Optoelectronic Devices: III Nitrides

Preface Wide bandgap Ill-nitrides differ from the rest of III-V compound semiconductors in many aspects. They cover a b...

Author: Mohamed Henini | M Razeghi

205 downloads 3139 Views 34MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form

DOWNLOAD PDF

Preface Wide bandgap Ill-nitrides differ from the rest of III-V compound semiconductors in many aspects. They cover a broad range of optical spectrum including the ultraviolet portion; their most thermodynamically stable structure is wurtzite; they exhibit strong piezoelectric fields; they are chemically and physically strong, etc. In addition to these inherent peculiarities, there are many technical challenges associated with the growth and processing of these materials. Most specifically, lack of commercially available native substrates has forced the researchers to use substrates with mismatched lattice constants and thermal expansion coefficients. New ideas have had to be developed in order to reduce the number of dislocations present in the resulting material grown atop these substrates. Doping problems also, especially p-type doping, had plagued the development of these materials for a long time. All these substantial differences have separated Ill-nitrides from the rest of III-V materials and they are generally discussed independently. In the past few years, due to the great number of potential applications (Figure 1), Ill-nitrides have been studied vigorously and as a result significant advancements have taken place in this field. InGaN-based blue LEDs have found their market in full color displays and white lighting; blue laser diodes will be used in the next generation highcapacity CD/DVD read/write systems; UV LEDs and photodetectors have seen tremendous improvements in their properties and it will not be long before they are commercially available to be used for detection of biochemical agents, early missile threat warning, air/water purification, or even higher quality white lighting. Also these materials

High capacity optical data storage

Early missile threat warning

Detection of biochemical agents

Water purification

White lighting

Power line monitoring

Figure 1. Some of the potential applications of Ill-nitrides.

Mobile wireless communications

I

Flame monitoring

vi

Preface

have been successfully utilized to constitute high-frequency high-power transistor for RF transmission purposes. The aim of this book is to gather some of the most crucial information on properties and applications of Ill-nitrides in one place in the form of a handbook for the benefit of the nitride community. The book chapters cover a broad range of subjects from the material growth technology to devices such as light emitters, photodetectors, and high-power highfrequency applications. In addition, some of the other interesting properties of Ill-nitrides such as ferromagnetism, negative differential resistance, and phonon/electron-phonon interactions will be discussed. Several distinguished scientists have contributed their knowledge to this book and they have tried not only to present a historical review of the Ill-nitrides, but also to offer up-to-date advancements and future directions towards better understanding of these materials. Manijeh Razeghi Northwestern University, USA Mohamed Henini Nottingham University, UK

Optoelectronic Devices: Ill-Nitrides M. Razeghi and M. Henini (Eds.) © 2004 Elsevier Ltd. All rights reserved.

Chapter 1

Introduction M. Razeghi^ and M. Henini'' ^Department of Electrical and Computer Engineering, Center for Quantum Devices, Northwestern University, Cook Room 4051, 2220 Campus Drive, Evanston, IL 60208-3129, USA ^School of Physics and Astronomy, University of Nottingham, NG7 2RD, UK

1.1.

INTRODUCTION

Wide bandgap Ill-nitrides, including (AI,Ga,In)-N, have seen enormous success in their development especially in the latest stages of the 20th century. Many substantial problems had to be overcome before these materials could constitute useful devices. High density of dislocations due to the lack of lattice-matched substrates and low doping efficiency were the most challenging problems that researchers in this area had to face. At the beginning, it was hard to believe that a material with a dislocation density in the order of 10^10^^ cm~^ would become the building block of many viable devices. However, thanks to the hard work of researches in this field, today blue/violet light-emitting diodes and laser diodes based on (Al,In,Ga)-N have been successfully commercialized. Blue/green LEDs have already found their market in full-color LCD displays and traffic lights, while blue LDs are expected to shortly replace red lasers in the current CD/DVD read/write systems. The unique properties of Ill-nitrides lead to a range of applications from optoelectronic devices to high-power electronics. The wide bandgap of GaN makes this material suitable not only for light emitting source but also for high-temperature applications. GaN and its alloys have the potential to form high power electronics such as transistors or thyristors. UV solar-blind photodetectors based on AlGaN have been demonstrated by several groups [1]. These detectors have applications in early missile threat detection and interception, chemical and biological threat detection, UV flame monitoring, and UV environmental monitoring. Due to the polar nature of the Ga-N bond, GaN does not possess inversion symmetry. Thus, when GaN is subject to an alternating electric field, the induced polarization is not symmetric. This property of GaN can be used in non-linear optics applications such as second-harmonic generation [2]. The same lack of inversion symmetry results in a huge piezoelectric field. There are some other conceivable

E-mail addresses: [email protected] (M. Razeghi), [email protected] (M. Henini).

2

Optoelectronic Devices: Ill-Nitrides

applications for Ill-nitrides such as surface acoustic wave generation [3], acousto-optic modulator [4], and devices that utilize negative electron affinity [5]. Group Ill-nitride materials are different from some conventional semiconductors such as silicon (Si) and GaAs in a sense that under ambient conditions, the thermodynamically stable structure is wurtzite. Although zinc blende structure for GaN or InN could exist by forcing the film to grow on {001} crystal planes of cubic substrates, the intrinsic tendency of Ill-nitrides is to form a wurzite structure with a hexagonal symmetry. The large difference in electronegativity between the group III elements and nitrogen (Al = 1.18, Ga = 1.13, In = 0.99, N = 3.0) leads to very strong chemical bonds in Ill-nitride material system which together with a wide direct energy gap is the origin of some interesting properties of Ill-nitrides [6]. Ill-nitrides have a bandgap energy tunable from 6.2 eV for AIN to 3.4 eV for GaN to 2 eV or below for InN (The energy gap of InN has been the subject of many debates and it is not agreed upon yet. There have been some recent reports on the bandgap energy of InN being as narrow as 0.7 eV) [7,8]. This corresponds to a wavelength range of 200-650 nm or higher, covering a broad spectral range, from UV to visible. Figure 1.1 demonstrates where Ill-nitride compound semiconductors stand compared to other semiconductors in the space of lattice constant-bandgap energy.

ZnS ZnO

0

•

Direct Bandgap

•

Indirect Bandgap

MgSe

A

C UJ

CdTe

InN* InSb — I —

3.0

I

3.5

4.0

I

1

4.5

1

1

-

5.0

7.0

Lattice Constant (A)

Figure 1.1. Bandgap energy vs. lattice constant of Ill-nitride family and some of the other ffl-V and II-VI compound semiconductors. Note: Bandgap of InN has been reported to be as narrow as 0.7 eV (InN* on the graph).

Introduction Table 1.1. Some of the important physical parameters of Ill-nitrides and sapphire Parameter Lattice constant, a (A) Lattice constant, c (A) Energy gap (eV) Thermal conductivity (W/cm K) Thermal expansion coefficient. a, (10-^K-^) Thermal expansion coefficient, a, (10"^K"^) Electron effective mass, m* (mg) Heavy hole effective mass, m^h (mo) Dielectric constant (so)

GaN

AIN

3.189 5.186 3.44 1.3 4.3 (17-477°C)

3.112 4.982 6.2 2.0 5.27 (20-800°C)

4.0 (17-477°C)

4.15 (20-800°C)

0.2 0.8 8.9

0.4 3.5 8.5

InN 3.545 5.703 1.9

AI2O3

4.758 12.991

-

-

0.3

5.6 (280°C)

-

3.8 (280°C)

7.7 (20-500°C)

0.11 1.6 15.3

-

Direct and indirect bandgap semiconductors are represented by triangles and circles, respectively. The direct bandgap of Ill-nitrides is one of their most beneficial features for optoelectronic device applications. In addition, the wide bandgap of Ill-nitrides results in a low intrinsic carrier density which in turn leads to low leakage and low dark current, especially important for photodetectors and high-temperature electronics. It is also known that Ill-nitrides are very robust materials with high melting points and mechanical strength. Adding to the list the ability to resist radiation damage yields a material system suitable for high-frequency, high-power and high-temperature applications. Some of the physical properties of Ill-nitrides are listed in Table 1.1. The data are complied from Refs. [9-11]. Some of the parameters for sapphire (the most widely used substrate for growth of Ill-nitrides) are also listed in this table. Owing to their direct bandgap, Ill-nitrides are promising candidates for optical devices such as laser diodes (LDs), light-emitting diodes (LEDs), and photodetectors. Not long ago, blue/violet LDs with lifetime of over 10,000 h were successfully demonstrated [12] and consequently commercialized. Implementation of blue LDs into CD/DVD read/write systems in computers will increase the data storage capacity by five times. This enhancement will become even larger when using UV LDs instead. The diameter of the focused laser beam is directly proportional to the laser wavelength. A, and inversely proportional to the numerical aperture, NA, of the imaging lens (Figure 1.2). The spot area is the square of the spot diameter:

A oc

VNAJ


' Spot size Figure 1.2.

Illustration of a compact disc and a focused laser beam for read/write purpose.

