close

Вход

Забыли?

вход по аккаунту

?

Microwave power aluminum gallium nitride/gallium nitride heterojunction field effect transistor for X-band applications

код для вставкиСкачать
UNIVERSITY OF CALIFORNIA
Los Angeles
Microwave Power AlGaN/GaN Heteroj unction
Field Effect Transistor for X-band Applications
A dissertation submitted in partial satisfaction of the
requirements for the degree of Doctor of Philosophy in
Electrical Engineering
by
Shujun Cai
2002
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
UMI Number: 3088996
UMI
UMI Microform 3088996
Copyright 2003 by ProQuest Information and Learning Company.
All rights reserved. This microform edition is protected against
unauthorized copying under Title 17, United States Code.
ProQuest Information and Learning Company
300 North Zeeb Road
P.O. Box 1346
Ann Arbor, Ml 48106-1346
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The dissertation o f Shujun Cai is approved.
Frank M. Chang
Dee-Son Pan
Ya-Mong Xie [
Kan]
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
/
Dedicated to
My Wife, Yanling
&
My Daughters, XueQing and Maylynn
iii
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Contents
Page
List of Figures
v"
List of Tables
xiii
Acknowledgments
xiv
Vita
xv
Abstract
xix
Chapter
1. Introduction
1
1.1. Why GaN based AlGaN/GaN HFET?
1
1.2. Research background
6
1.3. Synopsis
9
Chapter Bibliography
10
2. AlGaN/GaN Basic characteristics
2.1. Basic material characteristics for AlGaN/GaN system
13
14
2.1.1. Piezoelectric and spontaneous polarization effects
14
2.1.2. Conduction band discontinuity
20
2.1.3. Mobility
23
2.2. AlGaN/GaN HFET device structure design
25
2.3. Simulation
31
iv
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
2.3.1. Role of piezoelectric and spontaneous polarization induced
32
charges
2.3.2. Surface compensation
35
2.3.3. A1 content effect to the sheet charge
37
2.3.4. Spacer layer
39
2.3.5. Total sheet charge available in the channel
40
2.4. Summary
42
Chapter Bibliography
43
3. Experiments and device fabrication
3.1. Experiments
45
3.1.1 Hall measurement from different A1 contentstructures
45
3.1.2 Transport study
48
3.2. Device fabrication techniques
4
45
51
3.2.1. Isolation
51
3.2.2. Schottky contact
54
3.2.3. Ohmic contact
56
3.3. Summary
64
Chapter Bibliography
65
Basic AlGaN/GaN HFETs
67
4.1. Testing AlGaN/GaN HFETs structure onconductive SiC substrates
67
4.2. AlGaN/GaN HFET with 25% A1 content on SI SiC substrate
75
4.2.1.
Device DC performance
77
V
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
4.2.2. High temperature performance
79
4.2.3. RF performance
81
4.3. Device with 40% A1 content on SI SiC substrate
83
4.4. Summary
87
5. High performance AlGaN/GaN HFETs
88
5.1. Gamma gate process
89
5.2. Device performance
94
5.3. Simulation and discussion
98
5.3.1. Gamma shape effect
101
5.3.2. Dielectrics effect below the gamma gate
103
5.3.3. Effect o f the top layer doping and thickness
104
5.4. Summary
106
Chapter Bibliography
107
6. Proton irradiation study
108
6.1. Experimental details
108
6.2. Results and discussion
110
6.3. Summary
116
Chapter Bibliography
117
7. Conclusion and suggested future work
7.1
Conclusion
118
118
7.2 Suggested future work
120
Appendices
Appendix I
Appendix II
121
121
124
vi
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
List of Figures
Figure 1.1 Breakdown voltage vs cutoff frequency for InP and GaAs based
devices.
Figure 2.1 (a) Schematic of the wurzite structure with Ga-polarity for AlN/GaN
heterostructure. Ppe stands for the PZ polarization, and Psp stands for
the SP polarization, (b) Polarization induced 2DEG is shown at the
interface.
Figure 2.2 Calculated piezoelectric induced and spontaneous polarization induced
bound charge at the AlGaN/GaN interface vs A1 content.
Figure 2.3 Schematic of AlGaN/GaN band diagram.
Figure 2.4 Calculated sheet charge density of 2DEG vs A1 content for HFET
structure without gate due to AEC, for Nd=5xl0 17 cm '3, 5 x l0 18 cm '3
and 5 x l0 19 cm ' 3 cases ( for undoped spacer of d; = 30 A, 2DEG
mean distance to the interface Ad=80 A and correction 5=25 meV).
Figure 2.5 Total 2DEG density (the sum of SP & PZ effect induced and
conduction band discontinuity caused sheet charges) vs A1 content
for AlGaN/GaN structure. For 30 % A1 content, 8x10 *8 cm "3 doping,
13
-2
2 x 1 0 cm sheet charge at the heterojunction can be induced.
Figure 2.6 Calculated mobility for AlGaN/GaN material vs temperature. 2DEG
mobility is obtained by ignoring the impurity scattering. Parameters
used: m =0.228me, Energy gap: Eg = 3.4 eV, Polar optical phonon
energy: Epo = 91.2 meV, Piezo constant: epz = 0.5 C / itT , Elastic
constants: C l = 2.6 5 x l0 u N/m2, Cp = 4.42 xlO 10 N/m 2 and
deformation potential, Eds= 8.3 eV .
Figure 2.7 Critical thickness of AlxGa i_x N layer as a function of aluminum mole
fraction: superlattice (solid line), SIS structure (dashed line)
Figure 2.8 Undoped semiconductor energy band diagram: (a) Non polarized
Semiconductor, (b) Strongly polarized semiconductor with surface
inversion.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
AlxGai_xN layer design guideline: the surface inversion on top of
AlGaN layer caused by strong PZ and SP effect must avoid.
28
Figure 2.10 The maximum cap layer thickness to avoid the surface inversion for
two cases: 1) fully relaxed (SP only); 2) fully strained (SP + PZ).
Inset is the electric field intensity for the two corresponding cases.
Assuming no surface charge compensation and cap layer was heavily
doped.
29
Figure 2.11 Model for PZ and SP effects. Charge density at each interface is
calculated for 25% A1 content based on the analysis in previous
chapter. The sheet charge, a l stands for trapped negative charges
while o2 and o3 for positive trapped charges (assuming uniformly
distributed in the 10 A region).
32
Figure 2.12 Simulated sheet charge versus cap layer thickness for the standard
structure. When the cap layer thickness is too thin, no channel sheet
charge is present.
33
Figure 2.13 Comparison of our simulation results with the experimental results
from Koley et al. (cap layer undoped). Inset showing the minimum
19
9
thickness of the caplayer to have a 2 x 1 0 cm" sheet charge in the
channel for a structure of 25% Al content, 30 A undoped spacer
layer, and surface uncompensated .
34
Figure 2.14 Channel, parallel, hole and total sheet charge densities as the
function of compensation ratio for the standard structure with a cap
doping density of lxl O 19 cm ' 3 . We notice the hole will be present
on the top surface if there is a large a l resulting from the PZ and SP
effects and if the surface is not compensated.
36
Figure 2.15 Channel sheet charge vs the surface compensation ratio for different
doping densities in the cap layer. Lower doping has a more sensitive
surface. Different doping cases all saturated at around the same sheet
charge level (~ 1.6 xlO 13 cm "2 ) indicating the PZ and SP effects
dominate the sheet charge.
37
Figure 2.16 Maximum available channel sheet charge vs the Al content. The
charge density increases due to the PZ and SP contributions, almost
independent of the doping density of the AlGaN layer. Here we
assumed that all the surface charges were fully compensated for
different dopings in the cap layer. (Parallel conduction was not
included).
38
Figure 2.9
viii
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 2.17 Sheet charge density as a function o f the space layer thickness. Here
only the discontinuity o f the conduction band offset was taken into
account ( no PZ and SP effects and no gate metal were used in
calculations).
39
Figure 2.18 Sheet charge versus doping in the cap layer when the surface is
uncompensated.
41
Figure 2.19 Sheet charge versus doping in the cap layer when the surface is fully
compensated. The sheet channel charge is almost independent to the
doping in the cap layer, dominated by charges induced by the PZ and
SP effects.
41
Figure 3.1 (a) Material structure designed for PZ and SP experiments. Three
samples prepared with different Al contents: x=5%, 15%, 25%. (b)
showing the hall bars with Pd/Au as gate metal for transport study.
46
Figure 3.2 Hall measurement results from three samples: at 10K, mobility o f
4000 cm2/Vs and concentration o f 2 .1 x l0 12 cm '2 for 5% Al content
sample, 3900 cm2/Vs and 7.1 xlO 12 cm ’2 for 15% sample , and 1700
2
13
2
cm /Vs and 1.04 xlO cm' for 25% sample, respectively.
47
Figure 3.3 Typical magnetoresistivity vs magnetic field B. Data o f sample A at
0.42K in (a) and o f sample B at 0.54K in (b). The integer quantum
Hall effect is clearly seen in both figures and filling factors (v) o f the
corresponding quantum Hall states are labeled.
49
The logarithmic o f [Apxx/T)[l-exp(-2xr)] is plotted against
temperature
T for the temperature dependence o f pxx at four
different values o f magnetic field corresponding to four extrema in
the low-field regime. An average o f 0.228 me was obtained for
effective electron mass in AlGaN/GaN system.
50
Figure
3.4
Figure 3.5 AFM picture for ECR etched mesa.
53
Figure 3.6 Isolation resistance against high temperature storage (HTS) time
showing the degradation of the isolation performance, indicating that
this isolation method was not idea for GaN HEMT for long term
reliability performance.
54
Figure 3.7 Forward Schottky characteristics for Au/Pt/AlGaN/GaN junction in
log scale (semilog scale shown as in s e t). The gate size is o f 1 pm x
2 0 pm.
55
ix
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 3.8 Normalized transconductance versus contact resistivity for various
channel resistance.
57
Figure 3.9 Normalized extrinsic gm vs contact resistance for various source and
gate separation distance
58
Figure 3.10 1/RC vs Nsp. Ohmic contact was easier to form for higher Ns p
product epi layers.
60
Figure 3.11 Picture of transmission line patterns.
61
Figure 3.12 Contact resistance vs RTP annealing time for different samples.
62
Figure 3.13 Contact resistance vs elapsed time after surface treatment. The
contact performance degraded after the samples exposed in air for
longer time before metal deposition.
63
Figure 3.14 Ohmic specific resistance measured by transmission line(TML) for
p i sample. For this sample, a record ohmic contact number of
4.42x1 O' 7 12.cm 2 was achieved. The sample had 30% Al content,
and a cap layer doped with 5x10 cm’ .
64
Figure 4.1 Testing structure used to fabricate devices. Four samples were
prepared for different Al contents: x=0.15, 0.2, 0.3 and 0.4.
68
Figure 4.2 Hall measurement results showing that sheet concentration increases
as Al content increases.
69
Figure 4.3 Fabricated AlGaN/GaN HFET device.
69
IQ
O
DC characteristics for a test device with 15 % Al content, showing
the peak g m of 80 mS/mm and Ids of 140 mA/mm.
71
Figure 4.5 DC characteristics for the test device with 20 % Al content, showing
the peak g m of 170 mS/mm and Ids of 500 mS/mm.
72
Figure 4.6 DC characteristics for the test device with 30 % Al content, showing
the peak g m of 190 mS/mm and Ids of 700 mS/mm.
73
Figure 4.7 DC characteristics for the test device with 40 % Al content, showing
the peak g m of 195 mS/mm and Ids of 850 mS/mm.
74
Figure 4.8 Material structure for our basic HFET study (sample 817d).
76
Figure 4.4
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 4.9 Mercury CV profiling for the structure shown above, indicating the
peak 2DEG located at 200 A where the nominal AlGaN/GaN
interface is located.
76
Figure 4.10 Breakdown voltage of the device between the gate and the drain.
(Sample 817d, Lg= l pm, W g=100 pm, the distance between gate and
drain is 1 pm)
77
Figure 4.11 On-wafer measured DC output IV characteristics for 817d sample
(gate dimension: 1 pm by 2 0 pm).
78
Figure 4.12 On-wafer measured transfer curve and transconductance for 817d
sample. A 800mA/mm current density and a 165 mS/mm peak
transconductance were obtained, (gate dimension: 1 pm by 2 0 pm)
78
Figure 4.13(a) High temperature IV characteristics taken at 3 1 1°C.
79
Figure 4.13(b) High temperature IV characteristics taken at 592°C.
80
Figure 4.14 High temperature saturation current(IdSS) and transconduction (gm) of
the AlGaN/GaN HFET.
80
Figure 4.15 The cutoff and maximum frequencies of the sample: 10 GHz and 30
GHz, respectively.
81
Figure 4.16 Performance of a power amplifier made from the 5mm single device.
A CW power level of 8 W at 9 GHz CW was obtained.
82
Figure 4.17 Picture of the assembled amplifier.
83
Figure 4.18 IV characteristics for sample F2 (gate size: 0.8 pm x 20 pm).
84
Figure 4.19 IV characteristics for sample F2 with a Vds bias of 40 V.
85
Figure 4.20 High temperature performance of F2 device (gate dimension: 0.8 pm
by 2 0 pm).
86
Figure 4.21 DC characteristics for F2 sample with a total gate width of 2.56 mm.
86
Figure 4.22 Power amplifier data measured from a 1.28 mm device (sample F2).
87
Figure 5.1 Gamma gate fabrication process.
90
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 5.2 Gamma gate length is defined by the tilt angle which can be easily
controlled.
Figure 5.3 SEM picture of the gamma gate illustrated in Fig 5 .1(d).
Figure 5.4 Negligible leakage current of the SiCE dielectrics below the gamma
gate plate(IV was measured from a capacitance with an area of 2 cm
by 2 cm).
Figure 5.5 Schematic of a gamma gate device with optimized material structure
for sample 818A.
Figure 5.6 Gamma gate device DC characteristics(gate width: 100 pm).
Figure 5.7 Breakdown voltage of the device vs inverse of temperature.
Figure 5.8(a) RF performance of a normal gate device(Fg = 1 pm).
Figure 5.8(b) RF performance of a gamma gate device(Fg = 0.3 pm).
Figure 5.9 Simulation results of the potential and electrical field distributions in
the channel for gamma gate and normal gate GaN HFET when Vdg =
50 V. Dash line is for conventional device and solid like is for the
gamma gate device.
Figure 5.10 Peak electric field distribution in the barrier layer vs the gamma gate
(field plate) length (Log). (Based on structure shown in Fig.5.5 and
Vds = 50 V)
Figure 5.11 Peak electric field distribution in the barrier layer vs the gamma gate
height. (Based on structure shown in Fig.5.5 and Vds = 50 V)
Figure 5.12 Peak electric field distribution in the barrier layer for different
dielectrics under the gamma gate. (Based on structure shown in
Fig.5.5 and Vds = 50 V)
Figure 5.13 Peak electric field distribution in the barrier layer as a function of
doping in the barrier layer. (Based on structure shown in Fig.5.5 and
Vds = 50 V)
Figure 5.14 Peak electric field distribution in the barrier layer as a function of the
barrier layer thickness. (Based on structure shown in Fig.5.5 and Vds
= 50 V)
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 6.1 Device schematic under proton irradiation study.
Figure 6.2 Comparison of the I-V curves of the HFET before (a) and after (b)
proton irradiation. The HFET gate width used is 50 pm.
Figure 6.3 Transconductance and saturation current of the HFET measured at
V gs= 1 0 V , and V g= 0 .5 V versus annealing temperature. Before
irradiation, g m0= 8 0 mS/mm, I<iso=260 mA/mm.
Figure 6.4 Results of Hall measurement for the HFET structure after irradiation
and annealing. Before irradiation, Hall results are flo=705 cirT/Vs
and Nso=1.55xl0 13cm"2.
Figure 6.5
Typical Raman spectrum of an irradiated sample at room
temperature. Marked P I, P2 and P3 in the insert are the fitted
positions of the E2 peaks from different regions in the sample.
Figure 6.6
Raman shifts of the fitted E2 peak positions vs. annealing
temperature of the irradiated sample. Before irradiation, peak 1 in
Figure 5 was at 562.06 cm’1, the peak shifted to 561.17 cm ’1 after
irradiation.
List of Tables
Table 1.1 Basic properties for selected semiconductor materials.
Table 1.2 Milestones for the development of AlGaN/GaN HFET.
Table 2.1 Spontaneous polarization (Psp) and piezoelectric constants (e 33 and e 3 i)
for several materials with wurtzite structure.
Table 3.1 Our early device results showing poor performance due to poor
contacts.
Table 4.1 Summary of the DC results with different Al contents.
Table 5.1 Breakdown voltage of the gamma gate device.
xiii
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Acknowledgments
Back to school becoming a PhD student at an ageof mid-30’s was a tough
decision for me, although it had been a dream to be entitled as a “doctor” since I was very
young. Owing to my advisor, Prof. W ang’s encouragement, and constant support, I thus
can accomplish my PhD study. I am forever grateful
for what I have learned from him.
Indeed, “quality is not a product but a habit”.
I also like to express my sincere thanks to Prof. Frank Chang, Prof. D. S. Pan and
Prof. Y.H. Xie for their kind help and precious time serving on my PhD committee.
I would like to acknowledge that many colleagues have been worked with me for
the project. They have been offering helpful insight discussions and contributions to this
dissertation.
Special thanks goes to Dr. C.P. Wen, Faiz Rahim, Jiang Li, Richard Li,
Lauren Wong, Kevin Wang, Hui Feng, Ruigang Li, Sung Choi and all other DRL
members. It has been a great time to work together in such a great group. In particular,
Dr. YaoHui Zhang and Jianlin Liu reviewed the draft version and provided substantial
inputs and comments.
There are many others outside who helped our project and me along the way.
Thanks to Dr. Joan Redwing, Eddie Pinner from ATMI, who grew samples for us.
Finally, I would like to express my great appreciation to my wife - Yanling Chen
for her invaluable direct support in the research, her understanding and constant love to
help me through all the difficulties in the past years.
xiv
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
VITA
Nov. 1964
Bom in Hunan, P.R. China
1981 ----- 1985
BS in Physics Department, Wuhan University, Wuhan, China
1986 ----- 1988
MS in Semiconductor Engineering, Hebei Semiconductor Research
Institute(HSRI), P.R. China
1989 ----- 1996
Senior engineer in HSRI
1996 ----- 1998
Visiting scholar in Electrical Engineering Department, UCLA
1998 — present
PhD in Electrical Engineering, University o f California at Los
Angeles
PUBLICATIONS AND PRESENTATIONS
1) S.J.Cai, R. Faez and K. L. Wang, “ understanding AlGaN/GaN HFET through
simulation” submitted to IEEE-T ED, 2002
2) Chung, Y.; Cai, S.; Lee, W.; Lin, Y.; Wen, C.P.; Wang, K.L.; Itoh, T. High power
wideband AlGaN/GaN HEMT feedback amplifier module with drain and feedback
loop inductances. Electronics Letters, vol.37, (no. 19), IEE, 13 Sept. 2001. p. 1199200. 4 references.
3) S.Y.Lee, B.A.Cetiner, H.Torpi, S.J.Cai, J.Li, K.Alt, Y.L. Chen, C.P. Wen, K.L.
Wang, and T.Itoh “An X-Band GaN HEMT Power Amplifier Design Using an
Artificial Neural Network Modeling Technique” IEEE Trans. Electron Devices,
March 2001, Vol. 48 No.3. pp495-502.
4) Li, J.; Cai, S.J.; Pan, G.Z.; Chen, Y.L.; Wen, C.P.; Wang, K.L. High breakdown
voltage GaN HFET with field plate. Electronics Letters, vol.37, (no.3), IEE, 1 Feb.
2001. p. 196-7. 6 references.
5) S.J. Cai, Y.S. Tang, R. Li, Y.Y. Wei and K.L. Wang "Annealing Behavior o f A
Irradiated AlGaN/GaN HEMT" IEEE Trans. Electron Devices. Feb. 2000, vol. 47,
No.2
XV
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
6)
R. Li, S.J. Cai, L. Wong, Y. Chen, K.L. Wang etc "An AlGaN/GaN Undoped
Channel Heterostructure Field Effect Transistor with Fmax o f 107 GHz" IEEE
Electron Devices letter. July 1999 vol.20 No.7 pp323-6
7) S.J. Cai, R. Li, Y.L. Chen, L. Wong, W.G. Wu, S.G. Thomas and K.L. Wang "High
Performance AlGaN/GaN HEMT with Improved Ohmic Contacts" IEE Electronics
Letters 26th November 1998 Vol.34 N o.24 pp2354-6
8)
J.L. Liu, S.J. Cai, G. Jin and K.L. Wang " growth of Si whiskers on Au/Si (111)
substrate by gas source molecular beam epitaxy(MBE)" Journal o f Crystal Growth,
200(1999) p p l0 6 -l 11
9) Balandin, A.; Morozov, S.; Wijeratne, G.; Cai, S.J.; Li, R.; Li, J.; Wang, K.L.;
Viswanathan, C.R.; Dubrovskii, Yu. Effect o f channel doping on the low-frequency
noise in GaN/AlGaN heterostructure field-effect transistors. A pplied Physics Letters,
vol.75, (no.14), AIP, 4 Oct. 1999. p.2064-6. 10 references.
10) A. Balandin, S. Morozov, S. Cai, R. Li, K. L. Wang, G. Wijeratne, and C.R.
