotating disk
electrode (RDE) measurements of J
PS
in ethanoic HF, which gave an activation
energy of 0.125 eV for D
HF
[Me14].
The product of the dissolution process of silicon electrodes in HF is fluosilicic
acid, H
2
SiF
6
. In contrast to HF, H
2
SiF
6
is mostly (75%) dissociated into SiF
6
2
and
2H
+
in aqueous solution at RT. The diffusion coefficient of the SiF
6
2
at RT de-
creases from 1.210
5
cm
2
s
1
for 0.83 mol l
1
to 0.45 cm
2
s
1
for 2.5 mol l
1
, with
values of activation energy around 0.2 eV [We7].
So far only aqueous solutions have been considered; however, mixtures of HF
and ethanol or methanol are quite common, because this addition reduces the
surface tension and thereby the sticking probability of hydrogen bubbles. While
substantial quantities of ethanol or methanol are needed to reduce the surface
tension, cationic or anionic surfactants fulfill the same purpose in concentrations
as low as 0.01 M [So3, Ch16].
If aluminum is present on the electrode (for example if used for interconnects),
an ammonium fluoride-based electrolyte is more desirable than HF, because Al is
only stable in the pH range of about 4 to 8.5 [Oh4]. Note that PS formation is ob-
served in ammonium fluoride-based electrolytes [Ku5], as well as in water-free
mixtures of acetonitrile and HF [Ri1, Pr7], but not in alkaline electrolytes.
1.5
The Electrodes
This section deals with the electrodes in the electrochemical set-up, with special
emphasis on the silicon electrode and its semiconducting character. An
electrochemical cell with its complete electrical connections, as shown in Fig. 1.3
a and b, is similar to the well-known four-point probe used for applying a defined
bias to a solid-state device. The two lines that supply the current are connected to
1.5 The Electrodes 11
the counter electrode and the working electrode, which is the silicon sample. One
probe contact is connected to the sample and the other to a reference electrode
that is placed close to the silicon surface in the electrolyte. By measuring the po-
tentials at the probe contacts, all potential drops caused by ohmic losses in the
counter electrode, the connections and the electrolyte are eliminated. The stan-
dard, four-terminal power supplies for electrochemical experiments are potentio-
stats, which are commercially available from various vendors [4]. It enables work-
ing under constant bias (potentiostatically, Fig. 1.3a) or under constant current
(galvanostatically, Fig. 1.3b). By a scan of the potential (current) the potentiostatic
(galvanostatic) IV characteristics of the electrode can be recorded.
The counter electrode is commonly realized by a platinum mesh or sheet, a car-
bon plate or a highly doped silicon wafer. The position and geometry of the coun-
ter electrode is of great importance for the resulting etched geometry. A pin-like
electrode, for example, can even be used to micromachine the working electrode
if short current pulses are applied [Sc19]. A homogeneous current distribution, as
desired in most applications, is best achieved by using a counter electrode of the
same size and in-plane orientation as the working electrode. If a Pt mesh is used
its total surface area must be comparable to that of the Si electrode and the size
of the mesh openings smaller than the distance to the Si electrode. Platinum
black coating of the mesh can reduce the required mesh area.
The measurement of potentials in electrolytes is not as easy as it is for solid-
state devices. Depending on the composition of the electrolyte and the electrode
material a monolayer of adsorbates or a thin passivation layer may be formed on
the electrode, and can significantly shift the electrode potential. These effects have
to be taken into account for the working as well as for the counter electrode. The
potential at the latter becomes irrelevant if a reference electrode is used. The refer-
ence electrode should be placed as close as possible to the Si electrode or it can
access the Si electrode via a capillary. The size of the reference electrode is not rel-
1 Introduction, Safety and Instrumentation 12
Working
electrode
Reference
electrode
Working
electrode
Reference
electrode
Fig. 1.3 An electrochemical cell with all its electrical connections in (a) potentiostatic mode
and (b) galvanostatic mode. The dashed meters are optional.
evant, because it carries no significant current. The internationally accepted pri-
mary reference is the normal hydrogen electrode (NHE), which consists of a Pt
electrode in a stream of hydrogen bubbles at 1 atm in a solution of unit hydrogen
ion activity. The most common reference is the saturated calomel electrode (SCE)
with a potential of 0.242 V versus NHE [5].
For a metal electrode, as a working electrode, its resistivity is in most cases neg-
ligible compared to the electrolyte resistivity and space charges in the electrode do
not to be taken into account due to a high number of free charge carriers. This is
in stark contrast to a semiconducting electrode where the number of free carriers
is orders of magnitude smaller than in a metal. Ohmic losses in the electrode
have therefore to be taken into account, especially for low doped substrates. Un-
der reverse bias, in addition, a significant part of the applied bias may drop across
an SCR. Another source for a potential difference is a surface passive film, for ex-
ample SiO
2
, which may be present in the anodic regime. The correct determina-
tion of potential distribution at the interface of a silicon electrode is therefore
complicated, even if a reference electrode is used. Fortunately, it is found that the
electrochemical condition of the silicon electrode is in most cases well described
by the current density across the interface. This quality of the current density is
probably due to the fact that a certain bias across the interface between electrolyte
and semiconductor surface corresponds to a certain current density under con-
stant values of temperature, doping density and electrolyte concentration. In the
few cases where the applied bias is a parameter independent of the current den-
sity, as in the case of macropore formation on n-type electrodes, larger potentials
are involved and a platinum wire as a pseudo-reference is usually found to be suf-
ficient.
Because the silicon working electrode is the focus of study in this book, the de-
tails of its preparation from a wafer are worth discussing. Silicon is commercially
available as a single-crystalline wafer in diameters of 100, 125, 150, 200 and
300 mm, or even larger [6]. The thickness is usually in the range of 0.40.7 mm.
The crystal orientation of the majority of wafer used today is (100). For certain ap-
plications (111) and (110) oriented wafers are available in diameters up to
150 mm. For commercial wafers, n-type doping is realized by P, As or Sb, while
for p-type doping only boron is used. Bulk dopant concentrations usually range
from 10
13
to 10
20
cm
3
, which corresponds to specific resistivities between 1000
and 0.001 X cm, according to the figure in the inner front cover of this book. Dop-
ing and crystal orientation of wafers below 200 mm diameter are marked by a pri-
mary and sometimes by a smaller secondary flat, as shown in Fig. 1.4. The crystal
orientation of 200 and 300 mm wafers is marked by an edge-notch. For many
electrochemical experiments the desirable sample size is considerably smaller
than the whole wafer. Single crystalline wafers can easily be cleaved along the
(110) planes using a diamond tip scribe [7] to scratch the wafer and two pairs of
tweezers to bend and cleave it, as shown in Fig. 1.4. Polycrystalline substrates are
mainly fabricated for solar energy conversion and have a square or rectangular
shape. Such wafers are difficult to cleave and dicing by a diamond saw is recom-
mended.
1.5 The Electrodes 13
Most electrochemical experiments need an electrical contact of some kind to the
silicon substrate. Because of the semiconducting nature of silicon a metallic tip or
clip attached to the surface will not produce an ohmic contact but constitutes a
Schottky junction. However, for some applications, like the ELYMAT (Section
10.3), where the contact is only operated under forward conditions at low current
densities, such a contact is sufficient. For silicon samples with a doping concen-
tration in excess of 10
19
cm
3
the contact to a metal becomes ohmic. An ohmic
contact to a silicon sample with a doping concentration below 10
19
cm
3
can be
achieved in different ways:
1. Rubbing GaIn eutectic (24% In, 76% Ga) with a piece of fine grinding paper
on the backside of the sample.
2. Deposition of a metal film on the backside by evaporation or CVD techniques
and subsequent annealing.
3. Doping the back of the sample in excess of 10
19
cm
3
in order to produce an
nn
+
or a pp
+
structure.
4. Providing the backside with an electrolyte contact, which is forward biased or
alternatively reverse biased and intensely illuminated.
Method 1 is quick but contaminates the samples with Ga and In, which usually
prohibits further processing with standard semiconductor equipment. The same
1 Introduction, Safety and Instrumentation 14
Fig. 1.4 The three common orientations of
single crystalline silicon wafers are indicated
by flats, while for wafers with diameters of
200 mm and larger the orientation is indi-
cated by a small notch. Cleaving along the
lines, as indicated for each wafer orientation,
can be performed by scratching the wafer at
the edge with a diamond tip and bending it
with tweezers, as shown on the lower left.
is basically true for method 2. Methods 3 and 4 circumvent this problem and in
addition provide a transparent contact that is advantageous for experiments with
backside illumination. For methods 1 to 3 a metallic clip or an aluminum foil are
sufficient to connect the conducting backside of the sample to the terminal of the
power supply. Note that the contacts realized by method 1 are not stable if ex-
posed to HF vapor.
The geometry of the contact and the electrode area exposed to the electrolyte
are crucial because ohmic losses, which may be significant for low or moderately
doped silicon electrodes, lead to potential gradients. A consequence of these gradi-
ents is an inhomogeneous distribution of current density. Silicon samples are
commonly platelets cleaved from a wafer. A current flowing parallel to the surface
of the plate, as, for example, in immersion cell designs, leads to an inhomoge-
neous current density. A large contact on the backside of the sample and a cur-
rent flow normal to the surface, as is the case for O-ring cells, produces a homo-
geneous current density at the front side, exposed to the electrolyte. For some ap-
plications, like the etching of trough-holes, current density gradients are desirable
[Le20]. In such cases small, local backside contacts, as depicted in Fig. 4.17a, can
be used.
1.6
Cell Designs
The design of an electrochemical set-up is governed by a multitude of different re-
quirements such as different sample sizes, low contact resistance, high current
density, laminar flow of electrolyte, easy observation of the sample surface, adjust-
able illumination conditions, etc. This multitude of requirements is in stark con-
trast to the small number of cells specialized on silicon electrochemistry commer-
cially available [8]. Doing electrochemistry with silicon therefore commonly begins
with designing a suitable cell, which is usually a source of frustration as a result
of broken samples, leaky set-ups and corroded contacts. The different cell designs
discussed in this section may give some inspiration as to how such problems can
be avoided. Emphasis is put on critical points such as materials, sealants, contacts
and easy handling.
The properties of different illumination sources that can be combined with the
set-ups discussed above are presented in Section 4.6.
1.6.1
The Cell Materials
Which materials are best for cell design depends essentially on the type of electro-
lyte used. Because HF acid is quite common in the electrochemistry of silicon,
materials resistant to HF are preferable. Polyvinyl chloride (PVC), polypropylene
(PP), polytetrafluoroethylene (PTFE) and polyvinylidene fluoride (PVDF) can be
used for the cell body. PVC is a good choice for most designs because it is inex-
1.6 Cell Designs 15
pensive, inert in HF, and its mechanical performance is superior to that of PTFE.
In addition PVC parts can easily be glued, which is not the case for PTFE. Note
that standard plastic screws are made of polyamide, which is not resistant to con-
centrated HF. They should be replaced by PP or PVC screws.
Materials used for transparent windows are clear PVC, Plexiglas (polymethyl-
methacrylate, PMMA) and sapphire. PMMA shows a good transparency in the
visible and the IR, it is easily machinable, and stable at low HF concentrations. In
concentrated HF (>10%), however, it becomes opaque after the initial contact.
Clear PVC, which is of lower transmission coefficient than PMMA, is therefore
preferable for high HF concentrations.
Standard black O-rings made of an acrylonitrile-butadiene copolymer (such as
Perbunan) have proved to be stable in HF at concentrations up to 50%. If contam-
ination of the silicon sample is an issue, the nitrile O-rings may be replaced by vi-
nylidene fluoride-hexafluoropropylene (Viton) O-rings [9].
It is important to provide a good ohmic contact to a semiconductor like silicon.
The ohmic contact is especially critical for open cell designs, like the immersion
cell, because it is exposed to HF vapors from the electrolyte, which are corrosive.
Platinum or gold are inert contact materials with respect to HF, however some kind
of spring or clip is needed to press the noble metal to the sample. Metals commonly
used to make springs, like stainless steel or brass, are found to corrode rapidly in HF
vapor. Tungsten shows a better performance, but a more elegant way to solve this
problem is to use the elastic properties of the sample holder material itself to effect
a non-metallic clip, as shown in Fig. 1.5a. For sample contacts that are not directly
exposed to HF, other metals, e.g. aluminum, can be used. Such contacts, however,
should be easily exchangeable in case of corrosion. For moderate currents, needle
tip contacts at the edge of the wafer are useful such contacts are commercially
available as spring contact probes for electronic testing of PCBs [10].
1 Introduction, Safety and Instrumentation 16
Fig. 1.5 (a) Sample fixture for an immersion
cell. The elasticity of the PVC body is used to
clamp the Pt wire contact to the wafer.
(b) Standard wafer container used as an im-
mersion cell for anodic oxidation experiments.
1.6.2
The Immersion Cell
The simplest way to realize a basic electrochemical cell is to partially immerse a
strip-shaped silicon sample (working electrode) and a platinum wire (counter elec-
trode) in a beaker filled with electrolyte. If a power supply is connected to the elec-
trodes, the cell is ready for operation. This simple set-up has several advantages. It is
a clean way of sample preparation, because the sample is not in contact with an O-
ring and the area contaminated by the contact can easily be removed by cleaving it
off. This is advantageous if subsequent high-temperature processing of the sample
is desirable. The flexibility of immersion cell designs is shown in Fig. 1.6.
An inevitable property of this cell concept is a current flow along the strip. This
causes an inhomogeneous potential distribution along the stripe due to ohmic
losses, especially for low doped substrates. Porous layers, as a result, often show a
thickness gradient along the stripe. The potential drop along the strip can usually
be neglected for silicon samples of a sufficiently high conductivity or for small ano-
dization currents. If, however, the transformation of the whole thickness of a strip
into mesoporous silicon is desired, a slight beveling of the strip or an immersion
scanning technique is required, even in the case of highly doped silicon [Ba4, Ju2].
Another drawback of the immersion cell concept is that the active area is badly
defined, because of the meniscus formed at the electrolyte-air interface. The form
of the meniscus greatly depends on whether the sample is hydrophilic or hydro-
phobic, which again is a function of applied potential. This problem can be cir-
cumvented, if the active area of the sample is defined by a window in an inert
layer, for example resist or CVD nitride, which is fully immersed into the electro-
lyte, as shown in Fig. 1.6a.
1.6 Cell Designs 17
Fig. 1.6 Two different immersion cell designs
optimized for special applications. (a) Set-up
for fast removal and rinsing of a strip-shaped
electrode by fast rotation of the shaft (solid
arrow). This set-up is useful for measure-
ments of transient electrode processes like
anodic oxide growth during electrochemical
oscillations. (b) PMMA immersion cell set-up
for in situ determination of stress by optical
measurement of the electrode curvature.
Stress is induced by the growth of anodic
films. After [Le4].
Counter
Electrode
The simple immersion cell design is most suitable for applications for which
the current densities involved are very low, such as anodic oxidation. In this case
ohmic losses in the substrate become negligible, even for moderate doping densi-
ties and large samples, like whole wafers. Position and geometry of the counter
electrode, however, become important, because the oxide thickness is sensitive to
spacing of the electrodes. Large counter cathodes of the same size and shape as
the oxidized wafer can be realized by two highly doped wafers, which avoids ex-
pensive platinum sheets or meshes. The slits of a standard wafer container pro-
vides an easy way of positioning of the wafers sufficiently accurately to produce
homogeneous anodic oxides. Holes in the top of the container allow for contact-
ing. Figure 1.5b shows such a simple set-up, where a wafer container is used as
an anodization cell. A more sophisticated cell for anodic oxide formation is de-
scribed in the literature [Ba13].
If the wafer is not fixed in the cell, a mechanical wafer support is advisable.
The ohmic contact can be an integral part of such a sample fixture, as shown in
Fig. 1.5a. During formation of mesoporous silicon on highly doped substrates at
low bias (01.5 V), it was found that such a contact can even be immersed into
the electrolyte without a significant degradation of its electrical properties. It is re-
markable that mesoporous silicon formation takes place under the contact, too,
without significant degradation of the contact properties.
1.6.3
The O-Ring Cell
The immersion cell design discussed above can be achieved with standard labora-
tory equipment within a few minutes; however, it suffers from an inhomogeneous
current density along the strip and the badly defined active area of the sample.
These drawbacks are overcome if the sample area exposed to the electrolyte is de-
fined by an O-ring seal. The backside of the sample is accessible in this case and
can be used to realize an ohmic contact. Now the current flows normal to the
sample surface, which reduces ohmic losses significantly and leads to a homoge-
neous current density distribution. However, at a distance of about the wafer
thickness from the O-ring the current flow is not normal to the surface and the
current density is therefore slightly enhanced there. This effect has been found to
be responsible for thickness inhomogeneities of porous layers [Kr3]. To reduce
such inhomogeneities the O-ring should be of a diameter in excess of a centi-
meter and its section thickness as small as possible.
The sample has to be pressed against the O-ring in order to seal the cell. This
can be done by various means, as shown in Fig. 1.7 af. Simple fixtures use the
weight of the upper part of the cell or screws to press the O-ring against the sili-
con sample. A fixture using magnets is advantageous if fast handling is required
[Ch14]. For whole wafers pneumatic pistons or a vacuum seal [Ba13] are prefer-
able. Note that the use of a vacuum chuck, as shown in Fig. 1.7e, requires sup-
port of the wafer backside by a chuck to reduce its bow to values below 0.5 mm,
in order to avoid fracture. An advantage of the set-ups shown in Fig. 1.7af is that
1 Introduction, Safety and Instrumentation 18
they can be filled like a beaker and allow for easy observation of the sample from
the top without a window. The option of electrolyte agitation, however, is limited.
The upper right figure on the front cover of this book shows the top view of a
simple O-ring cell according to Fig. 1.7b.
1.6.4
The Double Cell
An advantage of the electrochemical double cell is the possibility of replacing the
ohmic contact, usually established by some kind of metal pressed against the wa-
fer, by an electrolytic contact. This not only avoids a potential source of contami-
nation but also establishes a transparent contact. A disadvantage of this arrange-
ment is that the potential of the wafer is not known. The double cell shown in
Fig. 1.8 combines the simplicity of the immersion cell with the homogeneous cur-
rent distribution of the O-ring cell [La5, La9]. The two cells are separated by the
wafer carrier, which has to fit so tightly into the set-up that leakage currents be-
come negligible compared to the current across the wafer. O-rings can be used if
leakage currents are not acceptable, as is the case for anodic oxidation. Tight seal-
ing of double O-ring cells for whole wafers requires pneumatic elements or an
evacuable recess as shown in Fig. 1.7 d and f [Ba13]. Such a double cell designed
for ELYMAT measurements of 200 mm wafers with a sophisticated sample holder
is shown on the upper left of the front cover of this book. The electrical contacts
1.6 Cell Designs 19
Fig. 1.7 Cross-sectional views of various
types of O-ring cells. The O-ring can be
pressed against the sample (a) by the weight
of the upper part of the cell, (b) by screws,
(c) by magnets, (d, e) by vacuum or (f ) by
pneumatic pistons. These designs can be ex-
tended to double O-ring cells: this requires (g)
a vertical sample position or (h) a closed cell.
of this set-up are realized by pneumatic tungsten carbide needles outside the O-
ring, close to the wafer edge.
In order to produce significant currents across moderately doped wafers the re-
verse biased junction has to be illuminated. Hence the anode (for the case of p-
type substrates) or the cathode (for the case of n-type substrates) should be made
of a platinum mesh to be sufficiently transparent.
A special O-ring cell design is needed for in situ infrared (IR) vibrational charac-
terization of an electrochemical interface. The absorption of one monolayer (i.e.
<10
15
cm
2
vibrators) can be measured if the silicon electrode is shaped as an
attenuated total reflection (ATR) prism, which allows for working in a multiple-in-
ternal-reflection geometry. A set-up as shown in Fig. 1.9 enhances the vibrational
signal proportional to the number of reflections and restricts the equivalent path
in the electrolyte to a value close to the product of the number of reflections by
the penetration depth of the IR radiation in the electrolyte, which is typically a
tenth of the wavelength. The best compromise in terms of sensitivity often leads
to about ten reflections [Oz2].
1 Introduction, Safety and Instrumentation 20
Fig. 1.8 Immersion double cell separated by
the fixture of the silicon electrode. Note that
no ohmic contact to the silicon wafer is
necessary. Illumination is needed for moder-
ately doped samples, to generate a current in
the reversely biased junction. After [La5].
Fig. 1.9 For in situ IR vibrational characterization of an electrochemical interface
the silicon electrode in the double O-ring cell has to be shaped as an ATR prism.
Electrolyte
1.6.5
The Rotating Disk Electrode
None of the set-ups discussed so far provides stirring of the electrolyte for bubble
removal or for enhancement of the reaction rates. A standard set-up developed to
study kinetic electrode processes is the rotating disc electrode [11]. The electrode
is a small flat disc set in a vertical axle. The hydrodynamic flow pattern at the disc
depends on rotation speed and can be calculated. An additional ring electrode set
at a different potential provides information about reaction products such as, for
example, hydrogen. However, because this set-up is designed to study kinetic pro-
cesses and is usually equipped with a platinum disc, it becomes inconvenient if
silicon samples of different geometries have to be mounted.
1.6.6
Cells with Electrolyte Circulation
Circulation of the electrolyte is essential for many experiments, because it reduces
concentration gradients at the electrode surfaces. It can remove bubbles from the
electrode and it allows for better temperature control. A magnetic or mechanical stir-
rer can be integrated in open cell designs, as shown in Fig. 1.7ag. However, high
flow rates can only be obtained with closed cell designs, as shown in Fig. 1.7h. Poly-
tetrafluoroethylene membrane pumps [12] and non-metallic valves [13], as com-
monly used for pumping of HF in wafer fabs, are sufficient to provide good circula-
tion. Peristaltic pumps are not advisable because of their relatively low flow rates. In
order to produce a homogeneous flow of electrolyte from the intermittent pumping
action a partly air-filled reservoir is added for damping. A second reservoir at ambi-
ent pressure serves as a container for the reflow from the cell and for refilling the
set-up. Note that special safety regulations apply to pumping HF.
A closed O-ring double cell with in- and outlets for electrolyte circulation, as
shown in Fig. 1.10, is quite complex, but it shows superior experimental flexibility.
This set-up, which combines several features such as a window for illumination,
an optional second electrolyte contact on the back of the sample, and the possibili-
ty to enhance the electrolyte convection by pumping, will be described briefly. The
electrochemical double cell shown in Fig. 1.10 is based on a commercially avail-
able optical microbench system [14]. It allows for a maximum active sample area
of 260 mm
2
, if a mounting plate with a 35.5 mm opening is used. The front side
and the backside cells are symmetric and consist of a PVC body, shown on the
left of Fig. 1.10. The windows are easily exchangeable using the window screw.
An electrolyte inlet hole is located at the bottom of the cell body, the outlet is on
the top. A counter electrode and a pseudoreference electrode are realized by plati-
num wires that are electrolyte-tight fed through the PVC body using an epoxy re-
sin. The width of the rectangular-shaped sample is given by the distance of the
rods (23 mm), whereas the length is usually somewhat larger in order to have en-
ough space for contacting. The sample is introduced between the two cell bodies,
which are pressed together by four springs. Illumination sources such as a halo-
1.6 Cell Designs 21
gen lamp, LEDs or lasers, as well as filters and other optical parts, can easily be
fixed on the optical bench allowing for a high flexibility of front side or backside
illumination conditions.
1.6.7
Lithographic Patterning
All the cell designs discussed so far show active working electrode areas in the or-
der of cm
2
. If much smaller active areas are desirable, photolithographic pattern-
ing of a thin surface film, e.g. resin, silicon oxide or silicon nitride is required
[Na6, Kr3, Du6]. With this technique electrode areas in the order of lm
2
and be-
low can be achieved. However, some problems, known from the O-ring cell, also
apply to patterned electrodes. On the one hand the current distribution becomes
inhomogeneous for area diameters in the order of the sample thickness and be-
low, especially for low-doped substrates. This may, for example, produce inhomo-
geneities of PS layer thicknesses. On the other hand, undercutting of the pattern
cannot be avoided. In HF electrolytes the undercutting is isotropic for oxide or ni-
tride masking, while resin patterns show large undercutting which eventually
leads to lift-off. Figure 6.6 shows PS layer profiles obtained for different sub-
strates and masking layers.
1 Introduction, Safety and Instrumentation 22
Fig. 1.10 A closed double O-ring cell for elec-
trochemical experiments with silicon electro-
des, based on a standard optical microbench
system. Top and side views of a PVC half-cell
are shown on the left. Two identical PVC half-
cells are mounted on the four rods of the mi-
crobench such that the front cell is moveable
for sample exchange (right side).
2.1
The Basics of Wet Processing of Silicon
Chemical dissolution of silicon can be performed in liquid as well as gaseous me-
dia. The latter is known as dry etching or reactive ion etching (RIE) [ Ja3] and is
an irreplaceable technique in todays microelectronic manufacturing. However,
about 30% of the total number of process steps for the fabrication of todays inte-
grated circuits are still wet etching and cleaning procedures, illustrating the im-
portance of wet processing.
To treat all the different wet processes for silicon wafers developed in the last
five decades exhaustively would make up a book of its own. However, a few basic
aspects are important, because chemical etching of silicon is closely related to the
electrochemical behavior of Si electrodes, especially to the OCP condition. A brief
overview of the most common etching and cleaning solutions will be given, with
emphasis on the electrochemical aspects.
In the early days of silicon device manufacturing the need for surfaces with a
low defect density led to the development of CP solutions. Defect etchants were
developed at the same time in order to study the crystal quality for different crys-
tal growth processes. The improvement of the growth methods and the introduc-
tion of chemo-mechanical polishing methods led to defect-free single crystals with
optically flat surfaces of superior electronic properties. This reduced the interest
in CP and defect delineation.
Cleaning and the control of surface passivation then became a major issue, be-
cause traces of heavy metals in concentrations of less than a thousandth of a
monolayer on the surface of a silicon wafer are sufficient to degrade device perfor-
mance.
The high selectivity of wet etchants for different materials, e.g. Al, Si, SiO
2
and
Si
3
N
4
, is indispensable in semiconductor manufacturing today. The combination
of photolithographic patterning and anisotropic as well as isotropic etching of sili-
con led to a multitude of applications in the fabrication of microelectromechanical
systems (MEMS).
In the following sections the wet treatments most common in the manufacture
of silicon devices will be presented according to their main application:
23
2
The Chemical Dissolution of Silicon
Electrochemistry of Silicon: Instrumentation, Science, Materials and Applications.
Volker Lehmann
Copyright 2002 Wiley-VCH Verlag GmbH
ISBNs: 3-527-29321-3 (Hardcover); 3-527-60027-2 (Electronic)
1. Cleaning and passivation of silicon surfaces.
2. Silicon removal in an anisotropic manner.
3. Silicon removal in an isotropic manner.
4. Defect and junction delineation.
5. Selective etching of layers of different chemical composition.
2.2
Silicon Surface Conditions and Cleaning Procedures
The surface condition of a silicon crystal depends on the way the surface was pre-
pared. Only a silicon crystal that is cleaved in ultra high vacuum (UHV) exhibits a
surface free of other elements. However, on an atomistic scale this surface does
not look like the surface of a diamond lattice as we might expect from macroscopic
models. If such simple surfaces existed, each surface silicon atom would carry one or
two free bonds. This high density of free bonds corresponds to a high surface energy
and the surface relaxes to a thermodynamically more favorable state. Therefore, the
surface of a real silicon crystal is either free of other elements but reconstructed, or a
perfect crystal plane but passivated with other elements. The first case can be studied
for silicon crystals cleaved in UHV [Sc4], while unreconstructed silicon (100) [Pi2,
Ar5, Th9] or (111) [Hi9, Ha2, Bi5] surfaces have so far only been reported for a ter-
mination of surface bonds by hydrogen.
Under ambient atmospheric conditions a native oxide is formed on cleaved Si
surfaces. The properties of native and chemical oxides are discussed in Section
5.2. The well-defined surface conditions produced by wet processes like rinsing
and cleaning procedures will be discussed below.
All standard cleaning processes for silicon wafers are performed in water-based
solutions, with the exception of acetone or (isopropyl alcohol, IPA) treatments,
which are mainly used to remove resist or other organic contaminants. The most
common cleaning procedure for silicon wafers in electronic device manufacturing
is the deionized (DI) water rinse. This and other common cleaning solutions for
silicon, such as the SC1, the SC2 [Ke1], the SPM [Ko7] and the HF dip do remove
silicon from the wafer surface, but at very low rates. The etch rate of a cleaning
solution is usually well below 1 nm min
1
.
Chemomechanical polishing (CMP) solutions [16] for Si show somewhat higher
etch rates than cleaning solutions, as shown in Fig. 2.1a. CMP has been used since
the late 1960s to prepare smooth, defect-free silicon wafer surfaces of optical quality
[Me3]. CMP is based on the combined mechanical grinding action and chemical
etching action of an alkaline suspension of colloidal silica: stock removal rate and
the contact angle are shown in Fig. 2.1a and b. The silicon surface is mainly covered
with SiH groups when the removal rate peaks at pH=11. The dissolution reaction
is assumed to occur according to the mechanism shown in Fig. 4.1 with OH
, H
2
O
and O
2
being the active species [Pi3]. If ammonia or amines and traces of copper are
added to the CMP solution acceptor neutralization takes place because of incorpora-
tion of atomic hydrogen in the bulk silicon crystal [Pr2].
2 The Chemical Dissolution of Silicon 24
Different chemical treatments for silicon can be categorized depending on the
condition of the Si surface after the clean. The two basic surface conditions for a
silicon surface are hydrophobic and hydrophilic.
2.2.1
The Hydrophobic Silicon Surface
A characteristic feature of a hydrogen-terminated silicon surface is its water repel-
ling property. Such a surface exhibits a large contact angle for a drop of water
[He1, Ra1] and is therefore called hydrophobic. The dependence of the contact an-
gle of a water droplet on the chemical treatments applied to the silicon surface
are shown in Fig. 2.1b. A common procedure to establish a hydrophobic, hydroge-
nated surface condition is dipping the sample in 1% HF for 15 s (HF dip). Con-
centrated HF, mixtures of HF and NH
4
F or pure NH
4
F will show similar results
concerning contact angle. The microscopic flatness of the silicon surface, however,
depends on the type of etchant. While an HF dip usually induces a certain rough-
ness, a short treatment in 40% NH
4
F is reported to produce atomically flat sur-
faces for (111) as well as (100) oriented silicon surfaces [Hi9, Th9].
A hydrophobic Si surface condition is also observed after alkaline treatments,
like CMP at a pH of about 11 or after etching in alkaline solutions, as shown in
Fig. 2.1b. Hydrophobic Si surfaces are very susceptible to hydrocarbon contamina-
tion, for example from the ambient atmosphere or from hydrocarbon films float-
ing on a liquid. To avoid the latter case, water rinses and HF dips are often per-
formed in an overflow wet bench.
2.2 Silicon Surface Conditions and Cleaning Procedures 25
Fig. 2.1 (a) Stock removal rate during CMP
as a function of slurry pH for Si (100) sur-
faces (open circles) and Si (111) surfaces
(filled squares), respectively.
(b) Surface contact angles of a 5 ll droplet
for hydrophilic and hydrophobic cleaning pro-
cedures and for CMP of Si (100) surfaces
(open circles) and Si (111) surfaces (filled
circles), as a function of slurry pH. After [Pi2].
2.2.2
The Hydrophilic Silicon Surface
A hydrophilic surface condition has been related to the presence of a high density
of silanol groups (SiOH) or to a thin interfacial oxide film. Such an oxide can be
produced chemically by hot HNO
3
or by solutions containing H
2
O
2
. The three
most common cleaning solutions for silicon are based on the latter compound:
1. SPM or pirhanja-clean: 1: 1 (30% H
2
O
2
: 98% H
2
SO
4
), 10 min at 1108C.
2. SC-1 or RCA-clean: 1: 1: 5 (30% H
2
O
2
: 25% NH
4
OH: H
2
O) 10 min at 808C.
3. SC-2: 1: 1: 6 (30% H
2
O
2
: 37% HCl: H
2
O), 10 min at 808C.
A standard cleaning sequence for silicon wafers starts with SPM, which is only re-
quired if organic residues are present. Between the following three steps (SC-1,
HF dip, SC-2) the wafer should be removed from the solutions after 1 min of
overflow quench with DI water and brought wet into the next cleaning solution.
Final drying is done in a wafer centrifuge (spin-dryer) [Ke1]. Wafer handling
should be done with PTFE or PP tweezers. An alternative concept without alka-
line chemicals has been proposed [Me4].
Details of the chemical oxidation process are discussed in Section 5.2. The strin-
gent requirements concerning metal contamination and the trend to more envi-
ronmentally friendly processing are a constant force to improve cleaning proce-
dures in todays semiconductor manufacturing [Me4, Sa1, Oh1].
Note that a native oxide film also forms under dry conditions in ambient air;
the oxidation rate of this process can be enhanced by ultraviolet (UV)-ozone
photooxidation [Ta1, Vi1]. Oxide-covered Si surfaces exhibit low contact angles.
Only if the oxide surface is contaminated, for example by a monolayer of ab-
sorbed hydrocarbons, may larger contact angles be observed.
For a silanol-covered surface the contact angle depends on the relationship be-
tween SiOH and SiH groups, which gives rise to a dependence on pH, as
shown in Fig. 2.1b. A DI water rinse is sufficient to generate both basic Si surface
conditions, depending on temperature and dissolved oxygen concentration (DOC).
If a hydrogen-terminated silicon surface is exposed to DI water of moderate or
high DOC at RT for several minutes the SiH is slowly replaced by silanol groups
(SiOH) and a native oxide is formed for long rinsing times. This reduces the
contact angle, and the surface becomes hydrophilic. In DI water of very low DOC
(<0.004 ppm), in contrast, the reverse reaction is observed for elevated tempera-
tures (808C) and long etching times (60 min) [Wa2]; the thin oxide is removed
and a hydrogen-terminated surface is established again.
2 The Chemical Dissolution of Silicon 26
2.3
Chemical Etching in Alkaline Solutions
Alkaline etchants are mainly used for two purposes: for CMP, as discussed in Sec-
tion 2.2, or to remove silicon in an anisotropic manner, which is applied in the
micromachining of silicon. Alkaline etchants are aqueous solutions of inorganic
compounds like LiOH [Se3], NaOH [Al1, Al2], KOH [Se3, Sa6, Pr1, Xi1], RbOH
[Wa3], CsOH [Cl1, Ya2, Ya3] or NH
4
OH [Ke1, Go1, Sc4], as well as aqueous solu-
tions of organic compounds, such as ethylenediamine [Re1], hydrazin [Me1, Ce1,
Xu1], tetramethyl ammonium hydroxide (TMAH) [Ta2, Me2], cholin [As1], and
amine gallates [Li2]. All these etchants show a similar etching behavior, which has
been attributed to OH
and H
2
O being the active species, the cation being of mi-
nor importance. Details of the chemical dissolution reaction are discussed in Sec-
tion 4.1. Common features of the alkaline etchants are:
1. Alkaline etchants are anisotropic. The etch rate for the (111) crystal planes of
the Si crystal is smaller by about two orders of magnitude than the etch rate of
any other crystal plane. The etch rate ratio between other crystal planes like
(100) and (110) depends on etchant concentration and temperature, but doesnt
usually exceed a factor of two [Sa6]. Addition of oxidizing agents reduces the an-
isotropy. The etch rate of (100) Si and SiO
2
in KOH at different temperatures is
shown in Fig. 2.2.
2. The etch rate of highly p-doped silicon (>10
19
cm
3
) is found to be reduced by
orders of magnitude compared to the etch rate of low doped or highly doped n-
type silicon of (100) orientation. This effect is shown in Fig. 2.3 for KOH; note
that the etch rate is significantly affected by the amount of dissolved oxygen
2.3 Chemical Etching in Alkaline Solutions 27
Fig. 2.2 The etch rate of (a) (100) bulk
silicon and (b) thermal SiO
2
as a function
of KOH concentration and temperature.
In the temperature regime shown the etch
rate of (110) silicon is roughly 1.5 times that
of (100) silicon. After data of [Se3].
present in the alkaline solution. A silicon surface becomes resistant to alkaline
etchants if implanted with a high dose of C [Le2], N [Ac1] or Ge [Fi6]. Thin
films of SiO
2
, Si
3
N
4
or SiC are sufficient for masking in alkaline etchants.
3. The (100) etch rates show an Arrhenius dependency on temperature with activa-
tion energies of about 0.30.6 eV, depending on the alkaline solution used [Kr1,
Se3].
4. Etching of silicon in alkaline solutions occurs under evolution of hydrogen with
a ratio of two molecules H
2
per dissolved Si atom. This ratio is found to be re-
duced under positive bias [Pa6] or by addition of oxidizing agents like H
2
O
2
[Sc6]. If the anodic bias is increased beyond the passivation potential (PP), the
dissolution rate is reduced by orders of magnitude.
5. The morphology of alkaline-etched (100) and (110) silicon surfaces varies from
rough surfaces that exhibit micron-sized pyramids or ridges [Sc5] to smooth
orange peel-like surfaces, depending on the etchant composition and substrate
doping density. Mirror-like surfaces can be obtained on (111) crystal planes.
6. During the alkaline etching process the silicon surface is covered with SiOH
and SiH. The ratio of this surface species depends on pH and temperature
[Pi3, Al3].
Points 1 and 2 of the above list provide perfect conditions for the micromachining
of silicon. Using a SiO
2
or Si
3
N
4
layer patterned by photolithography, all kinds of
geometric structures can be realized, of which the basic ones are pyramids and V-
grooves on (100) wafers [Ba1] or deep U-grooves on (110) wafers [Ke2, Kr1, Ya3,
Ka5]. If deep structures have to be formed, patterning Si
3
N
4
or SiC are superior to
SiO
2
because their etch rate in alkaline solutions is orders of magnitude smaller
than that of thermal oxide, which is shown for the case of a KOH etchant in
Fig. 2.2b. Other common masking materials like organic films (resists) or metals
are rapidly dissolved or under-etched in alkaline solutions. Only epoxy-based
resists, like SU-8 [15], are found to be quite stable.
Highly p-doped layers can also be used as masking layers. If the p-type doping
level of silicon substrates is high enough to cause degeneracy (N
A
> 10
19
cm
3
), a
decrease in etch rate with doping density is observed in all alkaline solutions inde-
2 The Chemical Dissolution of Silicon 28
Fig. 2.3 Etch rate of (100) bulk silicon in
1M KOH of low and high DOC at 708C as
a function of p-type dopant concentration.
The DOC of the solution was controlled by
bubbling with pure O
2
or N
2
.
Substrate resistivity (X cm)
pendent of their composition. This low etch rate is ascribed to a passivating oxide
film present on the electrode [Pa5]. This interpretation is supported by the fact
that the etch rates of thermal SiO
2
and p
+
show qualitatively the same depen-
dence on KOH concentration. The absolute etch rate of p
+
Si in alkaline solu-
tions, however, is one or more orders of magnitude larger than that of thermal
SiO
2
[Se4]. Note that the etch stop on p
+
substrates is lost if made mesoporous.
In the porous skeleton acceptors are still present in high concentration, while the
free carrier concentration becomes negligible [Le14]. This indicates that the etch
stop is related to the Fermi level position and is not caused by the chemical stabil-
ity of the SiB bonds. Low-doped silicon germanium alloys show a similar de-
crease in etch rate if the Ge concentration is raised above 20%. This effect has
also been attributed to a passivating oxide [Fi6]. Note that the low etch rate of
(111) surfaces in alkaline solutions is not attributed to a passivating oxide but to
the slow dissolution kinetics caused by the bonding configuration of a monohy-
dride silicon surface, as discussed in Section 4.1.
Silicon etch rates in alkaline solutions commonly increase monotonically with
temperature. For KOH, for example, the etch rate r can be calculated according to:
r = K [H
2
O[
4
[KOH[
1=4
exp (E
a
=kT) (2:1)
with K=2480 lm h
1
(mol l
1
)
4.25
and E
a
=0.595 eV for a (100) surface and with
K=4500 lm h
1
(mol l
1
)
4.25
and E
a
=0.6 eV for a (110) surface [Se3]. Values of r
for a (100) silicon surface as calculated from Eq. (2.1) are shown in Fig. 2.2a. The
often reported aging of alkaline etching solutions, accompanied by reduced etch
rates, has been attributed to the formation of silica in the etchant, polymerizing
from the silicate monomer formed as initial etching product [Ni6, Le29].
The surface morphology of (100) silicon planes in alkaline solutions depends
strongly on concentration. High OH
HNO
2
H
2
O 2 h
(2:2)
This reaction is complicated and very likely requires the participation of HNO
2
as a
catalyst and of NO
2
as an intermediate [Ko14, Ve1]. The injected holes are consumed
in the dissolution reaction of silicon, which proceeds in the same way as the tetra-
valent electrochemical dissolution path, as shown in Fig. 4.4. If the HNO
3
concen-
tration is low, the formation of a thin PS layer is observed, indicating that hole sup-
ply might be the limiting factor, as is the case in the divalent electrochemical disso-
lution path, as depicted in Fig. 4.3. The final dissolution product is a tetravalent
hexafluoride complex (SiF
6
2
) as is the case for the electrochemical dissolution pro-
cess. The majority of the gas that evolves during silicon dissolution is hydrogen,
with considerable amounts of N
2
O and traces of NO and NO
2
[Ko14].
2 The Chemical Dissolution of Silicon 32
Fig. 2.6 Curves of constant etch rate
(in lm min
1
) as a function of etchant com-
position. Compositions in regionAgive smooth,
specular surfaces and rounded edges, while
rough surfaces and peaked corners are gener-
ated for compositions in region C. Region B
is intermediate. Redrawn from [Sc9].
The OCP etch rate of p-type and highly doped n-type Si electrodes in HFHNO
3
mixtures increases by an order of magnitude under sufficiently anodic bias [Le20]. In
the cathodic regime significant dark-currents are observed for p-type electrodes, as
shown in Fig. 4.12. This is ascribed to hole injection from the electrolyte [Ko14].
Note that hole injection is not observed in aqueous HF free of oxidants.
Another characteristic feature of this etchant is the autocatalytic nature of reaction
(2.2). As a result, the reaction must proceed first through an induction period in
which the build-up of HNO
2
increases the etch rate to its steady-state value [Ro2].
This delay depends greatly on the condition and the size of the silicon sample sur-
face and the volume and the convection of the etchant. It becomes short for a dam-
aged or a porous silicon surface, for small ratios of etchant volume to sample surface
and low convection rates. Under such conditions the etch rate increases fast, which
may eventually lead to a significant increase in temperature, which again increases
the etch rate. This effect is especially pronounced for porous layers, for which the
solution can heat up to the boiling point within seconds and may cause hazard to
the user. Note that mixtures of micro PS and an oxidizing agent show an explosive
interaction even at cryogenic temperatures [Mc3, Ko24]. The strong temperature de-
pendence and the autocatalytic effect limit the reproducibility of this etchant con-
cerning etch depth and etch geometry. The induction period is specific to HNO
3
and not observed for other oxidants like, for example, KBrO
3
[Se6] or CrO
3
[Me5].