The maximum areal density is the number of bits per spot, b, per spot area, A:

It has been several years since GaN/InGaN blue LEDs found their market in automotive dashboard lighting, full-color LED displays, indicator lighting, and LCD back lighting (Figure 1.3). The success in mass production of these highly efficient high-brightness InGaN/GaN blue LEDs basically put an end to II-VI (ZnSe/ZnS) blue LED research. UV solar-blind photodetectors have also been demonstrated in AlGaN/GaN material systems. Early missile threat warning, detection of chemical/biological agents, flame detection, power line monitoring, and non-line-of-sight (NLOS) communication (when combined with UV light emitters) are among the most important applications of solarblind photodetectors. In addition to optical devices, a lot of advances have taken place in the area of highpower, high-frequency power transistors for RF transmission applications, thanks to the high thermal conductivity, high melting point, low dielectric constant, and high breakdown voltage of Ill-nitrides. GaN HEMTs can alleviate many of the problems associated with the current LDMOS technology for mobile wireless communications due to their inherently higher transconductance, good thermal management and higher cut-off frequencies. The GaN HEMT technology aims at applications such as ship-board, airborne and ground RADARs, high-performance space electronics, base station transmitters, C-band Satcom, Ku-K band VSAT and broadband satellites, LMDS and digital radio. GaN HEMTs are capable of generating high power per unit area that translates into smaller devices with higher impedances. Also their high breakdown voltage enables them to operate at high voltages eliminating the need for voltage conversion. In addition, due to their high electron velocity, high-frequency operation is possible. These

Introduction

Figure 1.3. GaN-based blue/green LEDs have been used for traffic lights, LCD back lighting, and full-color LED displays (image of US cellular field in Chicago, courtesy of www.diamond-vision.com).

properties combined with the direct nature of their bandgap have made the GaN family a promising candidate for high-power high-frequency electronics. Although not studied in as much details as InGaN-based blue light emitters, AlGaNbased UV LEDs have also been investigated by several groups and emitters with wavelengths as short as 267 runs have been demonstrated [13]. There are numerous applications for ultraviolet light emitters. White LEDs for efficient, low-cost lighting, high-density optical data storage, water purification, portable chemical and biological agent detection/analysis systems, and NLOS communication are a few of them. Currently available incoherent UV light sources include high-intensity arc discharge lamps and fluorescent lamps. Table 1.2 summarizes advantages and disadvantages of these two technologies along with those of Ill-nitride-based UV LEDs. By combining UV LEDs and UV photodetectors, a compact system that can immediately identify certain biological agents is possible. In principle, many biological agents fluoresce when excited by a UV light source. The fluorescence emission is normally several nanometers longer than the excitation wavelength. For instance. Figure 1.4 shows the absorption and emission spectra of Phenylalanine. In this example, a UV light source with a monochromatic emission at 257 nm along with a UV photodetector with a band pass-like detection spectra peaked at around 280 nm can form a system capable of detecting this particular agent (Figure 1.5). A series of such

Optoelectronic Devices: Ill-Nitrides Table 1.2. Advantages and disadvantages of (Al,In,Ga)-N based UV LEDs compared to the existing technologies Technology

Disadvantages

Advantages

High-intensity arc discharge lamps

High power

Fluorescent lamps

Efficient

Ill-nitride UV LEDs

Compact, low power consumption, efficient, low turn-on, easy for system integration, resistant to radiation damage and magnetic fields, light, inexpensive

Requires bulky power supply, short lifetime, high power consumption Linear sources (hard to focus their light) Immature technology

LED/photodetector sub-systems each tuned at a particular wavelength of interest can be used to detect a range of biological agents. There is a need for a secure means to send messages in the field using low-power communication systems, a requirement that cannot be met with conventional RF radios. Covert communication is made possible by UV sources that exploit the solar-blind region

257.5 nm

eHjN^ ^coê 279.5 nm

Phenylalanine Absorption Fluorescence

*0^4Am I

300

I

400

>ii"^"»» »*'

I

500

600

700

Wavelength (nm) Figure 1.4.

Absorption and fluorescence curves of phenylalanine peak at 257.5 and 279.5 nm, respectively.

Introduction UV photodetector

UV light

Cloud of unknown agent Figure 1.5.

^^^

^

"

Fluorescence

Basic representation of a system for detection of biochemical hazards.

located at 280 nm and below. In this region of the spectrum, the terrestrial solar flux is essentially zero, and the very low background can be used for NLOS communications over distances up to 250 m. The strong extinction coefficient (high scattering, high absorption) of the signal in the UV makes it difficult to detect these emissions from a distance, particularly in the forward direction. The success of such a portable UV frequency communicator unit depends on the availability of a compact, powerful and energy-efficient optical source. Blue/UV LEDs are also currently being utilized to generate white light. LEDproduced white light is cheap, durable, and more efficient than currently used light sources. One of the important applications of white LEDs is in medical illuminators. Most surgical systems are based on incandescent Ught bulbs. These light sources are simple to use, easily dimmable, and have a low initial cost. However, they are also sensitive to shock and vibration, sensitive to voltage variations, have short lifetime, consume too much power, and are inefficient (10-20 ImAV). On the other hand, Hght sources based on LED technology have long lifetime, are compact, efficient, and durable, and require low maintenance. There are three common methods for generation of white light: one method is to combine red, green, and blue (RGB) colors in order to achieve white color. Although this method can be very efficient and allows for a very good color rendering, some problems such as color mixing and existence of the yellow-green gap are yet to overcome. Another way of generating white light is to use blue color together with yellow phosphorus. This method is simple and exhibits good color rendering. However, it is limited on efficiency due to phosphorus conversion efficiency and self-absorption. In addition, multi-phosphor versions are needed to improve color rendering. The third and last approach is to use UV LED pumped RGB phosphors. In this case, white light is determined by phosphors only (tolerant to LED variation), excellent color rendering is possible, and it is easy to manufacture.

8

Optoelectronic

Devices:

Ill-Nitrides

One of the newest GaN-related research topics is the GaN-based diluted magnetic semiconductors (DMS). Spin polarizers, spin transistors, and ultra-dense non-volatile semiconductor memory are among the new class of devices and circuits based on magnetic semiconductors. In this regard, GaAs DMS has been studied more extensively. However, the Curie temperature of (Ga,Mn)As is low ( ~ 110 K). (Ga,Mn)N DMS on the other hand, has been shown to have Curie temperatures exceeding room temperature [14,15], which shows promise for realization of room temperature spintronics devices based on GaN DMS. A number of renowned scientists have contributed their knowledge to this book. A variety of useful information on some of the important properties and applications of Ill-nitrides has been collected in this book and our hope is that the scientific community will find this information helpful.

REFERENCES [1] Razeghi, M. (2002) Proc. IEEE, 90, 1006. [2] Miragliotta, J. & Wickenden, D.K. (1998) in Non-linear Optical Properties of GaN 50 B, Eds. Pankove, I. & Moustakas, T.D., Academic Press, San Diego. [3] Duffy, M.T., Wang, C.C, O'Clock, G.D., McFarlane, S.H., III & Zanzucchi, P.J. (1973) /. Electron. Mater., 2, 359. [4] Lotsch, H.K.V. & Schroter, F. (1970) Das Laser-Farbfensehen. Laser, 2, 37. [5] Pankove, J.I. & Schade, H.E.P. (1974) Appl. Phys. Lett., 25, 53. [6] Razeghi, M. (1998) Int. J. High Speed Electron. Syst., 9, 161. [7] Davydov, V.Y., Klochikhin, A.A., Emtsev, V.V., Ivanov, S.V., Vekshin, V.V., Bechstedt, F., FurthmuUer, J., Harima, H., Mudryi, A.V., Hashimoto, A., Yamamoto, A., Aderhold, J., Graul, J. & Haller, E. (2002) Phys. Stat. Sol. (b), 230, R4. [8] Wu, J., Walukiewicz, W., Wu, K.M., Ager, J.W., IE, Haller, E.E., Lu, H., Schaff, W.J., Saito, Y. & Nanishi, Y. (2002) Appl. Phys. Lett., 80, 3967. [9] Madelung, O. Ed. (1991) Semiconductors: Group IV Elements and III-V Compounds, Springer, Berlin. [10] Madelung, O. (1996) Semiconductors—Basic Data, l""^ Edition Springer, Berlin. [11] Leszczynski, M., Suski, T., Perlin, P., Teisseyre, H., Grzegory, I., Bockowski, M., Jun, J., Porowski, S. & Major, J. (1995) J. Phys., D28, A149. [12] Nakamura, S., Senoh, M., Nagahama, S., Iwasa, N., Yamada, T., Kiyoku, H., Sugimoto, Y., Kozaki, T., Umemoto, H., Sano, M. & Chocho, K. (1997) Jpn. J. Appl. Phys., 36, L1568. Part 2. [13] Yasan, A., McClintock, R., Mayes, K., Darvish, S.R., Kung, P. & Razeghi, M. (2003) Appl. Phys. Lett., 83, 4701. [14] Dietl, T., Ohno, H., Matsukura, F., Cibert, J. & Ferrand, D. (2000) Science, 287, 1019. [15] Reed, M.L., El-Masry, N.A., Stadelmaier, H.H., Ritmus, M.K., Reed, M.J., Parker, C.A., Roberts, J.C. & Bedair, S.M. (2001) Appl. Phys. Lett., 79, 3473.

optoelectronic Devices: Ill-Nitrides M. Razeghi and M. Henini (Eds.) © 2004 Elsevier Ltd. All rights reserved.

Chapter 2

The Rise of Ill-nitrides: An Introduction Patrick Kung Department ofECE, Center for Quantum Devices, Northwestern University, 2145 Sheridan Road, Evanston, IL 60208, USA

The nitrides of group III metal elements or "Ill-nitrides" (or "III-N") commonly refer to aluminum nitride (AIN), gallium nitride (GaN), indium nitride (InN) and their alloys, all of which are compounds of nitrogen—the smallest group V element in the Periodic Table and an element with one of the highest values of electronegativity. Long regarded as a scientific curiosity, Ill-nitrides have now earned a most respected place in the science and technology of compound semiconductors, as well as modem electronic and optoelectronic devices. Unlike more conventional semiconductors, such as silicon (Si) or gallium arsenide (GaAs) which have a diamond or zinc-blende structure with a cubic symmetry, Ill-nitride semiconductors crystallize in their most stable form into a wurtzite crystallographic structure (Table 2.1) with nitrogen atoms forming a hexagonal close packed (hep) structure and the group III atoms occupying half of the tetrahedral sites available in the hep lattice [1,2]. Ill-Nitrides are polar crystals as they do not have a center of symmetry [3]. They thus possess many other potentially useful properties such as piezoelectricity [4], pyroelectricity [5] and non-linear optical properties [6,7]. The large difference in electronegativity between the group III and group V elements (Al = 1.18, G a = 1.13, In = 0.99, N = 3.1) results in very strong chemical bonds within the Ill-nitride material system, which is at the origin of most of the exceptional Ill-nitride physical properties (listed in Table 2.2); that have increasingly spurred interest within the research community. Most importantly, AlInGaN compounds exhibit a direct bandgap energy that can be continuously tailored between 0.7 and 6.2 eV, which corresponds to a wavelength range from 1.78 ixm to 200 nm and thus covers the near infrared, visible or near ultraviolet spectral bands. Figure 2.1 compares the bandgap energy versus lattice constants of Ill-nitrides with that of various semiconductors. Furthermore, because the intrinsic carrier concentration is an exponential function of the energy gap and the temperature, wide bandgap Ill-nitride semiconductors have a much lower intrinsic carrier concentrations over a larger temperature range than Si or GaAs. This results in lower leakage and dark currents, which is especially important in photodetectors

E-mail address: [email protected] (P. Kung).

