advanced polarization-based design of algan/gan...
TRANSCRIPT
UNIVERSITY of CALIFORNIA
Santa Barbara
Advanced Polarization-Based Design of AlGaN/GaN HEMTs
A dissertation submitted in partial satisfaction of the
requirements for the degree of
Doctor of Philosophy
in
Electrical and Computer Engineering
by
Likun Shen
Committee in charge:
Professor Umesh K. Mishra, Chair Professor Steven P. DenBaars Professor Evelyn L. Hu Dr. Stacia Keller
June 2004
The dissertation of Likun Shen is approved.
________________________________________________ Steven P. DenBaars
________________________________________________ Evelyn L. Hu
________________________________________________ Stacia Keller
________________________________________________ Umesh K. Mishra, Committee Chair
April 2004
Advanced Polarization-Based Design of AlGaN/GaN HEMTs
Copyright © 2004
By
Likun Shen
iii
Acknowledgements
I feel fortunate to be at UC Santa Barbara to pursue my Ph.D., not only for its
beautiful beaches, but also for the superior research environment. Over the past a few
of years, I have had the opportunity to work with some truly exceptional people. It is
my great pleasure to mention them here for posterity.
First and foremost, I am extremely grateful to have worked for my advisor,
Prof. Umesh Mishra. Umesh has provided such a great research environment in which
all group members work together and provide support for one another. The
completion of this dissertation would not have happened without his guidance and
encouragement. His help is not limited to research, but in many aspects of life. I
really appreciate his encouragement which is very important to me when facing
challenges. I was also fortunate to have Dr. Stacia Keller, Prof. Evelyn Hu, and Prof.
Steve DenBaars on my committee. They also play a vital role in my research
experience. Steve and Stacia manage the MOCVD lab so well that I was able to get
many samples quickly. I have benefited from their knowledge of MOCVD growth
and material characterization. Prof. Hu has been a valuable resource in helping me
understand processing issues of devices.
I would like to thank all the growers of MOCVD and MBE: Stacia, Sten, Lee
M. Arpan, Gia, Brendan, Huili, Yulia, DJ, and Dan. Without their hard work, any
novel concepts would have stayed on paper. Special thanks to Yulia, her pioneering
work is the foundation of part of this dissertation. I am proud of being a member of
the HEMT team: Dario, Rob, Ale, Sten, Tomas, Siddharth, Hongtao, Chris, Pete,
iv
Haijiang and past members Naiqian and Rama. Thanks to Naiqian for teaching me
how to process. Dario, Rob and Ale have given me much help in device processing
and characterization. Yifeng also deserves recognition for a lot of useful discussions.
I am grateful to Yuan and Prof. Speck for providing TEM analysis. I would also like
to acknowledge some of the other members of Prof. Mishra and York groups: Birgit,
Can, Pengcheng, Bruce, James, Jeff, Mike, Yingda, Nadia, Paolo, Amir, Justin, Vicki,
Yun, and Jane. All of us made the office a noisy but lively place. The years spent in
Room 5120A will be remembered forever. Also thanks the Rodwell and Hu groups,
whom I have worked very closely with over the last five years.
The Mishra group is a very large group that would not run as smoothly as it
does without the help of Umesh’s administrative assistants. Lee B., Cathy, Masika,
Pam, Emeka, and Laura all deserve a great deal of thanks for all the help they have
given me over the years. I would also like to thank Val who has answered every small
question patiently. I would like to thank the cleanroom staff and management of Jack,
Bob, Neil, Brian T., Don and Ning, and microscopy staff Mark and Jinping for the
excellent job they have done in keeping the facilities running as well as they do.
Special thanks to Pete, Nadia, Lee M., and Arpan. They spent many hours to
improve the English of my dissertation. Pete also has helped to revise several of my
papers.
I would also like to acknowledge the financial support from the Office of
Naval Research (ONR).
v
Finally, I would like to thank my parents and my wife, Xiaojie. None of this
would have been possible without their support and love.
vi
Curriculum Vita
Likun Shen
EDUCATION
Bachelor of Science in Physics, Fudan University, Shanghai, China, July 1995. Master of Science in Electrical Engineering, Fudan University, Shanghai, China, July 1998. Doctor of Philosophy in Electrical and Computer Engineering, University of California, Santa Barbara, April 2004 (expected). PROFESSIONAL EMPLOYMENT
1995.9 – 1998.7 Research assistant, ASIC and System State-Key Laboratory, Department of Electrical Engineering, Fudan University, Shanghai, China
1998.9 – 2004.4 Research assistant, Solid State Electronics Laboratory, Department of Electrical and Computer Engineering, University of California, Santa Barbara
PUBLICATIONS
L. Shen, D. Buttari, S. Heikman, A. Chini, R. Coffie, A. Chakraborty, S. Keller, S. P. DenBaars and U. K. Mishra, “Improved high power thick-GaN-capped AlGaN/GaN HEMTs without surface passivation,” Accepted by the 62st Device Research Conference, Jun. 2004. L. Shen, R. Coffie, D. Buttari, S. Heikman, A. Chakraborty, A. Chini, S. Keller, S. P. DenBaars and U. K. Mishra, “Unpassivated GaN/AlGaN/GaN power HEMTs with dispersion controlled by epitaxial layer design”, Journal of Electronic Materials, vol.33, no.5, pp.422-425, May 2004.
A. Chini, D. Buttari, R. Coffie, L. Shen, S. Heikman, A. Chakraborty, S. Keller and U. K. Mishra, “Power and linearity characteristics of field-plated recessed-gate AlGaN/GaN HEMTs”, IEEE Electron Device Letters, vol. 25, no. 5, pp. 229-231, May 2004.
vii
H. T. Xu, N. K. Pervez, P. J. Hansen L. Shen, S. Keller, U. K. Mishra and R. A. York, “Integration of BaxSr1-xTiO3 thin films with AlGaN/GaN HEMT circuits”, IEEE Electron Device Letters, vol. 25, no. 2, pp. 49-51, Feb. 2004.
L. Shen, R. Coffie, D. Buttari, S. Heikman, A. Chakraborty, A. Chini, S. Keller, S. P. DenBaars and U. K. Mishra, “High-power polarization-engineered GaN/AlGaN/GaN HEMTs without surface passivation”, IEEE Electron Device Letters, vol. 25, no. 1, pp. 7-9, Jan. 2004.
D. Buttari, A. Chini, T. Palacios, R. Coffie, L. Shen, H. Xing, S. Heikman, L. McCarthy, A. Chakraborty, S. Keller and U. K. Mishra, “Origin of etch delay time in Cl2 dry etching of AlGaN/GaN structures,” Applied Physics Letters, vol. 83, no. 23, pp. 4779-4781, Dec. 2003. R. Coffie, L. Shen, G. Parish, A. Chini, D. Buttari, S. Heikman, S. Keller and U. K. Mishra, “Unpassivated p-GaN/AlGaN/GaN HEMTs with 7.1W/mm at 10GHz,” IEE Electronics Letters, vol. 39, no. 19, pp. 1419-1420, Sep. 2003. L. Shen, R. Coffie, S. Heikman, D. Buttari, A. Chini, A. Chakraborty, S. Keller, S. P. DenBaars and U. K. Mishra, “Polarization-engineered GaN/AlGaN/GaN HEMTs with record high power without passivation,” Proceedings of the 61st Device Research Conference, Late News, pp.2-3, Jun. 2003. L. Shen, A. Chini, R. Coffie, D. Buttari, S. Heikman, S. Keller, and U. K. Mishra, “Temperature dependence of the Current-Voltage Characteristics of AlGaN/ GaN HEMT,” Proceedings of the 61st Device Research Conference, pp.63-64, Jun. 2003. L. Shen, S. Heikman, Y. Wu, R. Coffie, D. Buttari, A. Chini, L. McCarthy, S. Keller and J. Speck, U. K. Mishra, “GaN/AlGaN/GaN heterostructure and its application to the dispersion removal in HEMTs,”, presented at MRS Spring Meeting, San Francisco, CA, U.S.A., Apr. 2003. L. Shen, I. P. Smorchkova, D. Green, S. Heikman, U. K. Mishra, “GaN planar-doped-barrier electron emitter with piezoelectric surface barrier lowering,” Journal of Vacuum Science & Technology B, Vol. 21, No. 1, pp. 540–543, Jan. 2003 R. Coffie, D. Buttari, S. Heikman, S. Keller, A. Chini, L. Shen, and U. K. Mishra, “p-capped GaN-AlGaN-GaN high-electron mobility transistors (HEMTs),” IEEE Electron Device Letters, vol. 23, no. 10, pp. 588-590, Oct. 2002. S. Keller, S. Heikman, L. Shen, I. P. Smorchkova, S. P. DenBaars, and U. K. Mishra, “GaN-GaN junctions with ultrathin AlN interlayers: Expanding heterojunction design,” Applied Physics Letters, vol. 80, no. 23, pp. 4387 – 4389, Jun. 2002.
viii
A. Jimenez, D. Buttari, D. Jena, R. Coffie, S. Heikman, N. Q. Zhang, L. Shen, E. Calleja, E. Munoz, J. Speck, and U. K. Mishra, “Effect of p-doped overlayer thickness on RF-dispersion in GaN junction FETs,” IEEE Electron Device Letters, vol. 23, no. 6, pp. 306-308, Jun. 2002. D. Buttari, A. Chini, G. Meneghesso, E. Zanoni, P. Chavarkar, R. Coffie, N. Q. Zhang, S. Heikman, L. Shen, H. Xing, C. Zheng, and U. K. Mishra, “Systematic characterization of Cl2 reactive ion etching for gate recessing in AlGaN/GaN HEMTs,” IEEE Electron Device Letters, vol. 23, no. 3, pp. 118-120, Mar. 2002. D. Buttari, A. Chini, G. Meneghesso, E. Zanoni, B. Moran, S. Heikman, N. Q. Zhang, L. Shen, R. Coffie, S. P. DenBaars, and U. K. Mishra, “Systematic characterization of Cl2 reactive ion etching for improved ohmics in AlGaN/GaN HEMTs,” IEEE Electron Device Letters, vol. 23, no. 2, pp. 76-78, Feb. 2002. I.P. Smorchkova, L. Chen, T. Mates, L. Shen, S. Heikman, B. Moran, S. Keller, S. P. DenBaars, J. S. Speck, and U. K. Mishra, “AlN/GaN and (Al,Ga)N/AlN/GaN two-dimensional electron gas structures grown by plasma-assisted molecular-beam epitaxy,” Journal of Applied Physics, Vol. 90, No. 10, pp. 5196–5201, Nov. 2001. S. Keller, S. Heikman, I. Ben-Yaacov, L. Shen, S. P. DenBaars, and U. K. Mishra, “Indium surfactant assisted growth of AlN/GaN heterostructures by metal-organic chemical vapor deposition,” Wiley-VCH. Physica Status Solidi A, vol.188, no.2, pp.775-778, Nov. 2001. S. Keller, S. Heikman, I. Ben-Yaacov, L. Shen, S. P. DenBaars, and U. K. Mishra, “Indium-surfactant-assisted growth of high-mobility AlN/GaN multilayer structures by metalorganic chemical vapor deposition,” Applied Physics Letters, vol.79, no.21, pp.3449-51, Nov. 2001. L. Shen, S. Heikman, B. Moran, R. Coffie, N. Q. Zhang, D. Buttari, I. P. Smorchkova, S. Keller, S. P. DenBaars, and U. K. Mishra, “AlGaN/AlN/GaN high-power microwave HEMT,” IEEE Electron Device Letters, vol. 22, no. 10, pp. 457-459, Oct. 2001. H. Xing, S. Keller, Y-F Wu, L. McCarthy, I. P. Smorchkova, D. Buttari, R. Coffie, D. S. Green G. Parish, S. Heikman, L. Shen, N. Q. Zhang, J. Xu, B. P. Keller, S. P. DenBaars, and U. K. Mishra, “Gallium nitride based transistors,” Journal of Physics-Condensed Matter, vol. 13, no. 32, pp. 7139-7157, Aug. 2001. L. Shen, I. P. Smochkova, D. Green, S. Heikman, U. K. Mishra, “GaN planar-doped-barrier electron emitter with piezoelectric surface barrier lowering,” Proceedings of the 14th International Vacuum Microelectronics Conference, pp.223-224, Aug. 2001.
ix
J. Wu, L. Shen, L. Zhou, “Nonlinear optical study of ER fluids,” Journal of Intelligent Material Systems & Structures, vol.7, no.5, pp. 565-568, Sept. 1996.
x
Abstract
Advanced Polarization-Based Design of AlGaN/GaN HEMTs
By
Likun Shen
During the past few years, enormous progress has been made in the
development of GaN-based devices. Due to high breakdown field, high sheet charge
density and high electron saturation velocity, GaN-based HEMTs have great potential
for high frequency high power applications. Extensive research has being carried out
on the material growth and the device structure. This dissertation focuses on the
efforts to develop novel epitaxial structures to improve the electron mobility and
suppress the dispersion without surface passivation. Relying on the utilization of
strategic band engineering and polarization charge, unpassivated high power GaN-
based HEMTs with minimal dispersion have been demonstrated.
The application of AlN in GaN-based HEMT is discussed. AlN is a binary
material, thereby alloy disorder scattering is eliminated which improves the electron
mobility. The high polarization field in AlN also results in high carrier concentration.
Low sheet resistance is observed in AlN/GaN heterostructures. The incorporation of a
thin AlN layer in an AlGaN/GaN HEMT is investigated, resulting in an
AlGaN/AlN/GaN structure. Due to the absence of alloy disorder scattering, and the
reduction of wavefunction penetration into AlGaN, the electron mobility is improved.
xi
Carrier concentration is also improved slightly due to the high polarization effect of
AlN. The DC and RF performances are presented.
Dispersion at different temperatures is presented. Increased dispersion is
observed at lower temperature. Hopping conduction and de-trapping/band conduction
models are discussed.
The concept of a thick GaN cap on top of an AlGaN/GaN HEMT is proposed
to reduce dispersion at epitaxial level without passivation. This approach utilizes a
thick cap layer to increase the distance between the channel and surface, thereby
screening the surface potential fluctuations. A GaN/AlGaN/GaN heterostructure is
investigated. Dispersion is suppressed without passivation. In order to decrease the
leakage current and increase the breakdown voltage, several variations of device
structures are discussed. By employing a SiO2 insulating layer, lowering Si doping
sheet density and utilizing a thick graded AlGaN cap layer, leakage current is reduced
and breakdown voltage is increased. These improvements resulted in a record output
power density of 12W/mm at 10GHz for GaN-based HEMTs without passivation.
xii
To,
My parents, Hongguang Shen and Zhanmei Sun,
and my wife, Xiaojie.
xiii
Table of Contents 1 Introduction………………………………………………….. 1
1.1 Advantages of GaN microwave power devices ……………………... 11.2 Research background of GaN-based transistors……………………... 41.3 Motivations for the research presented in the dissertation…………... 61.4 Synopsis of the dissertation………………………………………….. 81.5 References……………………………………………………………. 10
2 AlN and its Application in Improving 2DEG
Transport Properties……………………………………...... 12
2.1 Introduction…………………………………………………………. 122.2 Scattering mechanisms in AlGaN/GaN HEMTs……………………. 152.3 AlN/GaN heterostructures................................................................... 222.4 GaN/AlN/GaN heterostructures........................................................... 312.5 AlGaN/AlN/GaN HEMTs................................................................... 352.6 Summary.............................................................................................. 482.7 References............................................................................................ 49
3 DC-to-RF Dispersion………………………………………… 51
3.1 Introduction………………………………………………………….. 513.2 Dispersion at low temperature………………………………………. 553.3 Models………………………………………………………………. 643.4 Summary…………………………………………………………….. 703.5 References…………………………………………………………… 71
4 Thick GaN Capped AlGaN/GaN HEMTs…………………. 73
4.1 Introduction………………………………………………………….. 734.2 Solutions to dispersion control in the epitaxial structure……………. 784.3 Thick GaN capped AlGaN/GaN HEMTs……………………………. 864.4 Summary……………………………………………………………… 1204.5 References…………………………………………………………..... 121
xiv
5 Improvements of GaN Capped AlGaN/GaN HEMTs…….. 124
5.1 SiO2 insulating layer ………………………………………………… 1255.2 Effects of Si doping sheet density…………………………………… 1405.3 Thick graded AlGaN capped AlGaN/GaN HEMTs………………… 1555.4 Summary…………………………………………………………….. 1635.5 References…………………………………………………………… 165
6 Summary, Conclusions and Future work………………….. 166
6.1 Summary and conclusions…………………………………………… 1666.2 Future work…………………………………………………………... 1716.3 References……………………………………………………………. 175
xv
Chapter 1
Introduction
1.1 Advantages of GaN microwave power devices
With the recent upsurge of wireless communication market, as well as the
steady but continuous progress of the traditional military applications (for example,
radar system), microwave transistors are playing critical roles in many aspects of
human activities. The requirements for the performance of microwave transistors are
becoming more and more demanding. In the personal mobile communication
applications, next generation cell phones require higher frequencies. The
development of satellite communications and TV broadcasting requires higher
frequencies (from C band to Ku band, further to Ka band) and higher power to reduce
the antenna size of terminal users. The same requirement holds for broadband
wireless internet connection as well because it requires high speed data transmission
rate. Being the key component in the technical development, microwave transistors
and amplifiers have attracted much attention in the recent years. Several existing and
developing technologies are Si/SiGe, GaAs, SiC and GaN. Table 1.1 lists the major
1
parameters of these materials and the Johnson’s figure of merit calculated to compare
the power-frequency limits of different materials.
Table 1.1 Material properties related to the power performance at high frequencies for various materials
Si GaAs 4H-SiC GaN Diamond
Eg (eV) 1.1 1.42 3.26 3.39 5.45
ni (cm-3) 1.5×1010 1.5×106 8.2×10-9 1.9×10-10 1.6×10-27
εr 11.8 13.1 10 9.0 5.5
µn
(cm2/V s) 1350 8500 700 1500(2DEG) 1900
vsat
(107cm/s) 1.0 2.0 2.0 2.5 2.7
Ebr
(MV/cm) 0.3 0.4 3.0 3.3 5.6
Θ (W/cm
K) 1.5 0.43 4.9 1.3 20
JM 1 2.7 20 27.5 50
Johnson’s figure of merit (JM) is defined as [1]:
2br satE vJMπ
= (1.1)
It gives the power-frequency limit based solely on material properties and can be used
to compare different materials for high frequency and high power applications.
Si is able to satisfy applications requiring low power and at the lower end of
microwave frequency. In addition to sophisticated processing and device design, the
incorporation of SiGe allows Si-based technology to keep improving. However, the
2
limited intrinsic material properties (e.g. low breakdown electric field of 0.3MV/cm
and low electron saturation velocity of 1×107cm/s) make it difficult to satisfy high
frequency and high power applications. GaAs has a high saturation velocity of
~2×107cm/s and a very high low field mobility of 8500cm2/V s, but the narrow band-
gap results in a low breakdown field of 0.4MV/cm. It prevents the application of high
operation voltage, thereby limiting the output power density to 1-2W/mm. The
application of high power and high frequency requires both large breakdown voltage
and high electron velocity. From this point of view, wide bandgap materials, like GaN
and SiC, with higher JM are preferable. The wide bandgap results in higher
breakdown voltage because the ultimate breakdown field is the field required for
band-to-band impact ionization. Moreover, both have high electron saturation
velocity which allows high frequency operation. The ability of GaN to form
heterojunctions makes it superior compared to SiC, in spite of having similar
breakdown field and saturation electron velocity. GaN can be used to fabricate
HEMTs whereas SiC can only be used to fabricate MESFETs. The advantage of the
HEMT is its higher electron mobility due to reducing ionized impurity scattering,
because the electrons in the channel remain separated from the dopants [2]. The
combination of high carrier concentration and high electron mobility result in a low
on-resistance Ron, which is especially important for power switching applications. It
can be noted in table 1.1 that diamond has good properties in almost every aspect.
However, the immaturity of the doping technique in diamond, especially the absence
of a shallow n-type donor, has made it difficult to fabricate a microwave transistor.
3
From the amplifier point of view, GaN-based HEMTs have many advantages
over contemporary technologies (e.g. GaAs) [3]. The high output power density
allows the fabrication of much smaller size device with the same output power.
Higher impedance due to smaller size allows for easier and lower loss matching in
amplifiers. The operation at high voltage due to its high breakdown electric field not
only reduces the need for voltage conversion, but also provides the potential to obtain
high efficiency which is a critical parameter for amplifiers. The wide bandgap also
enables it to operate under high temperature. At the same time, the HEMT topology
gives device better noise performance than that of MESFET topology.
In conclusion, the high carrier concentration, the high electron saturation
velocity and the large breakdown voltage make GaN-based HEMT a very promising
candidate for microwave power applications.
1.2 Research background of GaN-based device
Compared with commercialized GaN-based optical devices, GaN-based
electronic devices are relatively immature. Although all kinds of typical electronic
devices were investigated (for example, HBT [4], MESFET [5], MISFET [6],
HEMT), most of the research work has been focused on HEMTs, because HEMTs
have better carrier transport properties than MESFET and the difficulty of p-doping in
GaN impedes the development of bipolar transistors.
The first observation of a Two-Dimensional Electron Gas (2DEG) with a
carrier concentration of the order of 1011cm-2 and a room temperature mobility of
4
400-800cm2/Vs in an AlGaN/GaN heterojunction was reported by Khan et al. [7].
The first DC performance of AlGaN/GaN HEMT was reported in 1993 by Khan et al.
with the saturation drain current of 40mA/mm [8]. In 1994, small signal RF
performance of AlGaN/GaN HEMT was reported by Kahn et al. with the current-gain
and power gain cutoff frequencies of 11 and 35GHz, respectively, for a 0.25µm-gate-
length device [9]. First RF power data of 1.1W/mm at 2GHz for an AlGaN/GaN
HEMT was reported by Wu et al. in 1996 [10]. Since then, the output power density
has increased steadily. The performance progress is due to greatly improved growth
techniques, material qualities and enhanced processing technologies. Two innovations
of the latter are especially important. One was the introduction of the SiN passivation
in 2000 [11], which effectively reduced DC-to-RF dispersion caused by surface trap
states, thereby resulting in a big increase in output power to more than 11W/mm on
SiC substrate [12]. Another was the adoption of the field plate in 2003 [13] [14]. In
addition to the tradition function of the field plate to increase the breakdown voltage,
it also reduced the dispersion beyond what SiN passivation offered. Output power
density of 18W/mm on SiC and 12W/mm on sapphire at 4GHz were reported [14]
[15]. The latest record for power is ~32W/mm at 8GHz with a drain bias larger than
100V [16]. Encouraging performance at mm-wave frequencies, e.g. 3.5W/mm at
30GHz, have also been demonstrated recently [17].
GaN-based HEMT have also exhibited better linearity than the existing
devices. 2.4W/mm with PAE of 53% with a carrier to third-order intermodulation
5
ratio of 30 dBc at 4GHz was reported [15], which was much better than that of GaAs-
based HEMTs.
1.3 Motivations for the research presented in the
dissertation
Although significant progress has been achieved recently, all these devices
needed SiN passivation to reduce dispersion. DC-to-RF dispersion has been identified
as a major cause for the discrepancy between the measured load-pull power and the
prediction from the static I-V characteristics. Gate-lag and RF I-V measurements
revealed that the current collapse and knee-voltage walkout occurred at high
frequency, due to the deep level traps in GaN. The introduction of the SiN passivation
reduced the dispersion effectively and greatly improved power performance.
However, there were several disadvantages. The effect of SiN surface passivation was
very sensitive to the deposition conditions as well as surface conditions prior to the
passivation. This sensitivity combined with the lack of understanding of the
passivation mechanism have resulted in poor reproducibility of the breakdown
voltage, gate leakage and dispersion reduction. These disadvantages have motivated
research into dispersion reduction at the device epitaxial level. Several remedies have
been proposed: regrown drain access region HEMT, p-GaN capped AlGaN/GaN
6
HEMT [18] and thick UID GaN (or graded AlGaN)-capped AlGaN/GaN HEMT [19].
In this dissertation, we will focus on the thick UID GaN cap solution.
Another topic that is discussed in this dissertation is the introduction of a thin
AlN layer in the channel to improve the 2DEG transport properties. Due to the large
effective electron mass in GaN, the electron mobility is lower than GaAs. Similar to
other III-V materials, at room temperature, phonon scattering is the most important
scattering mechanism whereas at low temperature, interface roughness scattering and
alloy disorder scattering are dominant. Studies revealed that because of high carrier
concentration (>1013cm-2) in AlGaN/GaN HEMT, alloy disorder scattering also plays
an important role at room temperature. The introduction of a thin AlN layer between
the AlGaN and the GaN channel was shown to improve both charge and mobility due
to enhanced electron transfer to the channel and reduced alloy disorder scattering,
thereby resulting in a lower on-resistance and higher current.
The final goal of this dissertation is to demonstrate high power GaN-based
HEMTs with minimal low temperature and room temperature dispersion without
surface passivation, relying only on the use of strategic band engineering and the
utilization of polarization charge.
7
1.4 Synopsis of the dissertation
In chapter 2, the application of AlN in GaN-based HEMT to improve the
2DEG transport properties is discussed. After a short review of scattering
mechanisms, the understanding of growth of AlN/GaN heterostructure, and its affect
on carrier density and mobility is presented. The behavior of the carrier concentration
in a GaN/AlN/GaN heterostructure is then explained by considering induced positive
charge accumulation. An AlGaN/AlN/GaN heterostructure with a thin AlN interfacial
layer is then introduced. Its effect on 2DEG density and mobility are discussed in
detail. The improvement on device performance is demonstrated.
Dispersion is discussed in Chapter 3. The concept of virtual gate is briefly
reviewed. The dependence of the pulsed I-V characteristics on the temperature
(varying from 300K to 77K) is reported. The trends of the changes in DC current,
pulsed current and dispersion are explained. Two models, one involving band-
conduction of de-trapped electrons and another involving hopping conduction, are
introduced to explain the dispersion behavior. Their applicability at different
temperatures is discussed and matched to experimental observation.
Based on the virtual gate model, the solutions to reduce the dispersion at
epitaxial level are presented in Chapter 4. Different mechanisms and their advantages
and disadvantages are studied. A thick UID GaN cap AlGaN/GaN heterostructure is
discussed in detail which uses a thick cap layer to screen the surface potential
fluctuations. The behaviors of carrier concentration and pinch-off voltage are checked
by both simulation and experiments. The effects of the epitaxial parameters on the
8
device performance are investigated. Processing issues are studied as well. The first
GaN/AlGaN/GaN HEMT with the reduction of dispersion without surface passivation
is demonstrated. Its limitations (high leakage current and low breakdown voltage) are
also discussed.
The improvements of the novel GaN/AlGaN/GaN HEMT are reported in
Chapter 5. The leakage path of the high gate leakage is identified. The application of
the insulating material in the leakage path to reduce leakage is discussed and
implemented, resulted in much improved power performance. The effect of doping in
the graded AlGaN is investigated. Its relation to the leakage is revealed and
explained. The introduction of a thick graded AlGaN layer as the cap to reduce the
leakage is also discussed and its effects on leakage current, breakdown voltage and
output power are studied experimentally.
In Chapter 6, summary and conclusions are presented. Possible future
improvements in growth, processing and device structure are discussed as well,
including the optimization of growth on SiC substrate, etch-stop technique, ion
implantation and epitaxial field plate.
9
1.5 References [1] E. O. Johnson, “Physical Limitation on Frequency and Power Parameters of
Transistors,” RCA Rev., vol. , pp. 163- 176, June 1965.
[2] W. Liu, Fundamentals of III-V Devices – HBTs, MESFETs, and HFETs/HEMTs,
Chapter 5, A Wiley-Interscience Publication, John Wiley & Sons, Inc., 1999.
[3] U. K. Mishra, P. Parikh, Y.-F. Wu, “AlGaN/GaN HEMTs – An overview of deivce operations and applications,” Proccedings of the IEEE. vol. 90, No. 6, pp. 1022- 1031, June 2002.
[4] H, Xing, D. S. Green, L. McCarthy, I. P. Smorchkova, P. Chavarkar, P. Mates, S.
Keller, S. Denbaars, J. Speck, and U. K. Mishra, “Progress in Gallium Nitride-based Transistors,” Proccedings of the 2001 BIPOLAR/BiCMOS Circuits and Technology Meeting, pp. 125-130, 2001.
[5] M. Asif Khan, J. N. Kuznia, A. R. Bhattarai, and D. T. Olson, “Metal
semiconductor field effect transistor based on single crystal GaN,” Appl. Phys. Lett. Vol. 62, no. 15, pp. 1786-1787, April 1993.
[6] A. Chini, J. Wittich, S. Heikman, S. Keller, S. P. DenBaars, U. K. Mishra UK,
“Power and linearity characteristics of GaN MISFETs on sapphire substrate,” IEEE Electron Device Letters, vol.25, no.2, pp.55-7, Feb. 2004.
