Journal of Alloys and Compounds 575 (2013) 17–23

Contents lists available at SciVerse ScienceDirect

Journal of Alloys and Compounds

journal homepage: www.elsevier.com/locate/jalcom

Effects of annealing in oxygen on electrical properties of AlGaN/GaN

heterostructures grown on Si
A.Y. Polyakov a, N.B. Smirnov a, Min-Woo Ha b, Cheol-Koo Hahn b, E.A. Kozhukhova a, A.V. Govorkov a,
R.V. Ryzhuk c, N.I. Kargin c, Han-Su Cho d, In-Hwan Lee d,⇑
a
Institute of Rare Metals, B. Tolmachevsky, 5, Moscow 119017, Russia
b
Korea Electronics Technology Institute, Seongnam, Gyeonggi 463-816, Republic of Korea
c
National Research Nuclear University (MIFI), Kashirskoe Ave. 31, Moscow 115409, Russia
d
School of Advanced Materials Engineering and Research Center for Advanced Materials Development, Chonbuk National University, Jeonju 561-756, Republic of Korea

a r t i c l e i n f o a b s t r a c t

Article history: We investigated the effects of oxygen annealing on electrical characteristics of AlGaN/GaN heterostruc-
Received 8 January 2013 tures grown on Si by metalorganic chemical vapor deposition. It was shown that such treatment sup-
Received in revised form 5 April 2013 presses parasitic tunneling leakage current due to surface states or dislocations, decreases the
Accepted 6 April 2013
threshold voltage corresponding to depletion of the two-dimensional electron gas at the AlGaN interface,
Available online 12 April 2013
and increases the effective interdevice isolation resistance. The suppression of the leakage current is
explained by oxygen passivation of the empty states at the surface of the AlGaN barrier. Oxygen diffusion
Keywords:
along dislocations and passivation of dislocation related states also could play a role. The decreased
AlGaN/GaN HEMT
Oxygen annealing
threshold voltage is attributed to the increased concentration of possibly oxygen related deep traps near
Leakage current the AlGaN/GaN interface. The increase in the GaN buffer resistivity after annealing is ascribed to the anni-
Deep traps hilation of the 0.38 eV deep electron traps believed to be introduced at the surface of GaN during mesa
formation by dry etching.
Ó 2013 Elsevier B.V. All rights reserved.

1. Introduction one chip, as demonstrated in some recent papers (see e.g. Ref.
[4]). There are, of course, very serious problems related to a
AlGaN/GaN high electron mobility transistor (HEMT) struc- large lattice mismatch and thermal expansion coefficient mis-
tures grown by metalorganic chemical vapor deposition and match of the GaN layers and Si substrates. These problems
molecular beam epitaxy (MBE) on Si substrates are now under caused a high density of extended defects and easy cracking of
intense studies for applications in high-frequency, medium the GaN films on Si in the early experiments. However, thanks
power microwave devices (see e.g. a recent review in Ref. [1] to intensive research and a lot of ingenuity demonstrated by var-
and references therein). (It could be mentioned here that alter- ious groups these problems have been to a large extent solved.
native approaches based on growth of two-dimensional metal Among the remedies offered one can name the use of AlN nucle-
oxides or chalcogenides have recently yielded rather promising ation layers, AlGaN transition buffer layers, low temperature AlN
results (see e.g. Refs. [2,3]), but the properties are still a far insertions, growth through a nanomask of SiNx and other tech-
cry from required by practical devices.) This interest is caused nological advancements (see a detailed discussion and relevant
by the low cost and ready availability of good crystalline quality, references in Ref. [1]). As a result, the two-dimensional electron
large diameter Si substrates. This is in contrast to extremely gas (2DEG) concentration and mobility in AlGaN/GaN hetero-
expensive and hard to come by semi-insulating SiC substrates junctions (HJs) on Si are on par with good quality HJs on sap-
or ‘‘native’’ bulk or quasi-bulk GaN and AlN substrates grown phire, the thickness of the GaN buffers in HEMTs of AlGaN/
by hydride vapor phase epitaxy (HVPE) or physical vapor trans- GaN on Si can be made as high as several microns which is suf-
port (PVT) techniques. Additional bonuses of growth on Si are ficient for achieving device breakdown voltages of about 200 V
the compatibility of many device processing operations with and devices net power of 20 W/mm [1]. At that, the thermal
the standard well established Si device technology and the pos- conductivity of Si, albeit lower than that of SiC, has been shown
sibility to integrate Si device circuits and GaN device circuits on through modeling [5] and experiment [6] to be sufficiently high
for operation of medium-power microwave devices. Neverthe-
⇑ Corresponding author. Tel.: +82 63 270 2292. less, some serious problems still remain to be addressed. For
E-mail address: ihlee@jbnu.ac.kr (I.-H. Lee).
example, for AlGaN/GaN on Si HJs and HEMTs, the leakage

