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