It is known that HFHNO
3
-based solutions etch highly doped substrates faster
by a factor of about three compared to moderately doped ones. A higher selectiv-
ity is reported on addition of chemicals that reduce the HNO
2
concentration, like
H
2
O
2
or NaN
3
[Mu1]. However, this report suffers from the fact that the etch rate
was measured for separate wafers of a homogeneous doping density. For pp
+
or
nn
+
structures, which are not spatially separated, only a low selectivity is ob-
served, because of the autocatalytic behavior of the etchant.
Masking is required for many micromechanical applications. While Si
3
N
4
is
only suitable for a small etching depth because of its significant etch rate in HF,
noble metals like gold are sufficient mask materials. In contrast to alkaline
etchants, organic materials like certain resists or even some adhesive tapes are
well suited to protect the silicon surface in isotropic etchants.
2.5
Defect and Junction Delineation
There are several techniques for characterizing lattice defects in a silicon crystal.
Transmission electron microscopy (TEM) gives very detailed information about a
single defect because of its high spatial resolution, as shown in Fig. 2.7a, but the
method is rather time-consuming. A projection of all defects in a wafer is given
by an X-ray topography (XRT) picture. XRT is non-destructive and the preparation
time is in the order of hours. Small angle neutron scattering (SANS) provides in-
formation about the average size, form and density of lattice defects. This method
is especially suited for small defects of about 10 nm. The high penetration depth
2.5 Defect and Junction Delineation 33
of neutrons allows us to investigate slugs several centimeters thick. Preferential
chemical etching is the fastest characterization technique, but it is destructive and
reveals only defects located in, or close to, the wafer surface. An advantage of pre-
ferential etching is its magnifying effect. As shown in Fig. 2.7a, the thickness of
the actual lattice defect is a few nanometers, while the etch pit it causes is mea-
sured in micrometers, as shown in Fig. 2.7b, and is therefore detectable by optical
methods, as shown in Fig. 10.6f.
Etchants for defect and junction delineation are usually composed of HF and an
oxidizing agent such as HNO
3
[Da1, Gr4, Ka4, Ne1], K
2
Cr
2
O
7
[Se5] or CrO
3
[Si1,
Je1, Sc7, Ya4, Me5]. Alkaline solutions are rarely used for defect delineation [Ma12].
An etch pit will form on a silicon surface if the dissolution rate is enhanced lo-
cally. Enhancement of the etch rate may occur for various reasons:
1. The higher chemical energy of the elastic strain field present around a dislocation.
2. A high impurity concentration caused by gettering of impurities by a dislocation.
3. An impurity that has formed a precipitate, for example a metal silicide.
4. A different doping density.
5. Bubbles sticking randomly at the surface.
Etch pit formation as a result of the factors given in 14 above can be used to
characterize silicon materials. A summary of common defect etchants for silicon
is given in Table 2.1.
Single etch pits can be inspected and counted under an optical microscope, as
shown in Fig. 2.7b. The scattering of light shining onto a wafer surface under a
2 The Chemical Dissolution of Silicon 34
Fig. 2.7 (a) Nickel silicide precipitates in
disks parallel to the (111) Si crystal planes.
Such a NiSi
2
platelet in bulk Si is revealed by
high-resolution electron microscopy (HREM);
the tip of the platelet is penetrating the ther-
mal oxide layer.
(b) After Secco etch the platelet and the sur-
rounding silicon are preferentially etched.
Note the different magnifications. A high den-
sity of etch pits produce a haze-like appear-
ance of the wafer surface.
small angle gives a macroscopic image of the etch pit density across the wafer.
Areas of high etch pit concentration show a bright haze-like picture while areas of
no defects remain dark, as shown in Fig. 10.6f.
The sensitivity of this haze method is about 10
12
cm
3
for Cu and Ni [Gr4],
while contamination by Fe can only be detected if the iron forms precipitates,
which is only the case in wafers of a sufficiently high oxygen concentration. This
property of the haze test has been exploited as a qualitative indicator of gettering
efficiency of oxygen-related defects in CZ silicon [Fa2, Fa3]. Because it has been
shown that there is no difference between the results obtained with Cr-based solu-
tions and HNO
3
-based ones, the latter are preferred for environmental reasons
[Ch1, Gr4, Ka4]. A study that compared images of polycrystalline substrates ob-
tained by electron beam induced current (EBIC), by chemical defect etching, and
by anodic etching showed that electrically active defect sites are preferentially
etched under applied bias [Fo1].
Methods of characterizing the dopant distribution based on metal plating or se-
lective chemical etching of doped layers and subsequent microscopy have been
used since the early days of device fabrication [Si5, Wu2]. Despite the fact that
other methods, for example secondary ion mass spectroscopy (SIMS) or spreading
resistance profiling (SRP), are sensitive over a wider range and give better calibra-
2.5 Defect and Junction Delineation 35
Tab. 2.1 Composition and etch rates of common defect etchants for silicon.
Etchant Composition in parts of volume r (lm/min) Ref.
HF (ml) HNO
3
(ml) CH
3
COOH (ml)
50% 70% 100%
HF 1 <0.0001 [Hu2]
John 0.15 100 0.018 [ Jo1]
Graff 1 20 4 1.85 [Gr4]
Dash 1 3 10 3 [Dal4]
CP4 3 5 3 5075 [Vo1]
Sirtl 1 Vol (100 ml H
2
O+50 g CrO
3
) 3.5 [Si1]
1 Vol (48% HF)
Schimmel 1 Vol (100 ml H
2
O+7.5 g CrO
3
)
2 Vol (48% HF)
1.73.2 [Sc7]
Yang 1 Vol (100 ml H
2
O+15 g CrO
3
)
1 Vol (49% HF)
1.5 [Ya4]
Secco 1 Vol (100 ml H
2
O+45 g CrO
3
)
2 Vol (48% HF)
1.5 [Se5]
Wright 1 Vol (90 ml H
2
O+45 g CrO
3
)
1 Vol (180 ml H
2
O+6 g CuNO
3
)
1 Vol (90 ml HNO
3
+180 ml CH
3
COOH+180 ml HF)
1 [ Je1]
Seo 12 M HF+0.05 M KBrO
3
0.150.5 [Se6]
tion, chemical etching is still indispensable, because it is time efficient and gives
a two-dimensional image. HF- and HNO
3
-based solutions show an enhanced etch
rate for n-type layers with doping densities in excess of 10
18
cm
3
if compared to
the etch rate on low doped n-type or p-type Si [Ne1]. For the delineation of p-type
layers alkaline solutions can be used because the etch rate is significantly reduced
for dopant densities in excess of 10
19
cm
3
[Se4]. Differences in doping concentra-
tion below this level, for example moderately doped pn junctions, can also be de-
lineated, but these methods are electrochemical in nature and are therefore dis-
cussed in Sections 4.5 and 4.6.
2.6
Selective Etching of Common Thin Film Materials
One of the properties that makes wet etchants indispensable in silicon device
manufacturing today is their high selectivity for layers of different chemical com-
position. Most prominent are the oxide etchants based on HF. In semiconductor
manufacturing thin oxide layers are commonly etched by diluted HF [Ki1]. For
thicker oxide layers a buffered oxide etchant (BOE), a mixture of HF, NH
4
F and
water is preferred. The commonly used ratio of NH
4
F to HF in BOE is 7 to 1.
The addition of NH
4
F increases the etch rate of HF [Pr3] on oxide, while the etch
rate on aluminum interconnects remains low. Equation (4.5) gives the overall dis-
solution reaction of SiO
2
in HF. Note that the reaction product fluosilicic acid
(H
2
SiF
6
) still shows a significant etch rate for silicon dioxide [Th8]. The simple as-
sumption that 6 moles of HF remove 1 mole of SiO
2
according to Eq. (4.5) is
therefore not correct. The etch rates of thermal oxide for different concentrations
of aqueous HF at RT are well studied [ Ju1, Ki1, Ve1]. The activation energy of
chemical dissolution of SiO
2
in 1% HF is determined to be 0.34 eV [Ki1].
Data from different authors are summarized in Fig. 2.8. Equation (2.3) is a fit
of these experimental values (for r in nm min
1
and c
HF
in % of aqueous HF):
r
oxide
= 0:25 5:5 c
HF
0:047 c
2
HF
0:0065 c
3
HF
(2:3)
The etch rates of other oxides such as borophosphosilicate glass (BPSG), phospho-
silicate glass (PSG), tetraethylorthosilicate (TEOS), spin-on glass (SOG), anodic
oxides and oxidized PS in HF are generally larger by factors of up to two orders of
magnitude compared to the value for thermal oxides [So2, Bu6, Mo5]. Compre-
hensive reviews of SiO
2
etching in HF are available in the literature [Bu6, Mo6].
The etch rate of CVD silicon nitride in HF is sensitive to the details of the de-
position process. Values measured for the etch rate of Si
3
N
4
deposited at 8508C
are shown in Fig. 2.8. The best fit to these etch rates is found to be (for r in
nm min
1
and c
HF
in % of aqueous HF):
r
nitride
= 0:18 c
HF
0:0013 c
2
HF
(2:4)
2 The Chemical Dissolution of Silicon 36
The reaction rate of silicon nitride dissolution has been found to be mainly deter-
mined by the concentrations of F
, while for an n-type semiconductor the reverse is true. An analog to the SCR
in the semiconductor is an extended layer of ions in the bulk of the electrolyte,
which is present especially in the case of electrolytes of low concentration (typi-
cally below 0.1 mol l
1
). This diffuse double layer is described by the Gouy-Chap-
man model. The Stern model, a combination of the Helmholtz and the Gouy-
Chapman models, was developed in order to find a realistic description of the
electrolytic interface layer.
Now we have discussed the electrostatics dealing with where the charges stay,
but how about the electrodynamics, how do they move in space and energy? In
contrast to a semiconductor where a current is caused by the movement of free
electrons and holes, an electrolyte can be understood as a phase through which
charge is carried by the movement of ions. Hence the mobilities of electrons or
holes in silicon (l
e
, l
h
> 100 cm
2
V
1
s
1
) and ions in solutions (l
H
+
=36.210
4
cm
2
V
1
s
1
, l
OH
=20.610
4
cm
2
V
1
s
1
, l
F
=5.710
4
cm
2
V
1
s
1
) differ by or-
ders of magnitude. However, the electrical conductivities of a common electrolyte
and a moderately doped silicon electrode are of the same order of magnitude, be-
cause the ratio of density of charge carriers in an electrolyte (10
21
cm
3
for
1 mol l
1
) and in moderately doped silicon (10
14
10
17
cm
3
) is about the inverse of
the ratio of mobilities.
Any transfer of a charge between the electrode and the host ion in the electro-
lyte is accompanied by a chemical reaction. If an electron moves from the elec-
trode to the solution, the host ion is reduced. If it moves in the other direction
the host ion is oxidized. The electron is a term used in physics as well as chemis-
try. This is in contrast to the concept of a hole, a positive charge carrier, which de-
scribes the behavior of an electron vacancy in the periodic potential of a semicon-
ductor crystal. This term, however, loses its meaning at the interface to the elec-
trolyte. Concerning the chemical reaction, any hole crossing the interface can be
replaced by an electron going in the other direction. Energetically, however, there
is a difference, because the injection of an electron from the host ion into the sili-
con conduction band (CB) requires an energy 1.1 eV larger than that required for
a hole in the valence band (VB) going in the other direction. In many cases the
interaction of holes with the electrode surface can be understood if two holes at
the interface are replaced by one broken bond of a surface atom.
Not only the movement of charges in space but also changes in energy occur in
different ways in the semiconductor and the electrolyte. The well established
semiconductor band model where an excited electron may jumps from one en-
ergy level to another is not applicable to an electrolyte. In a solution the energy
levels itself fluctuate in energy due to thermal fluctuations of the surrounding sol-
vent dipoles, carrying the electron with it. A detailed description of the latter mod-
el is given in the literature [Mo2, Ge1, Ge3].
The current-voltage (IV) characteristic of a semiconductor-electrolyte contact is
determined by both the semiconducting nature of the electrode, as well as by the
ionic and molecular species present in the electrolyte. The current density at the
electrode for a certain potential is limited by the reaction kinetic at the interface,
or by the charge supply from the electrode or the electrolyte.
3 The Semiconductor-Electrolyte Junction 40
If the charge transfer is limited by charge supply from the silicon electrode the
basic IV curve of the silicon-electrolyte junction is very similar to that of a sili-
con-metal junction, a Schottky diode, as shown in Fig. 3.1a and b. A p-type elec-
trode, for example, is reversely biased in the cathodic regime. Therefore the cur-
rent is minimal in the dark and increases upon illumination or for a bias high en-
ough to initiate breakdown (Fig. 3.1a). In the anodic regime the junction is under
forward conditions and the current is described by the diode equation. The case
of an n-type electrode is only different insofar as the forward and the reverse re-
gime are exchanged in the IV curve, as shown in Fig. 3.1b.
A Schottky diode is always operated under depletion conditions; flat-band condition
would involve giant currents. A Schottky diode, therefore, models the silicon electro-
lyte interface only accurately as long as the charge transfer is limited by the electrode.
If the charge transfer becomes reaction-limited or diffusion-limited, the electrode may
as well be under accumulation or inversion. The solid-state equivalent would nowbe a
metal-insulator-semiconductor (MIS) structure. However, the IV characteristic of a
real silicon-electrolyte interface may exhibit features unlike any solid-state device, as
3.1 Basics of the Semiconductor-Electrolyte Contact 41
Fig. 3.1 The IV characteristic of (a) a p-type
and (b) an n-type silicon electrode under the
assumption that the current is dominated by
the properties of the semiconductor and is
not limited by interface reactions or by diffu-
sion in the electrolyte. (c) The characteristic
IV curve in an alkaline electrolyte under the
assumption that the current is not limited by
charge supply from the electrode (which is
the case for a highly doped or strongly illumi-
nated electrode). (d) The IV curve under the
same assumption as in case (c) is shown for
an acidic electrolyte without (full line) and
with fluoride (broken line).
shown for example in Fig. 4.7. This is due to details of the chemical reaction at the
interface, like for example the formation of an anodic oxide film for a certain bias or
current density. In Fig. 3.1 c and d the basic IV curve in alkaline and acidic electro-
lytes respectively are shown for a silicon substrate doped in excess of 10
19
cm
3
. The
same characteristics are observed if low doped substrates are strongly illuminated.
Under such conditions the blocking effect of the reversed biased Schottky diode is
lost and the basic characteristic of the IV curve is the same for all silicon substrates
independent of doping type and density. This is a consequence of the fact that the
chemical reactions at the silicon electrode are independent of the minute traces of
dopants. However, while the chemical dissolution reaction is not sensitive to doping
density the Fermi energy is. As a result, there is a significant potential shift of the IV
curves and the open circuit potentials with doping density, as shown in Fig. 3.3.
The IV characteristic of a real silicon electrode is usually superpositions of the
basic potentiostatic curves shown in Fig. 3.1 ad. The galvanostatic IV curves are
deductible in many cases from the potentiostatic ones. This holds true for a con-
stant potential or a very slow scan of the applied potential, for fast changes of poten-
tial or current, however, the IV curve may show transient contributions. Transient
effects at the silicon electrodes can be due to chemical or capacitive effects as well as
to changes of the interface morphology for example by pore formation. Chemically
caused effects are usually slow and current or bias show transients on a time scale in
the order of seconds. The most prominent example for such effects are electroche-
mical oscillations observed at silicon electrodes anodized in HF, as discussed in Sec-
tion 5.5. Transient currents due to oxide passivation and electron injection during
oxide dissolution are shown in Fig. 4.14. Pore formation does significantly change
the surface area and morphology of the electrode and it thereby affects the IV char-
acteristic. The reverse current of an n-type electrode, for example, increases by or-
ders of magnitude due to formation of mesopores of the breakdown type, as dis-
cussed in Section 8.4. However, even hydrogen bubbles sticking to the electrode sur-
face are found to generate a considerable noise pattern in the IVcurves. The charge
values involved in charging interfacial capacities, due to the SCR, an anodic oxide or
the Helmholtz layer, are commonly much smaller than the charges exchanged dur-
ing chemical reactions, such as anodic oxide formation. The measurement of inter-
facial capacities is discussed in Section 10.2.
In the next section the charge states of the electrode and the electrochemical re-
actions are discussed for acidic, especially fluoride-containing electrolytes followed
by a section dealing with alkaline electrolytes.
3.2
The IV Characteristics of Silicon Electrodes in Acidic Electrolytes
In this section the IV characteristic of a silicon electrode in acidic, and specially
hydrofluoric electrolytes is discussed with emphasis on the different charge states
of the semiconductor. The accompanying chemical reactions are briefly men-
tioned, but are discussed in detail in Chapter 4.
3 The Semiconductor-Electrolyte Junction 42
While the cathodic characteristic of the IV curve is dominated by hydrogen re-
duction and therefore not very sensitive to the kind of ions present in an aqueous
electrolyte, the anodic part is. If a silicon electrode is anodized in an acidic electro-
lyte that is free of fluoride, a passivating oxide layer is formed. Details of the for-
mation and the properties of anodic oxides are given in Chapter 5. An IV curve
in the common sense does not exist for such electrolytes, because a certain anodic
bias does not correspond to a constant current density, but rather to a constant
charge of a few mAs cm
2
which is consumed during the formation of an oxide
layer. In this respect there is no current-voltage curve, but a charge-voltage curve
for the anodic regime, as shown in Fig. 5.1. Only for short anodization times and
small current densities in the lA cm
2
range the IV curve is not dominated by
oxide formation but by the semiconducting properties of the electrode [Ch10].
A passivating oxide is formed under sufficiently anodic potentials in HF, too.
However, there are decisive differences to the case of alkaline and fluoride-free
acidic electrolytes. For the latter electrolyte the steady-state current density prior to
passivation is zero and it is below 1 mA cm
2
for alkaline ones, while it ranges
from mA cm
2
to A cm
2
in HF. Furthermore, in HF silicon oxide formation does
not lead to passivation, because the anodic oxide is readily etched in HF. This
gives rise to an anodic IV curve specific to HF, it shows two current maxima and
two minima and an oscillatory regime, as for example shown in Fig. 4.7.
There are several methods to investigate the charge states of a semiconductor
electrode, for example high-frequency resistometry (HFR) [Ot1]. Below a transis-
tor-like set-up, as shown in the inset of Fig. 3.2, will be discussed because it
shows in an exemplary way the similarities and differences of solid-state junctions
and liquid junctions.
The electrochemical set-up, as depicted in the inset of Fig. 3.2, consists of a 0.5 mm
thick silicon slice in an electrolytic double cell, as shown in detail in Fig. 1.10. This
establishes a transistor-like configuration with the backside electrolyte as collector,
the front side electrolyte as emitter and the silicon electrode as base in between. How-
ever, this transistor is not operated in a common amplifying transistor circuit. A
constant reverse bias V
BC
of 3 V is applied between collector and base in order to
collect the minority carriers diffusing through the base, while the emitter-base volt-
age is switched from V
OCP
to values of V
EB
between 10 Vand +10 V (I
EB
according to
the solid line in Fig. 3.2). Prior to switching the electrode has been cleaned by elec-
tropolishing to ensure a virgin interface of the electrode free of contaminants or PS.
The measurements are performed with and without illumination of the front elec-
trode. The IR light of the light source was blocked by a filter, to insure that minority
carriers are only generated in the vicinity of the front surface. Note that the base-col-
lector junction could as well be a solid-state junction, for example a Schottky junc-
tion. Only the electrolytic nature of the emitter-base junction is probed by this set-up.
If minority carrier current (I
BC
, dotted line, symbols in Fig. 3.2) is detected at the
collector, it can be concluded that the emitter is no sink for minorities. The absolute
value of I
EB
depends not only on the charge state of the emitter-base junction and
surface recombination velocity, but as well on bulk diffusion length and on sample
thickness. However, the latter two parameters are constants for a given sample.
3.2 The IV Characteristics of Silicon Electrodes in Acidic Electrolytes 43
Four different regimes of the IV curve for moderately doped silicon electrodes
in an HF electrolyte are shown in Fig. 3.2. These regimes will now be discussed
in terms of the charge state of the electrode, the dependence on illumination con-
ditions, the charge transfer, the mass transport, and accompanying chemical reac-
tions. Transient effects are indicated in Fig. 3.2 by a symbol with an arrow.
3 The Semiconductor-Electrolyte Junction 44
Regime 1
This is the regime of cathodic currents. The silicon atoms of the electrode do not
participate in the chemical reaction in this regime. An n-type electrode is under
forward bias and the current is caused by majority carriers (electrons). The fact
that photogenerated minority carriers (holes) are detectable at the collector indi-
cates that the front is under flat band or accumulation. A decrease of I
BC
with
cathodization time is observed. As Fig. 3.2 shows, the minority carrier current at
the collector after switching to a cathodic potential is identical to that at V
OCP
in
the first moment, but then it decreases within seconds to lower values, as indi-
cated by arrows in Fig. 3.2. This can be interpreted as an increase of the surface
recombination velocity with time under cathodic potential. It can be speculated
that protons, which rapidly diffuse into the bulk of the electrode, are responsible
for the change of the electronic properties of the surface layer [Al7]. However, any
other effect sufficient to produce a surface recombination velocity in excess of
100 cm s
1
would produce similar results.
A p-type electrode shows only a small reverse current density in the cathodic re-
gime, if it is kept in the dark as shown in detail in Fig. 4.11. If defects are present
in the electrode surface, this small dark current increases by orders of magnitude
[Wa4]. If V
EB
is increased breakdown of the junction is observed, at a bias, which
depends on the doping density of the substrate, as displayed in the figure of the
front inner cover of this book. If the electrode is illuminated a photocurrent is ob-
served and the number of electrons transferred to the electrolyte per absorbed
photon is one, if losses by recombination in the bulk of the electrode are ne-
glected, as shown in Fig. 4.13. On the other hand, current multiplication effects,
as observed for anodized n-type electrodes, are not present in the cathodic regime.
Under illumination two different cases can be distinguished. In one case the bias
is sufficiently cathodic to keep the front side of the electrode under depletion and
so a cathodic emitter current proportional to the photon flux and independent of
applied bias is generated. In this case no electrons are detected at the collector
(region 1a in Fig. 3.2). In the other case the bias is closer to V
OCP
and not all
photogenerated electrons are immediately consumed in the cathodic reaction.
Electrons now accumulate and it can be concluded that the electrode is under in-
version or in flat-band condition. A certain fraction of the photogenerated elec-
3.2 The IV Characteristics of Silicon Electrodes in Acidic Electrolytes 45
Fig. 3.2 The inset (center left) shows the
electrochemical double-cell set-up with the
two applied potentials V
EB
and V
BC
, which
constitute a circuit similar to a solid-state bi-
polar transistor. The emitter-base current I
EB
(full line) of a moderately doped p-type elec-
trode illuminated corresponding to a photo-
current of 10 mA cm
2
is shown in the upper
part of the figure. Below I
EB
is shown for an
n-type electrode illuminated with an intensity
corresponding to 90 mA cm
2
. The base-
collector current I
BC
for a constant reverse
bias of U
BC
=3 V is monitored without (filled
symbols) and with illumination of the emitter-
base junction (open symbols, dotted line as a
guide to the eye). For each regime of the I
EB
V
EB
characteristic (indicated by numbers and
broken lines) the proposed band diagram is
shown (center right) for the illuminated case
assuming a quasi Fermi level to be present in
the electrolyte.
3
trons diffuse through the base and can be detected at the collector (region 1b in
Fig. 3.2).
Regime 2
This is the region of the OCP. For an HF electrolyte without an oxidizing agent
the electrode is inert, because no chemical reaction occurs at the front (emitter) at
this potential range. The OCP depends on illumination condition, substrate dop-
ing density, illumination condition, HF concentration and DOC [Ot1]. For moder-
ately doped Si substrates in 5% aqueous HF the OCP is usually close to 0.6 V
versus SCE in the dark. Under illumination a small negative (cathodic) shift to
0.64 versus SCE is observed for n-type electrodes, while the OCP for p-type sub-
strate shifts significantly in positive (anodic) direction to 0.2 V versus SCE [Be9].
Because the emitter-base current is zero in this regime, the electrode behaves
similar to an MIS contact. It can be assumed that the number of thermally gener-
ated minority carriers is not sufficient to produce inversion; the electrode is there-
fore under depletion if kept in the dark. Under illumination the generation rate is
high and minority carriers accumulate at the emitter because there is no chemical
reaction path for them to enter the electrolyte, at OCP. Therefore the emitter is
under inversion or under flat-band conditions. Under such conditions the collec-
tor shows a high and stable minority charge carrier collection efficiency for n-type
as well as p-type doping. This property of the electrode under OCP is exploited
for bulk diffusion length measurements as discussed in Section 10.3. A prerequi-
site of the high collection efficiency is a very low surface recombination velocity
in the order of 1 cm s
1
, which has been reported for a silicon surface in contact
with HF at OCP [Ya1]. By using different electrolytes, the presence of an inversion
layer and not a reduced density of electrical trap sites on the silicon surface has
been found to be primarily responsible for these low values of surface recombina-
tion velocity [Ro9].
Regime 3
This is the regime of anodic current densities below J
PS
. A hole approaching the
interface initiates the divalent electrochemical dissolution of a silicon surface
atom at the emitter. The dissolution proceeds under formation of H
2
and electron
injection, as shown in Fig. 4.3. The formation of PS structures is confined to this
region.
Let us now consider the charge state of the electrode. The emitter is positively
biased. A p-type silicon electrode is therefore under forward conditions. If the loga-
rithm of the current for a forward biased Schottky diode is plotted against the ap-
plied potential (Tafel plot) a linear dependency with 59 meV per current decade is
observed for moderately doped Si. The same dependency of I
EB
on V
EB
is observed
at a silicon electrode in HF for current densities between OCP and the first current
peak at J
PS
, as shown in Fig. 3.3 [Ga1, Ot1]. Note that the slope in Fig. 3.3 becomes
less steep for highly doped substrates, which is also observed for highly doped
Schottky diodes. This, and the fact that no electrons are detected at the collector, in-
dicates that the emitter-base interface is under depletion. This interpretation is sup-
3 The Semiconductor-Electrolyte Junction 46
ported by capacitance measurements as well as IR measurements [Ga1, Se7, Oz2,
Oz4]. The formation of PS shows little effect on the basic IV characteristic. How-
ever, a transient current-time behavior after a potential step from OCP to a potential
in region 3, due to pore initiation has been observed [Po3].
As shown in Fig. 3.3, the IV curve in this regime shows a cathodic potential
shift and a slight change of slope, if the doping density is increased. Compared to
p-type substrates the IV curve of p
+
is shifted cathodically by about 0.1 V and
that of n
+
by about 0.2 V [Ga1, Zh5]. This shift can be exploited for etch stops and
selective formation of PS, as discussed in Section 4.5.
An n-type electrode is under reverse bias in this anodic regime, and so only a
small dark current density in the order of a few lA cm
2
is observed, as shown in
Fig. 4.11. The dark current increases by orders of magnitude if the breakdown
field strength is reached at the electrode surface or at a pore tip present in the
electrode, as discussed in Section 8.4.
If the n-type electrode is illuminated, no holes are observed at the collector in
regime 3. This and the Tafel characteristic of I
EB
are indications of a depletion of
the emitter-base junction. Depletion is obvious for regime 3a, because V
EB
gener-
ates a large SCR sufficient to collect all photogenerated holes and to drive them to
the emitter. However, in region 3b no holes are detected at the collector, too. This
is in stark contrast to the corresponding region 1b of the p-type electrode, where
electrons not consumed at the emitter are allowed to diffuse through the base to
3.2 The IV Characteristics of Silicon Electrodes in Acidic Electrolytes 47
Fig. 3.3 The IV curves, as recorded and
compensated for ohmic losses (iR
cor.
), of
Si electrodes in aqueous HF (1M HF, 0.5M
NH
4
Cl) are found to shift cathodically with in-
creasing p-type doping density. In a V versus
log(i) plot (inset) a p-type electrode (1 X cm,
full circles) shows a slope of 59 mV per
current decade, while a p
+
-type electrode
(5 mX cm, open circles) shows a slope
increasing from 80 to 115 mV per current
decade. After [Ot1].
the collector. A high value of surface recombination velocity for photogenerated
holes at n-type electrodes in regime 3b would explain this behavior. A similar
high surface recombination velocity also appears to be present at the p-type elec-
trode in regime 3, because electrons, known to be injected during the divalent dis-
solution reaction, as shown in Fig. 4.3, are not detected at the collector.
If V
EB
is increased, I
EB
increases and the current density at the electrode even-
tually becomes equal to J
PS
. It has been speculated that this first anodic current
peak is associated with flat-band condition of the emitter-base junction. However,
data of flat-band potential of a silicon electrode determined from Mott-Schottky
plots show significant scatter, as shown in Fig. 10.3. However, from CV measure-
ment it can be concluded that all PS formation occurs under depletion conditions
independent of type and density of doping of the Si electrode [Ot1].
Regime 4
For anodic current densities above J
PS
tetravalent electrochemical dissolution of the
electrode is observed. The chemical reaction proceeds in two steps, first the electrode
is anodically oxidized, and then the oxide is chemically dissolved in HF. The overall
reactions, as given in Eqs. (4.2) and (4.5), show that the dissolution of one silicon
atom consumes four holes. In regime 4 the current density exceeds J
PS
and the sim-
ple diode behavior disappears. The IV characteristic shows a more complex struc-
ture with another current maximum and a region of current oscillations, the latter
is indicated by a splitting of the solid line in Fig. 3.2. These features of the IVcurve
are related to the properties of a thin anodic oxide present on the electrode surface,
details are discussed in Chapter 5. The dependence of I
EB
on electrolyte convection
in regime 4 indicates that the current is not limited by charge supply from the elec-
trode any more, but by diffusion in the electrolyte. The solid-state analog for this
regime would be an MIS structure with a leaky insulator.
For a bias positive of the flat-band potential holes accumulate at the interface of
a p-type electrode. These free holes produce a sizable absorption at wavenumbers
below 2000 cm
1
, which offers another way to measure the charge state of the
electrode. The charge state of p-type electrodes in an HF electrolyte has been
probed by such in situ IR absorption measurement [Oz2, Oz4]. Negligible absorp-
tion at negative potentials and below the first current peak at J
PS
support the ob-
servation that the electrode is in depletion in regime 3. At positive potentials be-
tween J
PS
and the second current peak at J
3
(Fig. 4.7) the free-carrier absorption
remains weak. This has been ascribed to an oxide of low blocking capability cover-
ing the electrode. For more positive potentials, at and above the second current
peak, the absorption increases rapidly, indicating significant accumulation due to
a dense oxide of low defect density. In the regime of current oscillations the free-
carrier density is found to oscillate, too. This is shown in Fig. 5.12.
For the p-type substrate a significant number of electrons are collected at the
backside, as shown in the top part of Fig. 3.2. This is true not only for the illumi-
nated p-type electrode but also if the electrode is kept in the dark, which indicates
that electrons are injected during the tetravalent dissolution reaction. In the re-
gime of oscillations the electron injection current is found to oscillate, too [Ca10].
3 The Semiconductor-Electrolyte Junction 48
For higher V
EB
(>7 V) electron injection increases and I
BC
oscillates when I
EB
does. Capacitance measurements indicate a surface charge of holes at the front,
this means a p-type electrode is under accumulation.
A current of photogenerated holes observed in regime 4 at the collector at low V
EB
of about 1 V for an illuminated n-type substrate indicates that no significant SCR is
present at the front side; the n-type electrode is in inversion. If the bias is increased
this current disappears indicating an SCR or an increased surface recombination
velocity at the emitter. In regime 4a all photogenerated holes are consumed by
I
EB
. Breakdown of the junction in regime 4 does not lead to pore formation.
Note that in contrast to a solid-state transistor, hole injection from the HF elec-
trolyte into the base is not observed independent of V
EB
(full diamonds in
Fig. 3.2). Hole injection becomes sensible if an oxidizing agent is added to the
HF, as shown in Fig. 4.12.
3.3
The IV Characteristics of Silicon Electrodes in Alkaline Electrolytes
There are fewer studies devoted to the electrochemistry of silicon in alkaline elec-
trolytes than is the case for HF. This can partly be ascribed to the fact that pore
formation is not observed in alkaline electrolytes, which limits the field of applica-
tions. This section gives a brief overview of the characteristic features of IV
curves of silicon electrodes in alkaline electrolytes.
In contrast to acidic electrolytes, chemical dissolution of a silicon electrode pro-
ceeds already at OCP in alkaline electrolytes. For cathodic potentials chemical dis-
solution competes with cathodic reactions, this commonly leads to a reduced dis-
solution rate and the formation of a slush layer under certain conditions [Pa2].
For potentials slightly anodic of OCP, electrochemical dissolution accompanies
the chemical one and the dissolution rate is thereby enhanced [Pa6]. For anodic
potentials above the passivation potential (PP), the formation of an anodic oxide,
as in the case of acidic electrolytes, is observed. Such oxides show a much lower
dissolution rate in alkaline solutions than the silicon substrate. As a result the
electrode surface becomes passivated and the current density decreases to small
values that correspond to the oxide etch rate. That the current density peaks at PP
in Fig. 3.4 are in fact connected with the growth of a passivating oxide is proved
using in situ ellipsometry [Pa2]. Passivation is independent of the type of cation.
Organic compounds like hydrazin [Su1], for example, show a behavior similar to
inorganic ones, like KOH [Pa8]. Because of the presence of a passivating oxide the
current peak at PP is not observed for a reverse potential scan.
The potential separation between PP and OCP, as well as the anodic current
density corresponding to PP, increase with temperature. Current density values in
the order of 100 lA cm
2
for example are found for a (100) Si electrode in 2 M
KOH at RT. This current increases by one order of magnitude if the temperature
is increased from RT to 608C [Pa5]. A similar temperature dependence is ob-
served for chemical etching of Si in KOH, as shown in Fig. 2.2a.
3.3 The IV Characteristics of Silicon Electrodes in Alkaline Electrolytes 49
The anodic current density corresponding to PP shows no significant dependence
on doping for n-type and moderately doped p-type Si electrodes. This indicates that
the anodic currents at n-type electrodes are caused by electron injection [Pa6, Xi1]. If
the p-type doping density approaches 10
19
cm
3
the PP current density becomes re-
duced and for 10
20
cm
3
the passivation peak disappears. This has been interpreted
as a passivating oxide being already present on highly p-doped electrodes at OCP.
That the separation of a few hundred mV between the PP and the OCP decreases
with increasing doping density and becomes negligible at p-type doping densities
in excess of 10
20
cm
3
, supports this interpretation [Fa7, Pa5, Sm6].
As expected from the anisotropy of chemical etching of Si in alkaline solutions,
the electrochemical dissolution reaction shows a strong dependence on crystal ori-
entation. For all crystal orientations except (111) a sweep rate independent anodic
steady-state current density is observed for potentials below PP. For (111) silicon
electrodes the passivation peak becomes sweep rate dependent and corresponds to
a constant charge of 2.4 0.5 mC cm
2
[Sm6]. OCP and PP show a slight shift to
more anodic potentials for (111) silicon if compared to (100) substrates, as shown
in Fig. 3.4.
It has been speculated that there is a common origin of the reduced chemical
etch rate for (111) oriented silicon substrates and for highly p-type doped sub-
strates. But the electrochemical investigations discussed above indicate that the
passivation of highly doped p-type Si can be ascribed to an oxide film already pre-
sent at OCP, while no such oxide film is observed on (111) silicon below PP. This
supports models that ascribe the reduced chemical etch rate on (111) planes to a
retarded kinetic for Si surface atoms with three backbonds, present at (111) inter-
faces [Gl1, Al2], as discussed in Section 4.1.
The charge states of the silicon electrode in alkaline electrolytes have not been
investigated in detail. It can be assumed that the electrode represents an MIS
structure above PP, while it behaves similarly to a Schottky junction for potentials
below PP.
3 The Semiconductor-Electrolyte Junction 50
Fig. 3.4 The electrochemical IV
characteristic of n-type and p-type
silicon electrodes of (100) and
(111) orientation in 40% aqueous
KOH solution at 608C in the dark.
The potential is scanned from
cathodic to anodic at a sweep rate
of 1 mV s
1
. Redrawn from results
of [Sm6].
4.1
Electrochemical Reactions
This section and the next are dedicated to the basics of the silicon-electrolyte con-
tact with focus on the electrolyte side of the junction and the electrochemical reac-
tions accompanying charge transfer. The current across a semiconductor-electro-
lyte junction may be limited by the mass transport in the electrolyte, by the ki-
netics of the chemical reaction at the interface, or by the charge supply from the
electrode. The mass transport in the bulk of the electrolyte again depends on con-
vection as well as diffusion. In a thin electrolyte layer of about a micrometer close
to the electrode surface, diffusion becomes dominant. The stoichiometry of the ba-
sic reactions at the silicon electrode will be presented first, followed by a detailed
discussion of the reaction pathways as shown in Figs. 4.14.4.
4.1.1
Overall Reactions
While the electrochemical reaction in the cathodic regime is similar for most com-
monly used aqueous electrolytes, the anodic reaction depends on composition and
pH of the electrolyte.
In the cathodic regime the silicon atoms of the electrode do not participate in
the chemical reaction. Therefore, an n-type or a strongly illuminated p-type silicon
electrode behave like a noble metal electrode and hydrogen evolution or metal
plating reactions are observed. For the case of an aqueous electrolyte free of metal
ions the main reaction is electrochemical hydrogen evolution according to:
2e
2H
3H
2
4:1
It is shown that the rate-limiting step in the photoelectrochemical evolution of
hydrogen in an HF electrolyte is linearly dependent on the excess electron concen-
tration at the surface of the p-type silicon electrode. The rate of this step does not
depend on the electrode potential and the H
+
concentration in the solution, but is
sensitive to the surface pretreatment [Sc11]. The plateau in the IV curve, slightly
51
4
The Electrochemical Dissolution of Silicon
Electrochemistry of Silicon: Instrumentation, Science, Materials and Applications.
Volker Lehmann
Copyright 2002 Wiley-VCH Verlag GmbH
ISBNs: 3-527-29321-3 (Hardcover); 3-527-60027-2 (Electronic)
cathodic of OCP, as shown in Fig. 3.2 (regime 2) can be ascribed to the hydrogen
evolution overpotential. This assumption is supported by the fact that upon addi-
tion of H
2
O
2
to the electrolyte a convection-dependent cathodic current density
can be observed in this regime.
Under cathodic bias hydrogen or deuterium may be incorporated in p-type elec-
trodes in concentrations of up to the doping level and depth in the order of a mi-
crometer. In the hydrogenated region the acceptors become passivated, causing a
dramatic increase in resistivity [Mi5]. Acceptor compensation is also observed dur-
ing CMP if copper ions are present in the slurry [Pr2]. Since precleaning in the
electrochemical investigations, mentioned above [Mi5], was also performed in a
copper-containing solution it can be speculated that the presence of Cu is a prere-
quisite for acceptor compensation.
Under anodic potentials in acidic electrolytes free of fluoride, silicon is passi-
vated by formation of an anodic oxide under comsumption of four holes (h
+
), ac-
cording to the reaction:
Si 2H
2
O4H
3SiO
2
4H
4:2
Details of the formation process and the properties of anodic oxides are dis-
cussed in Chapter 5.
In alkaline electrolytes silicon is dissolved chemically at OCP. From the observed
stoichiometric ratio of approximately 2H
2
/Si, a overall reaction can be derived [Fi1]:
Si 2OH
4H
2
O 3SiOH
2
6
2H
2
4:3
For moderate anodic potentials the dissolution rate becomes enhanced and the
ratio of H
2
/Si is reduced due to the contribution of the electrochemical reaction
path [Pa6]. At the electrode surface SiH as well as SiOH groups are present. For
higher concentrations of the silicate monomer produced by reaction (4.3) silicate
polymerization takes place [Ni6]. Passivation takes place for more anodic poten-
tials, due to formation of SiOSi bonds according to reaction (4.2).
In acidic electrolytes with fluoride, silicon is stable at OCP, while electrochemi-
cal dissolution takes place for anodic potentials. For anodic current densities be-
low the critical current density J
PS
PS is formed and the electrolyte-electrode inter-
face is found to be SiH covered. Species active in the dissolution process are HF,
(HF)
2
and HF
2
2
h
3SiF
2
6
2HF H
2
e
4:4
If the anodic potential is increased, the current density becomes larger than J
PS
and dissolution occurs via an intermediate anodic oxide film. Hence the reaction
can be separated into electrochemical oxide formation according to reaction (4.2)
and chemical dissolution of the oxide due to HF, (HF)
2
or HF
2
[So2]:
SiO
2
2HF
2
2HF 3SiF
2
6
2H
2
O 4:5
4 The Electrochemical Dissolution of Silicon 52
The overall reactions described above give no information about the actual reac-
tion pathways on the atomic scale. Models proposed for the dissolution pathways
in alkaline and in hydrofluoric electrolytes are discussed below.
4.1.2
Chemical Dissolution of Si in Alkaline Solutions
The common etching behavior of different alkaline solutions, as discussed in Sec-
tion 2.3, is attributed to H
2
O and OH
, as shown in step
1 of Fig. 4.1. An SiOH bond is established under formation of H
2
. In step 2 the
second SiH bond of the Si surface atom is replaced by SiOH under formation
of another H
2
. The attraction of electrons by the ligands now weakens the silicon
backbonds sufficiently that water can break these bonds as shown in steps 3 and
4 of Fig. 4.1. The silicon surface atom is dissolved with SiO
2
(OH)
2
2
being the fi-
nal etching product and the surface remains covered with SiH bonds [Pa4]. If
step 1 is rate limiting the silicon surface will be hydrogen passivated and hydro-
phobic. Whereas a rate limitation by step 3 will lead to a silanol-covered, hydro-
philic surface. As shown in Fig. 2.1, the highest degree of hydrophobicity has
been observed at pH=11, indicating the highest concentration of SiH surface
groups. Under anodic potentials below PP the alkaline dissolution process is en-
hanced as a result of an electrochemical contribution. While the dissolution rate
increases, the ratio of H
2
per dissolved Si decreases. Significant anodic currents
observed for n-type electrodes kept in the dark indicate electron injection during
anodic dissolution [Pa6].
In the frame of this model the anisotropy of alkaline etchants, the low etch rate
observed for (111) oriented silicon surfaces, can be interpreted as an insufficient
polarization of three silicon backbonds by only one SiOH surface bond that can
4.1 Electrochemical Reactions 53
Fig. 4.1 Reaction scheme proposed for the chemical dissolution of (100) oriented silicon
surfaces in alkaline solutions.
be established on a monohydride surface, as shown in Fig. 4.2. Whereas two Si
backbonds are polarized by two SiOH bonds for the dihydride surfaces present
for (100) or (110) surfaces, as shown in Fig. 4.1. The weak polarization induced by
only one SiOH efficiently reduces the probability of nucleophilic attack of the
backbonds [Gl1, Al2]. Scanning tunneling microscopy (STM) investigations re-
vealed that the etching of (111) surfaces proceeds at kink-sites present at the
edges of terraces, while the surface of the terrace itself is not attacked [Al1]. There-
fore, etch rates observed for (111) silicon surfaces are sensitive to minute misor-
ientations of the substrate. At high DOC, high p-type doping or if holes are avail-
able at the interface, due to an applied potential, direct attack of the terrace sur-
face seems to be more likely [Be23] and the anisotropy becomes significantly re-
duced, as for example indicated in Fig. 2.4. Why the etch rate ratio between (100)
and (110) planes is affected by temperature or additions of organic or inorganic
agents is not fully understood [Zu1].