0925-8388/$ - see front matter Ó 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.jallcom.2013.04.020
18 A.Y. Polyakov et al. / Journal of Alloys and Compounds 575 (2013) 17–23

current of the gate Schottky diodes has been shown [7] to be re- The masks used in these processes allowed the formation of a pattern of
Schottky diodes with diameters 0.5, 0.6, 0.7, 0.8, 1, and 1.5 mm and the distance
lated to some resonant tunneling process involving either dislo-
to the ohmic contact ring of either 20 lm or 50 lm. Table 1 summarizes the sets
cation related tunneling (see e.g. one of the recent reports in of combinations of Schottky diode diameters and distances to the ohmic contact
Refs. [8,9]) or tunneling via the surface states in AlGaN respon- ring studied for control and oxygen treated samples.
sible for the formation of the so called ‘‘virtual gate’’ during The characterization procedures for Schottky diodes involved Hall/Van der
pulsed device operation [10]. This situation is similar to the case Pauw two-dimensional electron gas (2DEG) concentration and mobility measure-
ments, current–voltage I–V measurements at various temperatures in the 85–
of AlGaN/GaN, AlGaN/AlN/GaN HJs and HEMTs grown on other
400 K temperature range, capacitance–frequency C–f measurements at different
substrates: sapphire, SiC [11–13]. The threshold voltage of temperatures and different applied biases, capacitance–voltage C–V measurements
AlGaN/GaN HJs and HEMTs on Si has been shown to be the at various temperatures, admittance spectra AS (see e.g. Ref. [20]) measurements,
function not only of the strain in the structure and the thickness and deep levels transient spectroscopy DLTS (see e.g. Ref. [21]) measurements.
We also looked in some detail into the effect the application of reverse voltage dur-
and composition of the barrier and buffer, as follows from the
ing cooling down had on low temperature C–V characteristics. The changes affected
theoretical models describing the ‘‘piezoelectric doping’’ (see by illumination at low temperature on C–V characteristics of Schottky diodes were
e.g. Refs. [14,15]), but also to be determined by the presence also studied (these latter experiments followed the procedure outlined in detail in
of deep negatively charged states in the AlGaN barrier or at some of our recent papers [7,12,13]). I–V data was collected using the HP4140B
the AlGaN/GaN interface [7,12,13]. Experimental studies of the picoamperemeter, C–V, C–f, AS measurements were performed using the HP4192
impedance analyzer at frequencies 5 Hz to 13 MHz, DLTS spectra were taken using
behavior of such states suggest that they could be due to
HP4280A C–V/C–t meter with external pulse generator HP8112A. Illumination was
complexes of native defects with some major technological done using a set of high-power light emitting diodes with the peak wavelength
impurities [12,13]. ranging from 625 nm to 365 nm.
In what follows we show that annealing of AlGaN/GaN HJs on Si In addition, the GaN buffer conductivity was determined by I–V measurements at
different temperatures performed between ohmic contacts located on adjacent me-
in oxygen has a marked effect on both the leakage current of Scho-
sas with similar mesa diameter. The temperature dependence of current through
ttky diodes and on the threshold voltage of AlGaN/GaN HEMTs. such ohmic contacts was measured to determine the position of the Fermi level pin-
Previously, oxygen annealing has been used to produce a thin insu- ning in the GaN buffer. The spectra of deep traps present in the GaN layer was ob-
lating layer of Al2O3 on the surface of the 2DEG supplying barrier. tained from photoinduced current transient spectroscopy (PICTS) [22,23]
With this, under optimized conditions, it has been shown that the measurements. In all experiments the temperature of the samples was varied and
controlled by the automatic system including a gas-flow liquid nitrogen cryostat
normally-off transistor structures with low leakage and relatively
(Oxford Instruments CF-4) and a computer controlled HP6038A programmable cur-
low density of interface states between the dielectric and the semi- rent source driving the heater. Detailed descriptions of the experimental setups can
conductor in such metal–insulator–semiconductor (MIS) structure be found in our earlier papers [24,25].
could be prepared [16]. Experimental studies of AlGaN/GaN struc-
tures in various degrees of oxidation showed that the magnitude of
band-bending at the surface of AlGaN/GaN caused by the surface 3. Results and discussion
Fermi level pinning is a strong function of the state of oxidation
and that the surface barrier height can substantially vary depend- We start this presentation with summarizing the changes in-
ing on the sate of oxidation (see Ref. [17]). Theoretical studies of duced by high temperature annealing in oxygen for the most prac-
the problem in Ref. [18] explain these observations by taking into tically important electrical properties and then proceed to the
account changes in surface reconstruction and bonding upon oxi- analysis of the data that helps to better understand the underlying
dation. They also predict an increase with oxidation of the density processes. Consider first the results of Hall/van der Pauw measure-
of defect-related surface donors in AlGaN. Such donors are believed ments on the samples before and after annealing in oxygen. Prior to
to be responsible for the virtual gate formation in AlGaN/GaN annealing the room temperature 2DEG concentration was
HEMTs during pulsed operation (see Ref. [10]). In our work we 9  1012 cm 2 and the 2DEG mobility was 1150 cm2/V s. This is in
tried to avoid the MIS structure formation upon oxygen treatment, reasonable agreement with the values often observed for good qual-
but to study the effects of oxygen atoms penetration into the bar- ity AlGaN/GaN HEMT structures on Si prepared by MOCVD (see e.g.
rier and to the AlGaN/GaN interface. Refs. [1,13]). After oxygen annealing and removing the oxide in buf-
fered HF the values were slightly higher, 1.15  1013 cm 2 for con-
2. Experimental centration and 1260 cm2/V s for mobility.
Fig. 2a and b presents the room temperature I–V characteristics
AlGaN/GaN high electron mobility transistor structures were grown by metalor-
ganic chemical vapor deposition on Si (1 1 1) substrates using high temperature AlN
of the reference and annealed samples. In the figure, the current
nucleation layer and a series of transition layers to relieve the high lattice mismatch density is plotted versus bias for different Schottky diode diame-
and thermal expansion coefficients mismatch between the Si substrate and the GaN ters and two different distances between the Schottky diode and
layer (see Ref. [1] and references therein). The GaN buffer layer was undoped and the ohmic contact ring, 20 lm and 50 lm. If the current is deter-
had the thickness of 1.7 lm, the thickness of undoped AlGaN barrier layer was
mined solely by the diode area these plots of current density versus
20 nm and the Al concentration in it was 23 atomic %. A detailed description of
the growth process and the structure of transition layers will be presented in a sep- voltage should form a dense group with little spread from diode to
arate paper. Here we would only like to note that the GaN layer was crack-free and diode. This is indeed the case for the oxygen treated samples in
the threading dislocation density in it, as measured by the density of dark spot de- Fig. 2b where the data are presented for a large variety of diode
fects in bandedge microcathodoluminescence (MCL) images (see e.g. Ref. [19]), was diameters and the two distances to the ohmic contact.
109 cm 2.
These HEMT structures were processed into test Schottky diodes by first
For the reference sample, the situation is quite different. The
defining 300-nm-high mesa structures by photolithography and dry etching. Then data are compared for three diode diameters of 0.6, 0.7, and
circular ohmic contacts were prepared on each mesa by electron beam (e-beam) 1 mm with 20 lm separation from the contact ring. Also shown
evaporation of Ti/Al/Mo/Au (20/100/20/200 nm), photolithography, lift-off, and is the data for the diode diameter of 1 mm for the separation to
rapid thermal annealing at 950 °C for 30 s. For control samples not subjected to oxy-
the ohmic contact of 20 lm and 50 lm. It can be seen that the leak-
gen annealing, circular Ni/Au (50/500 nm) Schottky diodes concentric with the oh-
mic contact rings were then prepared by photolithography and e-beam age current density increases considerably with the diode diameter
evaporation. Oxygen treatment was performed at 800 °C for 5 min by furnace for the fixed distance to the ohmic contacts and decreases substan-
annealing in dry oxygen. This treatment followed ohmic contacts deposition and tially for increased distance to the ohmic contact. This suggests a
preceded Schottky diodes deposition for oxygen annealed samples. After oxygen serious contribution of the surface leakage current for reference
annealing the samples were dipped in buffered oxide etcher (BOE; HF:H2O 1:6 solu-
tion) for 5 min to remove the oxide layers possibly formed during annealing. The
samples and a strong suppression of this parasitic leakage after
schematic representation of the used structure and the process flow stages is de- oxygen annealing at high temperature. It should also be pointed
picted in Fig. 1. out that the series resistance of the diodes determined from
 A.Y. Polyakov et al. / Journal of Alloys and Compounds 575 (2013) 17–23 19