Viswanathan, "Low Flicker-Noise GaN/AlGaN Heterostructure Field Transistors for
Microwave Communications" IEEE Trans, on Microwave Theory and Techniques,
August 1999 Vol. 47, no .8 ppl413-8
11) A. Balandin, S.J. Cai, R. Li, K.L. Wang, V.R. Rao and C.R. Viswanathan "Flicker
Noise in GaN/AlGaN N Doped Channel Heterostructure Field Effect Transistors"
IEEE Electron Device Letters, December. 1998 V ol.19 N o.12 pp475-8
12) L.W. Wong, S.J. Cai, R. Li, Kang Wang, H.W. Jiang and Mary Chen" Magnetotransport Study on the two-dimensional electron gas in AlGaN/GaN Heterostructures"
Applied Physics Letters, Vol.73, N o.10, sept.1998, ppl391-3
13) Y.S. Tang, S.J. Cai, G. Jin, K.L. Wang, H.M. Soyes, B.S. Dunn " Direct MBE
growth o f SiGe dots on ordered mesoporous glass-coated Si substrate" Thin Solid
Films 321(1998) pp76-80
14) Y.S. Tang, S.J. Cai, G. Jin, J. Duan and K.L. Wang " SiGe quantum dots prepared
on an ordered mesoporous silica coated Si substrate" Applied Physics Letters 71(17)
pp2448-2450, Oct. 1997
15) J.L. Liu, S.J. Cai, G. Jin and K.L. Wang " Wire-like growth o f Si on Au/Si (111)
substrate by gas source MBE” Electrochemical and solid-state Letters, Vol. 1(4),
Oct.1998 ppl88-90
16) S.J. Cai "Recent Progress in Si Power Transistors" Electronics Review, No. 6 , 1996
xvi
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
17) S.J. Cai "Design of the Ballast Resistance for Silicon Microwave Power Transistor"
Semiconductor Information, pp.11-15, No.4,1994
18) S.J. Cai "Study on S-Band 100W Pulsed Power Transistor" Proceedings o f the
Eighth National Conference on IC and Si Materials,pp449-550,Oct.l993
19) Q. He, C. Wang, S.J. Cai "Study on Inter-matching And Power Combination for Si
Microwave Power Transistor" Semiconductor Information, p p .14-21 N o.3,1992
20) S.J. Cai "Study on 2.3GHz P-1=1.5W Linear Power Transistor" Semiconductor
Information, pp.85-89 No.2,1990
21) S.J. Cai "Study o f a High Voltage Radiation Hardened Power Transistor-SEBISIT"
Research & Progress o f Solid State Electronics, pp330-333, No.4,1988
22) S.J. Cai, Jiang Li, Y. L. Chen, Sang Lee, C.P. Wen and K. L. Wang “ X-band
AlGaN/GaN HEMT mini-module with 8 W output” IEEE Topical Workshop on
Power Amplifiers for Wireless Communications”. Sept. 11-12, 2000 San Diego,
pp77-78
23) S. Y. Lee, B. A. Cetiner, S. J. Cai, J. Li, K. Alt, Y. L. Chen, C. P. Wen, K. L. Wang,
and T. Itoh “An X-band GaN HEMT Power Amplifier Design Using an Artificial
Neural Network Modeling Technique” IEEE Topical Workshop on Power
Amplifiers for Wireless Communications”. Sept. 11-12, 2000 San Diego ,pp 60-62
24) Liu, J.L.; Cai, S.J.; Jin, G.L.; Tang, Y.S.; Wang, K.L. Gas-source MBE growth of
freestanding Si nano-wires on Au/Si substrate. Superlattices and Microstructures,
vol.25, (no. 1-2), (11th International Conference on Superlattices, Microstructures and
Microdevices, 1998, Hurgada, Egypt, 27-31 July 1998.) Academic Press, 1999.
p.477-9. 5 references.
25) S.J. Cai, R. Li, Y.L. Chen, L. Wong and K.L. Wang "High Performance AlGaN/GaN
HEMT with a cutoff frequency o f 60 GHz" Symposium o f "the Leading Edge in
Southern California Solid State Research" Sept.23, 1998
26) Tang, Y.S.; Cai, S.; Wang, D.; Jin, G.; Duan, J.; Wang, K.L.; Soyez, H.M.; Dunn,
B.S. (Edited by: Polman, A.; Coffa, S.; Soref, R.) Control o f sizes and optical
emission o f SiGe quantum dots prepared on ordered mesoporous silica coated Si
wafer. (Materials and Devices for Silicon-Based Optoelectronics. Symposium,
Boston, MA, USA, 1-3 Dec. 1997.) Warrendale, PA, USA: Mater. Res. Soc, 1998.
p.255-60. xiii+409 pp. 10 references.
27) S.J. Cai "Study on S-Band 100W Pulsed Power Transistor" Proceedings o f the
Eighth National Conference on IC and Si Materials,pp449-550,Oct.l993
x v ii
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
28) S.J. C ai "Effect o f Radiation on the Silicon Dioxide and Its Interface" Proceedings of
the First National Conference on Reliability and Radiation Hardening, March, 1990
xviii
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
ABSTRACT OF THE DISSERTATION
Microwave Power AlGaN/GaN Heterojunction Field Effect Transistor
for X-band Applications
by
Shujun Cai
Doctor of Philosophy in Electrical Engineering
University of California, Los Angeles, 2002
Professor Kang L. Wang, Chair
GaN material has been considered in recent years an attractive candidate for
microwave power applications owing to its wideband gap, high saturation velocity and
strong piezo-electric (PZ) and spontaneous polarization (SP) effects. AlGaN/GaN
Heterojunction Field Effect Transistor (HFET) is chosen to overcome the disadvantage of
low mobility in wide bandgap materials so that both high power and high speed are
feasible.
Analysis and simulation are performed to understand the enhancement of sheet
charge density due to the PZ and SP effects in the AlGaN/GaN material system. Major
factors affecting the sheet channel charge density are discussed. To verify the PZ and SP
charge effects, testing structures of AlGaN/GaN with various Al contents for Hall
measurement are then fabricated. Results support our analysis.
xix
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
GaN-based HFET devices with 25% Al content are fabricated after solving
process issues. An external transconductance of 200 mS/mm, a saturation current density
of 800 mA/mm and a breakdown voltage of 40 V to 50 V are achieved. A CW power
amplifier with the output of
8
W at 9 GHz is achieved from a single 5 mm AlGaN/GaN
HFET device.
A novel process, referred to as Gamma gate process, is developed to realize high
breakdown performance as well as small gate length. As a result, a 0.3 pm gate length
device with an integrated field plate is fabricated using
1
pm conventional optical
lithograph techniques. Improvements of breakdown voltage and RF performance by a
factor of over
2
have been achieved.
High temperature storage and measurement show that the AlGaN/GaN HFET
devices can survive at an environment temperature to close to 600 °C. The devices also
survive after exposing to proton irradiation at a dosage of lxl O 14 cm"2, indicating its
intrinsic resistance to radiation.
XX
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 1
INTRODUCTION
§1.1 Why GaN based AlGaN/GaN HFET?
Microwave power transistor is one of the most important and costly components
in today’s wireless communications and radar systems which are necessary in almost all
major aspects of human activities. Based on the high electron mobility and saturation
velocity, GaAs MESFETs have been played a key role in the field for the past 20 years.
However, because of the limited breakdown voltage, a typical commercial GaAs power
MESFET has only a power density of around 300 mW/mm.
In the mid 90’s, GaAs based HBTs have achieved much progress by using the socalled Low Thermal-Impedance Techniques, in which, the efficient heat removal is made
possible by the fact that thick metallization (such as emitter airbridges with a thickness up
to
22
pm ) can reduce the effective heat dissipation path to less than
1
pm within the
device. Most of the heat is generated at the collector junction by electrons which are
accelerated by the high electric field. Using the thick metal airbridge sit right on top of it
(for vertical HBT structure), a low resistance path for heat flow is formed. With this
1
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
technique, a power density of 12.5W/mm of AlGaAs/GaAs HBT with a total CW power
of 25W was reported1, but at a relatively low frequency of 2 GHz.
The high-frequency performance of GaAs HBT was improved by reducing the
base-collector capacitance. With undercut collectors , a HBT unit cell with 10 fingers
(2.8 pm by 50 pm) was reported to exhibit a CW output power 2.09 W, a power density
of 4.18 W/mm, a power-added efficiency of 62.2%, and a gain of 7.13 dB at 10 GHz
when the collector was biased under a voltage of 10 V.
In the late 90’s, InAlAs/InGaAs based metamorphic MHEMTs on GaAs
substrates have shown a remarkable improvement in cut-off frequency. The metamorphic
structure is implemented with a thick metamorphic buffer layer between the GaAs
substrate and the active layer in order to relax the lattice mismatch strain. This enabled a
device designer to design an InGaAs layer used as the channel of HEMT, which the
composition of In can be chosen in a wide range in terms of the requirements of device
performance. A MHEMT device was demonstrated to have a cut-off frequency of
235GHz, an extrinsic transconductance of 1050 mS/mm from a 0.13 pm gate length
device'. However, the power density of MHEMT was not high enough for some
advanced applications such as base stations of wireless phone systems.
InP is another material for fabricating microwave power transistors because of its
high electron saturation velocity, low ionization coefficients, and good thermal
conductivity. In 1980’s, InP MESFETs with a gate length of 1.5 pm, fabricated from
VPE layers, were reported to exhibit a power output density of 1.15 W/mm and a gain of
4 dB at 9 GHz .4 Later, in order to suppress the gate leakage current and to allow higher
2
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
voltage operation, MISFET structure was used with an insulated gate layer formed by
low temperature pyrolytic S i0 2 deposition. At 9 GHz, the output power density per unit
gate width was improved to be 4.2 W/mm with a gain of 4 dB and a power added
efficiency of 40%.5
The high-frequency performance was improved further by using HEMT structures
so as to make full advantage of 2DEG with a higher mobility and a higher sheet density
of carriers. A InP-HEMT with a gate width of 50 pm was shown 6 to have an extrinsic
transconductance (gm) of 1000 mS/mm under a drain current of 880 mA/mm. The
channel breakdown voltage was improved to 6.0 V after passivation, which was
significantly higher than that of pseudomorphic InP-HEMTs. At 94 GHz, the output
power density of the device was 300 mW/mm with a power added efficiency of 21 %.
All the achievements mentioned above were very encouraging. However, the
issues resulting from poor thermal management and overwhelming interface traps made
the characteristics of power transistors unstable and it was difficult to put those power
transistors into production. Yet extensive studies have been continuing for improving the
performance of power devices and progress has been made continuously. In order to have
a new generation of power transistors whose performance can meet the requirements of
systems in the future, recent research efforts have been in addition pointed to wide band
gap semiconductors.
According to Johnson’s figure of merit7, the limit of power-frequency
product(PFP) was determined by the material parameters: the breakdown electric field
intensity (Ebk) and saturation velocity of electron (Vs):
3
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Where Dg is the thickness of the channel, Lg is the gate length. Ebk is the electric field
under which band-to-band impact ionization occurs, and it mainly depends on the band
gap Eg. Vs is primarily limited by phonon scatterings. Major parameters of materials
used for power devices relating to Johnson’s figure of merit are listed in Table 1.1.
T a b le 1.1 B a s ic p r o p e r tie s f o r s e le c te d s e m ic o n d u c to r m a te r ia ls
8
Si
GaAs
GalnP
4H-SiC
GaN
Eg(eV)
1.1
1.4
1.9
3.2
3.4
Ebk(V/cm)
3x10s
4x10s
6 x 10s
2 0 x 10s
(20-30)xl0s
v s (cm/s)
6 x 10 b
lOxlO 6
10x l0 b
2 0 x 10 b
2 0 x 10 b
\l (cm2/ Vs)
1000
8000
2000
500
1000
150
43
52
490
130
1
7
16
282
282-634
k
(W/m°C)
PFP
Power-frequency product (PFP) numbers based on John’s Figure of Merit are
normalized to that of silicon, k is thermal conductivity and |i is mobility of electron. It
was seen in the table that both wide bandgap materials of SiC and GaN can achieve a
Johnson’s figure of merit more than two orders larger than that from conventional
semiconductors because of their intrinsic large breakdown field and electron saturation
velocity.
4
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Essentially, SiC is a good candidate for power device. In recent years, SiC power
transistor has been achieved a high CW power density of 3.3 W /mm at 10 GHz through
intensive research efforts9. R. A. Sadler et al. demonstrated that a single SiC transistor
with a gate width of 12 mm exhibited an output power of over 30 watts at X band under
pulsed-mode conditions10. One of the major issues of SiC MESFET is the unstability of
its electrical characteristics. Another issue is the power density, still low in CW operation
case for the device. Both issues seem to be strongly related to the poor quality of
substrate and epitaxy layers, and it is now clear that the quality of SiC semi-insulating
substrates and the epitaxy layers are critical to fabricate SiC MESFETs.
Unlike SiC
devices whose performances depend on the bulk mobility, GaN HFETs (or HEMTs) are
able to take advantages of the two-dimensional-electron gas (2DEG) at the AlGaN/GaN
heterojunction. The velocity of the 2DEG in the channel of GaN HFET can be larger than
that of carriers in bulk SiC, because the bulk mobility of GaN doubles that of SiC, and
moreover, the mobility of 2DEG at the AlGaN/GaN heterojunction can be enhanced
further with respect to bulk GaN mobility. The large carrier velocity of 2DEG at the
AlGaN/GaN heterojunction can allow GaN HFET to achieve a larger current density and
a higher cut-off frequency than SiC device. Furthermore, the effect of piezoelectric and
spontaneous polarization of AlGaN/GaN can result in a much higher carrier density in the
channel of GaN HFET than the carrier density originated from the dopants in
conventional non-polar semiconductors, which is another important advantage over other
candidates for fabricating high-frequency power transistors.
5
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Based on data given in reference11, Fig 1.1 shows breakdown voltage versus
cutoff frequency (fT) of reported devices fabricated by using different semiconductor
materials. In order to achieve high power output, a large breakdown voltage is necessary.
However, as we can see, the breakdown voltage of the high-frequency transistors made
from conventional materials is limited to less than 15 Volts due to the intrinsic properties
of the material and the device structure. For comparison, one data point (GaN HEMT)
corresponding to reference 12 is also shown, indicating GaN based device can have much
higher fy* Vbr product than conventional material based devices as we expected.
100
AGaAs
vOJj
HBT
i ( InP
HB T
^InP
H E MT
X Ga N H EMT
a
"©
10
a
£
o
\
a«
sPQ
0
50
100
1 50
200
250
Cutoff Frequency (GHz)
F ig 1 .1 . B r e a k d o w n v o lta g e v s c u to f f f r e q u e n c y f o r In P a n d G a A s b a s e d d e v ic e s
ii
§1.2 Research background
The first GaN-based HFET was reported by Khan et al. in 199313. The structure
was deposited over a basal plane sapphire substrate using low pressure metalorganic
chemical vapor deposition (MOCVD) with triethylgallium, triethylaluminum, and
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
o
ammonia as the precursors. It consisted of a 0.6 pm thick n-GaN channel with a 1000A
thick cap layer of n-type Al 0.i4 Ga
0.86
N. A thin AIN buffer layer was used as a nucleus
layer to enhance the quality of the GaN films. The gate length and width of the HFET
were 4 and 50 pm, respectively, and the channel opening was 10 pm. The device used
Ti/Au as the ohmic metallization for source and drain contacts, followed by 250 ° C for 1
min annealing in flowing forming gas. The gate was formed using TiW as the contact
metal. The device had a transconductance of ~ 23 mS/mm. The first RF measurement
result from GaN HFET was reported by the same group one year later14. For the later
case, the epi-structure was also prepared by MOCVD on a sapphire substrate. The cap
layer thickness was reduced to 250 A for roughly the same Al content. The device had a
drain-source spacing of 1.75 pm and a gate length of 0.25 pm. The fabricated device had
a gm of 27 mS/mm, which was believed to be suffered from the poor ohmic contacts (28
Q/mm). The measured cutoff frequency, and the maximum oscillation frequency of the
device were 11 GHz and 35 GHz, respectively. These values were superior to the results
of reported FETs based on other wide band gap semiconductors such as SiC at that time,
demonstrating microwave application potential using AlGaN/GaN HFETs.
In 1995, the gm of the HFET device was improved to 120 mS/mm by Ozgur et
al . 15 with a better ohmic contact (2.5 Q.mm). The device had a gate length of 3 pm and
had a current density of 300 mA/mm. They also reported the high sheet charge density of
1 xlO
13
-2
cm" available in this AlGaN/GaN structure system, which was much higher than
those obtained in the AlGaAs/GaAs or InGaAs/InAlAs system, and experimentally
demonstrated the high current density capability of GaN-based HFETs .
7
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
We started our research effort on GaN based HFET at UCLA in June, 1996. At
that time, there were little techniques that were available to fabricate GaN HFETs. Our
primary goal was to study this new material system, especially to study the influence of
PZ and SP effects on saturation current density (sheet charge concentration) and
breakdown voltage of the device. Then develop a viable technology to design, fabricate
high performance power AlGaN/GaN HFET for x-band applications.
As a summary, Table 1.2 reviews the milestones in this research field, showing
the exciting progress made in the last
6
years.
T a b le 1 .2 M ile s to n e s f o r th e d e v e lo p m e n t o f A lG a N /G a N H F E T
1994
27 mS/mm 0.25pm L„, fT, fmax were 11GHz and 35 GHz, Asif Khan et
al . 14
1995
120 mS/mm, 3pm Lg, Ozaur, H. Morkoc et al . 15
1996
120 mS/mm, 1pm Lg, UTmax 6 GHz, 15GHz l.IW /m m @2GHz, by
Y.F.Wu et al. atU C S B 16
1997
240 mS/mm, 0.2pm Lg, fT 50GHz, 1.7W/mm @ 10GHz by Y.F.W u17;
UCLA: 1pm Lg, 65 mS/mm, fx 10 GHz
1998
200 mS/mm, 0.45pm, 6 .8 W/mm @ 10GHz, by S.Sheppard et al. of Cree,
total P 0 4W @ 10GHz 18;
UCLA 210 mS/mm, 0.8pm Lg, fx 60 GHz on sapphire 19
1999
230 mS/mm, 0.5 pm, 9.1W/mm @ 8.2 GHz, 11.7W @ 8.2GHz total for
4mm by W u 20 ;
UCLA 270 mS/mm, f MAX 107 GHz on SiC21, P0 7W @ 9.4 GHz 4x1.28 mm
2000
Pulsed 51 W @ 6 GHz 8 mm by Y. F, W u 22 ;
UCLA 8 W @ 9 GHz CW mini- module single 5 mm device 23
8
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
§1.3 Synopsis
The principal objective of this dissertation was to develop a viable technology for
GaN-based FETs with excellent DC, small signal RF and especially microwave power
performances predicted for such a wide band gap material.
Chapter 2 presents basic analysis, calculation and simulation of AlGaN/GaN
structures. Efforts focus on the piezoelectric and spontaneous polarization effect of this
material structure. Induced sheet charge density has been calculated using analytical
method as well as simulation. An important concept of surface compensation is
introduced to explain the structure behavior. Heterostructure design rules for this material
system are proposed.
Chapter 3 describes testing structures and experiments to verify our analysis to the
structure. Then followed by the development of process techniques.
Chapter 4 presents basic device structure and results. High temperature
performance was evaluated at a temperature of up to 592 °C. DC and RF power amplifier
results from the baseline devices are illustrated.
Chapter 5 introduces the development of the gamma gate process. As a result, a
0.3 pm gate length and a field plate are achieved simultaneously using 1 pm conventional
optical lithograph techniques, and the RF performance and breakdown performances are
improved significantly.
Chapter
6
depicts preliminary results of proton irradiated AlGaN/GaN HFET,
showing that AlGaN/GaN device is a very good candidate for radiation hardening
application.
9
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Finally, we discuss two major issues for suggested future work. One is the
extremely high power dissipation density related performance stability improvement, and
the other is the PZ and SP effects related surface charge control. These two issues must
be solved in order to commercialize this GaN HFET device.
Bibliography
1 Hill, D .; Kim, T.S. 2 8 -V low therm al-im pedance H B T w ith 2 0 -W C W ou tpu t p o w er. IEEE
T ransactions on M icro w a ve Theory an d Techniques, vol.45, (no. 12, pt.2), (1 9 9 7 IEEE M TT-S
International M icro w a ve Sym posium , D enver, CO, USA, 8-13 June 1997.) IEEE, D ec. 1997.
p . 2224-8. 8 references.
2 H in-F ai Chau; W ilcox, G.; W enliang Chen; Tutt, M .; H enderson, T. H ig h -p o w er high-efficiency
X -b a n d A lG a A s/G a A s h eterojunction b ip o la r tran sistors with undercut collectors. IEEE
M icro w a ve a n d G u id ed W ave Letters, vol. 7, (no.9), IEEE, Sept. 1997. p .288-90. 11 references.
3 D. Dum ka, G .C ueva, W. Hoke, PI L em onias an d I. A desida, 0.13 mu m gate-len gth In/sub
0.52/A l/sub 0.48/A s-In /su b 0 .5 3 /G a/su b 0.47/A s m etam orphic HEM Ts on G aA s substrate. 58th
D RC. D e v ic e R esearch Conference. C onference D ig e st P iscataw ay, NJ, USA: IEEE, 2000. p .834. x ii+ 1 7 6 pp.
4 A rm and, M .; C hevrier, J.; Linh, N.Y. M icrow ave p o w e r am plification w ith InP FETs.
E lectron ics L etters, vol.16, (no.24), 2 0 Nov. 1980. p.906-7.
5 Arm and, M .; Bui, D. V.; C hevrier, J .; Linh, N. T. InP f ie ld effect tra n sisto r f o r m icro w a ve p o w e r
am plification. R evue Technique Thomson-CSF, vol.16, (no. I), M arch 1984. p.47-98.
6 H wang, K .C .; Ho, P .; Kao, M .Y.; Fu, S.T.; Liu, J.; Chao, P .C .; Smith, P.M .; Swanson, A.W. Wb a n d high p o w e r p a ssiv a te d 0 .1 5 mu m InAlAs/InGaAs H E M T device. C onference P roceedings.
Sixth In ternational C onference on Indium P hosphide and R ela ted M a teria ls (Cat. N o.94C H 33696), Santa B arbara, CA, USA, 27-31 M arch 1994. N ew York, NY, USA: IEEE, 1994. p. 18-20.
v iii+ 6 5 6 pp.
7 E. O. Jonson, “P h ysica l lim itations on frequency a n d p o w e r para m eters o f tra n sisto rs”, RCA
Rev., pp . 163-77, 1965
10
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
8 Y. F. Wu, “A lG a N /G a N M ic ro w a ve P o w e r F ligh -M obility-T ran sistor”, dissertation , ch apter
one, p2.
9 S.Sriram , T.J.Smith e t al. ” Fligh p o w e r operation o f 4H -S iC M ESFET a t 10 G H z ”, 5 5 th D R C
1997
10 Sadler, R .A .et al. S iC M ESF E T h ybrid am plifier w ith 3 0 -W output p o w e r a t 10 GHz.
P ro ceed in g s 2 0 0 0 IE E E / C ornell C onference on H igh P erform ance D e v ic e s IEEE, 2000. p. 173-7.