[7] M. Asif Khan, J. N. Kuznia, J. M. Van Hove, N. Pan, and J. Carter, “Observation of a two-dimensional electron gas in low pressure metalorgnic chemical vapor deposited GaN-AlGaN heterojunctions,” Appl. Phys. Lett., Vol. 60, no. 24, pp. 3027-3029, June 1992.
[8] M. Asif Khan, A. Bhattarai, J. N. Kuznia, and D. T. Olson, “High electron
mobility transistor based on a GaN-AlxGa1-xN heterojunction,” Applied Physics Letters, vol. 63, no. 9, pp. 1214-1215, Aug. 1993.
[9] M. Asif Khan, J. N. Kuznia, and D. T. Olson, W. J. Schaff and J. W. Burm, M. S.
Shur, “Microwave performance of a 0.25µm gate AIGaN/GaN heterostructure field effect transistor,” Applied Physics Letters, vol. 65, no. 9, pp. 1121-1123, Aug. 1994.
10
[10] Y.-F. Wu, B.P. Keller, S. Keller, D. Kapolnek, S.P. Denbaars, U.K. Mishra “Measured microwave power performance of AlGaN/GaN MODFET,” IEEE Electron Device Letters, vol. 17, no. 9, pp. 455-457, Sept. 1996.
[11] B. M. Green, K. K. Chu, E. M. Chumbes, J. A. Smart, J. R. Shealy, L. F.
Eastman, “The effect of surface passivation on the microwave characteristics of undoped AlGaN/GaN HEMTs,” IEEE Electron Device Lett., Vol. 21 no. 6, pp. 268-270, June 2000.
[12] J.R. Shealy, V. Kaper, V. Tilak, T. Prunty, J.A. Smart, B. Green and L.F.
Eastman, “An AlGaN/GaN high-electron-mobility transistor with an AlN sub-buffer layer,” J. Phys.: Condens. Matter, 2002, vol. 14, p.3499.
[13] Y. Ando, et al.: ‘10W/mm AlGaNGaN HFET with a field modulating plate’,
IEEE Electron Device Lett., Vol.24, No. 5, pp. 289–291, 2003 [14] A. Chini, D. Buttari, R. Coffie, S. Heikman, S. Keller, U. K. Mishra, “12 W/mm
power density AlGaN/GaN HEMTs on sapphire substrate,” Electronics Letters, vol.40, no.1, 8 Jan. 2004, pp.73-4. Publisher: IEE, UK.
[15] A. Chini, D. Buttari, R. Coffie, L. Shen, S. Heikman, A. Chakraborty, S. Keller,
U. K. Mishra, “Power and Linearity Characteristics of Field-Plated Recessed-Gate AlGaN-GaN HEMTs,” Electron Device Letters, Accepted for future publication, 2004
[16] Y.-F. Wu, A. Saxler, M. Moore, R. P. Smith, S. Sheppard, P.M. Chavarkar, T.
Wisleder, U. K. Mishra, P. Parikh, “30-W/mm GaN HEMTs by Field Plate Optimization,” IEEE Electron Device Letters, Vol. 25, No. 3, pp.117-119, March 2004.
[17] Y.-F. Wu, M. Moore, A. Saxler, R. P. Smith, P.M. Chavarkar, P. Parikh 3.5-watt,
“AlGaN/GaN HEMTs and amplifiers at 35 GHz,” IEEE International Electron Devices Meeting, 2003. IEDM '03 Technical Digest., pp. 23.5.1 - 23.5.3, Dec. 2003
[18] R. Coffie, D. Buttari, S. Heikman, S. Keller, A. Chini, L. Shen, and U. K. Mishra, “p-capped GaN-AlGaN-GaN high-electron mobility transistors (HEMTs),” IEEE Electron Device Letters, vol. 23, no. 10, pp. 588-590, Oct. 2002.
[19] L. Shen, R. Coffie, S. Heikman, D. Buttari, A. Chini, A. Chakraborty, S. Keller,
S. P. DenBaars, and U. K. Mishra, “High-power polarization-engineered GaN/AlGaN/GaN HEMTs without surface passivation,” IEEE Electron Device Letters, vol. 25, no. 1, pp. 7-9, Jan. 2004.
11
Chapter 2
AlN and its application in improving 2DEG transport properties
2.1 Introduction
Microwave power is an important application for GaN-based HEMTs. From
equation (2.1), it can be seen that the maximum linear output power is proportional to
the maximum current of a device. Equation (2.2) describes the well-known long-
channel current–voltage characteristics of HEMTs [1]. It can be seen that for fixed
bias conditions, the current density increases when the product of the sheet charge
density of 2DEG, ns, and the electron mobility, µn, increases.
, max ,1 I (4out lin DS bias kneeP V= ⋅ − )V (2.1)
' 2
[( ) ]2
OX n DSD GS T DS
WC VI V V VLµ
= − ⋅ −
' 2
[2
OX n DSs DS
WC Vn VLµ α= ⋅ ⋅ − ] (2.2)
12
where W is the gate width, L is the gate length, COX is the unit-area
capacitance, and VDS is the drain bias voltage. The 2DEG density ns is proportional to
the difference between VGS, the gate-source voltage, and VT, the threshold voltage
[1].
Therefore, increasing the product of nsµn is critical to obtaining high power
density. Moreover, in applications such as power switching, a high nsµn product is
also desirable as it is inversely proportional to on-resistance for the device, which is a
key figure of merit for power switching devices.
The nsµn product represents the conductivity of the 2DEG. The sheet
resistance extracted from Hall measurements is often used to evaluate this property.
For most of the AlGaN/GaN HEMT structures, sheet resistance ranges from 200 to
500Ω/ٱ.
1sh
s n
Rqn µ
= (2.3)
The relationship between ns and µn has to be considered when optimizing the
nsµn product. For example, it is well-known that increasing the Al mole fraction of
AlGaN barrier can increase 2DEG density. In an AlGaN/GaN HEMT, the 2DEG
sheet charge density approaches the net polarization charge density at AlGaN/GaN
interface with increasing AlGaN thickness. The polarization charge density increases
in turn with Al mole fraction of the AlGaN barrier. However, this increase of the
13
2DEG density is accompanied with the dropping of the electron mobility, which is
due to the higher scattering rate since the centroid of the 2DEG moves closer to the
AlGaN/GaN interface at higher charge density. Therefore, some new features need to
be employed to maintain a high mobility at high charge concentration.
In this chapter, we will focus on the application of AlN to improve both the
2DEG density and electron mobility, utilizing its high polarization charge density and
the lack of alloy disorder scattering due to its binary nature. First, the scattering
mechanisms in AlGaN/GaN HEMTs will be reviewed. Then an AlN/GaN structure
will be discussed. It has both high 2DEG density and mobility, thereby resulting in
very high channel conductivity. As a variation, a GaN/AlN/GaN heterostructure will
be investigated as well. Its unique property of carrier concentration change as a
function of GaN cap thickness will be explained. Finally, a novel AlGaN/AlN/GaN
HEMT structure that included the insertion of a thin AlN layer at the AlGaN/GaN
interface will be proposed and its effects on the mobility and sheet charge density will
be discussed in detail. Superior 2DEG properties and output power density will be
presented to show the improvement of the device performance after adopting this new
structure.
14
2.2 Scattering mechanisms in AlGaN/GaN HEMTs
There are several scattering mechanisms which play a role in limiting the
mobility of 2DEG in AlGaN/GaN HEMTs. In addition to the traditional phonon
scattering, interface roughness scattering, charged impurities scattering and alloy
disorder scattering, dislocation scattering and dipole scattering are unique to GaN-
based HEMTs, due to the large dislocation density and strong polarization effects in
the GaN-based materials system.
Phonon scattering
Phonon scattering is the most important limiting factor at high temperature.
Three types of phonons are usually considered: acoustic phonons caused by the
deformation potential coupling and the piezoelectric coupling and polar optical
phonons [2].
The energy of acoustic phonons is very low and the scattering is essentially
elastic. A relaxation time can be defined. The coupling of the electron transport to
acoustic phonons can be through deformation potential or piezoelectric components.
Studies have suggested that piezoelectric component of acoustic phonon scattering is
weaker than that of the deformation potential and can be neglected [3]. Usually, only
the longitude-mode acoustic phonon is considered in calculations.
Polar optical phonons have a large energy and play a very important role in
the mobility at high temperature. Fig. 2.1 shows the dependence of mobility on
15
carrier density due to optical and acoustic phonon scattering at room temperature [2].
The mobility decreases when the carrier density increases.
Fig.2.1. Room temperature mobility due to optical andacoustic phonon scattering as a function of 2DEG density.Mobility drops when charge increases [2] (courtesy of Dr.Jena).
16
Interface roughness scattering [4]
Because the sheet charge density of the 2DEG in AlGaN/GaN HEMTs is very
high, the scattering at rough interface can be severe due to the high interface electric
field. At low temperature it can be one of the major causes of decreased mobility.
This limits transport even in the presence of a binary barrier.
Remote ionized impurities
Unlike AlGaAs/GaAs HEMTs, AlGaN/GaN HEMTs are polarization-doped.
That means there are no intentionally doped impurities in the barrier layer. Instead,
there are positive charged donor-like surface states at the surface of the AlGaN layer
[5]. Studies have revealed that the effect on mobility due to this mechanism is not a
strong function of the sheet charge density, but rather the barrier thickness. In most of
the practical AlGaN/GaN HEMTs, the thickness of the AlGaN barrier layer is around
30 nm, which is thick enough to make this scattering a relatively weak one.
Background residual impurities
Although the GaN and AlGaN are not intentionally doped, there are still
unintentional residual background donors in them due to a non-ideal growth process.
These donors are believed to be unwanted oxygen and silicon atoms that incorporate
during the growth. State of the art AlGaN/GaN structures have a background impurity
density of about 1016cm-3. Therefore the contribution of this mechanism to scattering
is relatively weak. Even if the background impurities density is high (>1018cm-3), this
17
contribution to scattering is still weak. Because of the high sheet charge density of the
2DEG, the effect of scattering due to background impurities is largely screened.
Dislocation scattering
Due to the unavailability of bulk GaN substrates, GaN-based devices are
usually grown on SiC or sapphire substrates. The resulting large lattice mismatch
leads to a much higher dislocation density compared to other well-known III-V
materials. State of the art AlGaN/GaN HEMTs have a dislocation density varying
from 108 to 1010cm-2. This fact makes dislocations play a much more important role
in GaN-based devices than those in GaAs-based devices. The charged dislocations
can scatter electrons in the 2DEG [6]. Moreover, even uncharged dislocations can
change the strain field around them with atoms displaced from their equilibrium
positions, thus scattering electrons.
Dipole dislocations
Dipole dislocations are another kind of unique scattering mechanism to GaN-
based materials due to the strong polarization effects [7]. AlGaN is a ternary alloy
and is a disordered system with Al and Ga atoms arranged in a random array. The
difference of polarization charges between AlN and GaN leads to dipoles of randomly
fluctuating magnitude, which contribute to the scattering of the electrons in the
channel.
18
Alloy disorder scattering
Alloy disorder scattering originates from the randomly varying alloy potential
in the barrier [8] [9]. Although the centroid of the 2DEG is in the binary material
GaN, there is a penetration of the wavefunction into the ternary AlGaN barrier. As
expected, the higher the Al mole fraction, the lower the probability of the penetration
because the discontinuity between the conduction band of GaN and that of higher Al
mole fraction AlGaN, i.e. effective barrier height, is larger. It is also noted that the
penetration becomes larger when the carrier density increases. The shift of the
wavefunction towards to the interface can partially explain the decrease of the
mobility when the sheet charge density increases. The deeper penetration implies
more alloy disorder scattering. As the wavefunction moves closer to the interface of
the 2DEG the interface roughness scattering becomes more obvious.
Although this effect in other III-V material systems is weak, it is an important
one in AlGaN/GaN heterostructures, due to the large electron effective mass, the high
sheet charge density and the large alloy scattering potential. Fig.2.2 shows the alloy-
disorder-scattering-limited electron mobility as a function of sheet charge density and
Al mole fraction [2]. It is important to note that the mobility decreases when the
carrier density increases, because more electrons push the wavefunction closer to the
interface resulting in deeper penetration as discussed above. When the Al content
decreases, the mobility is decreased. This is not surprising when considering deeper
penetration of the wavefunction into the AlGaN when the effective barrier height
reduces as a result of a lower Al mol fraction as discussed above.
19
Fig.
mechanism
very impor
Other scat
mobilities a
Fig.2.2. Alloy-disorder-scattering-limited electron mobility fora range of 2DEG densities and alloy compositions. Mobilitydrops when charge increases or Al composition increases [2](courtesy of Dr. Jena).
2.3 shows the calculated electron mobilities limited by different scattering
s at low temperature. It can be found that alloy disorder scattering plays a
tant role at low temperature as well as interface roughness scattering.
tering mechanisms are relatively weak. Different scattering-limited
lso show different dependence on carrier density.
20
Fig.2.3. Calculated electron mobilities limited by differentscattering mechanisms vs. carrier concentration at lowtemperature [2] (courtesy of Dr. Jena).
21
2.3 AlN/GaN heterostructures
In the AlGaN/GaN HEMT structure, optical phonon scattering is the major
contribution to scattering at room temperature. At low temperatures, since phonon
scattering is largely suppressed, alloy disorder scattering and interface roughness
scattering become the two major dominating factors, depending upon the nature of
the barrier. Alloy scattering dominates for AlGaN barriers for all Al compositions. It
is also important to note that the effect of the alloy disorder scattering is very
significant even at room temperature when the carrier density becomes very high. The
alloy-scattering-limited mobility approaches the limits set by optical phonon
scattering.
High quality of growth can improve the mobility; for example, reduction of
the interface roughness can decrease the interface roughness scattering. However,
here we will try to discuss how the mobility can be improved at the epitaxial level,
i.e. how the mobility can be increased by adopting novel epitaxial structures. We will
only focus on alloy disorder scattering in this chapter.
As discussed above, there are two essential reasons for alloy disorder
scattering in AlGaN/GaN heterostructures: one is that the barrier is AlGaN, which is a
ternary material and has alloy disorder. The second reason is that there is a
wavefunction penetration into the AlGaN barrier, which is a quantum effect. To
reduce or remove this effect, either a binary material is used for the barrier, or the
penetration is suppressed. To suppress penetration, increasing the Al mole fraction of
the AlGaN barrier can be helpful, since when the Al composition increases the
22
AlGaN has a larger conduction band discontinuity. This implies a larger barrier for
the 2DEG, therefore reducing the penetration. Fig.2.4 shows the different
wavefunction penetration into the barrier in AlGaN/GaN HEMT structures with
various Al compo itions. The trend is very obvious: the higher
Fig.2integAl0.22
the Al mole fract
amount of penetr
composition is. T
the practical barri
completely, we h
ternary one, in th
is a binary materi
member of the A
the wavefunction
discussed below,
s
24 26 28 30 32 34 36
0.00
0.02
0.04
0.06
0.08 Al0.22Ga0.78N/GaN AlN/GaN
AlGaN/GaNinterface
Prob
abili
ty
Distance (nm)
.4. Wavefunction penetration into barrier layer. Therated probability or electrons in barrier is 7.2% forGa0.78N/GaN and 4.8% for AlN/GaN.
ion is, the smaller the penetration is. However, there is still a small
ation of wavefunction into the barrier no matter how high the Al
he probability of an electron in the barrier cannot be zero because
er height cannot be infinite. Therefore in order to remove this effect
ave to consider another method, using a binary material, instead of a
e barrier. Fortunately, there is an ideal candidate – AlN. Since AlN
al, there are no alloy disorder effects in it. Moreover, AlN is an end
lGaN system and the Al composition is therefore 1. This implies that
penetration is suppressed substantially. In the AlN/GaN structure
this is not so important because there is no alloy disorder scattering
23
since the barrier consists of a binary material. However, in another novel structure
which will be discussed in the next section, this suppression introduced by AlN is
helpful. Also, the large conduction band discontinuity has the potential to reduce hot
electron injection into the barrier during device operation.
Not only from the mobility point of view, but also from the sheet charge
density point of view, the AlN/GaN heterostructure is highly desirable since the
polarization charge density (~5.6×1013cm-2) in AlN can lead to a very high 2DEG
sheet charge density. Fig. 2.5 shows the calculated channel carrier density as function
of AlN barrier thickness. The saturation value of the 2DEG density can be as high as
5×1013cm-2 when the thickness of the AlN barrier exceeds 10nm. This value is almost
3-4 times higher than the carrier density usually measured from Al0.3Ga0.7N/GaN
HEMTs.
0 5 10 15 20 25 300
1
2
3
4
5
6
2DEG
den
sity
(1013
cm-2)
AlN barrier thickness (nm)
How
epitaxial gro
Fig.2.5. Calculated 2DEG density in AlN/GaN heterostructureas a function of AlN thickness. More than 5×1013cm-2 chargecan be obtained due to the huge polarization effect in AlN.
ever, such a high 2DEG density is very difficult to achieve in reality. The
wth of AlN (even for high Al composition AlGaN) is problematic due to
24
the large tensile strain in the AlN layer. The lattice mismatch between GaN and AlN
is about 2.4%. As a result, the AlN layer cracks easily. The number of reports on the
transport properties of high Al composition AlxGa1-xN/GaN (x>0.5) is very limited.
Binari et al [10] reported on the MOCVD growth and characterization of the
AlN/GaN structure, but the sheet charge density was rather low, only 4.8×1012cm-2.
The value of the low temperature mobility reported was quite low as well: 720cm2/V
s. Alekseev et al. [11] made an attempt to use a low-pressure MOCVD technique to
grow an AlN/GaN heterojunction field-effect transistor. A higher 2DEG density of
2×1013cm-2 was obtained but the room temperature mobility was only 320cm2/V s
with an 11nm AlN barrier. The lower-than-expected charge density and poor mobility
could be attributed to the low quality of the AlN.
Recently, the growth of an AlN/GaN structure by MBE was demonstrated at
UCSB by Smorchkova et al [12]. A high quality AlN epitaxial layer was successfully
grown and promising transport properties were achieved. In this section, the work in
[12] by Dr. Ioulia Smorchkova will be reviewed and some new explanations will also
be discussed.
The AlN/GaN structures were grown by plasma-assisted MBE on top of either
semi-insulating or unintentionally doped GaN templates prepared by MOCVD on
(0001) sapphire substrates. The MBE grown films consisted of 0.25-0.3µm-thick GaN
layers followed by extremely thin AlN layers with different thicknesses. The growths
were performed at 730-740°C under Ga-stable growth conditions to obtain very
25
smooth surface morphologies that is essential for achieving good quality interfaces in
heterostructures.
The transport properties were studied by Hall measurements performed using
the Van der Pauw geometry, with indium dots as ohmic contacts. Fig.2.6 shows the
2DEG sheet charge density and electron mobility at 300K and 77K as a function of
AlN barrier thickness. The similar values at 300K and 77K clearly demonstrated that
the charge is due to the presence of a 2DEG in the channel. As expected, the charge
increased when the AlN barrier became thicker, from 1.5×1013cm-2
FasmD
with 2.4nm-th
large polariza
structure with
composition o
5×1013cm-2 co
2.0 2.5 3.0 3.5 4.0 4.5 5.0
1.5
2.0
2.5
3.0
3.5
4.0
Mob
ility
(cm
2 /V s
)
2DEG
den
sity
(1013
cm-2)
AlN thickness (nm)
ns 300K ns 77K µ 300K µ 77K
0
1000
2000
3000
4000
5000
ig.2.6. 2DEG density and electron mobility at 300K and 77K a function of Al barrier thickness. Charge increases andobility drops when AlN thickness increases [12] (courtesy ofr. Smorchkova).
ick AlN to 3.6×1013cm-2 when the AlN was 4.9nm thick. Due to the
tion-induced electric field in the barrier, the 2DEG density in the
4.9nm-thick AlN was much higher than standard HEMTs with an Al
f 0.2-0.4. Simulations showed that a 2DEG density saturation value of
uld be achieved when the AlN was thicker than 10nm. However,
26
because of the very large tensile strain in the AlN layer, it was easy for the AlN to
crack when the thickness exceeded 5nm. The mobility dropped when the charge
increased. This was partially due to the fact that the centroid of the 2DEG was pushed
closer to the interface, and the interface roughness could then significantly affect the
mobility. Another important reason for the mobility drop, especially when the AlN
was 50nm thick or above, was that the AlN layer started to crack when the AlN
thickness increased. Fig.2.7 shows AFM images of the AlN surface for samples of
different AlN thickness. The surface of the sample with a 3.7nm AlN barrier [fig.
2.7(a)] displayed a clear step structure with occasional spiral features associated with
threading dislocations having screw components. When the thickness of the AlN
barrier increased to 4.9nm, shown in fig.2.7(b), in addition to the step structure, lines
corresponding to cracks in the AlN appeared, pointing to the beginning of the tensile
relaxation process in the barrier. When the thickness increased further to 10nm, the
cracking density increased, as in fig.2.7(c). The defect formation process significantly
degraded the lateral transport characteristics.
27
Fig.2.7. Surface morphology of AlN films with thickness of (a)3.7nm; (b) 4.9nm; (c) 10nm [12] (courtesy of Dr.Smorchkova).
28
The relatively high dislocation density of the semi-insulating GaN template
(~1010cm-2) used above limited the improvement of the mobility. To reduce this
disadvantage, low dislocation density (5×108 – 109 cm-2) semi-insulating MOCVD
GaN templates were used. Fig.2.8 displays the temperature dependence of the
electron sheet density and electron mobility in the AlN/GaN structure with a 3.5nm
AlN barrier. A 2DEG density of 2.2×1013cm-2 was obtained, higher than the density
of 1.5×1013cm-2 in a typical Al0.3Ga0.7N/GaN HEMT structure. More importantly, a
relatively high room temperature electron mobility of 1600 cm2/V s was achieved.
Considering the high 2DEG density, this mobility value clearly displayed the
advantage of the AlN barrier. As a comparison, the typical mobility value of
Al0.3Ga0.7N/GaN with 1.5×1013cm-2 charge was only 1400-1500 cm2/V s. Due to the
both higher sheet charge density and mobility, the sheet resistance at room
temperature was only 180Ω/ٱ, much lower than the 250-350Ω/ٱ in normal
AlGaN/GaN HEMTs. Further measurement at low temperature showed that a
mobility of 13380 could be reached with 1.6×1013cm-2 charge in an AlN/GaN
structure, which was almost 3 times higher than the low temperature mobility in
AlGaN/GaN structures with a similar charge density. This improvement was due to
the removal of the alloy disorder scattering by the introduction of the binary AlN
barrier. As we know, alloy disorder scattering is one of the two major limiting factors
of the mobility at low temperature. Moreover, because of the high sheet charge
density, alloy disorder scattering plays an important role even at room temperature.
Thus, both low temperature and room temperature mobilities were improved.
29
Fig.2squathe Alow-d(dislo
0 10 20 30 40 501
10
µ (c
m2 /V
s)
n s (1013
cm-2)
1000/T (K-1)
1000
10000
.8. Temperature dependence of the Hall mobility (openres) and the sheet carrier concentration (dark squares) inlN/GaN structure with a 3.5nm AlN barrier grown on aislocation-density semi-insulating GaN templatecation density ~109cm-2) [12] (courtesy of Dr. Smorchkova).
30
2.4 GaN/AlN/GaN heterostructures
Although the AlN/GaN HEMT structure has very attractive carrier transport
properties, it has an obvious drawback – the barrier is too thin. Due to the cracking of
the AlN during growth, the thickness of the AlN barrier layer is limited to below
5nm. In this case, the distance between the surface and channel is so small that the
2DEG properties are sensitive to any processing applied to the surface. For example,
the plasma used during the RIE etching may penetrate the barrier layer and degrade
the mobility and carrier concentration. A barrier layer that is too thin may also lead to
high gate leakage due to the high tunneling probability
To increase the distance between the channel and surface while retaining the
advantages of the AlN/GaN heterostructure, a variation was proposed and
investigated: the GaN/AlN/GaN heterostructure [13]. Sample growth and
characterization were performed by Dr. Ioulia Smorchkova.
2.8
Fig.2.10. 2DEG density as a function of GaN cap thickness. 2DEG stopped decreasing when GaN cap is thicker than 20nm. The solid line was the calculated charge if only conduction band was considered [13] (courtesy of Dr. Smorchkova).
0 20 40 60 80 100 1200.0
0.4
0.8
1.2
1.6
2.0
2.4 T=20K
n s (10
13/c
m2 )
Thickness of GaN Cap (nm)
GaN / 3.5nm AlN / GaN
UID GaN
Substrate
UID GaNvarying from 0-100nm
3.5 nm AlN
UID GaN
Substrate
UID GaNvarying from 0-100nm
UID GaNvarying from 0-100nm
3.5 nm AlN
Fig.2.9. Epitaxial structure ofGaN/AlN/GaN heterostructure.
31
Fig.2.9 shows a schematic of the GaN/AlN/GaN heterostructure. To maximize
the 2DEG density while retaining the quality of the AlN layer, the thickness of the
AlN was fixed at 3.5nm. The GaN cap layer thickness was varied from 0 to 100nm.
The samples were still grown by RF-assisted MBE. The 2DEG charge density as a
function of GaN cap thickness obtained by Hall measurement is displayed in fig.2.10.
It shows a very interesting dependence. At the beginning, sheet charge density
dropped very quickly from 2.5×1013cm-2 to about 1×1013cm-2 when the thickness of
GaN cap increased from 0 to less than 20nm. Then the charge density stopped
decreasing and saturated around 1×1013cm-2 as the GaN cap layer became thicker.
Because this result was similar to the phenomenon observed in the
GaN/AlGaN/GaN heterostructure which is discussed in detail in chapter 4, just a brief
explanation is presented here.
Fig.2.11 shows the band diagrams of the GaN/AlN/GaN with different GaN
cap thicknesses. When the GaN cap is initially grown on top of the AlN/GaN
structure, the 2DEG density starts to decrease (fig.2.11(a)). The electric field in the
GaN cap, which is determined solely by the negative charge in the 2DEG, forces the
energy band at GaN/AlN interface to rise and thereby reduce the 2DEG density. This
trend continues until the valence band at the GaN/AlN interface contacts the Fermi
level, as shown in Fig.2.11(b). The contacting results in the accumulation of positive
charges at the top GaN/AlN interface. Now the electric field in the band cap is
determined by the difference of the negative charge in the 2DEG and the positive
charge at the top GaN/AlN interface. As a consequence, when the GaN cap becomes
32
thicker, the band diagram of AlN/GaN structure basically will not change (thereby,
2DEG density remains constant). This is because a small change of the valence band
can lead to a large enough change of the positive charges at GaN/AlN interface to
compensate the change of electric field in the GaN cap. This situation is shown in
fig.2.11(c). Considering this effect, the experiment agreed with the simulation very
well, as shown in fig.2.12.
0 10 20 30 40
-4
-2
0
2
4
6
AlN
GaN
GaN
Ene
rgy
(eV)
Thickness (nm)
0 10 20
-6
-4
-2
0
2
4AlN
GaN
GaN
Ene
rgy
(eV)
Thickness (nm)
(a (b)
Fig.2.11. Bdifferent Gband conta
)
0 10 20 30 40 50 60
-4
-2
0
2
4
6
Positivecharges
GaN
GaN
AlN
Ene
rgy
(eV)
Thickness (nm)
(c)
and diagrams of the GaN/AlN/GaN heterostructures withaN cap thickness (a) 2nm; (b) 20nm; (c) 40nm. Valence cts Fermi level when cap is thick.
33
FigGa(ex
0 20 40 60 80 100 120 140 160
0.8
1.2
1.6
2.0
2.4
2.8
n s (10
13/c
m2 )
Thickness of GaN Cap (nm)
GaN / 3.5nm AlN / GaN Simulation Experiment
.2.12. Simulated and experimental 2DEG density ofN/AlN/GaN structures with different GaN cap thickness.perimental data courtesy of Dr. Smorchkova)
34
2.5 AlGaN/AlN/GaN HEMTs
AlN was shown to be effective in removing the alloy disorder scattering,
therefore improving the 2DEG mobility in heterostructures. It was therefore useful to
incorporate it into the conventional AlGaN/GaN HEMT, i.e. by inserting an AlN
layer between the AlGaN and GaN to form a new heterostructure AlGaN/AlN/GaN
HEMT. This concept without including polarization effects was first proposed by Hsu
and Walukiewicz [14]. Smorchkova et al. experimentally demonstrated the structure
including effects of the polarization in 2001 [13]. Fig.2.13 displays the epitaxial
structure of a typical AlGaN/AlN/GaN structure grown by MBE. The inserted AlN
layer was 1nm thick. Our initial studies focused on using AlN layers to improve
mobilities, not carrier concentration, which is why such a thin AlN was implemented.
Moreover, the quality of thin AlN layer could be controlled better than the thick ones,
especially for MOCVD growth.
1 nm AlN
UID GaN
Substrate
25 nm UID Al0.33Ga0.67N
1 nm AlN
UID GaNUID GaN
Substrate
25 nm UID Al0.33Ga0.67N
Fig.2.13. Typical epitaxial structure ofAlGaN/AlN/GaN heterostructure. A thin AlN layer(~1nm) is inserted to remove alloy disorder scattering,thereby improving mobility.