Fig. 1. Schematic drawing illustrating the sample structure and different processing stages in our experiment.

forward I–V curves is considerably, by 1.3–1.5 times lower for the reasonably linear for both the reference and the oxygen annealed
oxygen treated samples (in agreement with Hall measurements samples, but the actual sheet resistivities were substantially differ-
above) and the ideality factor for these samples is generally lower, ent: 1012 X/square for the oxygen annealed sample and
close to 1.5 compared to higher values of 1.7–2 for the reference 6.7  108 X/square for the reference sample (the actual I–V charac-
sample. teristics are not shown here to save space). The temperature depen-
Room temperature capacitance–frequency characteristics mea- dence of this sheet resistivity gives the energy location of the deep
sured at zero bias for the reference and annealed samples are com- traps pinning the Fermi level in respective buffer layers. Fig. 5 com-
pared for several diode diameters in Fig. 3. At zero bias, one pares the temperature dependence of the ohmic dark current be-
measures the ‘‘accumulation’’ capacitance of the reference samples tween the mesas for the reference sample (measurements done at
when the space charge region boundary is located within the trian- 3 V) and the sample after oxygen annealing (measurements at
gular quantum well near the AlGaN/GaN interface. Thus, the capac- 30 V). This temperature dependence is equivalent to the tempera-
itance is determined mainly by the width of the AlGaN barrier. It ture dependence of the sheet conductivity of the buffer. The respec-
can be seen that this capacitance is slightly lower for the reference tive activation energies are 0.38 eV for the reference sample and
sample than for the annealed sample indicating a measurably low- 0.59 eV for the oxygen annealed sample and correspond to the
er (20 nm versus 23 nm) thickness of the AlGaN barrier in the latter depth of the traps pinning the Fermi level in respective GaN buffers
case, most likely due to partial oxidation of the top portion of the (it should be mentioned that the photosensitivity of the buffers is
barrier during annealing in oxygen and etching the oxidized part quite measurable and the measurements above were performed
off after subsequent treatment in buffered HF. upon cooling down from 400 K after heating up in the dark to this
Fig. 4 compares the low-frequency (1 kHz) C–V characteristics temperature to remove the effects of deep traps filling at room tem-
of the Schottky diodes with diameter 1 mm for the reference and perature by ambient light). Thus, oxygen treatment improves the
annealed samples. The data shows that the reverse voltage corre- interdevice insulation in AlGaN/GaN HEMTs and increases effective
sponding to the full depletion of the 2DEG region is shifted by sheet resistance of the GaN buffer.
0.3 V to more positive voltages after annealing. This voltage is
equivalent to the threshold voltage Vth in HEMT I–V characteristics.
In the Schottky diodes I–V curves in Fig. 2a and b at this voltage the Table 1
Dimensions of the Schottky diodes pattern prepared on the reference and oxygen
reverse current levels off because the space charge region moves
annealed samples: d is the diode diameter, D is the distance from the edge of the
from the 2DEG region into the GaN buffer. The threshold voltage Schottky diode to the ohmic contact ring.
thus determined has the same value as the one estimated from
Diode number d (mm) D (lm)
C–V data and shows the same trend after oxygen treatment. Thus,
annealing in oxygen is expected to quite measurably shift the d01 1.5 20
d02 1 20
threshold voltage of AlGaN/GaN HEMTs to more positive values.
d03 1 50
Finally, it was interesting to compare the current flowing be- d04 0.5 50
tween the two ohmic contacts located on adjacent mesas. This cur- d05 0.5 20
rent characterizes interdevice insulation in HEMT structures and is d06 0.7 20
determined by the resistivity of the GaN buffer. Current–voltage d07 0.8 20
d08 0.6 20
characteristics measured between such two ohmic contacts were
20 A.Y. Polyakov et al. / Journal of Alloys and Compounds 575 (2013) 17–23