9

10

Optoelectronic Devices: Ill-Nitrides Table 2.1. Crystallographic data of Ill-nitrides III-N Structure type Crystal system Space group Origin

Wurtzite Hexagonal P6T,mc (No. 186) 3ml; (N^~): positions: 2b site symmetry: 3m» (ai,a2,«3) = (0,0,0.375) (III^^): positions: 2b site symmetry: 3m« (ai,a2,«3) = (0,0,0)

Coordination

and electronic devices. Another consequence of the strong chemical bonding is the physical (high melting points, mechanical strength) and chemical stability of these materials. These also enjoy high thermal conductivity. Their effective masses are higher than conventional semiconductors, thus leading to lower carrier mobilities, but this is made up for by the highsaturated electron drift velocities predicted for this material system. The refractive indices of Ill-nitrides are lower compared to narrower gap semiconductors, which results in a lower

Table 2.2. Physical properties of Ill-nitride materials

300 K energy gap (eV) Lattice constant, a (A) Lattice constant, c (A) Thermal expansion coefficient a^ (10~^K~*) Thermal expansion coefficient a, (10"^ K~') Electron effective mass, mg (mo) Hole effective mass, m^ (mo) Refractive index, n

AIN

GaN

6.2 3.112 4.982 5.27 (20-800°C)

3.44 3.189 5.186 4.3 (17-477°C)

0.7 3.545 5.703 5.6 (280°C)

4.15 (20-800°C)

4.0 (17-477°C)

3.8 (280°C)

0.2

0.11 0.5 (mhh) 0.17 (mih) 2.56 (1,0 fxm) 3.12 (0.66 |xm)

1100 -23.0 10.0

e(0)

2.2 (0.60 \xm) 2.5 (0.23 iJim) 9.14

e(oo)

4.84

Thermal conductivity, K (W/cm K) Melting point (°C) ^(f (kcal/mol) Heat capacity. Cp (cal/mol K)

2.0

0.8 2.35 (1.0 (Jim) 2.60 (0.38 |xm) 10.4 (ElIc) 9.5 (E 1 c) 5.8(Ellc) 5.4 (E 1 c) 1.7-1.8

3000 -68.2 7.6

>1700 -33.0 9.7

InN

9.3

The Rise of Ill-nitrides: An Introduction

11

A Direct Bandgap

6 1

•

Indirect Bandgap

5 H

>

ZnS

4 -\

ZnO

MgSe

• (D C LJJ Q. CD D) T3 C TO GQ

3 H

InN*

~-|— 3.0

I

3.5

4.0

'

I 4.5

'

\ 5.0

5.5

7.0

Lattice Constant (A) Figure 2.1. Bandgap energy versus lattice constant of various semiconductors, including Ill-nitrides. The bandgap energy of InN was recently reported to be 0.7 eV instead of earlier reported 2 eV.

reflectivity at the interface. This is an advantage for photodetector efficiency, but a disadvantage when trying to achieve lasers with low threshold currents. The historical development of Ill-nitrides can be punctuated by the series of milestones listed in Table 2.3, grouped into four phases. For a large part, the research work has often taken example from the development of GaAs-based compounds, first for the fundamental reason that GaN and GaAs are both III-V semiconductor compounds with a direct bandgap and secondly for the practical reason that the technology of GaAs could be adapted to manufacture GaN compounds—up to a certain extent. Phase I corresponds to the major part of the 20th century, before 1960. Although the first AIN, GaN and InN compounds were synthesized as early as 1907 [8], 1910 [9] and 1932 [10], respectively, i.e. much earlier than conventional GaAs and Si semiconductors, no significant progress could be reported until the end of the 1960s. This was mainly due to the fact that Ill-nitride crystals had been very difficult to synthesize. Thanks to the development of modem epitaxial growth techniques, it was possible during the Phase II (1960- 1970s) period to demonstrate the first growth of GaN thin films by hydride vapor phase epitaxy (HVPE) in 1969 [11]. This was quickly followed in 1971 with the first metalorganic chemical vapor deposition (MOCVD) [12] and then the first

12


Table 2.3. Historical milestones in the development of Ill-nitride semiconductors Phase I

1907 1910 1932

Synthesis of AJN Synthesis of InN Synthesis of GaN

Phase II

1969 1971

1974 1975 1976

Epitaxy of GaN, by HVPE MOCVD of GaN Stimulated emission in GaN (needles) by optical pumping Metal-insulator-semiconductor GaN-based LED MBE of GaN MBE of AIN Bulk ALN crystals

Phase III

1983 1986 1989

Concept of low temperature buffer layer MOVPE of crystalUne GaN using AIN buffer layer P-type GaN using LEEBI

Phase IV

1990 1991

Stimulated emission by optical pumping of high-quality GaN GaN buffer by MOCVD GaN p - n junction blue LED P-type AlGaN 2DEG at AlGaN/GaN interface High-quality InGaN GaN photoconductor P-type GaN using thermal anneaUng AlGaN/GaN HEMT or HFET GaN MESFET, MISFET Schottky barrier GaN photodiode Conmiercial candela class blue InGaN LEDs Bulk GaN crystals GaN/SiC HBT (GaN/SiC heterointerface) Microwave HFET, MISFET Microwave GaN MESFET Conunercial candela class "bluish-green" InGaN LED P-type InGaN GaN-AIN SIS junction detector AlGaN/GaN HFET photodetector GaN p - n junction photodiode Pulsed operation blue laser diode 410 nm InGaN yellow LED Commercial "pure green" InGaN LEDs Ion implanted GaN JFET Doped channel AlGaN/GaN HEMT AlGaN photoconductor White LED using phosphors First continuous wave blue laser diode Complete range of AlGaN photoconductors Ultraviolet LED

1992

1993

1994

1995

1996

1997

(continued)


13

Table 2.3. Continued

1998

1999

2000

2001 2002

2003 2004

10,000 h lifetime RT CW InGaN violet LD Schottky barrier AlGaN photodiode AlGaN/GaN HBT GaN MOSFET GaN p - i - n 32 X 32 focal plane array Visible blind AlGaN p - i - n photodiode Sample shipment of blue-violet laser diode Solar blind AlGaN p - i - n photodiode GaN p - i - n 128 X 128 focal plane array GaN avalanche photodiodes MOSHFET in AlGaN (metal-oxide semiconductor heterostructure FET) Commercialization of blue-violet laser diode InN bandgap determined to be 0.7 eV 256 X 256 AlGaN solar blind focal plane array AlGaN UV LEDs at 280 nm Sample shipment of UV laser diodes at 375 nm AlGaN UV LEDs at 265 nm AlGaN UV laser at 350.9 nm

molecular beam epitaxy (MBE) of GaN in 1974 [13]. The first epitaxial growth of AIN was reported in 1975 [14]. The quality of these GaN crystals was sufficient to enable the study of their optical properties. In particular, the first observation of stimulated emission from GaN needles was observed in 1971 [15], prefiguring the demonstration of a blueultraviolet wavelength laser diode 20 years later. However, as it was still impossible to realize bulk Ill-nitride crystals, the epitaxial process had to be performed on non-native substrates and the quality of the GaN films was very poor. Phase III of the development stage of Ill-nitrides, spanning the 1980s, can be considered as the critical period when two fundamental breakthroughs were made. The first one was the introduction of the concept of a low-temperature nucleation layer which permitted the growth of smooth GaN thin films on a foreign substrate in 1983 [16]. This was followed by the demonstration of high-quality crystalline GaN films in 1986 [17]. The second breakthrough was the demonstration of p-type GaN films through low-energy electron beam irradiation (LEEBI) to activate p-type dopants in 1989 [18]. The p-type activation process was later refined in 1992 using thermal annealing [19]. Phase IV corresponds to the period from 1990 until today. It experienced the most dramatic and spectacular research and development work in Ill-nitride based compound semiconductor science and technology to date [20-33]. It is clear that understanding and controlling the optical properties of Ill-nitrides have been the first and primary driving force behind the research work in an effort to realize blue and green light-emitting diodes (LEDs) and laser diodes (LDs). Such components had long been desired in order to achieve bright full color electroluminescent displays, traffic lighting, automobile lighting, and higher