35
Compared to the thick AlN discussed previously, the role of thin AlN is
different. When AlN is thicker than a critical thickness (for the 2DEG formation), the
AlN layer itself is the major contributor to the formation of 2DEG and the addition of
the GaN (or AlGaN) cap on top of it just decreases the 2DEG density. However,
when the AlN is thin (~1nm in this case) and below the critical thickness, it cannot
form the 2DEG directly; instead the AlGaN layer on top of it is the major contributor.
The role of the thin AlN layer to the properties of 2DEG can be describe as a larger
effective ∆Ec (∆Ec,eff) than the ∆Ec in a standard AlGaN/GaN HEMT, which is
discussed in detail in the following.
Simulation was performed to stu
when a thin (~1nm) AlN layer w
Fig.2.14. Band diagrams of the hHEMT; (b) conventional AlGaNAlN layer results in a higher effe
(a)
0 10 20 30 40 50-1
0
1
2
3
Thickness (nm)
Ener
gy (e
V)
Thin AlN
Effective ∆
0 10 20 30 40 50-1
0
1
2
3
Thickness (nm)
Ener
gy (e
V)
Thin AlN
Effective ∆
----
++++
----
++++
GaN AlGaN
dy the effects on the charge and mobility
as inserted. The band diagram of an
eterostructures. (a) AlGaN/AlN/GaN /GaN HEMT. The insertion of thin
ctive ∆EC.
(b)
0 10 20 30 40 50
0
1
2
3
Ener
gy (e
V)
0
1
2
3
Ener
gy (e
V)
EC
EC
Thickness (nm)
∆EC
AlGaN GaN
0 10 20 30 40 50
Thickness (nm)
∆EC
AlGaN GaN
36
AlGaN/AlN/GaN heterostructure simulated by 1D Poisson Solver is shown in
fig.2.14(a). As a comparison, the band diagram of the standard AlGaN/GaN HEMT is
also displayed in fig.2.14 (b). A very obvious change observed after the insertion of
the thin AlN layer is that the energy band has a sharp peak. This is due to both the
wide band gap of the AlN and the very strong polarization effect in AlN. If the
conduction band offset between AlGaN and GaN at the both sides of AlN is defined
as the effective ∆Ec (∆Ec,eff), this value is larger than the ∆Ec in the standard
AlGaN/GaN HEMT. The conduction band discontinuity ∆Ec in a typical
Al0.3Ga0.7N/GaN is about 0.6eV. After the insertion of a thin AlN layer, the potential
across the AlN layer has to be included:
2
, ,0
( )AlN sc eff c AlGaN AlN
q nE E tσε ε
⋅ −∆ = ∆ + ⋅
⋅ (2.4)
where ∆Ec,AlGaN is the conduction band discontinuity of AlGaN, σAlN is the
net polarization charge density of the AlN (subtracting the polarization charge density
of the GaN), tAlN is the thickness of the AlN layer, and ns is the 2DEG density.
Although the AlN is only 1nm thick, the potential drop across it is almost 0.9V due to
the very strong polarization effect, leading to a ∆Ec,eff almost 1.4eV. The ∆Ec,eff is
double the ∆Ec of a standard HEMT. The larger ∆Ec,eff is good for both mobility and
carrier concentration, as will be discussed in the following.
There are two factors that improve the mobility of 2DEG in the structure. One
is the reduction of the alloy disorder scattering due to the binary nature of the AlN
interfacial layer. Secondly, the introduction of AlN also reduces the penetration of the
37
electron wavefunction into the AlGaN barrier due to the larger ∆Ec,eff, which acts as a
higher potential barrier for electrons, compared to the standard structures. Fig.2.15
shows the probability distribution of an electron in both AlGaN/AlN/GaN and
AlGaN/GaN structures. The penetration into the AlGaN layer in the new structure is
largely reduced, as compared to the standard structure. The integrated probability of
an electron residing in the AlGaN barrier in the conventional Al0.33Ga0.67N/GaN is
about 7.3%. A ter the insertion of AlN, this value is almost 0.
Fig.2.1and Areduceshows
The la
band slope in
change in the
can be obtaine
f
24 26 28 30 32 34 36
0.00
0.01
0.02
0.03
0.04
0.05 AlGaN/AlN/GaN AlGaN/GaN
Ener
gy (e
V)
Prob
abili
ty
Distance (nm)
0
1
2
3
AlN/GaNinterface
5. Probability distribution of an electron in AlGaN/AlN/GaNlGaN/GaN HEMTs. The insertion of the thin AlN layer greatly s the wavefunction penetration into the AlGaN. The gray line
the conduction band of the AlGaN/AlN/GaN structure.
rger ∆Ec,eff also affects the 2DEG density. The change of the energy
AlGaN cap layer in fig.2.13, i.e. the electric field, already implies a
carrier density. The 2DEG density in the AlGaN/AlN/GaN structure
d by:
38
,0
0
( )C AlGaNAlGaN AlGaN AlN AlN B
sAlGaN AlN
Et t
q qnt t d
εεσ σ φ∆
⋅ + ⋅ − −=
+ + (2.5)
'0 0,2
0
AlGaN AlGaN B c eff
AlGaN AlN
t Eq q
t t d
εε εεσ φ− + ∆=
+ + (2.6)
where
2', ,
0c eff C AlGaN AlN AlN
qE E tσεε
∆ = ∆ + (2.7)
Compared to the ∆Ec,eff defined in (2.4), the ∆E’c,eff defined in (2.7) is slightly
different. The effect of the negative 2DEG charge is not taken into account in
equation (2.7) because it is used to calculate the 2DEG density. But these two terms
directly relate with each other. Larger ∆Ec,eff leads to a larger ∆E’c,eff and vice versa.
Recall that the 2DEG density of a standard AlGaN/GaN HEMT can be written
as
0 0,2
0
AlGaN AlGaN B C AlGaN
sAlGaN
t Eq qn
t d
εε εεσ φ⋅ − + ∆=
+ (2.8)
Equation (2.8) is very similar to equation (2.6), except ∆E’c,eff is used in (2.6),
instead of ∆Ec,AlGaN (tAlN is very thin and can be ignored). Therefore, the behavior of
2DEG in the new structure can be attributed to the larger ∆E’c,eff (or ∆Ec,eff) which is
caused by the insertion of the thin AlN layer.
39
Because the AlN layer is very thin, only ~1nm, the effect of ∆E’c,eff to the
2DEG density is still limited, compared with the term σAlGaNtAlGaN when the AlGaN
is thick . For the structure in fig.2.13, the charge increase caused by the insertion of
the AlN layer is only about 10-15%. This increase also can be explained by
considering the band diagram in fig.2.14. The larger ∆Ec,eff results in a decrease of
the electric field in the AlGaN layer and the decrease can only be implemented when
the 2DEG densit increases.
0.1 0.2 0.3 0.40.0
0.5
1.0
1.5
2.0
2.5
Mob
ility
(104 cm
2 /V s
)
Al composition
AlGaN/GaN AlGaN/AlN/GaN
Th
Fig.2.16 d
and AlGa
was 1nm
the AlGa
3000-400
low temp
Fig.2.16. Mobilities at T=17K of the standard AlGaN/GaN HEMTsand novel AlGaN/AlN/GaN HEMTs. The novel structure showshigher mobility when the carrier concentrations are similar [13](courtesy of Dr. Smorchkova).
e theo
isplaye
N/GaN
thick. W
N/AlN/
0 cm2/V
erature
y
retical speculation was supported by the experimental results.
d the low temperature (T=17K) mobility for both AlGaN/AlN/GaN
HEMTs with different Al mole fraction. In all structures, the AlN
hen Al compositions were the same (~0.3), the electron mobility in
GaN structure was more than 6000 cm2/V s, much higher than the
s in a standard HEMT. When the Al composition reached 0.45, the
mobility dropped to 4000cm2/V s. However, considering the 2DEG
40
density was as high as 2.5×1013cm-2, this number was very impressive. At room
temperature, mobility was improved as well. An Al0.37Ga0.63N/AlN/GaN structure
showed a mobility of 1500cm2/V s with a carrier density of 2.15×1013cm-2 at room
temperature, which achieved the low sheet resistance of 194Ω/ٱ. As a comparison,
the standard AlGaN/GaN HEMT structures had 250-350Ω/ٱ sheet resistance.
Fig.2.17 displays both experimental and simulated curves of the 2DEG
density as a function of Al mole fraction. It is noted that the charge density was still a
strong function of Al composition, similar to standard AlGaN/GaN HEMTs. When
the Al composition was varied from 0.27 to 0.45, the 2DEG density increased from
1.45×1013cm-2 to 2.5×1013cm-2. In fact, from equation (2.4), one could find that the
dependence was almost identical to the standard structure. Simulations performed by
1D Poisson Solver greed with the experimental data very well.
a
0.25 0.30 0.35 0.40 0.45
1.5
2.0
2.5
3.0
2DEG
Den
sity
(1013
cm-2)
Al mole fraction x
Experiment Simulation
Fig.2.17. Simulated and experimental data of 2DEG densities ofAlGaN/AlN/GaN structure as a function of Al mole fraction.(experimental data courtesy of Dr. Smorchkova)
41
Another consequence of a large ∆Ec,eff is that the dependence of the 2DEG
density on the AlGaN cap thickness is different from that in standard HEMTs.
Usually in AlGaN/GaN HEMTs with reasonable Al composition (e.g. 0.2-0.4), a
2DEG appeared when the AlGaN thickness reached 4-5nm, and then increased
steadily from 2-3×1012 to 1-2×1013cm-2 as the AlGaN became thicker. After the
AlGaN exceeded 20-30nm, the 2DEG density saturated, as shown in fig.2.18.
Therefore, charge density was a strong function of AlGaN barrier thickness when it
was thinner than 20nm. However, in AlGaN/AlN/GaN structure, ns was a weak
function of AlGaN thickness. As shown in fig.2.18, the carrier density remained
relatively constant as the AlGaN became thicker. In this Al0.37Ga0.63N/AlN/GaN
structure for which the AlN was 1nm thick, when the AlGaN was only 6nm, the
2DEG density in the channel already reached 1.9×1013cm-2. When it was thicker than
20nm, the carrier density only increased slightly to 2-2.1×1013cm-2. This was due to
the larger ∆E’c,eff. In equation (2.6), when tAlGaN is small (i.e. AlGaN is thin), the
∆E’c,eff can play an important role. It provides a relatively high initial value for the
carrier concentration. Therefore, the dependence of 2DEG density on AlGaN
thickness in the new structure was different from that of standard structures.
42
The grow
more difficult to
with different A
Instead of follo
the AlN was th
Similarly, mobi
below 700cm2/v
resulted from th
thick AlN in
disappearance o
0 5 10 15 200.0
0.5
1.0
1.5
2.0
2.5Al0.37Ga0.63N/AlN/GaN HEMT
conventioal Al0.37Ga0.63N/GaN HEMT
2DEG
Den
sity
(1013
cm-2)
Thickness of AlGaN (nm)
Experiment Simulation
Fig.2.18. Simulated and experimental data of 2DEG densities ofAlGaN/AlN/GaN structure as a function of AlGaN thickness. Thedata of a standard AlGaN/GaN HEMT is also shown for comparison.Different trends are observed when AlGaN is thin. (experimentaldata courtesy of Dr. Smorchkova)
th of thick AlN presents another difficulty. Compared to MBE, it is
grow thick AlN by MOCVD. The results of a series of experiments
lN thickness are shown in fig.2.19 as well as simulated numbers.
wing the theory-predicted curve, the charge became saturated when
icker than 1nm and then started dropping when it exceeded 2nm.
lity reached a maximum when the AlN was 0.5nm. Then it dropped
s when the AlN was 2nm thick. The severe degradation after 1nm
e relaxation of the AlN. Practically, it was more difficult to grow
MOCVD than in MBE. After relaxation and cracking, both
f piezoelectric charge and formation of defects contributed to the poor
43
performance. Therefore, for future MOCVD growth, the AlN thickness was chosen
between 0.5 to 1nm.
0.0 0.5 1.0 1.5 2.0 2.5 3.01.0
1.2
1.4
1.6
1.8
Thickness of AlN (nm)
2DEG
Den
sity
(1013
cm-2)
Charge(Simulation) Charge(Experiment) Mobility(Experiment)
600
800
1000
1200
1400
1600
Mob
ility
(cm
2 V-1 s
-1)
optimum thickness
0.0 0.5 1.0 1.5 2.0 2.5 3.01.0
1.2
1.4
1.6
1.8
Thickness of AlN (nm)
2DEG
Den
sity
(1013
cm-2)
Charge(Simulation) Charge(Experiment) Mobility(Experiment)
600
800
1000
1200
1400
1600
Mob
ility
(cm
2 V-1 s
-1)
optimum thickness
Fig.2.19. Simulated and experimental data of 2DEG densities andmobility of AlGaN/AlN/GaN structure grown by MOCVD as afunction of AlGaN thickness. Practically, the transport properties of2DEG degrades when AlGaN is thicker than 1nm.
MOCVD was used to grow AlGaN/AlN/GaN for HEMT devices. A series of
HEMT samples were grown: a) conventional structure: 25 nm Al0.3Ga0.7N /GaN; b)
novel structure with unintentionally-doped (UID) cap AlGaN: UID 25 nm Al0.3Ga0.7N
/1 nm AlN/GaN; c) novel structure with Si-doped cap AlGaN: 20 nm Si-doped
Al0.3Ga0.7N /5 nm UID Al0.3Ga0.7N /1 nm AlN/GaN. Doping density is approximately
1×1018 cm-3. All of the samples were grown on SiC substrates for good thermal
conductivity to reduce self-heating.
As a control sample, the standard AlGaN/GaN sample showed a carrier
density of 1.1×1013cm-2 and a mobility of 1200cm2/V s. The sample B with a thin
AlN layer and UID AlGaN had a higher room temperature mobility of 1520 cm2/v s
44
and a slightly higher 2DEG density of 1.22×1013cm-2. The result was completely
consistent with the theoretical analysis, which predicted an improvement of mobility
and a small increase of charge density. The sample C with Si-doped AlGaN and thin
AlN layer demonstrated the best performance, increasing charge to 1.48×1013cm-2
while maintaining a mobility higher than 1500cm2/V s. The additional charge can be
attributed to the Si dopants which ionized and donated electrons to the channel. The
sheet density of the Si doping was about 2x1012cm-2, very close to the increase of the
2DEG density.
Devices were then fabricated. The processing was the same as that of standard
HEMTs. Ti/Al/Ni/Au (20nm / 220nm / 55nm / 45nm) ohmic contacts were
evaporated by electron-beam (EBeam) evaporation and annealed at 870ºC for 30s in
N2. Mesa isolation was accomplished with Cl2 reactive ion etching. Ni/Au (30nm /
300nm) was evaporated by Ebeam for gate metallization. The final processing step
was a sputtered 100nm Si3N4 passivation layer deposited by sputtering, which has
been shown to eliminate DC to RF dispersion.
Typical DC output current-voltage characteristics of a 0.15-mm-wide Si-
doped AlGaN/AlN/GaN HEMT with gate length LG = 0.7µm and gate-drain spacing
LGD = 2µm are shown in fig. 2.20. The maximum saturation current, Imax, at VGS = 2V
was 1A/mm and the pinch-off voltage is –3.5 V. The peak value of the extrinsic
transconductance, gm, was approximately 200mS/mm near VGS = -1.5V. The
Schottky gate turn-on voltage was approximately 1.5V and gate-drain breakdown
45
voltage was typically 70 ~ 80V. The ohmic contact resistance ranged from 0.5-0.7Ω-
mm.
Fig.2C). NCurr
Small-s
frequencies (ft
device at VGS =
An ATN
measurements a
of both HEMTs
performance. It
and tuned for m
The associated
device without
0 2 4 6 8 10 12 14 16
0
200
400
600
800
1000
gm = 200 mS/mm
∆VG = 1 VVG = 2 V
I D (m
A/m
m)
VDS (V)
.20. DC characteristics of the AlGaN/AlN/GaN HEMT(Sampleo dispersion was observed up to gate pulse width of 200ns.
ent density of 1A/mm was obtained at gate bias of +2V.
ignal RF measurements yielded current-gain and power-gain cutoff
and fmax, respectively) of 28 GHz and 56 GHz for 0.7-µm gate-length
-2.5 V and VDS = 20 V.
load-pull system was used for the large signal continuous-wave(CW)
t 8 GHz. Fig.2.21 shows the un-cooled on-wafer measurement results
with and without Si doping. The device with Si doping showed better
was biased in class-AB mode at VDS = 45 V and IDS = 160 mA/mm
aximum power. Output power density of 8.4 W/mm was obtained.
power gain and PAE were about 7.5 dB and 28 %, respectively. The
Si doping showed an output power density of 8.1W/mm at drain bias
46
of 50V. No stability problems were observed, compared with the conventional
AlGaN/GaN HEMT [15].
PAE
(%)
0 5 10 15 20 25 300
5
10
15
20
25
30
358.47 W/mm
PAE
(%)
Pout Gain PAE
Pout
(dBm
), G
ain
(dB)
Pin (dBm)
0
51015202530
3540
0
5 10 15 20 25 300
5
10
15
20
25
30
358.1 W/mm Pout
Gain PAE
Pout
(dBm
), G
ain
(dB)
Pin (dBm)
0510
1520253035
40
(a) (b)
Fig.2.21. Power performance at 8GHz of passivated AlGaN/AlN/GaNHEMTs on SiC substrates. (a) UID AlGaN; (b) Si-doped AlGaN.Output power density of 8.5W/mm was achieved.
47
2.6 Summary
An approach to using AlN to improve the mobility in GaN-based HEMT was
discussed in this chapter. The scattering mechanisms in AlGaN/GaN HEMTs were
reviewed. Alloy disorder scattering could be removed by the adoption of the binary
material, AlN. The work of the AlN/GaN heterostructure was reviewed, showing that
high charge concentration and mobility could be achieved at the same time. A sheet
resistance of 180Ω/ٱ was obtained.
A novel GaN/AlN/GaN structure was discussed as well. The introduction of
the GaN cap layer could reduce the processing sensitivity of the AlN/GaN device due
to its thin AlN layer. A decrease of the carrier concentration and then remaining at a
constant value were observed. It was explained successfully by considering the
contact of the valence band to the Fermi level at GaN/AlN interface, which resulted
in an accumulation of the positive charges.
The thin AlN was also incorporated into the standard HEMT to form an
AlGaN/AlN/GaN heterostructure. The introduction of AlN layer improved the
mobility by removing the alloy disorder scattering in the AlN and reducing the
electron wavefunction penetration into the AlGaN barrier layer. The larger effective
∆Ec also increased the 2DEG density. It was verified by the experiment that the
AlGaN/AlN/GaN HEMTs showed higher mobility and carrier concentration,
compared to standard AlGaN/GaN HEMTs. An output power density of 8.5W/mm
was achieved at 8GHz for a HEMT on a SiC substrate.
48
2.7 References [1] William Liu, Fundamentals of III-V Devices HBTs, MESFETs, and
HFETs/HEMTs, John Wiley & Sons, Inc, 1999 [2] Debdeep Jena, Ph.D. Dissertation, “Polarization induced electron populations in
III-V nitride semiconductors Transport, growth, and device applications”, University of California, Santa Barbara, 2003.
[3] W. Knap, S. Contreras, H. Alause, C. Skierbiszewski, J. Camassel, M. Dyakonov,
J. L. Robert, J. Yang, Q. Chen, M. A. Khan, M. L. Sadowski, S. Huant, F. H. Yang, M. Goian, J. Leotin, and M. S. Shur, “Cyclotron resonance and quantum Hall effect studies of the two-dimensional electron gas confined at the GaN/AlGaN interface,” Appl. Phys. Lett., vol.70, No.16, pp.2123-2125, Apr. 1997.
[4] D. K. Ferry and S. M. Goodnick, Transport in Nanostructures. Cambridge, UK:
Cambridge University Press, 1st ed., 1999. [5] J. P. Ibbetson, P. T. Fini, K. D. Ness, S. P. DenBaars, J. S. Speck, and U. K.
Mishra, “Polarization effects, surface states, and the source of electrons in AlGaN/GaN heterostructure field effect transistors,”Appl. Phys. Lett., vol.77, No.2, pp.250-252, July 2000.
[6] D. Jena, A. C. Gossard, and U. K. Mishra, “Dislocation scattering in a two-
dimensional electron gas”, Applied Physics Letters, vol. 76, no. 13, pp.1707-1709, Mar. 2000.
[7] D. Jena, A. C. Gossard, and U. K. Mishra, “Dipole scattering in polarization-
induced III-V nitride two-dimensional electron gases”, Journal of Applied Physics, vol. 88, no. 8, pp. 4734-4738, Oct. 2000.
[8] G. D. Bastard, Wave-Mechanics applied to Semiconductor Heterostructures. Les
Ulis Cedex, France: Les Editions de Physique, 1st ed. [9] Y. Zhang, I. P. Smorchkova, C. R. Elsass, S. Keller, J. P. Ibbetson, S. DenBaars,
U. K. Mishra, and J. Singh, “Charge control and mobility in AlGaN/GaN transistors: Experimental and theoretical studies,”J. Appl. Phys. Vol.87, pp.7981-7987, June 2000.
[10] S. C. Binari, K. Doverspike, G. Kelner, H. B. Dietrich, and A. E. Wick-enden,
“GaN FETs for microwave and high-temperature applications,” Solid-State Electron., Vol.41, No.2, pp.177-180 , Feb. 1997.
49
[11] E. Alekseev, A. Eisenbach, and D. Pavlidis, “Low interface state density
AlN/GaN MISFETs,” Electron. Lett. Vol. 35, No. 24, pp.2145-2146, Nov. 1999.
[12] I. P. Smorchkova, S. Keller, S. Heikman, B. Heying, P. Fini, J. S. Speck, and U.
K. Mishra, “Two-dimensional electron-gas AlN/GaN heterostructures with extremely thin AlN barriers,” Appl. Phys. Lett., vol. 77, No. 24, pp. 3998-4000, Dec. 2000.
[13] I. P. Smorchkova, L. Chen, T. Mates, L. Shen, S. Heikman, B. Moran, S. Keller,
S. P. DenBaars, J. S. Speck, and U. K. Mishra, “AlN/GaN and (Al,Ga)N/AlN/GaN two-dimensional electron gas structures grown by plasma-assisted molecular-beam epitaxy,” J. Appl. Phys., Vol. 90, No. 10, pp.5196-5201, Nov. 2001.
[14] L. Hsu and W. Walukiewicz, “Effect of polarization fields on transport
properties in AlGaN/GaN heterostructures,” J. Appl. Phys., vol. 89, no. 3, pp. 1783-1789, Feb. 2001.
[15] L. Shen, S. Heikman, B. Moran, R. Coffie, N. Q. Zhang, D. Buttari, I. P.
Smorchkova, S. Keller, S. P. DenBaars, and U. K. Mishra, “AlGaN/AlN/GaN high-power microwave HEMT,” IEEE Electron. Dev.Lett., vol. 22, no. 10, pp. 457-459, Oct. 2001.
50
Chapter 3
DC-to-RF Dispersion
3.1 Introduction
Although GaN has long been known to have the potential to generate a much
higher output power density than other semiconductor materials (for example, GaAs),
it has taken many years for GaN-based HEMTs to realize this potential . In 1996, the
first microwave power data of 1.1W/mm was reported by Wu et al. [1]. Later, when
SiN passivation became widely used, the average power performance was improved
greatly. Output power densities of 10-12W/mm were reported [2]. However, even this
number was still far below the theoretical prediction. According to the well-know
linear and saturated output power equations [3]:
max, max1 I (8lin BD kneeP V= ⋅ − )V (3.1)
max,2
16sat linP P
π= (3.2)
a typical AlGaN/GaN HEMT with maximum current Imax of 1.2A/mm, knee voltage
Vknee of 5V and breakdown voltage VBD of 100V should produce maximum linear
power of 14W/mm while the saturation power should reach 22W/mm.
51
This discrepancy between the predicted output power from static I-V curves
and load pull measured output power has been referred as “DC-to-RF dispersion”.
Fig.3.1 shows typical dispersion behavior. Both DC and pulsed (gate-lag) I-V
characteristics are displayed. A lower current resulting from pulsed mode was
observed than that of the DC characteristic. Moreover, the shorter the pulse width
was, the lower the current. The collapse of the current greatly reduced the current
swing, causing the output power at high frequencies to be much lower than that
predicted from DC I-V curves. Investigations suggested that the dispersion was
related to traps. Although traps can be located in the buffer [4], barrier layer [5] and
surface [6], the fact that SiN passivation improved power performance significantly
implied that surface traps played a dominant role. A virtual gate model based on the
surface traps w s proposed to explain the dispersion [7].
a
0 5 10 15
0
200
400
600
800
1000
1200
T = 300K
I d (m
A/m
m)
Vds (V)
DC 20us 200ns
Fig.3.1. DC and pulsed I-V characteristics of an unpassivatedAlGaN/GaN HEMT on SiC substrate. Obvious currentcollapse was observed in the pulsed mode.
52
An illustration of how surface-states can cause dispersion is shown in fig.3.2.
In region I, a negative voltage below pinch-off is applied to the gate. The channel
under the gate is depleted and the device is off. Due to the high electric field at the
drain edge of the gate, there is a small amount of gate leakage between the gate and
drain. This current charges up the surface states, forcing the surface potential
becoming negative. As a result, the channel under this part of drain
VG
IDS
VP
I II IIIVG
IDS
VP
I II III
G DS ---
G DS-----
-
----------- -------
---- --------- - ------
G DS ----------- ------
-
-
I
II
III
IDS
IDS
AlGaNGaN
Charged Surface traps
channel
G DS ---
G DS-----
-
----------- -------
---- --------- - ------
G DS ----------- ------
-
-
I
II
III
IDS
IDS
AlGaNGaN
Charged Surface traps
channel
Fig.3.2. Illustration of the virtual gate model. Dispersion iscaused by the slow response of the surface traps on the drainaccess region or the slow movement of the electrons throughthe surface,
access region next to the gate is depleted too. In region II, the gate bias changes
towards positive very quickly. The channel under the gate can follow the quick
53
change and turns on almost instantly. However, the situation under the drain access
region is different. Due to the slow response of the deep level traps or low mobility of
the electrons along the surface, the channel underneath the drain access region cannot
turn on immediately after the gate bias becomes more positive. Instead, this region is
still depleted or has very low carrier concentration immediately following the gate
bias change. This results in a highly resistive region and a substantial portion of the
drain bias drops across it. Therefore, the drain current remains low just after the
changing of the gate bias. As the electrons are released from the traps and move back
to the gate or drain gradually, the carrier concentration in the channel under the drain
access region increases correspondingly and the drain current increases as well. If the
pulse is long enough, the drain current reaches the steady-state value, as shown in
region III.
54
3.2 Dispersion at low temperature
Because the dispersion is related to deep level traps, the temperature-
dependent measurement is a useful tool to provide relevant information. The trap
capture and emission rate with the charge trapping and de-trapping are strong
functions of temperature. The surface conduction, either through band conduction or
hopping conduction, is affected by temperature as well. In this section, the dispersion
performance within the temperature ranging from 77K to 300K will be reported.
A cryogenic micro-manipulated probe system provided the controlled
temperature from 77K to 300K. Liquid nitrogen was used as the coolant, ensuring a
good cooling capacity. Gate-lag measurements were utilized to characterize the
dispersion. These measurements were performed using a waveform generator, a DC
power supply and an oscilloscope which obtained the current by measuring the
voltage drop across the resistor.
Three different kinds of devices were investigated: an unpassivated device, a
passivated device on a sapphire substrate and a passivated device on a SiC substrate.
Fig.3.3(a) and (b) show the DC and pulsed I-V characteristics of the
unpassivated device at 300K and 100K respectively. The unpassivated device showed
obvious current collapse even at 300K. The collapse of the pulsed current became
larger when temperature decreased. Since both the DC current increased at lower
temperature and the AC current decreased, the dispersion was worse at 100 K.
55
0 5 10 1
0
200
400
600
800
1000
1200
T = 300K
I d (m
A/m
m)
Vds (V)
DC 20us 200ns
5
(a)
Fig.3.3. DC and pulsed curHEMT at (a) 300K; (b) 10temperatures. The DC currencurrent decreased.
The performance of the passiva
in fig.3.4 (a) and (b). Because of the SiN
temperature. The higher current in the p
by the poor thermal conductivity of sa
temperature. There was an obvious knee
A similar phenomenon was ob
substrate, as shown in fig.3.5 (a) and (b
SiC, no severe self-heating occurred a
showed similar values. At T=100K, the
Dispersion again occurred at low tempe
0 5 10 15
0
200
400
600
800
1000
1200
1400
T = 100 K
I d (m
A/m
m)
Vds (V)
DC 20us 200ns
(b)
rents of an unpassivated AlGaN/GaN0K. Dispersion was observed at botht increased at T=100K while the pulsed
ted device on the sapphire substrate is shown
passivation, there was no dispersion at room
ulsed mode was due to the self-heating caused
pphire. However, dispersion appeared at low
voltage walkout at T=100K.
served for the passivated device on the SiC
). Because of the good thermal conductivity of
t room temperature. DC and pulsed currents
pulsed current was lower than the DC current.
rature.
