Deep levels associated with dislocation annihilation by Al pre-seeding and silicon delta doping in GaN grown on Si(111) substrates

Article abstract of DOI:10.1002/pssa.200723481

The introduction of Si burst during the growth of GaN film on Si(111) substrate by MOCVD formed a Si_xN_y layer which leads to an effective reduction in the density of screw dislocations. The reduction is associated with bending of screw dislocations to form a square dislocation loop when neighbouring dislocations with opposite Burger's vector paired up. The concentration of electron traps E_c-E_t ~0.17-0.26 eV which is associated with screw dislocations is substantially reduced and a kink is left at the silicon rich position. The mixed-edge dislocation, however, is not annihilated by the Si_xN_y layer. Addition of TMAI burst for the AIN growth leads to a substantial reduction in trap concentration associated with the nitrogen vacancies, V_N, and antisite of nitrogen, N_Al, at E_c-E_t~0.10 eV and E_c-E _t ~ 0.60 eV respectively. This improves the quality of the subsequent layer of HT-GaN grown and is useful for device fabrication.

Full text of DOI:10.1002/pssa.200723481

phys. stat. sol. (a) 205, No. 2, 266–270 (2008) / DOI 10.1002/pssa.200723481
a
p s s
www.pss-a.com
applications and materials science
Deep levels associated
with dislocation annihilation
by Al pre-seeding and silicon delta doping
in GaN grown on Si(111) substrates
*
*, 1
1
1
1
*, 1, 2
C. B. Soh , K. Y. Zang , L. S. Wang , S. Y. Chow , and S. J. Chua
1
Institute of Materials Research and Engineering, A*STAR (Agency for Science, Technology and Research), 3 Research Link,
Singapore 117602, Singapore
2
Singapore–MIT Alliance, NUS, 4 Engineering Drive 3, Singapore 11576, Singapore
Received 18 July 2007, revised 14 September 2007, accepted 4 October 2007
Published online 11 January 2008
PACS 61.72.J–, 61.72.Lk, 68.55.Ln, 71.55.Eq, 81.15.Gh
*
*
*
*
Corresponding author: e-mail elecsj@nus.edu.sg
e-mail cb-soh@imre.a-star.edu.sg
The introduction of Si burst during the growth of GaN film and a kink is left at the silicon rich position. The mixed-edge
on Si(111) substrate by MOCVD formed a Si N layer which dislocation, however, is not annihilated by the Si N layer.
x
y
x
y
leads to an effective reduction in the density of screw disloca- Addition of TMAl burst for the AlN growth leads to a sub-
tions. The reduction is associated with bending of screw dis- stantial reduction in trap concentration associated with the ni-
locations to form a square dislocation loop when neighbour- trogen vacancies, V , and antisite of nitrogen, N , at E –E
t
N
Al
c
ing dislocations with opposite Burger’s vector paired up. The ~0.10 eV and E –E ~ 0.60 eV respectively. This improves
c
t
concentration of electron traps E –E ~0.17–0.26 eV which the quality of the subsequent layer of HT-GaN grown and is
c
t
is associated with screw dislocations is substantially reduced useful for device fabrication.
©
2008 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
1
Introduction The growth of III–V nitride het- the trap levels by deep level transient spectroscopy (DLTS)
erostructures on silicon substrate open up new frontier in would provide more information on the characteristics of
the field of Optoelectronics as it is now possible to inte- these defects. In this letter, we have investigated the influ-
grate Si electronics on the same chip. Si-based GaN LEDs ence of periodic Si delta doping on the nature of the trap
[
1, 2] are now commercially available and top-bottom elec- levels and its relation to the dislocations and other defects
trode can be fabricated due to well-confined p- and n-type in the GaN film grown on Si.