14


density optical data storage. Now, wide bandgap Ill-nitride semiconductors have truly become the cornerstone technology for such devices, supplanting technologies such as silicon carbide (SiC) for blue LEDs, AlInGaP for green LEDs, and zinc selenide for blue and green LDs, even though these had been much more mature and well developed and had already been commercialized at the time the GaN material quality was still poor. This impressive research and development work culminated with the commercialization of candela class GaN-based blue LEDs in 1993 [34] and the realization of the first GaN-based laser diode under pulse operation in 1995 [35,36]. Since then, the development of lasers has followed an accelerated course, with milestones such as the demonstration of continuous wave operation [37], a lifetime of 10,000 h in 1997 [38], which led to the successful conMnercialization of room temperature continuous wave blue-violet laser diodes in 2001 [34]. One instrumental element in realizing long lifetime lasers was the development of lateral epitaxial overgrowth as a method to reduce dislocations in heteroepitaxially grown GaN [39,40]. The performance and reliability of these lasers have been sufficient to allow mass production and start establishing the basic specifications of a standard for next generation high-density optical disk video recordings. One of the leading standard format currently being developed, called "Blu-ray Disc", is expected to make possible "the recording, rewriting and playback of up to 27 GB of data on a single-sided, single-layer 12 cm CD/DVD-size disc using a 405 nm blue-violet laser" [41,42]. These scientific and commercial successes in realizing blue LEDs and lasers, accomplished in such a short period of time, have since spurred a plethora of activity associated with Ill-nitrides. For example, the first yellow/amber LED based on InGaN have been reported in 1995 [43]. Subsequently, the most commercially significant development since then has been the demonstration of a novel light source in 1996 in the form of a white LED that combined a blue LED and a phosphor coating [44]. The potential of this technology as a cheap, low energy consumption, more environmentally friendly solid-state light source has led to a large effort in the US called the "National Next Generation Lighting Initiative" that involves industry, universities and national laboratories [45]. Another current challenge for light emitters is to push the Ill-nitride technology to shorter wavelengths (below 340 nm) for applications such as laser-induced fluorescence of chemical/biological agents, water purification and non-line-of-sight communications. For these purposes, AlGaN ultraviolet LEDs emitting at 340 and 280 nm with increasing powers have been consistently reported [46-50]. Furthermore, AlGaN LED emitting at a wavelength as low as 267 nm have been successfully demonstrated in 2003 [51]. Sample shipments of UV laser diodes with wavelength as short as 375 nm have been initiated in 2002 [34]. And the shortest wavelength laser diode emitting at 350.9 nm was also recently demonstrated [52]. A further area of research with growing interest is that of visible-blind and solar-blind ultraviolet (UV) photodetectors based on Ill-nitrides for use in many applications such as covert space-to-space communications, early missile threat warning, UV astronomy.


15

chemical and biological agent detection, flame detection, engine and furnace monitoring [53,54]. Such devices can also be used for autocorrelation measurements [55]. The visible or solar blindness is the property that the photodetector is sensitive to UV light while being (ideally) insensitive to visible or solar light, and is a key parameter for photodetectors which are expected to detect UV light in a strong visible and/or infrared background. Compared to existing solid-state or other technologies, Ill-nitride semiconductors can lead to devices that are cheaper, more efficient and more robust. After the first report of a GaN ultraviolet photodetector in 1992 [56], that of GaN p - n junction photodiode in 1995 [57], and the demonstration of metal-semiconductor-metal GaN photodetectors [58], the entire range of AlGaN photodetectors has been demonstrated in 1996 with cut-off wavelengths from 200 to 365 nm [59-61], followed by the demonstration of AlGaN p - i - n photodiodes [62-65] and by the realization of increasingly larger size focal plane arrays between 1999 and 2002 [66-68]. With the recent report that the bandgap of InN is only 0.7 eV, there is likely to be a growing interest in Ill-nitrides for high-efficiency solar cells [69]. In addition to optoelectronic devices, Ill-nitrides have a proven potential for RFmicrowave-millimeter wave as well as high-power electronic devices in order to revolutionize high-power electrical energy control, conversion, and distribution, as well as to support applications such as all electric vehicles, wireless communications and radar technology. Two figures of merit generally used to characterize materials for electronic devices can be used to explain the interest that the research community has for Ill-nitrides. The first one, the Johnson figure of merit JFM = {E^VJlTff, where E^ is the breakdown voltage and V^ is the saturation velocity, is related to the electronic properties of materials. It characterizes the frequency-power trade-off, since high frequency usually requires small device dimension while high power need large device size. The Keyes figure of merit, KFM = ajicvjlire)^^^, where OY is the thermal conductivity of the materials and 8 is its dielectric constant, is related to the material thermal properties. It characterizes the device size and thermal resistance trade-off. The values are generally normalized to Si. The values for the JFM and KFM, listed in Tables 2.4 and 2.5, respectively [70], confirm GaN-based semiconductors are very promising for electronic devices. Table 2.4. Comparison of the Johnson figures of merit of different materials [70] Material

EîWIcm)

n (cm/s)

Si GaAs GaN 6H-SiC 3C-SiC Diamond

3X10^ 4X10^ 50 X 10^ 40 X 10^ 40 X 10^ 100 X 10^

1.0 X 2.0 X 2.7 X 2.0 X 2.0 X 2.7 X

10^ 10^ 10^ 10^ 10^ 10^

[{E^V,)liTf

(VW)

9.1 X 10^^ 64.8 X 10^^ 18466 X 10^^ 6485 X 10^^ 6485 X 10^^ 73863 X 10^^

Ratio to silicon 1.0 7.1 2029 712 712 8117

16


Table 2.5. Comparison of the Keyes figures of merit of different materials [70] Material

o-T (300 K) (W/cm)

V, (cm/s)

er

ajiVJe.y^ (W/cm^V^^)

Ratio to silicon

Si GaAs GaN 6H-SiC 3C-SiC Diamond

1.5 0.5 1.3 5.0 5.0 20.0

1.0X10^ 2.0 X 10^ 2.7 X 10^ 2.0 X 10^ 2.0 X 10^ 2.7 X 10^

11.8 12.8 9 9.7 9.7 5.5

13.8 X 10^ 6.25 X 10^ 22.5 X 10^ 71.8 X 10^ 71.8 X 10^ 443.1 X 10^

1.0 0.45 1.6 5.2 5.2 32.1

The first experimental indication of this potential came with the reported observation of a two-dimensional electron gas at AlGaN/GaN heterointerface in 1992 [71]. The subsequent year saw the demonstration of AlGaN/GaN high-electron mobility transistors [72]. And in 1998, the first AlGaN/GaN heterojunction bipolar transistor was reported [73].

Table 2.6. Comparison of the lattice properties and energy gaps of ni-nitrides with potential substrate materials [69,70] Formula name (crystal symmetry)

GaN (hexagonal) AIN (hexagonal) InN (hexagonal) AI2O3 (trigonal) 4H-SiC (hexagonal) 6H-SiC (hexagonal) LiGa02 (orthorhombic) LiAlOi (tetragonal) ZnO (hexagonal) MgO (cubic) Si (cubic) GaAs (cubic) GaP (cubic)

Lattice constants (A)

a = 3.1891; c = 5.1855 a = 3.112; c = 4.982 a = 3.5365; c = 5.7039 a = 4.758; c = 12.991 a = 3.073; c = 10.053 fl = 3.081; c = 15.117 a = 5.402; b = 6.372; c = 5.007 a = 5.1715; c = 6.2840 a = 3.253; c = 5.213 4.216 5.4310202 5.65325 5.4512

Plane with nearest match to (0001) GaN

Effective a lattice constant (A)

Lattice mismatch with GaN (%)

Energy gap

(^GaN ~ '^subV'^sub

(0001)

3.1891

0

3.44

(0001)

3.112

2.47

6.2

(0001)

3.53656

-9.82

0.7

(0001) rotated 30° (0001)

2.747

16.09

>8.5

3.073

3.77

3.20

(0001)

3.081

3.51

2.86

(001)

3.119, 3.186

2.25, 0.10

(100)

6.81, 1.50

(0001)

2.986, 3.142 3.253

-1.97

3.44 (1.6 K)

(111) (111) (111) (111)

2.981 3.840 3.99745 3.8546

4.93 -16.96 -20.22 -17.26

7.9 1.1242 1.424 2.26


17

Table 2.7. Comparison of the thermal properties of Ill-nitrides with potential substrate materials, including their thermal expansion coefficient or TEC {d^), thermal conductivity, and melting point [70] Formula name

a-a(10-^K-^) 0-1000°C

Mismatch with GaN by cooling 1000°C (%) (TECcaN ~ TECsub) X (-1000 K)

Thermal conductivity (W/cm K)

Melting temperature (°C)

GaN AIN InN

5.6 (0 to 600°C) 5.7 5.7 (300°C) 8.6 4.46 4.44 -7

0 0.01 0.01 0.30 -0.11 -0.12 0.14

1.7-1.8 2.0 4.9 0.3

>1700 3000 1100 2015

AI2O3

4H-SiC 6H-SiC LiGaOs LiAlOs ZnO MgO Si GaP GaAs

7.8 13.85 3.90 4.7 6.7

0.22 0.83 -0.17 -0.09 0.11

4.9

0.36 1.3 1.1 0.5

-1600 -1700 -2000 2800 1412 1470 1240

Looking into the future, as InN exhibits the highest saturation velocity of all Ill-nitride compounds, as high as 4.2 X 10^ cm/s, InN is a highly potential material for the fabrication of high-speed field effect transistors. To support all these exceptional technological endeavors, epitaxial growth techniques have significantly improved. Indeed, instead of adapting existing equipment originally used for GaAs, which may have been useful for the demonstration stage, new MOCVD and MBE growth apparatus have been engineered to accommodate and facilitate the growth of Ill-nitrides, including high growth temperatures and in situ monitoring in MOCVD. Bulk GaN substrates have been successfully grown first in 1994 [74], using high-pressure and high-temperature growth techniques. Since then, easier methods have been developed which also led to large bulk crystals. Hundreds of micrometer thick GaN films are now grown by HVPE on a foreign substrate, followed by an epitaxial lift-off and polishing to obtain a quasi-bulk single crystal. AIN substrates are now also being developed, with higher quality than originally reported in the 1970s [75], in order to support the drive toward shorter wavelength optoelectronics. However, the quality, dimension, availability and cost of these substrates are still too high for widespread use. This is why, to date, the quasientirety of Ill-nitride thin films are still grown on non-native substrates and the choice of the most appropriate substrate material is still an issue. In Tables 2.6 and 2.7, the physical properties of several candidate substrate materials are compared with those of Ill-nitrides,

18

Optoelectronic

Devices:

Ill-Nitrides

including their lattice mismatch with GaN and the thermal mismatch when cooling by 1000°C. Three substrates stand out to be the most promising ones: silicon (Si), 6H silicon carbide (6H-SiC) and sapphire (AI2O3) [20-24,26,28,40]. The choice of Si remains a desirable approach because it is the most widely used electronic material, is widely available in large areas and is very cheap. However, it does not have a crystal structure and physical properties similar to Ill-nitrides, and exhibits a lattice mismatch of ~ 17% with GaN. Silicon carbide presents the closest match with Ill-nitrides (lattice mismatch with GaN is only 3.5%) out of the three candidates but SiC substrates are still expensive in comparison with other alternative substrates, although the technology is progressing rapidly to bring prices down thanks mainly to the commercial thrust from Ill-nitride based devices. Sapphire offers a compromise between these two extremes as it is the most widely used substrate to date; high-quality wafers are widely available and are cheaper than SiC, and it has a similar hexagonal crystal symmetry as Ill-nitrides. In summary, after nearly a century since the first synthesis of AIN, Ill-nitride semiconductors have become one cornerstone of modem optoelectronic and electronic devices thanks to a spectacular research and development in the last 20 years. It is still too early to estimate their scientific and economic impact, but one can confidently say that it is not less significant than that of gallium arsenide. REFERENCES [1] Kung, P., Sun, C.J., Saxler, A., Ohsato, H. & Razeghi, M. (1994) Crystallography of epitaxial growth of wurtzite-type thin films on sapphire substrates. J. Appl. Phys., 75, 4515. [2] Sun, C.J., Kung, P., Saxler, A., Ohsato, H., Razeghi, M. & Haritos, K. (1994) A crystallographic model of (00*1) aluminum nitride epitaxial thin film growth on (00*1) sapphire substrate. J. Appl Phys., 75, 3964. [3] Daudin, B., Rouviere, J.L. & Arlery, M. (1996) Polarity determination of GaN films by ion channeling and convergent beam electron diffraction. Appl. Phys. Lett., 69, 2480. [4] Bemardini, F., Fiorentini, V. & Vanderbilt, D. (1997) Spontaneous polarization and piezoelectric constants of III-V nitrides. Phys. Rev. B, 56, R10024. [5] Bykhovski, A.D., Kaminski, V.V., Shur, M.S., Chen, Q.C. & Khan, M.A. (1996) Pyroelectricity in gallium nitride thin films. Appl. Phys. Lett., 69, 3254. [6] Miragliotta, J., Wickenden, D.K., Kistenmacher, T.J. & Bryden, W.A. (1993) Linear- and nonlinear-optical properties of GaN thin films. J. Opt. Soc. Am. B, 10, 1447. [7] Hahn, D.N., Kiehne, G.T., Wong, G.K.L., Ketterson, J.B., Kung, P., Saxler, A. & Razeghi, M. (1999) Phase-matched optical second-harmonic generation in GaN and AIN slab waveguides. J. Appl. Phys., 85, 2497. [8] Fichter, F. (1907) Uber aliminiumnitride. Z Anorg. Chem., 54, 322. [9] Fischer, F. & Schroter, F. (1910) Berichte der Deutschen Chemischen Gesellschaft, 43, 1465. [10] Johnson, V.C, Parsons, J.B. & Crew, M.C. (1932) /. Phys. Chem., 36, 2588. [11] Maruska, H.P. & Tietjen, J.J. (1969) The preparation and properties of vapor-deposited singlecrystalUne GaN. Appl. Phys. Lett., 15, 327.


19

[12] Manasevit, H.M., Erdmann, P.M. & Simpson, W.I. (1971) The use of metalorganics in the preparation of semiconductor materials. IV. The nitrides of aluminum and gallium. /. Electrochem. Soc, 118, 1864. [13] Akasaki, I. et al. (1974) MITI report in Japanese only; Akasaki, I., Hayashi, I., 1976, Ind. Set. TechnoL, 17, 48. [14] Yoshida, S., Misawa, S. & Itoh, A. (1975) Epitaxial growth of aluminum nitride films on sapphire by reactive evaporation. Appl. Phys. Lett., 26, 461. [15] Dingle, R., Shaklee, K.L., Leheny, R.F. & Zetterstrom, R. (1971) Stimulated emission and laser action in gallium nitride. Appl. Phys. Lett., 19, 5. [16] Yoshida, S., Misawa, S. & Gonda, S. (1983) Improvements on the electrical and luminescent properties of reactive molecular beam epitaxially grown GaN films by using AlN-coated sapphire substrates. Appl. Phys. Lett., 42, 427. [17] Amano, H., Sawaki, N., Akasaki, I. & Toyoda, Y. (1986) Metalorganic vapor phase epitaxial growth of a high quality GaN film using an AIN buffer layer. Appl. Phys. Lett., 48, 353. [18] Amano, H., Kito, M., Hiramatsu, K. & Akasaki, I. (1989) P-type conduction in Mg-doped GaN treated with low-energy electron beam irradiation (LEEBI). Jpn. J. Appl. Phys., 28, L2112. [19] Nakamura, S., Mukai, T., Senoh, M. & Iwasa, N. (1992) Thermal annealing effects on p-type Mg-doped GaN films. Jpn. J. Appl. Phys., 31, L139. [20] Sun, C.J. & Razeghi, M. (1993) Comparison of the physical properties of GaN thin films deposited on (0112) and (0001) sapphire substrates. Appl. Phys. Lett., 63, 973. [21] Sun, C.J., Kung, P., Saxler, A., Ohsato, H., Bigan, E., Razeghi, M. & Gaskill, D.K. (1994) Thermal stability of GaN thin films grown on (0001) AI2O3, (0112) AI2O3 and (OOOl)si 6HSiC substrates. J. Appl. Phys., 76, 236. [22] Saxler, A., Kung, P., Sun, C.J., Bigan, E. & Razeghi, M. (1994) High quality aluminum nitride epitaxial layers grown on sapphire substrates. Appl. Phys. Lett., 64, 399. [23] Dovidenko, K., Oktyabrsky, S., Narayan, J. & Razeghi, M. (1996) Aluminum nitride films on different orientations of sapphire and silicon. /. Appl. Phys., 79, 2439. [24] Kung, P., Saxler, A., Zhang, X., Walker, D., Wang, T.C., Ferguson, I. & Razeghi, M. (1995) High quahty AIN and GaN epilayers grown on (00*1) sapphire, (100) and (111) sihcon substrates. Appl. Phys. Lett., 66, 2958. [25] Saxler, A., Walker, D., Kung, P., Zhang, X., Razeghi, M., Solomon, J., Mitchel, W. & Vydyanath, H.R. (1997) Comparison of trimethylgallium and triethylgallium for the growth of GaN. Appl. Phys. Lett., 71, 3272. [26] Zhang, X., Kung, P., Saxler, A., Walker, D., Wang, T.C. & Razeghi, M. (1995) Growth of Al^Gai_;,N:Ge on sapphire and silicon substrates. Appl. Phys. Lett., 67, 1745. [27] Zhang, X., Kung, P., Saxler, A., Walker, D., Wang, T. & Razeghi, M. (1995) Photoluminescence study of GaN. Acta Phys. Polonica A, 88, 601. [28] Kung, P., Saxler, A., Zhang, X., Walker, D., Lavado, R. & Razeghi, M. (1996) Metalorganic chemical vapor deposition of monocrystalline GaN thin films on p-LiGa02 substrates. Appl. Phys. Lett., 69, 2X16. [29] Zhang, X., Kung, P., Saxler, A., Walker, D. & Razeghi, M. (1996) Observation of room temperature surface-emitting stimulated emission from GaN:Ge by optical pumping. /. Appl. Phys.,m, 65AA. [30] Zhang, X., Walker, D., Saxler, A., Kung, P., Xu, J. & Razeghi, M. (1996) Observation of inversion layers at AIN-Si interfaces fabricated by metal organic chemical vapour deposition. Electron. Lett., 32, 1622.

20

Optoelectronic

Devices:

Ill-Nitrides

[31] Kung, P., Saxler, A., Walker, D., Rybaltowski, A., Zhang, X., Diaz, J. & Razeghi, M. (1998) GalnN/GaN multi-quantum well laser diodes grown by low-pressure metalorganic chemical vapor deposition. MRS Internet J. Nitride Semicond. Res., 3 (1). [32] Saxler, A., Mitchel, W.C., Kung, P. & Razeghi, M. (1999) Aluminum gallium nitride shortperiod superlattices doped with magnesium. Appl. Phys. Lett., 74, 2023. [33] Yasan, A., McClintock, R., Darvish, S.R., Lin, Z., Mi, K., Kung, P. & Razeghi, M. (2002) Characteristics of high quality p-type Al;,Gai_;,N/GaN superlattices. Appl. Phys. Lett., 80,2108. [34] http://www.nichia.co.jp/info/history.html. [35] Akasaki, L, Amano, H., Sota, S., Sakai, H., Tanaka, T. & Koike, M. (1995) Stimulated emission by current injection from an AlGaN/GaN/GaInN quanmm well device. Jpn. J. Appl. Phys., 34, L1517. [36] Nakamura, S., Senoh, M., Nagahama, S.I., Iwasa, N., Yamada, T., Matsushita, T., Kiyoku, H. & Sugimoto, Y. (1996) InGaN-based multi-quantum-well-structure laser diodes. Jpn. J. Appl. Phys., 35, L74. [37] Nakamura, S., Senoh, M., Nagahama, S.I., Iwasa, N., Yamada, T., Matsushita, T., Sugimoto, Y. & Kiyoku, H. (1996) Continuous-wave operation of InGaN multi-quantum-well-structure laser diodes at 233 K. Appl. Phys. Lett., 69, 3034. [38] Nakamura, S., Senoh, M., Nagahama, S.I., Iwasa, N., Yamada, T., Matsushita, T., Kiyoku, H., Sugimoto, Y., Kozaki, T., Umemoto, H., Sano, M. & Chocho, K. (1997) InGaN/GaN/AlGaNbased laser diodes with modulation-doped strained-layer superlattices. Jpn. J. Appl. Phys., 36, L1568. [39] Usui, A., Sunakawa, H., Sakai, A. & Yamaguchi, A.A. (1997) Thick GaN epitaxial growth with low dislocation density by hydride vapor phase epitaxy. Jpn. J Appl. Phys., 36, L899. [40] Kung, P., Walker, D., Hamilton, M., Diaz, J. & Razeghi, M. (1999) Lateral epitaxial overgrowth of GaN films on sapphire and silicon substrates. Appl. Phys. Lett., 74, 570. [41] http://www.blu-ray.com. [42] New Blu-ray DVD format uses blue-violet lasers to achieve 27 GB recording capacity, http:// www.compoundsemiconductor.net (20 February 2002). [43] Nakamura, S., Senoh, M., Iwasa, N. & Nagahama, S.I. (1995) High-brightness InGaN blue, green and yellow light-emitting diodes with quanmm well structures. Jpn. J. Appl. Phys., 34, L797. [44] Sato, Y., Takahashi, N. & Sato, S. (1996) Full-color fluorescent display devices using a nearUV Hght-emitting diode. Jpn. J Appl. Phys., 35, L838. [45] http://lighting.sandia.gov/Xhghtinginit.htm. [46] Yasan, A., McClintock, R., Mayes, K., Darvish, S.R., Kung, P. & Razeghi, M. (2002) Topemission ultraviolet light-emitting diodes with peak emission at 280 nm. Appl. Phys. Lett., 81,801. [47] Yasan, A., McClintock, R., Mayes, K., Darvish, S.R., Zhang, H., Kung, P., Razeghi, M., Lee, S.K. & Han, J.Y. (2002) Comparison of ultraviolet light-emitting diodes with peak emission at 340 nm grown on GaN substrate and sapphire. Appl. Phys. Lett., 81, 2151. [48] Yasan, A., McChntock, R., Mayes, K., Darvish, S.R., Kung, P., Razeghi, M. & Molnar, R.J. (2002) 280 nm UV LEDs grown on HVPE GaN substrates. Opto-Electron. Rev., 10, 67. [49] Yasan, A., McChntock, R., Mayes, K., Kim, D.H., Kung, P. & Razeghi, M. (2003) Photoluminescence study of AlGaN-based 280 nm ultraviolet light-emitting diodes. Appl. Phys. Lett., 83, 4083. [50] Mayes, K., Yasan, A., McClintock, R., Shiell, D., Darvish, S.R., Kung, P. & Razeghi, M. (2004) High power 280 nm AlGaN light emitting diodes based on an asymmetric single quantum well. Appl. Phys. Lett., 84, 1046.