4.1.3
Divalent Electrochemical Dissolution of Si in HF
The first models for the electrochemical dissolution process of silicon in HF as-
sumed a fluoride-terminated silicon surface to be present in electrolytes contain-
ing HF [Ge6, Du3]. However, by IR spectroscopy it was found that virtually the
whole surface is covered by hydride (SiH) [Ni3]. No evidence of SiF groups is
found in IR spectra independent of HF concentration used [Ch9]. This is surpris-
ing insofar as the SiF (6 eV) bond is much stronger than the SiH (3.5 eV) bond,
and so it cannot be assumed that SiF is replaced by SiH during the electro-
chemical dissolution. This led to the conclusion that if a silicon atom at the sur-
face establishes a bond to a fluorine atom it is immediately removed from the sur-
face.
This fast removal of SiF species can be ascribed to the weakening of the Si
backbonds induced by the strong polarizing effect of F [Ub1]. The weak back-
bonds are then attacked by HF or H
2
O. This reaction scheme for the dissolution
process is supported by quantum-chemical calculations [Tr1]. The observed disso-
lution valence of two for J <J
PS
[Tu1], the photocurrent doubling indicating injec-
tion of an electron [Me11, Ma1], and the observed proportionality between elec-
tron injection current and SiF bond density [Be22] are experimental findings that
are in support of the divalent dissolution mechanism, as shown in Fig. 4.3 [Le1,
Ge7, Ho6].
4 The Electrochemical Dissolution of Silicon 54
Fig. 4.2 The chemical dissolution rate of (111) orien-
ted silicon surfaces in alkaline solutions is negligible
as a result of insufficient polarization of the Si back-
bonds by only one OH group.
Divalent dissolution is initiated by a hole from the bulk approaching the silicon-
electrolyte interface which allows for nucleophilic attack of the Si atom (step 1 in
Fig. 4.3). This is the rate-limiting step of the reaction and thereby the origin of
pore formation, as discussed in Chapter 6. The active species in the electrolyte is
HF, its dimer (HF)
2
, or bifluoride (HF
2
to dif-
fuse through the oxide film to the interface and to establish an SiOSi bridge un-
der consumption of two holes, according to Fig. 4.4, steps 1 and 2. Details of
anodic oxide formation processes are discussed in Chapter 5. This oxide film pas-
sivates the Si electrode in aqueous solutions that are free of HF.
The dissolution of the anodic oxide in HF involves nucleophilic attack of Si by
fluorine, bound in HF, (HF)
2
or HF
2
], which
indicates that HF and HF
2
] and (f )
2[HF
2
1
0
1
4
c
m
2
,
c
o
r
r
e
s
p
o
n
d
i
n
g
t
o
a
(
1
0
0
)
S
i
s
u
r
f
a
c
e
,
a
n
d
a
b
u
l
k
d
e
n
s
i
t
y
o
f
S
i
a
t
o
m
s
o
f
5
1
0
2
2
c
m
3
.
T
h
e
c
h
e
m
i
c
a
l
c
o
m
p
o
s
i
t
i
o
n
o
f
t
h
e
p
o
r
o
u
s
l
a
y
e
r
f
o
r
a
d
i
h
y
d
r
i
d
e
s
u
r
f
a
c
e
c
o
v
e
r
a
g
e
a
n
d
f
o
r
a
0
.
5
n
m
t
h
i
c
k
(
s
t
o
i
c
h
i
o
m
e
t
r
i
c
)
n
a
t
i
v
e
o
x
i
d
e
l
a
y
e
r
a
r
e
a
l
s
o
g
i
v
e
n
.
be about 600900 m
2
cm
3
. Note that the latter value is comparable to the internal
surface of activated charcoal. Pores of similar diameters were also reported for PS
grown on highly doped p-type substrates, however, the specific surface was found
to be somewhat smaller (300 m
2
cm
3
) [He4, Ca8, Ru3].
Hydrogen, which covers the internal surface of PS, can also be used to estimate
its structural dimensions. IR measurements indicated a stoichiometry of roughly
SiH for electrochemically prepared micro PS [Be2]. If dihydride groups are as-
sumed to cover the internal surface, every second atom must be a surface atom.
This is the case for a cube of about 1000 atoms that has a diameter of approxi-
mately 2 nm. A stoichiometry of SiH
0.4
obtained by thermodesorption measure-
ments points to a crystallite diameter in the order of 4 nm [Pe2]. The chemical
composition for a hydride coverage surface and for a 0.5 nm thick native oxide
layer are given in Table 6.1.
6.5
Mechanical Properties and Drying Methods
The mechanical properties of an array of macro PS can be calculated based on the
properties of bulk Si and the specific array geometry. The internal surface of
macro PS is usually too small to contribute significantly to the mechanical proper-
ties of the material. The stress induced by a native oxide coverage, for example, is
negligible for macro PS, and so an as-prepared macroporous layer is found to be
free of intrinsic stress. However, processes such as diffusion doping, oxidation or
deposition of a nitride layer are sufficient to induce considerable stresses, which
in turn may lead to bowing of the wafer. The drying of macroporous material is
not critical. Only if the porosity becomes high enough for pores to join (d/i =1 in
Fig. 6.7), leaving free-standing silicon pillars or sheets, do these structures tend to
bend and stick together during drying.
The strain induced by surface forces increases if the internal surface is en-
larged, which is the case for meso- and microporous silicon. For as-prepared
meso PS an increase in lattice spacing between 10
3
and 10
4
[Ba8] compared to
the bulk value is reported, while for micro PS values of up to 1% have been re-
ported [Yo2, Ya7]. This strain leads to a macroscopic curvature of the anodized sili-
con substrate. Microscopically compressive as well as tensile stresses seem to be
present in the distorted network of highly microporous samples, which increases
or decreases the interatomic spacing by several per cent [Le10]. The strain has
been proposed to originate from interactions between the SiH groups covering
the internal surface of PS [Su4, Bu5].
The thermal expansion coefficient of bulk silicon is positive at RT (2.610
6
K
1
), but becomes negative below 120 K. The thermal expansion coefficient of mi-
cro PS for heating from 290 to 870 K is found to be negative (4.310
6
K
1
),
which can be ascribed to hydrogen desorption and oxidation of the inner surface
[Di7]. For meso PS the thermal expansion coefficient was found to increase with
porosity in the temperature regime between 90 K and 300 K, from 0.410
6
K
1
to
6 Electrochemical Pore Formation 114
1.610
6
K
1
[Fa8]. The latter effect, which has been ascribed to surface stress,
also, is found to be reversible, in contrast to measurements at higher tempera-
tures.
The elastic properties of PS depend on its microstructure and porosity. The
Youngs modulus for meso PS, as measured by X-ray diffraction (XRD) [Ba8],
acoustic wave propagation [Da5], nanoindentation [Be13] and Brillouin spectrosco-
py [An2], shows a roughly (1p)
2
dependence. For the same values of porosity
(70%), micro PS shows a significantly lower Youngs modulus (2.4 GPa) than
meso PS (12 GPa). The Poisson ratio for meso PS (0.09 for p=54%) is found to
be much smaller than the value for bulk silicon (0.26) [Ba8].
The hardness, defined as the resistance to plastic deformation, of microporous
silicon decreases with porosity p from the bulk silicon value of about 11.5 GaP to
values around 4 GaP for porosities in the order of 75% following a (1p)
2/3
depen-
dence. For porosities above 75% a further decrease in hardness is observed. The
hardness of PS formed on highly doped p-type substrates is found to be some-
what less than that observed for low doped substrates, which may be caused by
the more columnar structure of meso PS [Du5].
The thermal conductivity of bulk silicon (148 W K
1
m
1
) is dominated by pho-
nons; electronic contributions are negligible. Due to restrictions of the mean free
path of phonons in the porous network the thermal conductivity of micro PS is
reduced by two or three orders of magnitude at RT, compared to the bulk value.
Because of the larger dimensions of its network, meso PS shows a thermal con-
ductivity several times larger than that of micro PS, for the same value of porosi-
ty. Thermal oxidation at low temperatures (3008C) is found to decrease the ther-
mal conductivity of meso PS by a factor of about 0.5 [Pe9]. In contrast to bulk Si
the thermal conductivity of PS is found to decrease with decreasing temperature
[Be21, La4, Ge9, Ly1].
The surface tension at liquid-gas interfaces is responsible for the pressure in a
capillary liquid. The pressure increases linearly with decreasing pore diameter.
The cracking and peeling of meso and micro PS films of high porosity during
drying, as shown in Fig. 6.11, is a consequence of these capillary forces. This ef-
fect and the resulting morphology is the same as in a dried lake-bed. Cracking of
a PS film can be expected if the film exceeds a critical thickness D, which is ap-
proximated by:
D = (d=2c
L
)
2
E
Si
(1 p)
3
c
Si
(6:9)
with d being the pore diameter, p the porosity of the film, c
L
the surface tension
of the liquid, c
Si
the average surface tension of silicon (1 J m
2
) and E
Si
the
Youngs modulus (140 GPa). Figure 6.12 shows values of the critical thickness D
as a function of p for different d. The problem of film cracking due to capillary
forces can be reduced if the samples are not dried in water (c
L
=72 mJ m
2
) but in
a liquid of low surface tension, e.g. pentane (c
L
=14 mJ m
2
) [Be12]. In this case
pure ethanol should be used as a transition liquid between water and pentane,
which are not miscible.
6.5 Mechanical Properties and Drying Methods 115
Cracking of PS because of capillary forces can be circumvented if one avoids
crossing the liquid-vapor boundary in the phase diagram of the solvent. This is
the case for supercritical drying [Ca4] or freeze drying [Am1], as shown in the in-
set of Fig. 6.12.
Detailed measurements of stress have revealed that microporous layers on bulk
silicon are under compressive stresses that decrease with increasing porosity,
6 Electrochemical Pore Formation 116
Fig. 6.11 Optical microscopy of the surface
of a 25 lm thick micro PS layer on the Si sub-
strate, showing (a) crack formation in the wet
film (t =0 s) and (b) subsequent (t =3 s)
shrinkage of the micro PS pieces due to capil-
lary forces during drying of propanol [Le19].
(c) SEM micrograph of a disintegrated micro
PS film formed at high current density
(SEM of region A in Fig. 4.17).
Fig. 6.12 Critical PS thickness above which
cracking of the PS film during evaporation of
water is expected, according to Eq. (6.9). Note
that drying in pentane or ethanol increases the
critical thickness by a factor of about 26 or 11,
respectively. Inset: Cracking can be avoided by
freeze drying (1) or critical point drying (2),
which removes the solvent without crossing the
fluid-gas boundary in the phase diagram (3).
from about 5 MPa to values around 1 MPa if kept wet after formation. During the
drying process the stress is tensile up to values around 10 MPa, which leads to
cracking of highly porous samples. Dry PS samples are under compressive stress
again, however the compression is by about 12 MPa lower than that observed un-
der wet conditions. The latter effect is ascribed to attractive forces due to instanta-
neous dipole moments of the surface hydrogen atoms, known as Van der Waals
forces [Gr6]. The stress-time curve during the drying process is shown in
Fig. 6.13. Re-immersion of the sample and repetition of the cycle showed that the
changes in stress are fully reversible. It should be noted that freshly dried PS
samples are strongly hydrophobic and re-wetting is suppressed without addition
of a surfactant to the aqueous solution.
The compressive stress in PS decreases with annealing temperature and is re-
versed for temperatures above 3508C, indicating that the PS is under tension
[Su4]. These changes are reversible if the hydrogen coverage is restored by a short
HF dip; however, at annealing temperatures above 5008C the PS microstructure is
irreversibly changed [Ha4].
6.6
Chemical Composition and Ageing Effects
In micro PS a considerable number of silicon atoms are located at the inner sur-
face, as shown in Table 6.1. Therefore, oxidation or absorption of hydrocarbons in
the order of a monolayer will significantly change the chemical composition of
the material. It is therefore essential for the comparison of chemical data if they
have been obtained in situ, directly after preparation, or after longer storage peri-
6.6 Chemical Composition and Ageing Effects 117
Fig. 6.13 Stress-time evolution during the drying
of a PS film grown on a low doped p-type sub-
strate in ethanol [Gr6].
ods. In the first part of this section the freshly prepared sample will be discussed,
while ageing effects are discussed in the second part.
Infrared (IR) spectroscopy is a perfect tool for investigating the chemical com-
position of PS because it can be used in situ and the frequency positions of the
characteristic absorption band indicate not only the element but also its binding
configuration, as shown in Fig. 6.14. The in situ IR investigation of bulk silicon
surfaces in dilute HF solutions has revealed that the silicon surface is terminated
with hydrogen. Hydrogen-associated Si fluorides, such as SiH
2
(SiF), SiHF
3
and
SiH
2
F
2
, expected from the dissolution reaction as shown in Fig. 4.3, have not
been detected by IR. If they exist, their lifetime is shorter than 0.3 ms [Ve4]. The
total fluorine concentration in PS is usually below 1% [Ea1] and is below the de-
tection limit after a short DI water rinse. In the latter case only the SiH, the Si
H
2
and the SiH
3
vibrational modes are observed. No SiO surface species are de-
tectable [Ni3]. The same is basically true for PS if it is formed in aqueous HF: no
oxygen is detectable in electrochemically formed PS during, or right after, the
preparation [Ve4, Li3]. Only in very dilute solutions ([HF] < 0.1 M) are submono-
layer oxide films found to be present at the porous-bulk interface during forma-
tion at current densities well below J
PS
[Be20]. The PS stoichiometry is reported
to be between SiH and SiH
2
. These results for electrochemically formed PS are in
contrast to PS formed chemically in an HFHNO
3
solution. In this case the com-
position is reported to be between H
2
SiO and HSiO
1.5
[Be2]. Significant oxygen
incorporation is also observed for PS formed by anodization in HF under UV illu-
mination, while addition of H
2
O
2
to the HF electrolyte showed no increase in the
oxygen content of PS [Ya11].
On storage in ambient air or immersion in H
2
O
2
the SiOSi vibrational mode
becomes detectable for electrochemically formed PS, without a significant change
in the number of SiH
x
bonds, as shown in Fig. 6.15 [Th5]. This supports the as-
sumption that oxygen penetrates the Si lattice and breaks the backbonds of the
6 Electrochemical Pore Formation 118
Fig. 6.14 Characteristic dielectric functions
(imaginary part), as observed by IR spectros-
copy for the solid component of as-prepared
PS (left) and an oxidized sample (right),
with the most important vibrational modes
assigned as indicated. After [Th6].
SiH surface groups, as observed for native oxidation of bulk Si surfaces [Gr5].
The oxidation at ambient conditions is accompanied by a desorption of traces of
SiH
4
from the surface [La1, Ca2]. An investigation of the reaction of atomic hydro-
gen with hydrogenated PS revealed that trihydride is the surface precursor of
SiH
4
. As far as the etch rate in atomic hydrogen is concerned (30 nm min
1
,
10 mTorr, RT), the inner PS surface still behaves like the single crystal silicon sur-
face [Gl2].
As observed for the case of hydrophobic bulk silicon surfaces [Li11], PS attracts
organic matter from the ambient atmosphere. The hydrocarbon coverage may be
in the order of a monolayer, which leads to C/Si atom ratios as high as 1 [Hi6].
Upon longer storage in ambient air at RT the hydrogen-passivated surface of PS
gradually converts to a heavily hydrocarbon-contaminated native oxide [Ca3]. It is
worth mentioning that these processes, known as ageing, usually increase the PL
efficiency of the sample [Ma6].
6.6 Chemical Composition and Ageing Effects 119
Fig. 6.15 The increase in the
SiOSi vibrational modes
(9801200 cm
1
) upon storage of
a high porosity (75%) micro PS
layer formed on a p-type substrate
(0.2 X cm) in ambient air. The
SiH
2
scissors mode (905 cm
1
)
shows little time dependence.
After [Th7].
Fig. 6.16 The normalized integrated absor-
bance of SiH and SiH
2
species (filled sym-
bols), the PL intensity (open circles) and the
specific surface area (open diamonds) as a
function of annealing temperature for a
porous layer formed on a low doped p-type
substrate. After data of [Ha4].
The absorption of water in the microporous silicon network was evaluated by
exposing the porous samples to H
2
O and D
2
O vapor at low pressure (10
5
Torr).
FTIR spectral analyses revealed that water dissociates at the PS internal surface to
SiH (SiD) and SiOH (SiOD). Upon annealing to 3508C, SiOSi is formed from
the SiOH (SiOD) groups while H
2
(D
2
) is desorbed [Gu3].
If a freshly prepared micro PS sample is annealed, hydrogen desorbs from the
inner surface, as shown in Fig. 6.16. Thermodesorption measurements using a
low ramping rate of 1 K min
1
showed that most hydrogen is initially bound as
SiH
2
, which converts to monohydride SiH at 610 K, producing a first effusion
peak [Pe2]. A detailed IR investigation revealed that the monohydride SiH is pre-
sent in a dimer structure (SiH)
2
formed by a surface reconstruction after the first
desorption peak [Og3], while at 720 K the rest of the hydrogen from the dimer is
desorbed, resulting in a second peak. A stoichiometry of SiH
0.4
has been obtained
by these measurements. In an earlier study, however, the initial concentrations of
monohydride and dihydride in PS (measured by FTIR) were reported to be the same
within 5%, while desorption peaked at 680 and 810 K respectively for fast ramping
(8 K s
1
). Hydrogen on the PS surface was observed to be very similar to hydrogen
on Si (100) 21. A desorption activation barrier of 1.86 eV (SiH
2
) and 2.82 eV
(SiH) has been observed. These activation barriers yield upper limits of 3.19 eV
(SiH
2
) and 3.67 eV (SiH) [Gu2]. For temperatures above the second desorption
peak, a decrease in the specific surface area by more than one order of magnitude
is found, which indicates a significant change in the PS microstructure [Ha4].
6.7
Electrical Properties of Porous Silicon
The focus of this section is charge transport in PS, and electrical properties such
as resistivity, carrier mobility, capacitance and photoconductivity are discussed.
6.7.1
Contacts to Porous Films
The measurement of properties such as the resistivity or dielectric constant of PS
requires some kind of contact with the PS layer. Evaporation of a metal onto the
PS film-covered silicon sample produces a metal/PS/Si sandwich, which behaves
like an MIS structure with an imperfect insulator. Such sandwich structures
usually exhibit a rectifying behavior, which has to be taken into account when de-
termining the resistivity [Si3, Be14]. This can be circumvented by four-terminal
measurements of free-standing PS films, but for such contacts the applied electric
field has to be limited to rather small values to avoid undesirable heating effects.
An electrolytic contact can also be used to probe PS films, but the interpretation
of the results is more complicated, because it is difficult to distinguish between
ionic and electronic contributions to the measured conductivity. The electrolyte in
the porous matrix may short-circuit the silicon filaments, and wetting of PS in-
6 Electrochemical Pore Formation 120
creases the average dielectric constant and leads to screening of surface charges,
which may change the values of the electrical properties of PS by orders of magni-
tude.
6.7.2
Definition of Conductivity and the Macroporous Case
The conductivity of porous structures depends on the size of the conducting fila-
ments making up the silicon skeleton [An5] and will therefore be discussed for
each of the three size regimes separately.
For a homogeneous porous layer a specific conductivity r can be defined:
r = ID=VA (6:10)
where I is the current due to an applied potential V across the PS film of thick-
ness D and area A. The specific resistivity of this film is 1/r.
According to the macropore formation mechanisms, as discussed in Section
9.1, the pore wall thickness of PS films formed on p-type substrates is always less
than twice the SCR width. The conductivity of such a macroporous silicon film is
therefore sensitive to the width of the surface depletion layer, which itself depends
on the type and density of the surface charges present. For n-type substrates the
pore spacing may become much more than twice the SCR width. In the latter
case and for macro PS films that have been heavily doped after electrochemical
formation, the effect of the surface depletion layer becomes negligible and the
conductivity is determined by the geometry of the sample only. The conductivity
parallel to the pores is then the bulk conductivity of the substrate times 1p,
where p is the porosity.
6.7.3
Conductivity of Mesoporous Silicon
For mesoporous silicon, where QC effects are still negligible, depletion of the po-
rous skeleton due to charges located at the large internal surface becomes the
dominant factor. A mesoscopic void like an etched pore compensates charges
from shallow dopants by capture at dangling bond (DB) centers and a depletion
region is formed around such a void. Calculations show that a spherical void of
10 nm diameter can compensate about 25 charges of either sign [No2]. Even if we
assume that the majority of surface bonds will reconstruct and that the effective
DB density in meso PS is as low as 10
12
cm
2
, the large internal surface area of
about 100 m
2
cm
3
will still generate an effective density of states (DOS) in the or-
der of 10
18
cm
3
. This high density of surface states is compatible with the pin-
ning of the Fermi level at mid-gap. As a result, the mesoporous sample becomes
insulating because of a complete trapping of free majority carriers at the inner
surface, which produces a depletion of the porous volume. This idea is supported
by optical transmission measurements, which show that the sub-bandgap absorp-
6.7 Electrical Properties of Porous Silicon 121
tion due to free carriers is significantly suppressed in the porous material com-
pared to the bulk substrate, as shown in Fig. 7.8. Detrapping of the charge car-
riers has been observed for a meso PS film exposed to a polar gas, for example
NO
2
[Bo7]. This effect has been interpreted as a screening of surface traps by ab-
sorbed polar species [St11, Le14]. Hall effect measurements show that the concen-
tration of free electrons and their mobility in meso PS formed on highly doped n-
type silicon is about 10
13
cm
3
and 30 cm
2
V
1
s
1
respectively [Si2], and so the
material behaves similarly to intrinsic Si. It should be noted that the reduction of
free carriers is not caused by a reduction of dopant atoms due to a preferential
etching of these atoms compared to silicon atoms during the PS formation pro-
cess [Le14]. The resistivity of mesoporous silicon varies between 10
4
and
10
7
X cm, depending on substrate doping density and anodization conditions.
Screening of the surface states by ions present in an electrolyte may increase the
conductivity of wet meso PS to values comparable to those of the bulk substrate.
The temperature dependence of meso PS resistivity can be understood if Cou-
lomb blockade effects are taken into account. For a silicon nanocrystallite to be-
come conducting, it must first overcome the Coulomb blockade barrier, E
B
[Le14,
Ha14, Mi8]:
DE
B
= e
2
=2C (6:11)
where e is the electron charge and C the self-capacitance of an isolated, spherical
Si nanocrystal, which is given by:
C = 2pe
r
e
0
d (6:12)
Here d is the crystallite diameter, e
r
and e
0
are the dielectric constants of the
surrounding medium and a vacuum, respectively. The probability of electrons
overcoming the barrier shows an Arrhenius dependency on temperature. A signif-
icant increase in conductivity of the PS film, however, is only expected if the den-
sity of unblocked sites in the porous network is high enough to form continuous
pathways from one electrode to the other. The continuous connection of un-
blocked crystallites is expected to produce an abrupt increase in conductivity at a
critical temperature T
c
, as shown in Fig. 6.17. A Coulomb blockade indicated by a
plateau in the IV characteristic at low applied bias is shown in the inset of this
figure. Above T
c
, the IV characteristic becomes linear and the activation energy
approaches the bulk silicon value of about 0.5 eV for large crystallites, while below
T
c
the activation energy is significantly lower. T
c
is expected to increase with de-
creasing crystallite mean size [Mi8].
6.7.4
Conductivity of Microporous Silicon
The resistivity of micro PS in dry air at RT is observed to increase with porosity
and shows high values ranging from 10
7
to 10
14
X cm, with the maximum of the
6 Electrochemical Pore Formation 122
distribution around 10
11
X cm. For micro PS-based EL devices a mobility of the
order of 310
4
cm
2
V
1
s
1
has been reported [Co5]. The conductivity shows a V
1/2
dependence on applied bias and is thermally activated, with an activation energy
of 0.30.7 eV. The resistivity increases upon cooling to 200 K by five orders of
magnitude and is found to be roughly constant below 150 K, as shown in
Fig. 6.18 [Ko2], which is similar to the behavior observed for meso PS, as shown
in Fig. 6.17. The observed anti-correlation between resistivity and PL intensity is a
result of the long lifetime of the luminescent state. If the probability of a charge
carrier leaving the crystallite is higher than that of the radiative recombination
channel, the PL becomes quenched.
For micro PS a decrease in the specific resistivity by two or three orders of mag-
nitude is observed if the dry material is exposed to humid air [Ma8] or vapors of
polar solvents, e.g. methanol [Be6]. This sensitivity of PS to polar vapors can be
used to design PS-based gas sensors, as discussed in Section 10.4. This change in
resistivity with pore surface condition becomes dramatic if the pores are filled
with an electrolyte. From the strong EL observed under low anodic as well as low
cathodic bias in an electrolyte it can be concluded that micro PS shows a conduc-
tivity comparable to that of the bulk substrate under wet conditions [Ge8]. Diffu-
sion doping has been found to reduce the PS resistivity by more than five orders
of magnitude, without affecting the PL intensity [El1].
The interpretation of these different experimental observations is complicated
by the fact that for micro PS, not only do the Coulomb blockade effects have to be
6.7 Electrical Properties of Porous Silicon 123
Fig. 6.17 Temperature dependence of the
current through a metal-meso PS-silicon
sandwich for different applied potentials
(3 mX cm n-type, 100 mA cm
2
, 24 lm,
8 mm
2
evaporated aluminum contacts,
freshly prepared). The IV curves for 220 and
240 K, shown in the inset, indicate a sharp
transition from a Coulomb blockade
characteristic to a linear one.
taken into account, but also quantum size effects. The non-linear electric field de-
pendence has been interpreted in terms of the Onsager model: an electron at the
Fermi level can jump from its crystallite and leave behind a positive trap, which
gives rise to a Poole-Frenkel-like functional behavior for temperatures above 200 K
[Be8]. AC measurements indicate hopping of charge carriers and no polarization
of atoms or particles [Be7]. It has been argued that the activation energy of 0.5 eV
is too large for hopping, and other transport channels such as thermionic emis-
sion above energy barriers [Ko10] and transport in extended states have been sug-
gested [Lu2]. For a more detailed discussion of conduction mechanisms in PS, see
[Be15].
6.7.5
Photoconductivity of Porous Silicon
So far all values of resistivity have referred to measurements performed in the
dark. If the porous layer is illuminated, a photoconductivity is found that is al-
most independent of polarity and value of the applied bias, but is sensitive to the
wavelength of the light. The photoresponse of micro PS is largest for wavelengths
between 500 and 600 nm, because at longer wavelengths the light is not effec-
tively absorbed, while short-wavelength illumination will only excite the topmost
region of the porous layer [Se13]. Even for moderate illumination intensities the
resistivity of dry micro PS is reduced by two or three orders of magnitude com-
pared to the value observed in the dark [Ko2]. This is because the free charge car-
rier concentration in PS is very low compared to the bulk material. It increases by
orders of magnitude under illumination.
From photocarrier grating measurements at low laser intensities (100 mWcm
2
),
the value of the ambipolar diffusion coefficient of meso PS is deduced to be about
6 Electrochemical Pore Formation 124
Fig. 6.18 Temperature dependence of
the micro PS resistivity (filled circles)
and the PL intensity (open squares)
for a self-supporting PS film of 40 lm
thickness. Anodization conditions in
ethanoic HF (1:1) are indicated on the
lower left. After [Ko2].
10
6
cm
2
s
1
, which gives a small diffusion length of a few tens of nanometers be-
tween the repeated trapping of photogenerated carriers [Sc13]. This is in contrast
to diffusion constants of 550 cm
2
s
1
reported for micro PS. The latter results were
obtained for excitation energies of up to 170 mWcm
2
, where bipolar plasma-like
diffusion prevails [To1]. The use of porous polysilicon as cold cathode, as discussed
in Section 10.6, indicates that charge drift lengths in the order of 1 lm are realistic at
high fields (10
5
Vcm
1
) [Kl4, Se13]. Measurements of the photoconductivity of PS
using transparent liquid contacts show that the porous structure is not necessarily
short-circuited by a conducting electrolyte inside the pores [Se10].
6.7.6
Carrier Mobility in Porous Silicon
The application of an electric field E to a conducting material results in an aver-
age velocity v of free charge carriers parallel to the field superimposed on their
random thermal motion. The motion of charge carriers is retarded by scattering
events, for example with acoustic phonons or ionized impurities. From the mean
time s between such events, the effective mass m
+
of the relevant charge carrier
and the elementary charge e, the velocity v can be calculated:
v = esE=m
+
(6:13)
At low electric fields v is proportional to E and a mobility l=e s/m
+
can be de-
fined. The mobility of electrons and holes in bulk silicon is shown in the figure
on the inner front cover of this book.
Measurements of mobility in PS suffer from the fact that the number of free
charge carriers is usually small and very sensitive to illumination, temperature
and PS surface condition. Hall measurements of meso PS formed on a highly doped
substrate (10
18
cm
3
, bulk electron mobility: 310 cm
2
V
1
s
1
) indicated an electron
mobility of 30 cm
2
V
1
s
1
and a free electron density of about 10
13
cm
3
[Si2].
Values reported for effective mobility of electron and hole space charges in micro
PS are about five orders of magnitude smaller (10
3
to 10
4
cm
2
V
1
s
1
) [Pe10].
The latter values are much smaller than expected from theoretical investigations
of square silicon nanowires [Sa9]. For in-depth information about carrier mobility
in PS see [Si6].
6.7.7
Dielectric Constant of Porous Silicon
The macroscopic dielectric constant of PS is very sensitive to the frequency range
investigated. For very high frequencies the dielectric constant can be obtained by
optical measurements. In this regime the effective medium approximation has
been found to be an adequate description [Th3]. For lower frequencies the dielec-
tric constant e is commonly obtained by capacitance measurements of metal/PS/
Si structures. AC conductivity measurements of such structures indicate a hop-
6.7 Electrical Properties of Porous Silicon 125
ping of charge carriers and no polarization of atoms or particles [Be7]. The resis-
tivity is found to decrease with increasing frequency, as shown in Fig. 6.19, a be-
havior typical of disordered solids. The full line in Fig. 6.19 corresponds to a di-
electric constant of about 2.3, which is in agreement with values of refractive in-
dex obtained by IR ellipsometry. This indicates that the Bruggeman effective me-
dium model is still valid in the MHz regime for micro and meso PS of porosities
greater than 50%. The relative dielectric constant in this regime can therefore be
approximated by e =1+10.4(1p). For lower frequencies and low porosity, how-
ever, higher values of the dielectric constant may be observed. Note that for par-
tially oxidized micro and meso PS films the dielectric constant may reach values
in the order of 10
3
for the low-frequency regime (1 kHz) [Un1]. Completely oxi-
dized PS, in contrast, shows low values of e, which correspond to SiO
2
(e =3.9) of
a certain porosity. For more information about PS layer capacitance see [Co3].
6 Electrochemical Pore Formation 126
Fig. 6.19 Frequency dependence of the con-
ductivity at different temperatures for a micro
PS film (5 X cm p-type, 30 mA cm
2
, ethanoic
HF). The transition from a frequency regime
of 0.95 slope (full line) to a regime of 0.5
slope (broken line) is indicated by arrows.
After [Be7].
7.1
Micropore Formation Mechanism
This section is devoted to the mechanism responsible for the formation of micro-
pores on silicon electrodes. Micropore formation on silicon electrodes illustrates
the importance of the pore position with respect to the neighboring pores, i.e. the
pore wall thickness.
If the dimensions of a silicon crystallite are reduced to a few nanometers, its
bandgap increases as a result of QC effects, as discussed in detail in Section 7.5.
The same holds true for the thin silicon walls separating the micropores. The in-
crease in bandgap energy DE
G
in the wall region compared to the bulk electrodes
produces an energy barrier for holes DE
V
, as schematically shown in Fig. 7.1a. If
DE
V
is larger than the bias-dependent energy of holes, the porous region becomes
depleted of holes and therefore passivated against further dissolution. It is energe-
tically more favorable for a hole to enter the electrolyte at the pore tip directly,
than via the porous structure [Le1]. Note that for the case of pore formation due
to QC no additional effect is required to explain the active state of the pore tip,
which is in contrast to passivation by an SCR. The pore tip is active, because it is
in contact with the unconfined bulk electrode, while the pore walls are passive
due to the close spacing of the pore tips. Direct consequences of this condition
are a distinct porous-bulk interface and significant values of porosity.
A few further conclusions can be drawn from this model. If, for example, the for-
mation current density is increased, the bandgap of the porous structure is expected to
increase, too. A higher current density requires an increase in the anodization bias,
thus holes have more energy to cross the barrier DE
V
. Therefore, the nanometer-sized
structures that form the porous skeleton will be etched further, until their bandgap is
large enough to become passivated again. The same is true for illumination during
anodization. By illumination with sufficiently energetic photons, electron-hole pairs
are generated in the porous structure. The photogenerated holes initiate further dis-
solution of the pore walls, as shown in Fig. 7.1b. This reduces their size and thereby
increases their bandgap energy until it is higher than the photon energy, which estab-
lishes passivation again. The increase in bandgap results in a blue shift of the PL from
PS formed under illumination, as shown in Fig. 7.10 [Ni2, Le10]. However, it is not
127
7
Microporous Silicon
Electrochemistry of Silicon: Instrumentation, Science, Materials and Applications.
Volker Lehmann
Copyright 2002 Wiley-VCH Verlag GmbH
ISBNs: 3-527-29321-3 (Hardcover); 3-527-60027-2 (Electronic)
only during the formation process of micro PS that the charge transfer is expected to
be sensitive to the degree of confinement. Generally any charge transfer to and in the
confined structures in micro PS depends on DE
V
and DE
C
. These effects dominate the
electrical and EL properties of the material, as discussed in Sections 6.7 and 10.5,
respectively.
An extension of this QC model, including tunneling probabilities between the
confined crystallites and the bulk, has been developed [Fr6]. The QC model for
microporous silicon formation, however, is still qualitative in character, and a
quantitative correlation between anodization parameters and the morphology and
properties of the porous structure is at yet beyond the capability of the model.
7.2
Morphology of Microporous Silicon
The dimensions of the microporous silicon skeleton are too minute to be resolved by
microscopy techniques that generate a direct image of the morphology, e.g. SEM.
However, an understanding of the morphology of microporous silicon has emerged
from a superposition of results from a multitude of investigations carried out by dif-
ferent techniques such as TEM, Raman spectroscopy (RS), XRD, BETor small angle
X-ray scattering (SAXS). A short summary of these investigations and a conclusive
idea of the morphology of the material are presented in this chapter.
7 Microporous Silicon 128
Fig. 7.1 (a) Schematic illustration of the in-
terface between PS and bulk silicon (top) and
the corresponding band diagrams for charge
transfer from the bulk to the porous skeleton
(middle) and from the bulk to the electrolyte
in the pore tip (bottom). Note that hole tran-
sition at the pore tip, indicated by a solid
arrow, is energetically favorable compared
to the transition into the pore wall (broken
arrow). (b) Illumination during anodization
leads to further thinning of the porous skele-
ton, because photons of an energy higher
than E
PS
are absorbed. After [Le24].
DE
V
E
G
+DE
G
7.2.1
Transmission Electron Microscopy
Even under magnifications of the order of 10
5
, micro PS looks rather homoge-
neous under the transmission electron microscope, as shown in Fig. 7.2a. This is
in contrast to the microstructure of meso PS, which can be revealed at this mag-
nification, as shown in Fig. 7.2b. It requires HREM techniques to reveal the struc-
7.2 Morphology of Microporous Silicon 129
Fig. 7.2 TEM (a, b) and HREM (c, d) micro-
graphs showing the interface between PS
(formed at 100 mA cm
2
in 1: 1 ethanoic HF)
and the (100) substrate for the cases of micro
PS (left) and meso PS (right). The insets
(e, f ) show the corresponding electron diffrac-
tion patterns. In the bottom figures lines are
drawn around crystallites as guides to the
eye. Note that micro PS crystallites may be
misaligned with the substrate, as indicated
by the lines in (c). Only crystallites rotated
around the solid axis, as shown in (g), reveal
lattice fringes in the image (c). After [Le15].
tural details of microporous material. In Fig. 7.2c lattice fringes of crystallites of
about 24 nm in diameter can be identified in a matrix that looks like an amor-
phous material. A closer inspection of the crystallites in Fig. 7.2c reveals a rota-
tion to each other by several degrees. Only crystallites that are rotated around an
axis parallel to the electron beam will reveal lattice fringes in the HREM image,
as shown in Fig. 7.2g. Crystallites rotated around the other two axes, indicated by
the broken axis in the inset to Fig. 7.2g, are imaged with a contrast like amor-
phous silicon. HREM images obtained under different angles to the sample
showed no anisotropy in the rotation of crystallites. These HREM investigations
[Le10] indicate that a high density of Si microcrystallites is present in micro PS.
However, no conclusions can be drawn from HREM images if an amorphous frac-
tion is present in the sample. It should be noted that the thickness of TEM and
HREM samples as shown in Fig. 7.2 is a few tens of nanometers, and so the elec-
trons of the beam penetrate several crystallites.
The transmission electron diffraction patterns of meso PS, as shown in
Fig. 7.2f, shows sharply defined Bragg diffraction spots that are characteristic of
crystalline bulk silicon. The diffraction spots of micro PS, as shown in Fig. 7.2e,
still show the material to be single crystalline, however, the spots are less well de-
fined, higher orders are missing and much diffuse scattering is observed. In con-
clusion, TEM observations indicate that microporous silicon can be visualized as a
sponge-like silicon crystal with average dimensions of a few nanometers [Ph1,
Be5, Le15]. Some morphological investigations of PS have been performed using
AFM and STM [Yu1]. These techniques, however, suffer from the fact that the as-
pect ratio of the tip used for inspection is usually not sufficient to reproduce the
micropore morphology.
7.2.2
Raman Spectroscopy
RS is another method that has been used to investigate the morphology of micro PS.
In single crystalline silicon only first-order scattering at the optical phonon in the
center of the Brillouin zone is allowed, which produces a sharp (FWHM 3.3 cm
1
)
Raman peak at 520 cm
1
. Confinement of the phonon leads to a momentum uncer-
tainty, and so the Raman peak broadens and shifts to lower energy. This effect can be
used to determine the size of silicon microcrystallites using Raman measurements
[Pa14]. The analysis of Raman data, however, is complicated by the fact that the
Raman signal is also sensitive to temperature and stress, as well as shape and sur-
face conditions of the crystallite. Nevertheless, dimensions observed by RS agree
well with values of size obtained by other techniques such as HREM or XRD. Broad-
ening and shift of the Raman peak is also observed for acoustic phonons (at 20 cm
1
)
[Fu2]. But nearly all investigations of PS refer to optical phonons.
Raman measurements of free-standing PS films have indicated crystallite di-
mensions as small as 2 nm [Ka3]. Investigations of PS films grown on substrates
of different doping density showed a broad size distribution (210 nm) for p
+
-type
substrates, while the size distribution for PS films on low doped p-type substrates
7 Microporous Silicon 130
showed only contributions below 3 nm, as shown in Fig. 7.3. The presence of an
amorphous contribution in the Raman spectra of PS is still controversial [Go3,
Zh4]. It is not clear whether the observed broad peak is due to an amorphous
phase or if the insertion of oxygen atoms leads to a relaxation of the lattice expan-
sion and to its distortion, which could result in the relaxation of the Raman selec-
tion rules and the appearance of an enhanced amorphous-like component of the
spectrum [Ta7]. The amorphous contribution in the Raman spectra is found to be
larger by nearly one order of magnitude near the surface of the sample. This is in-
terpreted as being caused by progressive chemical etching of the topmost region
of the PS layer [Mu3].
7.2.3
X-Ray Analysis
X-rays with a wavelength comparable to interatomic distances are diffracted by the
crystal lattice. The diffraction signal contains information about crystal orienta-
tion, particle size and strain. The XRD signal of an as-anodized micro PS layer
that is still on the substrate shows no significant misorientations of the crystal-
lites to the substrate and so PS can be imagined as a single crystalline sponge-like
structure [Fa8]. A significant amount of amorphous material in PS would produce
a broad peak at low diffraction angles. The fact that such a peak is not observable
for freshly prepared PS limits the maximal amount of amorphous silicon or sili-
con oxide present in PS to values below 10% [Le10].
The broadening of the characteristic peaks of the silicon XRD signal provides
information about stress and size of the crystallites. Figure 7.4 shows the diffrac-
tion pattern of microporous silicon powders scraped from p-type Si electrodes and
of a bulk silicon powder sample. The peak broadening increases with increasing
formation current density. For low formation current densities a superposition of
7.2 Morphology of Microporous Silicon 131
Fig. 7.3 (a) Raman spectra (dotted lines) of
PS films of 65% porosity formed on moder-
ately (1 X cm, micro PS) and highly
(0.01 X cm, meso PS) doped substrates,
together with calculated fits (full lines),
assuming the distributions of crystallite
diameters as shown in (b). Redrawn from
[Mu2].
a sharp peak and a broad peak is observed. In the light of HREM and Raman ob-
servations this is interpreted as being due to a long-range order that produces co-
herent diffraction and not as an indication of the presence of two different crystal-
lite sizes. The average crystallite diameter is in the order of 34 nm. For high for-
mation current densities a change in lattice constant of the order of Da/a=0.1 is
found [Le10], which corresponds to high values of stress. Values below Da/a=0.01
have been reported for lower current densities [Mu2].
7 Microporous Silicon 132
Fig. 7.4 XRD signal for powder samples pre-
pared from bulk Si (top) and micro PS films
produced on p-type electrodes using different
anodization current densities, as indicated in
the figure. From the broadening of the diffrac-
tion peaks the crystallite size was calculated
to be 3.5 nm (middle) and 2.8 nm (bottom).
After [Le10].
SAXS is sensitive to variations in the electronic density in a material, and so pro-
vides information about the shape and size of clusters in micro PS. In contrast to the
quantum wire structure proposed in early publications [Ca1, Le1], the crystallites in
micro PS are found to be almost spherical. There has been some evidence that the
dimensions in the growth direction are somewhat smaller than those parallel to the
surface [Fr2]. The latter result, however, is still controversial because investigations
by spectroscopic techniques indicate an opposite elongation [Mi4]. A combination
of grazing incidence diffraction (GID) and SAXS measurements on various free-
standing micro PS films showed crystallite diameters from about 1.5 to 4 nm, de-
pending on formation conditions. A good correlation between size reduction and
blue shift of the PL peak position has been observed [Bi3].