56
0 5 10 15
0
200
400
600
800
1000
1200
T = 300KI d (
mA
/mm
)
Vds (V)
DC 20us 200ns
(a)
Fig.3.4. DC and pulsed currenon sapphire substrate at (a) 30at room temperature. Howtemperature.
0 5 10 1
0
200
400
600
800
1000
1200
T = 300K
I d (m
A/m
m)
Vds (V)
DC 20us 200ns
5
(a)
Fig.3.5. DC and pulsed currenon SiC substrate at (a) 300K;room temperature. However, d
The first conclusion was that dis
all three cases. Even the devices that d
showed dispersion at low temperature.
(b)
0 5 10 15
0
200
400
600
800
1000
1200
1400
T = 100 K
I d (m
A/m
m)
Vds (V)
DC 20us 200ns
ts of an passivated AlGaN/GaN HEMT0K; (b) 100K. There was no dispersionever, dispersion appeared at low
p
i
T
0 5 10 15
0
200
400
600
800
1000
1200
T = 100 K
I d (m
A/m
m)
Vds (V)
DC 20us 200ns
(b)
ts of an passivated AlGaN/GaN HEMT (b) 100K. There was no dispersion atispersion appeared at low temperature.
ersion became worse at low temperature in
d not have dispersion at room temperature
hose with dispersion at room temperature
57
had worse performance at low temperature. To obtain more information, DC and
200ns-pulse-width currents were measured as a function of temperature.
Fig.3.6 shows the change in the current of the device on the sapphire
substrate. The DC current increased by 38%, from 930mA/mm to 1300mA/mm, when
the temperature decreased from 300 K to 77 K. The improvement was due to the
higher effective electron velocity in the channel at lower temperature. Moreover, the
improved thermal conductivity of sapphire at lower temperature also contributed to
the reduction of the self-heating effect. However, the increase of the pulsed current
was much smaller, only about 9%, from 1130mA/mm to 1230mA/mm. The smaller
increase was partially due to the dispersion. Another reason was that the temperature
change in the channel was smaller than that in the DC case because self-heating was
much less in the 200 ns-pulse-width mode. The dispersion became worse at lower
temperature because the pulsed current increased less than DC current.
58
0 2 4 6 8
0
200
400
600
800
1000
1200
1400
77K 100K
150K200K
250K
300K
I d (mA/
mm
)
Vds (V)
DC VG=+1V
(a)
Fig.3.6. DC and pulsed currenton a sapphire substrate as a funpulse-width. When temperaturBut the increase of the DC curre
The performance of the passivat
fig.3.7. Although the DC currents increa
increase is only about 9%, from 1100m
small increase was similar to that of the
good thermal conductivity of SiC, the t
than that for the devices on the sapphi
currents, the maximum currents almos
decreased, while the on-resistance decr
temperature because the DC current i
constant.
0 2 4 6
0
200
400
600
800
1000
1200
1400
200K,150K,100K,77K250K
300K
I d (m
A/m
m)
Vds (V)
200ns pulse-width VG=+1V
(b)
s of a passivated AlGaN/GaN HEMTction of temperature. (a) DC (b) 200ns-e decreased, both currents increased.nt is larger than that of pulse current.
ed device on a SiC substrate is shown in
sed with the decreasing of temperature, the
A/mm to 1200mA/mm. The reason for this
pulsed current explained above. Due to the
emperature change in the channel was less
re substrate. As for the 200ns-pulse-width
t remained unchanged when temperature
eased. More dispersion occurred at lower
ncreased but the pulsed current remained
59
0 2 4 6 8
0
200
400
600
800
1000
1200 150K, 100K, 77K 200K250K
300K
I d (
mA/
mm
)
Vds (V)
DC VG=3V
(a) Fig.3.7. DC and pulsed currents
on SiC substrate as a function ofwidth. When temperature decreapulse-width current remained co
As for the unpassivated device,
devices at low temperature. However, th
temperature was lowered, as shown in
200K, the knee voltage walkout was very
0 2
0
200
400
600
800
1000
I d (m
A/m
m)
20µs pulse-wid
Fig.3.8. 20us-pulse-width currHEMT as a function of tedecreased, the current increasesignificantly when temperature
0 2 4 6 8
0
200
400
600
800
1000
1200 200K,150K,100K,77K
250K300K
I d (
mA/
mm
)
Vds (V)
200ns VG=3V
(b)
of an passivated AlGaN/GaN HEMT temperature. (a) DC (b) 200ns-pulse-sed, DC current increased but 200ns-
nstant.
the DC current increased as in the other
e pulsed current decreased greatly when the
fig.3.8. When the temperature was below
severe.
4 6 8 10 12
77K100K
150K
200K250K
300K
Vds (V)
th VG=+1V
ent of an unpassivated AlGaN/GaNmperature. When the temperatured a small amount, and then droppedwas below 150K.
60
Fig. 3.9 displayed the ratio between the pulsed and DC current at drain bias of
knee voltage as a function of temperature, which could be used to evaluate the
dispersion. A value less than 1 implied dispersion. Obviously, the dispersion became
worse at lowe temperature in all three cases.
Fig.3.drainthis ratempe
The co
also measured
the contact re
to 300K, whic
a tunneling ef
160K, it was l
resistance is m
temperature [8
r
50 100 150 200 250 300
0.2
0.4
0.6
0.8
1.0
1.2
1.4
VG=+1V VD=Knee Voltage
passivated on sapphire passivated on SiC unpassivated
I D(Pu
lse) /
I D(DC
)
Temperature (K)9. The ratio between the pulsed current and DC current at a bias of the knee voltage. When the temperature was decreased,tio dropped in all three of these cases. The lower value at lowerrature represented more dispersion.
ntact resistance and sheet resistance as a function of temperature were
using an MMR temperature controlled system. As shown in fig.3.10,
sistance did not change much within the temperature range from 160K
h was consistent with the fact that the ohmic contact was dominated by
fect. The sheet resistance decreased substantially at low temperature. At
ess than half of that at 300K. This was not surprising because the sheet
ainly determined by low-field mobility, which is a strong function of
].
61
160 200 240 280
150
200
250
300
350
R SH (O
hm/S
qr)
Temperature (K)
RSH RC
0.2
0.4
0.6
0.8
RC (
Ohm
-mm
)
Fig.3.10. The ratio between pulsed current and DC current at drainbias of knee voltage. When temperature decreased, it dropped in allthese three cases. The lower value at lower temperature representedmore dispersion.
From these measurements, we observed:
1) DC currents increased in all cases when the temperature decreased.
2) Pulsed currents showed different trends in different cases, i.e. it is sample-
dependent. In some cases, it increased at low temperature while in other cases it
remained constant, or even decreased.
3) Dispersion increased at low temperature in all cases whether the pulsed currents
increased or not.
The increase of the DC current is expected. DC current is proportional to the
product of the 2DEG density and electron drift velocity. Because the electron velocity
increases at low temperature while the 2DEG density remains constant, the DC
current increases correspondingly. The behavior of pulsed current can be explained
by two competing factors: electron concentration and velocity. The conductivity of
the channel underneath the drain access region is proportional to both the electron
62
velocity v, and the carrier concentration ns. At low temperature, electron velocity
increases while the carrier concentration in the drain access region decreases due to
more severe dispersion. The final low temperature conductivity GLT can be either
higher or lower than the room temperature conductivity GRT for different samples.
Therefore, the pulsed current, determined by the conductivity of the channel, is
sample-dependent and difficult to predict.
63
3.3 Models
Although traps were found to be located at the surface, barrier and buffer, our
discussion will be on the surface traps and the virtual gate concept in this section.
This focus is reasonable because the effect of the SiN passivation proves that surface
traps play an important role in dispersion. Several models have been put forward to
understand the dispersion phenomenon, for example: capture and emission of deep
centers [9], and charging and discharging of states through charge transport related
delay [10]. Two physical pictures are applicable. One is that traps release electrons to
the conduction band and then electrons move back to the gate contact through band
conduction. The other is that a high density of traps forms a mini-band in the band
gap and electrons in the deep level move back to the gate contact through hopping
conduction.
EC
EV
ET
Detrapping
Band ConductionEC
EV
ET
Detrapping
Band Conduction
(a)
Fig.3.11. Illustrations of dispersion mconduction model; (b) Hopping condu
6
E C
E V
E T Mini - band
Hopping conduction
E C
E V
E T Mini - band
Hopping conduction
(b)
odels. (a) De-trapping and bandction model.
4
De-trapping and band-conduction model
In this model, the physical procedure can be divided into two steps: the trap
has to release an electron to the conduction band, and then the electron moves back to
the gate contact, as illustrated in fig.3.11(a).
Trap capture and emission rates are the two important trap parameters, which
determine how fast a trap can respond to a signal [11]. The capture rate, which is
defined as the probability per unit time that an electron is captured, is:
n n nc v nσ= (3.3)
where σn is the capture cross section, vn is the thermal velocity of the electron, and n
is the free electron density. The emission rate, which is defined as the probability per
unit time that an electron is emitted from a trap, is:
exp( )t cn n n c
E Ee v NkT
σ −= (3.4)
where Nc is the density of states of the conduction band, Et is the trap energy level,
and Ec is the conduction band edge.
From equation (3.4), a rough estimate can be obtained if the de-trapping time
is considered as 1/en. If we assume that σn is on the order of 10-16cm2, vn is on the
order of 107cm/s, and Nc is on the order of 1018cm-3, then Ec-Et should be ~7-20 kT
(~0.18-0.52eV at room temperature) for a trap occupancy relaxation time from 10-6s
to 1s.
The second step corresponds to the movement of the electron in the
conduction band. Kohn et al. reported a wide range of dispersion frequencies from
65
10-3Hz to 10GHz [12]. This indicates that this phenomenon may not be related to
deep traps alone, but also conduction between the gate and virtual gate. Kohn
introduced the concept of the lossy dielectric [12] [13]. It is assumed that the time
constant associated with the emission of an electron dominates the transient time
constant.
Fig.3.12
conduction throu
The resistor repr
the surface perm
This mod
to the variation
model is not lim
as well. Under t
Fig.3.12.. Lumped approximation of Kohn’s lossy dielectricconnection between the metal gate and surface-state induced virtualgate. σsur and εsur represent the surface conductivity and surfacepermittivity [11] [12].
shows an illustration of the lossy surface dielectric model. The
gh the surface is modeled as a resistor in parallel with a capacitor.
esents the surface conductivity, σsur, while the capacitor represents
ittivity, εsur. The dielectric relaxation time τ is given by:
ετσ
= (3.5)
el attributes the large range of time constants observed for dispersion
of surface conductivities. It is noteworthy that the lossy dielectric
ited to band conduction, but applies to other conduction mechanisms
his physical picture, the surface conduction is assumed to be band
66
conduction. The conductivity is proportional to the charge density and electron
mobility.
If band conduction is dominant at both high and low frequencies, the
dispersion will be improved at low temperature because the electron mobility
increases when temperature decreases. Unfortunately, this is in conflict with
experimental observations. Therefore, at least at low temperatures, the de-trapping
time constant should dominate because of its longer time response at lower
temperatures, as shown in equation (3.5), if this physical picture is applicable.
Hopping conduction model
Some very deep level traps have been reported in AlGaN and GaN. For
example, the study of the 2DEG density as a function of AlGaN thickness found that
the surface barrier height is 1.42eV for an AlGaN layer with a 0.27 Al mole fraction
[14]. This strongly suggests the presence of donor-like surface states located 1.42eV
below the conduction band edge. It was not expected that such a deep trap could
response to high frequency signals. In the model discussed in the previous section,
there is insufficient time for these deep traps to release electrons to the conduction
band at high frequencies. In order to incorporate the deep level traps into a
description of the origins of dispersion, a model of hopping conduction is proposed.
Instead of being released to the conduction band, the electrons just hop from one trap
state to another, as shown in fig.3.11(b), thereby achieving the movement of electrons
back to the gate contact.
67
Hopping conduction is a phonon-assisted tunneling of carriers directly
between localized states. For a given pair of localized states separated by a distance R
and an energy spacing ∆E, the probability per unit time p, of a carrier tunneling
between the two states is given by [15]:
2exp( )exp( )phR Ep va kT
∆≈ − − (3.6)
where a is the characteristic fall-off length of the localized electron wavefunction and
vph is the jump attempt frequency. The first exponential term takes into account the
wavefunction overlap between the electron’s initial and final positions, and the
second term accounts for the phonon that is required due to the energy difference
between the initial and final states. The density of the surface trap states in an
AlGaN/GaN HEMT is so high (~1013cm-2) that a mini-band can form in the band-gap,
as shown in fig.3.11(b). Therefore, the hopping conduction occurs in the mini-band
instead of between two discrete states. Considering all these effects, the final hopping
conductivity, σhopping, can be described as [15]:
0
2exp( )exp( )eff H
hopping
R Ea
σ σ= − −kT (3.7)
where Reff is the characteristic hopping length and EH is the characteristic hopping
energy. Reff can be treated as an average hopping length. Larger trap densities have
smaller Reffs because the trap states are ‘closer’ to each other. Usually this results in a
higher hopping conductivity. Large values of the characteristic hopping energy result
in low values of the hopping conductivity; the hopping conductivity has been
68
interpreted as a measure of the separation between the Fermi level (at T=0K) and the
peak in the trap density of states.
If equation (3.7) is considered at low temperatures, it can be found that the
hopping conductivity becomes smaller because of its kT dependence. This is
consistent with the observation of increased dispersion at lower temperatures.
69
3.4 Summary
In this chapter, the gate-lag measurements of different samples over a range of
temperatures were reported. The dispersion was found to be worse at lower
temperatures; the DC current increased due to higher electron velocities when
temperature decreased. The behavior of the pulsed current at low temperature was
explained by two competing factors: lower carrier concentration in the drain access
region next to the gate due to increased dispersion, and higher electron velocity at
lower temperatures.
Two models were discussed: the de-trapping and band-conduction model and
the hopping conduction model. The first model involves the release of a trapped
electron to the conduction band followed by the electron’s return to the gate by band-
conduction. In the second model the electron moves by hopping from one trap state to
another. Both models explain the increase in dispersion at low temperatures. Further
investigation is required to distinguish these two models.
70
3.5 References [1] Y.-F. Wu, B.P. Keller, S. Keller, D. Kapolnek, S.P. Denbaars, U.K. Mishra
“Measured microwave power performance of AlGaN/GaN MODFET,” IEEE Electron Device Letters, vol. 17, no. 9, pp. 455-457, Sept. 1996.
[2] J. R. Shealy, V. Kaper, V. Tilak, T. Prunty, J. A. Smart, B. Green, and L. F.
Eastman, “An AlGaN/GaN high-electron-mobility transistor with an AlN sub-buffer layer,” J. Phys.: Condens. Matter, Vol. 14, no. 13, pp. 3499-3509, April 2002.
[3] J. B. Walker, High-Power GaAs FET Amplifiers, Chapter 1 (Artech House, Inc.,
Norwood, MA 1993). [4] P. B. Klein, S. C. Binari, K. Ikossi-Anastasiou, A. E. Wickenden, D. D. Koleske,
R. L. Henry, and D. S. Katzer, “Investigation of traps producing current collapse in AlGaN/GaN high electron mobility transistors,” Electron. Lett., Vol. 37, no. 10, pp. 661-662. May 2001.
[5] A. Tarakji, G. Simin, N. Ilinskaya, X. Hu, A. Kumar, A. Koudymov, J. Yang, and
M. Asif Khan, “Mechanism of radio-frequency current collapse in GaN-AlGaN field-effect transistors,” Appl. Phys. Lett., Vol. 78, no. 15, pp. 2169-2171, April 2001.
[6] T. Mizutani, Y. Ohno, M. Akita, S. Kishimoto, and K. Maezawa, “A Study on
Current Collapse in AlGaN/GaN HEMTs Induced by Bias Stress,” IEEE Trans. Electron Dev., Vol. 50, no. 10, pp. 2015-2020, Oct. 2003.
[7] R. Vetury, N. Q. Zhang, S. Keller, and U. K. Mishra, “The impact of surface
states on the DC and RF characteristics of AlGaN/GaN HFETs,” IEEE Trans. Electron. Dev., Vol. 48, no. 3, pp. 560-566, March 2001.
[8] U. V. Bhapkar, M.S.Shur, “Monto Carlo calculation of velocity-field
Characteristic of wurtzite GaN”, Journal of applied Physics, Vol. 82, No.4, pp.1649-1655, Aug. 1997.
[9] P. B. Klein, J. A. Freitas, Jr., S. C. Binari, and A. E. Wickenden, “Observation of
deep traps responsible for current collapse in GaN metal semiconductor field-effect transistors,” Appl. Phys. Lett., vol. 35, pp.4016–4018, 1999.
[10] E. Kohn, I. Daumiller, P. Schmid, N.X. Nguyan, and C.N. Nguyan, “Large
signal frequency dispersion of AlGaN/GaN heterostructure field effect transistors,” Electron. Lett., vol. 35, pp.1022–1024, Dec. 1999.
71
[11] R. Coffie, Ph.D. dissertation, “Characterizing and Suppressing DC-to-RF Dispersion in AlGaN/GaN High Electron Mobility Transistors”, University of Californina, Santa Barbara
[12] I. Daumiller, D. Theron, C. Gaquiere, A. Vescan, R. Dietrich, A. Wieszt, H.
Leier, R. Vetury, U. K. Mishra, I. P. Smorchkova, S. Keller, N. X. Nguyen, C. Nguyen, and E. Kohn, “Current Instabilities in GaN-Based Devices” IEEE Electron Device Lett., Vol. 22 no. 2, pp. 62-64, Feb. 2001.
[13] E. Kohn, I. Daumiller, M. Kunze, M. Neuburger, M. Seyboth, T. J. Jenkins, J. S.
Sewell, J. Van Norstand, Y. Smorchkova, and U. K. Mishra, “Transient Characteristics of GaN-Based Heterostructure Field-Effect Transistors,” IEEE Trans. Microwave Theory Tech., Vol. 51, no. 2, pp. 634 – 642, Feb. 2003.
[14] I. P. Smorchkova, C. R. Elsass, J. P. Ibbetson, R. Vetury, B. Heying, P. Fini, E.
Haus, S. P. DenBaars, J. S. Speck, and U. K. Mishra, “Polarization-induced charge and electron mobility in AlGaN/GaN heterostructures grown by plasma-assisted molecular-beam epitaxy,” J. Appl. Phys., vol. 86, pp. 4520–4526, Oct. 1999.
[15] J. P. Ibbetson, Ph.D. dissertation, “Electrical Characterization of
Nonstoichiometric GaAs Grown at Low Temperature by Molecular Beam Epitaxial”, University of Californina, Santa Barbara
72
Chapter 4 Thick GaN capped
AlGaN/GaN HEMTs
4.1 Introduction
GaN-based HEMTs demonstrated high sheet carrier densities and high DC
current densities, however, high RF output power was not readily attainable. The first
AlGaN/GaN HEMT, reported in 1993 by Khan et al., had a saturation drain current of
40mA/mm [1]. Small signal RF performance of AlGaN/GaN HEMT was first
reported 1n 1994 by Khan et al. with current-gain and power-gain cutoff frequencies
of 11 and 35GHz, respectively, for 0.25µm-gate-length devices [2]. RF power
measurements were not obtained until 1996, when RF power of 1.1W/mm at 2GHz
was reported by Wu et al. [3]. Prior to the integration of SiN in 2000, most of the
unpassivated power densities were below 7W/mm. These power densities were much
lower than the theoretical prediction from DC I-V curves.
This discrepancy between the predicted RF output power from static I-V
curves and load pull measured output power is called “DC-to-RF dispersion” or
current collapse. The origin of this dispersion has been attributed to traps in GaN.
73
Extensive work has been done to identify the properties of these traps, which are
thought to be in the GaN buffer [5], AlGaN barrier [6] and surface [7] [8].
Gate-lag measurements show very obvious current collapse, implying the
possibility that the surface traps play an important role in dispersion, because gate-lag
measurements are considered to be sensitive to surface traps [9] [10]. Higher output
power densities can be achieved with the elimination or reduction of the effects of
these surface traps. As with the other two kinds of traps, surface trap density can be
reduced by more mature growth techniques. However, the effort to control the surface
trap density by epitaxial growth proved to be very difficult. Although one or two
groups reported good power performance of as-grown HEMTs [11] [12], most groups
report poor power performance. Dispersion has been found to be very sensitive to the
growth conditions, processing and so on. It has been known to vary based on the
particular growth chamber used, or even from run to run with the same chamber. For
example, in 1999 MOCVD reactor change at UCSB resulted in much poorer power
performance of unpassivated devices grown in the new reactor as compared those
from the old one. So far no group has reported a very reliable and repeatable growth
technique to reduce the effect of surface traps.
Aside from pursuing more mature growth, another way to reduce the
dispersion caused by surface traps is by surface passivation. Although surface
passivation has been used for a long time for semiconductor devices, usually the
purpose was to provide the protection for the devices. In 2000, the first application of
SiN surface passivation to GaN-based HEMTs was reported [13]. In that paper, the
74
passivation resulted in a doubling of the output power density. Although the value of
the power density was not high (~4W/mm) because of the poor thermal conductivity
of the sapphire substrate, the experiment demonstrated the very promising potential of
SiN passivation to reduce dispersion and improve device power performance. Since
then, SiN passivation has been widely used, resulting in a big increase in average
output power density. A maximum of 11.2W/mm was reported by Shealy et al. in
2002 [14]. SiN passivation significantly reduces the DC-to-RF dispersion. As shown
in fig.4.1, current collapse was effectively reduced or eliminated in gate-lag
measurement after SiN passivation, leading to the higher output power.
DC and 80us
Fig.4passmeas
Much
passivation wh
proposed:
(a) (b)
.1ivaur
wo
ile
DC
Sitioem
rk
no
80us
N passivation reduced dispersion effectively. (a) Beforen, current collapse was observed in gate-lagent; (b) After dispersion, no current collapse.
has been done to investigate the mechanism of the surface
conclusion has been made yet. There have been several theories
75
1) SiN passivation reduces the density of surface traps [15].
2) Positive charge located in the SiN neutralizes the effects of filled surface
states [16].
3) SiN stiffens the surface, reducing piezoelectric charge resulting from the
gate bias-induced non-uniform strain in the AlGaN barrier layer [17].
UCSB was one of the first groups to study the effect of SiN surface
passivation. In 2000, sputtered SiN was first employed for surface passivation and
successfully improved the device power performance [18]. The best performance
was obtained when the SiN films had an index of refraction of 2.025 and an etching
rate less then 2nm/s in buffered HF. Typical film thickness ranged from 50-100nm.
Annealing the deposited film at 350°C for 5 minutes in an N2 atmosphere sometimes
improved the dispersion reduction.
Plasma Enhanced Chemical Vapor Deposition (PECVD) of SiN was also
investigated. Good power performance, greater than 9W/mm, was obtained using this
technique. A dependence on SiN thickness was observed; thin films (less than 20nm)
did not reduce the dispersion. Usually SiN films thicker than 70nm were needed to
effectively eliminate the dispersion.
Although the introduction of SiN led to a big improvement in the GaN-based
HEMTs, there were several disadvantages. The effect of SiN surface passivation was
very sensitive to the deposition conditions. For example, the condition to get the best
passivation using sputtered SiN at UCSB often varied, making it unsuitable for
commercial process integration. Further, after relocation of the sputtering machine to
76
another lab, the deposited SiN films had minimal impact on dispersion despite the
same deposition conditions were used (e.g. gas flow, RF power etc.). The effect of the
surface passivation was also dependent on the surface condition prior to the
passivation; some samples had a large improvement in gate-lag and power
performance while others had small or even no improvement. It was also found that
leakage current increased and breakdown voltage decreased after passivation.
The sensitivity to both surface and deposition conditions combined with the
lack of understanding of the passivation mechanism have resulted in poor
reproducibility of the breakdown voltage, gate leakage and dispersion reduction.
These disadvantages have motivated research into dispersion reduction at the device
epitaxial level.
77
4.2 Solutions to dispersion control in the epitaxial
structure
Based on the virtual-gate model [18], when surface traps next to the gate in
the drain access region are charged, the surface has a negative potential which can
modulate the 2DEG in the channel, acting as a ‘virtual’ HEMT connected serially to
the ‘real’ one. Fig.4.2 shows the schematic of a conventional AlGaN/GaN HEMT.
For a standard AlGaN/GaN HEMT, the AlGaN layer has the same thickness and Al
composition in both the gate and access regions, which also makes the 2DEG
densities in the channel in the different regions the same (the gate metal on the
surface may change the surface potential, thereby affecting 2DEG density, but this
change is relatively small). At the same time, the pinch-off voltage is determined by
the carrier density in the channel and the distance between the surface and channel. It
can be found that for a standard AlGaN/GaN HEMT, both the real and virtual devices
have the same pinch-off voltage, i.e. the ratio between these two pinch-off voltages is
1. This fact explains the severity of dispersion. The virtual gate has the almost same
negative potential as the real gate when the real device is pinched off. Since the ratio
of the pinch-off voltages is 1, the virtual device must also be pinched off. The channel
in the drain access region is completely depleted as is the channel underneath the
gate. Due to the slow response of the surface traps, the virtual device is too slow to
follow the real device at high frequencies. When the real device is turned on, the
virtual device is still off. The depleted channel under the drain access region keeps
78
the drain current from increasing fast enough to follow the gate voltage change; this is
the cause of the dispersion.
-G DS - - -- - - - - - - - - - - - - - -
-AlGaN
GaN channel
Real Device Virtual DeviceCharged Surface traps
-G DS - - -- - - - - - - - - - - - - - -
-AlGaN
GaN channel
-G DS - - -- - - - - - - - - - - - - - -
-AlGaN
GaN channel
Real Device Virtual DeviceCharged Surface traps
Fig.4.2. The surface states in the drain access region forms avirtual gate. The virtual device has the same pinch-off voltageas the real device. The slow response of the virtual gate causesthe dispersion.
However, if the ratio is greater than one, the pinch-off voltage of the virtual
device (i.e. the pinch-off voltage of the drain access region) is much larger than that
of gate region, thus reducing the dispersion. When the real gate completely depletes
the channel underneath gate region, if the pinch-off voltage of the virtual gate is large
enough, e.g. larger than the gate-drain bias, the virtual device can still be on, allowing
for the channel underneath the drain access region to remain open. When the channel
underneath the gate opens again, the always-open channel underneath the drain access
does not impede the increase of the drain current as severely as in the standard device,
even though the surface potential of the drain access region adjacent to the gate still
can not follow the real gate potential quickly enough. Obviously, the larger the pinch-
off voltage ratio is, the better the dispersion reduction.
79
Therefore, in order to reduce dispersion, we need to make the pinch-off
voltages underneath the real gate and virtual gate different. It means two things:
1) The pinch-off voltage in the drain access region must be increased
2) The pinch-off voltage in the gate region must remain similar to the value in
a standard device.
From the epitaxial structure point of view, there are two solutions to
increasing the pinch-off voltage while keeping the charge approximately constant, by
employing either a doped cap layer or a thick unintentionally doped (UID) layer.
Doped cap layer
The first solution is to grow a doped cap layer on top of the standard
AlGaN/GaN HEMT structure. The doping density must be high so that at zero bias
sufficient unionized dopants are present in the cap layer. With this doped cap layer,
surface potential fluctuations modulate the doped layer instead of the channel. Only
after the doped layer is depleted, will the 2DEG in the channel be affected. This
results in a great reduction in the modulation from the surface to the channel.
Dopants can be either donors or accepters. Fig.4.3 shows a schematic epitaxial
structure of an n+ GaN-capped AlGaN/GaN HEMT. The GaN cap is 100nm thick and
is doped by Si of 1019cm-3. At zero bias, only the regions close to the surface and
GaN/AlGaN interface are depleted. Fig.4.4 shows the band diagrams at zero bias and
when a negative bias was applied to the surface. When the surface potential becomes
80
more negative, the n-GaN cap depletion deepens, but the 2DEG density in the
channel is not affected. The pinch-off voltage of the virtual gate can be estimated as:
21 2
0 0
( )2
d AlGaN Alp
q N t q t tσ σε ε ε ε
⋅ ⋅ ⋅ ⋅ + ⋅= +
⋅ ⋅ ⋅3NV (4.1)
where Nd is the Si doping density, σAlGaN is the net polarization charge density of the
AlGaN (subtracting the polarization charge density of GaN ), σAlN is the net
polarization charge density of the AlN, t1 is the thickness of the Si doped GaN cap
layer, t2 is the thickness of AlGaN layer and t3 is the thickness of AlN layer.
0.7 nm AlN
UID GaN
Substrate
100 nm UID GaN:Si: 1019cm-3
30 nm UID Al0.33Ga0.67N0.7 nm AlN
UID GaN
Substrate
100 nm UID GaN:Si: 1019cm-3
30 nm UID Al0.33Ga0.67N
t1
t2 t3
Fig.4.3 Epitaxial structure of the n+-GaN-capped AlGaN/GaNHEMT. 100nm cap layer is heavily doped with Si.