11 Chanh N guyen a n d M iro sla v M icovic, ” The State-of-the-A rt o f G aA s an d InP P o w e r D evices
a n d A m p lifie rs”, IEEE TRANSACTIONS O N E LE C TR O N DEVICES, VOL. 48, NO. 3, M A R C H
2001, p p 4 7 2 -8
12 S.J. Cai, R. Li, Y.L. Chen, L. Wong, W.G. Wu, S.G. Thom as an d K.L. W ang "High P erform ance
A lG a N /G a N H F E T with Im p ro ved O hm ic Contacts" IEE E lectron ics L etters 26th N ovem ber 1998
Vol. 34 N o .2 4 PP2 3 5 4 -6
13 M .A.Kahn, A .R .B hattarai, J.N .Kuznia, an d D ..T.O lson, “H igh electron m obility tra n sistor
b a se d on a G aN -A lG aN heterojunction ” Appl. Phys. Lett., vol. 63(9), p p 1214-5, 1993
I4M .A.Kahn, J.N .K uznia, a n d D ..T .O lson “M icrow ave perform an ce o f a 0 .25 p m gate
A lG a N /G a N H E E T ” A ppl.P hys.L ett. 65 (9), p p l 121-3, 1994
15 Ozgur, A .; Kim, W.; Fan, Z.; B otchkarev, A.; Salvador, A.; M oham m ad, S.N.; Sverdlov, B.;
M orkoc, H. “H igh transcon ductance n o rm ally-off G a N M O D F E T s”. E lectron ics L etters, vol.31,
(n o .16), 3 Aug. 1995. p . 1389-90. 8 references.
16 Y.F.Wu, B. K eller, S. Keller, D. K apolnek a n d U.K. M ishra “ M easu red m icro w a ve p o w e r
p erform an ce o f A lG a N /G a N M O D F E T ” IEEE E D L -17 (9) 1996, p p 4 5 5 -7
17 Y.F. Wu, B.P. K eller, S. K eller, N.X. Nguyen, M. N guyen an d U.K. M ishra, “ Short channel
A lG a N /G a N M O D F E T ’s w ith 5 0 G H z jT an d 1.7 W/mm output p o w e r a t 10 G H z ” IEEE E D L -18
(9) 1 9 9 7 p p .4 3 8 -5 1
18 S.T.Sheppard, K. D overspike, W. P rib b le an d J. W. P a lm o u r,”H igh p o w e r m icrow ave
G aN /A lG aN HFETs on S iC ” in late news, 5 6 th D RC, C harlottesville, VA, June 1998
19 S.J. Cai, R. Li, Y.L. Chen, a n d K.L. Wang ,IEE E lectron ics L etters 26th N o vem b er 1998 Vol.34
N o .24 p p 2 3 5 4 -6
20 Y.F .W u,D .K apolnek, J. Ibbetson, N. Zhang a n d U.K. M ish ra ” H igh A l con ten t A lG aN /G aN
HFETs on S iC su b stra tes w ith very high p o w e r p e rfo rm a n c e ” IED M 99-925, 1999
21R. Li, S.J. Cai, L. Wong, Y. Chen, K.L. Wang etc IEEE E lectron D e v ic e s letter. July 1999 vo l.20
N o .7 p p 3 2 3 -6
11
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
22 Wu, Y.-F.; C havarkar, P .M .; M oore, M .; Parikh, P .; Keller, B .P .; M ishra, U.K. A 50 -W
A lG a N /G a N H F E T am plifier. International E lectron D e vic es M eeting 2000. Technical D igest.
IE D M (C at. N o.00C H 37138), P iscataw ay, NJ, USA: IEEE, 2000. p .375-6. 871 pp. 5 references.
23 S.J. Cai, Jiang Li, Y. L. Chen, a n d K. L. Wang ,IEEE T opical W orkshop on P o w e r A m plifiers
f o r W ireless C om m u n ication s”.
Sept. 11-12, 2 0 0 0 San D iego, p p 7 7 -7 8 .
12
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 2
AlGaN/GaN BASIC CHARACTERISTICS
This chapter discusses the basic material characteristics o f the AlGaN/GaN
structure. A very important feature o f this material system is the piezoelectric and
spontaneous polarization effects. It is due to the polarization effects that the super high
density of 2DEG is available in this structure. The bound charge densities induced by the
polarization have been evaluated. The part o f the 2DEG density in the heterostructure
resulted from the conduction band offset o f AlGaN/GaN is also calculated using a
traditional heterojunction theory. Mobility o f electrons in the structure is calculated using
a set o f analytical equations as well, showing that the polar optical phonon’s scattering
and piezoelectric scattering are dominant factors in degrading mobility. In order to
understand more about the impact o f bound charge on the 2DEG density in the channel
and hence on performance o f the device, 2D simulation is performed based on the model
we proposed by using device simulator SILVACO®. Detailed simulations show that the
surface status, thickness o f the AlGaN cap layer and doping density in the cap layer are
important factors affecting the 2DEG density in the channel. These results can be used as
a guideline to design and fabricate high performance AlGaN/GaN HFETs.
13
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
§2.1 Basic characteristics for AlGaN/GaN material system
2.1.1 Piezoelectric(PZ) and spontaneous polarization(SP) effect
The natural structure o f the III-V nitrides is wurtzite, a hexagonal crystal structure
defined by the edge length a o f the basal hexagon, the height c o f the hexagonal prism,
and an internal parameter u defined as the anion-cation bond length along the [0 0 0 1 ] axis
in units o f c , as shown in Figure 2.1(a)(after Ambacher et al.1). The AIN layer was
epitaxied on the top of GaN forming a AlN/GaN heterojunction.
G«(A!Mace
tensile
strain
A------
|P si* | P it
~CT
▲
AlGaN
2DEG
SP
I
GaN
AIN or A lG aN
(a)
1
(b)
AI2O 3
substrate
Figure 2.1 (a) Schematic of the wurzite structure with Ga-polarity for AlN/GaN
heterostructure. Ppe stands for the PZ polarization, and Psp stands for the SP
polarization, (b) Polarization induced 2DEG is shown at the interface (after /7/).
Wurtzite has a unique axis, and it is a structure with the highest symmetry
compatible with the existence o f spontaneous polarization even in the absence o f strain.
However, the spontaneous polarization Psp in a solid is not well defined. Only those
14
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
differences in Psp
between two phases which can be linked by an adiabatic
transformation that maintains the insulating nature o f the system throughout are well
defined2. For example, one phase can be considered unstrained and the other strained.
Vanderbilt et al .3,4 proved that the polarization difference AP between the wurtzite and
zincblende phases can be calculated by considering an interface between the two phases
and by defining Psp to be zero in zinc blende because zinc blende is cubic and cannot
have a spontaneous polarization in an infinite bulk periodic crystal.
The direction of the polarization depends on the polarity o f the crystal, i.e.
depends on whether the bonds along the c-direction are from cation sites to anion sites or
vise versa. When the bonds along the c-direction are from cation (Ga) to anion (N), the
polarity is said to be Ga polarity, and the polarization is negative, i.e. the polarization
direction is along [OOOl]. In this case, the crystal surface is terminated with Ga atoms. In
our work, we are only interested in the Ga polarity case as shown in Figure 2.1, such that
donor-like charges (+a) are resulted at the AlGaN/GaN interfaces from the polarization,
and hence the 2DEG can be formed at the interface.
The detailed calculation for the spontaneous polarization and piezoelectric
constants for wurzites nitride was done by Bernardini et al.3, and the results are listed in
Table 2.1. To facilitate further comparison, a few other III-V materials and II-V wurtzite
oxides are also listed.
Table 2.1 Spontaneous polarization (Psp) andpiezoelectric constants
(ej3 and esi) for several materials with wurtzite structure3.
15
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Psp
d33
63 1
(C/m2)
(C/m2)
(C/m2)
AIN
-0.081
1.46
-0.60
GaN
-0.029
0.73
-0.49
InN
-0.032
0.97
-0.57
ZnO
-0.057
0.89
-0.51
GaAs
-0 .1 2
0.06
InP
0.04
-0 .0 2
Two important features o f the data in the above Table are the following:
1. The absolute value of the piezoelectric constants is about more than ten times
larger than in conventional III-V compounds. AIN is the one with the largest
value.
2. The spontaneous polarization is also very large in the nitrides. That o f AIN, in
particular, again is the largest among the listed.
In the absence of external fields, in the linear regime, the piezoelectric polarization
Ppz, is related to the strain s by :
P- = X
j
eu e J
(2-1)
Which defines the components o f the piezoelectric tensor eij and the strain tensor s,.
When forming a AlGaN/GaN structure, two mechanisms lead to the strain in the
AlGaN layer. The first is the lattice constants mismatch ( 2.41% mismatch between AIN
and GaN at room temperature) and the second is the thermal strain caused by the thermal
16
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
expansion coefficient difference between the substrate and the epi-layer. Since the lattice
constant for AIN is smaller than that for GaN, AlGaN experiences tensile strain in the
structure. When the material structures grow along [0001], the piezoelectric tensor of
wurtzite has three non-zero independent components. In this case, the piezoelectric field
aligned along the [0 0 0 1 ] direction, and here only two piezoelectric tensor components e 3 i
and
633
are important while the third one, e5i is related to shear strain and can be ignored
in our case. The index 3 o f the piezoelectric tensor corresponds to the direction of the c
axis. The piezoelectric polarization can be simplified as1,5:
P pe — 2 ( e 3j - e 33 C13/C33) s xx
(C /m )
( 2 .2 )
Where e 3 i and e 33 are piezoelectric tensor components, C 13 and C 33 are AlxGai_xN elastic
constants. exx is the strain component in the interface plane. All o f them are related to Al
mole fraction x in the AlxGai_xN layer. By using linear interpolation, we can get the
elastic constants as the following:
C 13= 1.087+0.021x (10u N/m2)
(2.3)
C 33 = 3.485+0.37x (10 11 N/m2)
(2.4)
Where the elastic constant data used were taken from five references and averaged6.
Similarly, we have the piezoelectric tensor components:
e3, = -0.1 lx-0.49 (C/m2)
(2.5)
e 33 = -0.73x+0.73 (C/m2)
( 2 .6)
Also the strain o f AlGaN and the lattice constants o f the two are:
£xx
(2.7)
<kjaN/U AlxGal-xN“l
o
( 2 .8)
aGaN - 3.189 (A)
17
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
a AixGai-xN= 3 .1 8 9 -0 .0 7 7 x (A)
(2.9)
We evaluated the spontaneous polarization in the structure. In AlxGai_xN material, using
Vegard’s law and the value listed in Table 2.1, the spontaneous polarization is:
Psp= -0.052 x - 0.029 (C/m2)
(2.10)
We notice that the spontaneous polarization o f GaN and AIN is negative (i.e., in the
layers grown in the [0001] direction on Ga face, Psp is opposite to the growth direction)
and increases in magnitude by going from GaN to AIN, as shown in Figure 2.1.
Therefore, at the top surface between AlxGai_xN and air, we have the difference o f Psp:
SPspi = -0.052 x-0.029 (C/m2)
(2.11)
At the AlxGai_xN /GaN interface, again, we have the difference o f Psp:
&PsP2 = -0.052 x (C/m2)
(2.12)
In the absence o f external electric fields, the total macroscopic polarization P at the
interface is the sum of the spontaneous polarization Psp in the equilibrium lattice, and the
strain-induced or piezoelectric polarization PPE. The gradient o f polarization in space is
the charge density given by
p
= -V
• P
(2.13)
Thus considering a slab-shaped structure, polarization is perpendicular to the surface. The
sheet charge density can be expressed as:
a = (Psp(top) + Ppe(top)) - (Psp( bottom) + Ppe{bottom))
(2.14)
If the polarization induced bound charge at the interface of the AlGaN/GaN is positive
(+ g), as in the case o f Ga polarity here, free electrons tend to accumulate and compensate
18
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
the polarization-induced charge, during the cooling process after growth. These electrons
then form a 2DEG with a sheet carrier concentration o f ns. .
By using the formula listed above, the induced bound charges at the interface thus
can be calculated and shown in Figure 2.2. Be aware that the density o f electrons in the
2DEG at the interface departs somewhat from the bound charge density, because a net
5.1*10
4.6*10
su
4.1*10
e
3.6*10
o
3.1*10
i-
aa
w
a©
u
V
V
JS
!Z3
2 . 6*10
2 . 1*10
1. 6 * 1 0
1 . 1*10
6*10
1*10
13
13
To al induced
sht set charges '
13
13
13
13
@ 30% A l,
1 .7 x l0 13cm 2
charges could
be induced
------
Sp. Polariza :i°n
induced cha ges
i
J
13
13
13
\
^
PZ ind jc e d
charge s
12
■
12
0.2
0.4
0.6
0.8
Al Content
Figure 2.2. Calculated piezoelectric induced and spontaneous polarization induced
bound charge at the AlGaN/GaN interface vs Al content.
charge density is required to terminate the electric field in the AlGaN layer. This electric
field is fixed by the potential drop in the AlGaN determined by the Fermi level positions
at the surface and at the interface. Flence, the bound charge density gives us an upper
limit for the 2DEG density. In the next section, we will see that the 2DEG density in the
interface obtained from simulation is very close to this upper limit if the structure is
carefully designed.
19
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
2.1.2 Conduction band discontinuity
In addition to the polarization-induced 2DEG charge, another contribution of
sheet charges accumulated at the interface comes from the energy conduction band edge
discontinuity.
Figure 2.3 shows a standard hetero-junction structure diagram.
The
energy band gap for AIN is 6.13 eY, for GaN is 3.42 eV. We assume 70% alignment
factor, using linear interpolation, for AlxGai.xN/ GaN system, the discontinuity o f the
*7
conduction band is :
AEc(x)
= 0 . 7 ( E g ( x ) - E g (0))
(2.14)
Eg (x) = 6.13x + (1 - x)3.42 - x b (l - x)
(2.15)
Where x is Al content, b is the bowing factor, varying from 0.7 to 1.2. We chose 1 for
the bowing factor in our calculation.
In tr in sic G a N
Figure 2.3 Schematic ofAlGaN/GaN band diagram
Assuming the first sub-band filled with electrons only, Fermi energy Ep(x), first
sub-band in the triangle well Eo(x) and sheet charge density ns ( x ) 8,9,10 can be expressed
as the following:
20
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
EF(x) = E0(x) +
7ih
m* ( x
)
(2.16)
n,(x)
2/3
97ihq2
E0(x) =
nfx)
(2.17)
8s0^m*(x) £(x)
f
) 1/2
ns(x) = | l£{x)m.[Ep (x) - AEc(x) + S]+ Nd2(di + Ad)2j
- Nd(di + Ad)
(2.18)
Where m* is the effective electron mass, which we take it as a constant, m*= 0.228 me' *.
By solving the self-consistent equations listed above using Mathcad®, we obtained the
results shown as in Figure 2.4 for different doping concentrations.
5*10
S
w
4*10
C
o
5x10
3*10
i.
a<u
«
ES
2*10
5x10
C
o
1*10
5x10
o
0.1
0.2
0.3
0.4
0.5
Al content
Figure 2.4 Calculated sheet charge density of 2DEG vs Al content for HFET structure
1
7
?
l
f
i
?
I
Q
?
without gate due to AEC, for Nd=5x10 cm , 5x10 cm' and 5x10 cm' cases ( for
undoped spacer of dj=30A, 2DEG mean distance to the interface Ad=80A and correction
6=25 meV9).
21
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
It’s worth noting that the 2DEG contribution from the conduction band gap
discontinuity is not significant compared to the bound charges from PS and PZ
contribution. For example, for 30% Al content case, we have 1.7xl013 cm’2 sheet charges
from PS and PZ, but only around 1.7 xlO 12 cm'2 from the contribution o f the band
discontinuity, one order of magnitude smaller. By taking the bound charge as the upper
limit o f the available 2DEG density from the SP and PZ effects, plus the contribution
from the band offset, the total charges available to the structure is depicted in Figure 2.5.
As can be seen, for 30 % Al content, with 8 x l0 l8cm_3 doping in the AlGaN cap layer, 2x
1013 cm-2 sheet charge in the heterojunction channel can be obtained, which is at least 5
40
a
w
30
For 30 % Al structure,
2 x l0 13cm'2 sheet
charge can be induced.
20
s-
10
JS
0
0.15
0.30
0.45
0.60
A lcontent
Figure 2.5 Total 2DEG density ( the sum of SP & PZ effect induced and conduction
band discontinuity caused sheet charges) vs Al content for AlGaN/GaN structure.
For 30 % Al content, 8 x l0 18 cm3 doping, 2xl0*3 cm2 sheet charge at the
heterojunction can be induced.
22
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
times higher than that obtained from AlGaAs/GaAs material. Therefore, in AlGaN/GaN
material system, piezoelectric effect and spontaneous polarization effect play a very
important role. In another word, we must be aware that any effects which could result in
the change of piezoelectric effect and/or spontaneous polarization effect would have a
direct impact on the device performance.
2.1.3 Mobility
Another fundamental parameter o f interest is mobility. Since AlGaN/GaN has PZ
and SP effects, we expect that in this material system, the PZ scattering and polar optical
phonon scattering play an important role.
By applying semi-analytical expressions,
referring to appendix I, by using Mathcad®, the calculated results for the mobility are
shown in Figure 2.6 (constants used partially after M. Shur12).
In the figure, four scattering mechanisms are taken into account. They are
acoustic phonon scattering, polar optical phonon scattering, ionized impurity scattering
and piezoelectric scattering mechanism. Each scattering mechanism limited mobility is
shown against temperature. Also shown is the combined mobility o f the material with
and without ionized impurity scattering, corresponding to the bulk material and the
2DEG case, respectively. Here we assumed the ionized impurity scattering is totally
suppressed in 2DEG case. In practice, this will be very difficult to achieve in
AlGaN/GaN. However, the calculation gives us an upper limit for the 2DEG mobility. In
real case, the ionized impurity may more or less have some scattering effects, hence the
real curve for the mobility o f the electrons in the AlGaN/GaN channel would lie in
23
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
between the two curves o f the “bulk” case and the “2DEG” case. We predict that, the low
temperature mobility o f the HFET structure is around 10,000 cm
■j
/Vs and room
temperature mobility is around 1500 cm2 /Vs. In this case, at temperature o f lower than
200 °K, the piezoelectric scattering plays a major role, while at temperature o f higher
than 200 °K range, the polar optical scattering dominants. Since higher Al content
material has higher polar optical and piezoelectric scattering effect, we expect those
samples with higher Al content will have lower mobility. This has been verified
experimentally in the next chapter.
105
lonized Impurities
^
2DEG
Piezo
104
E
u
£
IS
o
s
>•
Bulk
Polar
Acoustic
io 3
io 2
0
100
200
300
400
500
Temperature (K)
Figure 2.6 Calculated mobility for AlGaN/GaN material vs temperature.
2DEG
mobility is obtained by ignoring the impurity scattering. Parameters used: m*=0.228
me, Energy gap: Eg = 3.4 eV, Polar optical phonon energy: Epo = 91.2 meV, Piezo
constant: epz = 0.5 C/m2, Elastic constants: Cl = 2.65xlOn N/m2, Cr= 4.42 xl0in N/m2
and deformation potential, Ec/S= 8.3 eV.
24
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
§2.2 AlGaN/GaN HFET device structure design
Spontaneous polarization is independent of the thickness o f the layer. On the
other hand, the piezoelectric polarization is thickness related: once the thickness exceeds
the critical thickness, the strain is relaxed, and hence the piezoelectric effect disappears.
Repeatability o f the HFET channel sheet charge density can be greatly improved if
mechanical strain relaxation (with an increase misfit dislocation density) is alleviated. The
effect o f strain relaxation is found to be acceptable when the AlGaN layer is less than oneand-half times (1.5X) the critical thickness, which is a function o f aluminum mole
fraction. The estimated critical thickness o f AlGaN on GaN is shown
1^
as a function of
mole fraction, x, in Figure 2.7. The strong piezoelectric and the spontaneous polarization
60
S
a
0
0.2
0.4
0.6
0.8
1
Al Mole Fraction
Figure 2.7 Critical thickness ofAlxGa j.x N layer as a Junction of aluminum
mole fraction: superlattice (solid line), SIS structure (dashed line)xvu.
25
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
property o f the AlGaN/GaN HFET can lead to a very high sheet charge density at the
AlGaN-GaN interface in excess of 1013/cm2, as we calculated in the previous section.
However, the PZ and SP induced dipole in the AlGaN layer is placing an additional
constraint on the transistor structure design. In the above section, we ignored the possible
effect o f the bound charges on the top surface (by assuming the surface is fully
compensated - we will discuss this fully in next section). In fact, the “piezoelectric
induced surface inversion” is a possible reason for premature power saturation o f the
transistor amplifier, which must be avoided.
The energy band diagrams o f an intrinsic semiconductor with and without a builtin dipole are shown in Figure 2.8. The Fermi level remains flat in the absence of an
external electric field, and the slope of the band edges reflects the built-in electric field
whose magnitude is proportional to the piezo-electric dipole moment in the material. Both
conduction band edge and valence band edge intersect the Fermi level if the product o f the
built-in dipole electric field and the thickness o f the semiconductor slab exceeds the band
gap energy as illustrated in Figure 2.8b. In the case o f a HFET structure with a thick,
inversion
(b )
Figure
2.8
Undoped semiconductor energy band diagram: (a) Non polarized
Semiconductor, (b) Strongly polarized semiconductor with surface inversion.
26
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
undoped AlGaN layer, there will be “mobile” positive charges (holes due to surface
inversion) at the surface o f the AlGaN layer in addition to electrons at the heterojunction
interface. The positive charges are not easily replenished when the gate voltage is
increased from pinch-off, leading to the hysteresis in Ids vs. Vgs characteristics. The
surface inversion could be the main culprit o f premature transistor RF power saturation.
The negatively charged surface on the top o f AlGaN does not allow the channel to become
fully open when the RF gate voltage swings in the positive direction, limiting the
maximum channel current Imax needed for high power amplifier operation.