The X-ray absorption fine structure (EXAFS and NEXAFS) provides information
about the next neighbor atoms. A comparison of data from small oxidized silicon
spheres of known size with micro PS indicate that the average dimension of the
microstructures in luminescent PS is about 1.52 nm [Sc12]. Photoemission and
X-ray absorption spectroscopy gives information about the binding energy of elec-
trons. This method enables us to measure separately the change in energy of the
conduction and valence bands. As shown in Fig. 7.16, the VB shift DE
V
has been
found to be about a factor of two larger than the CB shift DE
C
[Bu2, Mo8]. In ad-
dition a ratio of 1: 3 has been observed between the increase in DE
C
and the in-
crease in PL emission energy [Da6]. These results have been interpreted in terms
of QC present in the microporous silicon structures, as discussed in Section 7.5.
7.3
Absorption, Reflection and Nonlinear Optical Effects
The interaction of photons with electrons in the porous structure, which is the
topic of this section, provides information about the PS band structure, while the
excitation of vibrational modes of molecules present in the porous structure by
photons gives information about its chemical composition, as discussed in Sec-
tion 6.6.
The dependence of the optical properties of silicon on its microstructure is
shown by the color figure in the lower left corner of this books cover. Despite the
fact that the thickness of all PS samples shown has been chosen to correspond to
a 20 lm thick film of bulk silicon, the difference in light transmission is dramatic
[Le27].
The structural dimensions of macro PS are of the same order as the wavelength of
photons in the visible regime and in the near IR. Therefore macro PS is a strongly
scattering material. As a result, a two-dimensional photonic bandgap has been ob-
served for certain regular pore patterns under illumination with IR light in a plane
orthogonal to the macropore axis. For illumination parallel to the macropore axis
transmission through the pores is observed. With decreasing pore diameter the
transmittance becomes reduced for long wavelength, and the material acts as an op-
tical short-pass filter. Therefore a macropore sample appears blue if illuminated by
7.3 Absorption, Reflection and Nonlinear Optical Effects 133
white light from the backside, as shown on the lower left of the books cover. These
effects have been exploited for optical applications, as discussed in Section 10.5.
The structural dimensions in meso PS and micro PS, in contrast, are too small to
scatter photons of energies in the visible or below, and the transmittance becomes
independent of the pore growth direction. This allows us to treat such structures
in the framework of the effective medium concept. Now the electronic properties,
namely the bandgap of the material, dominate the optical properties. Bulk Si, as
well as micro PS and meso PS, show an optical low pass filter characteristic. The
films shown in the lower left of the books cover, however, appear in different col-
ors, despite the fact that the total amount of silicon in the beam path is roughly con-
stant. The microporous film appears bright yellow, showing an even higher trans-
mittance than a 7 lm thin bulk Si film, which appears orange. A mesoporous film
appears red, which indicates a transmittance between that of the 7 lm thick Si film
and the 20 lm thick bulk Si grid. This blue shift of the absorption edge has been
interpreted in terms of QC, as discussed below [Le1].
The propagation of light waves inside the effective medium established by the
porous structure as well as the reflected and transmitted fractions can be de-
scribed by the complex refractive index [Th3]. The real part of the function de-
scribes the difference between the vacuum wavelength of a photon and its wave-
length in the material, while the imaginary part of the dielectric function gives
the exponential decay (or absorption) of the wave in the material. A reliable meth-
od of determining the refractive index is to measure the interference fringes of a
free-standing PS layer or a PS layer on the Si substrate. A reflectance spectra of a
PS film on the Si substrate that exhibits several interference maxima and minima
is shown in Fig. 7.5a. The complex refractive index of the PS film, as shown in
Fig. 7.5b, can be obtained by a complete simulation of the measured spectra
(broken line in Fig. 7.5a). The refractive index of bulk silicon is shown in
Fig. 7.5b for reference. The distinct maxima of the refractive index of bulk silicon
correlated to direct transitions of bulk Si (at 3.4 and 4.3 eV) can be identified in
reflectance spectra of meso PS and micro PS samples of low porosity, but disap-
pear for micro PS if the porosity increases above about 70% [Ko17, Th7].
A certain anisotropy of the refractive index along specific crystallographic axes
indicates that the microstructures in the porous network are not spherical but
somewhat elongated along the PS growth direction [Mi4]. This birefringence is be-
low 1% for micro PS, while it may reach values in the order of 10% for meso PS
films formed on (110) oriented silicon wafers [Ko22].
The refractive index of micro PS, n
PS
, for wavelengths in the visible or near IR
regime is found to decrease with porosity p [Lo1] and can be roughly approxi-
mated if the refractive index of bulk silicon n
Si
for the wavelength in question is
known, using
n
PS
= (n
Si
1) (1 p) 1 (7:1)
This dependence of n
PS
on p can be used to fabricate optical superlattices, e.g. Bragg
filters, by modulating the porosity of a PS layer, as discussed in Section 10.5.
7 Microporous Silicon 134
The imaginary part of the dielectric function describes the optical absorption in
PS and thereby gives information about the bandgap. Details of the optical transi-
tions responsible for absorption and emission of photons in Si are shown in
Fig. 7.12 and will be discussed in the next section. The absorbed fraction P(x) of
the non-reflected light intensity P depends on the sample thickness x and on the
absorption coefficient a according to
P(x) = P[1 exp (ax)[ (7:2)
This function is shown for bulk silicon and different values of light wavelength in
Fig. 10.4a. The transmitted fraction T(x) is T(x) =PP(x):
T(x) = P exp (ax) (7:3)
The optical absorption coefficient a as a function of wavelength for free-standing
micro- and mesoporous silicon layers determined by transmission measurements
in comparison with that of bulk silicon and amorphous silicon is shown in
Fig. 7.6. The absorption coefficient of the mesoporous silicon sample fits closely
that of bulk silicon over the entire investigated range of photon energies if the
data are corrected for porosity (effective medium approximation). The bandgap of
the meso PS sample, defined by extrapolation of the data shown in Fig. 7.6 to
zero absorption, is found to be 1.13 eV, which is only 10 meV higher than ob-
served for bulk Si [Ko15]. For the micro PS sample results are different. While for
photon energies above the direct bandgap (3.3 eV) the effective medium approxi-
mation describes a well, a is significantly reduced for lower energies. In this re-
gion a depends exponentially on the photon energy. This blue shift of the absorp-
tion edge of micro PS compared to bulk silicon can be interpreted in terms of QC
[Le1]. An observable blue shift is only expected if the majority of crystallites that
form the porous skeleton are small enough to show a significant QC, otherwise
7.3 Absorption, Reflection and Nonlinear Optical Effects 135
Fig. 7.5 (a) Measured reflectivity of a micro
PS film (1 lm thick, 71% porosity) on the Si
substrate (solid line) together with the simu-
lated reflectivity (broken line). (b) Real part
(bold solid line) and imaginary part (bold
broken line) of the refractive index obtained
from the simulation, together with the refrac-
tive index of bulk Si for comparison. Redrawn
from [Th7].
the non-confined crystallites would dominate the absorption spectrum. However, a
small fraction, in the order of 5% of the total volume of the sample of larger crys-
tallites, seems to be present and to dominate the absorption at energies below
1.5 eV [Ko4].
Between 7 and 470 K a is found to increase with temperature. This increase is
reversible and corresponds well with that of bulk silicon [Ko4]. If PS is annealed
in nitrogen at higher temperatures (6008C), hydrogen desorption takes place,
which changes the condition of the inner surface drastically. At these tempera-
tures an irreversible increase in a is observed for micro PS. A similar increase in
a after annealing is found for meso PS. Changes that affect the core of the crystal-
lites, e.g. stress effects [Ko4], as well as surface-related effects like the formation
of surface states [Ko5, Ba10], are proposed to be responsible for the observed in-
crease of a with T.
The optical transmission of free-standing micro PS shows a non-linear behavior
on the picosecond time-scale for high intensities. The increase in absorption with
increasing incident power can be interpreted as being caused by additional absorp-
tion by photogenerated free charge carriers [Ma9]. Carrier thermalization [Ow1] or
excitation transfer from core states to surface states [Ma10] is proposed to be re-
sponsible for the observed fast decay in the first picoseconds. At high carrier den-
sities (>10
18
cm
3
), fast Auger recombination dominates the following 200 ps
[Ma9], followed by trapping of the free carriers, while the radiative recombination
7 Microporous Silicon 136
Fig. 7.6 Spectral dependence of the absorp-
tion coefficient for free-standing micro PS
(72% porosity), meso PS (45% porosity),
bulk Si and amorphous Si:H at RT. While the
meso porous film can be roughly fitted to a
Bruggeman effective medium calculation for
Si spheres embedded in a vacuum with 45%
nominal porosity [Ko15], micro PS shows a
significantly lower absorption in the visible
than expected from a calculation for 72%
nominal porosity. After [Ko4].
process is much slower. The latter observation supports the idea that PS behaves
like an indirect bandgap material [Be10]. PL measurements in the regime of Au-
ger saturation can be used to determine the optical absorption cross-section of the
Si nanocrystals in micro PS. This parameter is found to vary by over five orders
of magnitude, depending on excitation energy, as shown in Fig. 7.7 [Ko19].
As a result of carrier localization the nonlinear optical response is expected to
be strongly enhanced for micro PS compared to bulk silicon. This property is de-
scribed by the third-order nonlinear susceptibility, whose real part is proportional
to the nonlinear refractive index, while the imaginary part is proportional to the
nonlinear absorption coefficient. For weak excitation at RT micro PS shows a
large nonlinear absorption change. The third-order nonlinear susceptibility is
found to be in the order of 10
12
esu cm for high switching frequencies (10
12
s
1
)
and increases to values in the order of 10
3
esu cm for the kHz regime [Kl2, Ka7,
Ma13].
While the above-bandgap absorption of meso PS roughly resembles that of the
substrate, as shown in Fig. 7.6, the below-bandgap absorption shows pronounced
differences to that of the substrate. The free carriers in heavily doped silicon ab-
sorb photons below the fundamental bandgap with increasing absorption towards
small photon energies, as shown in Fig. 7.8. If such substrates are made mesopor-
ous by anodization, the absorption coefficient of the dry layers decreases by orders
of magnitude with increasing porosity, as observed by photothermal displacement
spectroscopy (PDS) [Ch7]. Because dopant atoms are still present in the porous
layer, as proved by SIMS, the reduced a for the IR regime in mesoporous silicon
has been interpreted as being caused by trapping of free charge carriers by deep
levels relating to surface states [Le14].
7.3 Absorption, Reflection and Nonlinear Optical Effects 137
Fig. 7.7 Absorption cross-section of Si nanocrystals versus detection
energy. Excitation energies are indicated in the figure. After [Ko19].
7.4
Luminescence Properties
Bulk silicon is a semiconductor with an indirect band structure, as schematically
shown in Fig. 7.12c. The top of the VB is located at the center of the Brillouin
zone, while the CB has six minima at the equivalent 100) directions. The only al-
lowed optical transition is a vertical transition of a photon with a subsequent elec-
tron-phonon scattering process which is needed to conserve the crystal momen-
tum, as indicated by arrows in Fig. 7.12c. The relevant phonon modes include
transverse optical phonons (TO: 56 meV), longitudinal optical phonons (LO:
53.5 meV) and transverse acoustic phonons (TA: 18.7 meV). At very low tempera-
ture a splitting (2.5 meV) of the main free exciton line in TO and LO replicas can
be observed [Ko15].
Because of its indirect bandgap, bulk crystalline silicon shows only a very weak
PL signal at 1100 nm, as shown for RT and 77 K in Fig. 7.9. Therefore opto-
electronic applications of bulk silicon are so far limited to devices that convert
light to electricity, such as solar cells or photodetectors. The observation of red PL
from PS layers at room temperature in 1990 [Ca1] initiated vigorous research in
this field, because efficient EL, the conversion of electricity into light, seemed to
be within reach. Soon it was found that in addition to the red band, luminescence
in the IR as well as in the blue-green region can be observed from PS.
7 Microporous Silicon 138
Fig. 7.8 The below-bandgap transmission of
free-standing mesoporous silicon compared
to the bulk transmission of the corresponding
n
+
and p
+
substrates. The strong increase in
IR transmission for the porous samples has
been ascribed to a reduced free carrier con-
centration in the porous material due to trap-
ping in surface states. After [Le14].
7.4.1
The Red Photoluminescence Band
More than 1000 scientific publications address the PL of PS and make this single
property of PS probably its best studied one [Cu2]. Of the different luminescence
bands the red PL band is usually the most intense. This band is the subject of
most investigations and is usually addressed if the wavelength of PL is not speci-
fied explicitly. The PL peak intensity of the red band extends from about 1.3 eV
(950 nm) to 2.1 eV (590 nm). Towards the low-energy side the PL intensity be-
comes small even for high excitation intensities. This is in contrast to the high-en-
ergy side, for which efficiencies of up to 10% have been reported [Be11]. The PL
peak intensity of the red band can be shifted by a variation of the anodization con-
ditions. A blue shift is observed for an increase in formation current density, for a
decrease in HF concentration and for intense illumination during anodization.
However, a PL peak energy of 2.1 eV seems to be an upper limit for as-anodized
PS samples exposed to air. The decay times at RT are of the order of several mi-
croseconds, and the red band is therefore also termed the slow-band (S-band). For
most of the luminescent PS samples the red PL peak can be fitted to a Gaussian.
7.4.2
Dependence on Formation Current Density
The formation of luminescent PS requires HF to be present in the electrolyte, while
the presence of water is not essential [Pr7]. The intensity as well as the peak energy
of the PL emission increases with the PS formation current density J for a fixed elec-
trolyte concentration. If various electrolytes are compared, the ratio between forma-
tion current density J and the critical current density J
PS
is more relevant than the
absolute value of J. Because the porosity itself depends on J/J
PS
, in many studies
7.4 Luminescence Properties 139
Fig. 7.9 The PL spectra of various silicon-based materials (a) at RT and (b) at 77 K
(note the logarithmic scale of the PL intensity) (a-Si: amorphous Si, lc-Si: micro-
crystalline Si). After [St8].
the PL intensity is plotted versus the porosity. At this point it should be emphasized
that the PL depends not directly on porosity, but on the structural dimensions of the
porous network, which usually decrease with increasing porosity. This relationship,
however, holds true only for a certain doping density. If for example PS is formed on
highly doped substrates in the dark, the PL is red shifted and its intensity is reduced
by orders of magnitude compared to low doped p-type samples of the same porosity.
For low doped as well as highly doped substrates the PL intensity increases by more
than three orders of magnitude for an increase in porosity from 70% to 94% [Be11].
7.4.3
Dependence on Illumination
If the electrode is illuminated during anodization, the PL of the resulting PS de-
pends strongly on the wavelength of the light source. If the wavelength is short en-
ough to be absorbed by PS, a higher porosity and a higher PL efficiency at shorter PL
wavelength are observed, as shown in Fig. 7.10 [Le24, Su7]. This has been ascribed to
further etching of the porous network producing a smaller structure, as shown in
Fig. 7.1b. That the same effect can also be observed for prolonged soaking of the
PS in the HF electrolyte in the dark [Ca1] or under illumination [Mi3] following
the anodization, supports the idea that the increase in PL intensity and peak energy
can be ascribed to a size reduction of the porous skeleton.
Surprisingly it was found that PS luminescence excited by polarized light
emerges from the sample preferentially with the same sense of polarization. This
memory effect has been observed despite the fact that the electron-hole pair loses
energy in the order of 1 eV in elastic processes with lifetimes in the order of
7 Microporous Silicon 140
Fig. 7.10 PL spectra (RT) of micro PS sam-
ples anodized in the dark and under illumina-
tion using the set-up shown in the upper
right of the figure (1 X cm p-type, (100),
50 mA cm
2
). A blue shift and an increase
in PL intensity by one order of magnitude is
observed for short wavelength illumination.
After [Le24].
10 ls. The polarization memory has been ascribed to elongated crystals in the po-
rous network which act as dipole radiators. This idea is supported by experiments
in which micro PS was formed under illumination with polarized light. In this
case luminescence excitation with unpolarized light led to a partly polarized PL.
These investigations point to extended states as recombination centers, in a sense
that wave functions occupy a major part of the microcrystallite and depend there-
fore on their shape and orientation. For local defect states of atomic dimension no
polarization memory would be expected [Ko11].
7.4.4
Dependence on Magnetic Field and Hydrostatic Pressure
Anodization of Si in HF under an applied magnetic field produces an enhance-
ment of the PL efficiency at RT, accompanied by an enhanced porosity compared
to PS samples prepared without an applied field. The degree of polarization of the
emitted PL is reduced for field-assisted preparation [Na3]. At low temperatures
(4.2 K), the Stokes shift and the decay time of the PL are found to be increased, if
compared to PS formed under zero magnetic field. This has been interpreted as
Zeeman splitting of the spin-triplet exciton states. It indicates that the ground
state of the luminescing silicon crystallite is a triplet state [Ko13].
The dependence of PL on hydrostatic pressure has been investigated in a dia-
mond anvil cell up to pressures of 17 GPa [Ze1]. At moderate pressures of up to
3 GPa the PL peak shows a blue shift followed by a significant red shift and a de-
crease in PL intensity at higher pressures. However, it needs pressures in the or-
der of 1718 GPa to make the PL vanish. This is astonishing at first sight, be-
cause the diamond phase silicon crystal undergoes an irreversible phase transition
to the metallic b-tin phase at 1213 GPa, followed by a transition to the primitive
hexagonal phase at 1617 GPa. However, measurements of the PL and Raman in-
tensity for hydrostatic pressures up to 21 GPa indicated that the phase transitions
of silicon are sensitive to cluster size [Pa13].
7.4.5
Dependence on Chemical Environment
The intensity and wavelength of PL depend greatly on the chemical environment of
the luminescing crystallites. This addresses the medium present in the pores, as well
as the effects of species chemically bonded to the surface [Le25]. A reversible de-
crease in the luminescence intensity, for example, is found if the pores are filled
with a vapor or a liquid solvent. This quenching increases with increasing gas phase
dipole moment of the liquid or the vapor [La7]. Desorption of the hydrogen from the
internal PS surface, which starts at temperatures as low as 3508C, is sufficient to
reduce the PL intensity by a factor of 10, as shown in Fig. 6.16 [Ts3, Ha4]. The de-
crease in PL intensity associated with annealing to the first hydrogen desorption
peak is fully reversible by a short HF dip, while annealing above the temperature
range of the second peak leads to an irreversible loss of the PL [Ha4].
7.4 Luminescence Properties 141
The dependence of PL on the presence of SiH led to the idea that the lumines-
cent species is the silicon hydride itself [Pr6]. However, in situ PL and FTIR stud-
ies during annealing of PS from RT to 3508C revealed that the quenching of the
PL signal significantly precedes H
2
desorption from most of the SiH
2
surface spe-
cies. In addition, no correlation was found between the IR absorbance due to
SiH
x
and the PL intensity during chemical etching of PS in HF [Ro6]. Further-
more, the PL is restored if oxidized PS is formed by annealing above 6008C in
oxygen [Pe1, Ro6]. The role of hydrogen became clearer when the anticorrelation
between the PL intensity and the defect density of PS during the thermal treat-
ment was observed. Electron paramagnetic resonance (EPR) measurements re-
vealed that the defect density increases from 10
16
cm
3
for as-prepared PS to den-
sities close to 10
19
cm
3
for samples annealed at 6008C, as shown in Fig. 7.20.
The latter value of defect density corresponds to roughly one defect per crystallite,
if the crystallites are assumed to be 5 nm in diameter [Me9]. The defects in as-pre-
pared PS are mainly dangling bonds of the P
b0
type [Ro7, Ba12]. Further anneal-
ing under the presence of oxygen forms a thin oxide film that sufficiently passi-
vates the crystallites to restore the PL, as discussed in Section 7.6.
Deuterium-terminated PS shows the same microstructure and VB energy shift,
but the PL is significantly blue shifted if compared to hydrogen-terminated PS.
This indicates that the surface vibration of terminated atoms couples to the ex-
cited energy states of quantum-confined charge carriers [Ma11].
7.4.6
Photoluminescence Excitation Spectroscopy
The PL spectrum of a PS sample is usually recorded using a certain excitation
wavelength well above the PL signal range. However, if the excitation energy is re-
duced to values close to the emission energy or even below, the absorption pro-
cesses connected with light emission and the bandgap energy can be investigated.
This technique is called photoluminescence excitation spectroscopy (PLE). It mea-
sures the PL signal intensity for a certain wavelength in the PL spectrum while
the excitation wavelength is scanned.
A typical PLE spectrum, together with the PL spectrum of a red-emitting PS
sample, is shown in Fig. 7.11a. A quadratic increase in intensity with excitation
energy is observed for most PLE spectra of micro PS. This can be interpreted as a
reminiscent of indirect absorption in the crystallite core and allows us to extrapo-
late a bandgap. By extrapolation to zero absorption a bandgap of about 1.8 eV for
crystallites emitting at the detection energy of 1.55 eV is determined, for example
[Ku4]. Values for the bandgap energy in micro PS extrapolated from PLE corre-
spond well to values calculated from the IR PL band, as shown in Fig. 7.13. The
energy loss between absorption and emission of about 0.4 eV for the red band
and of 1.1 eV for the IR band, as shown in Fig. 7.11, is called the Stokes shift.
PLE investigations show that the red emission from dry PS and OPS, as well as
the green emission of PS in HF are significantly Stokes-shifted. Only the blue
emission from OPS can be excited close to the emission energy.
7 Microporous Silicon 142
At low temperatures the PLE of hydrogen-terminated PS reveals that phonons
and the exciton exchange splitting contribute significantly to the observed Stokes
shift [Ca6, Ku4, Ro5, Ka8, Ko13]. For oxidized PS the picture is not usually so
clear, due to a recombination path that may involve surface states.
7.4 Luminescence Properties 143
Fig. 7.11 (a) PL and PLE spectra of the IR
and the red S-band for a micro PS sample of
80% porosity at 4.2 K. (b) Double-logarithmic
plot of the decays of the IR and red PL inten-
sities at emission energies of 1.181 eV (IR)
and 1.907 eV (red) for different temperatures
of the sample [Ma17].
In indirect bandgap semiconductors, like silicon, optical transitions are allowed
only if phonons are absorbed or emitted to conserve the crystal momentum. Pho-
nons are not involved in luminescent recombination processes at surface mole-
cules. The observation of phonon-related structures in the PL signal is therefore a
fingerprint for the absorption and emission pathways, as shown in Fig. 7.12c and
d. Time-resolved, resonantly excited PL spectroscopy of PS and OPS at low tem-
peratures revealed momentum-conserving phonon satellites, as shown in
Fig. 7.12a and b. Each localized exciton state is split by the exchange interaction
into a spin singlet and a spin triplet, with the triplet lower in energy. The ex-
change splitting of excitons is found to be enhanced by QC, compared to the bulk
Si value. That the optical transitions are phonon-assisted supports the idea of crys-
talline silicon being the luminescing species in PS and indicates that the indirect
7 Microporous Silicon 144
Fig. 7.12 The resonant PL spectra of (a)
naturally and (b) heavily oxidized micro PS at
1.8 K. The arrows show the energy position
of silicon TA and TO momentum-conserving
phonons with respect to the triplet exciton
ground state.
The band structure of bulk silicon, with
possible optical transitions for (c) absorption
and (d) emission of a photon, together with
(e) the dispersion curves of phonon branches,
is shown on the right. After [Ko15].
P
L
I
n
t
e
n
s
i
t
y
(
a
r
a
b
.
u
n
i
t
s
)
Energy (eV)
character of silicon is preserved in PS. However, the spatial confinement of elec-
trons and holes inside a small crystallite increases the uncertainty of their crystal
momentum, thus allowing optical transitions in which phonons are not involved.
This breakdown of the k-conservation rule with increasing confinement energy is
theoretically predicted [Hy2] and observed experimentally [Ko13]. The crossover
from phonon-assisted to no-phonon recombination occurs at emission energies
around 1.81.9 eV.
7.4.7
Auger Saturation
The PL intensity is proportional to the excitation energy over several orders of
magnitude. At high excitation levels an occupation of a nanocrystal by two elec-
tron-hole pairs can be achieved. In this case fast, non-radiative Auger recombina-
tion of one pair leads to PL saturation. The excitation level needed for saturation
depends on lifetime and therefore decreases with temperature. The Auger satura-
tion phenomenon can be used to determine the optical absorption cross-section of
the Si nanocrystals in micro PS. As shown in Fig. 7.7, the absorption cross-section
varies over five orders of magnitude depending on excitation energy. In the light
of these findings the significant Stokes shift between absorption edge and the PL
peak energy becomes an artificial fact arising from the very small DOS in the
vicinity of the energy of the exciton ground state [Ko19].
7.4.8
Thermostimulated Luminescence
The red PL band of PS can not only be excited by above bandgap photons, but
also by an intense IR (1064 nm) pulse [Di6]. Such a thermostimulated lumines-
cence is known for the case of glasses. This observation was attributed to PS hav-
ing about 100 times the third-order nonlinear optical susceptibility of bulk Si, as
discussed in Section 7.3. Multiphoton excitation of the red PL band by resonant
pumping of the vibrational modes of surface groups like SiO [Di4] or SiH [Ch8]
provided evidence for excitation modes that involve the porous skeleton surface.
7.4.9
Photoluminescence Decay
The PL decay following a short excitation pulse provides valuable information
about absorption and emission processes. Therefore the temporal behavior of the
PL spectra has been studied at time-scales from picoseconds to tens of millisec-
onds for different temperatures.
Microcrystallites of direct semiconductors usually show a simple exponential de-
cay of the PL intensity P with time, with time constants s in the ps and ns range
at RT. A similar simple exponential decay (s=20 ms at 2 K) is observed for PL
from mesoporous silicon of high porosity, which shows a weak confinement effect
7.4 Luminescence Properties 145
[Po2]. For microporous silicon a more complex time behavior extending well into
the ms range is observed, as shown in Fig. 7.11b. The experimentally observed de-
pendence of intensity P on time t has been fitted to an exponential [Ho3, Fi4,
Ma10] or a stretched-exponential function according to Eq. (7.4) [Ka6, Pa12, Oo1].
P(t) = P
0
exp [(t=s)
b
[ (7:4)
The exponent b describing the distributions of PL decay rates is found to be in
the range 0.40.5 for most temperatures and PL energies investigated. In contrast,
the PL lifetime s decreases with increasing PL energy.
The decay on a picosecond time-scale, the so-called fast band, is understood as
a quasi-direct recombination process in the silicon crystallites or as an oxide-re-
lated effect [Tr2, Mg1]. This fast part of the luminescence requires an intense exci-
tation to become sizable; it then competes with non-radiative channels like Auger
recombination. The observed time dependence of the slow band is explained by
carrier recombination through localized states that are distributed in energy, and
dimensionally disordered [Gr7].
The PL lifetime values s obtained by fitting a stretched exponential function de-
crease with increasing PL peak energy E
PL
. For micro PS dried in a vacuum, for
which the PL energies range from 1.5 to 3.5 eV, this dependence can roughly be
fitted to the empirical relation:
s = 10
(52 E
PL
)
(7:5)
for the PL peak energy E
PL
in eV and the lifetime s in ls at RT [Mi3]. Equation
(7.5) fits well with the data shown in Fig. 7.19b, which indicate that the depen-
dence of PL lifetime on confinement energy can be traced smoothly down to the
fundamental bandgap of bulk silicon [Ko15].
7.4.10
Temperature Dependence
The variation of PL intensity, wavelength and decay with temperature is strongly
dependent on the history of the specific PS sample under investigation. In respect
of the PL intensity, three basic temperature regimes are observed. The PL inten-
sity is low at low temperatures (T <60 K). This can be interpreted as a PL being
dominated by temporary traps and the thermal activation of charge carriers from
these traps competing with non-radiative recombination. At medium tempera-
tures (60 K<T<250 K) the PL intensity shows a maximum. The thermal energy is
now sufficient for activation from shallow traps, but still too low to separate the
photogenerated electron-hole pair by thermally activated diffusion of one charge
carrier to the next crystallite. A further increase in temperature (T>250 K) results
in a decrease in the PL intensity. In this temperature regime hopping and diffu-
sion of charge carriers is activated and non-radiative recombination becomes effec-
tive. This basic behavior of PL intensity is shown in Fig. 6.18, together with the
7 Microporous Silicon 146
PS resistivity versus temperature. Note that the PL intensity reaches its maximum
at the onset of thermally-activated electrical transport. This supports the assump-
tion that the quenching of the PL at elevated temperatures is caused by a separa-
tion of the photogenerated electron-hole pairs by thermally-activated diffusion of
one charge carrier to the next crystallite [Oo1, Ko2]. The change in the PL
characteristic of PS upon cooling to temperatures below RT is usually fully revers-
ible, because it is a result of the temperature dependence of diffusion, trapping
and thermal activation processes of charge carriers. This is in contrast to the irre-
versible change in the PL observed for heating above RT.
The dependence of the PL spectra on temperature is again quite complicated.
The change from a Gaussian-shaped PL spectrum at RT to a multi-peak spectrum
at 5 K for low excitation energies was initially interpreted in terms of an interfer-
ence effect in the thin porous layer [Ho3]. Later work, however, showed that the
peak positions are independent of PS layer thickness, excitation wavelength and
sample rotation [Ke3]. The spacing of the peaks is about 61 meV. The origin of
this fine structure is still controversial [Ro4].
The picture is more consistent for the dependence of the PL decay time on tem-
perature. The dependence of s on T can be fitted to ln s=T
f
, with f increasing
with decreasing PL energy [Oo1]. In the temperature range 50150 K the average
lifetime becomes thermally activated by a characteristic energy of 1020 meV. This
energy can be interpreted as the relevant phonon energy in a phonon-mediated
transition and in terms of the singlet-triplet splitting of confined excitons. In the
latter case only the lower triplet state with a long lifetime is populated at low tem-
peratures. With increasing temperature the faster singlet state becomes populated
[Ho3, Ma14].
7.4.11
The Green-blue Photoluminescence Band
In contrast to the red PL band, the green-blue PL band of PS commonly shows
decay times in the nanosecond regime and is therefore termed the fast band or F-
band. This PL band can be observed under several conditions:
1. From porous samples that are still in the HF electrolyte a weak green PL (peak
energy 2.12.4 eV) is detected, while the red PL is quenched. The green PL
shows a blue shift and an increase in intensity with illumination time. If such
samples are dried they show the usual red PL band. This green-red switching is
reversible and only observed for polar liquids or vapors [Ku2, Li6, Oz3]. This ef-
fect has been attributed to a screening of luminescent impurity states by a high
dielectric constant of the embedding medium [Ts1].
2. Microporous samples that have been illuminated in the electrolyte after anodiza-
tion are found to exhibit a green or blue luminescence (peak energy from 2.1
3.0 eV) after drying in a vacuum or in an inert gas such as Ar. This PL red
shifts within seconds to the common red PL-band if the PS is exposed to oxy-
gen. This effect has been attributed to the formation of Si=O double bonds,
7.4 Luminescence Properties 147
which generate deep levels in Si crystallites of diameters below 2.5 nm [Mi3,
Wo2]. For anodization in an H
2
O
2
HF-based electrolyte, a more stable green PL
is reported for dried samples [Ya10], while for microcrystallites dispersed in so-
lution a blue PL, peaking at 390 nm, is found [Ak1].
3. A green-blue PL (peak energy 2.32.7 eV) is observed from PS samples that
have been exposed to some kind of oxidizing treatment, like formation in an
HFH
2
O
2
mixture, storage under ambient air for a prolonged period [Ha6,
Mg1], boiling in water [Ho5], oxidizing in H
2
O
2
[Ya9] or oxidizing by rapid ther-
mal oxidation (RTO) [Ko6]. The origin of this PL is still a matter of controversy.
The silanol group (SiOH) has been proposed as the luminescent center re-
sponsible for this blue band [Ta10, Ku3]. Other potential emission centers are
the neutral oxygen vacancy (2.7 eV) and the two-fold coordinated silicon lone
pair center (3 eV), which are known to cause efficient PL in silica [Yu2].
7.4.12
The Infrared Photoluminescence Band
In addition to the red PL band, an IR PL band between about 0.8 and 1.3 eV is
usually present in most PS samples [Fa6, Pi3, Ma17, Mo7, Ku4, Pe5], as shown in
Fig. 7.9. At RT the intensity of the IR band is weak. At cryogenic temperatures it
becomes much stronger and can even be more intense than the red band. The
PLE spectra for the red and the IR bands are identical despite their large differ-
ence in PL peak energy, as shown in Fig. 7.11a. Furthermore a correlation be-
tween the peak position of the two bands has been observed, as shown in
Fig. 7.13 [Ku6]. The PL decay, however, is found to be different. For the IR band a
s
2
dependence of PL intensity on time, with little dependence on temperature, is
observed, while a stretched exponential decay with strong temperature depen-
dence is observed for the red band, as shown in Fig. 7.11b.
The IR band has been ascribed to radiative recombination from the CB to a DB
center [Ku6]. Such a recombination process is sensitive to changes in the CB energy
only and is therefore expected to show a weaker dependence on confinement than
the red PL. This interpretation, together with the observation that the increase in
VB energy is about twice the increase in CB energy, as shown in Fig. 7.16, has been
used to approximate the bandgap energy in micro PS, as shown in Fig. 7.13.
7.4.13
Wet Electroluminescence
EL from PS is observed for electrolytic contacts as well as solid-state contacts. The
latter subject is discussed in Section 10.5 while this section discusses wet EL.
While the external quantum efficiency of PS reported for solid-state contacts is
usually low, wet contacts are found to give high EL efficiencies at low applied bias
under anodic [Vi2, It1, Ge2, Ha7] as well as cathodic conditions [Bs1]. An example
of bright EL from a micro PS sample in acetic acid under anodic bias is shown
7 Microporous Silicon 148
on the upper right of the books cover. The orange glowing PS region of the circu-
lar sample has been formed in HF at a higher current density (200 mA cm
2
)
than the red glowing PS part (10 mA cm
2
). According to the QC model of PS for-
mation, the higher formation current density leads to smaller crystallites that emit
at shorter wavelength [Le3]. Under anodic conditions the EL is transient, because
it is accompanied by irreversible oxidation of the PS network. During the anodic
oxidation of mostly meso PS a shift of the EL and PL bands to shorter wavelength
has been observed. This has been interpreted in the frame of the QC model as a
successive passivation according to the size of the Si crystallites in PS [Bi1]. The
anodic current flows through the largest and non-confined structures first. No
light emission is observed during this step. The oxide growth induces a potential
drop at their oxidized surface. Under galvanostatic conditions the bias will in-
crease and charge transfer to confined crystallites becomes possible. This pro-
duces red EL, which is blue shifted with further anodization. Finally oxidation
leads to electrical isolation of the PS from the substrate and light emission ceases.
An IR study of electro-oxidized PS showed a decrease in the OH signal and an
increase in the SiO signal with anodization time. This can be interpreted as oxide
formation on the PS surface and a removal of electrolyte from the pores. Further-
more a correlation in intensity of localized carrier IR absorption and lumines-
cence indicates that localized states are involved in the red EL [Du4].
In contrast to the anodic case, cathodic EL is persistent. Cathodic EL can be ob-
served from PS layers formed on n-type substrates in aqueous solutions contain-
7.4 Luminescence Properties 149
Fig. 7.13 Bandgap of PS for different types
of samples as determined from PLE measure-
ments with detection on the PL maximum
(filled circles) and from the position of the
IR PL band (open circles). The correlation
between the peak energy of the red PL band
and the IR PL band is shown in the lower
part of the figure (open circles). The diagonal
line indicates the case where the PL peak
energy is identical to the bandgap energy,
corresponding to zero Stokes shift.
After [Ku6].
ing the persulphate ion. The amplitude of the EL is found to be proportional to
the current, while the spectral position is only determined by the applied voltage.
The shift of the EL peak position with cathodic bias is shown together with the
corresponding shift for the anodic regime in Fig. 7.14. Note that the extrapolation
of the linear regimes to an EL energy of 1.1 eV, corresponding to the bulk silicon
bandgap, leads to a polarization gap of 1.1 V. This suggests that the injected en-
ergy level is directly determined by the polarization [Bs1]. A model for the voltage
tunability of wet EL based on the QC concept has been developed [Me10].
7.5
Quantum Confinement and Models of the Luminescence Process
This section gives a brief overview of theoretical investigations dealing with the
properties of quantum-confined silicon structures.
Every atom shows specific, discrete energy levels for electrons. These levels are
either empty or occupied by one or two (spin-paired) electrons according to the
Pauli exclusion principle. The energy of the levels can be found by solving the
Schrdinger equation. Exact solutions, however, can only be obtained for single
electron atoms (hydrogenic atoms).
If many atoms are bound together, for example in a crystal, their atomic orbi-
tals overlap and form energy bands with a high density of states. Different bands
may be separated by gaps of forbidden energy for electrons. The calculation of
electron levels in the periodic potential of a crystal is a many-electron problem
and requires several approximations for a successful solution.
The minute network structure of microporous silicon is between the two ex-
tremes of a single atom and a large crystal. A crystallite of a few hundred silicon
atoms is large enough to have a rich electronic band structure but is still small
enough to show an increase in the energy of an electron-hole pair (exciton) due to
7 Microporous Silicon 150
Fig. 7.14 EL peak energy from a 1 lm microporous Si film in an H
2
SO
4
, (NH
4
)
2
S
2
O
8
electrolyte
as a function of the applied (a) cathodic and (b) anodic polarization. Redrawn from [Bs1].
QC. The effect of an increase in fundamental bandgap due to size quantization is
not specific to silicon, but common to all semiconductors [He2, Br3, We1].
7.5.1
Bandgap Calculations
A very crude model to calculate the increase in bandgap energy is the effective-
mass particle-in-a-box approximation. Assuming parabolic bands and infinitely
high barriers the lowest conduction band (CB) level of a quantum wire with a
square cross-section of side length w is shifted by DE
C
compared to the value E
C
of the bulk crystal [Le1, Ho3]:
DE
C
= zh
2
=(8m
+
e
w
2
) (7:6)
where h is Plancks constant and m
+
e
denotes the effective electron mass. The
change in VB energy DE
V
is calculated analogously using the effective hole mass
m
+
h
. The total increase in the bandgap energy DE
G
is the sum of both changes
(DE
G
=DE
V
+ DE
C
). For the case of a two-dimensional confinement the factor z is
2. If the thickness of the wire fluctuates, the confinement energy will change
along the wire. As soon as the magnitude of these variations becomes comparable
to the thermal energy kT the mobility of carriers along the wire decreases. This
eventually leads to confinement in three dimensions, a quantum dot. However,
the approximation of Eq. (7.6) is too simplistic. It ignores the detailed shapes of
the valence and conduction bands, the influence of neighboring bands, excitonic
contributions, and surface effects. More sophisticated calculation methods give
more reliable results. An introduction to the different methods of band structure
calculation can be found in the literature [As3].
The main questions that are addressed in the theoretical studies are:
1. How does DE
G
depend on w?
2. Is the bandgap direct or indirect?
3. What are the lifetimes of excitons in the confined structures?
The methodology used to answer these questions can be classified as either semi-
empirical or based on first principles. The confined structure is assumed to be
two-dimensional (2D=quantum well), one-dimensional (1D=quantum wire) or
zero-dimensional (0D=quantum dot).
All authors who calculated the increase in bandgap energy DE
G
found that the
approximation of Eq. (7.6), based on an infinite energy barrier, overestimates the
size dependence. A linear [Re2, Bu1, De2, Hi4, Hi8] dependence between DE
G
and 1/w or a w
1.47
law [Pr4] has been found to be more realistic, as shown in
Fig. 7.15. In order to simulate the complicated structure of the sponge-like silicon
network the band structures for wires along different crystal orientations [Fi2], as
well as for wires with constrictions, have been calculated [Hi5]. The calculated en-
7.5 Quantum Confinement and Models of the Luminescence Process 151
ergy gaps are in good agreement with values determined by optical absorption
measurements.
High-resolution measurements of the L-edge absorption in electrochemically
prepared micro PS have shown that the absorption is shifted to higher energies
relative to the bulk Si value. The CB edge shift DE
C
increases with formation cur-
rent density and post-anodization soaking time in HF [Bu7]. By monitoring the
photoluminescent yield during such X-ray absorption fine structure measure-
ments (XAFS), it becomes possible to correlate the change in emission energy to
the relative CB shift. From these measurements a ratio DE/DE
C
of about 3 has
been observed [Da6]. In the above investigations, as well as for photoemission and
X-ray absorption near edge structure (XANES) spectroscopies, the ratio between
VB shift and CB shift DE
V
/DE
C
has been observed to be about 2 [Mo8].
Theoretical investigations of DE
V
and DE
C
have indicated that the ratio DE
V
/DE
C
is greater than 1 [Zh2, Vo2, Ne2] and may be as high as 3 for the case of hydroge-
nated silicon clusters [Re3]. In a similar calculation for thin silicon films, even-
odd oscillations of DE
V
according to the number of Si monolayers have been
found [Zh2]. For the latter case the ratio DE
V
/DE
C
showed values between 1 and 2.
The results of a study [Wa5] including the decrease in the static dielectric constant
with size are close to the experimentally [Bu2] observed values of about 2 for the
ratio DE
V
/DE
C
, as shown in Fig. 7.16. In this work [Wa5], it is concluded that the
electron-hole pair is confined by the physical dimension of the quantum dot, not
by Coulomb attraction.
The question of whether the bandgap of the quantum structures is direct or in-
direct has been discussed by several authors. However, it has been pointed out
that the electron and hole wave functions are spread in k space, breaking the
7 Microporous Silicon 152
Fig. 7.15 Increase in bandgap energy DE
G
with decreasing
diameter of a silicon cluster or dot (open symbols) or a silicon
wire (filled symbols), as predicted by various theoretical
investigations.
usual crystal momentum selection rules, so that it is no longer meaningful to de-
bate whether the gap is direct or indirect [Hy2]. Physically, the required momen-
tum derives from scattering from the crystallite surface. The question of whether
the luminescence occurs via a zero-phonon or a phonon-assisted transition can be
answered if the lifetimes for both mechanisms are calculated and compared. The
data shown in Fig. 7.18 suggest that, at least for the red luminescence band
(DE
G
=0.51.0 eV), phonon-assisted transitions are expected to dominate. Experi-
mentally no-phonon transitions have been found to dominate for DE
G
>1.9 eV
[Ko13].
The energy difference between optical absorption and emission, the Stokes shift,
is predicted to be below 0.1 eV for crystallites in the order of 1.5 nm diameter [Ma7].
However, values as high as 1 eV have been reported from experiments. This discrep-
ancy is explained by the localization of the electron-hole pair on one particular single
bond, leading to a longer bond length and a lower luminescence energy. Because the
elastic response of the environment is weaker at the surface than in the bulk, such a
self-trapped exciton is probably found at the crystallite surface [Al5]. Another expla-
nation for the large Stokes shift is based on interactions between neighboring quan-
tum wires. For closely packed quantum wires new interface-localized states appear
in the bandgap, reducing its value significantly (about 0.5 eV) [Os2]. The fact that
large Stokes shifts are observed for PS exposed to air, while a much smaller shift
is found for PS dried in vacuum or Ar, indicates that an oxygen bond is involved
in the recombination process [Mi3]. This assumption is supported by simulation
of the electronic structure of Si clusters with Si=O double bonds at the surface. This
model suggests that recombination in a silicon crystallite passivated solely by hydro-
gen is via free exciton states for all crystallite sizes, while in the presence of Si=O
7.5 Quantum Confinement and Models of the Luminescence Process 153
Fig. 7.16 CB edge shifts versus VB edge
shifts calculated for silicon clusters under dif-
ferent assumptions for the value of the dielec-
tric constant e (lines). After [Wa5]. For com-
parison the experimental results obtained by
photoelectron spectroscopy measurements on
electrochemical (filled circles) and stain-etch
(open circles) PS samples in comparison to
bulk silicon (open triangles) are shown.