(a) 10 0

d=1mm
4000

10
-1
3500 Oxygen treated
20 µm
Current density (A/cm )
2

-2
10 3000

Capacitance (pF)
-3
10 2500 Reference
-4
10
d=1 mm 2000
-5
10 50 µm 1500
-6
10
-7
d=0.6 mm, 20 µm 1000
10
-8 Reference sample d=0.7 mm, 500
10
20 µm 0
-9
10 -6 -4 -2 0
-5 -4 -3 -2 -1 0 1 2
Voltage (V) Voltage (V)

Fig. 4. Room temperature 1 kHz C–V characteristics of the d03 (d = 1 mm,

(b) 10 0
D = 50 lm) diodes of the reference sample (solid line, blue) and the oxygen treated
-1 sample (dashed line, red). (For interpretation of the references to color in this figure
10 Oxygen treated
Current density (A/cm )

legend, the reader is referred to the web version of this article.)

2

-2
10
-3
10
-4
10
10
-5
10
-8 Reference, +3V
-6 375 meV
10
-7 -9
10 10

10
-8 Current (A)
-10
10
-9 10 O2 annealed,
-5 -4 -3 -2 -1 0 1 2
Voltage (V) 30V, 590 meV
-11
10
Fig. 2. (a) Room temperature I–V characteristics of the Schottky diodes prepared on
the reference sample: diode diameter d = 1 mm, distance to the ohmic contact -12
D = 20 lm (solid line, black), d = 1 mm, D = 50 lm (dashed line, red), d = 0.7 mm,
10
2 4
D = 20 lm (dash-dotted line, green), and d = 0.6 mm, D = 20 lm (dash-dot-dot line, -1
blue); (b) room temperature I–V characteristics for the oxygen annealed sample; 1000/T (K )
the data are shown for the d = 1.5 mm, D = 20 lm diode d01 (black curve), the
d = 1 mm, D = 20 lm d02 diode (red curve), the d = 1 mm, D = 50 lm d03 diode Fig. 5. The temperature dependence of current through two ohmic contacts on
(green curve), the d = 0.8 mm, D = 20 lm d07 diode (blue curve), d = 0.7 mm, adjacent mesas for the reference sample (red line, measurements at +3 V) and the
D = 20 lm d06 diode (cyan curve), and d = 0.5 mm, D = 50 lm d04 diode (magenta oxygen annealed sample (blue line, measurement at +30 V). (For interpretation of
curve); the data shows very little spread in current density from diode to diode after the references to color in this figure legend, the reader is referred to the web version
annealing. (For interpretation of the references to color in this figure legend, the of this article.)
reader is referred to the web version of this article.)

Schottky diodes) is most likely due to the alleviation of the surface

8000 Fermi level pinning and decreased surface barrier. As discussed in
Refs. [17,18] it is difficult to avoid the formation of native oxide
1.5 mm, 20 µ m and surface pinning in AlGaN/GaN structures. We believe, how-
Capacitance (pF)

6000 ever, that intentional oxidation with subsequent etching in HF

leaves behind an unpinned clean surface passivated with hydrogen
(such phenomena have been reported for other semiconductors,
4000 1 mm, 50 µm
see e.g. one of the recent reports in Ref. [26]). The slight increase
of the 2DEG mobility could then come from suppressing the impact
2000 0.7 mm, 20 µ m of the band-bending variations across the sample. Mind that the
results of neutron irradiation studies on AlGaN/GaN structures do
suggest that the Fermi level at the surface could be strongly pinned
0
-1 0 1 2 3 4
(see Ref. [13]).
10 10 10 10 10 10
As for the reverse leakage current of the Schottky diodes some
Frequency (kHz) insight into its origin is provided by Fig. 6. The figure shows I–V
characteristics of the reference sample (1 mm diameter diode)
Fig. 3. Room temperature C–f characteristics of the reference (solid lines) and
oxygen treated (dashed lines) samples; d = 1.5 mm (black curves), d = 1 mm (green
measured at room temperature, at 400 K, and at 85 K. It can be
curves); d = 0.7 mm (blue curves). (For interpretation of the references to color in seen that, at room temperature, the reverse current reaches its
this figure legend, the reader is referred to the web version of this article.) peak at reverse voltage corresponding to the threshold voltage
and then shows a region of slight negative resistance. The leakage
current at 400 K is at high reverse voltages even slightly lower than
Let us now proceed to the analysis of the data that helps to at room temperature, while at 85 K the reverse current is the high-
somewhat understand the observed phenomena. The slight in- est of all. The latter, however, is only true when the sample is
crease of the 2DEG concentration and mobility after oxygen treat- cooled down at 0 V. After cooling at high reverse bias equal to or
ment and HF etching (and a decrease in the series resistance of the exceeding Vth ( 3.2 V in Fig. 6) the reverse current at 85 K
 A.Y. Polyakov et al. / Journal of Alloys and Compounds 575 (2013) 17–23 21

10
0
cooling down at high reverse bias measurably shifts the C–V char-
-1 Reference sample acteristic to more positive voltages, while illumination shifts the
10
C–V characteristic to more negative voltage. The effect of the bias
Current density (A/cm )
-2 85K, 0V
2