For AlGaN/GaN structure, in order to avoid the surface inversion, an important
design concept hence is developed. As illustrated in Figure 2.9a, the Fermi level remains
flat as long as there is no external applied electric field across an undoped AlGaN/GaN
heterojunction. As the Al content increases, the band-edges o f the strained AlGaN layer
start to tilt and the slope is equal to the electric field associated with piezoelectric induced
polarization. Higher Al content lead to a greater band bending shown in Figure 2.9c, and
a further increase in either the AlGaN layer thickness ormechanical strain will eventually
lead to undesirable surface inversion illustrated in Figure 2.9d.
Assuming 70% conduction band offset, , to avoid the Fermi level intercept into
the valence band, we have the design rule for the case o f heavily doped cap layer:
EpT< Egi ;
(2.19)
and for undoped cap layer,
EPT< 0.5Egi ,
(2.20)
27
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
top
interface
I
I
A lxG a ,.xN
G aN
E e l _____________ ____i
E c2
Ef-
7
-EgT " - •
Y
Figure 2.9a AlGaN/GaN band diagram
without PZ and SP.
Ev2
E'V I
Figure 2.9b Low aluminum mole fraction
leads to weakly polarized AlGaN layer with
tilted band edges; ground states in the
interface quantum well at interface will be
lightly populated because of the low AlGaN
layer (EnT) product.
Eel
Figure 2.9c Higher aluminum mole fraction
leads to increased AlGaN layer polarization
and more band bending; interface quantum
well heavily populated with piezoelectric
induced electrons as the Fermi level
approaches the bottom of the quantum well.
'C l
■VI
‘V2
Figure 2.9d Surface inversion caused by
excessive thick AlGaN layer (Ep T) product;
depletion of mobile positive charges (holes)
on AlGaN surface may lead to premature RF
power saturation of an amplifier.
inversion
Figure 2.9 AlxGaj.xN layer design guideline: the surface inversion on top of the
AlGaN layer caused by strong PZ and SP effects must avoid.
28
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
where Ep is the polarization related electrical field and T is the cap layer thickness. Egi is
the band gap o f AlGaN, given by Eq. (2.15)
The calculated results of the maximum thickness of the AlGaN cap layer to avoid
the surface inversion for heavily doped case are illustrated in Figure 2.10. When the
AlGaN is fully relaxed, the piezoelectric polarization disappears; only the spontaneous
polarization field exerts effect. In the case o f fully strained samples, both SP and PZ take
effect, the electric field inside is stronger, and the maximum thickness is even smaller. For
1000
%
u
-e
H
SP+PZ
•1 0 '
800
600
SPcfily
400
SP only
200
SP+PZ
0
0.2
0.4
0.6
A lcontent x
0.8
1
Figure 2.10 The maximum cap layer thickness to avoid the surface inversion
for two cases: 1) fully relaxed(SP onlyj; 2) fully strained(SP + PZ). Inset is
the electric field intensity for the two corresponding cases. Assuming no
surface charge compensation and cap layer was heavily doped.
29
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
example, with 30% A1 content, the maximum thickness is 200 A for fully relaxed cap
layer and only 110A for strain case. For a undoped cap layer, the corresponding thickness
is around half o f that for the heavily doped case. Therefore, doped cap layer is preferred
to prevent the surface from inversion.
The electric field in the cap layer resulted from the polarization is shown as the
figure inset. Due to the high electric field , the energy between the valence band edge and
the Fermi level on the top surface of the structure will change by approximately 70 mV
for an additional atomic layer thickness (assume 20 % A1 content, in which there is a E
field o f 2 x l0 6 V/cm and that each atomic spacing is ~ 3.5 A). The requirement for ultra­
precision thickness control is difficult for current CVD epitaxial technology and hence
makes the device structure non-reproducible. However, the stringent conditions (AlGaN
layer thickness and aluminum mole fraction) imposed on HFET structure preparation can
be partially alleviated if the barrier layer is intentionally doped as described earlier. The
Poisson equation gives A V, the potential difference across the AlGaN cap layer, as a
function o f the doping density and the thickness o f each layer for a general multiple layer
structure, as follows,
AV = E (E PiTi- ( p iTi2)/2) (V)
/
(2.2 1)
where Ep, is the polarization related electric field; T, is the thickness o f the ith layer and )j
is the charge density in the ith layer.
Details o f the band bending and the sensitivity to doping and layer thickness can
be simulated as in the following section.
30
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
§2.3 Simulation
Simulation is always a convenient way to perform a detailed analysis to explore
the effect o f structure parameter variations o f devices. For this new material system, in
which PZ and SP effects are extremely strong, it is important to understand the effect of
the piezoelectric and spontaneous polarization induced charges and the effect of the
surface condition to the DC performance o f the device in the future. There have been a
few papers studying AlGaN/GaN devices by simulation, but few took the PZ and SP
effects into account14,15. In this section, the results o f simulation including PZ and SP
effects, which obtained by using a device simulator o f SILVACO®, is presented for the
first time. Models o f bound charge resulted from the PZ and SP in AlGaN/GaN HFETs
were implemented in SILVACO. The results show that the condition o f the surface plays
an important role to the 2DEG density; hence the device performance is highly sensitive
to the charge density o f the surface. The surface condition, along with the formation of
the positive charge layer at the surface, is probably the reason for the irreproducibility of
performance o f AlGaN/GaN ElFETs reported in the past.
So far there is no commercial 2D device simulator, which can be able to include
the PZ and SP effects. In order to establish models of the PZ and SP effects in
SILVACO, we defined trapped charges inside the cap layer-the AlGaN barrier layer to
physically simulate the positive and negative bound charges induced by the PZ and SP
effects. The sheet densities o f these charges are calculated according to section 2.11, and
these charges are assumed to locate on the top surface and at the interfaces within 10 A,
respectively, as shown in Figure 2.11 (Assuming they are uniformly distributed in the 10
31
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
A region). In our analysis we define a standard material structure, which consists of a
semi-insulated SiC substrate, a GaN buffer layer of 2.0 pm on a nucleation layer (n-type
unintentionally doped layer with a background doping o f 1015 cm'3), a follow-on 30 A
undoped AlGaN spacer layer, a 150 A doped AlGaN cap layer, and finally a 50 A
undoped AlGaN top layer.
The A1 content o f all AlGaN is 25 % for the standard
structure.
ct1=
-3.237 x 1013 cm'2
A lG a N
a 2 = + 1 .4 2 5 x 1 0 ° cm'2
G aN
a 3 = +1.812x10° cm'2
Figure
2.11. Model for PZ and SP effects. Charge density at each interface is
calculated for 25% Al content based on the analysis in the previous chapter. The
sheet charge, al stands for negative trapped charges while o2 and a3 are for
positive trapped charges.
2.31 Role of Piezoelectric and Spontaneous polarization induced
charges
According to Gauss’ law, a l near the top surface shown in Figure 2.11 produces
an electric field, and this electric field has a direct impact on the charge density in the
channel.
The simulation results are shown in Figure 2.12, illustrating that for each
32
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
doping density there is a threshold of the cap layer thickness in order to have a minimum
sheet charge in the channel. This means that there is no 2DEG in the channel when the
thickness is small enough, even though strong PZ and SP effects are presented in
18
16
B
14
cm
12
w
s*3
Ao
10
8
6
5x10
cm
5x10
cm
4
o
H
2
0
0
100
200
300
400
500
Cap layer thickness (A)
Figure 2.12 Simulated sheet charge versus cap layer thickness for the
standard structure. When the cap layer thickness is too thin, no channel
sheet charge is present.
AlGaN/GaN material. This was evidenced by Koley et al.16 and Smorchkova et al.17.
Both o f these experimental works by these groups showed that when the undoped cap
layer thickness was below a certain value, no sheet charge was present in the channel. To
explain this, Smorchkova assumed the presence o f high density donor-like surface states,
whose energy level was located deep below the conduction-band edge o f the AlGaN
33
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
layer. When the cap layer is too thin, all surface states are occupied so there is no 2DEG
in the channel. As the thickness of cap layer increases, the Fermi level “hits” the surface
states energy level, these surface states start to empty and the channel then start to have
2DEG. They overlooked the fact that the bound charge on the top surface ( al ) should
always exist as long as polarization exists in the material. Our simulation based on the
physical models described previously can easily explain the presence o f this minimum
thickness. If the cap layer is too thin, the Coulomb force of the negative charges on the
top surface expels all 2DEG charges away, such that no 2DEG charges can exist in the
14
12
s
rt
O
w
J
a
<u
o
H3
W
Q
fS
Ref.
10
8
160
I- 120
6
4
2
1.E+17 1.E+18 1.E+19 1.E+20
Doping in cap (cm-3)
0
0
100
200
300
400
500
600
Cap layer thickness (A)
Figure 2.13 Comparison of our simulation results with the experimental results from
Koley et al. 16 (cap layer undoped). Inset showing the minimum thickness of the cap
layer to have a 2x10J2 cm2 sheet charge in the channel for a structure of 25% Al
content, 30 A undoped spacer layer, and surface uncompensated.
34
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
channel for this case. For comparison, we have also simulated the case in Ref. [16] using
our models, and our simulation results are in a good agreement with their experimental
results as illustrated in Figure 2.13. The minimum thickness o f cap layer, which induces
2x1012 cm"2 sheet charge in the channel, is depicted as an inset in Figure 2.13.
2.3.2 Surface compensation
The above simulation does not include any effect o f surface compensation.
In
fact, the negative bound charge at the top surface induced by the PZ and SP effects tend
to be compensated by positive fixed charge introduced by processing variations. Flere we
introduce a compensation ratio (y, y = A c l/a l, where Aal is the compensated charge
density and a l is the original bound charge density before compensation) to take the
compensation o f the surface bound charge caused by the PZ and SP effects into
consideration. The effect of the compensation is illustrated in Figure 2.14, showing that
the sheet charge in the channel increases when the compensation ratio y exceeds a certain
value. The knee point of the compensation ratio y occurs when the number of total
ionized dopants in the cap layer equals that o f the net surface charges. In the figure, we
can also see that the surface inversion hole charge density is very high when y is small.
These holes, which have a very low mobility in the surface inversion layer, might be
responsible for the RF dispersion effects.
Thus, any growth process or post-growth treatment (passivation, annealing,
chemical treatment... etc), which can fully compensate the surface charge, can result in
the highest channel sheet density for a given structure, and hence the highest saturation
35
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
30
ns2DEG
n
25
nsparallel
<s
20
nstotal
'ao
©
holes
w
4> 15
M
S.
a
JS
10
<w
JS
xn
5
0
0%
20 %
40%
60%
80%
100%
Compensation ratio
Figure 2.14 2DEG, parallel, hole and total sheet charge densities as the function of
compensation ratio for the standard structure with a cap doping density of 1x1019 cm'3.
We notice the hole will be present on the top surface if there is a large al resulting
from the PZ and SP effects and if the surface is not compensated. (The compensation
was done in the simulation by simply decreasing the bound charge a1).
current.
Similarly, any process, which can change the compensation, can change the
2DEG density and thus the available saturation current density. Indeed, AlGaN/GaN
wafers, which have not been grown under fully optimized conditions, often show a
degradation o f device performance after processing as observed by Eastman’s group l8.
Further simulation shows that this surface effect can be minimized by implementing
heavily doping in the cap layer, as depicted in Figure 2.15. It is shown that the channel
sheet charge is rather sensitive to the surface compensation ratio when the cap layer
doping is low. A heavily doped cap layer can screen the effect o f the bound positive
36
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
aa
o
16
><e
14 *
1x10
cm '
a
a
«
JS
u
0%
20 %
40%
60%
80%
100 %
Compensation ratio
Figure 2.15 Channel sheet charge vs the surface compensation ratio for different
doping densities in the cap layer. Lower doping has a more sensitive surface. Different
doping cases all saturated at around the same sheet charge level (~ 1.6 xlO13 cm'2 )
indicating the PZ and SP effects dominate the sheet charge.
charge on top and thus reduce the sensitivity. The doped charge, in essence, behaves like
positive compensating charge. In other words, these structures with higher doping in the
cap layer “inheritably” will have more stable performance. For low doping, a high
compensation ratio will result in a much higher sheet charge density in the channel than
that for a low compensation ratio.
2.3.3 Effect of Al content to the sheet charge
Since the PZ and SP effects increase as the Al content in the AlGaN layer
increases, we expect that the sheet charge density increase as a function o f the Al content
37
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
as well.
Figure 2.16 shows that the maximum 2DEG density in the channel, which
increases almost linearly with the Al content for different doping densities in the cap
layer. From the above discussion, we may conclude that whenever the measured sheet
charge density becomes higher than the maximum 2DEG density, the structure must have
parallel conduction from substrate or barrier layer, and hence it often suffers from a
mobility degradation due to the contributions from the doped layer and substrate. Our
simulations demonstrate that a measured mobility degradation in high sheet charge cases
is highly possible due to the parallel conduction. ( In contrast, the spillover effect from
the AlGaN barrier to the GaN well on the GaN side only plays a minor role 19 ).
35
xlO
cm '
" us
30
1x10
cm '
-o
25
4x10
cm
1x10
cm
be
I
**
dJ
20
15
a03
J3
10
So
5
H
0
0
10
20
30
40
50
60
Al content (%)
Figure 2.16 Maximum available channel sheet charge vs Al content. The charge density
increases due to the PZ and SP contributions, almost independent of the doping density oj
the AlGaN layer. Here we assumed that all the surface charges were fully compensated
for different dopings in the cap layer. (Parallel conduction was not included).
38
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
2.3.4 Spacer layer
A spacer layer is usually needed in HFET structure design in order to reduce the
ionized impurity scattering to channel electrons from the barrier layer. However, this
spacer layer could have an effect on the total channel carrier density as well. Figure 2.17
18
25% Al
sv
15
©
21
"O
w
s
-o
12
9
ja
61
S.
6
85
tU
OJ
-a
cn
3
0
0
30
60
90
120
Spacer thickness (A)
Figure 2.17 Sheet charge density as a function of the space layer thickness. Here only
the discontinuity of the conduction band offset was taken into account (no PZ and SP
effects and no gate metal were considered in calculations).
shows that the sheet density o f charges decreases as the spacer thickness increases as in
all the HFET structures (the PZ and SP effects were not included in the calculation).
When the doping density in the cap layer is large, the spacer layer may have an influence
on the total channel carrier density as in the case o f GaAs based HFETs. Channel carrier
density depends strongly on the spacer layer thickness. For the case o f low doping
39
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
( 5 x 10 17 cm'3) in the cap layer, the sheet density o f the channel is almost independent of
the spacer thickness. This result is in a very good agreement with that we obtained from
analytical equations. (It is also consistent with all HFET structures). The spacing between
the peak location of 2DEG in the channel and the interface is usually in the range from
50 A to 80 A when quantum effects take into consideration. Thus, for a nominal 30 A
spacer layer, we see from Figure 2.17, that only a density o f about 3x10
12
2
cm' is
available in the case o f 5x10 18 cm ' 3 doping density o f the cap layer. This carriers directly
contribute to the normal 2DEG. Compared with that induced by the PZ and SP effects,
the amount o f this usual 2DEG is not significant.
2.3.5 Total sheet charge available in the channel
Two extreme cases were simulated: One is surface completely un-compensated (y
= 0) and the other is the surface fully compensated (y = 1). The results are shown in
Figure 2.18 and Figure 2.19, respectively. When the surface is uncompensated, the total
sheet charge density is low, the channel loses its charge due to the effect o f a l on the top
surface. As the doping density in the cap layer increases, the dopants screen a l , and as a
result, the channel sheet density increases. The channel charge will increase until the total
sheet density of ionized impurities (Nd T product) equals the net surface charge. Beyond
which a further increase o f doping density will not further improve the screening. In case
that extremely high doping densities, the potential well in the barrier is formed such that
there are significant free electrons in the well, resulting in parallel conduction. However,
if the surface is fully compensated, the maximum channel sheet charge density o f the
40
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
35
30
s
25
o
ns channel
ns parallel
©
*
0)
w
s-o
A
A
o
a»
v
20
nstotal
15
10
A
C/2
5
Doping in cap layer (c m ')
Figure 2.18. Sheet charge versus doping in the cap layer when the surface is
uncompensated (y = 0).
50
ns channel
40
£w
30
ns parallel
nstotal
w
V 20
b£
u
S3
A
u
10
44>
>
A
cn
Doping in cap layer (c m ')
Figure 2.19 Sheet charge versus doping in the cap layer when the surface is fully
compensated(y = 1). The sheet channel charge is almost independent to the doping in
the cap layer, dominated by charges induced by the PZ and SP effects.
41
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
channel is almost independent o f the doping density in the cap layer, and reaches near a
constant value o f about 1.5xl0 13 cm ' 2 for our standard structure. In this case, the PZ and
SP effects, a2, dominate channel sheet charge.
§2.4 Summary
Based on the calculations and simulations, we understand that due to the PZ and
SP effects, the 2DEG density in the AlGaN/GaN structure is significantly enhanced. To
properly take advantage o f these effects, one should carefully design the structure so that
the surface inversion problem, which is the by-product o f the PZ and SP effects, can be
avoided by selecting a suitable doping and thickness o f the barrier layer. Heavily doping
is helpful to reduce the sensibility o f the surface problem, but too high the doping will
cause parallel conduction in the barrier layer. Another quantity needs to keep in mind is
the critical thickness o f the barrier layer, which is inversely proportional to the Al content
o f the barrier. Once the barrier thickness exceeds the critical thickness, the layer is
partially relaxed and the 2DEG density induced by the PZ part will decrease. Such a
relaxation process may possibly continue during the whole life o f the device, exhibiting
performance degradation all the time.
42
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Bibliography:
1 O. A m bach er,a) B. Foutz, J. Sm art, J. R. Shealy, N. G. Weimann, K. Chu, M. M urphy,
A. J. Sierakow ski, W. J. Schaff, a n d L. F. Eastm an, ‘‘Two dim en sion al electro n g a se s in du ced by
spon tan eou s a n d p ie zo e le c tric p o la riza tio n in u n doped a n d d o p e d A lG a N /G a N h e te ro stru c tu re s”
Jou rn al o f A p p lie d P h ysics Volume 87, N um ber 11 January 2000, p p 3 3 4 -3 4 5
2 H adis M orkoc, “N itrid e sem icon du ctors a n d d e vic es ” S prin ger se rie s in M a teria ls Science.
E ditors: R. Hull. R. O sgood, Jr. H. Sadaki. A. Zunger. G erm any 1999 p p 69.
3 M .P osternak, A. B aldereschi, a n d A. C atellani, R. R esta. “ A b Initio S tu dy o f the Spontaneous
P o la riza tio n o f P y ro e le c tric B e O ” Phys. Rev. Lett. 9 A p ril 1990 p p 1777-80.
4 F. B ernardini, V. F ioren tin i a n d D. Vanderbilt: ” Spontaneous p o la riza tio n a n d p ie zo e le c tric
con stan ts o f lll- V n itr id e s " Phys. Rev. B 56(1997) R 1 0 0 2 4 -7
3 R. G aska, J. Yang, M. Shur, “ P iezo effect a n d g a te cu rren t in A lG a N /G a N high electron m obility
tr a n s is to r ” A ppl. P hys.L ett. 71(25), 22, D ec. 1997, p p 3 6 7 3-5.
6 E.T.Yu, X .Z .D ang, P.M . A sb e c k a n d S.S.Lau “Spontaneous a n d p ie zo e le c tric p o la riza tio n effects
in III-V n itrid e h e te ro stru c tu re s” J. Vac.Sci.Technol. B 17(4), Jul/Aug. 1999, p p l 742-9
G .M artin, A. B otch k a rev a n d H . M orkoc, “ V a len ce-b a n d d isco n tin u ities o f w u rtzite GaN, AIN,
a n d In N h eteroju n ction s m ea su red by x -ra y ph o to em issio n s p e c tr o s c o p y ”,A ppl. Phys. Lett.
68(1996),p p 2 5 4 1
8 M. S. Shur a n d G aska, “ G a N -b a se d tw o-dim en sion al electron d e v ic e s ” Ioffe Institute, 6th Int.
Symp. S t P etersbu rg, Russia, June
9 M. S. Shur, M a te ria l resea rch so c ie ty sym posium Proc. 483, 15(1998)
10 S ig frid Yngvesson, M ic ro w a v e sem icon du ctor devices, K lu w er a ca d em ic p u b lish e rs P P 369 -3 7 3
11 L. Wong, S .C a i a n d K .L . W ang M ag n eto tra n sp o rt stu dy on the tw o -d im en sio n a l electron g a s
in A lG a N /G a N h etero stru ctu res A p p lie d P h ysics L etters, Vol. 73, No. 10, 7 S ep tem b er 1998, pp.
1391-1393
12 M.Shur, b.G elm ont, a n d M .A s if K han “E lectron M o b ility in T w o-dim en sion al electro n G as in
A lG a N /G a N h etero stru ctu res a n d in B ulk G a N ” J. Elec. M aterials. Vol. 25 (5), 1 9 9 6 p p 7 7 7 -8 5
13 A .D .B ykhovski, B.L. G elm ont, a n d M. S.Shur, “E lastic strain relaxation a n d p ie zo e ffe c t in G aN AIN, G a N -A lG a N a n d G a N -I n G a N su p e rla ttic e s” J. A. P. 8 0 (9), 1 m ay 1 9 9 7 p p 6 3 3 2 -8
14 C. M onier, Fan Ren; Jung H an; P ing-C hih Chang; R. Shul, K y u -P il Lee; A n pin g Zhang; A. G.
Baca, S. P earton , “Sim ulation o f npn a n d p n p A lG a N /G a N h eterojunction b ip o la r tran sistors
p erfo rm a n ces: lim itin g f a c to r s a n d optim um d e s ig n ”, IEEE Transactions on E lectron D evices,
v o l.48, (no.3), IEEE, M arch 2001. p p .4 2 7 -3 2 .
43
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
15 J. D eng, B. I n ig u e z , M. S. Shur, R. G aska ,M. A. Khan, a n d J. W. Yang “M icro w a v e Sim ulation
on the P erform an ce o f H igh P o w erG a N /A lG a N H eterostru ctu re F ie ld E ffect T ransistors ”, phys.
stat. sol. (a) 176, (1999) p p . 205-8.
16 G .K o ley a n d M .G .S p e n c e r” Surface p o te n tia l m easurem ents on G a N a n d A lG a N /G a N
h eterostru ctu res b y scan n in g K elvin p r o b e m ic ro sc o p y ”, J. A ppl.P hys. Vol 90(1), Ju ly 2001, p p
337-344.