Redrawn from [Bu2].
double bonds excitons with energies above 2.1 eV get trapped [Wo2]. In the light of
experimental findings, which show a variation of the absorption cross-section of PS
over five orders of magnitude, the large Stokes shift can be interpreted as an artificial
fact arising from a very small density of electronic states in the vicinity of the energy
of the exciton ground state [Ko19].
7.5.2
The Dielectric Constant and Doping of a Quantum Dot
The role of the dielectric constant of the microcrystallite and the medium in the
pores deserves further discussion. While the response of the porous layer to high
frequencies or light is adequately described by an effective medium approxima-
tion [Th3], this macroscopic model is not sufficient to describe the behavior of
charge carriers in the minute crystallites in the porous structure. Calculations
based on the Penn model [Ts1], pseudopotential calculations [Wa5], as well as self-
consistent linear screening calculations of hydrogenic impurities [Al4, La8], pre-
dict a reduction of the dielectric constant in a small crystallite compared to the sil-
icon bulk value. Figure 7.17 shows calculated values for the dielectric constant in
small silicon clusters with a hydrogen-saturated surface based on different mod-
els. A decreasing dielectric constant is reported to increase the energy splitting be-
tween the two lowest excitonic levels in asymmetric crystallites [Ma7].
In a study that addressed the effect of doping on quantum dots, the donor and
acceptor levels were found to be practically independent of particle size [De3]. In
other words, shallow impurities become deep ones if the dot size is reduced. Ex-
perimental observations show that the luminescence is not affected by doping if a
thermal diffusion process, for example using a POCl
3
source, is used [El1]. Im-
plantation, in contrast, is observed to effectively quench the PL [Ta14]. If the pores
are filled with a medium of a large low-frequency dielectric constant, such as
water or any other polar solvent, it is found that deep impurity states still exist,
7 Microporous Silicon 154
Fig. 7.17 Plot of the calculated dielectric
constant in silicon crystallites of different
size. The broken curve corresponds to
calculations based on the Penn model
[Ts1], the dotted line corresponds to
pseudopotential calculations [Wa5], while
the full line is based on self-consistent
linear screening calculation of hydrogenic
impurities [Al4]. Redrawn from [Al4].
but a free carrier can hardly get trapped into them, because of slow dielectric re-
laxation effects [Ch2, Br4, Ba3, Ts1]. This screening of the electric field of an ion-
ized impurity by the embedding medium is assumed to be responsible for the
green-to-red switching of the PL, as well as the reduction in the free carrier den-
sity, observed if the porous layer is dried [Ch2].
7.5.3
Lifetime Calculations and Surface Passivation
An approximation of the lifetime in PS at RT using an electron-hole pair density
equal to one pair per crystallite and the radiative recombination parameter of bulk
silicon give values in the order of 10 ms [Ho3]. The estimated radiative lifetime of
excitons is strongly size dependent [Sa4, Hi4, Hi8] and increases from fractions of
microseconds to milliseconds, corresponding to an increase in diameter from 1 to
3 nm [Hy2, Ta3], as shown in Fig. 7.18. For larger crystallites a recombination via
non-radiative channels is expected to dominate. The experimentally observed
stretched exponential decay characteristic of the PL is interpreted as a conse-
quence of the randomness of the porous skeleton structure [Sa5].
The dependence of lifetime on temperature in the range above RT shows an ac-
tivation energy in the order of 1025 meV [Bu3, Oo1]. This was proposed to be a
consequence of the exchange splitting of the exciton between the singlet and the
triplet state. While at RT both states are populated, only the lower triplet state is
populated at temperatures below 20 K. However, it has been shown that even for
crystallites of low symmetry the calculated values of the exchange splitting are too
low compared with experimental observations [De3]. Calculations of the radiative
lifetime of the triplet exciton that take into account spin-orbit interactions are re-
ported to be consistent with experimental results [Na1].
Values of DE
G
calculated on the basis of the effective mass model are sensitive
to the height of the barrier present at the surface of the confined structures [De4,
Xu2]. For the case of oxidized PS, it is obvious that the silicon crystallites are em-
bedded in a wide-gap material. In the case of a hydrogen termination of the crys-
7.5 Quantum Confinement and Models of the Luminescence Process 155
Fig. 7.18 The radiative recombination
time s as a function of the blue shift of
the photon energy DE from the bulk sili-
con band edge: zero-phonon transitions
(dots); TO phonon-assisted transitions
(line). This scatter plot shows the radia-
tive time for each member of an ensem-
ble uniformly distributed around a cubic
geometry. The top scale indicates the
equivalent cube size. Redrawn from
[Hy2].
tallite surface it is argued that the bonding-antibonding splitting of the SiH
bonds mimics a wide-gap material [De4]. However, the assumption of a finite bar-
rier gives more realistic results, especially for very small structures where the ef-
fective mass model overestimates the gap.
A hydrogen termination of the crystallite surface is assumed in most studies.
Calculations show that a small part of the hydrogen can be removed without the
formation of DBs, as shown for the transformation of Si
29
H
36
to Si
29
H
34
[Hi4].
Further dehydrogenation, however, leads to the formation of DBs that are efficient
recombination centers and thus effectively suppress the luminescence [De2, Hi4,
De5]. The observed IR luminescence has been proposed to be due to recombina-
tion via DB centers [De3].
The lifetime for Auger recombination was calculated to be in the 1 ns regime
[De3]. This may explain the saturation observed in the excitation of crystallites,
which seems to be limited to one exciton per crystallite [Mi1].
It should be noted that other structures, such as siloxene [De6] or polysilane
[Ta4], that have been proposed to play a role in the luminescence mechanism of
PS have also been studied on a theoretical level.
7.5.4
Conclusions from Calculations
It is possible to draw some conclusions about the physical properties of micropor-
ous silicon from the results of the theoretical studies:
1. The increase in bandgap energy DE
G
shows an approximately linear depen-
dence on 1/w.
2. The increase in VB energy DE
V
is expected to be considerably larger than the
increase in the CB energy DE
C
.
3. Fluctuations in the thickness of quantum wires can be assumed to be present
in a porous network and will drastically reduce the carrier mobility.
4. A visible red luminescence corresponds to a silicon cluster size between about
1 and 2.5 nm.
5. For the red luminescence band, lifetimes of about 10
3
to 10
6
s are expected
and the recombination is most probably phonon assisted.
6. A DB at the quantum structure surface is expected to efficiently suppress visi-
ble luminescence.
7. The dielectric constant in the crystallite is significantly reduced compared to
the bulk value.
8. Donor and acceptor levels are assumed to remain practically unaffected by par-
ticle size and thus become deep levels in confined crystallites.
9. If the pores are filled with a medium of a high dielectric constant, a screening
of impurity states is probable.
10. The large Stokes shift for confinement energies above 2.1 eV can be explained
by the electronic states produced by an Si=O double bond.
A detailed review of theoretical studies of PS is available in the references [ Jo3].
7 Microporous Silicon 156
7.5.5
Models of the Luminescence Process
With roughly 1000 atoms, the size of the silicon clusters that constitute the micro
PS network is between the bulk crystal and a molecule. Hence models of the lu-
minescence process based on size reduction of the crystal, as well as models
based on molecular structures, have been proposed, which are reviewed in detail
in [Ca7, Ju3]. Generally the various models of the luminescence of PS can be clas-
sified into three major categories:
1. The molecular recombination model. Absorption of a photon and its radiative
reemission occurs in chemical compounds such as siloxenes [Br6], polysilanes
[Ta4] or silicon hydrides [Pr6]. This luminescence mechanism is independent
of whether or not QC is present in PS.
2. The surface-state model. The absorption of a photon generates an exciton with-
in a quantum-confined silicon crystallite but its radiative recombination occurs
at localized electronic states on the surface of the crystallites [Ko5] or in defects
in the oxide coverage of the crystallites [Pr5].
3. The quantum recombination model. The photogeneration of an exciton and its
luminescent decay occur within a quantum-confined silicon crystallite [Ca6].
1. The molecular recombination model was proposed because the PL spectra of
some silicon compounds, e.g. siloxene, are very similar to the PL observed from
PS [Br6, St4, Mc3]. In addition aged PS is rich in H and O and the presence of
silicon hydrides, polysilanes or siloxene could therefore not be excluded. Detailed
investigations, however, showed that PS free of oxygen [Pr7], as well as OPS free
of hydrogen, do luminesce. The PL of micro PS is not very sensitive to its stoi-
chiometry, which would be the case if the luminescent species is molecular in na-
ture. For the special case of siloxene (Si
6
H
6
O
3
), detailed investigations of the PL
spectra revealed quantitative differences to the PL from PS, which led to the con-
clusion that light emission from PS is not directly associated with siloxenes [Sh3,
Mo4].
2. The surface-state model, in which the luminescent recombination occurs via
surface states, was proposed to explain certain properties of the PL from PS, for
example long decay times or sensitivity of the PL on chemical environment. In
the frame of this model the long decay times are a consequence of trapping of
free carriers in localized states a few hundred meV below the bandgap of the con-
fined crystallite. The sensitivity of the PL to the chemical environment is inter-
preted as formation of a trap or change of a trap level by a molecule bonding to
the surface of a PS crystallite. The surface-state model suffers from the fact that
most known traps, e.g. the P
b
center, quench the PL [Me9], while the kinds of sur-
face state proposed to cause the PL could not be identified.
3. A strong argument in favor of the QC model is the observation of step-like
features in the low-temperature PL spectrum of PS, as shown in Fig. 7.12, which
correspond well with the energies of momentum-conserving phonons in crystal-
7.5 Quantum Confinement and Models of the Luminescence Process 157
line silicon [Ca6, Ko13]. It has been argued that this observed weak luminescence
signal very close to the excitation energy is not identical to the strong red band ob-
served at room temperature [Ro5]. However, hole burning spectroscopy of PS
showed the same phonon-related structures, proving that the PL mainly originates
from quantum-confined states inside the Si crystallites [Ko12]. The observed pho-
non structures also exclude amorphous silicon as a potential source of the PL
[Va2]. In the frame of the confinement model the sensitivity of the PL on the sur-
face condition is explained as being due to polarization-induced changes in the
energy level of the quantum-confined states [Ts1].
The most fundamental properties of the confined exciton are confinement en-
ergy, the type of recombination process (indirect or quasi-direct) and the exciton
exchange splitting. As shown in Fig. 7.19, each quantity follows the expected
strongly nonlinear enhancement over a large dynamic range with increasing con-
finement energy and decreasing crystallite size. All these quantities can be traced
7 Microporous Silicon 158
Fig. 7.19 Influence of QC on the basic prop-
erties of excitons localized in Si nanocrystals.
Confinement energy dependence of (a) rela-
tive strength of no-phonon and TO phonon-
assisted recombination channels; (b) inverse
lifetime of the optically active exciton state;
(c) exciton exchange splitting. After [Ko15].
continuously and smoothly from bulk Si PL up to the visible spectrum, where the
confinement energy of more than 1 eV is as large as the fundamental silicon
bandgap itself [Ko13]. This smooth transition from bulk Si properties to the prop-
erties of micro PS are in favor of the QC model.
7.6
Oxidized Porous Silicon
In this section the properties of chemically and thermally oxidized and nitridized
PS will be discussed. Wet anodic oxidation of PS is commonly accompanied by lu-
minescence and is therefore discussed in Section 7.4.
Microporous and mesoporous silicon can be partly oxidized by chemical or elec-
trochemical methods, while thermal oxidation is needed for a complete conver-
sion of the porous layer to oxide. This complete conversion was the aim of the
first studies in this field.
Thermal oxidation of bulk silicon requires high temperatures and the maxi-
mum oxide thickness is limited to a few micrometers. The complete oxidation of
highly porous silicon, in contrast, can be performed in wet oxygen at tempera-
tures as low as 8509008C, independent of layer thickness. During the oxidation
the optical absorption edge of the material is blue shifted, which is ascribed to a
decreasing size of the microcrystallites [Pe1]. For films of several micrometers in
thickness this shift corresponds to a change in color from brown via red and yel-
low to a glassy appearance. This color change can be used as a simple indicator of
whether the oxidation was complete or not.
The porosity of the initial PS layer determines the density of the resulting ox-
ide. Oxidation of PS at temperatures below 10008C leads to porous oxides if the
initial porosities were in excess of 56%, while PS layers of lower porosity show re-
sidual silicon crystallites in an oxide matrix after oxidation. Porous oxides have a
lower dielectric constant than thermal oxides and are therefore favorable for re-
ducing parasitic capacitance [Ar4]. In addition the thermal budget of a manufac-
turing process can be reduced if the thermal oxide is not formed from bulk sili-
con but from PS. The resistivity of OPS is rather low for low oxidation tempera-
tures (6008C), which are known to produce a high density of DBs [Ha9], while
high resistivities (10
16
10
17
X cm) and low interface trap densities (10
12
cm
2
) are
observed for higher temperatures [Wu1].
The stress of oxidized PS layers is always compressive. For porous oxides, val-
ues below 10
8
N m
2
are reported [Ba5], which is nearly one order of magnitude
smaller than values of intrinsic stress generated by low-temperature thermal oxi-
dation of bulk silicon. The compressive stress in OPS has successfully been used
to lift up released mesoporous films and thereby fabricate 3D microstructures
[La9].
Annealing of porous oxides at temperatures above 10008C leads to a densifica-
tion. If the porosity of the initial PS was optimized to 56% the densification does
not affect the layer thickness, while otherwise the layer thickness increases (for
7.6 Oxidized Porous Silicon 159
p < 56%) or decreases (for p>56%). The stress in the densified oxides is caused by
the difference in thermal expansion coefficients between oxide and substrate. It is
therefore identical to that observed for high-temperature thermal oxidation. The
densified oxides are comparable to thermal oxides with respect to their etch rate
and refractive index. Fully oxidized PS is a good dielectric insulator, exhibiting rea-
sonable breakdown field strengths of around 5 MVcm
1
, low leakage currents and
moderate interface state densities [Wu1]. The densification process has an activa-
tion energy of 3 eV and is probably associated with the viscous flow of silica [Yo1].
While the dielectric constant for the densified oxide is identical to that of a ther-
mal oxide, it is one or two orders of magnitude larger for incomplete oxidized PS,
especially for oxidized meso PS. This effect has been ascribed to the polarizability
of the remaining silicon crystallites in the OPS film [Un1].
The doping dependence of the PS formation process and the possibility of
transforming PS to oxide at relatively low temperatures has been used to form
dielectrically isolated silicon islands, as shown in Fig. 10.23.
The observation of visible PL from as-prepared microporous silicon led to the
question of whether the PL is still present after partial oxidation of this material
[It2, Le15]. Investigations of this matter are not aiming at full conversion of PS to
SiO
2
but to a surface passivation of the silicon crystallites. Therefore, the effects
of soft oxidation methods such as storage of PS in dry or wet air at RT [Di2] and
boiling of PS in water [Ho5] or soaking with H
2
O
2
[Ko18] have been investigated.
It was found that backbond oxidation without an attack of the surface SiH bonds
is the initial step in the oxidation reaction at low temperatures [Ka2]. The PL in-
tensity is found to increase during this process [Ok1]. Full oxidation of PS at
room temperature is reported using ozone and UV light. This treatment, however,
quenches the visible PL effectively [Ho4].
Partial oxidation of PS can be performed thermally, if the oxidation rate is re-
duced by using dry oxygen, by dilution of the oxygen with an inert gas, by a low
oxidation temperature or by short oxidation intervals. Oxidation times in the order
of seconds can be achieved by RTO, for which oxidation rates and details of the
kinetics are given in the literature [Fu1, Go2].
The dependence of the PL intensity and peak position on oxidation temperature
for three different PS samples is shown in Fig. 7.20. Oxidation at 6008C destroys
the PL, while the initial PL intensity is restored or even increased after oxidation
at 9008C. This effect can be understood as a quenching of PL because of a high
density of defects generated during the desorption of hydrogen from the internal
surface of PS. Electron spin resonance (ESR) investigations show a defect with an
isotropic resonance (g=2.0055) in densities close to 10
19
cm
3
for oxidation at
6008C [Pe1, Me9]. This corresponds to one defect per crystallite, if the crystallite
diameter is assumed to be about 5 nm in diameter.
The spectral distribution of the PL from OPS is found to be similar to that of
as-prepared PS [Ta6]. In some cases a green band is found and has been ascribed
to point defects in SiO
2
[Ka9]. The slow red-orange band is dominant, while the
fast blue-green band contributes significantly to the PL intensity for highly oxi-
dized samples [Ko1]. While the red band is correlated with the presence of small
7 Microporous Silicon 160
silicon crystallites [Cu1], it is uncertain whether the blue PL is caused by QC ef-
fects [Mi7] or by hydroxyl groups absorbed on defects in the silica network [Ta5,
Ya6]. The fact that the blue PL remains even at high annealing temperatures
(10508C), for which the red PL is quenched, supports the latter proposal [Se12].
As in the case of as-prepared samples, the peak wavelength of the red band de-
pends on crystallite size, however, it cannot be blue shifted above 1.9 eV by pro-
longed oxidation. Hydrogen-related peaks in the FTIR spectra disappear during
RTO [Sh1], which allows us to conclude that polysilane species are not responsi-
ble for the visible PL. The observation of momentum-conserving phonon signals
in PLE measurements of OPS, as shown in Fig. 7.12b, indicate a common origin
of the red luminescence band for non-oxidized and oxidized PS.
Concerning long-term stability of the PL, OPS is superior to PS. Even under la-
ser illumination in air, no degradation of the PL was observed for OPS [Sh1]. OPS
has been found to have lower resistivity and higher photoelectric efficiency than
as-anodized material [Li4]. The quite high electrical conductivity of incompletely
oxidized OPS can be understood if we recall that the average oxide thickness be-
tween two crystallites is below 3 nm, which is sufficient for tunneling of charge
carriers.
Analogous to oxidized PS, nitridized PS can be fabricated by thermal treatment
of PS in N
2
or NH
3
[Ts2, Sm4, Li5]. The dependence of PL intensity on annealing
temperature is found to be similar to that of OPS. For low annealing tempera-
7.6 Oxidized Porous Silicon 161
Fig. 7.20 Luminescence intensity
and peak position versus RTO
processing temperature for PS
samples grown on p-type silicon
substrates (A: 1 X cm, B: 1 X cm,
C: 0.07 X cm). Note the anti-cor-
relation of the PL intensity and of
the ESR signal (taken for sample
series A). After [Pe1].
tures (6008008C) the PL is quenched, while the PL is restored for high anneal-
ing temperatures (80011008C) [Li5]. The optimal initial porosity of PS required
for a complete transformation of PS to Si
3
N
4
is 20%. Thick silicon nitride layers
can be formed in ammonia (at 0.1 bar) in 1 h at 7308C.
7.7
Related Materials
This section summarizes the properties of materials similar to PS but not formed
by anodization of a bulk silicon electrode. These materials fall into the following
categories:
1. PS formed chemically without an applied potential in a mixture of HF and an
oxidizing agent.
2. PS formed anodically on amorphous or polycrystalline silicon in HF.
3. Porous layers formed anodically on other semiconductors than silicon.
4. Luminescent silicon microstructures formed by other methods than anodiza-
tion and stain etching.
7.7.1
Stain Films
PS formed chemically without an applied potential in a mixture of HF and an oxi-
dizing agent, e.g. HNO
3
, is called a stain film. The similar nature of electrochemi-
cally formed micro PS and stain films was pointed out in 1960 [Ar1]. Stain films
are usually microporous. They are of predominantly monocrystalline character
and show visible PL [Sc16, Ji1]. If metal films are present on the silicon surface,
mesopore formation may also be observed [Li10].
The formation mechanism of stain films is not understood in detail. The chem-
ical identity of the oxidizing agent seems to be of minor importance, because
HNO
3
, NaNO
2
, CrO
3
, K
2
Cr
2
O
7
, KBrO
3
and KMnO
4
showed similar results [Be19,
Fa5, Na5, Ya9]. Cathodic and anodic sites present at the silicon interface have
been proposed to be responsible for stain film formation. In the case of a pn
junction, especially if it is illuminated in the electrolyte, the different sites are eas-
ily identified: the built-in potential between differently doped areas constitutes a
galvanic element. In the early years of semiconductor device manufacturing this
effect was exploited to determine the depth of a diffused pn junction.
That the assumption of such spatially separated sites is justified has been dem-
onstrated by experiments using evaporated metal films, acting as catalytic sites
[Li10]. In an electrolyte composed of aqueous HF, H
2
O
2
and ethanol, stain film
formation has been observed under and close to evaporated thin films of Au, Pt
and Pd, while silicon samples free of metal films showed no PS formation. The
metal is assumed to act as a cathodic site, where H
2
O
2
is reduced to H
2
O under
injection of two holes into the silicon VB. These holes are consumed by the for-
7 Microporous Silicon 162
mation of PS, according to the reaction shown in Eq. (4.4). If the metal films are
patterned, most holes are consumed directly under the metal film, however, a
fraction diffuses away and leads to PS formation in regions nearby. The doping
type and density of the silicon substrate determines pore morphology and thereby
film porosity.
For homogeneously doped silicon samples free of metals the identification of
cathodic and anodic sites is difficult. In the frame of the quantum size formation
model for micro PS, as discussed in Section 7.1, it can be speculated that hole in-
jection by an oxidizing species, according to Eq. (2.2), predominantly occurs into
the bulk silicon, because a quantum-confined feature shows an increased VB en-
ergy. As a result, hole injection is expected to occur predominantly at the bulk-po-
rous interface and into the bulk Si. The divalent dissolution reaction according to
Eq. (4.4) then consumes these holes under formation of micro PS. In this model
the limited thickness of stain films can be explained by a reduced rate of hole in-
jection caused by a diffusional limitation for the oxidizing species with increasing
film thickness.
Formation rate and morphology of stain films are strongly dependent on
etchant composition and on substrate doping. A delay of stain film formation of
up to 10 min has been observed, which is ascribed to the autocatalytic character of
the reaction in HFHNO
3
mixtures, as discussed in Section 2.4. In order to accel-
erate the process, the etchants are commonly primed by briefly etching a piece of
silicon in the concentrated solution prior to adding DI water. A mixture of
HF(50%) : HNO
3
(70%) : H
2
O for example produces rough surfaces consisting of
small hillocks with lateral dimensions of several hundred nanometers if the
HNO
3
concentration is high (mixing ratio 1: 5: 10) while for low HNO
3
concentra-
tions (e.g. mixing ratio 4: 1: 5) smoother surfaces are observed [Fa5, Sc18, Ca5,
Wi1]. In both solutions spontaneous staining is observed for Si samples of doping
densities in excess of 10
19
cm
3
, while lower doped samples show delayed staining
even for primed solutions. For very low HNO
3
concentrations (mixing ratio
1000: 1: 0) the staining behavior changes and all doping densities except highly
doped p-type Si show spontaneous PS film formation. In this case the growth rate
shows a parabolic rate law that is ascribed to the limiting diffusion of the oxidiz-
ing species. Growth rates of the order of 1 nm s
1
are observed. The porosity of
such stain films increases rapidly with thickness and saturates at values around
70% or more [Fr3, Sh6]. Acetic acid is sometimes used in place of water as dilu-
ent in order to relieve the surface tension, which reduces the sticking probability
for gas bubbles evolved during etching [ Je1]. In all etchants the thickness of
chemically formed porous films is limited to values below about a micrometer. In
some cases a double layer structure of the stain film has been observed; TEM
images have shown a 100 nm thick continuous layer of micro PS at the bulk in-
terface and oxide-rich flakes on top [Wi1]. The neutral P
b
center concentration is
an order of magnitude higher than in comparable anodically formed micro PS
films, indicating a higher degree of oxidation in the as-prepared stain film [Sc16].
Stain films show an intense PL at a peak wavelength of about 1.9 eV, indepen-
dent of substrate doping density [St7, Li10]. This is in contrast to anodically
7.7 Related Materials 163
formed PS, for which the visible PL intensity decreases rapidly if the substrate
doping density exceeds 10
18
cm
3
. The PL is observed from stain films grown on
single- as well as polycrystalline substrates, but not from stained amorphous thin
films [St9]. Visible EL is reported from solid-state devices using a stain-etched
film in between the bulk and an ITO (indium tin oxide) electrode [Xu3].
7.7.2
Porous Silicon on Amorphous and Polycrystalline Silicon
If a p-type polycrystalline silicon electrode is anodized in HF the dependence of
PS formation rate on the crystal orientation leads to thickness variations between
the grains, which become significant for highly doped polysilicon. In addition
pore formation is enhanced along grain boundaries, which may result in an un-
even interface and eventually lead to remaining non-anodized silicon grains in a
porous matrix [Ko21]. However, no significant difference is observed for optical
and electrical properties of PS formed on polycrystalline electrodes if compared to
PS formed on single-crystalline silicon [Ha8].
The picture is less clear if the electrode is made of amorphous Si. Device-grade
amorphous silicon is usually produced by radio-frequency glow discharge of SiH
4
at substrate temperatures between 100 and 3008C. Amorphous silicon is a highly
disordered material with an elastic scattering length below 1 nm for thermalized
carriers. Bulk amorphous silicon exhibits a more efficient PL (1.21.5 eV) than
crystalline silicon at low temperatures. The origin of PL from bulk amorphous sil-
icon, as well as from anodized porous amorphous silicon, is still under discussion
[Oh2, Al6, St3, Bu4, We2].
7.7.3
Pore Formation in Semiconductor Electrodes Other Than Si
The pore formation models discussed in Section 6.2 are based on the semicon-
ducting properties of the electrode but not on its chemical identity. Consequently
electrochemical pore formation has been observed for many semiconductors.
There is no solubility limit for Ge in Si, and so Si
1x
Ge
x
layers, with x varying
from a few per cent to 100%, have been investigated. Porous Si
1x
Ge
x
layers
(x=520%) formed by anodization in HF showed red luminescence under UV ex-
citation and the morphology seems to change from microporous to mesoporous
with increasing p-type doping level, as observed for PS [Sc14]. Even on pure p-
type Ge a porous layer is formed upon anodization in HF. This porous germa-
nium shows a broad PL band well above the bandgap at 1.17 eV, which blue shifts
to 2.15 eV after oxidation at 6008C [Mi2]. Mesoporous structures have been ob-
served on n-type 6H-SiC electrodes in HF under UV illumination [Sh5].
The electrochemical formation of porous structures is also observed for IIIV
semiconductors like GaP [An1, Er1], GaN [Pe7, My1], InP [Ki2, Ko16, Ta13, La10]
or GaAs [Be5, Fa4, Sc15]. Structural dimensions in the macroporous regime are
observed for n-type GaAs of moderate doping (10
17
cm
3
) anodized in KOH in the
7 Microporous Silicon 164
dark [Fa4] or in NH
4
OH under illumination [Mo9], as well as for GaN anodized
under UV illumination in KOH or H
3
PO
4
[Pe7]. Macroporous structures are also
observed for n-type InP after anodization in HCl [Ko16, Ta13, La10]. Mesoporous
films are found on highly doped GaAs anodized in HF [Be5]. Green PL (540 nm)
from porous GaAs, intense enough to be visible to the naked eye, has been re-
ported [Sc15]. Porous films have also been observed after photoelectrochemical an-
odization of IIVI semiconductors such as CdSe [Te1], CdTe [Mu5, Er2] and ZnSe
[Te2].
7.7.4
Luminescent Silicon Microstructures
The visible luminescence from PS has been ascribed to QC effects [Ca1]. This in-
terpretation is supported by the observation of a similar luminescence from sili-
con clusters in the nanometer regime or large silicon molecules that are not pro-
duced by anodization but by alternative methods.
Silicon microstructures can be categorized according to the dimensionality of
the confinement. Most PL studies deal with silicon structures confined in three
dimensions; such dot-like structures are designated zero-dimensional (0D). An
overview of size-dependent properties of silicon spheres is given in Table 6.1. Stan-
dard methods of generating such microstructures are gas-phase synthesis [Di3, Li7,
Sc12], plasma CVD [Ru2, Co1, Ta8] or conventional chemical synthesis [Ma15].
The visible blue and red EL, observed from silicon-rich oxides formed by CVD
of SiO
2
with excess Si, was attributed in 1984 to QC effects in the tiny Si precipi-
tates present in the films [Di3]. The blue luminescence was later attributed to OH
groups in the silica network, while the red luminescence is believed to be due to
QC effects [Ta5, Ko3]. Only the red PL band is observed for silicon microcrystal-
lites that are not in an SiO
2
matrix but covered by hydrogen, as is the case for mi-
crowave plasma decomposition of SiH
4
in hydrogen [Ta8]. In situ PL measure-
ments of hydrogenated silicon particles suspended in a silane plasma at RT
showed that the PL intensity is sensitive to the onset of agglomeration but not to
the oxygen content of the plasma [Co1]. Comparative studies of PS and oxide-pas-
sivated silicon microcrystallites, formed at 10008C in He by decomposition of disi-
lane and subsequent oxidation, indicate that particles of 12 nm in diameter are
present in red luminescing PS [Sc12, Br9].
High-frequency spark discharge is also found to be a sufficient method to gen-
erate a porous layer at silicon cathodes that exhibit visible PL bands in the green
(2.22.4 eV) and in the blue (3.2 eV). The PL shows a rapid decay (510 ns), a
quantum efficiency in the order of 1% and is found to be stable up to annealing
temperatures of 10008C. These properties of the PL, and the dependence of the
PL on the presence of oxygen and nitrogen during the spark processing, indicate
an SiO
2
-related PL, rather than a quantum dot-related one [Lu1].
Silicon wires or pillars are one-dimensional (1D) structures that are confined in
only two dimensions. Weak red PL as well as EL is observed from silicon pillars
7.7 Related Materials 165
with diameters below 10 nm fabricated by standard semiconductor manufacturing
techniques such as lithography, dry etching followed by thermal oxidation [Na2].
A sheet of silicon is a two-dimensional (2D) structure that is confined in only
one dimension. Such a sheet has been fabricated by magnetron sputtering [Su6]
or from a silicon-on-insulator (SOI) wafer if the thin superficial Si layer is further
thinned by thermal oxidation [Ta11]. An advantage of such SiO
2
/Si/SiO
2
sandwich
structures is that the thickness of the Si quantum well can be adjusted and mea-
sured with high precision. The PL of these 2D structures at 10 K is very similar to
that observed from PS concerning peak energy and blue shift with size reduction.
A 2 nm well for example shows a PL maximum at 1.65 eV, while for 1.4 nm a
maximum at 2 eV is observed. The temperature dependence, however, is very dif-
ferent. While the PL intensity of PS shows a maximum between 50 and 150 K,
the PL of the 2D structures decreases monotonically with increasing temperature.
Annealing of the sandwich structure at 11008C for 30 min is found to increase
the PL at RT by two orders of magnitude.
For inorganic and organic silicon compounds the same dimensional hierarchy
applies. Oligosilanes, for example, can be understood as quantum dots, polysi-
lanes as quantum wires and polysilynes as quantum planes. For all of these struc-
tures luminescence is quite a common phenomenon. The HOMO/LUMO gap de-
creases with increasing Si backbone dimension. Because of their smaller size, the
PL peak energy of these compounds is usually higher than observed for PS. Poly-
silanes, for example, show PL peak energies from 3 to 3.8 eV and Stokes shifts
ranging from 0.05 to 0.5 eV [Ma15, Hi8]. Octasilacubane, a cube-like molecule of
eight silicon atoms, which can be viewed as a model of the smallest three-dimen-
sionally bonded Si nanocrystal, shows a PL peaking at about 2.5 eV at RT [Ma16].
The PL properties of a few silicon-based compounds are discussed briefly in Sec-
tion 7.5.
7 Microporous Silicon 166
8.1
Mesopore Formation Mechanisms
This section is devoted to the mechanisms responsible for the formation of meso-
pores on p-type and n-type silicon electrodes. Pore formation in silicon electrodes
is only observed under anodic conditions in HF for current densities below J
PS
. In
this case a depletion region is present in silicon electrodes independent of type
and density of doping, as discussed in Section 3.2. With increasing doping den-
sity the electric field strength increases and the width of the depletion region de-
creases. This enables charge carriers to pass through the SCR by band-to-band
tunneling. For a planar silicon junction tunneling dominates the charge transfer
for doping densities in excess of 10
18
cm
3
, corresponding to a maximum field
strength of about 1 MVcm
1
, while avalanche breakdown dominates at lower dop-
ing densities. Note that in contrast to solid-state devices, where breakdown leads
to high currents and may eventually destroy the device by thermal heating, the
currents in the electrochemical cell are always limited by mass transport of chemi-
cal reactants in the electrolyte and by reaction kinetics.
The electric field depends not only on doping and on applied bias, but also on
geometry. Around a depression in the electrode the depletion region width is de-
creased and the field strength is increased, which increases the tunneling prob-
ability of charge carriers and thereby the local current density there [Zh3]. This ef-
fect becomes significant if the radius of curvature of the etch pit is smaller than
the SCR width. If the geometry of a pore tip is assumed to be hemispherical, the
breakdown bias can be approximated by calculating the breakdown bias for a
spherical junction, which is plotted as a function of its radius of curvature in
Fig. 8.1b. Breakdown at the pore walls can be approximated by calculating the
breakdown conditions for the cylindrical case, which is shown in Fig. 8.1a. In
these calculations breakdown by band-to-band tunneling is assumed to start at an
electric field strength in excess of 1.2 MVcm
1
at the junction, while avalanche
breakdown is assumed to occur if the ionization integral equals unity [Le23]. The
curves in Fig. 8.1 show that:
The bias needed for breakdown decreases with pore diameter and becomes as
low as 1 V independent of doping density for pore tip diameters below 20 nm.
167
8
Mesoporous Silicon
Electrochemistry of Silicon: Instrumentation, Science, Materials and Applications.
Volker Lehmann
Copyright 2002 Wiley-VCH Verlag GmbH
ISBNs: 3-527-29321-3 (Hardcover); 3-527-60027-2 (Electronic)
For moderately doped substrates the crossover from tunneling to avalanche
breakdown occurs at pore diameters of about 500 nm, corresponding to a bias in
excess of 10 V. Above doping densities of 10
17
cm
3
breakdown is always domi-
nated by tunneling. Tunneling is therefore expected to dominate all pore forma-
tion in the mesoporous regime and extends well into the lower macropore re-
gime, while avalanche breakdown is expected to produce structures of macropor-
ous size.
8 Mesoporous Silicon 168
Fig. 8.1 (a) Calculated breakdown bias for
a cylindrical Schottky junction as a function of
radius of curvature for tunneling (open sym-
bols) and avalanche multiplication (full sym-
bols) of charge carriers, for substrates of dif-
ferent doping density. The mechanism active
at lower bias dominates (full line). Experi-
mental results (in 1: 1 ethanoic HF) for pore
radius as a function of anodization bias are
plotted as crosses with error bars.
(b) Calculated breakdown bias for a spherical
Schottky junction. After [Le23].
The bias needed for breakdown at the cylindrical pore wall is a factor of 2 to 5
larger than for the pore tip. A comparison of the calculated results with experi-
mental observations shows that the bias dependence of the actual pore diameters
is well described by the calculation for the cylindrical case, as shown in Fig. 8.1a.
This can be understood if one recalls the current distribution at a real pore tip, ac-
cording to Fig. 9.12f. The current density decreases from the tip towards the wall
according to a cosine law and becomes zero when the pore wall is truly cylindri-
cal. At the wall segment that is close to being cylindrical, the current density be-
comes small, but not zero. Because any current flow is associated with the break-
down condition, the pore diameter increases until the limiting value for tunneling
at the pore wall, according to Fig. 8.1a, is reached.
Now the consequences of type and density of substrate doping on pore geome-
try and pore spacing will be addressed. Below the concept of a homogeneous SCR
with a constant width W, as described in Eq. (1.2), is used, independently of the
absolute value of the doping density. However, it should be noted that for doping
densities of 10
19
cm
3
and 1 V applied bias the SCR width (10 nm) exceeds the
average distance of dopants (6 nm) only by a factor of two, as shown in the dia-
gram on the inner front cover of this book. The width and the electric field in the
SCR is therefore inhomogeneous and it can be assumed that the growth of a me-
sopore depends on the local field distribution.
If the distances between the pores become significantly larger than twice the
SCR width W, the pore wall regions will no longer be depleted. This enables
branching at the pore tips and side pores would penetrate this region until deple-
tion is established again. As a result of this mechanism, pores tend to space
themselves at distances below 2W, as shown by the broken line in Fig. 6.10. In
highly doped p-type or n-type silicon the pore diameter is comparable to W, and
so mesoporous layers of significant porosity are formed. For low doped n-type
electrodes, on the other hand, the pore diameter is much smaller than W, and so
mesoporous layers of very low porosity are preferably formed in this case.
Not only the doping density but also the type of doping is important, because it
determines whether the silicon electrode is under forward or reverse conditions. A
p-type silicon electrode is under forward conditions in the anodic regime. The sol-
id-state equivalent is a forward-biased non-planar Schottky junction. Depending
on doping density the forward current of a Schottky diode is dominated by diffu-
sion, thermionic emission over the barrier, or by tunneling through the barrier. At
room temperature and for planar junctions the crossover between thermionic
emission and tunneling occurs at a doping density of about 10
18
cm
3
. The field
at a spherical pore tip is significantly larger than in the planar case and mesopore
formation due to tunneling is observed for p-type doping densities up to
3 10
17
cm
3
. The forward bias of a Schottky diode does not exceed 1 V. Therefore
only small mesopores with radii around 5 nm are expected in this doping regime,
according to the calculations shown in Fig. 8.1a. This is in accordance with experi-
mental observation, as shown in the SEM micrographs of Section 8.2. Avalanche
breakdown in p-type electrodes is, of course, confined to the cathodic regime, and
is therefore not relevant to pore formation.
8.1 Mesopore Formation Mechanisms 169
In contrast, n-type electrodes are under reverse conditions in the anodic regime.
However, no significant morphological differences to the p-type case are expected,
as long as the n-type substrate doping density is in excess of about 10
18
cm
3
. In
this case the formation bias is in the order of 1 V, which again produces meso-
pore radii around 5 nm, as shown in the SEM micrographs in Section 8.3. With
decreasing doping density, however, the bias required for breakdown increases. In
this regime the dependence of pore wall radius on applied bias becomes clearly
observable, as shown by the measurement in Fig. 8.1a (crosses). Because the tip
radius is always less than or equal to the wall radius, the pore tip current density
is not limited by charge carrier tunneling. This leaves reaction kinetics or ionic
diffusion as limiting mechanisms for the pore tip current. This fact, together with
the observation that the dissolution valence is greater than 2 for macropore forma-
tion supports the assumption that the pore tip current density is identical to the
critical current density J
PS
. The dependence of J
PS
on crystallographic orientation
now becomes important. J
PS
is highest along the h100i directions of the Si crystal,
as shown in Fig. 4.9, and so mesopores tend to grow in this direction. Mesopor-
ous films therefore exhibit the maximum thickness on (100) planes, as shown in
Fig. 6.3. In addition side pores are formed orthogonal to the main pore along the
other h100i directions. This mechanism is called spiking. If the main pore grows
deep into the bulk of the electrode the diffusional limitations become significant
and the pore growth rate may decrease to values close to zero. Mesopore forma-
tion and spiking on low doped n-type Si are discussed in Section 8.4.
For n-type doping densities below 10
17
cm
3
and an anodization bias above
10 V, avalanche breakdown becomes relevant. The interface morphology generated
in this regime is very complex and shows large etch pits, macropores and meso-
pores. The formation of this structure is not understood in detail. A hypothetical
model will be discussed in Section 8.5.
A good example of the consequences of the pore formation mechanisms dis-
cussed so far is the pore morphology formed in a p/p
+
multilayer structure. In or-
der to manufacture interference mirrors, a stack of 75 nm thick homogeneously
p-doped (10
17
and 10
19
cm
3
) Si layers have been grown by LPCVD, as shown in
the inset of Fig. 10.14. After anodization of the stack at 50 mA cm
2
in ethanoic
HF, the TEM micrograph (Fig. 10.14), as well as a simulation of the measured
spectral reflectivity (Fig. 10.15), indicated a triple layer structure with a significant
porosity gradient [Be16]. That three layers of different porosity have been formed
from only two doping densities can be understood in the frame of the mesopore
formation mechanisms:
The electric field in an anodized silicon electrode of a p-type doping density of
10
17
cm
3
is too small for breakdown, and so the charge transfer is dominated by
thermionic emission over the barrier and only micro PS of high porosity (67%) is
formed. The bias across the junction can be approximated to 0.1 V, as discussed
in Section 9.2. This corresponds to an SCR width of about 35 nm. If the micro-
pores reach a depth of 40 nm the SCR reaches the highly doped p-region. For sol-
id-state pin diodes this effect is known as punch-through. Because of the high
doping density in the p
+
region (10
19
cm
3
), the SCR width now monotonically de-
8 Mesoporous Silicon 170
creases with the increasing micropore depth. This in turn increases the field
strength at the pore tips until it approaches the value required for junction break-
down by tunneling. Let us assume that this value is reached at a total micro PS
thickness of 50 nm. The bias required for tunneling now becomes less than the
bias for thermionic emission and micropore formation stops, while a few meso-
pores grow by tunneling through the remaining 25 nm of low doped p-Si until
they reach the interface to the p
+
region. Note that the entire low doped region is
depleted, pore wall passivation is therefore established there, even for pore separa-
tions of tens of nanometers, and so a layer of very low porosity (5%) is formed. In
the p
+
region in contrast pore wall passivation requires a pore wall thickness of
the order of 3.5 nm. As a result, the pores start to branch when they reach the p
+
region and a network of closely spaced mesopores of moderate porosity (55%) is
formed. Due to the fact that the SCR now only extends 3.5 nm, the next low
doped p-type region does not affect the field at the pore tips until they are as
close as 3.5 nm. The sequence is then repeated.
8.2
Mesopores in Highly Doped p-Type Silicon
The morphology of mesoporous structures can be investigated by TEM, as shown
in Fig. 7.2b and d. However, sample preparation for TEM is time-consuming. For-
tunately the mesoporous regime can also be studied by todays most advanced
SEMs. For SEM inspection no sample preparation is required, and even sputter-
ing of the sample is dispensable, as shown by the SEM micrographs displayed in
this and the next section.
The initiation process of mesopores shows similarities to macropore initiation.
The growth of mesopores can be initiated by a depression in the electrode, for ex-
ample an alkaline etch pit [Le23]. On flat electrodes a doping-dependent density of
etch pits is generated during the first seconds of anodization in HF, as shown in
Fig. 8.2. A certain fraction of these etch pits develop into mesopores, for which
the density is again a function of doping density, as shown in Fig. 6.10. The
charge consumed during pore nucleation on a flat electrode until stable pore
growth is reached can be determined by impedance measurements. During the
initiation period of mesopores on highly doped p-type Si a charge of about
100 mAs cm
2
is consumed, which is significantly larger than the charge observed
for micropore initiation on low doped p-type Si (25 mAs cm
2
) [Po3].