81
0 50 100 150-5
-4
-3
-2
-1
0
1
2
3 E
nerg
y (e
V)
Thickness (nm)
GaN:Si
AlGaN
0 50 100 150-5
-4
-3
-2
-1
0
1
2
3 E
nerg
y (e
V)
Thickness (nm)
GaN:Si
AlGaN
0 50 100 150
-5
0
5
10
15
20
25
Ene
rgy
(eV)
Thickness (nm)
VG=-20V
depletedregion
(a) (b)
Fig.4.4 Band diagrams of the n+-GaN-capped device at (a)zero bias (b) surface potential of –20V. The Si-doped layerscreened the surface potential fluctuations.
It can be found from equation (4.1) that higher doping density or increased
cap thickness can increase the pinch-off voltage, resulting in reduced dispersion. Si
doping during epitaxial growth is a relatively mature technique and easy to
implement. However, the biggest disadvantage of this technique is the low
breakdown voltage resulting from the high doping density in the cap layer.
Accepters, like Mg, could also be used as dopants in the cap layer. Robert
Coffie investigated this approach in detail [19] [20]. The p-capped AlGaN/GaN
HEMTs were shown to reduce dispersion without surface passivation and significant
RF power was obtained. A typical epitaxial structure is shown in fig.4.5. The
advantage of the p-GaN cap HEMT is that the conduction of the Mg doped cap was
less than that of Si doped cap, resulting in lower gate leakage and a higher breakdown
voltage. Nevertheless, there were also some disadvantages to this technique: Mg
82
doping is not as mature as Si doping in GaN growth, and the dispersion control
mechanism was more complicated than that of n-type cap HEMT. The detailed
analysis can be found in Robert Coffie’s dissertation [21].
Fig.4HEM
Thick UID
The second
thick cap layer on
uses the thick cap
According to the c
channel is inversel
surface potential
Unlike the doped
0.7 nm AlN
UID GaN
Substrate
100 nm UID GaN:Mg: ~1020cm-3
30 nm UID Al0.33Ga0.67N10nm graded AlGaN:Si
0.7 nm AlN
UID GaN
Substrate
100 nm UID GaN:Mg: ~1020cm-3
30 nm UID Al0.33Ga0.67N10nm graded AlGaN:Si
.5. Epitaxial structure of the p-GaN-capped AlGaN/GaNT. 100nm cap layer is heavily doped with Mg.
cap layer
solution to increasing the pinch-off voltage is to grow an undoped
top of an AlGaN/GaN HEMT, as shown in fig.4.6. This method
layer to increase the distance between the surface and the channel.
harge control model [22], the modulation ability from surface to
y proportional to their separation. A thick cap layer can reduce the
fluctuations, thus reducing dispersion caused by surface traps.
cap solutions, the surface still modulates the channel directly, but
83
the surface to channel modulation is largely decreased. In this thesis, we will focus on
this method. The detailed analysis will be presented in the following sections.
0.7 nm AlN
UID GaN
Substrate
250 nm UID GaN
30 nm UID Al0.33Ga0.67N0.7 nm AlN
UID GaN
Substrate
250 nm UID GaN
30 nm UID Al0.33Ga0.67N
Fig.4.6 Epitaxial structure of the thick-GaN-cappedAlGaN/GaN HEMT. 250nm cap layer largely reduces themodulation from surface to channel, therefore decreasingdispersion.
All of the solutions discussed can reduce dispersion caused by surface in
theory. They all have both benefits and drawbacks. The undoped cap solution does
not employ any dopants in the cap layer, theoretically giving it a higher breakdown
voltage and lower leakage. Nevertheless, from a dispersion reduction point of view, a
much thicker cap layer is required because there are no dopants to deplete when
surface potential fluctuates. This thicker cap makes processing more complicated. On
the other hand, the doped-cap methods employ thinner cap layers, but the low
breakdown voltages and high gate leakage caused by the high doping concentrations
limit the applicability.
84
The second requirement is similarity of the pinch-off voltage and
transconductance to those of conventional AlGaN/GaN HEMTs in the gate region. If
the whole epitaxial structure is grown at one time, gate recessing is needed to obtain
an appropriate distance between the gate and channel, to ensure a reasonable pinch-
off voltage and transconductance. An alternative way to achieve this goal is through
regrowth: grow an AlGaN/GaN HEMT first, put gate and source/drain contacts on
and then regrow the UID or doped cap on top of drain access region. The advantage
of this approach is that the gate-channel distance can be controlled much more
accurately and the processing is simpler because no deep recessing is required. The
drawback is that the growth process is more complicated because of the regrowth. In
this thesis we will focus on the method of one-time growth with deep-recessing.
85
4.3 Thick GaN capped AlGaN/GaN HEMTs
GaN/AlGaN/GaN HEMT
As discussed in the previous section, we will focus on the epitaxial structure
with a thick undoped cap layer. When the distance between surface and channel
increases, the modulation from surface to channel is reduced, thereby suppressing the
dispersion.
One direct way of achieving this goal is to make the original AlGaN barrier
thicker. In theory, this is the most attractive solution because of its simplicity. The
carrier density is close to that of a standard HEMT because the 2DEG density
saturates when the AlGaN layer is thicker than 30nm. The 2DEG is introduced by the
polarization effects so no dopants are needed. The AlGaN layer can sustain high
electric fields so a high breakdown voltage can be expected. In fact, the shape of the
band diagram of thicker AlGaN barrier HEMT is very similar to that of the standard
30nm-AlGaN HEMT except for the AlGaN thickness. However, since AlGaN has a
large lattice mismatch relative to GaN (increasing with increasing Al mole fraction to
~2.4% for AlN), thick layers can dislocate and crack. In general, for an AlGaN layer
on GaN with an Al composition between 0.2 and 0.35, it starts to crack when its
thickness exceeds 50-60nm. This thickness is still not enough to effectively reduce
dispersion. For this reason, this method cannot be adopted at present. It could become
promising if a lattice matched cap layer could be incorporated. For example,
AlInGaN could a potential candidate.
86
A thick cap is necessary for effective dispersion reduction. From the thickness
point of view, GaN is the ideal candidate because it has no thickness limitation; the
buffer is GaN so there is no lattice mismatch problem. The first conceptual structure
with a UID GaN cap is shown in fig.4.7. It consisted of a UID GaN cap, 30nm
Al0.33Ga0.67N, 0.7nm AlN and a UID GaN buffer. Basically, it just added a GaN cap
layer on top of the standard AlGaN/GaN HEMT.
0.7 nm AlN
UID GaN
Substrate
UID GaN
30 nm UID Al0.33Ga0.67N0.7 nm AlN
UID GaN
Substrate
UID GaN
30 nm UID Al0.33Ga0.67N
Fig.4.7 Epitaxial structure of the thick-GaN-cappedAlGaN/GaN HEMT, consisting of a UID GaN cap, 30nm UIDAl0.33Ga0.67N, 0.7nm AlN and GaN buffer.
Recalling the GaN/AlN/GaN heterostructure discussed in the previous
chapter, this GaN/AlGaN/GaN can be considered a generalization of that structure,
although the purposes of the two structures are different. It is not surprising that this
structure has a similar dependence of carrier density on GaN cap thickness: the 2DEG
density decreases as the thickness of the GaN increases, until it stops decreasing at a
specific value.
87
The simulation of the band diagram is a good way to study the change in the
carrier density. Fig.4.8 shows the band diagrams of the GaN/AlGaN/GaN with
different cap thickness simulated by 1D Poisson Solver. When the GaN cap is
deposited on the AlGaN, the 2DEG in the channel starts to decrease. Because the
polarization charges of AlGaN at the two AlGaN/GaN interfaces cancel each other,
the electric field in the GaN cap is determined by the negative 2DEG only and points
from the surface to the GaN/AlGaN interface. The voltage drop across the cap layer
increases when the cap becomes thicker. Since the surface potential of GaN is
assumed to be fixed, the increasing voltage drop leads to raising the band at
GaN/AlGaN interface. This is equivalent to applying a negative bias on the standard
AlGaN/GaN HEMT, resulting in a reduced 2DEG. Note that the magnitude of the
electric field also decreases along with the decrease in the 2DEG density. The 2DEG
density can be obtained by solving the equation (4.2):
1 2 3
0 0 0
( ) ( )s AlGaN s AlN sB
q n t q n t q n t q n d0
0
sσ σφε ε ε ε ε ε ε⋅ ⋅ ⋅ − ⋅ ⋅ − ⋅ ⋅ ⋅
= − + + −⋅ ⋅ ⋅ ε⋅ (4.2)
02 3
1 2 3 0
AlGaN AlN B
s
t tqn
t t t d
εεσ σ φ⋅ + ⋅ −=
+ + + (4.3)
88
(a)
0 10 20 30 40 50 60
-4
-2
0
2
4 E
nerg
y (e
V)
Thickness (nm)
0 50 100 150 200 250 300-10
-8
-6
-4
-2
0
2
4
+ps+σsur=+(ns-ps) -sAlGaN
+sAlGaN
-ns
Ene
rgy
(eV)
Thickness (nm)
0 50 100 150
-4
-2
0
2
4
Eg-φB
φB
Ene
rgy
(eV)
Thickness (nm)
positive charges
100nm GaN
AlGaN
250nm GaN
AlGaN
7.5nm GaN AlGaN
(b)
)
Fig.4.8 The band diagraAlGaN/GaN HEMTs with di(b) 100nm; (c) 250nm. Thinterface contacts the Fermi
8
(c
ms of the thick-GaN-cappedfferent cap thickness: (a) 7.5nm;e valence band at GaN/AlGaNlevel when GaN cap is thick.
9
where ns is the 2DEG density in the channel, σAlGaN is the net polarization charge
density of the AlGaN (subtracting the polarization charge density of GaN ), σAlN is
the net polarization charge density of the AlN, t1 is the thickness of GaN cap layer, t2
is the thickness of AlGaN layer, t3 is the thickness of AlN layer, d0 is the distance
between centroid of the 2DEG and AlGaN/GaN interface, and φB is the surface
potential. In the right side of equation (4.2), the first term represents the voltage drop
across the GaN cap, the second term represents the voltage drop across the AlGaN
layer, the third term represents the voltage drop across the thin AlN layer, and the last
one represents the potential drop induced by the 2DEG based on a simplified triangle
potential well model. Equation (4.3) clearly demonstrates that the decrease in 2DEG
density is inversely proportional to the GaN cap thickness.
As the equation predicts, a decrease in 2DEG increases the field in the
AlGaN, raising the band at the GaN/AlGaN interface. When the thickness of the GaN
cap reaches a specific value, which makes the 2DEG density drop to a critical value,
the valence band at the GaN/AlGaN interface contacts the Fermi level, as shown in
fig.4.8(b). The bands at this interface no longer rise because even a small rise in the
band can lead to a large amount of charge variation for the semiconductor is
degenerate. This results in a near freezing of the 2DEG density because the energy
band of AlGaN/GaN structure remains unchanged even when the GaN cap becomes
thicker, as shown fig.4.8(b) and (c). The final value of the 2DEG density is given by:
90
02 3 ,
02 3 0
AlGaN AlN g AlGaN
s
t t Eqn
t t d
εεσ σ⋅ + ⋅ − ∆=
+ + (4.4)
where ∆Eg is the band gap of the AlGaN. Even though the energy band of the
AlGaN/GaN structure is fixed, the energy band of the GaN cap still continues to
change when the cap gets thicker. The decrease of the band’s slope indicates a
smaller electric field in the thicker cap (in fig.4.8(c)), which is reasonable because the
total voltage drop across the cap is now constant at (Eg-φB). Since the 2DEG density
cannot change, the system has to compensate in a different manner. It can be noted
that there is an accumulation of positive charges, which are holes in this simulation,
when the valence band contacts the Fermi level. At this point the electric field in the
GaN cap is determined by the difference between the 2DEG and hole concentrations.
The decrease of the electric field occurs because of the increase of the hole
concentration, not because of the decrease of the 2DEG density as before. The hole
sheet density can be written as:
0 ,0
1
( )g AlGaN Bs s
Ep n
q tε ε φ⋅ ⋅ ∆ −
= −⋅ (4.5)
where ns0 is as given in equation (4.4).
Fig.4.9 shows the 2DEG and hole densities as a function of GaN cap
thickness, based on the structure shown in fig.4.7. When 2DEG density becomes a
constant, the hole density gradually increases. The net charge continues to decrease,
which is consistent with the reduction of the electric field in the GaN cap. When the
91
GaN cap becomes very thick, the hole density approaches the 2DEG density,
resulting in a v ry small electric field in the cap layer.
This G
shows the cha
growing a ser
etching was em
and a 6nm Al
used to obtai
measurements
density with d
well. The exp
gate to control
e
0 50 100 150 200 250
0.4
0.6
0.8
1.0
1.2
1.4
Hole
Electron
n s, p s (
1013
cm-2)
Thickness of GaN Cap (nm)
Fig.4.9 Simulated electrons and holes concentration as afunction of GaN cap thickness in the structure shown infig.4.7.
aN cap thickness dependence was verified experimentally. Fig.4.10(a)
nge in the 2DEG density as function of cap thickness. Rather than
ies of samples with different GaN cap thickness, in this experiment
ployed to vary the cap thickness. A sample with a 50nm UID GaN cap
0.6Ga0.4N barrier layer was grown by MOCVD. RIE dry etching was
n different cap thickness. 2DEG densities were obtained by Hall
. The experimental data clearly demonstrated an increase of carrier
ecreasing GaN cap thickness, which agreed with the simulation very
eriment also proved that RIE etching could be used for recessing the
the pinch-off voltage in this structure.
92
An experiment varying the GaN cap thickness in a series of growths of
samples with different GaN cap thickness was reported in the last chapter. This
experiment involved a GaN/AlN/GaN structure, a limiting case of the general
GaN/AlGaN/GaN heterostructure. As expected, it showed similar behavior in the
change of 2DEG density as a function of cap thickness. Fig.4.10(b) shows the
experimental and simulation data, which are in agreement.
0 20 40 60 80 100
1.0
1.5
2.0
2.5
n s (10
13/c
m2 )
Thickness of GaN Cap (nm)
Simulation Experiment
0 20 40 60 80 100 120 140 160
1.2
1.6
2.0
2.4
n s (10
13/c
m2 )
Thickness of GaN Cap (nm)
Simulation Experiment
(a) (b)
Fig.4.10 Simulated and experimental 2DEG concentration as a functionof GaN cap thickness. (a) cap thickness determined by etching; (b) capthickness determined by growth.
This behavior of charge with GaN thickness can provide device designers
more freedom to adjust the carrier density in different regions by simply etching the
GaN cap to meet their specific requirements. For instance, the GaN cap in the source
access region can be etched to increase the 2DEG density in the corresponding
channel, thus reducing source access resistance.
93
The ultimate goal of the thick cap layer is to increase the pinch-off voltage in
the drain access region, so that the effect of surface potential fluctuations can be
decreased, thus reducing dispersion.
Simple electrostatics predict that the larger the distances between the channel
and the surface the larger the pinch-off voltage given the fixed charge in the channel.
This was the basis of the discussed capped HEMT concept. Fig.4.11 shows the
simulated band diagrams of two AlGaN/GaN HEMTs with 30 and 300nm (assuming
it can be grown) AlGaN barriers, respectively, when the channels are depleted.
0 10 20 30 40 50
-10
-5
0
5
10
+sAlGaN
-sAlN
+sAlN
Ene
rgy
(eV)
Thickness (nm)
0 50 100 150 200 250 300 350-20
0
20
40
60
80
100 E
nerg
y (e
V)
Thickness (nm)
(a) (b)
Fig.4.11 Band diagram of the devices when the channels are depleted.(a) 30nm-thick AlGaN barrier; (b) 300nm-thick AlGaN barrier. Thepinch-off voltage is almost proportional to the barrier thickness.
The pinch-off voltage of the 300nm-AlGaN-barrier HEMT is almost 10 times
larger than that of the 30nm-AlGaN. Since the 2DEG densities of these two HEMTs
are similar due to the charge saturation for AlGaN barrier thicker than 30nm, the
pinch-off voltage is basically proportional to the AlGaN thickness, as shown in
equation (4.6). From the simulated band diagrams in fig.4.11, it can be noted that
94
when the channel is depleted, the electric field is solely determined by the net
negative polarization charge at AlGaN/GaN interface. Therefore, the thicker the
AlGaN barrier is, the larger the pinch-off voltage.
0
AlGaNp AlGaN
qV tσε ε⋅
≈ ⋅⋅ (4.6)
Unfortunately, the situation in a GaN/AlGaN/GaN HEMT is more
complicated. It is necessary to check whether this rule remains applicable. Studies
have revealed that the positive charges introduced by band bending at the
GaN/AlGaN interface plays an important role in the saturation of the 2DEG density.
In simulations, it is assumed that all of these positive charges are holes. Let us
assume that the holes are fast enough to follow signals at high frequency. An epitaxial
structure of 100nm GaN/30nm AlGaN/0.7nm AlN/GaN was used for the simulation.
Fig.4.12 shows the band diagram at zero bias and pinch-off. The band diagram of a
similar structure with a 7.5nm GaN cap was also included as a reference and is shown
with dash lines. At pinch-off condition, the simulation shows that the holes were
depleted as well. This result is reasonable because there is no confinement for the
mobile positive charges at pinch-off. It is surprising to find that the pinch-off voltage
of the 100nm-GaN-cap HEMT is the same as that of the 7.5nm-cap device. The thick
cap layer does not increase the pinch-off voltage, i.e. it does not screen the surface
potential fluctuations.
95
0 20 40 60 80 100 120 140 160
-4
-2
0
2
4
Thickness (nm)
Ener
gy (e
V)
AlGaNGaN
0 20 40 60 80 100 120 140 160-10
-5
0
5
10
-sAlGaN
+sAlGaN
Thickness (nm)
Ener
gy (e
V) AlGaNGaN
(a) (b) Fig.4.12 Band diagrams of the devices with a 100nm cap (solid line) and7.5nm cap (dash line). (a) zero bias; (b) pinch-off. These two deviceshave the same pinch-off voltage.
From the band diagram, it can be seen that the energy band of the GaN cap is
flat when the channel is depleted. This is because in a GaN/AlGaN/GaN HEMT, the
electric field in the cap layer is determined by the 2DEG, ns, alone, unlike in the
standard AlGaN/GaN HEMT where the electric field in the barrier layer is
determined by the difference between the net positive polarization charge and
negative 2DEG in the channel.
0
sCap
q nEε ε⋅
≈⋅ (4.7)
The opposite polarity polarization charges at the two AlGaN/GaN interfaces
cancel with each other. When the device is pinched off and the channel is completely
depleted, the electric field in the GaN cap is zero and the energy band is flat. No
96
matter how thick the cap is, the pinch-off voltage is always the same as that of a thin
cap device. The pinch-off voltage can be written as
0
AlGaNp AlGaN
qV tσε ε⋅
≈ ⋅⋅ (4.8)
Notice that equation (4.8) is the same as (4.6). The thickness of the GaN cap
does not appear in the equation. Therefore, the pinch-off voltage is independent of the
cap thickness if all of the positive charges at the GaN/AlGaN interface are mobile. If
this were true, the GaN/AlGaN/GaN structure would not help in reducing the
dispersion because the modulation from surface to channel would remain the same
regardless of the cap thickness.
If a contrary scenario is proposed with all of the positive charges at the
GaN/AlGaN interface being fixed (e.g. holes trapped near the valence band), totally
different results are obtained. In fig.4.13, the band diagram with a depleted channel
shows that the pinch-off voltage increases with increasing thickness of the GaN cap.
0 50 100 150 200 250 300 350-20
0
20
40
60
80
100
Ene
rgy
(eV)
Thickness (nm)
Fig.4.13 Band diagram of the device at pinch off. The polarization-induced positive charges at GaN/AlGaN interface were assumed to befixed. The pinch-off voltage increases with the increasing of the capthickness.
97
In this case, the electric field in the GaN cap and pinch-off voltage can be
described as:
,
0
( )s fix sCap
q p nE
ε ε⋅ −
≈⋅ (4.9)
,
0 0
s fixAlGaNp AlGaN GaN
q pqV tσε ε ε ε
t⋅⋅
≈ ⋅ +⋅ ⋅
⋅ (4.10)
The difference between the fixed positive charge and the 2DEG determines
the electric field. Therefore, when the 2DEG in the channel is depleted, there is still
an electric field in the cap, which is the critical factor for the presence of a thickness
dependent pinch-off voltage.
The properties of these polarization-induced positive charges are still under
investigation. They may be holes, or trap-related fixed charges, or trap-related
‘mobile’ charges (i.e. deep levels can trap and de-trap charges). Based on the
discussions above, the pinch-off voltage of the un-recessed sample without Si doping
could give some information. If these positive charges were holes, the pinch-off
voltage should be similar to a standard HEMT. Otherwise, if they are fixed, the
pinch-off voltage should be very high. A sample with epitaxial structure similar to
fig.4.7 was processed and measured. DC pinch-off voltage of about –20V was
observed, which was between the two calculated values, -10 and -40V, respectively.
This result implied that most of the positive charges were either related with the deep
levels with relatively fast de-trapping time, or mobile holes. Further investigations are
needed in the future.
98
Introduction of graded AlGaN layer doped with Si
From the previous simulations and experiments, it is clear that some amounts
of fixed positive charges are required because the thickness-dependent pinch-off
voltage is desirable for dispersion reduction. Moreover, the positive charges need to
be controllable. One way to do this directly is to dope the structure with donors
because the ionized donors are fixed positive charges.
The positive charges introduced by the band bending at the GaN/AlGaN
interface are due to the negative polarization charges attracting positive charges. We
can therefore engineer charge balance during growth by introducing donors. One
possible doping profile is delta doping at the GaN/AlGaN interface, as shown in
fig.4.14(a). The band diagram is shown in fig.4.14(b).
0.7 nm AlN
UID GaN
Substrate
250 nm UID GaN
30 nm UID Al0.33Ga0.67N+++++++++++++++++++++++++
0.7 nm AlN
UID GaN
Substrate
250 nm UID GaN
30 nm UID Al0.33Ga0.67N+++++++++++++++++++++++++
Positive ionized donors
0 50 100 150 200 250 300
-4
-2
0
2
Thickness (nm)
Ener
gy (e
V)
AlGaNGaN
Fig.4.14 Epitaxial structure and band diagram of the GaN-cappedAlGaN/GaN HEMTs with delta Si doping at GaN/AlGaN interface.
99
It clearly shows that the Si doping lowers the valence band much below the
Fermi level, so that mobile positive charges are not presented. Meanwhile, the
positive ionized donors produce the electric field in the GaN cap which ensures the
thickness dependent pinch-off voltage. However, some disadvantages of that
approach are that the amount of delta-doping can not be controlled very accurately
and the doping density is usually very high which may degrade the growth quality of
the epitaxial layer above it. Therefore, a doping profile over a limited distance with a
more reasonable doping density may be preferred. To achieve point by point
neutrality, the polarization charge should be distributed over the same distance. This
can be achieved by grading from AlGaN to GaN, instead of using an abrupt junction.
The epitaxial structure is shown in fig.4.15(a) with a corresponding band diagram in
fig.4.15(b).
100
0 50 100 150 200 250 300 350
-8
-6
-4
-2
0
2
+sSi-sAlGaN -ns
+sAlGaN
Thickness (nm)
Ener
gy (e
V)
0.7 nm AlN
UID GaN
Substrate
250 nm UID GaN
20 nm UID Al0.33Ga0.67N
20 nm graded AlxGa1-xN(x=0-0.33):Si
0.7 nm AlN
UID GaN
Substrate
250 nm UID GaN250 nm UID GaN
20 nm UID Al0.33Ga0.67N
20 nm graded AlxGa1-xN(x=0-0.33):Si
AlGaN GaN
graded AlGaN
Fig.4.15 Epitaxial structure and band diagram of the GaN-cappedAlGaN/GaN HEMTs with a Si-dopde graded AlGaN layer.
The new structure has a 20nm graded AlGaN layer over which the Al
composition varies from 33% to 0%, i.e. GaN. This layer is doped by Si with a
doping density of 8.2×1018cm-3, which exactly compensates the negative polarization
charge, and is much lower than that of delta-doping, as shown in fig.4.15(b). There
are positive polarization charges at the AlGaN/GaN interface. The valence band in
the graded region is pulled down far below the Fermi level, so there are no longer
mobile positive charges present there. This structure has a thickness-dependent pinch-
off voltage:
1 2 3 20
1[ ( ) (2p Si AlGaN AlGaN Si
qV t t t tσ σ σε ε
= ⋅ + + ⋅ − ⋅ −⋅
)]σ (4.11)
where t1 is the thickness of GaN cap layer, t2 is the thickness of graded AlGaN
layer, and t3 is the thickness of AlGaN layer. Therefore, this epitaxial structure can
satisfy the initial requirement and is used as the basis for our reduced-dispersion
device.
101
Growth
The growth of the thick (250nm) UID GaN cap layer on top of Al0.33Ga0.67N
was very successful. High quality material was obtained. AFM and TEM were used
to characterize the material. Fig.4.16 shows the surface morphology of the
GaN/AlGaN/GaN HEMT measured by AFM. It shows a very clear step growth and
no cracking is observed. The TEM image in fig.4.17 also demonstrated that the
growth has a high quality.
Fig.4.16. Surface morphology of the GaN/AlGaN/GaN sample by AFM.
buffer
thickGaNcap
Fig.4.17. TEM image of the GaN/AlGaN/GaN sample.
102
Device Structure
Fig.4.18 shows the schematic of the device structure. It is very similar to the
standard HEMT, except that a deep recess is required for both the ohmic and gate
metallization, in order to get low source/drain contact resistance, reasonable
transconductance and a desirable pinch-off voltage. The active region can be divided
into four areas: the source access region, the gate region, the drain access region and
the source/drain contact region. The requirements for these regions are now discussed
presently.
G DS
Bulk and graded AlGaN
S.I. GaN
GaNGaN G DS
Bulk and graded AlGaN
S.I. GaN
GaNGaN
Gate Drain access
Source access
Fig.4.18 Device structure. Ohmic and gate deep recessing are needed toobtain good S/D contact resistance and reasonable pinch-off voltage.
Source access region: Because the dispersion is related to the drain access
region, there are no specific requirements for source access region from a dispersion
point of view. However, a low source resistance is desirable in order to increase the
extrinsic transconductance and the current-gain cutoff frequency ft. One possible
103
approach is to increase the carrier density just in the channel underneath the source
access region. In our novel structure, the charge concentration can be increased by
simply etching the GaN cap layer. For example, fig.4.10(a) shows the simulated
charge density as a function of GaN cap thickness. The charge increased from
0.8×1013cm-2 to more than 1.1×1013cm-2 when only 10nm cap layer remained. One
disadvantage of this approach is that the charge increases greatly only when the
remaining cap layer is thin, otherwise the charge remains roughly constant. This
technique requires very accurate etching which is not easy to achieve.
Gate region: This region has to be etched or else the pinch-off voltage will be
very large, and transconductance will be very small, which is undesirable for practical
devices. In order to obtain a reasonable pinch-off voltage (~ -4 to -8V) and
transconductance (~200mS/mm), only a layer as thick as a standard AlGaN/GaN
HEMT barrier layer needs to be retained, about 30nm. Moreover, the etching needs to
be done very carefully in order to keep the damage as low as possible.
Drain access region: This region will remain as it is, because the high pinch-
off voltage introduced by the thick cap layer is the key to dispersion control.
Source/drain contact region: Direct application of the standard S/D ohmic
metallization could lead to bad ohmic contacts because of the thick cap. Although the
alloyed metal penetrates into the semiconductor after annealing, the presence of a
very thick cap layer (>70nm) degrades the contact substantially. Therefore, recessing
is required to obtain good ohmic contacts. Since the present Ti/Al/Ni/Au recipe is
104
optimized for a 30nm AlGaN barrier layer, leaving 30nm layer left on top of the
channel after etching is desirable.
105
Processing
The processing flow is similar to that of the standard AlGaN/GaN HEMTs.
1) Ohmic lithography and deep recessing by RIE dry etching: ~30nm AlGaN left
2) Ohmic metallization by Ebeam deposition: Ti/Al/Ni/Au and RTA annealing
3) Mesa lithography and etching by RIE dry etching
4) Planarization with SiO2 by Ebeam deposition
5) Gate lithography and deep recessing by RIE dry etching
6) Gate metallization by Ebeam deposition: Ni/Au
Filling the top of the mesa-etched areas with SiO2 is important for the device.
The mesa etching is very deep (~400nm); without this filling, there is a ~ 400nm step
between the gate and the gate feed (on top of mesa-etched area), which can result in
the disconnection of the metal. Fig.4.19 shows the SEM image of the broken metal.
Broken metal
Fig.4.19 SEM image of the broken metal across the step.
106
Deep recessing is a key processing step for this novel device; it is used for
both ohmic and gate formation. It is essential for obtaining good contact resistance, a
reasonable pinch-off voltage, and a reasonable transconductance as well. RIE is
chosen due to its simplicity and relative maturity.