10 290K, 0V
-3
voltage saturates for voltages close to or higher than Vth, it is only
10 400K, 0V
-4
observed if the cooling down under applied bias starts at temper-
10 atures above room temperature. The effect of Vth change after illu-
-5
10 mination is persistent at low temperatures, it cannot be removed
10
-6 85K, -3.2V by application of forward bias at low temperatures, the photon en-
10
-7 ergy threshold for the observation of this effect is close to 1.5 eV,
10
-8 the return to the dark C–V characteristic occurs only after heating
-9 the sample up to about 320–330 K. These features again are quite
10
-10
common for many types of AlGaN/GaN, AlGaN/AlN/GaN, and In-
10
-5 -4 -3 -2 -1 0 1 2 AlN/GaN heterostructures [7,11–13] and have been shown by us
Voltage (V) to be due to the presence of deep acceptor centers in the AlGaN
barrier or at the AlGaN/GaN interface. These centers have a rela-
Fig. 6. I–V characteristics measured for the d03 (d = 1 mm, D = 50 mm) diode on the tively high barrier for capture of electrons which explains persis-
reference sample; measurements at room temperature (black line), at 400 K (red tent behavior at low temperatures. The occupation of these traps
line), at 85 K after cooling down at 0 V (cyan line), and at 85 K after cooling at
can be enhanced when cooling down under high reverse bias be-
3.2 V (blue line). (For interpretation of the references to color in this figure legend,
the reader is referred to the web version of this article.) cause of electrons tunneling from the Schottky metal into the
empty traps (see e.g. Ref. [28]). This increases the negative charge
trapped in the barrier or at the interface and shifts the C–V charac-
decreased by two orders of magnitude and became lower than at teristics to more positive voltage (the saturation of the effect for
room temperature. This kind of behavior is indicative of some sort biases exceeding Vth is explained by saturation of the reverse cur-
of resonant tunneling involving localized states and sensitive to the rent with voltage, as shown in Figs. 2 and 6). Illumination at low
band bending. The observed pattern is typical for many kinds of temperature removes electrons from such negatively charged cen-
AlGaN/GaN, AlGaN/AlN/GaN, InAlN/GaN HEMT heterojunction ters and persistently shifts the C–V characteristic to more negative
structures [7,11–13] and for actual transistor structures [6,7]. The voltage. As discussed in Refs. [12,13], the charge on these traps, Q,
authors of Ref. [10] have rather convincingly attributed it to the can be estimated from the threshold voltage shift, DVth, occurring
tunneling involving surface states in the AlGaN barrier. These after illumination as Q = DVth  Cacc. Here Cacc is the capacitance in
states are normally empty and can participate in the formation ‘‘accumulation’’, i.e. when the triangular quantum well near the
of a so called virtual Schottky diode (see e.g. Ref. [27]) in AlGaN/ AlGaN/GaN interface is filled with 2DEG electrons. The average
GaN HEMTs. It seems reasonable to assume that the combination areal density of the traps can then be simply calculated by dividing
of annealing in oxygen and subsequent etching helps to at least the charge by the product of electronic charge and the area of the
partly suppress the contribution of such states. Alternately, the Schottky diode. These calculations give for the reference sample in
kind of behavior illustrated in Fig. 6 has been attributed to the Fig. 7 the areal density of negatively charged acceptors as
leakage via dislocation related states (see e.g. Refs. [8,9]), in which 1012 cm 2 which is in good agreement with our estimates for
case it would suggest that oxygen can move along the dislocation AlGaN/GaN HEMTs and heterostructures on Si (1 1 1) substrate
cores during annealing and produce some form of oxide subse- grown by MOCVD by another group [7]. For the annealed sample
quently removed by HF etching. This question needs further study. the density of these traps was calculated by the same procedure
The nature of the Vth shift to positive voltages after annealing to be 1.6  1012 cm 2, i.e. considerably higher than in the reference
can be better understood by inspecting Fig. 7. It compares C–V sample. This increased density of negatively charged traps is
characteristics measured at 85 K on the reference sample after responsible for the observed shift of the threshold voltage in room
cooling down at 0 V, after cooling down at 3.2 V, and after cooling temperature C–V characteristics of annealed samples in Fig. 4
at 3.2 V and illumination with 365 nm LED. It can be seen that above. It seems to stand to reason to identify the traps in question
with some complexes involving oxygen.

3000
0V 20
2500 0.6 eV
DLTS signal (Arb. units)

15 O2 annealed
Reference, 85K
Capacitance (pF)

2000 C-V hysteresis 10

-3V->0V
5
1500 PPC -2.7V->0V
0

1000 -5 Reference
-10
500
-15 0.1 eV -3.2V->0V
-3.2V
0 -20
-6 -4 -2 0 100 150 200 250 300 350 400
Voltage (V) Temperature (K)
Fig. 7. 1 kHz 85 K C–V characteristics for the reference sample measured after Fig. 8. DLTS measurements for the reference sample with bias of 3.2 V (close to
cooling at 0 V (black line), at 3.2 V (blue line), after cooling down at 3.2 V and Vth) and bias pulse of 0 V (black line), for oxygen annealed sample with bias of 3 V
illumination with 365 nm LED (red curve, marked as ‘‘PPC’’; measurements under (close to Vth) and bias pulse 0 V (blue line), and the same sample for the bias voltage
illumination and after illumination and application of +2 V forward bias give the of 2.7 V (partial depletion) and bias pulse of 0 V (red line); time windows 100 ms/
same result). (For interpretation of the references to color in this figure legend, the 1000 ms. (For interpretation of the references to color in this figure legend, the
reader is referred to the web version of this article.) reader is referred to the web version of this article.)
22 A.Y. Polyakov et al. / Journal of Alloys and Compounds 575 (2013) 17–23