In order to investigate the mesopore morphology versus pore depth, samples
are cleaved along a (110) plane after anodization. The dependence of mesopore
morphology on doping density and formation current density is shown in Fig. 8.3
for anodization in ethanoic HF and in Fig. 8.4 for 6% aqueous HF. Values of po-
rosity, growth rate and dissolution valence are given for each sample above the
micrograph and are plotted as a function of doping density in Fig. 6.9. A common
feature of all mesoporous layers formed on highly doped substrates is a distinct
interface between the mesoporous region and the bulk. This interface is easily
8.2 Mesopores in Highly Doped p-Type Silicon 171
8 Mesoporous Silicon 172
Fig. 8.2 SEM micrographs of the initially polished (100) silicon electrode
surfaces after galvanostatic anodization in ethanoic HF. After [Le23].
Fig. 8.3 SEM micrographs of the interface between bulk and meso PS for p-type doped (100)
silicon electrodes anodized galvanostatically in ethanoic HF. After [Le23].
identified even by optical microscopy if the sample is cleaved, as shown in
Fig. 6.3. This distinct interface is an indication of a depletion of charge carriers
which promote dissolution in the pore walls, as is the case during microporous
silicon formation.
For very high doping densities and large formation current densities, the pore
dimensions approach the macroporous regime, as shown in the upper right of
Fig. 8.3. In this regime the pore diameter depends approximately exponentially on
current density. For p-type substrates of 1 mX cm anodized in ethanoic HF at
600 mA cm
2
, pore diameters of 1 lm and porosities above 90% have been ob-
served [ Ja4].
Using SIMS or elastically recoiling particle detection spectroscopy (ERDA) the
remaining concentration of acceptors and donors was measured in meso PS
formed on highly doped substrates. No preferential dissolution of dopant atoms
has been observed [Le14, Po2]. As expected from the dimensions of the mesopor-
ous skeleton, which show no or minimal QC, the optical behavior of meso PS is
similar to that of bulk silicon for photon energies above the Si bandgap. The lu-
minescence efficiency in the visible regime is usually negligible. Only meso PS
samples of high porosity, which contain a certain fraction of small crystallites,
show a weak PL signal one- or two-tenths of eV above the Si bandgap energy
[Po2]. However, the optical absorption for below bandgap photons is significantly
reduced for dry mesoporous layers, if compared to the highly doped substrate, as
shown in Fig. 7.8. This has been ascribed to a depletion of free charge carriers in
8.2 Mesopores in Highly Doped p-Type Silicon 173
Fig. 8.4 SEM micrographs of the interface
between bulk and meso PS for p-type doped
(100) silicon electrodes anodized in 6%
aqueous HF. For high current densities and
high doping densities (upper right) the PS
layers disintegrate during drying. After [Le23].
the porous region due to a trapping in surface states present at the large inner
surface [Le14]. For wet mesoporous silicon these surface states appear to be partly
passivated, because the conductivity of wet mesoporous silicon is higher than for
the dry case. The passivation of p
+
type substrates in alkaline solutions is lost if it
is made mesoporous, as discussed in Section 2.3.
8.3
Mesopores in Highly Doped n-Type Silicon
An overview of the morphologies of mesopores grown under galvanostatic condi-
tions on n-type substrates of different doping densities is shown in Figs. 8.6 and
8.7. Note that the mesopore morphology depends on doping density, not on the
elementary nature of the dopant. Even implanted protons activated by a post-im-
plant anneal are reported to form a highly n-doped layer sufficient for mesopore
formation [Tu5].
The decisive difference to the p-type case is that n-type silicon electrodes are un-
der reverse conditions during mesopore formation. This becomes apparent if the
pore morphologies are compared. While the morphology of mesoporous silicon
formed on n-type substrates of doping densities in excess of 10
19
cm
3
is very sim-
ilar, as can be seen by comparison with Figs. 8.3 and 8.6, the difference becomes
apparent at lower doping densities. For p-type substrates the SCR field becomes
insufficient for breakdown below doping densities of 10
17
cm
3
and only micro PS
or macro PS formation is observed in this regime. For n-type electrodes, in con-
trast, mesopore formation is not quenched for doping densities below 10
17
cm
3
.
The bias required for breakdown at the flat n-type electrodes of this doping den-
8 Mesoporous Silicon 174
Fig. 8.5 SEM micrographs of an Si electrode
(n-type 10
18
cm
3
(100), 300 m A cm
2
, 60 s)
anodized galvanostatically in ethanoic HF.
An increase in pore diameter and porosity
with depth is observable. After [Le23].
sity in the dark is in excess of 10 V. It decreases with decreasing radius of curva-
ture of the pore. The pore diameter in this doping regime can therefore be ad-
justed by anodization bias, as shown by the experimental data plotted in Fig. 8.1a.
In contrast to electrochemical macropore formation, spiking and formation of
side pores cannot be avoided, because of the high applied bias. The technological
impact of this method of tuning the mesopore diameter is therefore limited.
Another consequence of the SCR width being significantly larger than for the
corresponding p-type case, is the lower porosity observed for meso PS on n-type
electrodes, as shown in Fig. 6.9c. The porosity increases with current density and
decreases with HF concentration [Na8, Le23]. But porosity is also a function of
meso PS layer thickness, especially for high current densities [It3, Th4, Le23]. The
morphology of mesopores changes with depth, from narrow, strongly branched
pores to wider unbranched ones, as shown in Fig. 8.5. This can be understood as
a consequence of diffusion-limited mass transfer in the pores. Because of the dif-
fusion gradient the HF concentration decreases with increasing pore depth. This
again reduces the limiting pore tip current density J
PS
according to Eq. (4.9),
which results in an increase in pore diameter under galvanostatic conditions. An
8.3 Mesopores in Highly Doped n-Type Silicon 175
Fig. 8.6 SEM micrographs of the interface between bulk and meso PS for n-type doped (100)
silicon electrodes anodized galvanostatically in ethanoic HF. After [Le23].
9, 154, 2.66, 50-70
10, 16.3, 2.25,<10
increase in applied current density may even lead to a separation of the mesopor-
ous layers from the substrate, which is a common method of producing free-
standing films.
The dependence of mesopore formation on crystal orientation becomes appar-
ent in Figs. 8.6 and 8.7. For moderate doping densities or high current densities
preferential growth in the h100i direction is observed. This anisotropy is en-
hanced further if anodization is performed under a magnetic field [Na8]. On (111)
substrates the h113i direction can be identified as a preferred growth direction, as
shown in Fig. 8.8. Note that the same orientation dependence has been observed
for macropore formation on (100) and (111) substrates, as shown in Figs. 9.7 and
9.14, respectively. The preferential direction of mesopore growth is reduced for
high doping densities and moderate to low current densities due to severe branch-
ing of pores, as shown in the lower part of Fig. 8.6. It can be speculated that the
growth direction of a pore at high doping densities becomes dominated by the
local electric field distribution at the pore tip, which depends on the position of
individual dopant atoms in reference to the pore tip. The average growth rate of
mesoporous films, however, is still found to be highest on (100) substrates, as
illustrated by the etched sample edge in Fig. 6.3 or by the data displayed in
Fig. 6.4.
8 Mesoporous Silicon 176
Fig. 8.7 SEM micrographs of the interface between bulk and meso PS for n-type doped (100)
silicon electrodes anodized galvanostatically in 6% aqueous HF. After [Le23].
8.4
Mesopore Formation and Spiking in Low-Doped n-Type Silicon
The formation of pores during anodization of an initially flat silicon electrode in
HF affects the IV characteristics. While this effect is small for p-type and highly
doped n-type samples, it becomes dramatic for moderate and low doped n-type
substrates anodized in the dark. In the latter case a reproducible IV curve in the
common sense does not exist. If, for example, a constant potential is applied to
the electrode the current density usually increases monotonically with anodization
time [Th1, Th2]. Therefore the IV characteristic, as shown in Fig. 8.9, is sensitive
to scan speed. The reverse is true for application of a certain current density. In
this case the potential jumps to values close to the breakdown bias for the flat
electrode and decreases to much lower values for prolonged anodization. These
transient effects are caused by formation of pores in the initially flat surface. The
lowering of the breakdown bias at the pore tips leads to local breakdown either by
tunneling or by avalanche multiplication. The prior case will be discussed in this
section while the next section focuses on the latter.
According to Fig. 8.1a the diameter of mesopores decreases with decreasing
substrate doping density for anodization at a certain bias. The pore spacing, on
the other hand, depends on SCR width and therefore increases with doping den-
sity. As a result, the porosity of the porous layer on low doped n-type silicon is
very low and, in contrast to other types of PS, no clear interface between the me-
soporous region and the bulk can be identified in microscopic images. Simple op-
tical microscopy, even at magnifications of 1000, is usually not sufficient to reveal
individual mesopores. In the optical micrograph of an n-type Si electrode cleaved
after anodization, as shown in Fig. 8.10a, only a few larger pores normal to the
electrode surface are visible. If interference contrast is used, the porous region
8.4 Mesopore Formation and Spiking in Low-Doped n-Type Silicon 177
Fig. 8.8 SEM micrograph of the interface be-
tween bulk and meso PS for an n-type doped
electrode of (111) crystal orientation anodized
galvanostatically in ethanoic HF (510
18
cm
3
,
50 mA cm
2
). After [Le23].
can be identified better because of the increase in roughness of the cleaved sur-
face it induces, as shown in Fig. 8.10b. By a short etch in a silicon defect etchant
the pores are sufficiently enlarged to become visible by optical microscopy. Now
mesopores spiking out of the main pore in the other h100i directions become visi-
ble, as shown in Fig. 8.10c and d. Under the high magnification of a TEM the
needle spiking pores can be directly revealed, as shown in Fig. 8.11. However, it
requires formation of an oxide replica and etchback to give a clear idea of the me-
sopore morphology [Le9]. This technique is used in Fig. 8.12 to reveal the mor-
phology of mesopores spiking out of macropore walls. Such spiking is usually un-
desirable during macropore array fabrication. The formation of spiking mesopores
can be suppressed if low bias, low substrate doping density and high array porosi-
ty are used.
Pore growth in silicon electrodes shows a crystal anisotropy [Hi7, Ch11, Le8,
Ro8, Ch17]. This effect is most pronounced for mesopores in low doped n-type
electrodes and will therefore be discussed here. Mesopores in low doped n-type
silicon are found to always grow along the h100i directions of the silicon crystal.
The resulting pore morphology is shown for a (100) and a (111) Si substrate in
Fig. 8.13. This orientation dependence has been ascribed to the anisotropy of the
critical current density J
PS
[Le9]. As shown in Fig. 4.9, J
PS
is found to show a max-
imum on (100) substrates, independent of the HF concentration used. As dis-
cussed in Section 8.1, the current density at the mesopore wall is limited by tun-
neling, while at the pore tip charge supply is not the limiting factor. Therefore the
tip current approaches J
PS
and the anisotropy of J
PS
becomes decisive. The disso-
lution rate is maximum for the h100i directions and so pore tips are formed pre-
ferentially along these directions. This orientation dependence increases with in-
creasing anodization bias. If pores grow into the depletion region of neighboring
8 Mesoporous Silicon 178
Fig. 8.9 Current density versus potential for
n-type electrodes of different doping density
anodized in 5% aqueous HF. The current
density has been measured 30 s after a step-
wise increase in bias. Redrawn from [Th2].
pores, the electric field at the tip is reduced and eventually the pore stops grow-
ing. This leads to morphologies as depicted in Fig. 8.13. Mesopore formation on
low doped n-type substrates has been observed to occur exclusively along the
h100i directions of the Si crystal, independent of substrate orientation [Ro8]. This
is in contrast to macropores (Fig. 9.14) or mesopores on highly doped substrates
(Fig. 8.8), for which pores along the h113i directions are found on (111) oriented
electrodes.
8.4 Mesopore Formation and Spiking in Low-Doped n-Type Silicon 179
Fig. 8.10 (a) Optical microscopy (magnifica-
tion 1000) of a cleaved (100) n-type Si elec-
trode after 300 s at 10 V in ethanoic HF.
(b) The roughness in the porous region of
the cleaved interface becomes visible if inter-
ference contrast mode is used. (c) The high
density of breakdown pores becomes appar-
ent after pore widening by chemical etching
in KOH. (d) Same area as in (c) with interfer-
ence contrast.
Fig. 8.11 (a) TEM micrograph of mesopores
(in a plane parallel to the sample surface) in-
itiated by a surface pattern of etch pits (5 V,
3% HF, RT, 10
15
cm
3
n-type, pattern according
to inset of Fig. 9.9). (b) While the center pore
shows a diameter of about 40 nm, the dia-
meter of the smallest spiking pores is not
resolved (<5 nm). After [Le9].
a b
8.5
Etch Pit Formation by Avalanche Breakdown in Low-Doped n-Type Silicon
From the calculated curves in Fig. 8.1 it can be concluded that breakdown in n-
type substrates of a doping density below 10
17
cm
3
at an applied bias in excess of
10 V is dominated by avalanche breakdown. Anodization in this regime produces
a complex morphology, dominated by large spherical etch pits with diameters in
the order of tens of micrometers, as shown for different substrate doping densi-
ties in Fig. 8.14. The density of these etch pits increases with increasing doping
density, as indicated by downward-pointing triangles in Fig. 6.10. If the cross-sec-
tion of such a pit is inspected by SEM, macro- as well as mesopores are found in
the outer perimeter of the pit [Le23]. The inner perimeter shows significant etch-
ing but no pore formation, except the large central cavity. The position of the etch
8 Mesoporous Silicon 180
Fig. 8.12 (a) SEM micrograph of the cleaved
wall area between two macropores that have
been anodically oxidized (70 nm) after their
formation. The existence of breakdown spikes
is only indicated by a small surface roughness
of the cleaved macropore wall. (b) After Si
etchback the oxide replica of a high density
of spiking breakdown mesopores is revealed.
Note that the spikes appear distorted because
of stress in the oxide film.
Fig. 8.13 Dependence of the mesopore
morphology on the crystal orientation of
the silicon substrate. After [Le8].
pits is random on a polished electrode surface, but can be predetermined by a de-
pression, as for example produced by alkaline etching.
The regime of avalanche breakdown has not been studied in detail and only a
rough picture of the formation mechanism can be drawn [Le23]. In contrast to
solid-state devices where breakdown results in high current densities that may de-
stroy the device by thermal heating, the current density in electrolyte-filled pores
and pits is limited by mass transport of chemical reactants to values below J
PS
. It
can therefore be assumed that avalanche breakdown occurs in a transient manner.
Prior to breakdown only a small dark current density is present, which is too
small to generate a concentration gradient in the HF. The local current density
then increases dramatically by avalanche multiplication. It is, however, limited by
electrolyte conductivity and ionic diffusion. Most of the generated charge is now
consumed to form the central etch pit, which reduces the HF concentration in
and around the pit considerably and drives this interface area into the electropol-
ishing regime. The potential drop across the anodic oxide film and in the electro-
lyte now efficiently reduces the field in the SCR and thereby stops the avalanche
multiplication process. The circular region of dissolution and macropore forma-
tion around the large etch pit can be understood as being a result of etching
caused by charge carriers generated by avalanche multiplication in the bulk of the
electrode but not consumed to form the large central pit. These remaining charge
carriers diffuse to the surface, generating a current density that decreases with the
radial distance to the central pit, thus generating a pore-free inner ring of a disso-
lution rate close to or in the electropolishing regime and a surrounding ring
showing macropore and mesopore formation.
8.5 Etch Pit Formation by Avalanche Breakdown in Low-Doped n-Type Silicon 181
Fig. 8.14 The density of avalanche etch pits
depends on substrate doping density, as
shown by optical micrographs of two n-type
electrodes anodized at high bias in 6% aque-
ous HF in the dark (a) 50 V, 30 s, 1.5 X cm;
(b) 100 V, 30 s, 5 X cm. (c, d) SEM micro-
graphs of the sample shown in (b) reveal a
circle of macro- and mesopores around the
central pit. After [Le23].
9.1
Macropore Formation Mechanisms
This section is devoted to the formation mechanisms that have been proposed as
being responsible for formation of macropores on p-type and n-type silicon elec-
trodes.
First models for macropore formation on p-type Si electrodes were based on
surface passivation by organic molecules [Po1] or on electrostatic considerations
[We5]. These models, however, were found to contradict experimental observations
[Le21]. Thermionic emission over the Schottky barrier and Fermi level pinning at
the silicon electrode surface were originally proposed to cause micro PS forma-
tion. It has been argued that because the barriers height and not its width con-
trols the current density, pores can grow into each others depletion region and
the structural sizes may become very small [Be5]. The observation of micropore
formation on n-type electrodes contradicts this model, because in this case no in-
terfacial barrier for minority carriers (holes) is present. But the proposed mecha-
nism can account for macropore formation on p-type Si electrodes. As discussed
below, pore formation can be understood as a consequence of charge transfer
across the Schottky barrier, if the non-planar interface constituted by a pore is
taken into consideration [Le21].
A p-type electrode is under forward conditions during macropore formation.
The basic charge distribution around pores in a p-type electrode is shown in
Fig. 9.1. No electric field is present in the bulk electrode and neutrality requires
the hole concentration n
h
to equal the acceptor concentration N
A
there. Close to
the interface to the electrolyte an SCR exists in the semiconductor electrode. The
width W of the SCR depends on the difference between the Fermi levels of elec-
trolyte and semiconductor, called the built-in potential V
bi
, on the p-type doping
density, on the applied bias V (according to Eq. (1.2)) and on the interface geome-
try. The latter factor is key to the understanding of pore formation and will be dis-
cussed in some detail. By solving the Poisson equation, W can be calculated for
different geometries. Assuming a doping density of N
A
=10
15
cm
3
and a potential
(V
bi
V2kT/e) of 1 V, W is 1150 nm for the planar interface. At the walls of a cylin-
drical pore of radius r =500 nm, W is reduced to 950 nm, due to geometrical field
183
9
Macroporous Silicon
Electrochemistry of Silicon: Instrumentation, Science, Materials and Applications.
Volker Lehmann
Copyright 2002 Wiley-VCH Verlag GmbH
ISBNs: 3-527-29321-3 (Hardcover); 3-527-60027-2 (Electronic)
enhancement. At a hemispherical pore tip of the same radius, W is further re-
duced to 800 nm. W becomes even smaller for conical pore tips. In conclusion, W
is minimal at the pore tip, while the barrier height E
b
=e(V
bi
V)2kT is indepen-
dent of geometry.
After having discussed the electrostatics, charge transfer is the next topic. A p-
type silicon electrode anodized in HF behaves basically like a solid-state Schottky
diode under forward conditions. The forward current of a Schottky diode is domi-
nated by diffusion, or thermionic emission, or tunneling of holes. At zero bias
and under forward conditions the SCR is not fully depleted of holes and so a dif-
fusion current I
diff
exists that is caused by the concentration gradient of holes
dn
h
/dr, according to Schottkys theory. At thermal equilibrium and with no ap-
plied bias (OCP), I
diff
is compensated by the field current I
field
and I
diff
=I
field
is
fulfilled at the pore walls as well as at the pore tip. However, the absolute values
of I
diff
and I
field
are larger at the pore tip, because the concentration gradient dn
h
/
dr, as well as the electric field strength, increase with decreasing W, as shown in
the left part of Fig. 9.2. If a forward bias V is applied, I
field
decreases, while I
diff
in-
creases. The higher current density values at the pore tip now become decisive.
The total current density at the tip, I
tip
=(I
diff
I
field
)
tip
, is always larger than the
current density at the wall I
wall
=(I
diff
I
field
)
wall
. Therefore a depression in the elec-
trode etches faster than a planar area a pore tip develops. If the distance be-
tween two neighboring pores becomes less than 2W their SCRs overlap and the
pore wall becomes passivated due to the depletion of holes.
The above model is sufficient to qualitatively understand the basic properties of
macroporous layers formed on p-type substrates, for example their porosity, which
is usually high. However, the observed suppression of macropore formation in
9 Macroporous Silicon 184
Fig. 9.1 The charge distribution around
pores in a low doped p-type silicon electrode.
Two types of charge carriers are present in
the electrode: holes, the majority charge car-
riers (crosses) and immobile ionized impuri-
ties (circles).
aqueous electrolytes for p-type doping densities in excess of N
A
=10
16
cm
3
cannot
be understood using the Schottky diffusion theory. It has been proposed that the
change from a diffusion-dominated charge transfer to a thermionic emission-
dominated is responsible for this effect [Le21]. According to the theory of Bethe,
the thermionic emission current I
therm
is due to holes that have sufficient energy
to overcome the potential barrier E
b
and are traversing normal to the interface.
Thermionic emission is sensitive to barrier height but not to barrier width, if the
width is less than the main free path of holes. If the charge transfer is dominated
by thermionic emission, depressions show no increase in current density com-
pared to planar areas, and so pore formation is not initiated. However, the upper
limit of p-type doping density for macropore formation calculated from this mod-
el (10
17
cm
3
) is about one order of magnitude larger than the value observed ex-
perimentally in aqueous electrolytes (10
16
cm
3
). At doping densities in excess of
10
18
cm
3
, tunneling becomes dominant, which leads to the formation of meso-
porous silicon, as discussed in Section 8.1.
In contrast to p-type electrodes, an n-type electrode is under reverse conditions
in the anodic regime. This has several consequences for pore formation. Signifi-
cant currents in a reverse biased Schottky diode are expected under breakdown
conditions or if injected or photogenerated minority carriers can be collected.
Breakdown at the pore tip due to tunneling generates mainly mesopores, while
avalanche breakdown forms larger etch pits. Both cases are discussed in Chapter
8. Macropore formation by collection of minority carriers is understood in detail
and a quantitative description is possible [Le9], which is in contrast to the pore
formation mechanisms discussed so far.
Holes, which initiate the dissolution process, are minority charge carriers in n-
type electrodes. The concentration of holes n
h
is very low in n-type Si under equi-
librium conditions. The hole concentration can be increased by illumination or by
9.1 Macropore Formation Mechanisms 185
Fig. 9.2 (a) Equilibrium (V= 0) field currents
and diffusion currents across the SCR for a
macropore geometry in p-type Si.
(b) Field currents and diffusion currents
under forward bias (V>0). Note that because
of field enhancement around the pore tip the
tip currents are always larger than the pore
wall currents.
a b
injection. Three kinds of charge carriers are now present in the bulk electrode:
holes (of concentration n
h
), electrons (n
e
) and the ionized donors (N
D
). Neutrality
is established if n
e
=N
D
+ n
h
. Concentration gradients of holes in the bulk semi-
conductor may therefore exist in n-type electrodes, because their charge can al-
ways be compensated by the electron distribution, as shown in Fig. 9.3. Hence the
strict requirement for p-type macropore formation, i.e. that the wall thickness w
needs to be less than 2W for passivation, no longer applies to n-type electrodes.
Note that the condition w=2W has been proposed to be valid for macropore for-
mation on n-type Si, also [Le8, Zh3, Se8]. But is has been experimentally shown
that macropores on n-type Si may be separated by as much as 10W [Le21, Ri4]. In
this case the concentration of holes becomes close to zero between the pore walls,
due to their diffusional distribution, as shown in Fig. 9.3. The distance between
two n-type macropores required for passivation is therefore limited by the minor-
ity carrier diffusion length [Le21] rather than by the SCR width [Se8]. In terms of
the DLA model [Me8] discussed in Section 6.2, the minority carrier diffusion
length L
D
becomes the parameter L of the model, rather than the SCR width W.
Having discussed the causes of pore wall passivity, we will now focus on the ac-
tive state of the pore tip, which is caused by its efficiency in minority carrier col-
lection. Usually the current density at the pore tip is determined by the applied
bias. This is true for all highly doped as well as low doped p-type Si electrodes
and so the pore growth rate increases with bias in these cases. For low doped, illu-
minated n-type electrodes, however, bias and current density become decoupled.
The anodic bias applied during stable macropore formation in n-type substrates is
9 Macroporous Silicon 186
Fig. 9.3 The charge distribution around
pores in a low doped, illuminated n-type sili-
con electrode. Three types of charge carriers
are present: electrons, which are the majority
carriers (dashes), holes, which are the minor-
ity carriers (crosses), and immobile ionized
impurities (circles). Note that a concentration
gradient of holes is possible, because of com-
pensation by electrons.
sufficient to produce current densities well in the electropolishing regime, while
the current density remains low because it depends on minority carrier genera-
tion rate only. Under this condition of relatively high bias and a limited supply of
holes, the current density at every pore tip increases until the pore tip is close to
electropolishing. A further increase in current density would cause a decrease in
the dissolution rate because of the increase in the dissolution valence from 2 to 4,
as discussed in Section 4.2, and a change in the pore tip condition from depletion
to inversion. Holes would accumulate around the tip and eventually enter the
pore walls and thereby destroy their passivity. The crossover from depletion to in-
version for Si electrodes in HF is characterized by the critical current density J
PS
,
which is reached at any macropore tip during stable growth [Le9]. J
PS
is a func-
tion of the HF concentration and the temperature only, as shown in Eq. (4.1). A
fixed value of current density for every pore tip produces a very stable dissolution
condition. If, for example, the hole supply for a single pore is increased, the pore
has to increase its diameter, because the tip current density is fixed. Another con-
sequence of J
PS
being the pore tip current density, is a tip growth rate indepen-
dent of tip collection efficiency. This growth rate, common to all macropores in
the electrode, produces a condition of coexistence for the pore tips, rather than a
survival of the fittest. Hence the pore density is usually constant during stable
macropore growth and changes in this value are mainly observed during pore ini-
tiation.
Pore initiation mechanisms and the initial state of the Si electrode prior to anod-
ization have so far been neglected. This is on the one hand because the investigation
of pore initiation in the micro- and mesoporous regimes requires high-resolution
techniques. Simulations, on the other hand, show that mechanisms supporting
pore formation are based on an interfacial instability, which is also sufficient to ex-
plain pore initiation, even on surfaces without lateral inhomogeneities. As a result,
micro- and mesoporous films, discussed in the preceding sections, have usually
been formed by anodization of single crystalline, defect-free, polished and homoge-
neously doped Si electrodes. This leaves open the question of in what way is pore
formation affected the topography of the electrode. The macropore regime, because
of its larger dimensions, is well suited for such investigations. For all kinds of macro-
pores it has been found that pits or depressions, generated artificially prior to ano-
dization, are sufficient to initiate macropores [Le8].
9.2
Macropores in p-Type Silicon
Macropore formation on p-type silicon electrodes was first observed for anodiza-
tion in water-free mixtures of anhydrous HF and an organic solvent [Pr7, Ri1].
Later it was observed that organic HF electrolytes with a certain fraction of water
[Po1, We5], or even non-organic, aqueous HF electrolytes [We2, Le21], are also suf-
ficient for the formation of macropores on p-type Si electrodes. This indicates that
macropore formation on such electrodes cannot be ascribed to the chemical iden-
9.2 Macropores in p-Type Silicon 187
tity of a certain electrolyte. Only the presence of HF in the electrolyte is required,
as is the case for the other pore size regimes. The pore morphology and stability
of pore growth, however, depend on electrolyte composition and concentration, as
well as on additives such as surfactants [Le21, Ch16].
The morphology of macroporous layers formed on p-type Si electrodes can be
understood as a consequence of the stringent requirement for pore wall passiva-
tion in low doped p-type substrates: pore walls are depleted if their thickness is
less than twice the SCR width W. W depends on doping density and applied bias.
The change in pore morphology with p-type doping density is shown in Fig. 9.4,
together with an n-type sample for comparison. If the average pore wall thick-
nesses from Fig. 9.4 ad are plotted versus the substrate doping density, the
square root dependence, expected from Eq. (1.2), is observed, as shown in
Fig. 9.5. The SCR width obtained from Fig. 9.5 is found to correspond to a bias of
about 0.1 V.
For n-type as well as p-type Si, the dependence of macropore density on doping
density is linear. However, the absolute values of pore density observed on non-
structured p-type substrates is about one order of magnitude larger than for n-
type electrodes, as shown in Fig. 6.10. This produces high values of porosity
(p > 0.5) for random p-type macropore structures as well as for arrays [We5, Le21].
The pore walls become fragile if the doping density is increased above 10
16
cm
3
and usually only a roughening of the electrode surface is observed for anodization
in aqueous HF.
For macropore formation on low doped p-type substrates, the applied bias and
current density are coupled: a change in the applied bias produces a correspond-
ing change in anodization current density. For macropore growth on p-type Si
9 Macroporous Silicon 188
Fig. 9.4 (ad) Optical micrographs of p-type
substrates of different doping densities po-
lished under a small angle to the electrode
surface, after anodization under identical
conditions in the dark (10% aqueous HF,
3 mA cm
2
, 90 min, RT). (e) An n-type sub-
strate anodized under similar conditions is
shown for comparison (10% HF, 5 X cm,
3 mA cm
2
, 30 min, RT). After [Le21].
electrodes an increase in the current density has little effect on pore diameter, but
produces a roughly proportional increase in pore growth rate [Le21]. Note that
just the opposite is true for macropore formation in n-type Si. Concerning growth
rate and the dissolution valence (n
v
=2.02.3), macropore formation on p-type Si
shows similarities to micro PS formation. This can be understood as a conse-
quence of the fact that depletion in p-type substrates is only present for an ap-
plied bias significantly lower than that corresponding to the critical current den-
sity J
PS
. Thus J
PS
is never reached at the pore tips or pore walls during macropore
formation on p-type Si electrodes. This is supported by the observation that micro-
porous silicon covers not only the macropore walls but also the pore tips. For
higher HF concentrations the whole pore volume can be filled with micro PS, as
shown in Fig. 9.6.
The dependence of pore morphology on crystal orientation is found to be weak
in aqueous HF. The tendency of the pores to branch is enhanced on (110) and
(111) substrates if compared to (100) substrates. Some faceting can be observed at
the pore tips. In organic solvents the orientation dependence becomes stronger
and the h100i and h113i directions are observed to be preferred for pore growth.
At high HF concentrations or in mixtures with organic solvents, most of the pore
volume is found to be filled with micro PS. The presence of an oxidizing species,
like water, reduces the amount of pore filling [Ch15].
The growth of a macropore on a p-type substrate can be initiated by artificial
etch pits. The growth of predefined pore arrays is observed to be more stable than
the growth of random pores on flat electrodes [Ch16, Le21]. If a slit is used for
pore initiation the formation of trenches separated by thin walls has been ob-
served on (100) p-type substrates [Oh5]. Note that for slits along the h110i direc-
tion the walls become (110) planes, in contrast to trenches produced by alkaline
etchants, for which only (111) oriented walls can be formed on (110) oriented sili-
con substrates.
In conclusion it can be said that the flexibility of pore array design on low
doped p-type Si is less than that for macropore formation on n-type substrates, be-
cause of the limitations in array porosity and substrate doping range.
9.2 Macropores in p-Type Silicon 189
Fig. 9.5 Values of pore wall thickness from
Fig. 9.4 ad as a function of doping density
(error bars), together with the SCR width
for a potential (V
bi
+V2kT/e) of 0.1 V (line).
After [Le21].
9.3
The Phenomenology of Macropore Formation in n-Type Silicon
A specific feature of macropore formation in n-type silicon is the possibility of
controlling the pore tip current by illumination and not by applied bias. This adds
another degree of freedom that is not available for mesopore or macropore forma-
tion on p-type substrates. The dark current density of moderately doped n-type Si
electrodes anodized at low bias is negligible, as shown in Fig. 4.11, therefore all
macropore structures discussed below are formed using illumination of the elec-
trode to generate the flux of holes needed for the dissolution process. Illumina-
tion, however, is not the only possible source of holes; for example, hole injection
from a p-doped region is expected to produce similar results.
If the sample is illuminated from the side that is exposed to the electrolyte,
hole generation in the depleted pore walls is inevitable. Under this frontside illu-
mination, a full passivation of the pore walls is not obtained and for short wave-
lengths (<700 nm), macropore growth even becomes suppressed. Under illumina-
tion with long wavelength light (7001100 nm) the resulting pores are conically
shaped, as shown in the center column of Fig. 9.7. An electrode surface covered
with such tapered pores appears pitch black to the eye. This is due to light trap-
ping in the porous layer, which reduces the reflectivity to values below 1% for the
visible regime. The minority carrier density profile generated by frontside illumi-
nation is shown in Fig. 10.4b.
9 Macroporous Silicon 190
Fig. 9.6 Cross-sectional optical micrographs
of a macropore formed by anodization of
(100) p-type Si electrodes in (a) 2 M HF in
acetonitrile (20 X cm, 3 mA cm
2
, 90 min, RT)
and in (b) 30% aqueous HF (20 X cm,
40 mA cm
2
, 7.5 min, RT). (c) A SEM micro-
graph of the macropores in (b) reveals that
most of the pore volume is filled with micro-
porous silicon.
Under backside illumination carrier generation in the pore walls can be
avoided, if the wavelength of the light source is significantly lower (900 nm) than
the wavelength corresponding to the bulk silicon bandgap (1100 nm). For the lat-
ter wavelength the absorption coefficient of bulk silicon becomes very low, as dis-
played in Fig. 7.6, and the light penetrates deep into the bulk, as shown in
Fig. 10.4a. A significant fraction may even reach the porous region and produce
an undesirable generation of minority carriers in the pore walls. Light sources re-
commended for macropore formation are discussed in Section 4.6. Backside illu-
mination, however, is only applicable for samples of moderate surface recombina-
tion velocity and a bulk diffusion length of about the sample thickness. Otherwise
the majority of photogenerated holes are lost by recombination before they reach
the pore tips to support dissolution. A low surface recombination velocity can be
realized by a diffused or implanted n
+
layer on the backside of the electrode. A
benefit of such a layer is that it serves well as a transparent and homogeneous oh-
mic contact. The minority carrier density profile generated by backside illumina-
tion is shown in Fig. 10.4c.
The way in which pore initiation occurs and how it can be controlled will be
considered next. If a polished, illuminated n-type Si electrode of (100) orientation
9.3 The Phenomenology of Macropore Formation in n-Type Silicon 191
Fig. 9.7 SEM micrographs of surface and
cross-section of an n-type Si electrode ano-
dized for the indicated times under white
light illumination of the front side (0.4 X cm,
(100), 2.5% HF, 10 mA cm
2
, 14 V). Micro-
porous silicon covering the macropores, as
shown in the top row, has been removed by
alkaline etching for better visibility of the
macropore initiation process in the center
and bottom rows. After [Le8].
is anodized in aqueous HF, the topmost few micrometers of the electrode are
etched homogeneously until tiny etch pits appear, which develop into stable pores,
as shown in Fig. 9.7. Only a certain fraction of the initial etch pits develop into
macropores by a kind of Ostwald ripening process, which can easily be observed
for a sample polished under a small angle to the electrode surface, as shown in
Fig. 9.4e. The decrease in pore density causes an increase in pore tip current for
the surviving pores, which leads to a bottleneck at the pore opening due to the de-
pendence of pore diameter on current, according to Eq. (9.3). After this initiation
and ripening period pore growth becomes stable and a pattern of randomly dis-
tributed pores penetrates the electrode without a further significant change in the
number of pores, as shown in Fig. 9.8. The average pore density of such random
structures increases with doping density, as shown in Fig. 6.10. The absolute val-
ues, however, are significantly lower than for macropore formation on p-type sub-
strates of the same doping density [Ri4].
The question arises as to what effects are responsible for the generation of the
very first tiny etch pits on the surface. However, this question is misleading in
some respects. A flat n-type surface anodized in HF is in an unstable condition.
Tiny inhomogeneities of flatness in the electrode surface, inevitably produced by
the dissolution process, are amplified, because the concave areas will focus the
electric field and thereby become more efficient in collection holes than flat areas.
An increase in anodization bias accelerates pore initiation at flat electrodes.
This understanding of pore initiation led to the idea of producing an ordered
rather than a random array of pores by using an artificial pattern of etch pits
formed prior to anodization. This approach proved to be successful [Le8]. Artificial
pits can, for example, be generated by standard lithography of an oxide film and
subsequent alkaline etching, which leads to the well-known inverted pyramids, as
shown in the inset of Fig. 9.9. Anodization of such a textured electrode leads to a
well ordered array of macropores, as shown in the same figure. The etch pit ge-
ometry is important. For example, incompletely etched pyramids may lead to for-
mation of four pores per pit, located at the corners of the flat pyramid bottom.
Note that the same principle of a pre-textured electrode has successfully been ap-
plied to the electrochemical formation of ordered pore arrays in alumina [As5].
If the pore density of arrays as well as random structures is plotted versus the
doping density, a linear dependency is observed over several orders of magnitude,
as shown in Fig. 6.10. If the pore density, pre-set by the pitch of a predefined pore
pattern, is mis-adjusted to the natural pore density determined by substrate dop-
ing density, to a certain extent the system will change its pore density. An in-
crease in the number of pores, as shown in Fig. 9.10a, has been called branching,
while a decrease, as shown in Fig. 9.10c, has been called dying of pores [Le17].
Note that branched pores are still macropores, because of an illumination-induced
photocurrent, while spiking pores, in contrast, are mesopores caused by break-
down of the SCR at the pore tip. Branching or dying of pores does not affect
properties like porosity or pore growth rate.
It has been proposed that depletion of the pore walls requires their thickness to
be 2W [Zh3, Se8]. However, this does not apply for n-type electrodes. Because the
9 Macroporous Silicon 192
porosity depends on the applied current density, according to Eq. (9.1), pore walls
can be much thinner than 2W. On the other hand, in low doped substrates wall
dimensions as large as 10W are observed, as shown in Fig. 9.11b. In the latter
case the pore wall region is not fully depleted. Holes diffuse into the wall region,
but their numbers are small, because the pore tips drain most of the hole current.
If the number of holes in the pore walls is not sufficient to initiate branching,
only a roughening of the pore walls is observed, as shown in Fig. 9.11b.
Because the SCR width depends not only on doping density but also on bias,
the average pore density is expected to decrease with the square root of bias. This,
however, is not observed. An increase in bias often leads to the formation of
breakdown-type mesopores at the macropore walls. Because these spiking pores
show diameters on the order of a few tens of nanometers they are hard to identify
even in an SEM. Spiking pores can be identified if they are enlarged by subse-
quent chemical etching, as shown in Fig. 8.10. Details of their branched morphol-
ogy become visible by formation of an oxide replica and after etchback of the sub-
9.3 The Phenomenology of Macropore Formation in n-Type Silicon 193
Fig. 9.8 SEM micrograph of the surface,
cross-section and a 458 bevel of an anodized
n-type silicon sample (5 X cm, (100), 6% HF)
showing a random pattern of macropores.
Pore initiation was enhanced by applying 10 V
bias in the first minute of anodization fol-
lowed by 149 min at 3 V. The current density
was kept constant (10 mA cm
2
) by adjusting
the backside illumination. After [Le9].
Fig. 9.9 SEM micrograph of an n-type silicon
electrode with an etched macropore array
(5 X cm, (100), 3 V, 350 min, 2.5% HF). Pore
growth was induced by a square pattern of
pits produced by standard lithography and
subsequent alkaline etching (inset upper
right). In order to measure the depth depen-
dence of the growth rate, the current density
was periodically kept at 5 mA cm
2
for 45 min
and then reduced to 3.3 mA cm
2
for 5 min.
This results in a periodic decrease in the pore
diameter, as indicated by the white labels on
the left-hand side. After [Le9].
9 Macroporous Silicon 194
Fig. 9.10 SEM micrographs showing the de-
pendence of macropore morphology on n-type
doping density (2.5% HF, 5 mA cm
2
, 2 V).
(a) High doping leads to branching, while
(c) low doping results in dying of pores.
(b) Strong branching is observed at the bor-
der to an unpatterned area producing under-
etching in an isotropic manner (white broken
line). After [Le17].
Fig. 9.11 Optical micrograph of macropore
arrays with (a) densely and (b) widely spaced
pores (n-type, 40 X cm, 4% HF, 143 min, 2 V).
The broken line indicates the interface between
bulk and SCR. The dots represent the hole
distribution. After [Le 21].
strate, as shown in Fig. 8.12b. Spiking leads to pore cross-sections as in
Fig. 9.12e. Because spikes are mesopores formed by breakdown, their formation
current is independent of photogenerated carriers, as discussed in Section 8.1.
This is the decisive difference from branching pores, as for example shown in
Figs. 9.10a and 9.14, which are still macropores etched by the photocurrent. A
consequence of severe spiking is an increase in the anodic current density with
etching time under constant illumination conditions. Note that the decrease in re-
maining bulk thickness during macropore growth also leads to an increase in cur-
rent density with etching time under constant illumination conditions according
to Eq. (10.10). Both cases can be easily distinguished if the illumination is
switched off, as spiking produces a considerable dark current. Spiking, which is
usually undesirable in macropore array fabrication, puts an upper limit to the
bias. This limit is as low as 2 V for high doping densities (> 10
16
cm
3
), while for
low doped substrates a bias in the order of 10 V is required to initiate spiking.
A lower limit of bias is given by the onset of unstable macropore formation.
This is shown in Fig. 9.13, which shows the pore morphology and the correspond-
ing formation conditions in the current density-voltage plot. The current density J
is held constant by the intensity of the backside illumination, so that the influ-
ence of the applied bias can be studied independently. At 0.4 V, versus a plati-
num wire as a pseudoreference electrode, the current density is constant over the
9.3 The Phenomenology of Macropore Formation in n-Type Silicon 195
Fig. 9.12 By an increase in bias or doping
density the round (a) or slightly faceted (b)
cross-section of macropores becomes star-
shaped by branching (c, d) or spiking (e)
along the h100i directions orthogonal to the
growth direction. (f ) The current density dis-
tribution at a pore tip. (g) SEM micrograph
of a macropore grown on (100) n-type Si.
The macropore walls are covered by micro
PS, indicated by a broken line at the interface
to bulk Si. (i, h) Sectional views of the pore
shown in (g) at positions h and i. (i) Note
that the pore tip is free of micro PS.
entire sample surface and only homogeneous formation of microporous silicon is
observed. At 0.2 V the current density at depressions in the electrode may reach
the value indicated by the open circle in the center diagram; macropores are ini-
tially formed but their depth is small compared to the thickness of the micropor-
ous layer. For a bias of 0.0 V the macropore formation is more pronounced but
not all pores have the same depth. Stable macropore formation is observed at a
bias of 2 V, as shown in Fig. 9.13d. In general stable macropore formation re-
quires an applied bias sufficient to generate the current density J
PS
(about 1 V).