conductivity. However, due to the large lattice and thermal
mismatched between GaN and Si [3, 4], growth of GaN 2 Experimental procedures The MOCVD growth
film on Si substrate leads to generation of high dislocation was performed with an EMCORE D125 turbo disk reactor
density [5] and thermally induced cracks. The introduction on Si(111) substrate. Monosilane (SiH ) diluted in hydro-
4
of low temperature AlN multilayer [6], in-situ Si N [7, 8] gen (50 ppm) was used as Si dopants. 100 nm of AlN
x y
masking and Si delta doping (δ-doping) [9, 10] have re- buffer was first deposited at 1020 °C on Si(111) substrate
sulted in the significant improvement in the crystalline for 90 min in sample A, B and C. For the growth of the
qualities of GaN film grown on silicon. Wang et al. [10] HT-AlN buffer, the flow rate of NH and TMAl is kept at
3
has reported that the addition of silicon burst for δ-doping 0.45 mol/min and 200 sccm respectively with a pressure of
reduces the in-plane tensile stress while Contreras et al. [9] 50 Torr. For sample B and C, it has a seed layer of Al pre-
reported the phenomenon of screw dislocation reduction. deposited at 1010 °C using Trimethyl Aluminium (TMAl)
However, the electrical properties of GaN film grown on as the source gas for 6.0 s before the growth of the HT-
Si substrate have not been well addressed. Identification of AlN buffer, but not for sample A. During the growth of the
©
2008 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim

Original
Paper
phys. stat. sol. (a) 205, No. 2 (2008)
267
2
.0 µm thick HT-GaN layer for the three samples, the flow
1
4
3
3
(a)
(b)
1
.2x10
A1
Sample A
A2
A5
rate of TMGa is kept at 150 µmol/min and reactor pressure
maintained at 200 Torr with a temperature of 1010 °C.
However, for sample C, Si periodic delta doping was done
for 4 periods to form the SiN interlayers after the growth of
each 0.4 µm of HT-GaN interval. The flow of TMGa was
1
8
.0x10
A4
1
4.0x10
0
.0
1
3
3
3
9
.0x10
A2
interrupted while SiH was introduced at 0.167 µmol/min
4
Sample B
1
6
.0x10
into the reactor for 36 s at constant NH flow. The silane
3
A1
1
'
3
.0x10
A
flow was kept at 10 sccm and this gives an estimated
growth rate of 0.2 nm/s, giving a SiN interlayer of thick-
ness ~7.0 nm. After each cycle of Si δ-doping, purging
was carried out for 60 s to eliminate any memory effect in
the system and to avoid residual dopants. For sample A, no
Si-delta doping or TMA burst was carried out. Before con-
ducting DLTS measurement, C–V data was used to estab-
lish the carrier profile and to ascertain that the carrier con-
centration is much greater than the trap concentration at all
2
0.0
1
3
3
3
4x10
(
c)
A3
Sample C
A4
1
3x10
'
1
A
2x10
2
A1
13
x10
1
0
5
0
100
150
200
250
300
T(K)
16 –3
temperature. Carrier concentrations n ~ 5.0 × 10 cm ,
s
1
6
–3
16
–3
.0 × 10 cm and 8.5 × 10 cm is obtained for sample A,
4
0.20
.18
.16
.14
0.12
.10
.08
.06
0.04
.02
.00
B and C respectively.
A1
0
A2
A3
A4
A5
0
3
spectra from temperature scan from 50 K to 300 K carried
Result and discussion Figure 1 shows the DLTS
0
out for the four samples with a period width t of 2.0 s un-
w
der a quiescent reverse bias voltage of –2.0 V (to a depth
0
of ~0.5 µm) and pulsed at t = 50 ms. Typical DLTS
p
0
spectra as shown in Fig. 1(a) for the growth of undoped
GaN on Si substrate gives four prominent electron traps
labeled as A , A , A and A with energy position at
0
1
2
4
5
E –E ~ 0.10 eV, 0.17 eV, 0.45 eV and 0.60 eV. The A
c t 1
0
level would be related to that of the nitrogen vacancies as
the activation energy for these donors found from Hall
0
-6
0
-5
10
-4
10
-3
10
-2
10
-1
10
0
10
1
(
d)
measurement, E ~ 70 meV. The discrepancy in the value
D
Filling pulse, t (s)
p
is partially accounted to the error in the fitting of the
Arrhenius plot. The well-defined spectra peak, A5 at
E –E ~ 0.60 eV has been reported to be related to nitrogen
Figure 1 (online colour at: www.pss-a.com) DLTS spectra for
temperature scan performed on (a) sample A, GaN grown on Si,
c
t
antisite N , in GaN grown on sapphire [11]. In sample B, and (b) sample B, with the addition of TMAl pre-treatment before
Ga
with a seed layer of Al pre-deposited at 1010 °C by the AlN buffer growth while (c) sample C includes the addition of Si
TMA burst, it was observed that A level has been effec- δ-doping at 70 sccm. (d) DLTS peak amplitude versus the filling
5
tively quenched from the peak spectra in Fig. 1(b). There is pulsewidth ‘t ’ for the trap levels observed in Si-doped GaN.
p
Table 1 Trap parameters for the deep levels in GaN layer grown on Si determined from the Arrhenius plot.