I S m orch kova IP, E lsa ss CR, Ib betson JP, Vetury R, H eyin g B, F ini P, H aus E, D en B aars SP,
S p eck JS, M ish ra UK. P o la riza tio n -in d u ced ch arge a n d electron m o b ility in A lG a N /G a N
h eterostru ctu res g ro w n b y p la s m a -a ssiste d m olecu lar-beam ep itaxy Jou rn al o f A p p lie d Physics,
vol.86, no.8, 15 Oct. 1999, p p .4 5 2 0 -6
18 G reen BM, Chu K K, C hum bes EM, S m art JA, S h ealy JR, E astm an LF. The effect o f surface
p a ssiv a tio n on the m icro w a ve ch a ra cteristics o f u n doped A lG a N /G a N HFETs. [J o u rn a l P a p e r]
IEEE E lectron D e v ic e L etters, vol.21, no. 6, June 2000, p p .268-70.
19 R. G aska, M.S. Shur, A .D . Bykhovski, A.O . O rlov, G.L. Snider, “E lectron m o b ility in
m o d u la tio n -d o p ed A lG a N -G a N h e te ro stru c tu re s”, A p p lie d P h ysics L etters, vol. 74, (no.2), AIP,
I I Jan. 1999.
44
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 3
EXPERIMENTS AND DEVICE FABRICATION
§3.1 Experiments
3.1.1 Hall measurement from different Al content structures
First o f all, based on the analysis above, we would like to see the PZ and SP
effects in AlGaN/GaN structure experimentally.
For this purpose, we designed the
structure as shown in Figure 3.1(a). Three samples were grown by MOCVD according to
the design. For the three samples, only difference was the Al content in the cap layer,
varying from 5% (sample A), 15% (sample B) to 25% (sample C). All samples were
undoped in the cap layer. Both Hall square and Hall bar (Figure 3.1(b)) were prepared.
Ohmic contacts were formed by depositing Ti/Al/Pt/Au (310A /1000A/300A /1000A)
multilayers o f metal, followed by a 900 °C, 35 seconds RTP annealing in N 2. Typical
specific ohmic resistance was 0.5 Q.mm. For the Hall bar, mesas were formed by using
ECR etching. Process details are to be discussed in the next section.
Hall results are shown in Figure 3.2 against temperature. For 5% Al, sample A,
the carrier concentration is 2 .1 x l0 12 cm'2; for 15% sample B, the concentration is 7.1
xlO
12 cm'2; for sample C the concentration is 1.04 xlO 13 cm’2 . The near constant carrier
45
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
concentration obtained in the measurement temperature range for all three samples
indicates that the 2DEG in the channel is the main contribution o f the sheet charge. The
measured concentration numbers are lower than what we predicted in the previous
Chapter by around 30 %. This may be resulted from the partial relaxation o f the cap layer
due to the fairly large thickness, hence the PZ contribution decreases compared to the
ideal un-relaxed case. Contrary to the case o f the concentration, the mobility decreases as
the Al content increases. At 10 K, the mobility is 4000 cm 2/Vs, 3800 cm2/Vs and 1600
cm2/Vs for sample A, B and C, respectively, indicating that the polar optical phonon and
PZ scatterings play an important role as predicted in the previous chapter. The mobility is
lower than the calculated value, because it is likely the interface is not perfect, which
causes extra scattering to the carriers, hence reduces the mobility.
50 nm, undoped
A lG a N
3 nm. und op ed A K ia N spacer
50 nm. undoped GaN
1.2 uni. u n d o p e d G a N b u f f e r
0.1 p m , u n d o p e d A I N
(a)
1
Figure 3.1 (a) Material structure designed for PZ and SP experiments. Three samples
prepared with different Al contents: x=5%, 15%, 25%>. (b) showing the hall bars with
Pd/Au as gate metalfor transport study.
46
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
10000
5% Al
1 5 % Al
nc
au
Mk—ir-**r-*inir
1000
*
-D
O
100
100
0
200
300
Temperature (K)
10 14
au
25% Al
C
o
'O
10 13
15% Al
JS
to
5% Al
10
12
50
100
150
200
250
300
Temperature (K)
Figure 3.2. Hall measurement results from three samples: at 10K, mobility of 4000
cm2/Vs and concentration of 2.1xl012 cm2for 5% Al content sample, 3900 cm2/Vs and
7.1 xlO12 cm2for 15% sample , and 1700 cm2/Vs and 1.04 xlO13 cm2for 25%> sample,
respectively.
47
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
3.1.2 Transport study1
Hall bars were used to perform magnetotransport study. Figure 3.3 shows typical
traces o f our magnetotransport data o f sample A and sample B. Here, the longitudinal
resistivity and the Hall resistivity are plotted against magnetic field B up to 30 T at
temperature T = 0.420 K. In the longitudinal resistivity, one series o f the Shubnikov-de
Hass oscillations is seen in magnetic field B below 10 T. In high magnetic field B, the
integer quantum Hall states are seen. The observation o f the quantum Hall effect gives
direct evidence o f the existence o f a 2DEG at the AlGaN/GaN heterostructure interface.
The electronic density was determined to be «/r=2.07xl012 cm
'2
from the low-field Hall
resistivity and n sdH=2.08 xlO 12 cm ' 2 from the Shubnikov-de Hass oscillations for sample
A. For sample B, the electronic density was determined to be n h =5.48 xlO 12cm
'2
from
the low-field Hall resistivity and n sdH =5.47 xlO 12 cm ' 2 from the Shubnikov-de Hass
oscillations.
Since the AlGaN epilayer was undoped, again, we believe that the 2DEG
channel is induced by the piezoelectric strain caused by the lattice mismatch at the
AlGaN/GaN interface and the spontaneous polarization as analyzed in previous Chapter.
All electrons are in the lowest subband. This can be seen since the Shubnikov-de Hass
oscillations have one period and n His equal to n SdH within our experimental uncertainty.
Thus, the undoped layers are semi-insulating at low temperature and showing good
sample quality.
The property that all electrons occupy one subband with such high density is ideal
for high mobility and high power HFETs. It is known that intersubband scattering
48
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
0.5
0.05
Sample A
0 .0 4
(a)
■ **—
<£
0.02
▲
II
>
0.03
s \
, vz l / '
,
V
T
A
a
0.01
\
*■
0.4
0.3
0.2
*
x.
X
a.
0.1
Sample B
0.010
<u
v = 3*
0.0 08
0 .0 06
0 .0 04
0.002
10
15
20
25
Magnetic field B(T)
Figure 3.3 Typical magnetoresistivity v.S' magnetic field B. Data of sample A at 0.42K in
(a) and of sample B at 0.54K in (b). The integer quantum Hall effect is clearly seen in
both figures and filling factors (v) of the corresponding quantum Hall states are
labeled.
reduces carrier mobility when more than one subband is occupied in a 2DEG. When all
electrons are in the lowest subband, HFETs can achieve high mobility.
One fundamental parameter in transport is the effective mass o f the 2DEG.
• 2
Magneto-optical studies
, such as cyclotron resonance (CR) have been applied to obtain
the value o f the effective mass o f the 2DEG in AlGaN/GaN heterostructures. However,
its value has not been checked directly by transport measurements. Here, we derived the
49
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Temperature (K)
Figure 3.4 The logarithmic of (Apxx/T)[l-exp(-2xr)] is plotted against temperature
T for the temperature dependence of pxx at four different values of magnetic field
corresponding to four extrema in the low-field regime. An average of 0.228 me was
obtainedfor effective electron mass in AlGaN/GaN system.
effective mass from the temperature dependence o f the low-field Shubnikov-de Hass
oscillations. In our approach, the oscillations are analyzed using the conventional Ando
formula3:
)
ZT=
2 7C2kRT
%C0c
eB
—
m
(3.1)
(3.2)
Where coc is the cyclotron frequency, and xq is the quantum lifetime. Thus, the effective
mass can be obtained by fitting the temperature dependence o f the oscillations with above
50
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
equation. We find that the average m*=0.228 me , where me is the free electron mass.
This value is in very good agreement with those obtained from magneto-optical studies
and the theoretically calculated CR mass 4 when the band non-parabolicity and the
screening o f the electron-phonon interaction are taken into account.
§3.2 Device fabrication techniques
The basic processing techniques to fabricate an AlGaN/GaN HEMT include mesa
isolation, source/drain ohmic contacts and formation o f the Schottky gate. Since at that
time (1997), not much processing data were available in the area, we spent many efforts
in exploring and developing proper techniques for the device fabrication. For power
devices, an air-bridge technique is also needed to connect multiple-cells together to
obtain high current. The following sections mainly present our base line process we
developed.
3.2.1 isolation
Owing to the chemical inertness o f AlGaN and GaN, proper wet etching can not
be easily obtained. UV photo-assisted wet etching o f GaN in KOH was possible5, but
usually resulted in a rough surface. However, many dry etching methods have been
successfully developed. Among them, Pearton’s group 6 first reported the smooth,
anisotropical dry etching of GaN layers using low-pressure (1-30 mTorr) CH 4 or C f, H 2
ECR discharges with additional DC biasing o f the sample and obtained about a 400
A/min etching rate. Adesida group 7 applied HBr reactive ion etching (RIE) to etch GaN
51
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
and achieved a 600 A/min etching rate and smooth etched surface. Chemically assisted
0
ion beam etching (CAIBE) was used by the same group 8 and a 2100 A/min rate was
achieved using a 500 eV Ar ion beam directed onto GaN sample in a C b ambient. Later
H. P. Gillis 9 reported the use o f low energy electron enhanced etching (LE4) on GaN and
very good surface quality was achieved.
In LE4, electrons at energies 1-15 eV and
reactive species at thermal velocities arrive at the surface. These electrons carry
negligible momentum to the etching surface, and thereby avoid the ion bombardment
damage intrinsic to RIE, ECR, and CAIBE, while enhancing etch chemistry to give
anisotropic pattern transfer.
To realize our first AlGaN/GaN HFET in late 1997, we used C ^ /C H ^ ^ /A r ECR
(flow rates o f 10, 3, 15 and 10 seem, respectively). The process pressure was 1 mT; the
microwave power was 850W, and the substrate temperature was 170°C.
In order to
protect the active region during ECR etch, a 1000 A Cr layer was used to protect the
surface. To prevent the Cr from contacting directly to the AlGaN active surface (to avoid
any possible unknown contaminations or surface changing during the process of
removing Cr), 2000 A SiC>2 was deposited as a buffer layer between the Cr and AlGaN
cap layer by PECVD at 200 °C.
Mesa patterns were formed by a standard lift-off
process, then followed by BOE dip to etch away extra SiC>2 . By etching off AlGaN/GaN
by 1500 A to
2000
A, with a rate o f 400 A/min, a l x l 0 7 Q.mm isolation resistivity was
routinely achieved. A typical AFM picture after ECR etching o f the mesa is shown in
Figure 3.5.
Ion implantation was our second choice for isolation process since this planar
52
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
0 '
■■'•Q
Figure 3.5 AFMpicture for ECR etched mesa
implantation isolation approach did not result in any steps which could have a negative
impact for fine gate patterning. This approach was first report by Pearton group10. In our
process, a 5 pm AZ4620 photoresist was used to protect the active region from ion
bombardment. The implantation recipe we used in sequence is:
As+ implant:
1.27 x 10 1Vcm2 @ 75 keV
4.32 xlO n /cm 2 @ 375 keV, followed by
He+ implant: 5.43 xlO u /cm 2 @ 75 keV
Multiple energy schemes were used to create a uniform damage region from the surface
to 2500 A. As+ was chosen by the fact that it is a heavy ion (leading, possibly, to damage
with relatively high thermal stability). The choice o f helium was dictated primarily by its
large penetrating depth into GaN material. The initial isolation resistance was close to
lxlO 10 Ohm/square. Temperature stability was examined under 500°C N 2 ambient high
temperature storage test (HTS). The isolation resistivity was monitored during the
53
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
temperature burn-in and the results were shown in Figure 3.6. The degradation of
isolation resistance during the HTS was observed and finally it stabilized at about lxlO 5
Ohm/square. Nevertheless, this isolation approach is good enough to serve our
subsequence device study.
cr
G
108
c
CG
( C1 on S apphire s u b s tra te )
G
UJ
a
•omt
"©
0
50
100
150
200
HTS time (hours)
Figure 3.6 Isolation resistance against high temperature(500°C) storage (HTS) time
showing the degradation of the isolation performance, indicating that this isolation
method was not ideafor GaN HEMTfor long term reliability performance.
3.2.2 Schottky Contact
Schottky contact to GaN has been extensively studied by many investigators11.
Unlike GaAs where the Femi-level at the surface is pinned at mid-gap owing to the high
density o f surface states, AlGaN surface exhibits different Schottky barrier heights with
54
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
different metals, although not quite following simple calculations from work function
difference. In particular, Pd, Pt, Ni and Au can form a relatively high Schottky barrier of
0.8 ~ 1.4 eV on n-GaN and similar results were observed on AlGaN/GaN structures12. In
order to achieve a well behaved Schottky junction, proper surface treatment before gate
metal evaporation is necessary.
In our process, after standard photo lithograph, O 2
descum was done, followed by a 30 seconds BOE dip to get ride o f possibly a thin
surface oxide layer.
After BOE dip, a few minutes DI wafer rinse and N 2 blow dry
followed immediately to reduce the ideality factor. Our baseline process can achieve an
n factor o f about 1.5.
Figure 3.7 shows a typical Schottky junction forward current-
voltage curve measured at room temperature. The leakage current visible in log scale
curve is dominant at low voltage region, showing the material quality or the surface
,0
3.E-03
■2
S ' 2.E-03
2> 1.E-03
1.E-10
■4
v g (V)
•6
0
0.5
1.5
1
2
2.5
Vg (V)
Figure 3.7 Forward Schottky characteristicsfor Au/Pt/AlGaN/GaNjunction in log scale
(semilog scale shown as inset). The gate size is of ljum x 20 ym.
55
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
passivation is needed to improve the characteristics. From the slope o f the thermionic
emission part of the curve, the ideality factor o f 1.45 is obtained. Other metals, Pd, and
Ni were also used for Schottky junction formation. These metals also exhibited good
Schottky barriers, which are slightly different from that from Pt. The barrier height was
different from the number obtained from metal work function and the AlGaN vacuum
energy level. One reason for this deviation is that the A1 content in the cap layer is not
accurate; another reason is the AlGaN cap layer is not defect free. There were surface
states but the density is not high enough to pin the surface. Therefore, the barrier height
changed for different contact metals but was not determined solely by the metal itself.
After high temperature storage test, the results showed that Au/Pd/AlGaN/GaN Schottky
was very stable under 500 °C for 128 hours. Since the barrier height was not crucial to
our power HFETs performance, we decided to choose this Au/Pd metal scheme as our
base line process.
3.2.3 Ohmic Contact
Ohmic contact to the source and drain are important to achieve good microwave
performance since the contact resistance degrades the extrinsic DC gain, gm, and hence
1f
the cutoff frequency, fT as the following :
(3.3)
_ §mo (1 ~ g d s ( R s + R d ) )
£>n
"
~
1 "t" Rsgmo
(3.4)
fT =
(3.5)
^ ------------
2 ^ ( q , + c , v)
56
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Since gds is usually very small, Eq. 3.3 can be simplified as:
( 3 -6 )
§m ~ § m c /(l ~^Rs§mo)
This result shows that the series resistance o f the source, Rs, is a key factor. Be noticed
that since Rs is composed of two parts, one from the contact, the other from the channel,
it is necessary to examine the two Rs sources at the same time. Figure 3.8 shows the
normalized extrinsic gm decreases as the contact resistance increases for different sheet
channel resistance (assuming the intrinsic gmo=500 mS/mm, and the gate to source
separation (Lgs) 1pm). Figure 3.9 depicts the normalized extrinsic gm but takes the source
to gate distance as a parameter (Assuming the intrinsic gmo=500 mS/mm, and the channel
sheet resistivity 500Q/sq). From the Figures, it is seen that the contact resistivity could
seriously degrade device gm, depending upon the channel sheet charge and the device
0.6
«
0.4
S
s©
0.2
Contact resistivity (Q.mm)
Figure 3.8 Normalized transconductance versus contact resistivityfor various channel
resistance.
57
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
'O
<u
.a
13
S
u
o
Z
0.4
0.2
Contact resistivity
Figure 3.9 Normalized extrinsic gmvs contact resistance for various source and gate
separation distance.
structure. The lower the sheet charge, the bigger the impact o f the contact resistance on
device performance. To reduce the gm degradation within 10% caused by ohmic contact,
for a general case, with a 500 Q/sq sheet channel charge material and a source-to-gate
distance o f 1pm for example, one should reduce the Ohmic contact resistivity to less than
0.5 Q. mm.
According to our experience, making a good Ohmic contact to a wide band gap
semiconductor is o f great challenge, Table 3.1 list the device results we obtained during
our early stage processing (1997 to 1998), showing that gm’s were quite low. One reason
could be related to poor material quality; another main reason for the poor performance
was attributed to the poor ohmic contact, which is also listed in the Table.
Table 3.1 Our early device results showing poor performance due to poor
contacts
58
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Rc
Substrate (supplier)
(mS/mm)
55
Vt
(V)
@ gm= 0
-4.5
(Q.mm)
13.9
SAPPHIRE (SVT)
500
105
- 1 .8
4.4
SAPPHIRE (APA)
LA14
450
70
-4.0
5.5
SAPPHIRE (APA)
LA23
296
70
-1.5
5.6
SiC (APA)
LA7
350
55
-5.0
4.2
SAPPHIRE (APA)
LA25
260
50
-1.5
7.8
SiC (APA)
A T M Il
116
30
-2 .0
6.5
SiC (ATMI)
ATMI_2
140
48
-2 .2
21
SiC (ATMI)
Ids
gm
FET31
(mA/mm
500
LAI 1
In general, a material with a higher doping concentration cap layer can lead to a
lower contact resistivity and so does a material with a lower A1 content (or smaller band
gap). Otherwise, unless a metal semiconductor interaction takes place and/or the
semiconductor itself is altered with damage, nitrogen vacancies resulting in an increased
electron concentration, it is difficult to form ohmic contact to wide band gap AlGaN
material. If the damage or the interaction causes the surface electron concentration to
increase, significant band bending can occur. This may result in a very thin barrier and
increase tunneling. Figure 3.10 shows the inverse o f ohmic contact resistance versus N sp
products, depicting that samples with a higher Nsp product have better ohmic contacts.
Formation o f ohmic contacts to n-type AlGaN is possible by selecting a contact
metal with an appropriate work function since AlGaN did not exhibit Fermi level
59
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
1.4
S
1.2
a
1
B
&
LA8
L A 15
ATMI 2
0.8
0.6
0.4
LA11
0.2
I
F E T 3 2 FET31
R O CK293
*
LA7
♦ ♦
0
0
1
0.5
Ns*u
1.5
(10 16/Vs)
Figure 3.10 1/RCvs Nsju. Ohmic contact was easier to form for higher Ns p product
evi layers.
pinning, in contrast to many o f the III-V compound semiconductors.
Therefore, any
metal with a work function less than or equal to the electron affinity o f AlGaN (~4.2eV)
should form an ohmic contact to n-type AlGaN, if the interface was clean. TiN has a
work function o f 3.74 eV 14, making it a suitable candidate for ohmic contact formation to
n-type AlGaN. Metallic Ti has a work function o f 4.18eV, comparable to that o f GaN.
Therefore, Ti formed a near-ohmic contact.
For this process, Ti was deposited first,
followed by a rapid thermal annealing (RTA) at 900 °C for 35 seconds, and the resulting
resistivity dramatically decreased.
This was attributed to the interfacial reaction to
produce TiN as :
60
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
which produced a thin layer of TiN at the Ti/GaN interface.
Different rapid thermal annealing (RTA) conditions gave different ohmic results.
Transmission lines (TML) and arrays o f dots with a diameter o f 100 pm were used to
evaluate the contact resistance. Figure 3.11 depicts the TML patterns we used with 5 pm,
10 pm, 15 pm, 20 pm and 25 pm gaps for resistance measurement. Figure 3.12 shows the
ohmic contact changes with different RTA conditions. Contact resistance was measured
from constant distance dots against RTA time at 850 °C, 900 °C and 950 °C in N 2 ambient
for different samples. The result shows that 900 °C with 40 seconds annealing time lead
to the best contact resistance. Note that the RTA temperature plays a key role not only
1
Figure 3.11 Picture of transmission line patterns.
for the ohmic contact formation but also for the morphologic quality o f the resulted
contact. Lower temperature usually gave a better morphology but with a higher ohmic
contact resistance. New metal scheme such as Ta/Au was also explored in our process to
improve the morphology but was not pursued further due to the lift-off difficulties caused
61
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
by the high temperature evaporation o f Ta.
Pre-cleaning before metal deposition is also important. Prompt metal deposition
after surface treatment was required for good results. Figure 3.13 shows the ohmic
contact resistance degraded as a function o f time exposed to air after surface treatment.
The contact resistance was measured with constant distance dots; sample surface was
treated in O 2 plasma (100 W 60 sec descum) and
N H 4O H
solution for a 10 sec dip before
ohmic metal deposition. The degradation suggests that a thin insulating layer could have
formed on the surface, similar to the results obtained from the Schottky formation
process. The optimal pre-cleaning before metal deposition was determined to be an O 2
plasma descum (100W 60 sec) followed by a rinse in a diluted
N H 4O H
solution (1:20 for
10 sec). The optimal metal system and RTA conditions are Ti/Al/Ni/Au (200 A/800
A/400 A /1500 A) and 900 °C, 35 s RTA annealing in N 2 ambient, respectively, which
4000
3500
3000
♦ 900C(FET3.2)
—HI— 850C(FET3.2)
► 950C(FET3.2)
—O —900C (GaN RTD)
§
2500
S
2000
5
1500
__f
900C (FET3.1)
Q
1000
500
0
10
20
30
40
50
60
Annealing time (s)
Figure 3.12 Contact resistance vs RTA annealing timefor different samples.
62
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
100
t
£▲
a
50
50“
35 q
g
O h m ic C o n ta ct
33 q
Q
«
T i/A I/N i/A u
•Hi
'
10
I—
0.1
—........ — ;................