There are two requirements for the etching:
1) The etching rate should be repeatable and well controlled.
2) Etching should introduce little damage and the etched surface should be
smooth, especially for gate recessing.
Dario Buttari has done a comprehensive investigation of RIE etching. In order
to meet the first requirement, a standard cleaning process is employed before the
etching. O2 is used to clean the chamber and then the working gases (e.g. Cl2 and
BCl3) are introduced to do a dummy etch. This cleaning procedure effectively ensures
similar chamber conditions before each etch no matter what kind of etching was done
in the chamber before.
A plasma is generated during RIE etching. Accelerated ions can damage the
etched surface and sidewalls. This damage results in a high gate leakage which
degrades the device performance. Lowering the power setting was found to reduce
the damage [23]. In our processing, 15W was used to keep the damage at an
acceptably low level.
A smooth etched surface is a basic requirement for recessing, particularly for
gate recessing. However, traditional Cl2 etching did not work very well: a rough
surface was often obtained. Further investigation revealed that the oxide layer on the
107
surface, which was difficult to etch with Cl2 [24], was the cause of the roughness.
BCl3 was found to remove this oxide effectively. Therefore, a short BCl3 etch was
introduced prior to the major etching [24]. After that, a mixture of BCl3/Cl2 gases was
used for the deep etching. For a standard AlGaN/GaN HEMT, Cl2 alone (with BCl3 at
the beginning) proved to be effective [24]. However, it was not suitable for our device
process. Some experiments have implied that an unstable etching rate occurs
sometimes. One of the possible reasons is that the oxygen from the SiO2 filling the
mesa-etched region may have reacted with the GaN to form an oxide, preventing
further etching. For this reason, BCl3 was introduced with the Cl2 to screen the effect
of oxygen since the BCl3 would react with it.
The etch depth was controlled by time. The etch rate was carefully calibrated
at specific etching conditions where the etch depth was calculated based on the
thickness of epitaxial layers. Etch depth was measured by Atomic Force Microscopy
(AFM). An etch rate of 0.1nm/s was measured under the following conditions:
Cl2=5sccm, BCl3=20sccm, pressure=10mTorr, and power=15W. Devices were
processed successfully using these conditions.
However, one disadvantage of this technique is that it is difficult to control
accurately. The etching rate could have ~10% variance; the thickness of epitaxial
layers could have ~10% variance. There is also a non-uniformity of the epitaxial layer
thickness across the sample (e.g. the layer thickness close to the edge could be thinner
than in the middle of the sample). The combination of these uncertainties lowers the
yield. For this reason, improvements in etching technology are always being pursued
108
with the goal of developing effective selective etch and etch-stop layers. Achieving
large selectivity of the etching rate between GaN and AlGaN making the etch stop at
the AlGaN layer regardless of the etching rate variance or sample non-uniformity is
highly desirable.
Some other technologies which improve the processing are also under
investigation. For example, if the ohmic recessing is too deep or too shallow, high
ohmic resistance can result. However, if the ohmic region is highly doped, it would
be much easier to obtain a good ohmic contact in the absence of accurate recessing.
To achieve this goal, ion implantation has been studied by Haijiang Yu. The initial
results show that it does greatly reduce the requirement for accurate etching [25].
109
Issues of epitaxial parameters
The ultimate goal of our GaN-capped AlGaN/GaN HEMT structure is to
reduce dispersion without surface passivation. As discussed in the previous sections,
the mechanism of dispersion reduction in our novel device is to screen the surface
potential fluctuations by increasing the pinch-off voltage of the drain access region.
From this point view, a high drain access region pinch-off voltage is always
preferred. However, other factors such as processing and growth difficulties have to
be considered at the same time. Some trade-offs must be made to balance the
different requirements.
Given a structure similar to that in fig.4.15, the pinch-off voltage can be
written as:
(4.11) 1 2 3 20
1[ ( ) (2p Si AlGaN AlGaN Si
qV t t t tσ σ σε ε
= ⋅ + + ⋅ − ⋅ −⋅
)]σ
where σSi is the Si doping sheet density, σAlGaN is the net polarization charge density
of the AlGaN (subtracting the polarization charge density of the GaN), t1 is the
thickness of the GaN cap layer, t2 is the thickness of the graded AlGaN layer, and t3 is
the thickness of AlGaN layer.
Obviously, a thicker cap is desirable because it increases the pinch-off
voltage. Fig.4.20 shows pinch-off voltage as a function of cap thickness, given the
parameters in fig.4.15. It can be found that the pinch-off voltage can be as high as
200V when the GaN cap is 600nm, which is very good for dispersion reduction. A
typical DC drain bias is around 50V. The pinch-off voltage of 200V in the drain
110
access region implies that there will still be plenty of electrons in the channel. From a
dispersion control point of view, the thicker cap is better. However, if the cap is too
thick it can make processing more difficult. If a 5% variance is assumed for etching,
the etching error could be as large as 30nm for a 600nm-deep etch, which is greater
than the process tolerance.
Tb
Fig.4he e ac
Cons
thickness. T
the etching e
If etc
easier, a ve
0 100 200 300 400 500 6000
50
100
150
200
Thickness of GaN cap (nm)
Pinc
h-of
f Vol
tage
(V)
t .20 Simulated pinch-off voltage as a function of GaN cap hickness.epitaxial structure is shown in fig.x. A pinch-off voltage of 200V canhieved when the cap is 600nm thick.
idering all of these design aspects, 250nm was chosen as a standard
his cap thickness can provide a pinch-off voltage of 85V without pushing
rror beyond process tolerances.
h-stop technology were available, making the control of deep recessing
ry thick GaN cap would still preferred because it provides better
111
dispersion reduction. As etching technology progresses, sophisticated drain access
region design will be implemented.
Si doping density is another important parameter; it is directly related to the
2DEG density and dispersion control ability. From a dispersion reduction point of
view, a high doping density is preferred. Equation (4.11) clearly shows that a higher
Si doping density in the graded layer leads to a higher pinch-off voltage. This is
understandable because higher Si doping results in a higher carrier density in the
channel, which therefore increases the pinch-off voltage. A higher carrier density is
also desirable because it can support a higher current density, and therefore higher
power. However, excessive Si doping can also introduce parallel conduction in the
AlGaN. In order to study the effect of the Si doping density, the band diagrams of
structures with different Si doping densities are shown in fig.4.21. Table 4.1 lists
some of the simulated and experimental 2DEG densities for different structures with
different doping densities. If the Si sheet doping density is higher than the
polarization density in the graded AlGaN region, parallel conduction can occur, as
shown in fig.4.21(a), in which the Si density is 1.2 times the polarization charge. The
conduction band in the graded region is very close to the Fermi level, implying an
electron accumulation there. This was verified by the experiment, as listed in table
4.1. The 2DEG density of 2.1×1013cm-2 measured at room temperature was higher
than at low temperature (77K), 1.7×1013cm-2. The carrier freeze-out is a clear
indication of parallel conduction. The difference of 4×1012cm-2 was predicted by the
simulation. If the doping density is too low, e.g. the Si density is only half of the
112
polarization charge as shown in fig.4.21(c), the fixed positive ionized donors are not
numerous enough to lower the band diagram in the graded AlGaN region. The
valence band is then close to the Fermi level and there is the possibility of
accumulating positive charges. In order to avoid these two extreme cases, the Fermi
level should be at the middle of the band gap, as shown in fig.4.21(b). Based on the
simulation, a Si doping density of 70-80% of the polarization charge can satisfy this
requirement.
Table 4.1 Experimental and simulated 2DEG densities with different doping densities
Si doping (cm-2)
ns(300K) (cm-2)
ns(77K) (cm-2)
ns,pal (cm-2)
ns,sim (cm-2)
2.1×1013 2.12×1013 1.70×1013 0.42×1013 1.70×1013
1.4×1013 1.35×1013 1.35×1013 0 1.36×1013
1.14×1013 1.15×1013 1.15×1013 0 1.1×1013
113
0 50 100 150 200 250 300 3
-4
-2
0
2
Thickness (nm)
Ener
gy (e
V)
(a)
0 50 100-10
-8
-6
-4
-2
0
2
4
Ene
rgy
(eV)
Thi
Fig.4.21. Band diagrams of densities. The ratio between charge density is (a) 1.2; (b) 0.8graded region in (a). In (caccumulate in the graded region
0
0 50 100 150 200 250 300 350
-4
-2
0
2
graded AlGaN
AlGaNGaN
Ene
rgy
(eV)
Thickness (nm)5
(b)
150 200 250 300 350
+sSi-sAlGaN
+ps-ns
+sAlGaN
ckness (nm)(c)
the devices with different Si dopingthe Si doping density and polarization; (c) 0.5. There is parallel conduction in), small amount of positive charges.
114
Device performance
The epitaxial structure of the first generation of the GaN-capped device was
similar to that in fig.4.15, but the Si doping density was lowered to 6.0×1018cm-3 in
order to keep the Fermi level at the middle of the band gap. The device consisted of a
250nm UID GaN cap layer, a 20nm graded AlxGa1-xN (x=0-0.33) layer doped with Si,
a 20nm Al0.33Ga0.67N layer, a 0.7nm AlN interfacial layer, and a semi-insulating GaN
buffer. The band diagram is shown in fig.4.21(b). The graded AlGaN layer is doped
by Si to remove the accumulation of mobile positive charges and obtain a thickness
dependent pinch-off voltage, as discussed in previous sections. The thin AlN layer is
utilized to remove alloy disorder scattering, thus improving the 2DEG mobility [26].
The sample was grown by MOCVD on a sapphire substrate. The room temperature
sheet charge density and Hall mobility were 1.15×1013 cm-2 and 1750 cm2/V-s,
respectively.
DC and gate-lag pulsed I-V measurements were used to check the dispersion
at relatively low frequencies, as shown in fig.4.22(a). The shortest pulse width used
was 200ns which was limited by the measurement system. As expected, dispersion
was reduced greatly without any surface passivation. As compared to the DC current,
the higher pulsed current was due to the self-heating effect, caused by poor thermal
conductivity of the sapphire substrate. Shorter the gate pulses resulted in less heat
generation, so the current measured with a 200ns-pulse-width was the highest. A
current density of 1.2A/mm was measured at a gate bias of +1V while the
115
corresponding value of the DC current was only 970mA/mm. The pinch-off voltage
was around –5.5V.
0 5 10 15
200
400
600
800
1000
1200
1400VG = +1 V
Vp = -5.5 V
VG: -9 -> +1 V
I D (m
A/m
m)
VD (V)
DC 80us 200ns
0
(a)
Fig.4.22. Thick-GaN-capped passivation on sapphire (a) DCand pulsed mode transconduct200ns.
Fig.4.22(b) shows the transcond
pulsed mode and DC measurements. Be
temperature, the best transconductance
measurement: 230mS/mm at a drain bias
The concept of the thick cap was
of dispersion. The DC and pulsed I-V
standard SiN passivated AlGaN/GaN H
reduction between the standard passivate
lag measurements at low temperature w
dispersion becomes more severe when te
trapping and movement are slower. The
-8 -6 -4 -2 0 2
0
50
100
150
200
250VD = 6 V
g m
(mS/
mm
)
VG (V)
DC 80us 200ns
(b)
AlGaN/GaN HEMT without surface and gate-lag I-V characteristics; (b) DCance. No dispersion was observed up to
uctance as a function of gate bias for the
cause the mobility degraded with the rising
was obtained from the 200ns-pulse-width
of 6V.
proved to work with the successful removal
curves in fig.4.22(a) are also typical for
EMTs. In order to compare the dispersion
d device and the novel capped device, gate-
ere performed. As discussed in chapter 3,
mperature decreases because both charge de-
refore, the passivated HEMT which did not
116
have dispersion in gate-lag measurements at room temperature may have dispersion
at low temperature. Both the standard and novel devices were measured at room
temperature and low temperature (T=100K). Fig.4.23 shows the DC and pulsed I-V
curves of two samples at low temperature. Both had no current collapse at room
temperature. However, at low temperature, the passivated standard AlGaN/GaN
HEMT had dispersion while our novel HEMT did not. This implies that the effect of
dispersion reduction was better for the GaN capped HEMTs than for the standard
passivated ones.
0 5 10 15
0
200
400
600
800
1000
1200
1400
I d (m
A/m
m)
Vds (V)
DC 20us 200ns
(a)
Fig.4.23. DC and gate-lag measuremstandard SiN passivated HEMT; (bvoltage walkout for standard device
Unfortunately, the gate leakage
terminal leakage current was measured
standard AlGaN/GaN HEMT, the lea
higher. More than 10mA/mm of gate l
0 5 10 15
0
200
400
600
800
1000
1200
1400
I D (m
A/m
m)
VD (V)
DC 80us 200ns
(b)
ents at low temperature (T=100K) of (a)) GaN capped HEMT . There was a kneebut not for the capped one.
was found to be high. The gate-drain two
, and is shown in fig.4.24. Compared to the
kage current was 3-4 orders of magnitude
eakage was measured at a gate-drain bias of
117
20V while the standard AlGaN/GaN HEMT showed less than 10uA/mm under the
same bias condition. As a consequence, the breakdown voltage was very low. Two-
terminal destructive breakdown was as low as 20-35V, which severely limited the
application of the drain bias, as shown in fig.4.25.
Fig.HEone
Figlim
-20 -15 -10 -5 0
1x10-5
1x10-4
10-3
10-2
10-1
100
101
102
VGD (V)
I G (m
A/m
m)
GaN/AlGaN/GaN HEMT AlGaN/GaN HEMT
4.24. Two terminal gate-drain leakage currents. The GaN cappedMT has 3-4 orders of magnitude higher leakage than the standard.
.4.its
0 5 10 15 20
0
200
400
600
800
1000
I D (
mA
/mm
)
VD (V)
VD=15V VD=20V
25. The two terminal breakdown voltage is only 20-35V, which the application of high drain biases.
118
CW power was measured at 10GHz to check whether dispersion was reduced
at gigahertz frequencies because gate-lag measurements only provide information at
low frequencies. Fig.4.26 shows the power measurement results at drain biases of
10V and 15V. An output power density of 3.4W/mm with a peak PAE of 32% was
obtained at a drain bias of 15V. At 10V, 2.2W/mm with a peak PAE of 38% was
achieved. These results were very promising. The low values of PAE are clearly due
to early gain compression because of the low breakdown voltage. The standard
HEMTs grown by the same MOCVD machine without SiN passivation showed much
poorer and less performance. The power density of the unpassivated AlGaN/GaN
HEMTs varied from 0 to 2W/mm at a drain bias of 10V, usually falling below
1W/mm. The low breakdown voltage limited the application of higher drain biases so
no further power data was obtained.
0 5 10 15 20 250
5
10
15
20
25
30
2.2W/mm
Pin (dBm)
Pout
(dBm
), G
ain
(dB)
Pout Gain PAE
0
10
20
30
40
50
60
38%
(a)
PAE
(%)
0 5 10 15 20 25
5
10
15
20
25
30
32%
3.4W/mm
PAE
(%)
Pin (dBm)
P out (
dBm
), G
T (dB
)
Pout Gain PAE
0
10
20
30
40
50
60
(b)
Fig.4.26. Power performance at 10GHz of an unpassivated device on a sapphire substrate. (a) Bias conditions: VDS=10V, IDS=230mA/mm. SaturatedPout=2.2W/mm ; peak PAE=38%. Device dimension: 0.7µm×150µm. (a) Bias conditions: VDS=15V, IDS=230mA/mm. Saturated Pout=3.4W/mm ; peak PAE=32%. Device dimension: 0.7µm×150µm.
119
4.4 Summary
A concept of reducing dispersion at epitaxial structure level was proposed.
This approach utilized a thick cap layer to increase the distance between the channel
and surface, thereby decreasing the modulation from the surface. This resulted in the
screening of the surface potential fluctuations which reduced dispersion. Several
structures were investigated. A simple GaN/AlGaN/GaN structure was found not to
produce high pinch-off voltage due to the accumulation of the positive charges at
GaN/AlGaN interface, induced by the polarization effect. A graded AlGaN doped
with Si was inserted between GaN cap and AlGaN layers to remove the unfavorable
positive charges, thereby obtaining the thickness-dependent pinch-off voltage.
Processing issues were discussed. Deep gate recess was necessary to obtain the good
ohmic contacts, and decent pinch-off voltage and transconductance. The results of the
first generation of the GaN-capped AlGaN/GaN HEMT successfully proved the
concept. Dispersion was not observed in 200-ns-pulse-width gate-lag measurements
without SiN passivation. An output power density of 3.4W/mm with a peak PAE of
32% was obtained at a drain bias of 15V at 10GHz from an unpassivated HEMT on a
sapphire substrate. However, the large gate leakage and low breakdown voltage
limited the application of higher drain biases, which was a problem that remained to
be solved.
120
4.5 References [1] M. Asif Khan, A. Bhattarai, J. N. Kuznia, and D. T. Olson, “High electron
mobility transistor based on a GaN-AlxGa1-xN heterojunction,” Applied Physics Letters, vol. 63, no. 9, pp. 1214-1215, Aug. 1993.
[2] M. Asif Khan, J. N. Kuznia, and D. T. Olson, W. J. Schaff and J. W. Burm, M. S.
Shur, “Microwave performance of a 0.25µm gate AIGaN/GaN heterostructure field effect transistor,” Applied Physics Letters, vol. 65, no. 9, pp. 1121-1123, Aug. 1994.
[3] Y.-F. Wu, B.P. Keller, S. Keller, D. Kapolnek, S.P. Denbaars, U.K. Mishra
“Measured microwave power performance of AlGaN/GaN MODFET,” IEEE Electron Device Letters, vol. 17, no. 9, pp. 455-457, Sept. 1996.
[4] Y.-F. Wu, D. Kapolnek, J. Ibbetson, N.-Q. Zhang, P. Parikh, B.P. Keller, U.K.
Mishra “Hi Al-content AlGaN/GaN HEMTs on SiC Substrates With Very High Power Performance,” IEEE International Electron Devices Meeting, Technical Digest, pp.16.7.1-3, 1999.
[5] P. B. Klein, S. C. Binari, K. Ikossi-Anastasiou, A. E. Wickenden, D. D. Koleske,
R. L. Henry, and D. S. Katzer, “Investigation of traps producing current collapse in AlGaN/GaN high electron mobility transistors,” Electron. Lett., Vol. 37, no. 10, pp. 661-662. May 2001.
[6] A. Tarakji, G. Simin, N. Ilinskaya, X. Hu, A. Kumar, A. Koudymov, J. Yang, and
M. Asif Khan, “Mechanism of radio-frequency current collapse in GaN-AlGaN field-effect transistors,” Appl. Phys. Lett., Vol. 78, no. 15, pp. 2169-2171, April 2001.
[7] T. Mizutani, Y. Ohno, M. Akita, S. Kishimoto, and K. Maezawa, “A Study on
Current Collapse in AlGaN/GaN HEMTs Induced by Bias Stress,” IEEE Trans. Electron Dev., Vol. 50, no. 10, pp. 2015-2020, Oct. 2003.
[8] I. Daumiller, D. Theron, C. Gaquiere, A. Vescan, R. Dietrich, A. Wieszt, H. Leier,
R. Vetury, U. K. Mishra, I. P. Smorchkova, S. Keller, N. X. Nguyen, C. Nguyen, and E. Kohn, “Current Instabilities in GaN-Based Devices” IEEE Electron Device Lett., Vol. 22 no. 2, pp. 62-64, Feb. 2001.
[9] S. C. Binari, K. Ikossi-Anastasiou, J. A. Roussos, D. Park, D. D. Koleske, A. E.
Wickenden, and R. L. Henry, “GaN electronic devices for microwave power applications,” Proc. Int. Conf. GaAs Manufacturing Technology, pp. 201-204, 2000.
121
[10] S. C. Binari, P. B. Blein, and T. E. Kazior, “Trapping Effects in GaN and SiC Microwave FETs,” Proc. IEEE, Vol. 90, no. 6, pp. 1048-1058, June 2002.
[11] N. G. Weimann, M. J. Manfra, and T. Wachtler, “Unpassivated AlGaN-GaN
HEMTs with minimal RF dispersion grown by plasma-assisted MBE on semi-insulating 6H-SiC substrates,” IEEE Electron Device Lett., Vol. 24, no. 2, pp. 57-59, Feb. 2003.
[12] M. Manfra, N. Weimann, Y. Baeyens, P. Roux, and D. M. Tennant,
“Unpassivated AlGaN/GaN HEMTs with CW power density of 3.2 W/mm at 25 GHz grown by plasma-assisted MBE,” Ellectron. Lett., Vol. 39, no. 8, pp. 694-695, April 2003.
[13] B. M. Green, K. K. Chu, E. M. Chumbes, J. A. Smart, J. R. Shealy, L. F.
Eastman, “The effect of surface passivation on the microwave characteristics of undoped AlGaN/GaN HEMTs,” IEEE Electron Device Lett., Vol. 21 no. 6, pp. 268-270, June 2000.
[14] J.R. Shealy, V. Kaper, V. Tilak, T. Prunty, J.A. Smart, B. Green and L.F.
Eastman, “An AlGaN/GaN high-electron-mobility transistor with an AlN sub-buffer layer,” J. Phys.: Condens. Matter, 2002, vol. 14, p.3499.
[15] A. V. Vertiatchikh, L. F. Eastman, W. J. Schaff, and T. Prunty, “Effect of surface
passivation of AlGaN/GaN heterostructure field-effect transistor,” Electron. Lett., Vol. 38, no. 8, pp. 388-389, April 2002.
[16] T. R. Prunty, J. A. Smart, E. M. Chumbes, B. K. Ridley, L. F. Eastman, and J. R.
Shealy, “Passivation of AlGaN/GaN heterostructures with silocn nitride for insulated gate transistors,” Proc. IEEE/Cornell High-Performance Devices Conf., pp. 208-214, Aug. 2000.
[17] X. Hu, A. Koudymov, G. Simin, J. Yang, M. Asif Khan, A. Tarakji, M. S. Shur,
and R. Gaska, “Si3N4/AlGaN/GaN-metal-insulator-semiconductor heterostructure field-effect transistors,” Appl. Phys. Lett., Vol. 79, no. 17, pp. 2832-2834, Oct. 2001.
[18] R. Vetury, N. Q. Zhang, S. Keller, and U. K. Mishra, “The impact of surface
states on the DC and RF characteristics of AlGaN/GaN HFETs,” IEEE Trans. Electron. Dev., Vol. 48, no. 3, pp. 560-566, March 2001.
[19] A. Jimenez, D. Buttari, D. Jena, R. Coffie, S. Heikman, N. Q. Zhang, L. Shen, E.
Calleja, E. Munoz, J. Speck, and U. K. Mishra, “Effect of p-doped overlayer thickness on RF-dispersion in GaN junction FETs,” IEEE Electron Device Letters, vol. 23, no. 6, pp. 306-308, June 2002.
122
[20] R. Coffie, D. Buttari, S. Heikman, S. Keller, A. Chini, L. Shen, and U. K.
Mishra, “p-capped GaN-AlGaN-GaN high-electron mobility transistors (HEMTs),” IEEE Electron Device Letters, vol. 23, no. 10, pp. 588-590, Oct. 2002.
[21] R. Coffie, Ph.D. dissertation, “Characterizing and Suppressing DC-to-RF
Dispersion in AlGaN/GaN High Electron Mobility Transistors”, University of Californina, Santa Barbara
[22] M. B. Das, “Charge-control analysis of m.o.s. and junction-gate field-effect
transistors,” Proc. IEE, vol. 113, No. 10, pp. 1565-1570, Oct. 1966. [23] C.-H. Chen, S.Keller, E. D. Haberer, L. Zhang, S. P. DenBaars, E. L. Hu, U. K.
Mishra, and Y.Wu, “Cl reactive ion etching for gate recessing of AlGaN/GaN field-effect transistor,” J. Vac. Sci. Technol. B, vol. 17, pp. 2755–2758, Nov./Dec. 1999.
[24] D. Buttari, A. Chini, T. Palacios, R. Coffie, L. Shen, H. Xing, S. Heikman, L.
McCarthy, A. Chakraborty, S. Keller, and U. K. Mishra, “Origin of etch delay time in Cl2 dry etching of AlGaN/GaN structures,” Appl. Phys. Lett. Vol.83, No.23, pp.4779-4781, Dec.2003.
[25] H. Yu et al., “Low contact resistance to si implnated GaN and its application to
GaN/AlGaN HEMTs,” accepted by Electronic Materials Conference 2004. [26] L. Shen, S. Heikman, B. Moran, R. Coffie, N.-Q. Zhang, D. Buttari, I. P.
Smorchkova, S. Keller, S. P. DenBaars, U. K. Mishra, “AlGaN/AlN/GaN high-power microwave HEMT,” IEEE Electron Device Letters, vol. 22, No. 10, pp.457-459, Oct 2001.
123
Chapter 5 Improvement of GaN-Capped
AlGaN/GaN HEMTs
In the previous chapter, the concept of novel GaN capped AlGaN/GaN HEMT
was discussed and the prototype device was demonstrated successfully. The gate-lag
measurement showed no dispersion up to gate pulse width of 200ns. Output power
density of 3.4W/mm was obtained at 10GHz without SiN passivation on a sapphire
substrate. The initial results were promising. However, a large gate leakage current
and low breakdown voltage prevented the application of higher drain bias, therefore
limiting the output power. In this chapter, the causes of the high leakage and low
breakdown will be discussed. Based on this understanding, several solutions will be
discussed and much better power performance will be reported as a result of their
implementation.
124
5.1 SiO2 insulating layer
The GaN capped AlGaN/GaN HEMTs showed high gate leakage current,
which needed to be reduced. To investigate the nature of the leakage, one of the most
important issues is to locate the leakage path, which not only provides information
about the cause, but also gives the opportunity to solve the leakage problem by
blocking the leakage path. For instance, the leakage can be reduced if the insulating
material is put in an appropriate position in the device.
As shown in fig.4.16, the gate metal directly contacts the etched bottom
surface and sidewall. Either the gate/sidewall or gate/AlGaN interface, or both, can
contribute to the gate leakage. An experiment was designed to locate the leakage
path. Shown in fig.5.1, two devices with different gate configurations were
investigated. One of them had a gate which did not contact the sidewalls: a 1.3µm
long recess was etched first, then a 0.7µm long metal gate was deposited with a
spacing about 0.3µm between gate metal and the sidewalls. Therefore the gate
leakage could go through the etched bottom surface only. The second device had a
normal
125
(b)
GaN GaN G
AlGaN
(a)
GaN GaNG
AlGaN
Fig.5.1. Schematics of GaN/AlGaN/GaN HEMT cross-section. (a) Gate doesn’t contact sidewalls. 0.3µm spacing is left. (b) Gate contacts the sidewalls.
gate recessing with a 0.7µm opening. A re-aligned T-shaped gate was formed to
ensure the complete coverage of the sidewalls. If the leakage through the etched
bottom surface was dominant, these two devices should have shown similar leakage
current. Otherwise, if the major contribution was through the sidewall, the device
with the T-shape gate should have much higher leakage. The measured gate-drain
leakage currents are shown in fig.5.2. The gate leakage of the device with the gate
contacting the sidewalls was on the order of 10mA/mm, one order of magnitude
higher than that of the one without contacting the sidewalls. This result supported the
speculation that the leakage contributions from the sidewalls were dominant.
Moreover, the voltage bias between the gate and drain was much higher than that
between the gate and source in typical HEMT operation, so the majority of the gate
leakage current should flow through the sidewall of the drain side.
126
Figdevdevof m
Although
fact that most of
to propose a solu
path. There has
leakage in electro
been used undern
also successfully
shows the schem
chosen as the ins
sidewall towards
proved to be the
transconductance
semiconductor if
-20 -15 -10 -5 0
0.1
1
10
100
I G (m
A/m
m)
VGD (V)
Touching sidewall Not touching sidewall
.5.2. Two terminal gate-drain gate leakage current of theices with and without gate contacting the sidewalls. Theice with the gate touching the sidewalls showed one orderagnitude higher gate leakage.
the exact causes of the large leakage still needed to be clarified, the
the leakage went through sidewall of the drain side made it possible
tion that involved using an insulating material to block the leakage
been a long history of using insulating materials to reduce gate
nic devices. Even for GaN-based HEMTs, insulating materials have
eath the gate to decrease gate leakage [1] [2]. A similar method was
applied to the p-GaN capped device by Robert Coffie [3]. Fig.5.3
atic cross-section of the device with insulating material. SiO2 was
ulator. Unlike the normal MISFET, the insulating material is on the
the drain, instead of on the bottom surface, because the sidewall was
predominant leakage path It also avoids the problems of decreased
and potential interface states between the insulator and
there was SiO2 underneath the gate.
127
G DS
AlGaN
S.I. GaN
GaNGaN G DS
AlGaN
S.I. GaN
GaNGaN
SiO2 (40~50nm thick)SiO2 (40~50nm thick)
Fig.5.3. Schematic cross-section of the GaN-capped HEMTwith SiO2 layer on the sidewall. SiO2 insulator is used to blockthe leakage current path, therefore reducing leakage andimproving breakdown.