The behavior discussed above shows that in the regime of stable macropore for-
mation the pore current divided by the cross-sectional area of the pore is equal to
J
PS
. Because of the hemispherical shape of the pore, only the current density at
the center of the tip is equal to J
PS
, while the interfacial current density decreases
towards the pore wall with a cosine dependency, as shown in Fig. 9.12f [Zh3]. As
a result the pore walls become covered with micro PS film, while the center of
9 Macroporous Silicon 196
Fig. 9.13 SEM micrographs of pre-patterned
n-type electrodes anodized at different bias
(a: 0.4 V, b: 0.2 V, c: 0 V, d: 2 V, versus
Pt-reference, 5% HF, 12.5 mA cm
2
, 5 X cm,
(100)). The four micrographs correspond to
positions in the IV curve as indicated by bro-
ken arrows. The current density as adjusted
by illumination is indicated by full circles;
open circles indicate the pore tip current den-
sity. (d) Stable macropore formation requires
a bias sufficient to produce J
PS
at the pore
tip. After [Le9].
the pore tip is close to electropolishing and therefore free of micro PS for a suffi-
cient bias of 2 V, as shown in Fig. 9.12 gi. A more quantitative proof of the condi-
tion J
tip
=J
PS
is the measurement of the macropore growth rate, which is found to
depend only on HF concentration and temperature [Le9], but not on applied cur-
rent density, as shown by solid squares in Fig. 6.5. Note that the increase in pore
diameter induced by the death of a neighboring pore, as shown in Fig. 9.10c, or
the formation of bottlenecks at the pore openings during pore initiation without a
predetermined pattern [La2] are direct consequences of the condition J
tip
=J
PS
. If
the number of pores decreases, the tip current of dying pores is added to the tip
currents of the neighboring pores, thus increasing their diameter. An increase of
J
tip
to values in excess of J
PS
would cause carrier accumulation at the tip as dis-
cussed in Section 3.2, thus effectively destroying the depletion in the pore walls.
In addition, J
tip
>J
PS
would be accompanied by an increase in the dissolution va-
lence from 2 to 4, which would imply a reduction in the tip growth rate. The lat-
ter effect could explain the remarkable flatness of the plane that is formed by all
pore tips of a macropore array [Le17]. The consequences of the condition J
tip
=J
PS
for a quantitative description of macropore formation are discussed in detail in
the next section.
The extraordinary straightness of macropores is not only a consequence of the
lithographic pattern but also of the crystal nature of the substrate. The direction
of macropore growth depends on the crystal orientation of the substrate and on
the source of holes that support the dissolution. As observed for the case of meso-
pores in low doped n-type electrodes, macropores also have a tendency to grow
along the h100i directions of the crystal, as illustrated in Fig. 8.13 [Ha10, Ch11,
Le8]. This orientation dependence can be ascribed to the anisotropy of the pore
tip current density J
PS
. J
PS
is found to be maximum on (100) substrates, as shown
in Fig. 4.9. Note that J
tip
becomes smaller than J
PS
for an applied bias below
about 1 V, and so the orientation dependence as well as the homogeneity of
macropore growth is reduced at low bias. However, in contrast to the case of the
mesopores where the hole generation occurs directly by the high field present at
the pore tip, the source of holes, for example the illuminated backside of the sam-
ple, is spatially separated from the drain of holes, the pore tips, for the case of
macropore formation. Therefore the direction of pore growth is under mixed
control. If a h100i direction of the crystal is pointing towards the source of holes,
which is the case for (100) substrates illuminated homogeneously from the back-
side, the formation of perfectly straight pores is observed, as shown in Fig. 9.9.
However, even for a considerable misalignment of the substrate, the pores still
grow along h100i, but the tendency to branching is enhanced, as shown in
Fig. 9.14a. For (110) oriented substrates pores form along both h100i directions,
but branching along the h113i directions is also observed. For (111) oriented sub-
strates the h113i directions of the crystal become the preferred pore growth direc-
tions, producing a three-fold symmetry. However, such pores are strongly
branched, also, as shown in Fig. 9.14b [Ro8, Ch17].
The geometry of the pore cross-section in a plane parallel to the electrode sur-
face is also dominated by the crystal orientation. Usually shapes between a circle
9.3 The Phenomenology of Macropore Formation in n-Type Silicon 197
and a square are observed for (100) substrates as shown in Fig. 9.12ac. An in-
crease in formation bias leads to spiking. All spiking occurs along h100i direc-
tions and so the pore shape becomes distorted to a four-pointed star, as shown in
Fig. 9.12e.
The geometry of the pore tips is pyramidal, with facets formed by (111) planes,
for J
tip
< J
PS
. This is the case when the current density is limited by the applied
bias, as is the case for the samples shown in Fig. 9.13 b and c. If the bias is in-
creased, dissolution at the pore tip occurs partially in the isotropic electropolish-
ing regime (J
tip
=J
PS
). This reduces the tendency to form facets and the tip geome-
try becomes almost hemispherical, as shown in Fig. 9.13d.
9.4
Calculating Macropore Growth and Mass Transport
In contrast to the micro- and mesoporous regimes, for which only a few empirical
laws for the growth rate and porosity are available, the detailed pore geometry for
macropore arrays in n-type silicon can be pre-calculated by a set of equations.
This is possible because every pore tip is in a steady-state condition characterized
by J
tip
=J
PS
[Le9]. This condition enables us to draw conclusions about the porosi-
ty of a macroporous film. The etching current, controlled by illumination divided
by the initial sample surface area as defined for example by the O-ring of the cell,
gives the applied current density J. If temperature and HF concentration present
at the pore tip are known, J
PS
can be calculated according to Eq. (4.9). The porosi-
ty p is then simply given by the ratio of the applied current density J and J
PS
:
p J=J
PS
9:1
This equation holds true even for thick porous layers, if the decrease in the HF
concentration in the pores, which leads to a decrease in J
PS
at the pore tip, is
taken into account.
9 Macroporous Silicon 198
Fig. 9.14 While the preferred macropore growth direction on a (322) substrate is still h100i, it
becomes h113i for (111) substrates (4% HF, 5.6 mA cm
2
, 4 X cm n-type Si). After [Ro8].
Note that Eq. (9.1) applies to pore arrays as well as to randomly distributed
pores. For simple orthogonal or hexagonal arrays of macropores with one pore
per unit cell of the pattern, the porosity can be defined locally as the ratio be-
tween the cross-sectional area of the pore A
P
and the area of the unit cell A
U
, as
shown in Fig. 9.15a:
p A
P
=A
U
9:2
The cross-section of a macropore may have all shapes between a circle and a
four-pointed star, as shown in Fig. 9.12ae. In addition the pore walls are covered
with a microporous silicon layer, as shown in Fig. 9.12h, which makes the deter-
mination of A
P
difficult. In most cases, however, the approximation of the pore
cross-section by a square of size d is found to be sufficient. Under this assump-
tion and for a square pattern of pitch i, as shown in Fig. 9.15a, d becomes simply:
d iJ=J
PS
1=2
9:3
For arrays of complex geometry where the unit cell contains several pores, or
for random pore distributions, the cross-sectional area of each pore A
P
is propor-
tional to the area from which it collects charge carriers, indicated by broken lines
in Fig. 9.15b and c. Equation (9.3) offers an elegant way of tuning the pore diame-
ters during pore growth by a corresponding variation of J. Periodic changes of J,
for example, produce corresponding variations of the pore diameter, as illustrated
by Fig. 9.16. J is usually adjusted by the backside illumination intensity. Interest-
ingly, a modulation of pore diameter with anodization current has been observed
for pore formation in aluminum, too, and has likewise been interpreted in terms
of a pore growth velocity independent of applied current [He8].
For a square pattern and square pores the pore wall thickness w is given by:
w i1 J=J
PS
1=2
9:4
9.4 Calculating Macropore Growth and Mass Transport 199
Fig. 9.15 Cross-sections of macropore arrays
parallel to the electrode surface (a) for a
square, (b) for an ordered and (c) for a ran-
dom pattern. The pores (black squares)
collect holes from the area indicated by the
broken lines. The porosity (ratio of black area
to total area) is 0.25 for all patterns.
The steady-state condition (J
tip
=J
PS
) at the pore tip determines not only the
pore diameter but also the pore growth rate. The rate r
p
of macropore growth can
be calculated if the local current density at the pore tip is divided by the dissolu-
tion valence n
v
(number of charge carriers per dissolved silicon atom), the elemen-
tary charge e (1.60210
19
C) and the atomic density of silicon N
Si
(510
22
cm
3
):
r
p
J
PS
=n
v
eN
Si
9:5
J
PS
and n
v
are the only variables in this equation. The dissolution valence n
v
shows a steep increase from 2 to 4 at J
PS
, as shown in Fig. 4.5. Because J
PS
is a
function of temperature T and electrolyte concentration c only, the macropore
growth rate on n-type substrates is dependent on c, T and n
v
. Gravimetric mea-
surements of macroporous samples show that 2.2<n
v
<3 is valid in most cases
[Le9]. This is shown in Fig. 9.17; the experimentally observed macropore growth
rates that are plotted as a function of the HF concentration fit well with growth
rates calculated for n
v
=2 and n
v
=4 using Eq. (9.5). Assuming an average dissolu-
tion valence of n
v
=2.6, the macropore growth rate r
p
has been calculated accord-
ing to Eqs. (9.5) and (4.9); it is given as a function of HF concentration in the ta-
ble on the inner back cover of this book.
It should be emphasized that n
v
and J
PS
, and therefore c and T, refer to the con-
dition at the pore tip. The dissolution valence and the temperature can be as-
sumed to be independent of pore depth. This is not the case for the HF concen-
tration c. Because convection is negligible in macropores, the mass transport in
the pore occurs only by diffusion. A linear decrease in HF concentration with
depth and a parabolic growth law for the pores according to Ficks first law is
therefore expected, as shown in Fig. 9.18a. The concentration at the pore tip can
be calculated from the concentration in the bulk of the electrolyte c, the pore
length l, the diffusion coefficient D
HF
(Section 1.4) and the flow of HF molecules
F
HF
, which is proportional to the current density at the pore tip:
9 Macroporous Silicon 200
Fig. 9.16 A sine wave modulation of the etch-
ing current produces an array of macropores
with a corresponding modulation of diameter.
After [Le17].
c
tip
c F
HF
l=D
HF
9:6
The experimentally observed parabolic increase in pore depth and linear de-
crease in concentration shown in Fig. 9.18c indicate that Eq. (9.6) is valid [Le9].
The macropore growth rate decreases linearly with l according to Eq. (9.6). If,
therefore, a constant pore diameter is desired for a macropore array, a decrease in
etching current or illumination intensity, respectively, with time is required.
For deep and narrow pores, however, the slow etching of the pore walls, which
produces a small dark current density in the order of 110 lA cm
2
, as shown in
Fig. 4.11, cannot be neglected. Unfortunately the dissolution at the pore walls not
only depends on the total wall area but also on parameters such as temperature,
DOC and amount of surfactants added to the electrolyte. It is therefore difficult to
quantify this effect. The consumption of HF in the dissolution process at the pore
9.4 Calculating Macropore Growth and Mass Transport 201
Fig. 9.17 Calculated growth rates of macro-
pores for n
v
=2 and n
v
=4 (lines) and experi-
mentally obtained initial macropore etch rates
r
p
(squares) versus the HF concentration
c
HF
of the electrolyte ( J/J
PS
=0.25, 10
15
cm
3
n-type, RT, square pattern). After [Le9].
Fig. 9.18 (a) If the pore tip is the only reac-
tion site, a linear HF concentration gradient
and a parabolic pore growth is observed. (b)
If HF is consumed at the pore walls, also, the
concentration gradient is nonlinear and pore
growth becomes retarded. (c) Using depth
markers, as shown in Fig. 9.9, the depth of
macropores l is plotted versus the anodiza-
tion time t (squares). The decrease in HF
concentration c
HF
(circles) in the pore is cal-
culated from the observed l using Eqs. (9.5)
and (4.9). After [Le9].
walls leads to a non-linear decrease in the HF concentration along the pore, as
shown in Fig. 9.18b. As a result, a non-linear term is superimposed on the linear
decrease in growth rate with depth.
Note that during macropore formation in p-type silicon electrodes the pore tip
current density is usually well below J
PS
, and so Eqs. (9.1) to (9.5) are not applica-
ble to p-type macropore formation [Le21].
9.5
Design Rules and Limits of Macropore Array Fabrication
Not all desirable macropore array geometries can be achieved by the electrochemi-
cal etching process. This section gives the upper and lower limits for pore dimen-
sions and a few design rules. Macropores in n-type substrates will be discussed
first [Le17].
The realization of a desired pore pattern requires a certain doping density of
the n-type Si electrode. A good rule of thumb for the selection of an appropriate
substrate is to multiply the desired pore density given (in lm
2
) by 10
16
and take
this number as doping density (in cm
3
). This dependency is shown in Fig. 6.10.
A square pattern of 10 lm pitch, for example, produces a pore density of
0.01 pores lm
2
, which can best be etched using a n-type substrate doping density
of 10
14
cm
3
.
The number of possible arrangements for the pore pattern is only limited by
the requirement that under homogeneous backside illumination the porosity has
to be constant (Eq. 9.1) on a length scale above about three times the pitch. This
means it is possible to etch a pattern with a missing pore, a missing row of pores
or even two missing rows. Patterns as shown in Fig. 9.15b can also be etched; a
single pore, however, cannot be etched. It is also possible to enlarge or shrink the
pitch of a pattern across the sample surface by a maximum factor of about three.
But a pattern with an abrupt border to an unpatterned area will lead to severe un-
der-etching, according to Fig. 9.10 b, and random pore formation in the unpat-
terned area.
A local variation in porosity can be produced by an inhomogeneous illumina-
tion intensity. However, any image projected on the backside of the wafer gener-
ates a smoothed-out current density distribution on the frontside, because of ran-
dom diffusion of the charge carriers in the bulk. This problem can be reduced if
thin wafers or illumination from the frontside is used. However, sharp lateral
changes in porosity cannot be achieved.
Arrays with pore diameters d as small as about 0.3 lm have been achieved
[Le17]. The lower limit for the pore diameter is established by breakdown, accord-
ing to Fig. 8.1b, which leads to light-independent pore growth and spiking. There
seems to be no upper limit for the pore diameter, because the formation of
100 lm wide pores has been shown to be feasible [Kl3]. Array porosities may
range from 0.01 to close to 1. The porosity, which is controlled by the etching cur-
rent, determines the ratio between pore diameter and pitch of the pore pattern.
9 Macroporous Silicon 202
This means that for a square pattern, any pore diameter between a tenth of the
pitch and nearly the pitch can be realized.
The pore diameter can be varied over the length of the pore by a factor of up to
about 3 for all pores simultaneously by adjusting the current density, as shown in
Fig. 9.16. This means that the porosity normal to the surface can be varied. The
taper of such pore geometries is limited by dying of pores to values below about
308 for a pore diameter decreasing in growth direction, while values in the order
of 458 have been realized for an increase in pore diameter in the growth direction
[Mu6]. Note that narrow bottlenecks will significantly reduce the diffusion in the
pore and the formation of deep modulated pores becomes more difficult than for-
mation of straight pores. Bottlenecks at the pore entrance may result from the
transition of the pyramidal etch pit into a pore tip. They can be avoided by an in-
crease in the current density during the first minutes of pore array fabrication.
The pore cross-section under stable array formation conditions is usually a
rounded square, as shown in Fig. 9.12b and h. Following electrochemical pore for-
mation, the cross-section can be made round by oxidation steps or can be made
square by chemical etching at RT in aqueous HF or weak alkaline solutions like di-
luted KOH or NH
4
OH. Formation of side pores by branching or spiking, as shown
in Figs. 9.10a and 8.12, can be suppressed by an increase in current density or a
decrease in doping density, bias or HF concentration. The dying of pores, as shown
in Fig. 9.10c, is suppressed by an increase in current density, doping density or bias.
The pore length l can be as large as the wafer thickness (up to 1 mm). However,
the growth of deep pores requires low electrolyte concentrations, low temperatures
and etching times in the order of a day or more, because the etch rate in deep pores
is limited by HF diffusion to values in the order of 0.5 lm min
1
and below. Shorter
pores (l <0.1mm) can be etched much faster (5 lm min
1
). Under stable etching con-
ditions all pores have the same length. Pore arrays with through-pores can be rea-
lized by an increase in the etching current density into the electropolishing re-
gime, which separates a free-standing porous plate from the substrate. Macropores
penetrating the whole wafer thickness can also be etched, but pore formation be-
comes unstable in the vicinity of the backside. The formation of dead-end pores
and subsequent oxidation and alkaline etchback is technologically favorable.
Another effect that limits the obtainable pore length is shown in Fig. 9.19a.
This effect is characterized by a sudden drop in the growth rate at the pore tip to
negligible values and an increase in pore diameter close to the tip. This degrada-
tion of pore growth establishes an upper limit for the pore length for a given set
of anodization parameters. The fact that pore degradation is delayed for a reduced
formation current, which produces conical pores as shown in Fig. 9.19b, indicates
a diffusion-related phenomenon. The observed dependence of degradation on the
concentration of the dissolution product H
2
SiF
6
in the electrolyte points to a poi-
soning of the dissolution reaction. The maximum obtainable pore depth decreases
rapidly with increasing HF concentration, as shown in Fig. 9.20. This effect can
be ascribed to the rate of H
2
SiF
6
production being proportional to J
PS
, which
again depends exponentially on HF concentration (Eq. 4.9), while the diffusion of
H
2
SiF
6
is expected to show little dependence on HF concentration.
9.5 Design Rules and Limits of Macropore Array Fabrication 203
According to Eq. (4.4), hydrogen is another dissolution product. This raises the
question of whether hydrogen bubbles are present inside the pore. Direct experi-
mental observations relating to this question are not available. However, the
growth of perfect macropore arrays seems unlikely under the assumption of ran-
domly distributed microbubbles in the pores, which would hinder or even block
the continuous mass transfer. It can therefore be assumed that hydrogen is super-
saturated in the electrolyte inside the pores and that bubbles are formed at the
pore entrance. This assumption is supported by the observation that a change in
ambient pressure between 0.3 and 100 bar shows little or no effect on electro-
chemical pore formation in aluminum electrodes, which is also accompanied by
hydrogen evolution [Be24].
9 Macroporous Silicon 204
Fig. 9.19 Optical micrographs of (a)
cylindrical pores showing a degradation
effect at the pore tips, ascribed to a lim-
itation of mass transport (15% HF, 3 V,
1200 s). (b) Conical pores do not show
this degradation at the same depth
(15% HF, 3 V, 720 s). After [Le17].
Fig. 9.20 The maximum pore length for
cylindrical pores that can be achieved with-
out a degeneration of the pore tips is
shown as a function of the electrolyte con-
centration at RT (circles). The shaded area
indicates the degeneration depth for a
decrease or increase in pore diameter ver-
sus pore depth. After [Le17].
The pore growth direction is along the h100i direction and toward the source of
holes. For the growth of perfect macropores perpendicular to the electrode surface
(100), oriented Si substrates are required. Tilted pore arrays can be etched on sub-
strates with a certain misorientation to the (100) plane. Misorientation, however,
enhances the tendency to branching and angles of about 208 appear to be an
upper limit for unbranched pores. For more details see Section 9.3.
Pore array formation is also feasible for p-type substrates [Le21]. Under optimized
conditions a pore depth of up to 400 lm, at a growth rate of 0.60.8 lm min
1
has
been reported [Ch16]. A detailed investigation of the limits of pore geometry and
design rules are not available for p-type electrodes. The condition that the pore wall
thickness is always less than 2W, however, reduces the flexibility in array design sig-
nificantly. The fact that the pore growth rate depends on current density, on the
other hand, could be applied to form pores of different depth by lateral variations
of current density. The latter is not possible for stable macropore formation on n-
type substrates.
In conclusion it can be said that the limits of macropore array formation are in
some way complementary to the limitations of plasma etching [ Ja3]. The latter
technique gives a higher degree of freedom in lateral design, while the freedom
in vertical design and the feasible pore aspect ratios are limited.
An interesting question is whether such well-ordered pore arrays can also be
produced in other semiconductors than Si by the same electrochemical etching
process. Conversion of the macropore formation process active for n-type silicon
electrodes on other semiconductors is unlikely, because their minority carrier dif-
fusion length is usually not large enough to enable holes to diffuse from the illu-
minated backside to the front. The macropore formation process active in p-type
silicon or the mesopore formation mechanisms, however, involve no minority car-
rier diffusion and it therefore seems likely that these mechanisms also apply to
other semiconductor electrodes.
9.5 Design Rules and Limits of Macropore Array Fabrication 205
10.1
Overview
A complete review of the multitude of applications based on electrochemical prop-
erties of silicon is beyond the scope of this chapter and so selected examples re-
presenting the different categories of applications are given. Two fundamental cat-
egories of applications can be identified: in the first, the properties of the silicon-
electrolyte contact are exploited for analytical purposes [Fo2]. The HF tester and
the pin hole detector directly exploit special properties of the electrochemical reac-
tions at anodically or cathodically polarized silicon electrodes, while for diffusion
length mapping and carrier concentration profiling the electrolyte serves mainly
as a transparent, homogeneous and easy-to-use contact with the substrate. The
second category exploits the properties of electrochemically formed materials to
create new devices. An anodic oxide is an example of such an electrochemically
formed material. Because of its electrical performance, which is inferior to a ther-
mal oxide, however, device-related applications of anodic SiO
2
are limited to cases
where the thermal budget is crucial, e.g. nuclear radiation detectors [Me13]. This
is in contrast to the different kinds of porous silicon, for which anodization in HF
or stain etching are the only known fabrication techniques. PS-related applications
can be found in various fields. Electronic applications include cold cathodes and
capacitors, while optical applications include anti-reflective coatings, LEDs, short-
pass filters, interference filters, waveguides and photonic crystals. The high inter-
nal surface area of PS can be exploited for chemical and biological applications
such as biocompatible interfaces or biochips used in genomics, as well as for a
multitude of different sensing devices. The use of PS as a sacrificial layer opens
new doors in micromechanical engineering. A selection of the wide variety of PS-
based applications is given below.
207
10
Applications
Electrochemistry of Silicon: Instrumentation, Science, Materials and Applications.
Volker Lehmann
Copyright 2002 Wiley-VCH Verlag GmbH
ISBNs: 3-527-29321-3 (Hardcover); 3-527-60027-2 (Electronic)
10.2
AC Properties of Silicon Electrodes and Carrier Concentration Profiling
This section summarizes the AC behavior of a silicon electrode and introduces an
application of these properties, the carrier concentration profiling technique.
Several factors contribute to the impedance behavior of a silicon electrode. Fig-
ure 10.1a shows a proposal for an equivalent circuit that considers most of these
factors, which are:
1. A resistance R
U
due to the sum of the ohmic contact resistance, the silicon
bulk resistance and in part to the electrolyte resistance.
2. The capacitance C
SC
and the resistance R
SC
due to the SCR in the Si electrode.
3. The capacitance C
SS
and the resistance R
SS
connected parallel to C
SC
, which
are introduced to describe the effect of surface states.
4. An additional RC element C
S
and R
S
, to describe interfacial layers, such as an
anodic oxide.
5. The capacitance C
H
and the charge transfer resistance R
CT
representing the
Helmholtz layer.
At high electrolyte concentrations (>1 mol l
1
) the Helmholtz capacitance C
H
of
the electrode shows values in the order of C
H
/A=0.5 lF cm
2
[Na7] and is thereby
at least one order of magnitude larger than C
SC
/A. Therefore it can be neglected
10 Applications 208
Fig. 10.1 Equivalent circuits used to repre-
sent the semiconductor-electrolyte interface.
(a) A more complete approach taking into
account the series resistance (R
U
), the deple-
tion layer (C
SC
, R
SC
), an oxide surface film
(C
S
, R
S
), the Helmholtz layer (C
H
) with a
transfer resistance (R
CT
), and surface states
(C
SS
, R
SS
). Simplified equivalent circuits are
shown in (b) and (c). After [Ot1].
for AC measurements under depletion conditions. If the electrode is free of sur-
face films and shows a low surface state density it is sufficient to use the simpli-
fied equivalent circuit shown in Fig. 10.1b. In case leakage currents are negligi-
ble, the circuit can be further reduced to two elements, as shown in Fig. 10.1c.
The potential dependence of the SCR capacity C
SC
on applied potential V is de-
scribed by the Mott-Schottky relation:
1=C
2
SC
= 2(V V
fb
)=(eN
d
eA
2
) (10:1)
where A is the interfacial area, e the elementary charge, N
d
the doping density,
V
fb
the flat-band potential and e the dielectric constant of the semiconductor.
From Eq. (10.1) the flat band potential for an electrolytic contact can be calculated
from measurements of C
SC
(V) if N
d
is known; or V
fb
can be graphically deter-
mined by extrapolation of a C
SC
2
versus V plot to zero capacitance, as shown in
Fig. 10.2. The value of 0.75 V found for p-type Si, however, is unrealistically high.
By HFR the flat band voltage of a p-type and an n-type Si electrode in 0.5 M HF +
1 M NH
4
Cl are determined to be 0.14 V (SCE) and 0.54 V (SCE), respectively
[Ot1]. A similar value of 0.35 V (SCE) is observed for n-type Si in 1 M HF by mi-
crowave reflectivity measurements [Na7]. Figure 10.3 summarizes values of V
fb
ob-
tained by different methods. Note that the scatter in these data is much larger for
p-type silicon electrodes than for n-type. A similar scatter has been observed in
the determination of the OPC potential of p-type electrodes, which is found to be
more sensitive to parameters such as, for example, illumination intensity than
that of n-type electrodes, as discussed in Section 3.2.
Equation (10.1) can be used to determine the doping density of a silicon sub-
strate and its depth profile, even if the flat band potential is not known accurately.
Diffusion doping, ion implantation or the growth of an epitaxial layer are com-
mon methods of producing doped regions in semiconductor substrates. The do-
pant concentration close to the surface can be measured by SRP or capacitance
10.2 AC Properties of Silicon Electrodes and Carrier Concentration Profiling 209
Fig. 10.2 Mott-Schottky plots of
an n-type and a p-type silicon
electrode in an electrolyte com-
posed of 0.5 mol l
1
HF and
0.5 mol l
1
NH
4
Cl. Dotted lines
correspond to N
D
=1.110
15
cm
3
and N
A
=3.410
15
cm
3
.
After [Ot1].
voltage (CV) techniques. In order to determine the carrier concentration depth
profile, we need a method of removing material and continuously probing its do-
pant concentration. SIMS is now the standard analytical tool to investigate doping
profiles. However, in some cases, e.g. very deep or ultra-shallow dopant profiles,
electrochemical methods can be beneficial.
By varying the applied bias the CV technique itself gives a certain depth resolu-
tion, but the investigated depth is limited by reverse bias breakdown. It is there-
fore advantageous to use the electrolyte not only for measurement but also for
material removal. Silicon can be etched by anodic oxidation in HF-free electrolytes
and subsequent etching in HF or directly by electropolishing in HF. The prior
method is very accurate but slow, and most suited for shallow implants. Forma-
tion and removal rates of anodic oxides and silicon are given in Sections 5.3 and
5.6, respectively. Determination of dopant depth profiles by anodic sectioning of
silicon and other semiconductors is an analytical tool that has been applied for
many years in semiconductor manufacturing [Ta9, Qu1, Am2, Bl1, Sh2, Ba9,
Se11]. The dopant concentration N(W) at the edge of the SCR of width W to the
bulk can be calculated from the change in capacitance dC
SC
with bias dV, accord-
ing to the Mott-Schottky relation:
N(W) = (C
3
SC
=eeA
2
)=(dC
SC
=dV) (10:2)
W = eA=C
SC
(10:3)
The doping densities calculated from the slope of the C
2
versus V plots, as
shown for example in Fig. 10.2, agree well with values measured by other meth-
ods [Ot1]. Deviation between results obtained by spreading resistance measure-
ments and electrochemical CV measurements are usually found to be below 20%
for doping densities between 10
12
and 10
18
cm
3
[Pe3].
The carrier concentration depth profile can be monitored by repeated CV mea-
surement and controlled electrochemical removal of the substrate. An optimum
10 Applications 210
Fig. 10.3 Graphical representation of flat
band potentials of silicon electrodes as deter-
mined by different methods. The solid lines
show a linear regression, the dotted lines
represent the mean value and the broken
lines represent the standard deviation from
mean. After [Ot1].
for smooth and homogeneous removal of the substrate has been reported for an
electrolyte composed of ammonium bifluoride (0.1 M NH
4
F HF) with a few
drops of a non-ionic wetting agent and dissolution in the tetravalent regime
(n=3.7 0.1) at about 3 mA cm
2
[Pe4]. A computerized electrochemical set-up for
carrier concentration profiling has been commercialized [20]. A drawback of the
electrochemical CV technique is its limited spatial resolution, usually given by the
size of the O-ring, which cannot be reduced significantly below 1 mm
2
. This value
is comparable to the area needed for a spreading resistance measurement. The
doping density has to be homogeneous over this electrode area. This limitation
can be overcome by the use of photoresist patterning, which provides a lateral res-
olution of down to 0.01 mm
2
[Se11].
An extension of the CV technique to a method comparable to deep level transi-
ent spectroscopy (DLTS) is possible if the electrode is illuminated during the capa-
citance measurement. Plots of capacitance versus wavelength, for a sweep from
low to high photon energies, yield a series of plateaus or peaks, each correspond-
ing to the population or depopulation of a given deep level in the bandgap. The
onset energy of the capacitance change yields the energy of the state, relative to
the appropriate band edge. This technique is known as electrochemical photocapa-
citance spectroscopy (EPS) [Ha5].
10.3
Diffusion Length and Defect Mapping
Contamination of silicon wafers by heavy metals is a major cause of low yields in
the manufacture of electronic devices. Concentrations in the order of 10
11
cm
3
[Ha2] are sufficient to affect the device performance, because impurity atoms con-
stitute recombination centers for minority carriers and thereby reduce their life-
time [Sc17]. In addition, precipitates caused by contaminants may affect gate ox-
ide quality. Note that a contamination of 10
11
cm
3
corresponds to a pinhead of
iron (1 mm
3
) dissolved in a swimming pool of silicon (850 m
3
). Such minute con-
tamination levels are far below the detection limit of the standard analytical tech-
niques used in chemistry. The best way to detect such traces of contaminants is
to measure the induced change in electronic properties itself, such as the oxide
defect density or the minority carrier lifetime, respectively diffusion length.
10.3.1
Diffusion Length Mapping
The basic idea of most diffusion length measurement techniques is to generate a
certain number of minority carriers inside the bulk Si, for example by illumina-
tion, and to measure the fraction of these carriers that diffuse to a collecting inter-
face. This fraction can be determined capacitively [Bo6], as well as by measure-
ments of the steady-state photocurrent [Dr2, Le11]. The parameter obtained by
these measurements is the minority carrier diffusion length L
D
of electrons in
10.3 Diffusion Length and Defect Mapping 211
p-type Si or of holes in n-type Si, respectively. L
D
can be converted into the life-
time s of minority carriers using
s = L
2
D
=D
e;h
(10:4)
where D
e,h
is the diffusion coefficient for the relevant minority charge carrier in
silicon. The diffusion coefficient itself depends on the mobility l of the relevant
species according to
D
e;h
= (kT=e)l
e;h
(10:5)
The frontside of the sample will now be illuminated. Let us first assume that
the collecting SCR is located at the frontside, too, as shown in the inset of
Fig. 10.4b. This mode of operation will be termed frontside photocurrent mode
(FPC). The collected photocurrent I
FPC
is a result of minority charge carriers gen-
erated in the SCR producing the current I
SCR
and in the bulk of the electrode pro-
ducing I
DIFF
.
I
FPC
= I
SCR
I
DIFF
(10:6)
The number of charge carriers generated in the SCR depends on the absorbed
flux of incident photons per unit area P, the width W of the SCR and the wave-
length-dependent absorption coefficient a of bulk Si. The latter parameter is
shown in Fig. 7.6, while the resulting penetration depth for light of different
wavelengths is shown in Fig. 10.4a.
I
SCR
= eP[1 exp(aW)[ (10:7)
If the sample thickness is much larger than L
D
, the backside condition becomes
irrelevant and I
DIFF
is simply given by:
I
DIFF
= eP aL
D
exp(aW)=(aL
D
1) (10:8)
If the photon energy approaches the bandgap energy of Si (1.1 eV), I
SCR
be-
comes small and I
FPC
is dominated by I
DIFF
. In contrast, for illumination with
light of a short wavelength, the absorption coefficient a of Si is large and minority
carrier generation occurs close to the illuminated surface, in a region smaller
than the width of the SCR W. If we assume the latter case, all absorbed photons
contribute to the collected current, I
FPC
=I
SCR
=eP.
Let us now assume that the collecting SCR is located at the backside, as shown
in the inset of Fig. 10.4c. This mode of operation will be termed backside photo-
current mode (BPC). The collected photocurrent I
BPC
again depends on genera-
tion in the bulk and in the SCR:
I
BPC
= I
SCR
I
DIFF
(10:9)
10 Applications 212
For not too low doped samples (D W), however, the contribution of I
SCR
is
usually negligible. If the surface recombination velocity at the illuminated front is
low, I
BPC
then only depends on sample thickness D, illumination intensity eP,
and minority charge carrier diffusion length L
D
.
I
BPC
= eP=cosh[(D W)=L
D
[ (10:10)
10.3 Diffusion Length and Defect Mapping 213
Fig. 10.4 (a) Computed minority carrier gen-
eration rate in bulk silicon for different wave-
lengths of monochromatic illumination of an
intensity corresponding to a photocurrent
density of 10 mA cm
2
. (b) Bulk minority car-
rier density for carrier collection at the illumi-
nated frontside (k=950 nm, eP=10 mA cm
2
).
(c) Minority carrier density for carrier collection
at the backside (k=950 nm, eP=10 mA cm
2
,
DW=500 lm). (d) Computed values of L
D
as
a function of the ratio I
BPC
/eP and the sample
thickness (DW) according to Eq. (10.10).
Equation (10.10) is the solution of the one-dimensional continuity equation un-
der several assumptions, such as a lifetime independent of the carrier concentra-
tion, a negligible electric field inside the wafer, and no coupling between injected
electrons and holes. However, despite these crude approximations, the results for
L
D
correlate well with values obtained by other methods such as surface voltage
spectroscopy (SPV). The sensitivity of the method for iron, for example, has been
found to be better than 10
11
cm
3
[Be17, Wa9]. Plots for different values of L
D
over the ratio I
BPC
/eP and the thickness DW according to Eq. (10.10) are shown
in Fig. 10.4d. For large values of diffusion length, the measurement range in BPC
mode is limited to about four times the wafer thickness, while for small diffusion
length the limit is about one-fifth of the wafer thickness. The L
D
of electronic
grade silicon is usually well above the lower limit, but for solar silicon materials,
the diffusion length may be too low for the BPC mode. In the latter case, L
D
can
still be obtained if the FPC value measured with short wavelength illumination is
compared to the FPC value for long wavelength illumination, using Eqs. (10.7)
and (10.8).
Double-sided electrolytic contacts are favorable for this method of diffusion
length measurement because they are transparent and the required SCRs are easi-
ly induced by application of a reverse bias. Therefore homogeneously doped wa-
fers need no additional preparation, such as evaporation of metal contacts or dif-
fusion doping, to produce a pn junction. Furthermore, a record low value of sur-
face recombination velocity has been measured for silicon surfaces in contact with
an HF electrolyte at OCP [Ya1]. Note that this OCP value cannot be further de-
creased by a forward bias at the frontside, because any potential other than OCP
has been found to increase the surface recombination velocity, as shown in
Fig. 3.2. Note that contaminations in the HF electrolyte, such as Cu, may signifi-
cantly increase the surface recombination velocity. This effect has been used to de-
tect trace levels (20 ppt) of Cu in HF [Re5].
Double cells that allow the establishment of an SiHF contact to both interfaces
of the wafer are shown in Fig. 1.7g and h and Fig. 1.10. A photograph of an ELY-
MAT double cell designed for diffusion length measurement of 200 mm wafers is
shown on the upper left of the front cover of this book. Tungsten carbide tips
pressed to the edge of the wafer are found to produce a sufficiently ohmic contact
for the majority carrier current. The applied reverse bias at the collector side
should be low (13 V) in order to reduce contributions to the reverse current be-
cause of metal precipitates or other crystal defects present in the SCR, as shown
in Fig. 10.6c and d. The illumination intensity eP can either be measured by an
independent method or, for short wavelength and large values of L
D
, it can be re-
placed by I
FPC
in Eq. (10.10). Thus, L
D
can be determined by measurements of
I
BPC
and I
FPC
, if the sample thickness and doping are known.
The method described so far measures diffusion length but provides no infor-
mation about the element responsible for the contamination. This information
can for some cases be obtained indirectly, if the change in lifetime with tempera-
ture or injection level is measured. Injection levels in the FPC mode are generally
one or two orders of magnitude smaller compared to the BPC mode at the same
10 Applications 214
light intensity, as shown in Fig. 10.4 b and c. Combining both modes for injection
level spectroscopy, the accessible levels cover four orders of magnitude. This en-
ables us to distinguish, for example, titanium from iron contamination, because
in the case of Fe the lifetime increases by one order of magnitude for an increase
in injection level of three orders of magnitude, while the lifetime shows little de-
pendence on injection level in the case of Ti. In addition iron contamination in p-
type Si can be identified by the characteristic decrease in lifetime upon low-tem-
perature annealing (2108C). This effect is caused by a break-up of FeB pairs at
elevated temperatures. Other elements such as Ni can be identified, because they
produce a decrease in lifetime accompanied by a significant increase in the dark
current, because of nickel silicide precipitates in the SCR, as shown in Fig. 2.7
[Wa4, Zo1, Ei1, Wi2].
If different wavelengths are used for illumination, the effects of bulk and sur-
face recombination can be decoupled. Surface recombination velocities between
10
2
and 10
5
cm s
1
, as well as the bulk diffusion length L
D
, have been measured
simultaneously using this multi-color method [Os1]. Different wavelengths can
also be used to obtain a certain depth resolution. For the case of oxygen-rich CZ
wafers, for example, the depth of the denuded zone, the region free of oxygen pre-
cipitations, has been found to be measurable using an 820 nm and a 905 nm la-
ser diode for illumination [Ob1]. This technique is not only advantageous because
it is non-destructive, it also offers to map the lateral distribution of oxygen precipi-
tations with high resolution, as discussed below.
If the homogeneous illumination is replaced by a point-like source, a laser for
example, diffusion length mapping of the wafer becomes possible. In this case
the equations, discussed above, become approximations, because the carrier diffu-
sion profile is no longer one-dimensional. However, the use of point-like illumina-
tion is not intended to give a better quantitative determination of L
D
, but rather to
produce a lateral resolution. The maximum obtainable resolution is determined
by the diffusion profile of minority carriers. It is therefore in the order of the wa-
fer thickness for the BPC mode and in the order of 50 lm for FPC measurements
[Ca9]. This lateral resolution is much better than for any comparable technique,
such as SPV [Go4], photoconductive decay (PCD) [De7], or microwave absorption
[Na4, Ya8]. The use of a laser for illumination also enables fast scanning, which is
not possible with other methods, for example a moving Hg drop used as a Schott-
ky contact [Su5]. The lateral resolution of the technique was first demonstrated for
wafers that had been intentionally contaminated with different metals [Le11], as
shown, for example, in Fig. 10.6a. The diffusion length of the contaminated wafer
was then measured for every square millimeter of surface inside the O-ring, using
BPC mode. The observed values are represented by dots of different sizes. In the
diffusion length plot, as shown in Fig. 10.6b, the contamination pattern is clearly
visible.
The high spatial resolution of the technique enables identification of a source of
contamination by the characteristic fingerprint left on the wafer. This is illustrated
in Fig. 10.5, which shows lifetime maps of wafers contaminated in different ways.
The concentric circles shown in Fig. 10.5a are caused by oxygen precipitates. The
10.3 Diffusion Length and Defect Mapping 215
oxygen originates from the crystal growth process and is therefore incorporated in
a circular manner. Such precipitates can also be revealed by XRT, although diffu-
sion length mapping has been found to be more sensitive. In contrast to such
bulk contamination, surface contaminants show up only after a thermal drive-in
process. This is visualized by the imprints of vacuum tweezers shown in
Fig. 10.5b, which were detected on as-received wafers after an RTA step (12008C,
30 s). Other contamination sources present in a wafer fab and their characteristic
fingerprints are shown in Fig. 10.5cf. Even after the full device manufacturing
process, contamination incorporated during processing can be resolved, if the top-
most 10 lm of the wafer are removed prior to the diffusion length measurement.
The result described above generated sufficient interest for commercialization
of this technique. The ELYMAT (electrolytic metal tracer) [21] is equipped with la-
ser diodes (670 and 905 nm), two rotatable mirrors, and an electrolytic double cell.
The time needed for 10000 single measurements, which are necessary for one wa-
fer map, is in the order of a few minutes, with a lifetime measurement range of
0.01 to 10 ls in FPC mode and 10 ls to 1 ms in BPC mode [Ei1, Os1].
10 Applications 216
Fig. 10.5 Diffusion length maps characteristic
of certain sources of contamination. Each dot
corresponds to one measurement, with dark
regions representing areas of high diffusion
length. (a) Concentric rings indicate oxygen
precipitates caused by a high oxygen level
during wafer growth. Contamination induced
by wafer handling: (b) vacuum tweezers,
(c) transport belts, (d) wafer chucks, (e) boat
contamination in the furnace, and (f ) contam-
ination by particulates.
10.3.2
Defect Mapping
While the local variation of a laser-induced current can be used for diffusion
length mapping, as discussed above, the local formation of hydrogen bubbles or
metal plating can be used for defect mapping. The yield of a semiconductor fabri-
cation process depends critically on the defect density of insulating layers such as
an SiO
2
or an oxide-nitride-oxide (ONO) dielectric. Average defect densities are in
the order of 11000 cm
2
, while the size of the actual defect is usually below a mi-
crometer, as shown, for example, in Fig. 2.7a. The small size and the low density
make it very difficult to find such defects by microscopic inspection only. It is
more advisable to detect the electric current that flows through the defect. This
can be done, for example, by a liquid crystal, which changes its reflection due to
heating by the high local current. In an electrolyte the chemical reaction caused
by the current through the defect can be sensed. For anodic currents the reaction
at the defect is electrochemical etching, as discussed in Section 4.5, while for
cathodic currents the defect is detected by metal plating or hydrogen bubble for-
mation [Or1]. If an n-type wafer with an insulating oxide is immersed in an elec-
trolyte, for example 10% acetic acid, and cathodically biased, hydrogen bubbles
will form at defects in the oxide. For the case of a p-type wafer additional illumi-
nation is needed because it is under reverse bias. Such measurements can be per-
formed in simple O-ring cells or in double cells, as shown in Fig. 1.7. In Fig. 10.6
the technique is demonstrated using an intentionally contaminated wafer. Oxide
defects induced by mechanical damage or by copper contamination of the back-
side are clearly visible because they are decorated by hydrogen bubbles, as shown
in Fig. 10.6e. Microscopic investigations have revealed that copper silicide precipi-
tates penetrate the oxide [We4]. If a more permanent decoration is desirable the
Pt counter electrode can be replaced by a copper plate. In this case the cathodic
current at the defects leads to copper deposition and after a few minutes the cop-
per crystallites become large enough to be identified [Ok2].
Defects in a SCR, which is present under reverse bias, can be tested in a simi-
lar way. Figure 10.6c shows the same wafer as in Fig. 10.6e after removal of the
oxide and under cathodic polarization in the dark. Hydrogen bubbles caused by
the dark current now decorate nickel silicide precipitates that short-circuit the
SCR. Nickel precipitates are known to increase the dark current of a p-type Si
electrode under reverse bias by orders of magnitude [Wa4]. If the bias is increased
the copper silicide precipitates also become visible, as shown in Fig. 10.6d. This
method, like defect etching (Fig. 10.4f), is only sensitive to precipitated metals.