1
^ ----- ----—
10
100
Elapsed time after surface treatment (hours)
Figure 3.13 Contact resistance vs elapsed time after surface treatment. The contact
performance degraded after the samples exposed in airfor longer time before metal
deposition.
gave the best results yielding a contact resistivity as low as 0.14 Q.mm. The specific
contact resistance was 4.42xl0‘7 f lc m 2 for this case. This was one o f the lowest numbers
ever achieved for AlGaN/GaN material system15. The results are shown in Figure 3.14.
Similar results were also obtained from the Ti/Al/Pt/Au system.
We have tested tens o f samples. At any rate, we can confidently state that ohmic
contact resistivity o f below 10' 6 £ lcm 2 can be obtained for AlGaN /GaN structures for A1
content less than 30%; this number is suitable for power HFETs .
A process flow chart for the whole device fabrication process can be found in the
appendix II.
63
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
12
1 4 .7 7 6
y = 0 .4 6 3 7 X + 0 .1 4 3 2
R2 = 0 .9 9 9 9
0.143 Q.mm
2
4.42x1 O' 7 flc m '
0
0
20
10
30
TML gap (pm)
Figure 3.14 Ohmic specific resistance measured by transmission line(TML) for pi
sample. For this sample, a record ohmic contact number of 4.42x10' 7 Q.cm2 was
3
achieved. The sample had 30% Al content, and a cap layer doped with 5x1018cm'.
§3 .3 summary
We have characterized samples with different Al contents and the results showed
that higher Al content samples had a higher sheet concentration. Mobility for higher Al
sample was lower due to the stronger polar optical phonon scattering and piezoelectric
scattering in the structure.
The effective electron mass in the structure was measured from a magnetic
transport study. The effective mass equals to 0.228 me, which is in good agreement with
that obtained from other researchers.
64
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
A baseline process for fabricating AlGaN/GaN HFET was established.
Ohmic
contact formation and characterization were a special focus in the discussion, since it was
hard to achieve with the wide band gap material. Surface pre-cleaning and RTA are the
key processes for forming a good ohmic contact. A record low contact resistance of
4.42xl0‘7 Q.cm 2 was achieved using an optimal process. For the gate Schottky metal
contact, Pd/Au metal scheme was chosen.
In the next chapter, AlGaN/GaN HFET devices are to be fabricated using this
baseline process.
Bibliography:
1L. W. Wong, S. J. Cai, R. Li, and Kang Wang “Magnetotransport study on the two-dimensional
electron gas in AlGaN/GaN heterostructures’’ Applied Physics Letters 73, Number 10 Sept. 7
1998ppl 391-93
2Knap W, Alause H, Bluet JM, Camassel J, Young J, Asif Khan M, Chen Q, Huant S, Shur M.
The cyclotron resonance effective mass of two-dimensional electrons confined at the GaN/AlGaN
interface. [Journal Paper] Solid State Communications, vol. 99, no. 3, July 1996, pp. 195-9
3Ando T, Fowler AB, Stern F. Electronicproperties of two-dimensional systems. [Journal Paper]
Reviews of Modern Physics, vol.54, no.2, April 1982, pp.437-672.
4XWu, Peeters FM. Cyclotron-resonance mass of two-dimensional electrons in GaN/Al/sub
x/Ga/sub 1-x/Nheterostructures. Physical Review B, vol.55, no.23, 15 June 1997, pp. 15438-40.
3 Cho, H.; Auh, K.H.; Han, J.; Shul, R.J.; Donovan, S.M.; Abernathy, C.R.; Lambers, E.S.; Ren,
F.; Pearton, S.J. UV-photoassisted etching of GaN in KOH. Journal of Electronic Materials,
vol. 28, (no. 3), (Proceedings of the 40th Electronic Materials Conference (EMC), Charlottesville,
VA, USA, 24-26 June 1998.) TMS, March 1999. p.290-4
6 Pearton, S.J.; Abernathy, C.R.; Ren, F.; Lothian, J.R.; Wish, P.W.; Katz, A.; Constantine, C.
Dry etching of thin-filmInN, AINand GaN. Semiconductor Science and Technology, vol. 8, (no.2),
Feb. 1993. p. 310-12.
7 Ping, A.T.; Adesida, L; Asif Khan, M.; Kuznia, J.N. Reactive ion etching of gallium nitride
using hydrogen bromide plasmas. Electronics Letters, vol. 30, (no.22), 27 Oct. 1994. p.1895-7.
65
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
8 Adesida, I.; Ping, A.T.; Youtsey, C.; Dow, T.; Asif Khan, M.;Olson, D.T.; Kuznia, J.N.
Characteristics of chemically assisted ion beam etching of gallium nitride. Applied Physics
Letters, vol. 65, (no. 7), 15 Aug. 1994. p.889-91.
9Gillis, H.P.; Christopher, M.B.; Martin, K.P.; Choutov, D.A. (Edited by: Pearton, S.J.; Kuo, C.;
Wright, A.F.; Uenoyama, T.) Patterning III-N semiconductors by low energy electron enhanced
etching. GaN and Related Alloys. Symposium, (GaN and Related Alloys. Symposium, GaN and
Related Alloys. Symposium, Boston, MA, USA, 30 Nov.-4 Dec. 1998.) Warrendale, PA, USA:
Mater. Res. Soc, 1999. p.G8.2.1/9pp.. 1028pp. 33 references.
10S. C. Binari,a) H. 8. Dietrich,-G. Keiner, L. B. Rowland, and K. Dover “H, He, and N implant
isolation of n-type GaN”J. Appl. Phys. 78 (5), 1 September 1995 pp3008
11Lei Wang and M.I.Nathan “ High barrier height GaN Schottky diodes: Ti/GaN and Pd/GaN”
Appl. Phys. Lett. 68(9), 26 Feb 1996, ppl267
12Yu, E.T.; Dang, X.Z.; Yu, L.S.; Qiao, D.; Asbeck, P.M.; Lau, S.S.; Sullivan, G.J.; Boutros, K.S.;
Redwing, J.M. Piezoelectric enhancement of Schottky barrier heights in GaN-AlGaN HFET
structures. 56th Annual Device Research Conference Digest New York, NY, USA: IEEE, 1998.
p.116-17. viii+145 pp
1?
Yuan Taur and Tak H. Ning “Fundamentals of Modern VLSI devices” Cambridge
university press1998, pp222
14 Glass RC, Spellman LM, Davis RF. Low energy ion-assisted deposition of titanium nitride
ohmic contacts on alpha (6H)-silicon carbide. [Journal Paper] Applied Physics Letters, vol. 59,
no.22, 25 Nov. 1991, pp.2868-70.
15S.J. Cai, R. Li, Y.L. Chen, L. Wong, WG. Wu, S.G. Thomas and K.L. Wang "High Performance
AlGaN/GaN HEMT with Improved Ohmic Contacts" IEE Electronics Letters 26th November
1998 Vol. 34 No. 24 pp2354-6
66
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 4
BASIC AlGaN/GaN HFET DEVICE
This chapter describes the development o f our basic AlxGai_xN/GaN HFETs.
Testing structures on conductive SiC substrate are prepared to see the Al content
variation effect to the DC performance. It is verified that the higher saturation current
can be obtained from higher Al content structures. Then structures with 25% and 40% Al
content are selected to grown on semi-insulated SiC substrate for a complete study. DC
characteristics, high temperature performance and RF results are presented.
§4.1
Testing
AlGaN/GaN
HFETs
structure
on
conductive
SiC
substrates
Al content is one o f the key parameters in our studies o f AlGaN/GaN HFETs. To
study this, four samples were prepared with different Al contents: 15%, 20%, 30% and
40%.
For DC performance test purpose, the epi-layers were grown on conductive
substrates to reduce the cost. The structure schematic is shown in Figure 4.1.
Hall measurement results showed that the mobility for all four samples ranged
'y
from 700 to 900 cm /Vs in room temperature; there was little change as the Al content
67
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
AlxG a^N
AlxGa^xN
l-G a N
200 A
3x10lscm '3
30 A
U ndoped
1pm
B uffer layer
conductive S iC sub strate
Figure 4.1 Testing structure used to fabricate devices. Four samples were prepared for
different Al contents: x=0.15, 0.2, 0.3 and 0.4.
was varied. However, the sheet charges increased for higher Al content, as showing in
Figure 4.2. The concentrations o f three samples dropped at low temperature (77K)
compared to the room temperature value probably due to a parallel conduction as
discussed in section §2.3.
A picture o f a device fabricated is depicted in Figure 4.3. The DC characteristics
for four structures are shown in Figure 4.4 through Figure 4.7. All o f them show good
ohmic contact resistance, (The transmission line measurement yielded a contact
resistance o f around 0.4 Q.- mm). The device saturation current densities for structures
with 15%, 20%, 30% and 40% Al contents were 140 mA/mm, 500 mA/mm, 700 mA/mm
and 850 mA/mm, respectively, indicating that higher Al content devices had higher Ids,
as expected from the previous analysis based on the PZ and SP effects, and hence
68
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
4x10
13
♦ N(300K)
■ N(70K)
3x10 13
a
#o
*-*j
a
s-
-M
e<u
2 x 10
13
1 x 10
13
sj
S
ou
-g
4)
4>
J3
(Z5
0%
10%
20%
30%
40%
50%
Al contents
Figure 4.2. Hall measurement results showing that sheet concentration increases
as Al content increases.
Figure 4.3 Picture of afabricated AlGaN/GaN HFET device.
69
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
pointing out that higher Al content structures seem to be preferred for power
performance. Pinch-off voltages for these different structures were -0.3 V, -1 V, -1.6 V
and -3 V, respectively. The pinch-off voltage increased as the Al content increased. This
was resulted from a higher 2DEG in the channel for the higher Al content, which was in
turn resulted from the stronger PZ and SP effects. Breakdown voltage were generally
higher than what could be obtained from GaAs based HFET, but was still not quite
satisfactory, particularly for the 40% Al content device. One reason was that for higher
channel conductivity as the case o f 40%, a smaller part o f voltage was shared in the
channel. Thus we need to design a device structure to have both high current density and
high breakdown voltage at the same time. This topic will be discussed in detail in the
next chapter. Besides the breakdown issue, another problem with the higher Al content
structure was morphology problems. When the Al content was higher than 30%, the
visible particle density was very high after MOCVD growth. As a compromise, most
structures we designed had an Al content o f 25% to 30%.
Table 4.1 summarizes the DC results o f the devices o f the test structures.
Table 4.1 Summary of the DC results with different Al contents
15 %
20%
30%
40%
I ds (mA/mm)
140
500
700
850
g m (mS/mm)
80
170
190
195
Vth(V)
-0.3
-1
- 1 .6
-3
Vbr(V)
82
63
51
46
Al content
70
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Al 15%
Vgs=O V
s te p : - 1 V
(y) spi
0
4
2
6
Vds(V)
90
160
80
140
Vds=8v
70
60
100
50
40
30
Ids (mA/mm)
120
20
10
0
2
1
0
1
2
3
Vgs(V)
Figure 4.4. DC characteristics for a test device with 15 % Al content, showing
the peak g m of 80 mS/mm and f s of 140 mA/mm.
71
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
Al 20%
Step: -1 V
<
Ta3
0
4
2
0
6
v ds(V)
200
.8
t/3
J,
e
W
D
180
160
140
120
500
V ds = 8 V
400
300
100
80
60
40
200
Ids (mA/mm)
s
600
100
20
0
2
1
0
1
2
3
4
Vgs(V)
Figure 4.5. DC characteristics for the test device with 20 % Al content, showing
the peakgm of 170 mS/mm and 7dSof500 mS/mm.
72
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Al 30%
S tep :-1 V
0
4
2
6
Vds(V)
250
800
700
200
500
400
300
200
50
Ids (mA/mm)
600
100
2
1
0
1
2
3
4
Vgs(V)
Figure 4.6. DC characteristics for the test device with 30 % Al content, showing
the peak g mof 190 mS/mm and / ds of 700 mS/mm.
73
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Al 40%
1.4x1 O'2
Step: -1 V
1.0x1 O'2
2
£
6 .0 x 1 O'3
2 .0 x 1 O'3
0
0
4
2
6
V d s(V )
1000
250
900
200
V ds = 8 V
800
? 150
J
GO
S
600
500
400
100
E
6£
300
Ids (mA/mm)
700
200
100
50
0
5 - 4 - 3 - 2 - 1 0
1
2
3
4
Vgs (V)
Figure 4.7. DC characteristics for the test device with 40 % Al content, showing
the peak g n of 195 mS/mm and I ds of850 mS/mm.
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
§4.2
AlGaN/GaN HFET with 25% Al content on semi-
insulating SiC substrate
To establish a base line device for general study, we focused on a structure with
Al content o f 25% first, a relative safe number in avoiding strain relaxation and yet to
have a high current density. The structure is shown in Figure 4.8 with a lower doping
density of 4.15 x 1018 cm'3 being chosen to avoid parallel conduction.
The material
growth was performed at ATMI Inc., using MOCVD (Metallic Organic Chemical Vapor
Deposition).
The growth started with a 100 A AIN nucleation layer on a 4H semi­
insulated SiC substrate. This was followed by the 1.2 pm GaN buffer layer, with a
background doping o f about l x l 0 16 cm'3. The active layer was consisted o f a 30 A Al
Ga
0.75
0.25
N spacer layer, a 150 A Si-doped charge supply cap layer (n = 4.15x 1018 cm'3)
and a 50 A unintentionally doped top layer. The ohmic layer was formed by Ti/AI/Ni/Au
(200 A/800 A/400 A/1500 A) deposition followed by 900 °C 35 seconds rapid thermal
annealing in N 2 ambient; this yielded a typical specific resistance o f 0.4 Q.mm measured
by the transmission line method. Isolation was realized using As+ and He+ implantation in
sequence. Gate metalization with Pd/Au (200 A/6500 A) was carried out using e-beam
evaporation. The gate length was 1pm and was formed by optical lithograph. The gateto-source distance and the gate-to-drain distance were both 1pm.
CV profiling for the sample was done using a standard Mercury probe at a
frequency o f 100 KHz. The results are shown in Figure 4.9, revealing that the peak
carrier density was located at the AlGaN/GaN interface. Hall measurement results on the
75
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
structure showed a room temperature mobility o f 1050 cm2/Vs and a sheet charge density
of 8 x 10 12 cm'3.
A lx G a i_> N
A lxGa-i_xN
150 A
A !xG a i . xN
l- G a N
A Undoped
50
30
A
1pm
4 . 1 5 x 1 0 iac m '3
Undoped
B u ff e r la y e r
S I S iC s u b s tr a te
Figure 4.8 Material structure for our basic HFET study (sample 817d).
10
20
10 19
ao
fl
u
8
a
o
U
10 18
10 17
io
16
10 15
1 0 14
U .0 0 0 .0 2
0 .0 4 0 .0 6
0 .0 8
0 .1 0 0 .1 2
0 .1 4 0 .1 6
0 .1 8 0 .2 0
Depth (pm)
Figure 4.9. Mercury CVprofiling for the structure shown above, indicating the peak
2DEG located at 200 A where the nominal AlGaN/GaN interface is located.
76
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
4.2.1 Device DC performance
The breakdown voltage for a typical device is around 55 V, as shown in Figure
4.10. Note that the breakdown voltage is much higher than that obtained from a
conventional GaAs based HFET. The leakage current at low voltage was most likely
contributed from generation-recombination in the depletion region and/or the surface
leakage. The output I-V characteristics o f the device are shown in Figure 4.11. There is
no significant drain current drop due to the self-heating effect, suggesting that SiC
substrate serves as a very good thermal sink for this
20
pm small device before dicing.
Also evident is the low turn-on Ohmic contact resistance as seen from the IV curve near
the knee. A current density o f 800 mA/mm (at Vgs = 2 V) and a near linear
transconductance (gm) with the maximum value o f 165 mS/mm were obtained as depicted
in Figure 4.12. The pinch-off voltage o f the device was - 4 V.
-60
-50
-40
-30
-20
-10
0
Voltage (V)
Figure 4.10 Breakdown voltage of the device between the gate and the drain. (Sample
817d, Lg=l pm, Wg=100 pm, the distance between gate and drain is 1 pm).
77
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
0
4
2
0
6
10
8
Vds(V)
Figure 4.11. On-wafer measured DC output IV characteristics for 817d sample
(gate dimension: 1 pm by 20 pm).
180
900
ds
160
800
140
700
B
120
600
;-fi
100
500
£
80
400
60
300
40
200
20
100
J
£
B
a
WD
0
4
3
2
1
0
1
2
v gs (V)
Figure 4.12. On-wafer measured transfer curve and transconductance for 817d
sample. A 800mA/mm current density and a 165 mS/mm peak transconductance
were obtained, (gate dimension: 1 pm by 20 pm)
78
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
4.2.2 High temperature performance
The DC performance o f the GaN HFET was characterized at various temperatures
from 25 °C to 592 °C. IV curves at 311°C and 592 °C are shown in Figure 4.13(a) and
(b), respectively.
Impressively the device was still functional at 592 °C, showing the
intrinsic advantage o f this wide-band-gap GaN based device for high temperature
operation. The saturation current and gm dropped as the temperature increased as shown
in Figure 4.14. This data suggests that the saturation velocity o f carriers in the channel
decreased as the temperature increased due to increased phonon scattering. Besides, as
we can see from Figure 4.13(b), significant leakage current appeared at 592 °C. IV
measurements from isolation patterns showed that the leakage was mostly due to the
isolation degradation. As we pointed out before, mesa etch could be used for isolation
process instead o f ion implantation isolation, and this problem could be avoided.
Step = -1V
0
'
0
2
4
6
8
10
Vds(V)
Figure 4.13(a) High temperature IV characteristics taken at 311°C.
19
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
4x10 -3
3x10 -3
^
Step: -1 V
2x10‘3
1x10
-3
q
#■---*■
v ds(V)
Figure 4.13(b) High temperature IV characteristics taken at 592°C.
700
S
500
i-fi
500
s
BJD
ns 400
a93
300
B
B
|
200
100
Idss ( V g = l V )
■■
gm (max)
A A A
‘AA AAAA, Ak
200
400
A iA
600
Temperature (°C)
Figure 4.14. High temperature saturation current(Idss) and the maximum
transconduction (gm) of the AlGaN/GaN HFET.
80
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
4.2.3 RF performance
Small signal RF performance was characterized with a HP8722E Network
analyzer. The highest current cutoff frequency o f 10 GHz was measured at a VdS of 10 V
while the peak maximum oscillation frequency o f 30 GHz was obtained, as shown in
Figure 4.15. The frequency performance can be improved by shrinking the gate length,
as we will see in the next chapter.
Single large devices with a 5 mm gate length were used in designing power
amplifier.
At an operating voltage o f 27 V, a total power o f
8
W at 9.0 GHz was
60
50
40
30
20
o
10"
Frequency (Hz)
Figure 4.15. The cutoffand maximumfrequencies of the sample: 10 GHz and
30 GHz, respectively.
81
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
achieved, with a PAE o f 28% and a 4 dB gain as shown in Figure 4.16. When input
power Pin = 20 dBm, the small signal gain was 5 dB. Although it had a lower gain due to
the long optical gate length (1pm), the results o f
8
Watts at X band frequency shows
excellent RF power potential o f the AlGaN/GaN HFET. A picture o f this amplifier is
shown in Figure 4.17.
10
30
20
40
Input Power (dBm)
Figure 4.16 Performance of a power amplifier made from the 5 mm single
device. A CWpower level of 8 Wat 9 GHz CW was obtained.
82
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 4.17 Picture o f the assem bled amplifier.
§4.3 Device with 40% A1 content on SI SiC substrate
After improving growth techniques, Sample F2 with a similar structure as 817d,
but with a higher A1 content o f 40% in the cap layer was designed and grown using
MOCVD. Hall measurements showed a room temperature mobility o f 1339 cm /Vs and
concentration o f 1.23 x 10 13 cm"2. Compared to the 25% A1 content structure in the
previous section, the concentration indeed increased for this high A1 content device. The
mobility did not decrease for high Al. This improvement was attributed from the
improved interface quality. At low temperature o f 77 K, this structure had a mobility of
5365 cm2/Vs and the concentration o f 1.22xl0 13 cm"2. A small carrier concentration drop
at low temperature meant that almost all carriers were contributed from the 2DEG
channel. HFET devices were fabricated based on this layered structure using 0.8 pm
83
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
optical lithograph for the gate. The DC characteristics o f a small device with a 20 pm
gate width are depicted in Figure 4.18, showing a peak gm o f 160 mS/mm and a
saturation current density o f 0.85 A/mm(at Vgs=0 V). The current density reached 1.08
A/mm at Vgs = 1.5 V ; at that point, no significant gate leakage current was observed. It
further confirmed that higher A1 content structures can result in higher current density.
Vds was measured up to 40 V in I-V curve, as illustrated in Figure 4.19. Further
measurements showed that the device had a breakdown voltage o f about 45 V (not
shown). Fligh temperature DC performance was also evaluated up to 400 °C. The results
are shown in Figure 4.20. As can be see from the figure, the Ids and gm drop to half o f that
at room temperature. Compared to Sample 817d, the present F2 shows a better high
temperature performance. This is probably contributed from the material quality
improvement. Based on the high temperature shelf experiments for Hall samples, poor
2x10'
Step= - IV
......................
0
o
2
4
6
8
10
Vds (V)
Figure 4.18 IV characteristics for sample F2 (gate dimension: 0.8 pm x 20 pm).
84
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
quality materials had shown a faster degradation. Figure 4.21
shows the DC
characteristics for a device with 20 gate fingers, having a total gate width o f 2.56 mm.
Close to 2 A current (at Vgs = 0 V) was measured from this device, indicating an
excellent scaling-up capability o f the device. Due to high power dissipation and thus
heating during measurement, current dropping at high power levels was evident.
A power amplifier was designed using a device with a total gate width o f 1.28
mm. Results are illustrated in Figure 4.22. Due to the relatively low breakdown voltage
(44 V), the power density was limited to 1.5 W/mm at the operation voltage o f 18.2 V. In
order to improve RF performance, particularly the power-add-efficiency and the power
gain, further efforts are needed.
8x1 O'3
Vgs = -3 V, Step: -0.5 V
0
0
10
20
30
40
50
Vds(V)
Figure 4.19 IV characteristics for sample F2 with a Vjs bias of 40 V.