Additional processing steps are needed to implement this feature. To
minimize the amount of additional processing, the SiO2 deposition is done right after
gate recessing and before gate metallization. There is then no additional lithography
needed. SiO2 is deposited by Ebeam deposition and the thickness is usually around
40-50nm. This is a low temperature processing step and does not damage the
photoresist. Therefore the following gate metal deposition can share the same
lithography. Because only the sidewall of the drain side needs to be covered by SiO2,
the sample has to be tilted during deposition. Fig.5.4 shows the schematic of the
deposition. The sample is tilted to an angle so that the sidewall of the source side and
the bottom are shadowed by the photoresist. The tilt angle θ is determined by the
thickness of the photoresist and the length of the gate opening:
128
tan gLt
θ = (5.1)
where t is the photoresist thickness and Lg is gate length. For instance, if the
photoresist is 0.7µm think and gate length is 0.7µm too, the tilted angle is 45 degree.
SiO2 Source
SamplePR
PRSiO2
SiO2 Source
SamplePR
PR
SamplePR
PR
SamplePR
PRSiO2
θ
Fig.5.4. The sample is tilted to an angle during the SiO2deposition so that the sidewall of the source side and thebottom are shadowed by the photoresist.
The correct angle is important. A smaller or larger tilt angle can lead to the
incomplete coverage of the sidewall or extra coverage of the bottom. Ideally, only the
sidewall of the drain side should be covered and there should be no SiO2 left on the
bottom surface. However, it is very difficult to control the deposition so accurately.
To ensure the complete coverage of the sidewall, the tilt angle is set a little larger,
which can result in a small coverage of the bottom surface. The effect of the extra
coverage will be discussed later.
129
SiO2
Fig.5.5. The SEM image of the device with SiO2 insulating layer.
The epitaxial structure is as same as that in fig.4.15. The samples were grown
on sapphire substrates. The devices with SiO2 insulator covering the sidewall were
processed and measured. DC and gate-lag I-V characteristics are shown in fig.5.6.
There was still no dispersion observed up to 200ns. A current density of 1.2A/mm
was measured at gate bias of +1V at 200ns-gate-pulse-width measurement. The
higher current density of the pulsed-mode current was due to the severe self-heating
because of the poor thermal conductivity of the sapphire substrates.
130
0 5 10 15
0
200
400
600
800
1000
1200
VG = +1 V
Vp = -7.5 V
VG: -9 -> +1 V
I D (m
A/m
m)
VD (V)
DC 80us 200ns
Fig.5.6. DC and gate-lag I-V characteristics of the GaN-cappedAlGaN/GaN HEMT with SiO2 insulating layer. No dispersion wasobserved up to gate pulse width of 200ns.
Gate leakage was reduced and breakdown voltage was increased greatly.
Fig.5.7 shows the two terminal gate-drain leakage current of the HEMTs with and
without SiO2 insulating layer. The leakage current was reduced by more than one
order of magnitude, from 10mA/mm to 0.5mA/mm. Moreover, the two-terminal gate-
drain destructive breakdown voltage increased from 25-35V to more than 90V. This
ensured that much higher drain bias could be applied. The introduction of SiO2 was
proved very effective.
131
-20 -15 -10 -5
0.1
1
10
100
I G (mA/
mm
)
VGD (V)
Without SiO2 With SiO2
Fig.5.7. Two terminal gate-drain leakage current of the HEMTswith and without SiO2 insulating layer. The leakage was reducedmore than one order of magnitude by the introduction of SiO2.
The small signal performance of the devices was also evaluated. Fig.5.8
displays the current gain (h21) and Mason’s unilateral power gain (UPG) as a
function of frequency of a device with 0.7µm gate length. Current-gain and power-
gain cutoff frequencies (ft and fmax, respectively) of 21GHz and 39GHz were obtained
at the drain current of 280 mA/mm and drain bias of 15 V. These numbers were very
close to the typical values of 0.7um-gate-length standard AlGaN/GaN HEMTs,
indicating that the in roduction of SiO2 did not affect the device RF performance.
Fig.5.8. Cuas a functCurrent-ga39GHz wer
t
1 10 10
10
20
30
40
00
h21,
UPG
(dB)
Frequency (GHz)
h21 UPG
rrent gain (h21) and Mason’s unilateral power gain (UPG) ion of frequency of a device with 0.7µm gate length. in and power-gain cutoff frequencies of 21GHz and e obtained.
132
Because higher drain bias could be applied, the power performance was
improved greatly. The uncooled CW power performance of a 0.7µm×150µm HEMT
at 10GHz is shown in fig.5.9. Drain bias of 25V was applied to the device, instead of
the 15V to the device without the SiO2 insulator. A saturated power of 4.8W/mm was
achieved without SiN passivation, compared to the 3.4W/mm from the device without
SiO2. Peak PAE was 33% at which point the output power density was 4.2W/mm.
Fig.5.9. Power performance at 10GHz of an unpassivated deviceon sapphire substrate. Bias conditions: VDS=20V, IDS=200mA/mm.Saturated Pout=4.8W/mm ; peak PAE=33%. Device dimension:0.7µm×150µm.
0 5 10 15 200
5
10
15
20
25
30
35
Gain: 10dB
4.8W/mm
4.2W/mm
PAE
(%)
Pin (dBm)
Pout
(dBm
), G
ain
(dB)
Pout Gain PAE
0
10
20
30
40
Devices were also grown on SiC substrates. Since higher drain bias could be
applied, better power performance due to the good thermal conductivity of SiC was
expected. The epitaxial structure was slightly different from the previous one, as
shown in fig.5.10. An Al composition of 22% was used, instead of 33%. The low Al
mole fraction, therefore lower 2DEG density, showed higher breakdown
performance. The AlGaN structure was changed to 10nm graded AlxGa1-xN:Si(x=0-
0.22) / 40nm Al0.22Ga0.78N. Si doping density was 8×1018 cm-3, about 80% of the
133
polarization charge density in order to keep the Fermi level in the middle of the band
gap. Room temperature sheet charge density and Hall mobility were 8×1012 cm-2 and
2000 cm2/V-s, respectively.
0.7 nm AlN
UID GaN
Substrate
250 nm UID GaN
40 nm Al0.22Ga0.78N
10 nm graded AlGaN: Si
Fig.5.10. Epitaxial structure of the GaN-capped AlGaN/GaNHEMT with 40nm Al0.22Ga0.78N layer and 10nm graded AlxGa1-xN(x=0-0.22) layer doped by Si.
With the application of the higher drain bias to the devices on SiC substrates,
much better output power performance was achieved, shown in fig.5.11. At 4GHz,
output power density of 12W/mm was obtained with a peak PAE of 44% at the drain
bias of 50V. Further measurement at 10GHz demonstrated its excellent power
performance again: 12W/mm with a peak PAE of 41%, as shown in fig.5.12. In both
cases, device dimensions were 0.7µm×150µm. They were biased under class AB with
a DC drain current of 270mA/mm. It should be noted that there was no surface
passivation employed. 12W/mm was the highest output power density among all
GaN-based devices for several months. Although this record was broken by the
134
application of a field plate to non-GaN-capped (that is, standard) devices, it is still the
highest output power density without SiN passivation.
0 5 10 15 20
10
15
20
25
30
35
44%
12W/mm
PAE
(%)
Pin (dBm)
Pout
(dB
m),
Gai
n (d
B) Pout
Gain PAE
0
10
20
30
40
50
Fig.5.11. Power performance at 4GHz of an unpassivated deviceon SiC substrate. Bias conditions: VDS=50V, IDS=270mA/mm.Saturated Pout=12W/mm ; peak PAE=44%. Device dimension:0.7µm×150µm.
Fig.5.1on SiCSatura0.7µm
0 5 10 15 20 25 30
5
10
15
20
25
30
35
41.8%
12W/mm
PAE
(%)
Pin (dBm)
Pout
(dB
m),
Gai
n (d
B) Pout
Gain PAE
0
10
20
30
40
50
60
2. Power performance at 10GHz of an unpassivated device substrate. Bias conditions: VDS=45V, IDS=270mA/mm.
ted Pout=12W/mm ; peak PAE=42%. Device dimension:×150µm.
135
Power performance as a function of drain bias was also measured to
investigate the dispersion control. In fig.5.13, it can be seen that the PAE remained
relatively constant, between 43% and 50%, over the bias from 30 to 50V, indicating
minimal trapping in the device [4].
30 35 40 45 50
6
8
10
12
P out (W
/mm
)
Drain Bias (V)
PAE
(%)
0
10
20
30
40
50
60
70
Th
performan
density of
been show
affected t
may intro
difficult f
surface. T
on the dev
Fig.5.13. PAE and output power density as a function of drainbias. The relatively constant PAE indicated minimal trapingeffects.
e coverage of SiO2 on the bottom surface could affect the device
ce, because the interface between SiO2 and semiconductor may have a high
traps, depending on the deposition techniques. Ebeam-deposited SiO2 has
n to have a high density of traps at the interface [5]. Their existence
he modulation of the channel. For instance, the slow response of the traps
duce the similar phenomenon as dispersion. As discussed above, it was
or the processing to just cover the sidewall without touching the bottom
herefore, the effect of the coverage length of the bottom surface of the gate
ice performance was investigated. The different coverage length of SiO2 on
136
the bottom was obtained by setting different tilt angles during Ebeam deposition. The
coverage profiles were measured by Atomic Force Microscopy (AFM). Fig.5.14
shows the actual profiles after SiO2 deposition (no gate metal) and schematics of the
device structures. In fig.5.14(a), the coverage of the SiO2 on the bottom surface was
about 0.3µm, which was about half of the gate length. The gate-lag measurement for
this device is shown in fig.5.15(a) and a small but observable knee voltage walkout at
gate pulse width of 200ns was observed. This indicated that some dispersion
occurred. If the coverage was minimal and most of the gate region was not covered,
as shown in fig.5.14(b), the knee voltage walkout disappeared(fig.5.15(b)). However,
the power performance did not show much difference between these two cases at high
frequency. This was because 0.3µm was still short so that the dispersion was not
severe. It was consistent with the gate-lag measurement where only a small amount of
dispersion was observed.
(
Gate opening
SiO2
G
AlGaN
S.I. GaN
GaN GaN G
AlGaN
S.I. GaN
GaN GaN G
AlGaN
S.I. GaN
GaN GaN
a)
137
0
0 5 10 15
200
400
600
800
1000
1200
VG = +1 V
Vp = -7.
VG: -9 -> +1 V
I D (m
A/m
m)
VD (V)
DC 80us 200n
Fig.5.14. Actual SiO2 coveschematics. (a) 0.3µm SiO2SiO2 coverage on the bottom
SiO2
(a) Fig.5.15. DC and gate-lag
AlGaN/GaN HEMT with (small amount of knee voltSiO2 coverage on the bottgate pulse width of 200ns.
G
AlGaN
S.I. GaN
GaN GaN G
AlGaN
S.I. GaN
GaN GaN G
AlGaN
S.I. GaN
GaN GaN
(b)
rage profile measured by AFM andcoverage on the bottom (b) minimal.
400
600
800
1000
1200
VG = +1 VVG: -9 -> +1 V
I D (m
A/m
m)
Vp = -7.5 V
VD (V)
DC 80us 200ns
0 5 10 15
0
200
5 V
s
(b)
I-V characteristics of the GaN-cappeda) 0.3µm SiO2 coverage on the bottom,age walkout was observed. (b) minimalom, no dispersion was observed up to
138
This method can be extended to smaller gate length devices by using some
other techniques. Due to the smaller opening, SiO2 deposition directly by Ebeam does
not work for the sub-micron devices. However, there are other methods available, for
example, an internal sidewall formation using PECVD SiO2 deposition with dry
etching of SiO2 on the gate region could be used. Therefore the concept itself can be
scaled down.
At the same time, new techniques can be employed to improve the SiO2
quality. For example, Jet Vapor Deposition (JVD) was found to be effective to reduce
the interface trap density [6]. It may decrease the SiO2-induced dispersion.
However, there are some disadvantages. The processing becomes more
complicated because of the additional SiO2 deposition. The introduction of SiO2 may
also lead to potential reliability problem because it can trap hot electrons.
139
5.2 Effects of Si doping sheet density
In the previous section, the method of using SiO2 to block the leakage path
was discussed. The leakage was reduced successfully and high output power density
was achieved. However, the more complicated processing and the potential reliability
problem made pursuing the reduction of the leakage without SiO2 insulator
worthwhile.
Although SiO2 reduced the leakage, the actual reason for the high leakage still
remained unknown. Gate recessing has been considered as a major contributor,
because the exposed surfaces could be damaged by the accelerated ions during the
RIE etch [7]. To clarify whether deep etching was the main cause of leakage, two
different devices were processed and measured, as shown in fig.5.16. One was a
GaN-capped AlGaN/GaN HEMT with gate recessing, so that the gate contacted the
sidewalls and bottom surface. The other was a device without gate recessing, i.e. the
gate was deposited on the top of the GaN cap directly. Because there was no RIE
etching surface in the second device, the leakage current difference between the two
devices should give some clues as to whether RIE etching played an important role.
(a) (b)
AlGaN
G GaNGaNAlGaN
G GaNGaN
AlGaN
GGaN
AlGaN
GGaN
Fig.5.16. (a) Normal recessing device. Gate contacts the bottom surface andsidewalls. (b) Device with no gate recessing. There was no etched surface.
140
The two terminal gate-drain leakage current densities of these two devices
were displayed in fig.5.17. It could be found that the gate-drain leakage currents of
both devices approached a similar value, about 10mA/mm, at high gate-drain bias.
The discrepancy at low bias will be explained later. Since there was no dry etching
for the second device, the result clearly showed that etching-induced damage was not
the major reason r the high leakage.
Fig.5surfano et
Therefore
GaN cap was in
shown in fig.5.18
be described as:
where the
fo
-50 -40 -30 -20 -10 010-3
10-2
10-1
100
101
I G (m
A/m
m)
VGD (V)
Unrecessed device Recessed device
.17. (a) Normal recessing device. Gate contacts the bottomce and sidewalls. (b) Device with no gate recessing. There wasched surface.
, some ‘internal’ reasons had to be checked. The electric field in the
vestigated first. A schematic of the cross-section of the HEMT is
. Recalling the discussion in the last chapter, the electric field could
0
( )Si scap
q nE σεε
−= (5.2)
σSi is the Si doping sheet density and ns is the 2DEG density.
141
The maximum value is reached when the channel is depleted, i.e. ns=0:
,max0
Sicap
qE σεε
= (5.3)
G D
Graded AlGaN: Si(σSi)
S.I. GaN
GaN
AlGaN
- - - - - - - - - 2DEGDepletedchannel
A
B
+ + + + + + + + + + + + + +
GG D
Graded AlGaN: Si(σSi)
S.I. GaN
GaN
AlGaN
- - - - - - - - - 2DEGDepletedchannel
A
B
+ + + + + + + + + + + + + +
E =qσSi/εE =qσSi/ε
Fig.5.18. Schematic cross-section of the device. When the channelwas depleted, the electric field in the cap was determined by the Sidoping sheet density.
The electric field exceeded 1MV/cm when the Si density was higher than
5.5×1012cm-2. This high electric field was a possible reason for the high leakage. At
the same time, the Schottky barrier height on GaN is only 0.9V, lower than the barrier
height of about 1.5V on Al0.3Ga0.7N. This also exacerbated the leakage and
breakdown problems.
However, the maximum electric field in the GaN cap only occurred when the
channel in the drain access region was depleted. The pinch-off voltage of the drain
access region was more than 50V due to the thick cap layer. The leakage was already
very high (>1mA/mm) when the gate-drain bias was only 10V. The reason was that
142
the sidewall etched by RIE etching was not perfectly vertical. There was a small
angle, as shown in fig.5.18 (not to scale). Because the pinch-off voltage was
determined by the cap layer thickness, the corresponding pinch-off voltage at point A
was much less than that at point B. Therefore, the electric field already reached the
maximum value at point A when the gate-drain bias is only 10-20V, resulting in high
leakage current.
This also can explain the discrepancy of the leakage currents between devices
with and without gate recessing, as shown in fig.5.17. For a device without recessing,
the pinch-off voltage at every point was large. Therefore, the electric field in the cap
was low when the gate-drain bias was low, according to equitation (5.2). Only when
the 2DEG density in the channel was low at high drain bias, did the electric field
become high enough to increase the leakage current. In fig.5.17, it can be seen that
the leakage of the unrecessed device is close to that of recessed device above 40V.
Experiments were designed to verify this proposed theory. From the
discussion above, it was obvious to speculate that a lowering Si doping sheet density
could reduce the leakage since the maximum electric field is determined by the Si
doping, as shown in equation (5.3). A series of samples with different Si doping sheet
density was grown by MOCVD on sapphire substrates. The epitaxial structures were
similar to the previous ones, consisting of a 250nm UID GaN cap layer, 10nm graded
AlGaN layer doped by Si, a 40nm AlGaN layer and a SI GaN buffer, shown in
fig.5.10. The Si sheet densities of 3×1012cm-2, 5×1012cm-2, and 7×1012cm-2 were
chosen. The measured Hall data are listed in table 5.1. An interesting phenomenon
143
was observed: when the Si doping sheet density decreased from 7×1012cm-2 to
3×1012cm-2, the decrease in 2DEG density did not follow the same trend, saturating at
about 6×1012cm-2. It was similar to the phenomenon observed when the cap thickness
changed in the last chapter. In fact, the mechanisms behind both were the same: the
accumulation of positive charges in the graded AlGaN region, which will be
discussed in the following section.
Table 5.1 Measured and simulated charge densities
Si doping sheet density
(1012cm-2)
ns,Hall (1012cm-2)
ns,simulated (1012cm-2)
ps,simulated (1012cm-2)
3.0
6
6.4
3.4
5.0
6.2
6.5
1.5
7.0
7.3
7.0
0
Simulation was employed again to investigate the structures. Fig.5.19 displays
the simulated band diagrams and the simulated charge densities are also listed in table
5.1. When the Si doping density was 7×1012cm-2, the Fermi level was between the
conduction and valence band. There was not any charge accumulation in the graded
AlGaN region. All the Si dopants ionized and the 2DEG density in the channel was
equal to Si doping sheet density, which was verified by a Hall measurement.
However, when the Si density decreased to 5×1012cm-2, the simulated 2DEG density
is 6.5×1012cm-2. Among the electrons, 5×1012cm-2 still came from the donors while
144
the other 1.5×1012cm-2 was balanced by induced positive charges in the graded
AlGaN region. In the simulation those charges are holes. This was because when the
2DEG density decreased, the electric field in the AlGaN region increased, leading to
the contact of the valance band with the Fermi level and the accumulation of the
mobile positive charges, as shown in fig.5.19(b). Therefore, the decrease of the 2DEG
concentration stopped, the process being screened by accumulated holes. The
simulated 2DEG density agreed with the experimental data very well. The same thing
happened in the case of Si doping density of 3×1012cm-2 too, fig.5.19(c), except that
more mobile positive charges (~3×1012cm-2) were accumulated in the graded AlGaN
region. The 2DEG density was still around 6×1012cm-2. A comparison of the
simulated 2DEG and hole concentration along with the measured 2DEG
concentration is shown in fig.5.19(d).
145
-50 0 50 100 150 200 250 300 350-5-4-3-2-101234
Ene
rgy
(eV)
Thickness (nm)
Holes
2DEG
-50 0 50 100 150 200 250 300 350
-8
-6
-4
-2
0
2
4
+sSi
+ps
-sAlGaN -ns
+sAlGaN Ene
rgy
(eV)
Thickness (nm)
(a) (b) 2 )
-50 0 50 100 150 200 250 300 350
-5-4-3-2-101234
Ene
rgy
(eV)
Thickness (nm)
(c) Fig.5.19. Simulated band diagram oof (a) 3×1012cm-2; (b) 5×1012cm-2; experimental data of the 2DEG function of Si doping sheet densitlow, valence band in the graded AlGleading to the accumulation of the p
The trend of 2DEG concentration as
be divided into two parts.
When the Si doping sheet density
2DEG density is very close to the Si doping
graded AlGaN region does not contact wi
charge accumulation in the region. σSi0 ca
14
y
6
0 2 4 6 8 100
2
4
6
8
10
positive charges
2DEG
2DEG
and
hol
e de
nsiti
es (1
012cm
-
Si doping sheet density (1012cm-2)
(d)
f the devices with Si doping density(c) 7×1012cm-2; (d) simulation andand mobile positive charges as a. When the Si doping density wasaN region touched the Fermi level,
ositive charges.
a function of Si doping sheet density can
is higher than a specific value σSi0, the
sheet density σSi. The valence band in the
th the Fermi level. There is no positive
n be obtained by calculating the 2DEG
density when the valence band at the graded AlGaN/AlGaN interface just contacts
with the Fermi level (induced positive charge concentration is still zero).
0 01 2 , 1 ,
01 2 0 1
AlGaN AlN g AlGaN AlGaN g AlGaN
Si
t t E t Eq q
t t d t
εε εσ σ σσ
⋅ + ⋅ − ∆ ⋅ − ∆= ≈
+ +
ε
(5.4)
where σSi is the Si doping sheet density; σAlGaN is the net polarization charge density
of the AlGaN/GaN interface; σAlN is the net polarization charge density of the
AlN/GaN interface; t1 is the thickness of AlGaN layer; t2 is the thickness of AlN
layer; ∆Eg is the band gap of the AlGaN.
When σSi >= σSi0,
s Sin σ≈ (5.5)
when Si doping density is lower than σSi0, the 2DEG density remains at the
value of σSi0. In this case, the valence band in the graded region contacts with the
Fermi level, resulting in the accumulation of positive charges. These induced positive
charges balanced part of the 2DEGs while the others are provided by Si dopants.
when σSi < σSi0
0Sisn σ= (5.6)
Fig.5.19(d) shows the simulation and experimental data of the 2DEG and
mobile positive charges as a function of Si doping sheet density. Two different
147
dependences of carrier concentration on Si doping sheet density are shown by dashed
lines in the figure. They agreed with the experimental results very well.
However, although the measured 2DEG densities were the same when the Si
doping density was below a specific value, the pinch-off voltages were different.
Referring to the band diagram in fig.5.20 when the channel is depleted, the pinch-off
voltage Vp can be given as:
1 2 3 20
1[ ( ) (2p Si AlGaN AlGaN
qV t t t tσ σ σε ε
= ⋅ + + ⋅ − ⋅ −⋅
)]Siσ (5.7)
where σSi is the Si doping sheet density, σAlGaN is the net polarization charge
density at the AlGaN/GaN interface, t1 is the thickness of the GaN cap layer, t2 is the
thickness of the graded AlGaN layer, and t3 is the thickness of the AlGaN layer.
0 50 100 150 200 250 300 350
-10
0
10
20
30
40
+sSi-sAlGaN
+sAlGaN
Ene
rgy
(eV)
Ev
Ec
AlGaN
GradedAlGaN
GaN
Thickness (nm)
Fig.5.20. Simulated band diagram of the GaN-capped AlGaN/GaN HEMT when pinched off. Pinch-off voltage is proportional to the thickness of GaN cap and the Si doping sheet density.
148
Therefore, the pinch-off voltage was still a linear function of the Si doping
sheet density assuming all other parameters given. This led to an interesting result:
the samples with same 2DEG density had different pinch-off voltages. Fig.5.21 shows
the measured pinch-off voltages and 2DEG densities of the samples with different Si
sheet densities. The results confirmed the theoretical prediction. Although both
unrecessed devices with Si density of 3×1012cm-2 and 5×1012cm-2 had the same 2DEG
density of 6×1012cm-2, they had different pinch-off voltages: 19 and 30V,
respectively.
3 4 5 6 7
10
20
30
40
2D
EG D
ensi
ty (1
012cm
-2)
Measured Vp 2DEG Density
Pin
ch-o
ff vo
ltage
(V)
Si doping (1012cm-2)
6
8
10
Fig.5.21. The measured pinch-off voltage and 2DEG density as afunction of Si doping sheet density. Although 2DEG densitystopped reduction, pinch-off voltage still decreased.
As discussed previously, the pinch-off voltage gave an indication of the
dispersion control ability. The higher the pinch-off voltage was, the better the
dispersion reduction, because less charges in the channel in the drain access region
149
were depleted given a gate-drain bias. This implied that lower Si doping density could
affect the dispersion control.
HEMTs were fabricated and measured. Processing was similar to that
described in the last chapter and no SiO2 insulating layer was used. The two terminal
gate-drain leakage currents were shown in fig.5.22. As expected, the leakage current
decreases when the Si doping sheet density was reduced: from 3mA/mm (Si doping
density of 7×1012cm-2) to 0.3mA/mm (Si doping density of 3×1012cm-2). If compared
to more than 10mA/mm (Si doping density of 1.2×1013cm-2), the reduction was about
1-2 orders of magnitude. At the same time, the two terminal gate-drain destructive
breakdown voltage increased from 40-50V (Si of 7×1012cm-2) to more than 100V (Si
of 3×1012cm-2). Therefore, lowering the Si doping sheet density successfully reduced
the gate leakage current and improved the breakdown voltage.
However, the price paid was that the dispersion control ability was weakened,
because the low Si doping density resulted in low pinch-off voltage in drain access
region. Gate-lag measurements were performed and shown in fig.5.23. It could be
-25 -20 -15 -1010-2
10-1
100
101
102
Si doping sheet density
3x1012cm-25x1012cm-2
1.2x1013cm-2
7x1012cm-2
I G (m
A/m
m)
VGD (V)Fig.5.22. Two terminal gate-drain leakage current density. The leakage was reduced 1-2 orders of magnitude when Si doping density decreased from 1.2×1013cm-2 to 3×1012cm-2.
150
found that there was a small knee voltage walkout at gate pulse width of 200ns when
the Si doping density was reduced to 3×1012cm-2. No dispersion was observed in the
gate-lag measurements in the devices with the Si doping densities of 5×1012cm-2 and
7×1012cm-2. Another disadvantage of the low Si doping density was that the
maximum drain current was lowered. The devices with 3×1012cm-2 and 5×1012cm-2
showed a current density of around 800mA/mm at gate bias of +1V while the one of
7×1012cm-2 had 1A/mm.
0 5 10 15
0
200
400
600
800
1000
Vp = -7 V
VG: -8 -> +1 V
I D (m
A/m
m)
VD (V)
DC 80us 200ns
0 5 10 15
0
200
400
600
800
1000
Vp = -3.5 V
VG: -9 -> +1 VI D (
mA
/mm
)
V (V)
DC 80us 200ns
D
0 5 10 15
0
200
400
600
800
1000
Vp = -7 V
VG: -9 -> +1 V
I D (m
A/m
m)
VD (V)
DC 80us 200ns
(a) (b)
(c)
Fig.5.23. DC and gate-lag I-V characteristics of GaN-capped AlGaN/GaN HEMTs with Si doping density of (a) 3×1012cm-2; (b) 5×1012cm-2; (c) 7×1012cm-2. Small amount of dispersion appeared when Si doping densitywas 3×1012cm-2.
151
CW power measurements were performed at 4GHz to check the dispersion
reduction at high frequency. Fig.5.24 shows the power performance of a device with
the Si doping density of 5×1012cm-2 on sapphire substrate. An output power density of
7.7W/mm without SiN passivation was obtained with a peak PAE of 62%. DC drain
bias was 40V and drain current was 100mA/mm. The device was biased to deep
class-AB mode. Considering that the substrate was sapphire, this was a good power
number. Fig.5.25 shows the output power density and PAE as a function of drain bias.
The PAE remained relatively constant over the bias of 20 to 40V, implying that the
dispersion was suppressed well [4].
0 5 10 15 200
5
10
15
20
25
30
62%
7.7W/mm
PAE
(%)
Pin (dBm)
Pout
(dB
m),
Gai
n (d
B) Pout
Gain PAE
0
10
20
30
40
50
60
70
80
Fig.5.24. Power performance at 4GHz of an unpassivated devicewith Si density of 5×1012cm-2 on sapphire substrate. Biasconditions: VDS=40V, IDS=100mA/mm. Saturated Pout=7.7W/mm ;peak PAE=62%. Device dimension: 0.7µm×150µm.
152
20 25 30 35 402
4
6
8
PAE
(%)
VD (V)
Out
put p
ower
Den
sity
(W/m
m)
Output power density (W/mm) PAE (%)
40
50
60
70
80
Fig.5.25. PAE and output power density as a function of drain biasof an unpassivated device with a Si doping density of 5×1012cm-2
on sapphire substrate. The relatively constant PAE indicated good
As a comparison, the device with a Si doping density of 3×1012cm-2 displayed
poorer power performance. As shown in fig.5.26, an output power density of
6.4W/mm with the PAE of 55% was obtained. The DC bias condition was the same
as that of the sample with a 5×1012cm-2 Si doping density. Both the power density and
PAE were lower. Fig.5.27 shows the PAE as a function of drain bias. An obvious
drop of PAE was observed at drain bias of 40V, implying that dispersion occurred.
This is not surprising since the device with a Si density of 3×1012cm-2 had the lowest
pinch-off voltage in drain access region as discussed above. The dispersion affected
the power and PAE, especially at high drain bias.