Metals that stay in solution, like iron, do not show up in defect mapping and
have to be determined by other methods, for example diffusion length mapping.
If mapping of the defects is dispensable and only the average contamination
level is of interest, measurements of the reverse dark current are sufficient to pro-
vide this information [Wi2]. This method is also applicable to n-type samples,
which is in contrast to decoration of SCR defects by hydrogen bubbles, which is
not possible in the anodic regime.
10.3 Diffusion Length and Defect Mapping 217
10 Applications 218
Fig. 10.6 A p-type Si wafer with a 20 nm
thick thermal oxide has been contaminated
by scratching the backside with metal wires
(Ni, Cu, Fe), according to the pattern shown
in (a) and later annealed at 12008C for 30 s.
(e) Under cathodic bias in acetic acid, oxide
defects become decorated by hydrogen bub-
bles. (c, d) After oxide removal junction de-
fects caused by metal precipitates are deco-
rated by hydrogen bubbles, if sufficient catho-
dic bias is applied in the dark. (f ) The re-
moval of the precipitates by a defect etchant,
like Secco etch [Se5], produces a surface
roughness, visible as haze under small angle
illumination (wafer backside is shown).
(b) Non-precipitating metals, like Fe, have to
be traced by other methods, e.g. diffusion
length mapping. Scale: L
D
=50 lm (white) to
L
D
=250 lm (black).
10.4
Sensors and Biochips
Silicon-based sensors can be classified as physical, chemical or biological, depend-
ing on the parameter to be measured.
Sensors for measurements of physical parameters such as pressure, rotation or
acceleration are commonly based on elongation or vibration of membranes, canti-
levers or other proof masses. The electrochemical processes used to achieve these
micromechanical structures are commonly etch-stop techniques, as discussed in
Section 4.5, or sacrificial layer techniques, discussed in Section 10.7.
For the fabrication of chemical sensors, either the unique properties of the sili-
con electrode itself or a variation in PS properties with absorption of molecules at
the large internal surface, are exploited for sensing. Both principles are addressed
below.
10.4.1
Hydrofluoric Acid Sensor
The HF tester is a commercial safety tool for sensing whether an unidentified
liquid contains HF [2]. It shows in an exemplary way how the electrochemical
properties of a silicon electrode, namely its IV curve in HF, can be applied for
sensing. The ability to dissolve an anodic oxide layer formed on silicon electrodes
in aqueous electrolytes under anodic bias is a unique property of HF. HF is there-
fore the only electrolyte in which considerable, steady-state anodic currents are ob-
served, as shown schematically in Fig. 3.1. This effect has been exploited to real-
ize a simple but effective safety sensor, which allows us to check within seconds if
a liquid contains HF. This is useful for safety applications, because HF consti-
tutes a major health hazard in semiconductor manufacturing, as discussed in Sec-
tion 1.2.
The electronic circuit of the safety sensor consists of a p-type silicon electrode,
an LED, a resistor, two 3 V lithium batteries, and a platinum wire as a counter
electrode, connected in series, as shown in the right part of Fig. 10.7. These com-
ponents are assembled in a pen-like housing, optimized to measure even thin
layers of liquid on a flat surface, as shown in the left part of Fig. 10.7. This config-
uration is advantageous if a puddle, observed for example under a wet bench or
other equipment, is to be analyzed.
The working principle of the sensor is simple. If the tip of the sensor, which
contains the electrodes, is immersed in a liquid free of HF, an anodic oxide is
formed and the anodic current decreases within a second to very low values: the
LED is off. For the case of a liquid containing more than 5% HF, a constant anod-
ic current flows which is only limited by the series resistor and the LED emits
with its maximum intensity. If the liquid contains between 0.5% and 5% HF the
intensity of the LED becomes roughly proportional to the HF concentration. In
contrast to other chemical sensors where the electrodes are very sensitive to con-
tamination or drying, the HF sensor is quite robust. The sensor electrode can be
10.4 Sensors and Biochips 219
cleaned with acetone or even grinding paper if it is for example contaminated
with oil or resist. Because the resistor limits the current, the etch rate at the elec-
trode remains low even in highly concentrated HF and the lifetime of the elec-
trode is in the order of 1000 tests.
The safety sensor, however, gives only qualitative information. For a quantitative
determination of the concentration of HF in a solution, it is necessary to deter-
mine J
PS
, which can be done by scanning the anodic potential from about 3 V to
0 V and measuring the relative current maximum in a unstirred solution. If J
PS
and the temperature T are determined, the electrolyte concentration c can be cal-
culated using Eq. (4.9). This method of determining the concentration of HF is
superior to simple measurements of the conductivity of the solution, because it is
insensitive to dissolution products of Si or SiO
2
, or to other ionic species in the
analyte.
10.4.2
Porous Silicon-Based Gas Sensors
Chemical sensors based on changes in work function, refractive index, dielectric
constant, PL intensity or resistivity of PS with absorption of chemical species,
such as H
2
O, NO, NO
2
, Cl
2
, Br
2
, alcohols, ethers, aromatic and chlorinated hydro-
carbons, have been proposed by several groups [Sa7, Wa6]. The sensitivity to ad-
sorbates is usually discussed in terms of a change in surface charge by polar mol-
ecules, which produces a change in capacitance [An4] or resistivity [Bo7]. The ob-
served decrease in resistivity, for example, has been attributed to detrapping of
charge carriers from surface states, as discussed in Section 6.7 [Be6, St11, Le14].
A change in resistivity of a PS layer can easily be measured with a device struc-
ture as shown in the inset of Fig. 10.8. In this example of a sensor application, a
2030 lm thick meso PS film has been formed by anodization of a p-doped sub-
strate (10
19
cm
3
) in ethanoic HF at 10 mA cm
2
. An oxide, serving as an insulat-
ing layer, has then been sputtered on top. After deposition of two Pt or Au con-
tacts the resistivity can be measured between the two terminals. Figure 10.8
10 Applications 220
Fig. 10.7 The HF safety sensor (left) and its electronic circuit (right).
shows the response to NO
2
as a function of exposure time for meso PS layers of
different porosity. While a significant increase in conductivity has been observed
for an oxidizing species like NO
2
, a decrease has been observed for exposure to a
reducing gas like NO. At RT the sensor shows a high sensitivity around 1 ppm
for NO
2
and a moderate sensitivity of 5 ppm to NO. No response is observed to
non-polar gases like CO and CH
4
, up to levels of 1000 ppm. The ambient humid-
ity level, however, affects the sensitivity of the sensor [Bo7].
The inability to differentiate between different gases, e.g. NO
2
and water vapor,
is one of the drawbacks of the resistive sensor approach described above. This
problem can be circumvented by gas spectroscopy. The tunable light source
needed for this method can again be achieved using PS. Figure 10.9 illustrates
the working principle of the spectrometer. A broadband IR source illuminates a
gas cell. Depending on the specific gas inside the cell certain bands in the IR are
absorbed. By changing the tilt angle of the interference filter an IR spectrum is re-
corded by the thermopile detector.
The IR filter is realized by a PS layer with a modulation of porosity, which con-
stitutes an interference filter as described in detail in the next section. The 30 lm
thick porous layer is then released from the substrate by electropolishing, which
is easily done in situ by increasing the etching current density above J
PS
. This pro-
cess is commonly applied to form free-standing PS membranes and PS tubes
[Tj1]. The internal strain between the Si
3
N
4
layer used for masking and the po-
rous layer lifts the filter up to its rest position, as shown in Fig. 10.10. The filter
is suspended at two microactuator arms, which work as thermal bimorph actua-
10.4 Sensors and Biochips 221
Fig. 10.8 Dynamic response of meso PS gas
sensors of different porosities to low concen-
trations of NO
2
. Measurements were per-
formed at RT and 20% relative humidity. The
structure of the sensor is shown on the upper
left. After [Bo7].
tors if heated by a current applied to the CrNi film that has been deposited on the
arms. More than 908 deviation has been achieved with a DC power of 100 mW. It
has been shown that the system is able to differentiate between gases even at low
levels (for example 0.5% CO and 1% CO
2
in N
2
) [La9, La11].
The above examples show that it is possible to envisage the application of PS in
electronic artificial nose chips [Le26].
10 Applications 222
Fig. 10.9 A MEMS IR gas spectrometer, which measures the IR absorption of a gas at different
wavelengths selected by a tunable interference filter. After [La11].
Fig. 10.10 SEM micrograph of a tunable IR
Bragg filter (1.11.85 mm) of density-modu-
lated PS, suspended at two microactuator
arms. Its rest position is more than 1808
out of plane, showing the remarkable flatness
of the interface, which has been separated
from the substrate by electropolishing.
After [La11].
10.4.3
Biochips
While silicon has for decades been the dominant material for electronic and micro-
mechanical devices, it has only recently been used for medical and biological appli-
cations [St13]. PS is advantageous for some of these applications because of its large
surface area. Micro and meso PS, for example, are found to be biocompatible mate-
rials that bond directly with healthy bone by calcification, which is an advantage if
implantation in living tissue is desirable [Ca11] [23]. Because of its biodegradability
and non-toxicity, mesoporous silicon doped with trace elements has been proposed
to serve as an implanted long-term delivery source [Ca14] [23].
Other applications are aimed at using PS for sensing of biochemical com-
pounds. It is well known that the CV characteristics of an electrolyte-insulator-
semiconductor (EIS) structure can be used to sense the pH of a solution. A chem-
ical reaction can therefore be sensed if it results in a change in pH. This principle
has been successfully used to sense penicillin. First the enzyme penicillinase is
immobilized on an SiO
2
, Si
3
N
4
passivated silicon surface enlarged by macropores.
Then the analyte is brought into contact with the surface. The more penicillin pre-
sent in the analyte, the more molecules are converted to penicilloic acid, resulting
in a measurable decrease in pH near the sensor surface [Sc20].
10.4 Sensors and Biochips 223
Fig. 10.11 Schematic diagram of a drug
screening concept. Disease-associated genes
are identified using different display technolo-
gies. Probes for disease-associated genes are
designed and are spotted onto the porous
chip. Libraries of pharmaceutical compounds
are screened against a validated cell model
system. Messenger RNA is collected from the
cells and analyzed using the porous chip to
determine efficiency of the drug against multi-
ple gene targets. Hybridization events are de-
tected by optical readout of the chip, as
shown in Fig. 10.12. After [St12].
Another large area of biological applications of PS is micro array technology.
The concept of fabricating miniaturized molecular assay systems that are capable
of performing thousands of analyses simultaneously has its roots in the
electronics and semiconductor chip industry. The arrangement of transistors, re-
sistors and capacitors that make up electronic components on a silicon chip are
analogous to an array of molecular recognition elements on a biochip. The state
of each element on the biochip is determined by the abundance of the comple-
mentary analyte in a sample mixture. Molecular recognition events are translated
into an analytical signal via a signal transduction mechanism, for example fluores-
cence, which is measured by some detection device. Biochip recognition elements
and analytes can be small molecules, nucleic acids or proteins, such as ligand re-
ceptors antibodies and enzymes. A comparison of the sizes of important biologi-
cal building blocks and pore sizes in PS is given in Table 6.1.
Macro PS satisfies three major requirements of micro array technology: a large
surface area for attachment, an intense contact between analyte and recognition ele-
ments, and a geometry that supports the optical readout [St12] [19, 24]. For this ap-
plication macropores penetrating the full substrate thickness are required. Such
through-pores can be obtained by etchback in KOH, if a CVD nitride is deposited
inside the pores as etch stop layer prior to etchback. The recognition elements are
usually spotted onto the chip and are attached to the pore walls by linker mole-
cules. The analyte is then pumped through the pores, which guarantees intimate
10 Applications 224
Fig. 10.12 Chemiluminescence image of four
spotted probes on a macroporous silicon-
based flow-through chip after hybridization.
The total number of spots on a chip is in the
order of 1001000.
contact with the recognition elements. This reduces the time needed for hybridiza-
tion by an order of magnitude compared to flat substrates. A schematic diagram of
the application of a flow-through chip in drug discovery is shown in Fig. 10.11. For
the final readout the macroporous silicon chip is illuminated with an excitation
wavelength and the emanating fluorescent signal is recorded. An excitation source
is not necessary if chemiluminescence readout is used, as shown in Fig. 10.12.
10.5
Passive and Active Optical Devices
PS layers formed by anodization of silicon in HF show a variety of interesting optical
properties. For example, free-standing layers of micro PS behave like an optical long-
pass filter, as discussed in Section 7.3. However, high-quality low-pass filters based
on doped glasses are available for a multitude of cut-off wavelengths [17]. The
property of macroporous silicon to behave as a short-pass filter for light transmis-
sion along the pores is of particular value, because it is difficult to produce high-
pass filters for the visible regime. The different transmission properties of free-
standing micro-, meso- and macroporous silicon films are shown on the lower left
corner of the front cover of this book.
Optical filters with a well-defined transmission band can be manufactured by
periodic variations of the dielectric constant. PS offers two ways to realize such
periodic dielectric structures. Either variations in the optical density of meso- or
microporous films can be exploited or the contrast of the dielectric constant be-
tween the air-filled core of a macropore and its solid wall can be exploited. Both
approaches are discussed below.
However optical applications of PS are not restricted to passive devices. It has been
shown that electroluminescent solid-state devices based on silicon are feasible.
10.5.1
Optical Short-pass Filters
Bulk silicon constitutes an optical long-pass filter, as shown in Fig. 7.6. The same
is true for micro- and mesoporous silicon, for which the effective medium approx-
imation (EMA) is valid in the visible regime. The dimensions of macroporous sili-
con are in the visible regime and the EMA becomes invalid.
To calculate the optical transmittance of a single macropore, it can be modeled
by two opaque screens with an aperture of diameter d at a distance l, as shown in
the inset of Fig. 10.13. For such an arrangement the ratio n
t
of transmitted power
to incident power of a wavelength k can be approximated by
n
t
= p
2
(d=2)
4
=(k
2
l
2
) (10:11)
The short-pass characteristic of Eq. (10.11) is evident. Measurements of the
spectral transmittance of macropore arrays with the pore axis parallel to the inci-
10.5 Passive and Active Optical Devices 225
dent light beam confirm these calculations, as shown in Fig. 10.13. The dips in
the transmission characteristic at 290 and 370 nm have been interpreted as remi-
niscent of the direct transitions in bulk silicon, which indicates that the assump-
tion of zero reflection inside the pores is not justified.
Optical short-pass filters based on macropore arrays show several advantages
compared to conventional filters: macropore filters are true short-pass filters, be-
cause light is transmitted in empty space, not in matter. As a result, functionality
down to wavelengths in the order of a few nanometers is expected if operated in
vacuum. The cut-off wavelengths can simply be tuned by pore diameter and pore
length. Despite the fact that light is diffracted at the pore entrance and the pore
exit, the macropore filters are of optical quality because of constructive interfer-
ence between the high number of pores in an array [Le27].
10.5.2
Dielectric Thin Films
The structural features of micro- and mesoporous silicon are much smaller than
the wavelength of visible light, and so these materials may be treated according to
the EMA. The dependence of the porosity of micro- and mesoporous silicon on
formation current density and substrate doping density can be used to generate
layers with a single or a periodic change in the dielectric constant.
10 Applications 226
Fig. 10.13 The measured transmittance of
macropore arrays of pore diameter d and
pore length l, with the pore axis parallel to
the light beam. The data are corrected for
porosity (p=1). The transmittance of a single
macropore can be calculated assuming two
opaque screens with an aperture of diameter
d at a distance l, as shown in the inset.
After [Le27].
A single layer of a micro PS film on a silicon substrate always reduces its reflec-
tivity, because of its lower refractive index compared to bulk Si. Hence micro and
meso PS films of a thickness around 100 nm have been proposed as anti-reflec-
tive coatings for solar cells [Pr8, Gr9, Po1, Bi4, Sc18, St10].
The dependence of PS porosity on doping density has been exploited to fabri-
cate optical waveguides. Such structures require only a single variation of refrac-
tive index. If a p-type substrate with implanted p
+
lines is anodized and partially
oxidized, mesoporous lines of high refractive index are embedded in a low refrac-
tive index material. An obtained refractive index contrast between core and clad of
30% is promising for future applications [Ta12].
A substrate with a periodic variation of the doping density can be formed by
epitaxy. If such sandwich structures are anodized under constant current condi-
tions a surface film with periodic variations of refractive index is obtained. Such a
porosity superlattice formed by anodization of a sample with epitaxially grown
layers of high and low p-type doping density is shown in the TEM micrograph of
Fig. 10.14. The optical reflectance spectra of such periodic layered structures exhi-
bits characteristic transmission and reflection bands as shown in Fig. 10.15
[Be16]. It is worth noting that the TEM micrograph, as well as the simulation of
the density distribution from the reflectance spectra, as given in the inset of
Fig. 10.15, reveals a more complex structure than expected from the simple bi-
layer doping scheme, as discussed in detail in Section 8.1.
Another way of producing periodic porosities is a variation of anodization cur-
rent density. Drawbacks of this approach are a certain roughness of the different
227
Fig. 10.14 TEM cross-sectional micrograph
of a one-dimensional porosity superlattice.
Anodization was performed on a periodically
p-doped substrate (75 nm of 10
17
cm
3
, 75 nm
of 10
19
cm
3
) using a current density of
50 mA cm
2
in ethanoic HF. Values of porosi-
ty as labeled are determined by measure-
ments of spectral reflectivity, shown in
Fig. 10.15. After [Fr5].
layers, which increases with etch depth and the limited range of obtainable poros-
ities. For p
+
type substrates the porosity of the mesoporous layer can be varied
from about 30% to 75%, corresponding to current densities between 10 and
240 mA cm
2
, while for moderately doped p-type substrates, this range is consider-
ably smaller (6075%). The roughness has been ascribed to lateral variations of
PS growth rate [Fr5].
An optical microcavity produced by the latter process has been applied to tune the
emission from erbium-doped PS [Zh6]. Erbium compounds like Er
2
O
3
are known to
exhibit a narrow emission band at 1.54 lm, which is useful for optical telecommu-
nications. Several methods have been used to incorporate erbium in PS. A simple
and economical way is cathodic electrochemical doping. External quantum efficien-
cies of up to 0.01% have been shown from erbium-doped PS films under electrical
excitation [Lo2]. The emission band, however, is much broader than observed for
Er
2
O
3
. This drawback can be circumvented by the use of an optical cavity formed
by PS multilayers. In this case the band is narrowed and the intensity is increased
because emission is only allowed into optical cavity modes [Lo3].
In conclusion, it has been demonstrated that the manufacture of waveguides,
Bragg reflectors, Fabry-Perot filters and anti-reflective coatings can be based on PS
thin films. Such filters are applied for example in gas sensors, as described in the
preceding section.
10 Applications 228
Fig. 10.15 Measured reflectance spectrum
(solid line) of 10 periods of the superlattice
shown in Fig. 10.14. The fit (broken line) has
been calculated based on porosities shown as
in the inset. For the 67% porosity layer the
Bergman effective medium theory, in conjunc-
tion with a model dielectric function for the
particles, has been used. For the other layers
the Bruggeman effective medium theory was
used. After [Be16].
10.5.3
Photonic Crystals
The periodic structures discussed so far are one-dimensional. This is an inevitable
property of the formation method, which produces layers parallel to the surface.
The periodicity of macropore arrays offers a method for varying the dielectric
constant in two dimensions. This, and the high contrast in the dielectric constant,
which is 11.7 for the pore walls and 1 for the air-filled pores, enables us to fabri-
cate a photonic crystal with a complete two-dimensional photonic bandgap in the
IR regime [Gr8]. A photonic bandgap is a frequency range where photons are not
allowed to propagate. The propagation of photons in the periodic dielectric lattice
of a photonic crystal is analogous to the propagation of electrons in the periodic
potential of a semiconductor crystal, with the difference that the Coulomb interac-
tion of electrons has to be considered while photon interactions can be neglected.
The Brillouin zone of a triangular lattice, together with the calculated photonic
band structure for a lattice constant of 2.3 lm and H-polarization, is shown in the
center of Fig. 10.16.
The manufacture of a photonic crystal requires extreme process control because
a deviation from perfect periodicity in the order of a few percent of the wave-
length worsens the optical performance. Macroporous silicon is a potential candi-
date for the realization of such structures because of its photolithographic pattern-
ing. The precision of the macroporous structures is reflected in the transmission
measurements along the C-M and C-K directions, which exhibit a photonic band-
gap centered at 5 lm, as shown in Fig. 10.16. For measurement the macroporous
10.5 Passive and Active Optical Devices 229
Fig. 10.16 In the center part, the calculated
photonic band structure (TE-pol.) of a two-
dimensional photonic crystal of lattice constant
2.3 lm is plotted, while the inset shows the
first Brillouin zone. On the sides the mea-
sured transmission spectra of the macropor-
ous film for the C-M and C-K directions are
shown. After [Gr8].
layer has been structured into bars, as shown for example in Fig. 10.17, by a sec-
ond photolithography and subsequent plasma etching [Ot2].
The use of photolithography to determine the pore positions enables us to cre-
ate artificial defects in the photonic crystal. By leaving out a row of pores, for ex-
ample, propagating modes confined within the defect can be created for frequen-
cies within the photonic bandgap. Such a line defect can therefore act as a wave-
guide, with the confinement achieved by means of the photonic bandgap and not
by total internal reflection as in traditional waveguides. It has been shown that
the line defect as shown in Fig. 10.17 exhibits a single-mode transmission
characteristic [Le28].
A review of silicon-based photonic crystals is available in the literature [Bi6].
10.5.4
Electroluminescent Devices
The majority of application-related work exploiting the visible PL from PS is
aimed at the fabrication of electroluminescent solid-state devices. Only a few other
applications of the PL of PS, e.g. the use of luminescent PS for fluorescent labels
in biosensing [Ak1] or for chemical sensing [Le26], have been proposed. This sec-
tion therefore focuses on PS-based EL devices. Note that EL from porous struc-
tures using wet contacts is discussed in Section 7.4.
10 Applications 230
Fig. 10.17 Microstructured bar of macropor-
ous silicon with a waveguide oriented in the
C-K direction. The triangular lattice of pores
cladding the waveguide has a pitch of 1.5 lm,
producing an optical bandgap ranging from
3.3 to 5.5 lm. After [Le28].
Shortly after the observation of visible PL from micro PS at room temperature
[Ca1], the first EL from a solid-state device was reported [Ri2]. This initiated vigor-
ous research, because silicon-based optoelectronic devices seemed to be within
reach. After several years of intense research the potential and the main problems
involved with the EL from PS have been clarified.
The external quantum efficiency of the EL from PS-based devices has been in-
creased from low initial values of 0.001% [Ko9] to values close to 1% [Ni4, La6,
Co5]. This, however, is still about one order of magnitude smaller than the maxi-
mum quantum efficiency of state-of-the-art LEDs based on IIIV semiconductor
heterostructures.
For data transfer applications the modulation speed of an emitter is a decisive
parameter. Here the long lifetime of the excited state in PS becomes problematic.
The fall time of the EL is usually in the ls regime, while somewhat shorter val-
ues are reported for the rise time. Only for devices based on OPS has a signifi-
cantly shorter fall time (0.03 ls) been reported [Wa10]. For small signal modula-
tions of the EL from PS, frequencies in the order of 1 MHz are reported [Ts4,
Co5]. Such modulation frequencies are sufficient for display applications. For data
transfer via optical interconnects, however, they are much too low.
The requirements for spectral position and width of the EL emission depend
strongly on the application. While for display applications the three colors blue,
green and red are crucial, compatibility with the waveguide technology and the de-
tector are more important for data transfer. The FWHM of the EL from PS is
usually in the range 0.20.4 eV. Peak emission wavelengths in the red are usually
reported from electrochemically and stain-etched films [Xu3]. However, wave-
lengths from 450 to 700 nm have been reported for porous layers that have been
illuminated during anodization [St5].
The threshold voltage, the lowest voltage at which EL is observed, has been
drastically reduced from values in the order of hundreds of volts to values as low
as about 3 V for visual detection [Si4]. The threshold current density is usually in
the order of mA cm
2
, however values as low as 10 lA cm
2
have been reported to
be sufficient for visual detection of the EL [Si4].
The long-term stability of devices based on as-prepared PS films is generally
poor. Under emission conditions the degradation is measured usually in minutes
or hours. The stability of devices based on chemically [Ko18], anodically [Ge11] or
thermally oxidized PS films [It4] is usually better; however the standard lifetime
requirement for microelectronic devices (>10
5
h) has not been met.
The different parameters of EL discussed above illustrate that considerable ad-
vances have been made, but more are still needed to meet todays specifications
for EL devices. So far the basic properties of EL devices have been discussed. For
details of the EL device structure, including contact type, doping and geometry, a
comprehensive review is available [Co4].
10.5 Passive and Active Optical Devices 231
10.6
Porous Silicon-Based Electronic Devices
The material properties of PS offer new ways of making electronic devices. For the
manufacture of cold cathodes, for example, oxidized microporous polysilicon has
been found to be a promising material. The application of basic semiconductor pro-
cessing steps such as doping, oxidation and CVD to a macroporous material enable
us to fabricate silicon-based capacitors of high specific capacitance. Both devices will
be discussed below.
10.6.1
Cold Cathodes
In many electronic applications, e.g. vacuum tubes, an electron emitting cathode
is an indispensable part of the device. For many such devices cold electron emis-
sion is favorable because of its lower energy consumption.
In an early study of electron emission from PS it was proposed that the observed
emission is caused by extremely sharp silicon tips present at the interface between PS
and bulk silicon [Yu3], thus assuming PS to work by the same principle as conven-
tional field emitting arrays [Sp1]. However, later studies indicate that ballistic trans-
port through the porous layer is more likely [Ko20, Sh4, Ko21]. This is supported by
time-of-flight (TOF) measurements, which show that the drift length of carriers un-
der a strong electric field (10
5
Vcm
1
) within a PS layer may reach 1 lm [Kl4].
A schematic view of the cold cathode fabrication process is shown in Fig. 10.18.
The cold cathode is fabricated by low pressure chemical vapor deposition
(LPCVD) of 1.5 lm of non-doped polysilicon on a silicon wafer or a metallized
glass substrate. The topmost micrometer of polysilicon is then anodized
(10 mA cm
2
, 30 s) in ethanoic HF under illumination. This results in a porous
layer with inclusions of larger silicon crystallites, due to faster pore formation
along grain boundaries. After anodization the porous layer is oxidized (7008C,
60 min) and a semi-transparent (10 nm) gold film is deposited as a top electrode.
If a sufficient bias is applied between the metallized glass and the Au electrode,
an electron emission current I
e
can be observed, as shown in Fig. 10.19a. A Fow-
ler-Nordheim plot of the emission current, shown in Fig. 10.19b, indicates that
field-induced tunneling occurs during electron emission from the porous polysili-
con. The emission has been found to be homogeneous over the emitting surface
and to be stable over time. Spike-like fluctuations of the emission current, as com-
monly observed for field-emission devices, have not been found. The stability of
operation has been ascribed to a quasi-ballistic drift of electrons in the porous
layer. This may also be the cause of the low sensitivity of the emission current
against variations in vacuum quality, of up to 10 Pa for Ar, N
2
or O
2
. I
e
may reach
values in the mA cm
2
range and efficiencies in the order of 1%. The feasibility of
a 20 cm
2
multicolor matrix panel with a porous polysilicon cold cathode has been
demonstrated, which is promising for application of this technique in future large
flat panel displays [Ko23].
10 Applications 232
10.6.2
Silicon Capacitors
Since the invention of the Leiden jar in 1745 there has been tremendous progress
in capacitor technology; in particular, capacitance-to-volume ratios have been in-
creased by orders of magnitude. But todays capacitors still suffer from a few
drawbacks: either one selects a superior dielectric like polystyrene or SiO
2
and has
to cope with low surface-to-volume ratios of the electrodes or one increases the
electrode area by etching or sintering techniques and has to accept the disadvan-
tages of anodically formed oxides, which is the case for aluminum or tantalum ca-
pacitors. Ceramic dielectrics have been developed to solve this problem by increas-
ing the dielectric constant instead of the electrode area. A drawback of this
approach is the strong temperature and electric field dependence of the dielectric
constant in these materials. A solution to these problems is offered by a capacitor
10.6 Porous Silicon-Based Electronic Devices 233
Fig. 10.18 Schematic view of the
fabrication process of a porous
polysilicon-based cold cathode.
Note that metallized glass may
also serve as a substrate.
Redrawn from [Ko21].
Fig. 10.19 (a) Current-voltage characteristic
of the cold cathode according to the circuit
shown in Fig. 10.18. (b) A Fowler-Nordheim
plot indicates that the emission current is
caused by tunneling of electrons. After [Ko21].
concept based on macroporous silicon that combines an electrochemically en-
larged electrode surface with a superior dielectric; this device has been termed
SIKO [Le18] [19].
A macroporous silicon substrate with pores of about a micrometer and a pore
depth of a few tenths of a millimeter offers a surface area enhancement of about
two or three orders of magnitude compared to an unetched silicon surface. An ex-
ample of such a macroporous substrate used for fabrication of a silicon capacitor
(SIKO) is shown in Fig. 10.20 b.
All process steps after electrochemical etching are standard techniques in micro-
electronic manufacturing. The porous part of the low doped n-type substrate is
highly doped using a phosphorus silicate glass (PSG) diffusion process (n
+
region
in Fig. 10.20d). After removal of the PSG layer by wet etching, an ONO dielectric
is formed inside the pores. For the device as shown in Fig. 10.20, the ONO con-
sists of 5 nm of a thermally grown SiO
2
, 20 nm of CVD Si
3
N
4
and 5 nm of SiO
2
formed by thermal oxidation of the Si
3
N
4
. The second electrode of the capacitor is
realized by CVD of a phosphorus-doped polysilicon film. The two contacts to the
electrodes can either be realized separately on the front- and backside of the chip
or both on the top side using patterned Al or Au layers. In the latter case the sub-
strate contact is realized by removing the poly-Si and the ONO in the contact area
by lithography and etching. A few microns of aluminum are sputtered onto the
surface to provide the pads for bonding, as shown in Fig. 10.20a.
The electrical characteristics of the SIKO are summarized below. The specific
capacitance of the device shown in Fig. 10.20 is close to 4 lFV mm
3
, and with
smaller designs values of up to 20 lFV mm
3
are feasible. The range of manufac-
turable capacitance is not only a question of chip size but also of defect density of
the ONO. Surprisingly, a defect density of <10
3
cm
2
has been observed for the
porous structure, which is much smaller than the defect density of planar ONO
layers, which is in the order of 0.1 cm
2
. This low defect density can be under-
stood if the defect density of planar films is assumed to be due to particles that
do not penetrate into the pores.
The remarkable stability of the capacitance of the SIKO against variations in
bias, temperature, frequency and time of operation is a consequence of the superi-
or properties of its ONO dielectric. In contrast to aluminum and tantalum capaci-
tors, the SIKO is a symmetrical device. It shows no significant voltage depen-
dence of the capacitance, as the high e ceramic capacitors do. Only polymeric ca-
pacitors show a lower dependence of capacitance on bias than a SIKO.
The change of capacitance in relation to the temperature is very small and a lin-
ear function of the substrate temperature. Unlike the change in metal film capaci-
tors it is completely reversible. The maximum operating temperature of the capa-
citor chip (more than 2008) is determined by its aluminum gate. For encapsulated
systems the bond contacts and the material of the package will determine the
upper temperature limit.
The change of capacitance in relation to frequency is a matter of the polarizabil-
ity of the dielectric. This change is very large for ceramics and large for most poly-
mer dielectrics, but very small for Si
3
N
4
and SiO
2
.
10 Applications 234
For ceramic capacitors there is a continuous decrease of capacitance with time. The
higher the dielectric constant, the higher is this decrease. In metal film capacitors the
best values given for 63 Vcapacitors are changes of 0.5% in 2 years. The SIKO shows
the same long-term stability of the capacitance as the trench capacitors in DRAMs.
10.6 Porous Silicon-Based Electronic Devices 235
Fig. 10.20 The SIKO chip with its aluminum
contacts to the substrate and the active area
is shown together with the head of an ant for
size comparison. One side of the chip was
beveled in order to reveal the details.
(b) Ten-fold magnification of the marked area in
(a) shows a rectangular array of pores generated
by electrochemical etching. (c) Hundred-fold
magnification reveals the homogeneity of
pore diameters and spacings. (d) Thousand-
fold magnification shows the heart of the
capacitor: a highly doped substrate and
polysilicon layer with a thin ONO dielectric
in between.
The dissipation factor of capacitors at high frequencies is determined by the se-
ries resistance. For low frequencies there may be losses caused by leakage cur-
rents as well as by slow components in the polarizability, especially of high e
ceramics and polymer dielectrics. The dissipation factor of the SIKO at room tem-
perature is below 10
4
. At 2008C it is still very low (210
4
).
The ohmic series resistance is determined by the geometrical layout of the capa-
citor as well as by the conductance of its electrodes. The electrochemically etched
array of straight pores produces a comb-like electrode structure and gives very low
values of electrical series resistance (ESR) for the capacitor chip.
10.7
Sacrificial Layer Applications
Despite the fact that dry etching techniques have improved dramatically in recent
decades, the manufacture of microelectromechanical systems (MEMS) is still a do-
main of wet etching and silicon electrochemistry. The multiplicity of structures
that can be achieved with silicon, together with its excellent mechanical properties
[Pe6], have led to an immense variety of micromechanical applications.
In the manufacture of micromechanical devices electrochemistry is commonly
used to realize etch stop structures or to form porous layers. The first of these is
discussed in Section 4.5. In the latter case, the use of PS as a preserved layer or
as a sacrificial layer can be distinguished. In the first case PS is an integral part
of the ready device, as discussed in Sections 10.4 to 10.6, while in the latter case
the PS serves as a sacrificial layer and is removed during the manufacturing pro-
cess.
10.7.1
Microporous Silicon
Microporous silicon is suitable for sacrificial layer applications because of its high
etch rate ratio to bulk silicon, because it can be formed selectively, and because of
the low temperatures required for oxidation. PS can be formed selectively if the
substrate shows differently doped areas, as discussed in Section 4.5, or if a mask-
ing layer is used. Noble metal films can be used for masking as well as SiO
2
,
Si
3
N
4
and SiC. Oxidation conditions are given in Section 7.6, while the etch rates
of an etchant selective to PS are given in Fig. 2.5b.
The use of micro PS as a sacrificial layer is illustrated by the fabrication of a
free-standing polysilicon bridge, as used for example in a hot-wire anemometer
[St6]. The process sequence is depicted in Fig. 10.21. A sandwich of thermal SiO
2
,
n-type poly Si, NiCr and Au is deposited on a moderately doped p-type Si sub-
strate. After producing these layers, micro PS is formed under the poly Si bridges
by anodization in an electrochemical double cell, as shown in Fig. 1.8. The PS is
then removed by KOH etching at RT. The resulting structure is shown in
Fig. 10.22. By bonding another silicon wafer with a similar groove on top of the
10 Applications 236
structure a closed channel is formed. The use of a PS sacrificial layer is advanta-
geous compared to conventional alkaline etching because in the prior case there
are no geometrical restrictions due to crystal orientation. This decreases the re-
quired chip area and thereby reduces production costs [He5].
For applications where micro PS is transformed into SiO
2
it can be argued that
micro PS is not truly a sacrificial layer. However, the application described next
has been included in this section for the sake of completeness.
The doping dependence of the PS formation process and the possibility of
transforming PS to SiO
2
at relatively low temperatures can be used to form dielec-
trically isolated silicon islands [Ho1, Im1]. This process is also known as full isola-
tion by oxidized porous silicon (FIPOS). The process sequence is shown in
Fig. 10.23. First a p-type wafer with a nitride layer patterned by a resist is im-
planted by boron ions to form highly doped p-type regions in the unmasked
areas. After resist removal, protons are implanted through the nitride film. After
annealing at 5008C n-type regions are formed under the nitride film by the im-
planted protons. The proton dose, however, is not sufficient to reverse the doping
10.7 Sacrificial Layer Applications 237
Fig. 10.21 The process steps for
the fabrication of a free-standing
poly Si bridge. After [St6].
Fig. 10.22 SEM micrograph of the
500 nm thick polysilicon bridge (straight
line, center) that is the heart of the hot-
wire anemometer. The bridge crosses a
80 lm deep groove produced by a micro
PS sacrificial layer. After [St6].
of the heavily doped p-type regions. These areas, as well as the p-type substrate,
are then converted into PS by anodization in HF, while the n-type regions remain
unetched. Finally the PS is oxidized at 10508C in wet oxygen. At this temperature
the protons diffuse away and the isolated islands resume the initial p-type doping
of the bulk silicon wafer. The feasibility of the FIPOS approach has been demon-
strated by successful application of a complementary metal oxide semiconductor
(CMOS) process to such FIPOS substrates [Im1]. A drawback of this approach is
the limited size of the islands, because of the under-etching required.
10.7.2
Mesoporous Silicon
A method of producing an SOI structure that circumvents the shortcomings of
the FIPOS process described above, is based on the use of mesoporous silicon as
a sacrificial layer. In mesoporous silicon the perfect bulk lattice is preserved, while
the oxidation rate and the etch rate are only somewhat lower than for micro PS.
These properties favor meso PS as a sacrificial layer for production of SOI
material [Sa3, Sa8]. The process sequence is depicted in Fig. 10.24. A heavily
doped seed wafer is anodized to form 4 lm of low porosity meso PS on top of a
high porosity meso PS layer of about the same thickness. A pre-bake in H
2
(900
10408C) forms a continuous silicon film at the surface, by filling the pore open-
ings with migrating silicon surface atoms. This pre-bake reduces the defect den-
sity of the subsequently deposited epitaxial silicon layer (0.52 lm thick) signifi-
cantly. After forming a thin thermal oxide on top of the epitaxial layer, the wafer
is bonded to a handle wafer and annealed to increase the bonding strength.
Using a water jet the two bonded wafers are then cleaved at the weakest interface,
which is the high porosity meso PS. Thus the epitaxial layer has been transferred
to the handle wafer. By selective chemical etching, e.g. in a mixture of HF and
H
2
O
2
at RT, the remaining PS is removed. A final H
2
anneal smoothes the sur-
face of the epitaxial silicon film. Advantages of this process are the reusability of
the seed wafer and the freedom to choose the silicon film thickness, from ultra
10 Applications 238
Fig. 10.23 Fabrication of a silicon
island using the FIPOS process.
Redrawn from [Im1].
thin (< 50 nm) up to several micrometers. This technique has been commercia-
lized under the name ELTRAN (epitaxial layer transfer) [22]. A similar technique
has been proposed for the fabrication of low-cost monocrystalline Si photovoltaic
thin film cells [Br8].
10.7.3
Macroporous Silicon
One of the remarkable properties of macroporous silicon is the straightness of its
pores. This property makes macropore arrays a perfect template for forming other
materials with the same geometric precision. An example of such a process is the
fabrication of anti-scatter grids for radiology applications. In radiology, a collima-
tor located between the patient and the photographic plate is used to absorb scat-
tered X-ray photons, which otherwise would blur the image. Such collimators are
termed anti-scatter grids. Conventional grids consist of alternating layers of lead
as the absorbing medium, together with paper or aluminum coupled together. As
a result of this fabrication process the number of lines per millimeter and their
precision are limited, which reduces the resolution of the image and has become
a limiting factor in the performance of todays radiology system.
In order to circumvent these shortcomings, a fabrication process based on
macro PS as a sacrificial layer has been proposed [Le30]. The process sequence is
shown in Fig. 10.25. First etch pits in the desired pore pattern are formed on the
n-type silicon wafer surface by photolithography and subsequent alkaline etching.
Then deep macropores are formed by electrochemical etching according to the
10.7 Sacrificial Layer Applications 239
Fig. 10.24 Process for the fabrication
of the SOI wafer based on meso PS.
Redrawn from [Sa3].
10 Applications 240
Fig. 10.25 Schematic view of the
fabrication process for macro PS-
based anti-scatter grids. Redrawn
from [Le30].
Fig. 10.26 (a) SEM micrograph of a silicon
wafer with lead-filled macropores after etchback
of 265 lm of bulk Si. (b) The lead pillars are
arranged in closely spaced lines (pitch: 16 lm)
in one direction and a large spacing (pitch:
48 lm) in the other direction. Because of vol-
ume reduction of the lead during solidifica-
tion, the pore tips remain empty. After [Le30].
process described in Chapter 9. A nitride layer is then deposited by CVD. Next
the pores are filled with liquid lead under pressure. Then the wafer is cooled
down to RT and the lead is solidified. After removal of the nitride on the wafer
backside, about half of the wafer thickness is etched back in hot KOH. An SEM
image of a sample at this stage of the fabrication process is shown in Fig. 10.26a
and b. In this stage of the process the CVD nitride serves not only as an etch stop
but also as a mechanical support for the soft lead pillars. The fragile structure of
free-standing pillars is then further stabilized by an epoxy resin. If desired a repe-
tition of etchback and epoxy fill generates lead pillars in an epoxy matrix. This is
favorable for radiology applications in the low-energy regime (30 keV) where the
silicon absorption is undesirably high.
10.7 Sacrificial Layer Applications 241
List of Symbols and Constants
A area
a lattice constant (a
Si
=0.543095 nm)
a absorption coefficient
b exponential factor
C
PS
constant
C capacitance
c concentration
c
F
fluoride concentration
D substrate or film thickness
D
e
diffusion coefficient of electrons (in Si: 36.8 cm
2
s
1
)
D
h
diffusion coefficient of holes (in Si: 12.4 cm
2
s
1
)
D
HF
diffusion coefficient of HF (in H
2
O: 10
5
cm
2
s
1
)
d diameter (of pore or crystallite)
E electric field strength
E
BD
breakdown field strength (for Si: 310
5
Vcm
1
)
E
m
maximum field strength
E
Si
Youngs modulus (for Si: 140 GPa)
E
a
activation energy
E
B
Coulomb barrier energy
E
C
conduction band energy
E
G
bandgap energy
E
V
valence band energy
e elementary charge (1.60210
19
C)
e dielectric constant
e
0
dielectric constant of a vacuum (8.854210
12
AsV
1
m
1
)
e
r
relative dielectric constant
e
Si
relative dielectric constant of bulk silicon (11.9)
F
HF
flow of HF
f frequency
c
L
surface tension of liquid
c
Si
surface tension of silicon (1 J m
2
)
243
Appendices
Electrochemistry of Silicon: Instrumentation, Science, Materials and Applications.
Volker Lehmann
Copyright 2002 Wiley-VCH Verlag GmbH
ISBNs: 3-527-29321-3 (Hardcover); 3-527-60027-2 (Electronic)
h Plancks constant (6.626110
34
Js)
I current
i pitch
J current density
J
PS
critical current density
K constant
k Boltzmann constant (1.3810
23
J K
1
, 8.61810
5
eV K
1
)
L length
L
D
diffusion length (for e: L
e
, for h: L
h
)
l pore length
k wavelength
M molar weight
m mass
m
e