level
A1
A2
A′
2
A3
A4
A5
sample A
E –E (eV)
c
0.10
0.17
0.45
0.60
T
2
σ (cm )
–20
14
–18
13
–18
13
–16
13
5.7 × 10
1.1 × 10
1.8 × 10
9.0 × 10
8.3 × 10
3.0 × 10
1.5 × 10
7.6 × 10
N
–3
N (cm )
T
sample B
E –E (eV)
c
0.11
0.17
0.19
T
2
σ (cm )
–19
13
–18
13
–18
13
1.1 × 10
3.2 × 10
6.5 × 10
7.5 × 10
2.5 × 10
1.2 × 10
N
–3
N (cm )
T
sample C
E –E (eV)
c
0.10
0.19
0.39
0.45
T
2
σ (cm )
–20
12
–18
12
–17
13
–17
13
8.3 × 10
5.3 × 10
5.8 × 10
5.5 × 10
5.2 × 10
3.6 × 10
1.3 × 10
1.0 × 10
N
–3
N (cm )
T
www.pss-a.com
© 2008 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim

268
C. B. Soh et al.: Deep levels associated with dislocation annihilation
also a reduction in concentration of deep level A which is feature of point defects. The relation of A and A levels to
1 4 5
14 –3
associated with nitrogen vacancies from 1.1 × 10 cm to antisite of nitrogen, NGa and N , explains its saturation fill-
Al
13 –3
.2 × 10 cm as shown in Table 1. It is speculated that ing pulse at tens of milliseconds. On the otherhand, the
3
with pre-seeding of Al adatoms on Si substrate, it enhances deep level A are related to the linear arrays of defects due
2
the crystal orientation of the AlN buffer layer, with less to the dangling bonds along the dislocation core sites,
broken bonds and also reduces the concentration of point- where charge buildup governs the capture rate. The curve
like intrinsic defects for instance the vacancies, V , V and will only saturate at longer filling pulses (t > 1 s), when
Al
N
p
its antisites, N . Complexes of [N + V ] which are domi- the trap levels are effectively filled. Further increase in fill-
Al Al Al
nant in conventional AlN layer are less likely to be formed ing pulsewidth caused all the trap levels to emit during the
12]. In the HT-AlN buffer, the reduction in N adatoms oc- capture process, which accounts for the slight reduction of
cupying the vacant Al site to give antisite of nitrogen, NAl the DLTS peak amplitude.
leads to the elimination of deep level A with better bond Figure 2 shows the XRD Omega scan at (0002) plane
carried out for the samples using Philip X-Pert HRXRD
In sample C, Si δ-doping has been carried out to form system with the diffraction peak for GaN (0002) normal-
Si N layer. It has been reported by Wang et al. that with ized to zero. A comparison of the intensity with its width
[
5
formation of Al and N adatoms in the buffer layer.
x
y
Si δ-doping, better morphology has been obtained and for sample A, B and C shows that the Full width half maxi-
there is a reduction in void defects [10]. In Fig. 1(c), it was mum (FWHM) has been effectively reduced from 0.45° to
observed that there is a significant reduction in concen- 0.29° and 0.31° with the use of the Al pre-seeding layer
tration of deep levels, A (E –E ~ 0.15–0.17 eV) and A′ and Silane burst. This indicates that the defect density in
2
c
t
2
(E –E ~ 0.19–0.23 eV) with Si-delta doping where silane the GaN layer using Al preseeding and Silane burst has
c t
flow is kept at 0.167 µmol/min. It has been reported that been reduced and the crystal quality at the symmetry plane
A and A′ are related to the decoration of O and Si along has been much improved. This agrees with the finding
2
2
N
Ga
the screw dislocations respectively [14]. This is based on from DLTS that density of deep level A and A′ related to
2 2
the high thermal stability of the O and Si which can the decoration of O and Si along the screw-mixed dislo-
N Ga N Ga
only dissociate from the dangling bonds of screw disloca- cations have been reduced using Silane burst and Al pre-
tions with an annealing temperature of higher than 850 °C. seeding as (0002) symmetrical plane determines the de-
The deep levels also displayed extended defect-like behav- fects related to screw dislocations. The reduction in pits
ior such as asymmetrical broadening with increasing filling can be seen in the AFM images as reported by Zang et al.
pulse and logarithmic capture kinetic behavior [13–17]. [18] using the same technique but on different batch of
Similar observation was noted for this deep level A in samples.