85
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
1000
a
a
800
czi
B
B
be
600
400
B
a
a
>m
200
0
0
200
100
300
400
Temperature (°C)
Figure 4.20 High temperature performance ofF2 device (gate dimension: 0.8 pm by 20 pm)
2000
Vgs=-6
•« Vgs--5
_ i _ Vgs=-4
-»«—Vgs=-3
-*-V gs=-2
Vgs=-1
—i—Vgs=0
1600
<
1200
a
800
400
0
0.0
6.0
3.0
9.0
12.0
Vds (V)
Figure 4.21 DC characteristics for F2 device with a total gate width o f 2.56 mm.
86
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
20
35
30
=
«
2
w
u
*o
D
-*■<
3
C
3
o
15
25
20
10
15
* — G ain(dB )
5
«8
tt
w
#S
‘3
O
♦ — P o (d B m )
10
W
<
Oh
*-E ff(% )
0
0
5
10
20
15
25
30
Input power (dBm)
Figure 4.22 Power amplifier data measuredfrom a 1.28 mm device (sample F2).
§4.4 Summary
In this chapter, AlGaN/GaN HFET devices were fabricated. It is confirmed that
higher A1 content structures could deliver higher saturation current density.
Over 1
A/mm current density was achieved from the 40% A1 content AlGaN/GaN structure. The
results with good pinch-off voltage, saturation current and RF performance were
presented. A power amplifier o f
8
Watt output operating at 9 GHz in CW mode was also
achieved for a single 5 mm AlGaN/GaN HFET, translating into a CW RF power density
o f 1.6 W/mm at X band, which is about 3 times o f a GaAs based device. Higher A1
content structures offered a higher current density, but had a lower breakdown voltage.
Hence the power density improvement was limited. Further improvement o f the DC and
RF performance are to be discussed in the next chapter.
87
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 5
HIGH PERFORMANCE AlGaN/GaN HFET
High performance device requires high breakdown voltage, high current density
and high frequency performance at the same time. This chapter focuses on how to
improve the DC and RF performance o f AlGaN/GaN heterostructure field effect
transistor with the gamma shaped gate structure we developed. With such a gate, a sub
micron length gate and a field plate are formed simultaneously. Hence both DC
breakdown and RF performance are improved. With the gate-to-drain distance of 1.7 pm,
a breakdown voltage o f over 110 V is achieved, compared to that o f approximately 50 V
for normal gate devices. Both devices have a saturation current density o f 850 mA/mm.
To explain this, the 2D SILVACO® simulation is performed. The simulation results show
that the peak electric field in the cap layer o f the device is reduced from
6
MV/cm to 4
MV/cm at a drain-gate bias o f 50 V due to the presence o f the field plate. Meanwhile the
cutoff frequency (fr) is also increased from 10 GHz to 26 GHz. The temperature
dependence o f the breakdown behavior is studied showing that the avalanche breakdown
is the dominant mechanism.
Conventional methods to increase the breakdown voltage usually involve
increasing the gate-drain distance (to over 5 pm).
But in general, the distribution of
88
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
electrical field between the gate and drain is not uniform and thus increasing the gate to
drain distance works only to a certain extent.
Further increase o f the gate-drain
separation will increase the drain series resistance and hence degenerate the power
performance o f the device1,2. In addition, a small gate length is required in order to
improve the RF performance. The conventional approach is to use an E-beam writer or a
stepper to define the fine gate length.
In order to improve the breakdown voltage, a field plate was previously used for
Si high voltage devices 3 and for AlGaAs/GaAs HFET4. A similar technique has also
been applied to AlGaN/GaN HFET to increase the breakdown voltage o f the devices, but
only 20% of improvement in breakdown voltage was obtained 5. Due to the large
separation between the gate and the drain, the structures used are only good for relatively
low frequency applications. In this chapter, we will describe a simple process to
simultaneously form both a “gamma shaped” sub micron size gate and a field plate. With
this gate structure, a small gate o f 0.3 gm is achieved and breakdown voltage is also
improved. Both RF and I)C performance is shown to improve significantly, both in
simulation and experiments.
§5.1 Gamma gate process
The process flow-chart for the gamma gate fabrication is shown in Figure 5.1. A
sacrificial layer o f PECVD SiC>2 is first deposited to help the lift-off process. Then the 1
gm gate opening is patterned using conventional optical lithography, as shown in Figure
5.1a. A SiC>2 layer o f about 0.1 gm in thickness is then deposited in an angle
89
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 5.1(a) 1 [irrigate
opening with sacrificial layer
\
\
V
Figure 5.1(b) tilt
evaporation of dielectrics
~\
Figure 5.1(c) gate metal
deposition
A
\
Figure 5.1(d) lift-offand
gamma gateformed
Figure 5.1 Gamma gate fabrication process
*L,
Figure 5. 2 Gamma gate length defined by the tilt angle which can be easily controlled.
90
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
onto the wafer prior to the Pd/Au gate metal deposition (Figure 5.1b). After that, the gate
metal is deposited with normal angle (90 degree to the wafer surface) evaporation (Figure
5.1c).
Due to a shadow effect, by varying the angle, gates with different footprint sizes
can be easily fabricated. Referring to Figure 5.2, the effective gate length can be easily
found as:
Lg (5,0) = 5 tan'1 (0)
Where 5 is the height o f the total opening minus the height o f the deposited dielectrics
layer, and 0 is the tilt angle, as illustrated in Figure 5.2 .
In order to have a 0.3 pm gate length, for 5 = 1.6 pm, an angle o f 72° is used for
the shadow deposition. It is worth noting that the final gate length is independent o f the
gate opening, hence it is easy to achieve sub micro size gate. At the same time, the fieldmodulating plate is formed on top o f the SiC>2 layer and became an integral part o f the
gate metal. The length o f the field plate is the difference o f the gate opening and the
effective gate length, Lg (5, 0). This field plate is helpful to improve the breakdown
voltage as can be seen later. A picture o f the gamma gate is illustrated in Figure 5.3. The
gate metal is E-beam evaporated Au/Pd with thicknesses o f 6500 A/200 A, respectively.
The dielectrics below the gamma gate plate is evaporated SiC>2 . To examine the quality
o f the evaporated SiC>2, a big capacitor
(2
cm by
2
cm) is fabricated using the SiC>2 as the
insulator. It is seen from Figure 5.4 that there is no leakage current through the SiC>2 .
To verify the effect o f the gamma gate, both the gamma gate and normal gate
devices were fabricated on the same material. The material structure grown on semi-
91
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
insulated SiC substrate by MOCVD along with the layer thickness is shown in Fig.5.5,
which was a result o f optimization after analysis stated in the previous chapters. The
gamma gate structure is also shown. The material had a channel sheet resistance o f 560
Figure 5.3 SEM picture of the gamma gate illustrated in Fig 5.1(d).
lxlO "7
5x1 O' 8
5
S
o
§-
5
u
-5 x l0 "8
-lx lO ' 7
Figure 5.4. Negligible leakage current of the Si02 dielectrics below the gamma gate
plate (IV was measuredfrom a capacitance with an area of 2 cm by 2 cm).
92
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
_ J
L g
L o g
*
H Si02
x_______________
Source
5 nm undoped AlGaN (25%)
15 ran, 6E18 cm'3 AlGaN (25%)
Drain
3 nm, undoped AlGaN spacer (25%)
1.2 pm, undoped GaN
0.1 pm AIN
SI SiC Substrate
Figure 5.5 Schematic of a gamma gate device with optimized material structure for
sample 818A .
Q/sq. All fabrication process other than the gamma gate part was the same as described in
Chapter 4. To fabricate normal gate devices for comparison, we simply blocked part of
the wafer from SiC>2 tilt deposition and afterwards, the metal gate was evaporated at the
same time as the gamma gate metal deposition after removing the blocking material. The
resulting normal gate device had a gate length o f
pm, and the source-to-gate distance o f 1 pm.
1
pm, the gate-to-drain distance o f
1
For earlier works[v], devices with the
field-modulating plate had a gate-to-drain spacing o f 5 pm and greater. For the present
gamma gate device with the integrated field plate, the effective gate-to-drain spacing was
only 1.7 pm because o f our special processing technique developed.
For our gamma
gate, the drain-source distance was 3 pm; the gate-source distance was 1 pm, effective
gate length, Lg, was 0.3 pm, and the field plate length, Log, was 0.7 pm.
93
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
§5 .2 Device Performance
The device obtained had a typical transconductance o f 210 mS/mm and a
saturation current density o f 850 mA/mm (at Vg = 1.5 V). The DC characteristics are
shown in Figure 5.6. The DC-IV characteristics o f the normal gate device are similar to
0.10
0.08
S te p : -1 V
0.06
0.04
0.02
0.00
10
5
0
VdS(V)
250
1000
v ds = 8 v
200
800
150
600
100
400
50
200
£
e
SJD
a
a
a
0
4
3
2
1
0
1
2
V«(V)
Figure 5.6 Gamma gate device DC characteristics(gate width: 100 jum)
94
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 5.1 Breakdown voltage o f the gamma gate device
B V g s (V )
B V gd (V )
B V gd/B V g S
32
67
2 .0 9
30
83
2 .7 7
40
95
2.11
60
110
1 .8 3
40
110
2 .7 5
35
110
2.86
0
5
10
15
Termperature (1000/T) (1/K)
Figure 5.7. Breakdown voltage of the device vs inverse of temperature
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
these o f gamma gate device; only gm is slightly smaller for the normal gate device than
that for the gamma gate device(not shown). The breakdown voltage o f GaN HFET was
measured under the condition set at Ig = 1 mA/mm. For the case o f the 1.7 gm gate-drain
separation, the maximum breakdown voltage o f the gamma GaN HFET achieved was
over 110
V
at 1 mA/mm drain current leakage. The use of the asymmetric gamma gate
allowed us to conveniently assess the breakdown voltage equivalent to that with and
without the field plate using the very same device, by simply measure
B V gd
and
B V gs
(or
breakdown voltage o f the gate-to-drain and gate-to-source, respectively). By this way of
measuring, the breakdown voltage for the same device, we can eliminate the material
non-uniformity related errors. The ratio o f B V gd to
B V gs
reaches a maximum value of
2.86 (or an average o f 2.4 for six devices) for the field plate device, as shown in Table
5.1, while that o f the conventional GaN HFET is only 1.25 (or an average ratio o f 1.04).
The latter ratio o f the deviation from unity is probably due to some misalignment of the
symmetric gate. Figure 5.7 shows the breakdown voltage for GaN HFETs with and
without the field plate measured at different temperatures, varying from 300 K to 77 K.
At low temperatures, the breakdown voltage o f the gamma gate GaN HFET was also
higher than that o f conventional GaN HFET.
For both cases, the breakdown voltage
increased when the temperature was increased, indicating that the breakdown was caused
by impact ionization or avalanche breakdown.
The temperature dependence of the
breakdown voltage can be explained by the modified B araff s theory6. When hot carriers
pass through the depletion layer, they lose part o f their energy to optical phonons. The
electron-phonon mean free path Xis given by:
96
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
X= X0tanh(E 0p/2kT),
where Eop is the optical-phonon energy, and X0 is the low-temperature, high-energy
asymptotic value o f the phonon mean free path.
When the temperature increases, X
decreases as the carriers lose more energy along a given distance at a constant field, and
they need higher electric field to acquire sufficient energy to generate an electron-hole
pair. Hence, the breakdown voltage increases. The temperature coefficient is 0.033 V/ C
for the breakdown voltage. Comparing to the 0.024 V/ C for the Si pn junction case7,
GaN material exhibits a larger temperature coefficient.
Thus, in our simulation to be
discussed later, we assumed that the breakdown was caused by avalanche process.
For RF performance, small signal scattering parameters were measured for both
conventional and gamma gate devices at Yds = 15 V and Vgs = -1.5 V on devices with the
50
45
^
pa
3
40
35
S 2350
I
20
-l ft
O
1 0 11
Frequency (Hz)
Figure 5.8(a). RFperformance of a normal gate device(Lg = 1 pm).
97
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
CQ
w
W
xs
I
O
50
45
40
35
30
25
20
15
10
5
0
108
Frequency (Hz)
Figure 5.8(b). RFperformance oj a gamma gate device (Lg = 0.3 pm).
100 pm gate width. The results are illustrated in Figure 5.8(a) and Figure 5.8(b),
respectively. The cutoff frequency (fx) and the maximum oscillation frequency (fmax) of
the gamma gate device reached to 26 GHz and 50 GHz, respectively, while the fj and fmax
of the conventional device with the 1 pm gate length was about 10 GHz and 28 GHz,
respectively, showing a significant improvement to the RF performance by using the
gamma gate.
§5.3 Simulation and discussion
In order to understand the mechanism o f the improvement o f breakdown voltage,
the devices, with and without the field plate were simulated using the SILVACO®
simulator. A model including the piezo-electrical effect and spontaneous polarization
98
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
induced charge for AlGaN/GaN material was implemented as discussed in chapter 2.
Results related to breakdown performance are discussed as follows.
For the conventional gate device (referring to Figure 5.5, let Log=0), the potential
distribution in the structure along the channel at a breakdown voltage o f 50 V on Vdg is
shown in Figure 5.9 in dashed line, showing that the biggest voltage drop in the channel
occurs beneath the right corner of the gate where depletion happens. In this region, the
maximum electric field in the channel is about 2.5 MY/cm (not shown), far less than the
o
critical electric field suggested by Kunihiro et al. , which is around 4 MV/cm for GaN.
Hence we believe the breakdown does not happen in this region. However, as illustrated
in the inset in Figure 5.9, the electric field in the cap layer (AlGaN) shows a peak value
60
50
<u
a
S3
40
JS
30
Distance from source (um) t
G
20
is
aat
-M
»
o 10
Pm
0
0
1
2
3
Distance from source edge (nm)
Figure 5.9 Simulation results of the potential and electrical field distributions in the
channel for gamma gate and normal gate GaN HFET when Vdg = 50 V. Dash line is for
conventional device and solid line isfor the gamma gate device.
99
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
of
6
MV/cm for the conventional gate device. The peak is located near the gate corner
towards the drain in the cap layer. To have breakdown, the electron field reaches the
critical field (Ecr) first at this point. Since the scattering rate in AlGaN is higher than that
in GaN, we expect that the mean-free-path in AlGaN is shorter and electrons need to
travel further to gain energy than in GaN in order to have impact ionization. Moreover,
the thickness o f the AlGaN is small, and the band gap o f AlGaN is larger than that of
GaN, and thus AlGaN have a higher critical breakdown field.
When the gamma gate is used, there is an electric field below the levering
electrode o f the gate, which reduces the 2D electron concentration in the channel. As a
result, there is a bigger region along the channel with large resistance compared to that of
a conventional gate device. Also the voltage drops across in a wider region, as shown
with solid line in Figure 5.9, and hence the electric field in the channel drops to about 1.6
MV/cm(not shown). For the normal device, voltage drop occurs across a range o f about
0.85 pm. However, the range extends to 1.4 pm for the gamma device.
The inset in
Figure 5.9 shows that a single peak o f the electric field in the conventional gate device
becomes two peaks, such that the highest field is reduced to 4 MV/cm in the cap layer
when 50 V o f Vdg is applied. Therefore a higher breakdown voltage is expected in this
gamma gate structure. To reach a similar electric field value as that o f the conventional
gate structure, a Vdg o f over 100 V is required for the gamma gate structure, showing
good agreement with our experimental result o f 110 V (see Table 5.1).
In the following sections we will discuss how to optimize the gamma gate
structure to obtain highest breakdown voltage for a selected structure.
100
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
5.3.1 Gamma shape effect
The variation o f gamma gate, namely, the top plate length (or overlap length, Log)
and the gamma height, H, shown in Figure 5.5, will change the electric field distribution
and the peak electric field value. Figure 5.10 and Figure 5.11 show the peak electric field
distribution in the cap layer as functions o f field plate length and gap, respectively. Also
shown is for the normal gate case, or Log = 0, which has a single electric field peak
(Peakl). This latter single electric field peak is the highest among all cases and the use of
the field plate indeed reduce the peak electric field and improves the breakdown
performance. The second peak (Peak2) arises due to the use o f the field plate. As the
field plate length increases, the peakl decreases and peak2 increases.
As can be
8x106
P eak l
6x106
Peak2
4x106
2x106
0
0
2
1
3
Distance from source (pm)
Figure 5.10 Peak electric field distribution in the cap layer vs the gamma gate (field
plate) length (Log). (Based on structure shown in Fig 5.5, Vdg=50 V).
101
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
6x10'
P eak l
Peak2
\
S
(J
^
C
4x10'
2
"3
yj
1
2x10£
3
0
2
1
3
Distance from source (pm)
Figure 5.11. Peak electric field distribution in the cap layer vs the gamma gate height.
(Based on structure shown in Fig 5.5, Vdg=50 V).
seen, too long Log is unnecessary. The optimal situation is about where peakl and peak 2
have the same value, which happens when Log~ 0.7 pm for this case. Figure 5.11 shows
the dependence o f the electrical field on the height o f the gamma plate, i.e. the thickness
o f the dialectic layer, if there is any, to support the gamma plate. The closer the overlap
gamma plate to the top AlGaN layer is, the stronger the “gamma gate effect” in
improving breakdown voltage becomes. The optimal gap height for this case is
approximately 400 A when the two peaks again are equal (In Fig. 5.5, thicker SiC>2 was
chosen because the evaporated SiC>2 has a smaller sr).
As the gamma gate height is further reduced, the peak 1 begins to merge with the
peak 2 as if the entire plate were a single normal gate again. In this case, the use of the
plate loses its effectiveness.
102
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
For microwave application, the use o f the field plate may degrade the RF
performance due to the increase o f the gate capacitance if an excessively long plate
length and a small gap are used. Thus these numbers must be optimized for different
frequency applications.
5.3.2 Dielectrics effect below the gamma gate
Material with different dielectric constants may be used below the overlap gate.
They can change the electric field distribution according to Gauss’s law, since sri Ej = s r2
E 2 at the interface o f the dielectric and the AlGaN surface. For higher sri, Ei will be
smaller, and thus if the voltage is fixed, E 2 in the semiconductor will be larger; or in other
words, we have a stronger gamma gate effect. These results are shown in Figure 5.12, in
^ P eak l
0
w
Peak2
4x10'
s.
0
2
1
3
Distance from source (pm)
Figure 5. 12 Peak electric field distribution in the cap layerfor different dielectrics under
the gamma gate. (Based on structure shown in Fig 5.5, Vdg=50 V).
103
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
which three cases with different er,
air, SiC>2 , and SisN4 below the field plate were
studied. For air, the gamma gate effect is the smallest. The Si3N 4 case has the strongest
gamma gate effect, as evidenced by the highest peak2 among the three cases. Again, the
optimum case to maximize the breakdown voltage is to design a gamma structure such
that the value o f peakl almost equals that o f peak2. For a given chosen dielectric
material, we should carefully design the gamma height and gamma length to maximize
the breakdown voltage, provided that the microwave performance is not compromised.
5.3.3 Effect of the cap layer doping and thickness
Variations of the doping concentration in the cap layer also change the electric
field distribution since the voltage applied is shared by two regions in the semiconductor:
the depleted cap layer region below the gate and the channel region, whose resistance is
related to the doping concentration in the cap layer. When the channel resistance
decreases, the voltage drop in this region will also decrease, and the voltage-drop in the
cap layer will increase, resulting in an increase o f the electric field peak in the cap layer.
The results are shown in Figure 5.13, indicating that the peak electric field value is very
sensitive to the doping. For 5 x l0 17 cm '3 in doping, the highest electric field is 2.5
MV/cm, and increases to
6
MV/cm for 2 x l0 19 cm ' 3 with a given applied voltage o f 50 V.
The reason is that for low doping devices, the sheet charge density in the channel is about
5 x l0 12 cm ' 2 in comparison with 1.6xl0 13 cm ' 2 for the 2 x l0 19 cm ' 3 doped case (i.e. for the
structure shown in Fig 5.5 with 70% compensation ratio, y = 0.7). Therefore careful
trade-offs must be taken in order to obtain both high current and high voltage.
104
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Peakl
5x10
>
6x106
1x 1 0
Peak2
2 x 10
■O
o
■m
m*
+->
O
Q>
2x106
LU
0
0
1
2
3
D is ta n c e from s o u r c e (Mm)
Figure 5 . 1 3 . Peak electric field distribution in the cap layer as a function of doping in the
cap layer. (Based on structure shown in Fig 5 .5 ,
6x106
Vdg = 5 0 V).
Peakl
Peak2
Bw
2
c
0
0
1
2
3
Distance from source (fim)
Figure
5.14.
Peak electric field distribution in the cap layer as a function of the cap
layer thickness. (Based on structure shown in Fig 5 .5 ,
Vdg - 5 0 V).
105
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Changing the cap layer thickness has a similar effect on channel resistance as the
doping density does, hence affecting the breakdown performance. The results are shown
in Figure 5.14. For a cap layer of 350 A, the peak electric field value reaches 5.1 MV/cm.
When the thickness decreases to 120 A, the peak E value drops to 3.7 MV/cm. A thinner
cap layer can thus give a higher breakdown voltage for a given doping density in the cap
layer.
§5. 4. Summary
We have developed a novel process to fabricate sub-micron (0.3 pm) gamma gate
length AlGaN/GaN HFET with a field modulation plate using conventional optical
lithograph. This field plate was formed simultaneously with gate metallization. The high
breakdown voltage o f 110 V along with a high current density was achieved with the
gate-drain separation as short as 1.7 pm for the gamma gate GaN HFET.
improvement of breakdown voltage by a factor o f 2.8 was achieved.
An
Our simulation
results showed that for 25% A1 content AlGaN/GaN HFET structures, the peak critical
electric field was around
6
MV/cm, and located in the AlGaN cap layer.
The peak
electric field o f the gamma gate GaN HFET could be reduced through a careful optimized
design.
Due to the use o f the 0.3 pm gate length, the device also showed a significant
improvement o f the RF performance, i.e. cutoff frequency (fx) and maximum oscillation
frequency (fmax) increasing from 10 GHz, 27 GHz to 26 GHz, 50 GHz, respectively.