DLTS measurements help to clarify the energy position and the 1000
spatial location of the traps in question. Fig. 8 presents DLTS spec-
tra measured on the oxygen annealed sample at reverse biases 0V -2V -2.5V -2.7V
800
3 V and 2.7 V, with forward bias pulse of 0 V. In both cases

Capacitance (pF)
the applied bias is either close to or lower than the threshold volt-
600
age so that the full depletion of the 2DEG region is not achieved
(particularly for the 2.7 V bias) and the space charge region -2.8V
boundary is located near the AlGaN/GaN interface. It can be seen 400
that both spectra are dominated by the deep traps with activation
energy of 0.6 eV (the electron capture cross section of 200
6.8  10 15 cm2). The energy is very close to the activation energy
of the traps pinning the Fermi level in the GaN buffer after oxygen 0
-1 0 1 2 3 4
annealing. This makes us believe that the centers in question are 10 10 10 10 10 10
located at the AlGaN/GaN interface on the GaN side and are similar Frequency (kHz)
to the traps pinning the Fermi level in the semi-insulating buffer.
Since the trap in question is the most dominant one it stands to Fig. 9. Room temperature C–f characteristics measured for the reference sample at
reason to associate it with the deep acceptors responsible for meta- reverse biases of 0 V (black line), 2 V (red line), 2.5 V (green line), 2.7 V (blue
line), and 2.8 V (cyan line); the data are shown for the lowest diode diameter of
stability and threshold voltage shift in C–V characteristics dis- 0.5 mm and lowest distance to the ohmic contact ring of 20 lm. (For interpretation
cussed above. Similar traps have been observed in AlGaN/AlN/ of the references to color in this figure legend, the reader is referred to the web
GaN, AlGaN/GaN HEMT structures grown on sapphire [12] and in version of this article.)
AlGaN/GaN heterostructures and HEMTs grown on Si [7]. Measure-
ments on the reference sample at the reverse bias of 3.2 V and
forward bias pulse of 0 V still reveals the 0.6 eV traps as the dom- traps and traps in the bulk. The amplitude of this negative signal is
inant species (see Fig. 8). It is difficult to attribute an absolute value the higher the higher the series resistance, while the activation en-
to these interface traps concentration based only on DLTS data, but ergy of the negative, series-resistance-related features gives some
C–V measurements suggest that the concentration of the traps in- idea of the activation energy of the process (in Fig. 8 this activation
creases significantly after annealing in oxygen and prompt the energy is 0.1 eV). The absence of the negative signal (at least in the
attribution of the traps to oxygen-related complexes (see above). low temperature part of DLTS spectra) for the oxygen annealed
This should then be also the case for the centers in the GaN buffer samples seems to correlate with the lower series resistance in
layer and for other heterojunctions studied in Refs. [7,11,13]. The I–V characteristics and the reduced influence of the deep traps at
increased concentration of these interfacial traps has a beneficial the surface of the AlGaN barrier (see above).
effect as concerns the decrease of the threshold voltage of the In conclusion, let us discuss the origin of the apparent decrease
HEMT structures. However, in a recent paper devoted to studies of the sheet resistivity of the GaN buffer in the reference sample
of current collapse and degradation mechanisms in high-power Al- compared to the oxygen annealed sample. Consider for that PICTS
GaN/GaN HEMTs deep traps with very close energy were held spectra of deep traps measured on these two buffers (see Fig. 10
responsible for device degradation during operation [29]. We also showing the temperature dependence of the DI = I(t1) I(t2) tran-
observed an increase of the concentration of such traps after neu- sient PICTS signal measured for the two time windows t1 and t2
tron [13] and electron irradiations, which, in the latter case, corre- and normalized to the photocurrent Iph [22,23]). For the annealed
lated with degradation of AC performance of transistors (detailed sample, the spectrum consists of two peaks corresponding to traps
results will be reported in a separate paper). In that respect, more with activation energy of 0.6 eV (most likely the one pinning the
detailed studies of the actual device characteristics of oxygen an- Fermi level) and another, deeper trap with the apparent activation
nealed HEMTs are necessary and are currently under way in our energy of 0.82 eV. In the reference sample one still observes the
laboratories. 0.6 eV traps, but also the 0.38 eV traps pinning the Fermi level.
When inspecting Fig. 8 one can see that, for the reference sam- Obviously, what happens during annealing is the annihilation of
ple, the positive sign electron-trap-like 0.6 eV trap signal (during
the capacitance transient the capacitance increases with time
[21]) is superimposed on a strong almost temperature independent 35
negative hole-trap-like signal. The origin of this feature becomes
more clear from inspecting the C–f characteristics of the diodes 30 0.38 eV Reference
PICTS signal Δ I/Iph (%)