2
our GaN sample grown on Si substrate. This implies the
The analysis of the deep level defects in the samples
less availability of the dangling bonds from the threading suggests that the delta doping is effective in reducing the
dislocations to form bonds with SiGa and O antisite. On density of threading dislocations. The effectiveness in the
N
the other hand, the higher SiH flow introduces a deep reduction of dislocations density is further supported by
4
level, A (E –E ~ 0.36–0.39 eV) of trap concentration comparing the weak beam TEM image taken with diffrac-
3 c t
13
–3
~
3.6 × 10 cm in the GaN layer as illustrated in Fig. 1(c). tion vector g = [0002] with images at diffraction vector
This deep level is believed to be related to that of Si point g = [0110] along 〈1120〉 zone axis. In general, threading
defects as it displayed an exponential capture kinetic be-
havior.
To further study the nature of the deep levels in Si-
doped GaN, the dependence of the deep levels on filling
pulse time ‘t ’ was monitored. A saturation pulsewidth ‘t ’
40000
Omega scan at (0002) plane
30000
p
p
of 1 s, 30 ms and 60 ms is obtained for A , A and A lev-
2 4 5
els, respectively, as compared to 20 µs for the A level,
1
which is typical for a point-like defect [16]. The plots
20000
10000
0
of peak height, ∆C, versus the filling pulsewidth, t , in
p
Fig. 1(d) show that the magnitudes of A levels increase
2
o
FWHM=0.31
sample C
sample B
sample A
linearly with the logarithm of filling pulsewidth up to 1 s.
This is an indication of the nature of trap levels associated
with the extended defects [14–16]. The A , A and A level
o
FWHM=0.29
3
4
5
o
FWHM=0.45
-0.8 -0.6 -0.4 -0.2
has a shorter saturation pulse time and exhibit a logarith-
mic capture kinetic behavior only for lower pulsewidth. Its
C starts to decrease with further increase in t as the traps
0.0
0.2
0.4
0.6
0.8
∆
p
o
Degree ( )
are filled and re-emitted. We can attribute the deep levels
A to Si dopant, since it is not observed in sample A and B.
3
Figure 2 (online colour at: www.pss-a.com) High resolution
XRD scan at (0002) plane for sample A, B and C with its x-axis
normalized to zero.
This accounts for its anomalous capture kinetic behavior
and its low thermal stability as reported [13], which is a
©
2008 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.pss-a.com

Original
Paper
phys. stat. sol. (a) 205, No. 2 (2008)
269
dislocations which includes the screw, edge and mixed dis- density with delta doping. A number of voids (or pits) were
locations, with Burgers vectors b = [0001], b = 1/3[1120] observed at the region where silane burst was carried out.
and b = 1/3[1123] respectively, have been observed in epi- It has further been confirmed by electron energy loss spec-
taxially grown GaN. With diffraction condition, g = [0002] troscopy (EELS) that Si has been incorporated in the side-
as in Fig. 3(a), only screw and mixed dislocations are visi- walls of these voids [18]. The different structures of these
ble. It was observed that at Silane burst (as shown by the Si-rich precipitates would have account for the contrast as
dotted line), the silicon atoms pin the surface step edge and observed in the TEM image. With incorporation of SiN rich
x
forced the threading dislocation line to bend over the basal regions, large tensile stress field is developed [19] which
plane (as indicated by K in the figure). The screw disloca- leads to the dislocation bending as observed in Fig. 3(a).
tions can be paired to form a dislocation loop (L) or bends
4
Conclusions In summary, delta doping with silicon
back to form a kink (K). This accounts for the substantial
reduction of screw and mixed dislocations with delta dop-
ing and correlates well to the deep level analysis, where
trap concentration of A and A′ decreases with Si-delta
is an effective method for the reduction of the threading
dislocation density (especially screw dislocations) in GaN
epitaxy. The density of the deep level traps A and A′ has
2
2
2
2
been much reduced with Si delta doping. TEM images
show that propagation of screw dislocation has been ham-
pered with bending and kink formation. This accounts for
the reduction in the density of trap levels with energy posi-
tion at E –E ~ 0.17 eV. The reduction in the FWHM of
doping. With diffraction condition, g = [0110] as in
Fig. 3(b), where only edge dislocations is shown, its den-
sity is found to be much less as compared to the screw dis-
locations. The edge dislocations, does not show any bend-
ing or looping and there is no substantial reduction in its
c
t
XRD Omega scan at (0002) plane for the Si delta doped
sample with Al pre-seeding is an indication of the reduc-
tion in defect density in the GaN layer. However, the pin-
ning by the silicon impurities of the surface lattice steps
associated with screw dislocations generated a high
concentration of point-like defects of Si, and leads to a
substantial increase in the concentration of related traps at
E –E ~ 0.36–0.42 eV. With pre-seeding of Al adatoms on
(
a )A lN
g = [0002]
K
c
t
Si substrate through TMAl burst, it enhances the crystal
orientation of the AlN buffer layer and reduces the concen-
tration of point-like intrinsic defects, the vacancies (VN)
voids
and the antisites of nitrogen (N ) with energy position at
Al
E –E ~ 0.10 eV and E –E ~ 0.60 eV, respectively.