153
0 5 10 15 200
5
10
15
20
25
30
55%
6.4W/mm
PAE
(%)
Pin (dBm)
Pout
(dB
m),
Gai
n (d
B) Pout
Gain PAE
0
10
20
30
40
50
60
70
80
20 25 30 35 400
2
4
6
PAE
(%)
VD (V)
Out
put p
ower
Den
sity
(W/m
m)
Output power density (W/mm) PAE (%)
40
50
60
70
80
Fig.5.26. Power performance at 4GHz of an unpassivated devicewith Si density of 3×1012cm-2 on sapphire substrate. Biasconditions: VDS=40V, IDS=100mA/mm. Saturated Pout=6.4W/mm ;peak PAE=55%. Device dimension: 0.7µm×150µm.
Fig.5.27. PAE and output power density as a function of drain biasof an unpassivated device with Si density of 3×1012cm-2 onsapphire substrate. The drop of PAE at drain bias of 40V implieddispersion.
The device with a Si doping density of 7x1012cm-2 has relatively lower
breakdown voltage of 50V. Therefore, only a drain bias of 25V could be applied. The
low breakdown voltage limited the output power density.
154
5.3 Thick graded AlGaN capped AlGaN/GaN
HEMTs
In the previous section, the causes of the high gate leakage and low
breakdown were discussed. The sheet density of the Si doping was found to be a
critical factor. The high Si doping density could result in a high electric field in the
GaN cap when the channel was depleted. The experiments on a series of HEMTs with
different Si doping density proved that a lower Si doping density reduced the gate
leakage and improved breakdown voltage. However, lowering Si doping density also
lowered the pinch-off voltage of the drain access region, which weakened the
dispersion reduction. Therefore, a solution which reduces the leakage while retaining
a relatively high Si doping density is a preferable solution.
AlGaN was considered as the material for the cap layer again because AlGaN
can sustain higher electric field than GaN. Moreover, the Schottky barrier height is
higher as well. These properties are desirable for the reduction of leakage. However,
the dispersion reduction also requires a thick cap. Due to the nature of the
pseudomorphic growth, the thickness of the bulk AlGaN layer is limited to be less
than 50-60nm, which cannot meet the thickness requirement of dispersion reduction.
To avoid this problem, a graded AlGaN cap layer epitaxial structure was
introduced, as shown in fig.5.28. Compared to bulk AlGaN, the Al composition of the
graded AlGaN decreased along the growth direction. This resulted in reduced strain,
155
allowing the graded AlGaN to be grown much thicker than bulk AlGaN without
cracking.
0.7 nm AlN
UID GaNSapphire Substrate
250 nm graded AlxGa1-xN
(x=0.22 to 0.05): Si
40 nm Al0.22Ga0.78N
Fig.5.28. Epitaxial structure of the graded AlGaN cappedAlGaN/GaN HEMT with 250nm graded AlxGa1-xN (x=0.05-0.22)cap layer and 40nm Al0.22Ga0.78N layer.
In this new epitaxial structure, no GaN cap layer was employed. The AlGaN
layer was graded the full 250nm and from an initial maximum Al composition of 22%
to 5%. It was uniformly doped with Si. The doping density was chosen so that the
negative polarization charges were completely compensated. Given the parameters in
fig.5.28, the Si doping density was about 3.3×1017cm-3. The simulated band diagram
is shown in fig.5.29. The energy band of the graded AlGaN layer was linear because
the polarization charges were compensated by ionized donors and the background
doping density was only about 1016cm-3. The flatness of the energy band implied that
156
the electric field in the graded layer at zero bias was very small. The simulated 2DEG
density was about 9×1012cm-2.
0 5 1 1 200 25 3 35-1
T0 00 50 0 00 0
0
-8
-6
-4
-2
0
2
+sSi
-sAlGaN-ns
+sAlGaN Ene
rgy
(eV)
hickness (nm)
GaN
AlGaNGraded AlGaN
Fig.5.29. Simulated band diagram of the graded AlGaN cappedHEMT. Graded AlGaN layer was doped with Si to compensate thenegative polarization charges.
The sample was grown by MOCVD on a sapphire substrate. A 250nm graded
AlGaN cap layer was grown successfully without any cracking. A Hall measurement
gave a carrier density of 9×1012cm-2 at room temperature, which agreed with the
calculation very well.
The processing steps were the same as those mentioned before. The only
difference in the epi-structure was that the cap layer was AlGaN, instead of GaN.
Some references reported that the etching rate of AlGaN was slower than that of GaN
[8]. However, the latest investigations revealed that this apparent difference was due
to the surface oxidation [9]. After a pre-etching treatment of one minute BCl3 etching
in the RIE,, the following BCl3/Cl2 etching rate of AlGaN was similar to that of GaN.
157
The gate-lag pulsed I-V characteristics were measured and are shown in
fig.5.30. No dispersion was observed up to 200ns-gate-pulse-width. A current density
of 1A/mm was obtained at gate bias of +1V.
0 2 4 6 8 10 12 14 16
0
200
400
600
800
1000
∆VG = 1 V VG = +1 V
Vp = -6.5 V
I D (m
A/m
m)
VD (V)
DC 80us 200ns
Fig.5.30. DC and gate-lag I-V characteristics of the graded AlGaN capped AlGaN/GaN HEMT. No dispersion was observed up togate pulse width of 200ns. Current density of 1A/mm was obtainedat gate bias of +1V.
Good leakage and breakdown performances were achieved. Fig.5.31 shows
the two terminal gate-drain leakage currents of the different devices. The thick graded
AlGaN cap HEMT had the lowest leakage. It was one order of magnitude lower than
those of devices with SiO2 or a Si doping density of 3×1012cm-2, three orders of
magnitude lower than that of the device with Si doping density of 1.2×1013cm-2.
Moreover, the two terminal breakdown voltage exceeded 120V, which allowed the
application of high drain bias.
158
Fdw
CW
the HEMT
and drain c
with a PAE
GaN-based
highest nu
matter whe
displayed i
PAE dropp
Although t
channel co
bias, and d
-20 -15 -10 -5 010-5
10-4
10-3
10-2
10-1
100
101
102
DC
A
B
I G (m
A/m
m)
VGD (V) ig.5.31. Two terminal gate-drain leakage current. (a) Si doping
ensity of 1.2×1013cm-2. (b) Si doping density of 3×1012cm-2. (c)ith SiO2 insulating layer. (d) with thick graded AlGaN cap.
power was measured at 4GHz. Fig.5.32 shows the power performance of
on a sapphire substrate. The device was biased at a drain voltage of 50V
urrent of 100mA/mm, deep class-AB mode. Output power of 8.5W/mm
of 57% was achieved. This was the highest reported power density of the
HEMT without SiN passivation on a sapphire substrate. It was also the
mber among the devices on sapphire substrates without a field plate, no
ther SiN passivation was applied. The PAE as a function of drain bias was
n fig.5.33. The PAE remained relatively constant up to 40V. After that, the
ed from 62% to 57%. Dispersion was one of the reasons for this decrease.
he pinch-off voltage of the drain access region was larger than 50V, the
uld still be depleted when the voltage swing was also large at high drain
ispersion occurred. Another reason for the decrease in PAE was because of
159
the bad matching of the load pull. At high drain bias, the load matching point tends to
be at the edge of the Smith chart. Our present load pull setup can not provide
appropriate match ng.
Fig.5of anabovmatc
i
0 5 10 15 200
5
10
15
20
25
30
35
57%
8.5W/mm
PAE
(%)
Pin (dBm)
Pout
(dB
m),
Gai
n (d
B) Pout
Gain PAE
010
20
30
40
50
60
70
80
Fig.5.32. Power performance at 4GHz of an unpassivated deviceon sapphire substrate. Bias conditions: VDS=50V, IDS=100mA/mm.Saturated Pout=8.5W/mm ; peak PAE=57%. Device dimension:0.7µm×150µm.
20 25 30 35 40 45 502
4
6
8
10
PAE
(%)
VD (V)
Out
put p
ower
Den
sity
(W/m
m)
Output power density (W/mm) PAE (%)
40
50
60
70
80
.33. PAE and output power density as a function of drain bias unpassivated device on sapphire substrate. The drop of PAE e drain bias of 40V was due to both dispersion and badhing.
160
Power was also measured at 10GHz, as shown in fig.5.34. The bias condition
employed was a drain bias of 40V and drain current of 70mA/mm. Only 5.9W/mm
with a peak PAE of 40% was obtained on sapphire substrates. A similar output power
density was usually observed on devices on SiC substrates at both 4GHz and 10GHz
because our standard 0.7um-gate-length HEMTs had current-gain and power-gain
cutoff frequencies of 20GHz and 45GHz, respectively. Normally there is not a big
difference in power density at 4GHz and 10GHz whereas the PAE drops when the
frequency increases because of reduced gain. However, the substrate was sapphire in
this case, which has a lower thermal conductivity than SiC. When the PAE dropped at
higher frequencies, self-heating became more and more severe. This caused obvious
power degradation at 10GHz. If a SiC substrate is used, similar output power
densities at 4 and 10 GHz can be expected.
0 5 10 15 20 25
5
10
15
20
25
30
40%
5.9W/mm
PAE
(%)
Pin (dBm)
Pout
(dB
m),
Gai
n (d
B) Pout
Gain PAE
0
10
20
30
40
50
60
Fig.5.34. Power performance at 10GHz of an unpassivated deviceon sapphire substrate. Bias conditions: VDS=40V, IDS=70mA/mm.Saturated Pout=5.9W/mm ; peak PAE=40%. Device dimension:0.7µm×150µm.
161
Growth on a SiC substrate was also tried. However, the sample cracked due to
the large stress. Growth optimization needs to be done in the future.
162
5.4 Summary
In this chapter, solutions of reducing the gate leakage current and improving
the breakdown voltage were discussed and implemented. Much better power
performances were obtained.
The gate recess sidewall of the drain side was found to be the major path for
leakage. A SiO2 insulating layer was deposited on the sidewall to block the leakage
path. Gate leakage was reduced by more than one order of magnitude and the two
terminal breakdown voltage increased from about 30V to more than 90V. Output
power density of 12W/mm was achieved at 10GHz on SiC, which was the record
number for GaN-based HEMTs without surface passivation. The effect of the
coverage of SiO2 on the gate region was also investigated. Although long coverage
length (~0.3µm) increased the dispersion, its effect on the power performance was
very limited. This fact relieved the strict requirements on processing.
The effect of the Si doping sheet density on the leakage current was
investigated as well. The high electric field in the GaN cap caused by the ionized
donors was found to be the major contributor to the high leakage current. Lowering Si
doping density effectively reduced the leakage current and increased the breakdown
voltage. However, low Si doping density decreased the pinch-off voltage as well,
which was not good for dispersion control. Si doping sheet density of 5×1012cm-2 was
chosen to compromise these two conflicting requirements. An output power density
163
of 7.7W/mm with a peak PAE of 62% at 4GHz was obtained from an unpassivated
HEMT on a sapphire substrate at drain bias of 40V.
The solution of employing a thick (~250nm) graded AlGaN as the cap layer
was also discussed. The advantage of this approach was to reduce the leakage while
retaining the relatively high 2DEG density because AlGaN could sustain higher
electric field than GaN. 2DEG density of 9×1012cm-2 was measured. Two terminal
gate-drain leakage current was reduced by 2 to 3 orders of magnitude, and two
terminal gate-drain destructive breakdown voltage of more than 120V was achieved
as well. An output power density of 8.5W/mm with a peak PAE of 57% at 4GHz was
obtained from an unpassivated HEMT on a sapphire substrate at drain bias of 50V,
which was the highest output power density of the unpassivated GaN-based HEMTs
on sapphire substrates.
All of these solutions can effectively reduce the gate leakage current and
increase the breakdown voltage. However, they have their own advantages and
disadvantages. The employment of a SiO2 insulating layer on the sidewall allows the
high carrier concentration. But the processing is more complicated and the
introduction of SiO2 may induce potential reliability problem. The method with low
Si doping sheet density is SiO2-free while the carrier concentration is limited which
degrades the dispersion control ability. The utilization of a thick graded doped AlGaN
layer can reduce the leakage while retaining a relatively high carrier concentration,
which is the most promising among these methods. But the growth optimization of
the thick graded AlGaN layer on SiC substrate needs to be done as the future work.
164
5.5 References [1] M. Asif Khan, X. Hu, G. Sumin, A. Lunev, J. Yang, R. Gaska, M.S. Shur,
“AlGaN/GaN metal oxide semiconductor heterostructure field effect transistor,” IEEE Electron Devices Letters, vol. 21, no. 2, pp. 63-65, Feb. 2000.
[2] X. Hu, A. Koudymov, G. Simin, J. Yang, M. Asif Khan, A. Tarakji, M.S. Shur, R.
Gaska, “Si3N4/AlGaN/GaN-metal-insulator-semiconductor heterostructure field-effect transistors,” Applied Physics Letters, vol.79, no.17, pp.2832-4, Oct. 2001.
[3] R. Coffie, L. Shen, G. Parish, A. Chini, D. Buttari, S. Heikman, S. Keller, U.K.
Mishra, “Unpassivated p-GaN/AlGaN/GaN HEMTs with 7.1 W/mm at 10 GHz,” IEE Electronics Letters, vol.39, no.19, pp.1419-20. Sept. 2003
[4] Y.-F. Wu,, P.M. Chavarkar, M. Moore, P. Parikh, U.K. Mishra, “Bias-dependent
performance of high-power AlGaN/GaN HEMTs,” IEEE International Electron Devices Meeting, Technical Digest, pp.17.2.1-3, 2001.
[5] N.-Q. Zhang, B. Moran, S.P. DenBaars, U.K. Mishra, X.W. Wang, T.P. Ma,
“Effects of surface traps on breakdown voltage and switching speed of GaN power switching HEMTs,” IEEE International Electron Devices Meeting, Technical Digest, pp.25.5.1-4, 2001.
[6] B. Gaffey, L.J. Guido, X.W. Wang, T.P. Ma, “High-quality oxide/nitride/oxide
gate insulator for GaN MIS structures,” IEEE Transactions on Electron Devices, Vol.48, pp.458-464, 2000.
[7] C.-H. Chen, S.Keller, E. D. Haberer, L. Zhang, S. P. DenBaars, E. L. Hu, U. K.
Mishra, and Y.Wu, “Cl reactive ion etching for gate recessing of AlGaN/GaN field-effect transistor,” J. Vac. Sci. Technol. B, vol. 17, pp. 2755–2758, Nov./Dec. 1999.
[8] S. A. Smith, C. A. Wolden, M. D. Bremser, A. D. Hanser, and R. F. Davis, “High
rate and selective etching of GaN, AlGaN, and AlN using an inductively coupled plasma,” Appl. Phys. Lett. Vol.71, pp.3631-3633, Dec.1997.
[9] D. Buttari, A. Chini, T. Palacios, R. Coffie, L. Shen, H. Xing, S. Heikman, L.
McCarthy, A. Chakraborty, S. Keller, and U. K. Mishra, “Origin of etch delay time in Cl2 dry etching of AlGaN/GaN structures,” Appl. Phys. Lett. Vol.83, No.23, pp.4779-4781, Dec.2003.
165
Chapter 6
Summary, conclusions and future work
6.1 Summary and conclusions
This dissertation has focused on the development of the novel epitaxial
structures to improve the device performance. Two novel devices, AlGaN/AlN/GaN
HEMT and GaN/AlGaN/GaN HEMT, were analyzed, fabricated and characterized.
Carrier transport properties were improved and dispersion was reduced effectively.
High power GaN-based HEMTs have been successfully demonstrated with minimal
low temperature and room temperature dispersion without surface passivation, relying
only on the use of strategic band engineering and the utilization of polarization
charge.
AlN and the improvement of 2DEG transport properties
The scattering mechanisms of the 2DEG in the AlGaN/GaN HEMT were
reviewed. Alloy disorder scattering was found to be one of the dominant factors at
low temperature. Unlike in GaAs-based HEMT, it also plays an important role at
room temperature due to the high carrier concentration in the channel.
An approach of utilizing binary material AlN as the barrier layer to reduce the
alloy disorder scattering was discussed. AlN/GaN heterostructure was investigated by
166
Smorchkova et al. [1]. Because of the high polarization charge in the AlN and the
absence of alloy disorder scattering, both high 2DEG density and high electron
mobility were obtained, resulting in a low sheet resistance of 180Ω/ٱ.
In order to incorporate the advantage of the AlN into conventional
AlGaN/GaN HEMT, a novel AlGaN/AlN/GaN HEMT with a thin (~1nm) AlN
interfacial layer was investigated [2] [3]. The introduction of the thin AlN not only
removed the alloy disorder scattering at the interface, but also reduced electron
wavefuction penetration into the AlGaN barrier, resulting in improved electron
mobility. The larger effective ∆Ec due to the high polarization field in AlN also
improved the carrier concentration at the same time. An AlGaN/AlN/GaN with a
2DEG density of 1.22×1013cm-2 and an electron mobility of 1520cm2/v s was grown,
better than 1.1×1013cm-2 and an electron mobility of 1200cm2/v s from a conventional
AlGaN/GaN HEMT. Si-doping in AlGaN barrier further improved carrier density to
1.48×1013cm-2 while retaining electron mobility above 1500cm2/v s. An output power
density of 8.5W/mm was achieved at 8GHz from a passivated HEMT on a SiC
substrate.
DC-to-RF dispersion
Dispersions of different samples were characterized by gate-lag measurement
as a function of temperature. The DC currents increased due to the higher electron
velocity when temperature decreased, while the pulsed current showed an
unpredictable trend which either increased, or decreased, or remained relatively
167
constant, depending on the sample. However, dispersion became worse at lower
temperature in all the cases if the ratio between pulsed and DC current at DC knee
voltage was considered. This seemingly ‘random’ behavior of pulsed current could be
explained by considering both dispersion and channel velocity. At lower temperature,
dispersion was more severe, which decreased channel charge under the drain access
region. On the other hand, channel velocity increased. Depending on which of these
two competing factor is dominant, an increase or decrease in pulsed current could be
observed as temperature was reduced.
Two dispersion models based on the concept of surface virtual gate were
discussed: de-trapping and band-conduction model and hopping conduction model.
The first one required the trap release electron to conduction band and then electron
move back to the gate, while in the latter one electron moves by hopping from one
trap state to another. Both can explain the worse dispersion at lower temperature.
Further investigation needs to be done to distinguish these two.
GaN/AlGaN/GaN HEMTs
In order to overcome the drawback of the SiN passivation, research to reduce
the dispersion at epitaxial level was motivated. Based on the virtual gate model, some
solutions to reduce the surface modulation to the channel were discussed. The
concept of thick GaN cap on top of standard AlGaN/GaN HEMT was the focus of
this dissertation. This approach utilized a thick cap layer to increase the distance
between the channel and surface, thereby decreasing the modulation from the surface.
168
This resulted in the screening of the surface potential fluctuations which reduced
dispersion. A simple conceptual GaN/AlGaN/GaN heterostructure was investigated.
A decrease of the carrier density with increase in the GaN cap thickness followed by
a saturation was observed and was explained by the accumulation of the induced
positive charges in the GaN/AlGaN interface. However, it could not produce high
pinch-off voltage if the positive charges at the interface are mobile. A graded AlGaN
doped with Si was inserted between GaN cap and AlGaN layers to remove the
unfavorable positive charges, thereby obtaining the thickness-dependent pinch-off
voltage. Processing issues were discussed. Deep gate recess was necessary to obtain
good ohmic contacts, and decent pinch-off voltage and transconductance. The results
of the first generation of the GaN-capped AlGaN/GaN HEMT successfully proved
the concept. Dispersion was not observed in 200-ns-pulse-width gate-lag
measurements without SiN passivation. An output power density of 3.4W/mm with a
peak PAE of 32% was obtained at a drain bias of 15V at 10GHz from an unpassivated
HEMT on a sapphire substrate. However, the large gate leakage and low breakdown
voltage limited the application of higher drain biases, which was a problem that
remained to be solved [4].
Several methods were discussed to solve the high gate leakage and low
breakdown voltage problems.
The etched sidewall towards drain was found to be the major path of the gate
leakage. A SiO2 insulating layer was deposited on the sidewall to block the leakage
path. It effectively reduced the leakage current and increased the breakdown voltage.
169
The application of the high drain bias and the reduced dispersion resulted in a record
output power density of 12W/mm at drain bias of 45V at 10GHz on a SiC substrate
without SiN passivation [5]. The investigation of the effect of the coverage of SiO2 on
the gate region revealed that although long coverage length (~0.3µm) could degrade
dispersion reduction, its effect was very limited.
The effect of the Si doping sheet density on the leakage current was
investigated as well. The high electric field in the GaN cap caused by the ionized
donors was found to be the major contributor to the high leakage current. Lowering Si
doping density effectively reduced the leakage current and increased the breakdown
voltage. However, low Si doping density decreased the pinch-off voltage as well,
which was not good for dispersion control. The compromise between these two
conflicting requirements led to an optimized Si doping sheet density of 5×1012cm-2.
An output power density of 7.7W/mm with a peak PAE of 62% at 4GHz was
obtained from an unpassivated HEMT on a sapphire substrate at drain bias of 40V.
A variation of the structure which employed a thick (~250nm) graded AlGaN
layer to replace the GaN cap was investigated. The higher critical electric field of the
AlGaN allowed the higher Si doping sheet density, which resulted in higher carrier
concentration, while retaining low gate leakage. An output power density of
8.5W/mm with a peak PAE of 57% at 4GHz was obtained from an unpassivated
HEMT on a sapphire substrate at drain bias of 50V, which was the highest output
power density of the unpassivated GaN-based HEMTs on sapphire substrates [6].
170
6.2 Future work
The thick GaN (or graded AlGaN) capped AlGaN/GaN HEMTs were
successfully demonstrated. Highest output power density on both sapphire and SiC
substrates without SiN passivation was achieved. However, it is still under
development. More work can be pursued in different aspects to improve the device
performance in the future.
Growth
The HEMT with a thick graded AlGaN layer is the most promising candidate
among several structures. However, only the device grown on sapphire substrate has
been demonstrated so far. There is no device on SiC substrate yet. Due to the
different strain situation between sapphire and SiC substrates, the present growth on
SiC substrate resulted in cracking of the sample. In order to achieve higher output
power density, SiC is desirable. This is especially important for higher frequency,
because the drop of PAE at higher frequency can result in self-heating problem which
is much more severe on sapphire substrate than that on SiC substrate. Therefore, the
optimization of the growth of thick graded AlGaN on SiC substrate needs to be
investigated in the future.
171
Processing
The present gate recessing used in the processing is controlled by time,
therefore requiring accurate etch rate control which is difficult to achieve. The
variation of the etching rate due to unstable chamber condition lowers the yield.
Recently, etch-stop technique has been developed by Dario Buttari. BCl3/SF6 was
found to have a high etching selectivity, up to 29, between GaN and AlGaN. The
selectivity is expected to be even higher after the optimization. By adopting etch-stop
technique, the strict requirement to the gate recessing can be relived and the yield can
be improved. Some new epitaxial designs may be necessary to utilize the property
efficiently. For example, a thin AlN layer can be inserted into the structure as an
effective etch-stop layer.
Ion implantation was proved to be an effective way to improve ohmic contacts
[7]. A high density of implanted donors can lower the contact resistance. Moreover, it
reduces the strict requirement to accurate etching rate control for good ohmic contact,
thereby increasing the yield.
Device structure
In 2003, the application of field plates (fig.6.1(a)) made a tremendous impact
on the output power performance of AlGaN/GaN HEMTs [8]-[10]. More than
30W/mm was achieved at 8GHz which was more than double the highest reported
number for HEMTs without field plate. Field plate not only increases the breakdown
voltage, but also reduces the dispersion. The reduced electric field around the gate is
172
considered as one of the major causes of the improvement of the dispersion, because
the lowered electric field may decrease the electron injection to the surface states.
The same concept can be applied to the thick GaN-capped AlGaN/GaN
HEMTs, as shown in fig.6.1(b), referred as epitaxial field plate. The initial two-
dimension device simulation by Atlas showed that the electric field peak at the corner
of gate on the drain side was lower than that without field plate. The initial result of
adding a field plate to the thick graded AlGaN capped AlGaN/GaN HEMT was
promising. Output power density of EpitaxialField PlateEpitaxialField Plate
s
F
f
G DS G DS
Field PlateSiN
AlGaNGaN
G DS G DS
Field PlateSiN
G DS G DDSS
Field PlateSiN
AlGaNGaN
AlGaNGaN
S DG GaN
AlGaNGaN
SS DDG GaN
(a) (b)
Fig.6.1. (a) Field plate on standard AlGaN/GaN HEMT; (b) Epitaxialfield plate on GaN/AlGaN/GaN HEMT.
9.2W/mm with a PAE of 64% from an unpassivated HEMT on sapphire
ubstrate was obtained, as shown in fig.6.2, better than that without field plate.
urther investigations are needed to explore the epitaxial field plate technique in the
uture.
173
Induced positiv
In chapter 4
simulation. Howev
charges so far. Th
which they can res
that some of pos
investigation of th
frequency can prov
0 5 10 15 20
15
20
25
30
35
64%
9.2W/mm
PAE
(%)
Pin (dBm)
Pout
(dB
m),
Gai
n (d
B) Pout
Gain PAE
10
20
30
40
50
60
70
80
Fig.6.2. Power performance at 4GHz of an unpassivated gradedAlGaN capped AlGaN/GaN HEMT with field plate on sapphiresubstrate. Bias conditions: VDS=45V, IDS=100mA/mm. SaturatedPout=9.2W/mm ; peak PAE=64%. Device dimension:0.7µm×150µm.
e charge
, the appearance of the induced positive charges was discussed by
er, there has been no direct experimental evidence of these positive
eir properties need to be investigated, for example, the speed at
ponse. The higher-than-expected DC pinch-off voltage may imply
itive charges can response fast enough while some not. The
e transconductance or gate-source capacitance as a function of
ide more information of these charges.
174
6.3 References [1] I. P. Smorchkova, S. Keller, S. Heikman, B. Heying, P. Fini, J. S. Speck, and U.
K. Mishra, “Two-dimensional electron-gas AlN/GaN heterostructures with extremely thin AlN barriers,” Appl. Phys. Lett., vol. 77, No. 24, pp. 3998-4000, Dec. 2000.
[2] I. P. Smorchkova, L. Chen, T. Mates, L. Shen, S. Heikman, B. Moran, S. Keller,
S. P. DenBaars, J. S. Speck, and U. K. Mishra, “AlN/GaN and (Al,Ga)N/AlN/GaN two-dimensional electron gas structures grown by plasma-assisted molecular-beam epitaxy,” J. Appl. Phys., Vol. 90, No. 10, pp.5196-5201, Nov. 2001.
[3] L. Shen, S. Heikman, B. Moran, R. Coffie, N. Q. Zhang, D. Buttari, I. P.
Smorchkova, S. Keller, S. P. DenBaars, and U. K. Mishra, “AlGaN/AlN/GaN high-power microwave HEMT,” IEEE Electron. Dev.Lett., vol. 22, no. 10, pp. 457-459, Oct. 2001.
[4] L. Shen, R. Coffie, D. Buttari, S. Heikman, A. Chakraborty, A. Chini, S. Keller, S.
P. DenBaars and U. K. Mishra, “Unpassivated GaN/AlGaN/GaN power HEMTs with dispersion controlled by epitaxial layer design”, Journal of Electronic Materials, be published in June 2004.
[5] L. Shen, R. Coffie, D. Buttari, S. Heikman, A. Chakraborty, A. Chini, S. Keller, S.
P. DenBaars and U. K. Mishra, “High-power polarization-engineered GaN/AlGaN/GaN HEMTs without surface passivation”, IEEE Electron Device Letters, Vol.25, No.1, pp.7-9, Jan. 2004.
[6] L. Shen, D. Buttari, S. Heikman, A. Chini, R. Coffie, A. Chakraborty, S. Keller, S.
P. DenBaars and U. K. Mishra, “Improved high power thick-GaN-capped AlGaN/GaN HEMTs without surface passivation”, accepted by 62nd Device Research Conferenc, June 2004.
[7] H. Yu et al., “Low contact resistance to si implnated GaN and its application to
GaN/AlGaN HEMTs,” accepted by Electronic Materials Conference 2004. [8] Y. Ando, et al.: ‘10W/mm AlGaNGaN HFET with a field modulating plate’,
IEEE Electron Device Lett., Vol.24, No. 5, pp. 289–291, 2003 [9] A. Chini, D. Buttari, R. Coffie, L. Shen, S. Heikman, A. Chakraborty, S. Keller,
U. K. Mishra, “Power and Linearity Characteristics of Field-Plated Recessed-Gate AlGaN-GaN HEMTs,” Electron Device Letters, Accepted for future publication, 2004
175
[10] Y.-F. Wu, A. Saxler, M. Moore, R. P. Smith, S. Sheppard, P.M. Chavarkar, T.
Wisleder, U. K. Mishra, P. Parikh, “30-W/mm GaN HEMTs by Field Plate Optimization,” IEEE Electron Device Letters, Vol. 25, No. 3, pp.117-119, March 2004.
176