on the reference sample measured at various applied reverse volt-

ages (see Fig. 9 for the smallest diameter of 0.5 mm diode d04 from 25
Table 1). It can be seen that, with increasing the applied bias, the 0.82 eV
20
roll-off frequency characterizing the effective series resistance
markedly decreases. This happens because the filling of the 2DEG 15 0.6 eV
region with increasing voltage decreases and the layer underneath
the Schottky diode becomes more resistive thus increasing the ac- 10
cess resistance to the diode. In the extreme case of full 2DEG deple-
5
tion and when the substrate is semi-insulating, as in our case, the O2 annealed
only access for the current to the Schottky diode becomes lateral, 0
from the side of the space charge region. Upon application of the 200 250 300 350 400
forward bias pulse the series resistance decreases and the capaci- Temperature (K)
tance increases. Switching of the bias back to reverse bias causes
a decrease of capacitance with time due to the increase with time Fig. 10. PICTS spectra of the reference sample (red line) and oxygen annealed
sample (blue line); bias of 5 V; light pulse of 365-nm-wavelength high brightness
of the series resistance. The result is the appearance of the hole- LED (5-s-long), time windows 100 ms/1000 ms. (For interpretation of the references
trap-like sign of the DLTS signal superimposed on the ‘‘normal’’ po- to color in this figure legend, the reader is referred to the web version of this
sitive electron-trap-like sign due to the recharging of the interfacial article.)
 A.Y. Polyakov et al. / Journal of Alloys and Compounds 575 (2013) 17–23 23