106
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Bibliography:
1 Y-F. Wu, B.P. Keller, S. Keller, D. Kapolnet, P. Kozodoy, S.P. Denbaars, and U.K. Mishra,
“Very high breakdown voltage and large transconductance realized on GaN heterojunctionfield
effect transistors", Appl. Phys. Lett. Vol.69, No. 10, 2 September 1996, ppl438-1440
2Z. Fan, S.N. Mohammad, O. Atkas, A.E. Botchkarev, A. Salvador, and H. Morkoc, “Suppression
of leakage current and their effect on the electrical performance of AlGaN/GaN modulation
dopedfield effect transistors”, Appl. Phys. Lett., Vol. 69, No.9, ppl229-1231, Aug. 1996
3A.S. Grove, O. Leistiko, and W.W. Hooper, IEEEElectron Devices, ED-14, No.3, p!57, 1967
7 K. Asano, Y. Miyoshi, K. Ishikura, Y. Nashimoto, M. Kuzuhara and M. Mizuta, “Novel High
Power AlGaAs/GaAs HFET with a Field-Modulating Plate Operated at 35V Drain Voltage”,
IEEEIEDM, 1998, 3.3.1 ~3.3.4
5 N.-Q. Zhang, S. Keller, G. Parish, S. Heikman, S. P. DenBaars, and U. K. Mishra, “High
Breakdown GaN HEMT with Overlapping Gate Structure ” IEEE Electron Device Letters, Vol.
21, no. 9, September 2000, p421-423
6S.M. Sze, “Physics ofSemiconductor Devices ”, Second Edition, John Wiley & Sons, New York,
pp47-48.
A. Goetzberger, B. McDonald, R. H. Haitz and R. M. Scarlet, “Avalanche effects in Silicon p-n
junction. II. Structurally perfectjunctions" J. A.P., 34, 1591 (1963)
s Kunihiro, K; Kasahara, K; Takahashi, Y.; Ohno, Y. “Experimental evaluation of impact
ionization coefficients in GaN”. IEEEElectron Device Letters, vol.20, (no. 12), IEEE, Dec. 1999.
p. 608-10
107
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 6
PROTON IRRADIATION STUDY
The utilization o f microwave technology for satellite broadcasting and weather
forecast is now essential in our information technology society. For this reason, the
development o f semiconductor devices that can operate reliably in radiation environment
is o f interest. Since GaN is a wide band gap material, it is intrinsically a good candidate
for radiation hardening application. In this chapter we examine its performance under
proton irradiation. The influence o f proton irradiation (1.8 MeV, l x l 0 14 cm'2) on the
properties o f an AlxGai.xN/GaN HFET is studied. Current-voltage (I-V) and Raman
scattering are obtained as a function o f rapid thermal annealing (RTA) temperature after
irradiation.
The I-V curves show that the saturation current
(Ids) and the
transconductance (gm) o f the HFET are reduced by the irradiation, and the DC
characteristics o f the HFET can mostly be recovered by RTA at above 800°C. Compared
to the reported results o f GaAs based devices from the literature, GaN shows a much
better proton irradiation hardening capability.
§6.1. Experimental Details
108
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The device under study is a standard MBE grown HFET structure as shown in
Figure 6.1. A 500A GaN active channel layer was grown on a 1.2 (am GaN undoped
buffer layer on a C-plane sapphire substrate. The doping concentration o f the channel
layer was about 5x10 17 cm'3. A 300A Alo.15Gao.s5N top layer with a doping concentration
of 5 x l0 18 cm ' 3 was prepared on top o f a 30A Alo.15Gao.s5N undoped spacer layer. Hall
measurements showed a room temperature mobility o f 705 cm2/Vs and a sheet
concentration o f 1.55xl013 cm'2. The device fabrication is the same as that described in
section §3.2.
The gate length o f the device used is 1 pm and the space between the
source and the drain is 3 pm. One piece o f the sample o f the same layered structure was
prepared for Hall and Raman scattering measurements.
s our c e
gat e
dr ai n
3 00 A n - A l xG a , . xN
30A u n d o p e d A l xGai _xN
500A
1.2 p m
n - Ga N
G aN b uf f e r
Sapphire
F ig u r e 6 .1 D e v ic e s c h e m a tic u n d e r p r o to n ir r a d ia tio n stu d y .
109
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
All samples were exposed to 1.8 MeV proton irradiation with a dosage of
l.OxlO 14 cm"2. The protons penetrated through the entire epilayers with a projected range
o f about
20
pm.
After irradiation, the I-V, Hall, and Raman scattering measurements were
performed as a function of anneal temperature to evaluate the change o f material
properties and the performance o f the HFET device.
Annealing was carried out at
different temperatures in a N 2 RTA system for 40 seconds.
§6.2. Results and Discussion
Figure 6.2 shows the comparison o f the I-V curves of the HFET device before and
after irradiation. One can see that the saturation current o f the HFET was greatly reduced
by the proton irradiation, dropping from 260 mA/mm to about 100 mA/mm. This
revealed that free carriers were removed and/or trapped by defects induced by irradiation.
The device transconductance was also degraded from about 80 mS/mm to 26 mS/mm
after irradiation. The degraded device performance could be mainly due to the irradiation
induced defects or damage to the material itself, and a possible degradation o f the
contacts o f the device. However, as shown in Figure 6.3, the electrical performance was
gradually recovered with increasing the RTA temperature. It can be seen that the
saturation current and transconductance recovered significantly after annealing at
temperatures above 600°C. When the RTA temperature reached 800°C, the Es and gm
increased to 220 mA/mm and 56 mS/mm, respectively.
110
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 6.2 Comparison of the I-V curves of the HFET before (a) and after (b) proton
irradiation. The HFET gate width used is 50 pm.
Ill
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
0
a
I ds ( m A / m m )
Ia
gm (mS/mm)
200
75
65
&
us-
u
a
U
a
55
0
45
35
‘■♦3
«I 3
-M
«
0
25
C/3
a
a
Cfl
8,
V
w
C
A
w
3
■3
a
ou
5«
a
s-
H
15
0
300
400
500
600
700
800
Annealing Temperature (°C)
Figure 6.3 Transconductance and saturation current of the HFET measured at Vgs=10
V, and Vg=0.5 V versus annealing temperature. Before irradiation, gmo=80 mS/mm,
IdsO= 260
mA/mm.
Figure 6.4 shows the results from room temperature Hall measurements obtained
at different stages o f annealing. Combining with the results from Figure 6.3, both the
saturation current and transconductance apparently increased due to the recovery o f the
sheet concentration while the N s-p product remained almost the same. This indicates that
the transconductance and saturation current are more sensitive to the concentration of free
carriers than the low electrical field mobility.
In other word,
the free carriers
consistently contributed to and were dominantly responsible for the saturation current
112
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
700
ao
600
ro
O
w
a
Va
u
400
"§
300
a
O
S3
C
V
200
|
500
+*
0V
+0*)
M
Vi
'£
100
"St"'
0
0
250
500
750
1000
Annealing temperature (°C)
Figure 6.4 Results of Hall measurement for the HFET structure after irradiation and
annealing. Before irradiation, Hall results are jUo=705 cm2/Vs and Nso~l.55x1013 cm 2 .
while the mobility decreased due to increased scattering by defects (or dislocations) and
piezoelectric scattering in the region close to the channel. The increase o f scattering may
be a result o f annealing-induced strain relaxation in the AlxGai.xN region. The latter is
confirmed by Raman scattering measurements as shown in Figures 6.5 and
specific frequencies o f Wurtzite GaN Raman modes
1
6 .6 .
All
were observed in the Raman
measurements. To understand what actually occurred in the irradiated structure, we have
followed the GaN E 2 peak centered at 567 cm '1, which is actually composed o f three
peaks.
113
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
E2(T0)
^2 peak and fitting
p2
e
D
R a m an Shift (cm ' l)
J v
500
600
700
800
R am an Shift (cm '1)
Figure 6.5 Typical Raman spectrum of an irradiated sample at room temperature.
Marked PI, P2 and P3 in the insert are the fitted positions of the E2 peaks from different
regions in the sample.
From Raman studies o f other semiconductor systems, the characteristic Raman
peak from a tensile (compressive) strained semiconductor layer shifts to a lower (higher)
'y
wave-number relative to the unstrained one . In our case, there was a compressive strain
in the GaN region near the AlGaN cap layer ( the channel interface ), and a tensile strain
in the AlGaN layer close to the channel due to the lattice mismatch between the
GaN(with a lattice constant o f 3.18A) and AlGaN (3.10A for AIN). We therefore
assigned the decomposed three Ga-N E2 peaks to those from the strained region o f the
heterojunction - the Ga-N in AlGaN near the channel (PI), the unstrained GaN (P2), and
compressive strained GaN region including the channel region (P3), as shown in the
114
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
78
73
&
68
fl
A
a
03
Pi
63
58
0
200
400
600
800
1000
Annealing Temperature (°C)
Figure 6.6 Raman shifts of the fitted E2 peak positions vs. annealing temperature of the
irradiated sample. Before irradiation, peak 1 in Figure 5 was at 562.06 cm1 and the
peak was shifted to 561.17 cm1after irradiation.
insert o f Figure 6.5. By using a least square curve fitting procedure, we obtained the
evolution o f the three peaks as shown in Figure
6 .6 .
Due to the proton bombardment,
atoms were displaced, the Raman frequency o f Peak 1 red-shifted to 561.17 cm ' 1 from
562.06 cm ' 1 in the as-grown sample (not shown). After 400°C anneal, the lattice at least
partially recovered from the irradiation effect, showing Peak 1 blue-shifted. At above
400°C, the behavior o f Peak 1 suggests that atoms near the irradiated AlGaN/GaN
heterojunction began to rearrange and become more tensile strained as the lattice was
being restored, and hence the PZ effect became stronger, resulting an increase o f the
sheet concentration. On the other hand, the mobility decreased due to the increase o f the
115
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
PZ scattering. This behavior persisted until 850°C as observed in Figure 6.4. Peak 3 also
blue shifted slightly at above 400°C, showing the GaN crystal in the channel side become
more compressive, meaning the crystalline quality improved due to the RTA, as also
indicated by the narrowing o f the peak width (not shown). This relatively low recovering
temperature suggests that proton induced damage could be removed much easier than that
caused by other ions3. However, after 850°C thermal annealing, the lattice mismatch
induced strain relaxes, possibly creating additional dislocations at the interface, reducing
SP and PZ effects, hence the sheet concentration decreased. Due to the decrease o f SP
and PZ scatterings, the mobility increased after 850°C as shown in Fig 6.4.
§6.3. Summary
In summary, we have studied the influence o f proton irradiation on AlGaN/GaN
HFET in terms of post-irradiation thermal annealing. The DC performance o f the device
has been shown to recover remarkably by the RTA at above 800°C for GaN HFET. The
saturation current density dropped to about 38 % o f the original current density after
l xl O 14 cm '2 proton irradiation. For comparison, we also gave the experiments results for
AlGaAs/GaAs HFET device here by W. T. Anderson4: the dosage was 2 x l0 13 cm ’2 when
the saturation current density dropped to 38 % o f its original current density after proton
irradiation, which was 5 times lower than that for the GaN HFET case. This comparison
suggests that GaN based HFET had a much better radiation hardening performance.
116
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Bibliography:
1 Leah Bergman and Robert J. Nemanich, “Raman spectroscopy for characterization of hard,
wide-bandgap semiconductors: diamond, GaN, GaAIN, AIN, BN”, Annu. Rev. Mater. Sci., vol.26,
pp.551-579, 1996
2 B.Dietrich, Y.S.Tang & C.M.Sotomayor Torres, "Comparison of strain in Si/SiGe
pseudomorphic heterostructures and quantum dots", Solid State Phenomena, vol. 47-48, p.535538, 1996
3 Wilson,R. G. ; Pearton, S. J.; Abernathy, C.R.; Zavada, J. M. "Thermal Stability of Implanted
Dopants in GaN", Applied Physics Letters, vol.66, (no. 17), p.2238-40,24 April 1995.
W. T. Anderson, A. R. Knudson, A. Meulenberg, H. L. Hung, J. A. Roussos and G. Kiriakidis,
“Heavy Ion Total Fluence Effects in GaAs Devices” IEEE Trans. On Nucl. Sciences, Vol. 37,
(6), Dec. 1990, pp2065-70
4
117
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 7
CONCLUSION AND FUTURE WORK
§ 7.1 Conclusion
Based on the calculation and simulation, we understand that due to the PZ and SP
effects, the 2DEG density in the AlGaN/GaN structure can be significantly enhanced. To
properly take advantage of these effects for HFET, one should carefully design the HFET
structures. Optimized structure was also presented.
Samples with different A1 contents were grown using MOCVD.
Hall
measurement results showed that higher A1 content sample had higher sheet
concentration. Mobility for higher A1 sample was generally lower due to the stronger
polar optical phonon scattering and piezoelectric scattering in the structure.
Key processing for AlGaN/GaN HFET fabrication were developed and
established. Ohmic contact was particularly important for the device, since a good ohmic
contact is difficult to achieve with the wide band gap material. Surface pre-cleaning and
RTP were critical for a good ohmic contact. A record low contact resistance number of
4.42xl0’7 Q.cm 2 was achieved. Pd/Au metal scheme was chosen for the gate Schottky
metal contact.
118
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Basic AlGaN/GaN HFET devices were fabricated. It was confirmed that higher
A1 content structure could deliver higher saturation current density.
Over 1 A/mm
current density were achieved from 40% A1 content AlGaN/GaN structure. Results with
good pinch-off voltage, saturation current and RF performance were presented. Power
amplifier of
8
Watt output at 9 GHz was also achieved from a single 5 mm AlGaN/GaN
HFET, corresponding to a CW RF power density o f 1.6 W/mm at X band, which is about
3 times of that from a GaAs based device.
In order to further improve the DC and RF performance o f the AlGaN/GaN
HFET, We developed a novel process to fabricate sub-micro (0.3 pm) gate length with a
field modulation plate using conventional optical lithography approach. This field plate
was formed simultaneously with gate metallization. The high breakdown voltage o f 110
V was achieved with the gate-drain separation as short as 1.7 pm for the FP GaN HFET
with a high current density, and an improvement factor of
2 .8
to the breakdown voltage
was achieved. Meanwhile the RF performance o f the device also showed a significant
improvement: both cutoff frequency (fj) and maximum oscillation frequency (fmax) were
doubled.
We also studied the influence o f proton irradiation on AlGaN/GaN HFET device.
After l x l 0 14 cm ' 2 proton irradiation, the saturation current density o f the device dropped
to about 38 % o f the original current density. For a similar amount o f current density
drop for a GaAs based HFET, the proton dosage was about 2 xlO 13 cm'2, showing that
GaN based HFET potentially is a very good candidate for irradiation hardening
applications.
119
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
§7.2 Future suggested work
In order to commercialize the device, more work need to be done. Key issues are
listed as following:
1. Surface states evaluation and control. Based on the analysis in Chapter 2, the
condition o f the surface is critical to device performance. Although some
initial works had shown favorable results, the surface issue remains unsolved
fully.
2. Performance degradation. So far, more or less, all devices exhibited a
degradation history. This could be related to thermal dissipation problem. For
example, for a 1KW hot plate, its power density is about 0.1 W/mm2. But for
a
8
W power module shown in Figure 4.15, the dissipated power density was
10 W/mm . Therefore, die attachment techniques must be improved
revolutionarily. New heat sink material, i.e. diamond, may be needed for heat
removal.
3. Power density limitation exploring. Although we have achieved exciting
results compared to conventional GaAs HFET device, the potential o f GaN
HFET is far from exhausted. Particularly for our gamma gate device, it is
possible to output 10 W/mm power at a bias o f 50 V and 400 mA/mm. After
solving the issues listed above, we believe we can achieve this goal.
120
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Appendix I
1. Ionized impurity scattering:
For an arbitrary degeneracy o f the electron gas, the mobility, p h limited by the
ionized impurity scattering is given by1:
2
5 .5
Ml
2 , J
rr
X 1. 5
ttg v ( k ft T )
F3
(A l.l)
3 q3N, Jm^Fx
Where:
*oo
00
1/ 2
J
dx
JO
0 I1 + exp( x - e F)
=
(A 1.2)
f00 x 3 exp( x —e F) dx
JO O ( ( l + exp( x - e F) ) 2
<5>i = l n ( l + q ) - - 9 —
(A 1.4)
l + rj
r, =
3)
8 m ’s s {k BT Y F
'V /V
1
q 2h 2n F .
(A1.5)
j°°x1/2 exp ^ -e^ )<ix:
(A1.6)
•0 (l+expOc-e,,))2
In(n/Nc) | 9k * ( n ! N c f 3
1+0?/Afc)5
yyj
lr
71
N -2(^4-)
C
16
(A1 7)
1+n/JVc
3/2
(A1.8)
iT ti1 ’
1M.Shur, b.Gelmont, andM.Asif Khan “Electron Mobility in Two-dimensional electron
Gas in AlGaN/GaN heterostructures and in Bulk GaN” J. Elec. Materials. Vol. 25 (5),
1996pp777-85
121
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Here m* is the electron effective mass, £s is the static dielectric permittivity, Nj is the total
concentration o f ionized impurities. E f if the dimensionless Fermi level. Ec is the bottom
o f the conduction band, kn is the Boltzmann constant, and E f if the Fermi level and n is
the electron concentration. Nc is the effective density o f states n the conduction band.
2. Polar optical phonon scattering
According to the analytical theory developed by Gelmont et al , the electron
mobility limited by the polar optical phonon scattering is given by:
AttksJ i2(\-5k bT / qEg)
* P° = qNpm{2mqEr/,(\+ErJ E sr )
(A 1 '9>
where:
k-
£ I£ - £ I£
O
NP
(A1.10)
1
00
O
S
Qxp(qEpo / kBT)~ 1
( A l . 11)
here k is the coupling constant. ssand a, are the static and high frequency dielectric
constants, respectively, so is the vacuum dielectric permittivity. Epo is the polar optical
phonon energy, which is 91.2 meV. Eg is the energy gap in eV. Np is the Planck
function.
3. Acoustic scattering
2 B.
Gelmont, M. Shur and M. Stroscio, Polar Optical phonon scattering in three- and twodimensional electron gases J. Appl. Phys. 77, 657, Jan. (1995)
122
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Like in non-polar semiconductor, carriers in GaN also suffer from the scattering
o f acoustic phonon. The mobility limited by the acoustic phonon scattering can be
expressed as in a classic case:
2( 27ry,2C Lh 4q
(A1.12)
Where Eds is the deformation potential (shift o f the band edge per unit dilation in eV),
which is
8
eV~ 9.2 eY for GaN material. Be noticed that the longitudinal elastic constant,
Cl, is equal to the product o f the material density and the second power o f the sound
velocity and equals to 2.65x10 11 N/m2.
4. Piezoelectric scattering
The mobility determined by the relaxation time o f piezoelectric scattering is given
MPZ
\ 6 q ( 2 n y n CLh 2
3 (qe j e , ) \ m - y \ k , r r
(A 1.13)
Where epz is the piezoelectric coefficient, Cl is the longitudinal elastic constant.
3 H adis M orkoc, “N itrid e sem ico n du ctors a n d d ev ices ”, C h apter 8: C a rrie r Transport. Springer,
p p 239, (1999)
123
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Appendix II
GaN HFET processing flow
1. Wafer inspection via Alpha step
If moly exits, remove back moly
Protect front side with photoresist and bake
Etch moly:
5 Phosphoric: 3 Nitric: 2 DI water for around 2 min
DI water rinse 5 min and N 2 blow dry
Note: DI water rinse and N 2 blow dry are required after each o f following step.
2. First Photolithograph (mask: markers)
PR:AZ5214/ 4.5K rpm @30 sec / bake @95°C lm in/ Chlorobenzene 5 min
Carl Suss: 12 sec.
Development: AZ400k:DI=l:5 40sec
3. E-beam evaporation
Ti/Cr=
2 0 A /1 0 0 0 A
Lift off in Acetone and clean with IPA
4. Second Photolithograph (mask: isolation)
PR:AZ5214/ 4.5K rpm @30sec/ bake @95°C lm in/ Chlorobenzene 5 min
Carl Suss: 12 sec
Image Reverse:
bake @ 90-100°C lm in
124
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
flush exposure @80 sec
Development: AZ400k:DI=l:5 40sec
Note: If use implantation for isolation, AZ4620 is used without image reverse.
5. E-beam evaporation
Si0 2 /Cr=
2 0 0 0 A /1 0 0 0 A
Lift off in Acetone and clean with IPA
6 . ECR
mesa etch
ECR or ICP
Post clean: warm Chloroform, Acetone, IPA
7. Cr /S i02 etch:
Cr etchant solution lm in
BOE lm in/ DI water
* mesa inspection and mesa height measurement.
Note: if use ion implantation for isolation, skip steps 5-7.
8 . Third
Photolithograph (mask: ohmic contact)
PR:AZ5214/ 4.5K rpm @30sec/ bake @95°C lm in/ Chlorobenzene 5 min
Carl Suss: 12 sec
Development: AZ400k:DI=l:5 40sec
9. E-beam evaporation
Ti/Al/Pt/Au = 200A/800A/400A/1500A
Lift off in Acetone and clean with IPA
125
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
10. RTP Annealing (Program: 900CGaN)
Delay 120sec
Ramp 200sec@900°C
Stay
35sec@900°C
Delay 120sec
11. Inspection and ohmic contact resistance measurement
12. Fourth Photolithograph (mask: Gate)
PR:1AZ5214:1AZ1350 thinner/ 2.5K rpm @30sec:: thickness=4500A-5500A
Bake @95°C l min/ Chlorobenzene 2.5 min
Carl Suss: 12 sec@ Vacuum contact
Development: AZ400k:DI=l:5 @40sec
13.E-beam evaporation
Pd/Au = 200A/6500A
Lift off in Acetone and clean with IPA
14.Fifth Photolithograph (mask: FIC)
PR:AZ5214/ 4.5K rpm @30sec/ bake @95°C lm in/ Chlorobenzene 5 min
Carl Suss: 12 sec
Development: AZ400k:DI=l:5 40sec
15.E-beam evaporation
Ti/Pt/Au = 200A/200A/3600A
Lift off in Acetone and clean with IPA
126
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Документ
Категория
Без категории
Просмотров
0
Размер файла
5 912 Кб
Теги
sdewsdweddes
1/--страниц
Пожаловаться на содержимое документа