The Rise of lU-nitrides: An Introduction

21

[51] Yasan, A., McClintock, R., Mayes, K., Shiell, D., Gautero, L., Darvish, S.R., Kung, P. & Razeghi, M. (2003) 4.5 mW operation of AlGaN-based 267 nm deep-ultraviolet light-emitting diodes. Appl Phys. Lett., 83, 4701. [52] lida, K., Kawashima, T., Miyazaki, A., Kasugai, H., Mishima, S., Honshio, A., Miyake, Y., Iwaya, M., Kamiyama, S., Amano, H. & Akasaki, I. (2004) 350.9 nm UV laser diode grown on low-dislocation-density AlGaN. Jpn. J. Appl. Phys., 43, L499. [53] Razeghi, JVl. & Rogalski, A. (1996) Semiconductor ultraviolet detectors. /. Appl. Phys. Appl. Phys. Rev., 79, 7433. [54] Razeghi, M. (2002) Short wavelength solar-blind detectors: status, prospects, and markets. Proceedings of IEEE, Wide Bandgap Semicond. Devices: The Third Generation Semiconductor Comes of Age, 90, 1006. [55] Streltsov, A., Moll, K.D., Gaeta, A., Kung, P., Walker, D. & Razeghi, M. (1999) Pulse autocorrelation measurements based on two- and three-photon conductivity in a GaN photodiode. Appl. Phys. Lett., 75, 3778. [56] Asif Khan, M., Kuznia, J.N., Olson, D.T., Van Hove, J.M., Blasingame, M. & Reitz, L.F. (1992) High-responsivity photoconductive ultraviolet sensors on insulating single-crystal GaN epilayers. Appl. Phys. Lett., 60, 2917. [57] Zhang, X., Kung, P., Walker, D., Piotrowski, J., Rogalski, A., Saxler, A. & Razeghi, M. (1995) Photovoltaic effects in GaN structures with p - n junction. Appl. Phys. Lett., 67, 2028. [58] Walker, D., Monroy, E., Kung, P., Wu, J., Hamilton, M., Sanchez, F.J., Diaz, J. & Razeghi, M. (1999) High-speed, low-noise metal-semiconductor-metal ultraviolet photodetectors based on GaN. Appl. Phys. Lett., 74, 762. [59] Kung, P., Zhang, X., Walker, D., Saxler, A., Piotrowski, J., Rogalski, A. & Razeghi, M. (1995) Kinetics of photoconductivity in n-type GaN photodetector. Appl. Phys. Lett., 67, 3792. [60] Walker, D., Zhang, X., Kung, P., Saxler, A., Javadpour, S., Xu, J. & Razeghi, M. (1996) AlGaN ultraviolet photoconductors grown on sapphire. Appl. Phys. Lett., 68, 2100. [61] Walker, D., Zhang, X., Saxler, A., Kung, P., Xu, J. & Razeghi, M. (1997) A^Gai-^N (0 < X < 1) ultraviolet photodetectors grown on sapphire by metal-organic chemical-vapor deposition. Appl. Phys. Lett., 70, 949. [62] Walker, D., Saxler, A., Kung, P., Zhang, X., Hamilton, M., Diaz, J. & Razeghi, M. (1998) Solar blind GaN p - i - n photodiodes. Appl. Phys. Lett., 72, 3303. [63] Monroy, E., Hamilton, M., Walker, D., Kung, P., Sanchez, F.J. & Razeghi, M. (1999) Highquality visible-blind AlGaN p - i - n photodiodes. Appl. Phys. Lett., 74, 1171. [64] Walker, D., Kumar, V., Mi, K., Sandvik, P., Kung, P., Zhang, X.H. & Razeghi, M. (2000) Solar-blind AlGaN photodiodes with very low cutoff wavelength. Appl. Phys. Lett., 76, 403. [65] McClintock, R., Yasan, A., Mayes, K., Shiell, D., Darvish, S.R., Kung, P. & Razeghi, M. (2004) High quantum efficiency AlGaN solar-blind photodetectors. Appl. Phys. Lett., 84,1248. [66] Yang, B., Heng, K., Li, T., Collins, C.J., Wang, S., Dupuis, R.D., Campbell, J.C., Schurman, M.J. & Ferguson, I.T. (2000) 32 X 32 ultraviolet Alo.1Gao.9N/GaN p - i - n photodetector array. /. Quantum Electron., 36, 1229. [67] Brown, J.D., Boney, J., Matthews, J., Srinivasan, P. & Schetzina, J.F. (2000) UV-specific (320-365 nm) digital camera based on a 128 X 128 focal plane array of GaN/AlGaN p - i - n photodiodes. MRS Internet J. Nitride Semicond. Res., 5, 6. [68] Lamarre, P., Hairston, A., Tobin, S.P., Wong, K.K., Sood, A.K., Reine, M.B., Pophristic, M., Birkham, R., Ferguson, I.T., Singh, R., Eddy, C.R., Jr., Chowdhury, U., Wong, M.M., Dupuis, R.D., Kozodoy, P. & Tarsa, E.J. (2001) AlGaN UV focal plane arrays. Phys. Stat. Sol. (a), 188, 289.

22

Optoelectronic

Devices:

Ill-Nitrides

[69] Bhuiyan, A.G., Hashimoto, A. & Yamamoto, A. (2003) Indium nitride (InN): a review on growth, characterization, and properties. /. Appl Phys., 94, 2779. [70] Kung, P. & Razeghi, M. (2000) Ill-Nitride wide bandgap semiconductors: a survey of the current status and future trends of the material and device technology. Opto-Electron. Rev., 8, 201. [71] Asif Khan, M., Kuznia, J.N., Van Hove, J.M., Pan, N. & Carter, J. (1992) Observation of a twodimensional electron gas in low pressure metalorganic chemical vapor deposited GaNAl^Gai-;,N heterojunctions. Appl Phys. Lett., 60, 3027. [72] Asif Khan, M., Bhattarai, A., Kuznia, J.N. & Olson, D.T. (1993) High electron mobility transistor based on a GaN-Al^^Gai-^^N heterojunction. Appl. Phys. Lett., 63, 1214. [73] Ren, F., Abemathy, C.R., Van Hove, J.M., Chow, P.P., Hickman, R., Klaasen, J.J., Kopf, R.F., Cho, H., Jung, K.B., La Roche, J.R., Wilson, R.G., Han, J., Shul, R.J., Baca, A.G. & Pearton, S.J. (1998) 300°C GaN/AlGaN heterojunction bipolar transistor. MRS Internet I. Nitride Semicond. Res., 3, 41. [74] Teisseyre, H., Perlin, P., Suski, T., Grzegory, I., Porowski, S., Jun, J., Pietraszko, A. & Moustakas, T.D. (1994) Temperature dependence of the energy gap in GaN bulk single crystals and epitaxial layer. J. Appl. Phys., 76, 2429. [75] Gerlich, D., Dole, S.L. & Slack, G.A. (1986) Elastic properties of aluminum nitride. J. Phys. Chem. Solids, 47, 437.


Chapter 3

The Evolution of Nitride Semiconductors Satoshi Kamiyama, Hiroshi Amano and Isamu Akasaki Faculty of Science and Technology, High-Tech Research Center, and Nano-Factory, Meijo University, 1-501 Shiogamaguchi, Tempaku-ku, Nagoya 468-8502, Japan

It had been believed for long time that it was almost impossible to grow a high-quality GaN single crystal. This difficulty was overcome with low-temperature deposited buffer layer technology. This achievement opened up the pioneering path to the discovery of p-type conduction, control of n-type conductivity and verification of quantum size effects. These breakthroughs have led to the novel and high-performance devices such as high-efficiency blue, green and white light-emitting diodes, long-lived blue-violet lasers, low-noise UV detectors and high-speed transistors. Furthermore, UV-LEDs, which have recently been focused on as a leading edge of frontier application, have been improved by the reduction of threading dislocation density in an AlGaN layer. A device with external quantum efficiency of 1.4% at 363 nm peak wavelength has been demonstrated that could be used in such applications as lighting equipment in combination with three-color phosphors, exciting light sources of optical catalyst and so on. All of these nitride-based devices are able to operate in a harsh environment because of their toughness. They should also enable a great saving in energy and are suitable for the protection of the environment. Nitride semiconductors and their devices are expected to contribute significantly to the future of our world.