the 0.38 eV trap that could be, in fact, induced in the near surface [2] S. Balendhran, J. Deng, J.Z. Ou, S. Walia, J. Scott, J. Tang, K.L. Wang, M.R. Field, S.
Russo, S. Zhuikov, M.S. Strano, N. Medhekar, S. Sriram, M. Bhaskaran, K.
region during the low energy ions bombardment used to form the
Kalantar-Zadeh, Adv. Mater. 25 (2013) 109.
mesas. Similar traps have been reported for GaN irradiated with [3] Q.H. Wang, K. Kalanta-Zadeh, A. Kis, J.N. Coleman, M.S. Strano, Nature
protons (see recent reviews in Refs. [30–32]). Additional etching Nanotechnol. 7 (2012) 699.
of the thin oxide formed after annealing in oxygen could also be [4] T. Kazior, K.J. Herrick, J.R. La Roche, in: T. Li, M. Mastro, A. Dadgar (Eds.), III-V
Compound Semiconductors – Integration with Si-Based Microelectronics, CRC
conducive to effective removal of the damaged region with a high Press, Boca Raton, 2011 (Chapter 6).
density of radiation defects. [5] E.A. Douglas, F. Ren, S.J. Pearton, J. Vac. Sci. Technol. B 29 (2011) 020603.
[6] C.F. Lo, F. Ren, S.J. Pearton, A.Y. Polyakov, N.B. Smirnov, A.V. Govorkov, I.A.
Belogorokhov, A.I. Belogorokhov, V.Y. Reznik, J.W. Johnson, J. Vac. Sci. Technol.
4. Conclusions B 29 (2011) 042201.
[7] A.Y. Polyakov, N.B. Smirnov, A.V. Govorkov, E.A. Kozhukhova, S.J. Pearton, F.
We have shown that oxygen annealing at 800 °C of AlGaN/GaN/ Ren, L. Lui, J.W. Johnson, N.I. Kargin, R.V. Ryzhuk, J. Vac. Sci. Technol. B 31
(2013) 011211.
Si HJs leads to a partial suppression of the surface related leakage
[8] S.W. Kaun, M.H. Wong, S. Dasgupta, S. Choi, R. Chung, U. Mishra, J. Speck, Appl.
of the Schottky diodes, quite likely due to the passivation of empty Phys. Express 4 (2011) 024101.
defect sites at the AlGaN barrier surface and possibly also due to [9] Y.-Y. Wong, E.Y. Chang, T.-H. Yang, J.-R. Chang, J.-T. Ku, M.K. Hydait, W.-C.
oxygen diffusion via dislocations and passivation of dislocation re- Chou, M. Chen, K.-L. Lin, J. Electrochem. Soc. 157 (2010) H746.
[10] J.K. Kanshik, V.R. Balakrishnan, B.S. Panwar, R. Muralidharan, Semicond. Sci.
lated defect traps. This annealing decreases (by absolute value) the Technol. 28 (2013) 015026.
threshold voltage corresponding to the full depletion of 2DEG near [11] C.F. Lo, L. Liu, T.S. Kang, F. Ren, O. Laboutin, Y. Cao, W.J. Johnson, A.Y. Polyakov,
the AlGaN/GaN interface. The effect is attributed to the introduc- N.B. Smirnov, A.V. Govorkov, I.A. Belogorokhov, A.I. Belogorokhov, S.J. Pearton,
J. Vac. Sci. Technol. B 30 (2012) 011205.
tion of additional deep acceptor centers into the AlGaN barrier or [12] A.Y. Polyakov, N.B. Smirnov, A.V. Govorkov, E.A. Kozhukhova, S.J. Pearton, F.
at the AlGaN/GaN interface after oxygen annealing. This is clearly Ren, S.Y. Karpov, K.D. Shcherbachev, N.G. Kolin, W. Lim, J. Vac. Sci. Technol. B
demonstrated by the shift of the low temperature C–V characteris- 30 (2012) 041209.
[13] A.Y. Polyakov, N.B. Smirnov, A.V. Govorkov, E.A. Kozhukhova, S.J. Pearton, F.
tics of the Schottky diodes upon increased bias at which the sam- Ren, L. Liu, J.W. Johnson, W. Lim, N.G. Kolin, S.S. Veryovkin, V.S. Ermakov, J.
ples were cooled. Persistent reverse shift of the C–V characteristics Vac. Sci. Technol. B 30 (2012) 061207.
after illumination at low temperature also points in this direction. [14] I. Vurgaftman, J.R. Meyer, J. Appl. Phys. 94 (2003) 3675.
[15] L. Hsu, M. Valukiewicz, J. Appl. Phys. 89 (2011) 1783.
The concentration of these additional traps introduced by oxygen [16] M. Eikelkamp, M. Weingarten, L.R. Khosroo, N. Ketteniss, H. Behmenburg, M.
annealing is close to 1012 cm 2. DLTS spectra measurements in Heuken, D. Donoval, A. Chvala, P. Kordos, H. Kalisch, A. Vescan, J. Appl. Phys.
oxygen annealed heterojunctions revealed the electron traps with 110 (2011) 084501.
[17] M. Higashiwaku, S. Choudhury, B. Swenson, U. Mishra, Appl. Phys. Lett. 97
the activation energy of 0.6 eV as the main deep traps at the Al-
(2010) 222104.
GaN/GaN interface. The similarity of the trap energy to the energy [18] M.S. Miao, J.R. Weber, C.G. Van de Walle, J. Appl. Phys. 107 (2010) 123713.
of deep traps pinning the Fermi level in the GaN buffer, as deter- [19] A.Y. Polyakov, D.-W. Jeon, I.-H. Lee, N.B. Smirnov, A.V. Govorkov, E.A.
mined by the ohmic current temperature dependence measure- Kozhukhova, E.B. Yakimov, J. Appl. Phys. 113 (2013) 083712.
[20] A.Y. Polyakov, N.B. Smirnov, A.V. Govorkov, E.A. Kozhukhova, S.J. Pearton, F.
ments for adjacent mesas, suggests that the traps in question are Ren, S.Yu. Karpov, K.D. Shcherbatchev, W. Lim, ECS J. Solid State Sci. Technol. 1
on the GaN side of the AlGaN/GaN HJ and come from the GaN buf- (2012) P152.
fer. Similar traps have been previously reported for other AlGaN/ [21] Z.-Q. Fang, B. Claflin, D.C. Look, D.S. Green, P. Ventury, J. Appl. Phys. 108 (2010)
063706.
GaN/Si, AlGaN/GaN/sapphire, AlGaN/AlN/sapphire HJs and HEMTs [22] A.Y. Polyakov, N.B. Smirnov, A.V. Govorkov, E.A. Kozhukhova, A.M. Dabiran,
[7,11–13]. On the strength of that similarity it would seem that P.P. Chow, A.M. Wowchak, S.J. Pearton, J. Appl. Phys. 107 (2010) 023708.
they can also be ascribed to oxygen-related complexes. The in- [23] A.Y. Polyakov, N.B. Smirnov, A.V. Govorkov, A.V. Markov, Q. Sun, Y. Zhang, C.D.
Yerino, T.-S. Ko, I.-H. Lee, J. Han, Mater. Sci. Eng. B166 (2010) 220.
crease of the GaN substrate resistivity and the shift of the position [24] A.Y. Polyakov, I.-H. Lee, N.B. Smirnov, A.V. Govorkov, E.A. Kozhukhova, S.J.
of the Fermi level pinning state from Ec 0.38 eV to Ec 0.6 eV after Pearton, J. Appl. Phys. 109 (2011) 123701.
oxygen treatment is attributed to the annihilation of the 0.38 eV [25] A.Y. Polyakov, In-Hwan Lee, N.B. Smirnov, A.V. Govorkov, E.A. Kozhukhova,
N.G. Kolin, A.V. Korulin, V.M. Boiko, S.J. Pearton, J. Appl. Phys. 109 (2011)
states that are possibly due to the surface damage introduced by 123703.
dry etching during mesa formation. [26] K. Arima, K. Endo, K. Yamanuchi, K. Hirose, Y. Sano, J. Phys.: Condens. Matter
23 (2011) 394202.
[27] M. Faqir, M. Bouya, N. Malbert, N. Labat, D. Carisetti, B. Lambert, G. Verzellesi,
Acknowledgments
F. Fantini, Microelectron. Reliab. 50 (2010) 1520.
[28] S. Arulkumaran, G.I. Ng, C.H. Lee, Z.H. Liu, K. Radhakrishnan, N. Dharmasa,
The work at IRM was supported in part by the Ministry of ‘Z. Sun, Solid-State Electron. 54 (2010) 1430.
Education and Science of Russian Federation through a subcontract [29] A.R. Arehart, A. Sasikumar, S. Rajan, C.D. Via, B. Poling, B. Winningham, E.R.
Heller, D. Brown, V. Rei, F. Recht, U.K. Mishra, S.A. Ringel, Solid-State Electron.
to NRNU (MIFI). The work at NRNU (MIFI), Moscow, was supported 80 (2013) 19.
by the Ministry of Education and Science of Russian Federation un- [30] A.Y. Polyakov, A.S. Usikov, B. Theys, N.B. Smirnov, A.V. Govorkov, F. Jomard,
der contract #16.523.12.3010 of June 5 2012. The work at Chonbuk N.M. Shmidt, W.V. Lundin, Solid-State Electron. 44 (2000) 1971.
[31] A.Y. Polyakov, Radiation effects in GaN, in: S.J. Pearton (Ed.), GaN and ZnO-
National University was supported by the National Research Foun- Based Materials, and Devices, Springer Series in Materials Science, Springer,
dation of Korea (NRF) funded by the Korea government (MEST) Heidelberg, 2012, pp. 251–294.
(2010-0024973). The work at KETI was partly supported by the [32] D.W. Palmer, Electronic energy levels in Group-III nitrides, in: P. Bhattacharya,
R. Fornari, H. Kamimura (Eds.), Comprehensive Semiconductor Science and
IT R&D program of MKE/KEIT (2011101050017B). Technology, vol. 4, Elsevier, Amsterdam, 2011, pp. 390–447.

References

[1] W. Johnson, E.L. Piner, GaN HEMT Technology, in: S.J. Pearton (Ed.), GaN and
ZnO-based Materials and Devices, Springer, Heidelberg, 2012, pp. 209–238.