c t c t
0
.5 µm
References
[
1] J. W. Wang, A. Lunev, G. Simin, A. Chitnis, M. Shatalov,
M. A. Kahn, J. E. Van Nostrand, and R. Gaska, Appl. Phys.
Lett. 76, 273 (2000).
(
b)
g = [0110]
[
2] T. Egawa, B. Zhang, N. Nishikawa, H. Ishikawa, T. Jimbo,
and M. Umeno, J. Appl. Phys. 91, 528 (2000).
[
3] H. Ishikawa, G. Y. Zhao, N. Nakada, T. Egawa, T. Soga,
T. Jimbo, and M. Umeno, phys. stat. sol. (a) 176, 599 (1999).
voids
[4] A. Watanabe, T. Takeuchi, K. Hirosawa, H. Amano,
K. Hiramatsu, and I. Akasaki, J. Cryst. Growth 128, 391
1993).
5] S. Tanaka, Y. Kawaguchi, and N. Sawaki, Appl. Phys. Lett.
6, 2701 (2000).
(
[
[
[
7
6] A. Dadgar, J. Bläsing, A. Diez, A. Alam, M. Heuken, and
A. Krost, Jpn. J. Appl. Phys. Part 2, 39, L1183 (2000).
0
.5 µm
7] K. J. Lee, E. H. Shin, and K. Y. Lim, Appl. Phys. Lett. 85,
1502 (2004).
Figure 3 Weak-beam TEM images showing the effect of Si
δ-doping for sample C with diffraction condition at g = [0002] for
Fig. 3(a); g = [0110] for Fig. 3(b). The images show (a) disloca-
tions with screw components; Burgers vectors b = [0002] and
[8] A. Dadgar, M. Poschenrieder, J. Bläsing, K. Fehse, A. Diez,
and A. Krost, Appl. Phys. Lett. 80, 3670 (2002).
[9] O. Contreras, F. A. Ponce, J. Christen, A. Dadgar, and
A. Krost, Appl. Phys. Lett. 81, 4712 (2002).
b = 1/3[1123], (b) dislocations with edge component, particularly [10] L. S. Wang, K. Y. Zang, S. Tripathy, and S. J. Chua, Appl.
b = 1/3[1120]. The dotted line shows the point where Si delta
Phys. Lett. 85, 5881 (2004).
doping occurs and caused dislocations bending with kinks forma- [11] H. K. Cho, C. S. Kim, and C.-H. Hong, J. Appl. Phys. 94,
tion.
1485 (2003).
www.pss-a.com
© 2008 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim

270
C. B. Soh et al.: Deep levels associated with dislocation annihilation
[
[
[
12] N. Yu. Arutyunov, A. V. Mikhailin, V. Yu. Davydov, [15] D. Cavalcoli, A. Cavallini, and E. Gombia, Phys. Rev. B
V. V. Emtsev, G. A. Oganesyan, and E. E. Haller, Semicon-
ductors 36, 1106 (2002).
56(23), 14890 (1997).
[16] P. N. Grillot, S. A. Ringel, E. A. Fitzgerald, G. P. Watson,
and Y. H. Xie, J. Appl. Phys. 77(7), 3248 (1995).
13] C. B. Soh, S. J. Chua, H. F. Lim, D. Z. Chi, W. Liu,
and S. Tripathy, J. Phys.: Condens. Matter 16, 6305 [17] P. Omling, L. Samuelson, and H. G. Grimmeiss, J. Appl.
2004). Phys. 54, 5117 (1983).
14] C. B. Soh, S. J. Chua, H. F. Lim, D. Z. Chi, S. Tripathy, and [18] K. Y. Zang, Y. D. Wang, L. S. Wang, S. Y. Chow, and
(
W. Liu, J. Appl. Phys. 96, 1341 (2004).
S. J. Chua, J. Appl. Phys. 101, 093502 (2007).
19] F. Yun et al., J. Appl. Phys. 98, 123502 (2005).
[
©
2008 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.pss-a.com