3.1.

INTRODUCTION

Group-Ill nitride semiconductors have been recognized as among the most promising materials for optical devices in the short-wavelength region because of their wide bandgap with direct transition. Since the AlGaInN system can cover a very wide wavelength range, from 200 nm to more than 1700 nm, it is also applicable not only to the short-wavelength devices but also the long-wavelength ones, as shown in Figure 3.1. The high electronsaturation velocity in GaN is also suitable for application in high-speed and high-power electronic devices. The superior physical and chemical stability of the nitride

E-mail address: [email protected] (I. Akasaki).

23

24

Optoelectronic Devices: Ill-Nitrides 8

1

1

1

^sapphire

6 \ Q. CD O) •D C CD

4

1

//•

1

1

1

^Si

^GaAs

^1 GaN

\

BeSe n^

1

0.25

L

1—1

0.30

GaP ^

\

1

MgSe

\

' \ ' /f

- I\ 1

0

1

ZnSe \ ^ 1 cdS

6H-^

2

1

ÎnP

i •' i MgS

ZnS \ i ^ ;

\

; % \zno

00

1

AIN

BN

>

i

^SiC

\

InN

'

//•

0.35

i^ L

i Si i 1

1

0.50

^

CdSe

? InP

GaAs

L J — •

0.55

•

•'

1

1 —

0.60

In-plane Lattice Constant (nm) Figure 3.1. Direct bandgap energy as a function of in-plane lattice constant in AlGaInN system.

semiconductors will enable them to operate in harsh environments. Moreover, nitridebased devices are the most "environmentally friendly" ones available. To produce such novel devices, it is essential to grow high-quality nitride single crystals and to control their electrical conductivity. However, it has been quite difficult to grow high-quality epitaxial GaN films with a specular surface free from cracks. Moreover, its conductivity had never been controlled and, hence, many researchers retired from the nitride field. In the second half of the 1980s, there were two important breakthroughs: the development of extremely high-quality GaN single crystal with a specular surface free from cracks [1] and the discovery of p-type GaN together with the ability to fabricate a p - n junction light-emitting diode (LED) [2]. These breakthroughs have led to such developments as high-performance blue and green LEDs, violet laser diodes (LDs), ultraviolet (UV)-photodetectors (PDs) and field effect transistors (FETs). Very recently, the development of UV light-emitting devices has been focused on for frontier applications, such as lighting equipment in combination with three-color phosphors, light sources of high-density optical data-storage systems and excitation sources for optical catalysts. Many significant results concerned with the UV lightemitting devices have been reported in recent years. In this chapter, the evolution of group-Ill nitride semiconductors and blue lightemitting devices is reviewed. The recent advances of the crystal growth methods for

The Evolution of Nitride Semiconductors

25

low-dislocation-density and high-performance UV light-emitting diodes are also described.

3.2. NITRIDE RESEARCH IN THE EARLY DAYS

In 1969, Maruska and Tietjen succeeded in growing the first GaN single crystal on a sapphire substrate by hydride vapor phase epitaxy (HVPE) [3]. They also found that GaN has a direct-transition bandstructure with a bandgap energy of about 3.39 eV. This accelerated and inspired further research on GaN (Figure 3.2(a)). Dingle et al. demonstrated optically pumped UV stimulated emission from a GaN crystal at 2 K [4]. The first bluish-green LED having a metal-insulator-semiconductor (MIS) structure was developed by Pankove et al. in 1971 [5]. Ejder reported energy dispersion of the refractive index of GaN in 1971 [6]. In 1974, Monemar reported the temperature dependence of exciton recombination energy in GaN grown by HVPE [7].

1—'—\—'—\—'—r Source INSPEC Keyword GaN As of 1999 1000 h Laser diode (CW) h

y

Laser diode (pulse) •D

J) to C

500°C HBT commercial LED —•

100

o

FET-^"— LT-buffer /photodetector

1

MIS-LED/"

I /

10

/

study of V luminescence'

V •

I

stimulated e emission RT

p-type, (n-type) pn-junction LED

_ HVPE GaN W

(B)

31:

(C)

(D)

-rr*-»r

_L ^ L 1965 1970 1975 1980 1985 1990 1995 2000 Calendar year Figure 3.2. Number of publications concerned with nitrides and their devices over the years.

26

Optoelectronic Devices: Ill-Nitrides (b)

IN, Figure 3.3.

Surface morphology of (a) GaN directly grown on a sapphire substrate and (b) GaN grown using LT-buffer layer on a sapphire substrate.

In contrast to traditional III-V compounds such as GaAs and InP, however, it was quite difficult to grow high-quality epitaxial GaN film and in particular, film with a flat surface free from cracks (Figure 3.3(a)). This was mainly due to the large lattice and thermal mismatches between the GaN epitaxial layer and the sapphire substrate. Moreover, the high residual electron concentration in GaN made it quite difficult to achieve p-type conduction and to control the conductivity of n-type GaN. As a result, many researchers withdrew from the field of research on nitride semiconductors (Figure 3.2(b)). One of the authors (LA.) started working on GaN in 1973, and grew single-crystalline GaN by molecular beam epitaxy in 1974 [8]. In 1981, he developed fairly bright bluishgreen MIS LED based on GaN grown by HVPE [9]. The emission wavelength was 495 nm and the external quantum efficiency T7ext was about 0.12%, which was a new record. This LED had the first flip-chip-type electrode so that we were able to fabricate it more easily. This LED, however, was not commercialized because the operating voltage, which is determined by the thickness of the insulating layer, ranged from several to 10 V and could not be controlled. However, he determined to continue to struggle with this difficult system at that time.

3.3.

BREAKTHROUGHS IN CRYSTAL GROWTH

In 1985, an extremely high-quality GaN with a specular surface free from cracks on a sapphire substrate was achieved by pioneering a low-temperature-deposited (LT-)buffer layer technology in the metal-organic vapor phase epitaxy (MOVPE) [1]. The essence of this method is to insert a slightly softer material between the epitaxial layer and the highly mismatched substrate in order to reduce the interfacial free energy. The surface morphology of GaN was dramatically improved by using an LT-AlN-buffer layer. The GaN film grown with the LT-buffer layer on a sapphire substrate is quite transparent and it has a specular surface free from cracks, as shown in Figure 3.3(b). The X-ray diffraction profiles [10] and photoluminescence (PL) property [11] showed that the crystalline quality of GaN was also significantly improved by using the LT-buffer layer.

The Evolution of Nitride Semiconductors

27

The residual electron concentration of the GaN grown with the LT-buffer layer was reduced to the order of 10^^ cm~^ (and soon after, to the order of 10^^ cm~^). Simultaneously, the electron mobility was greatly improved [12]. All of these results show clearly that, by inserting the LT-buffer layer, not only the crystalline quality but also the electrical and optical properties of GaN can be dramatically improved at the same time. In 1989, for the first time, distinct p-type GaN with low resistivity was discovered in Mg-doped GaN on the LT-buffer layer with low-energy electron-beam irradiation (LEEBI) [2]. Immediately, the first p - n junction UV (edge emission) and violet LED [2] were demonstrated. A p-type AlGaN and a p-type GaInN were also achieved in 1991 [13] and 1994 [14], respectively. In 1992, Nakamura succeeded in making p-type GaN by thermal annealing in a N2 atmosphere of Mg-doped GaN using Cp2Mg [15]. Regarding n-type doping, researchers attempted doping with SiH4 in 1986 [16], but it was difficult to control the conductivity due to the high density of residual donors. We also succeeded in controlling the conductivity of n-type nitrides using high-quality GaN or AlGaN grown with the LT-buffer layer in combination with SiH4 doping [17,18]. The electron concentration could be linearly controlled from an undoped level to 10^^ cm~^ without deterioration of surface morphology when the SiH4 flow rate was varied. Thus, essential technologies for the achievement of nitride-based devices were established in this period. These breakthroughs caused the transition from decline to prosperity in research on nitrides, which can be called the "Renaissance" in research on nitrides, as seen in Figure 3.2(c).

3.4. EVOLUTION OF NITRIDE-BASED BLUE LIGHT-EMITTING DEVICES Since the accomplishments of two important milestones, everything has progressed very quickly and work in the field began to grow tremendously. The high-crystalline nitrides have led to a revolutionary change in the optical properties of nitride materials and devices. Figure 3.4 shows the chronological change of the external quantum efficiency, T7ext of nitride-based blue LEDs. 7]^^^ had been saturated at about 0.1% before the breakthroughs were reached. It began to increase steeply after the success in growing high-quality nitride crystal, which resulted in the p - n junction LED. In 1992, rj^^^ of 1.5% was achieved [19], and in 1993, the first nitride-based blue LED was commercialized. At present, blue and green LEDs have rj^^^ of more than 20 and 10%, respectively. The LEDs are much brighter than incandescent lamps. A i7ext of 32% for a violet LED offers a promising excitation source for phosphors. Figure 3.5 shows the threshold power, Pth^ for optically pumped stimulated emission from nitrides over the years. Before 1985, the stimulated emission using optical excitation was obtained only at low temperatures and Pth was very high. After the two breakthroughs

28


10 1 t

[

•

HVPE

• A

MOVPE MBE

:

1/ J1

/' Q LU

/

1 P /

—I 0 _3

:

(flip-chip

GO

S

A" '

0-1

/ L

/

r

/

/ //

•

t

/

\ M

- AL

)./

//

/

0)

/

0.01

• h

1E-3bL

/

W —J—1

1970

1975

1

1

—1

1

\L-X

L-U

1 1/ 1 _ j — i _ j — 1 _ \

1980/ 1985 / 1990

LT-buffer

p-n junction

1 1 1

1995\ 2000 Commercial LED

Calendar Year Figure 3.4. Chronological change of the external quantum efficiency, i7ext of nitride-based blue LEDs.

mentioned above, stimulated emission was achieved at RT and Pth began to decrease exponentially. Then the first lasing operation with pulsed current injection was achieved in 1995 [20] and a LD having continuous-wave operation was announced in 1996 [21]. By 1998, an estimated lifetime of 10"^ h had been achieved [22], becoming commercially available in 1999. One can see that the performance of light-emitting devices and the number of related publications (Figure 3.2) scale with each other.

3.5.

PROGRESS IN THE STUDY OF NITRTOE-BASED QUANTUM STRUCTURES

In a related activity, in 1991, a quantum size effect was verified by Itoh et al. [23] using an AlGaN/GaN quantum well, which was grown on high-quality GaN. A high electron

The Evolution of Nitride Semiconductors ^ •

29

Stimulated emission (low temperature) Stimulated emission (room temperature)

^

^^^^'"9

r

(S): surface mode

80K A

2120K

10

(S)

: w

[

•

i \

):

\

(D

i CL

\ \ \

r^ 2K

o

( S ) \ ( ^ idet) 0.1

•

\ \

h

:

current inj.

0.01 1970

L_|—1—1—L_ —J 1 1

1975

1980

LT-buffer

jf\

r 1 / f

L3u

1990

p-n junction

1995 \

2000

Laser

Calendar Year Figure 3.5. Threshold power, P^YI for optically pumped stimulated emission from nitrides over the years.

mobility of a two-dimensional electron gas in AlGaN/GaN heterostructure was reported by Khan et al. in 1991 [24]. In 1993, Sakai et al. reported the theoretical band lineups of binary alloys of AIN, GaN and InN [25], which indicated that both AlGaN/GaN and GalnN/GaN systems have the type-I heterostructure. In 1994, it was shown that the wavelength dependence of the refractive indices of AlGaN and GaInN was suitable for optical confinement in DH or SCH structures [26]. A large piezoelectric field of about 5X10^ V/cm, which is induced by lattice mismatch, was observed in a GaN/AlN/GaN heterostructure in 1993 [27].

30


The GalnN/GaN multi-quantum well (MQW), which is currently used as an active layer in nitride-based LDs, was first reported in 1995 [28-30]. Since then, peculiar properties of GaInN have been revealed. Despite the strong compressive strain of more than 1% in GaInN layers grown on GaN, it was found to be grown coherently, even in MQWs [31]. It was found that the bandedge photoluminescence intensity from GaInN MQW was greatly increased when the width of the wells was small [29,30], which was later qualitatively explained by the quantum-confined Stark effect (QCSE), due to the presence of the piezoelectric field [32].

3.6.

REDUCTION OF THREADING DISLOCATION DENSITY

3,6.1 GaN The reduction of threading dislocations in a GaN layer has been a focus of activity. The GaN grown on a sapphire substrate with the LT-buffer layer still contains dislocations, as many as 10^-10^^ cm~^.These dislocations are considered to be a major cause of the degradation of LDs which operate at high current density of 2 - 4 kA/cm^, and may pose an obstacle to the high-performance of PDs and TRs. Various technologies have been developed for the reduction of the dislocation density in a GaN layer, which are summarized in Figure 3.6. One is the application of a method

Akasaki's group Direct growth

Rough surface: Many cracks Dislocation density > 10'"'' cm'^

(until 1985)

Residual donor cone. > 10^^ cm-^ LT-buffer technology LT-buffer

Specular surface: Free from cracks

GaN

Dislocation density 10^-10''° cm"^

(1986) Sapphire

^

Residual donor cone. 10 Frequency (GHz)

Figure 5.6.

-

.^.^^ 100

(a) Current gain versus frequency; (b) power gain versus frequency.

76

Optoelectronic Devices: Ill-Nitrides 2.2 W/mm

15

5 10 Output Power (dBm) Figure 5.7.

20

Power performance of a 0.3 |xm X 100 jjum GaN MESFET on sapphire at 2 GHz.

10^

10^ Frequency [Hz]

Figure 5.8.

Input referred equivalent noise voltage spectral density for GaN MESFETs.

11

MOCVD Growth of Group III Nitrides

Figure 5.8 reports the noise spectra of the equivalent input voltage spectral density ^v for GaN MESFET devices. We have observed that all the noise spectra exhibit a llf noise shape and that pronounced generation-recombination {g-r) bumps are absent. On the other hand, it can be observed that the noise spectra exhibit a pronounced dependence on VGS- The noise magnitude at frequencies above 1 kHz decreases with decreasing VGSIn order to better address the device quality, the investigated devices were compared in terms of their Hooge's parameter, defined as follows: «H ^Vsat/d

where Sy is defined in Figure 5.8, g^ is the transconductance, Vsat is the electron saturation velocity in the channel, I^ is the drain current, L^ff is the length of the saturated velocity channel beneath the gate, / is the frequency and q is the elementary charge. Figure 5.9 compares the Hooge's parameters obtained in the present work with the values reported in the Hterature for different GaN-based FET's (HFET and HEMT) [7-9]. Moreover, it is worth noticing that the GaN MESFET devices exhibit values of the Hooge's parameter comparable, or at least lower, to the values reported in Ref. [7] for other HFET devices.

3i

• HEMT

I

o o 21 +

HFET

E cz I—

a. 11 + o o X

HFET

MESFET [This work]

'.S; 1 1997

Figure 5.9.

1998

1999 Year

2000

2001

Comparison of the Hooge's parameter for different GaN-based FET technologies.

78

Optoelectronic Devices: III-Nitrides

5.3. GaAlN/GaN HEMT DEVICES 53,1 Sapphire Substrate 5.3.1.1 Experimental Procedure, GaAlN bulk material and GaAlN/GaN HEMT heterostructures with Al content varying from 12.5 to 29% were grown in a single wafer AIXTRON reactor, on sapphire substrates, using triethylgallium (TEG), trimethygallium (TMG), trimethylaluminum (TMAl) and ammonia (NH3) as group III and group V precursors, respectively. The incorporation of aluminum in the solid phase has been studied as a function of the group III element molar fraction, V/III ratio and growth temperature. A parasitic reaction was clearly identified to occur in the gas phase between TMAl and NH3, which has strong influence on the growth rate and the aluminum incorporation. As a matter of fact, these two parameters were found to decrease as the NH3 flow (V/III ratio) is increased, as seen in Figure 5.10, while their variation with the Al/Al -|- Ga molar fraction appeared to be non-linear. 5.3.1.2 Material Property Results, GaAlN bulk material grown on GaN appeared to be highly tensilely strained. The strain, checked by HR-XRD was found to increase with Al incorporation (a =13

GPa; JCAI = 12.5% => a= 2.82 GPa; JCAI = 29%).

Plastic relaxation, associated with the generation of misfit dislocations and a strong decrease of the tensile strain, was observed in GaAlN alloy with 30% Al content, which correlates well with a critical thickness of 1000 A for this alloy composition [10].

— •

I

1

1

1

•

•

1 — 1

•

•

1

•

«

•

1

'

•

• — 1

1 ' 1

1.10

13-

1.05; J 1.00 - H

03^ 1 1 -

« 0.96 ,

"^ 1 0 -

0.85

81

2.0x10^

1

1

1

r — 1

2.5x10"

,

1

1

3.0x10'

1

1

1

1

3.5x10'

1

1

11

>

4.0x10^

'"1

1

(

" 1

1

4.5x10'

Motar fk>wNH^ (moi/min) Figure 5.10. Aluminum content and growth rate versus NH3 molar flow in GaAlN bulk material.


79

N-type doping of GaAlN material has been studied as a function of Al content, using Si from silane as dopant. The material appeared to be electrically compensated for Al content higher than 20%. However, SIMS analysis performed on such samples revealed silicon incorporation up to 10^^ cm~^ even in highly compensated samples, associated with a low oxygen concentration ([O] ~ 5 X 10^^ cm~^). The observed compensation could be explained by the amphoteric behavior of silicon (N type -^ P type material) due to the high nitrogen vacancy concentration generated in GaAlN bulk material with high Al content (low anmionia flow during the growth), or by the presence of a deep level bound to Al. More investigation studies have to be done on this specific aspect. The physical properties of the GaAlN/GaN bulk material we have grown are summarized in Table 5.1. GaAlN/GaN HEMT heterostructures under study consisted of a 3 ixm insulating GaN buffer layer followed by 0.03 (jim undoped and Si doped GaAlN layer with Al content varying from 22 to 29%. The surface morphology was observed to be "mirror like" for all the heterostructures. This roughly indicates that the high bandgap layer thickness is lower than the critical thickness in the range of the alloy composition studied. The transport properties of such heterostructures have been studied at low temperatures (10-300 K) and at high temperatures (300-500 K). The variation of the hall mobility and sheet carrier density for one of these heterostructures (xî ~ 20%) is illustrated in Figures 5.11 and 5.12. We observe a constant value of the Hall mobility from 100 to 10 K, which can be explained by the formation of a two-dimensional electron gas (2DEG) at the GaAlN/GaN interface of the HEMT heterostructure. Our analysis of undoped and doped HEMT structures revealed that the 2DEG sheet charge density is mainly due to the spontaneous and piezoelectric polarization, while the contribution of the introduction of Si doping into the GaAlN layer plays a secondary role in the formation of such 2DEG sheet charge density. Table 5.1. Physical properties of GaAlN bulk material Run number {R °) Al content XAL(%)

Strain a (MPa)

Sheet resistance

Rnm AEC791 AEC789 AEC792 AEC788 AEC790 AEC805

12.5 15.4 16 19 23 29

1.3 1.75 1.75 2.19 2.5 2.82

1150 1250 1500 2800 10,000 > 20,000

Carrier concentration Nd-Na (cm'^)

Hall carrier concentration

1.5X10^^ 7X10^^ 5 X 10^^ 5-7x10^^

-

-

3 X 10^^

-

Hall mobility fjL (300 K) (cmÂ^ s)

800

-


80 *tuuu •

3000-

1

—r

n ww « * « * * * * * * -f * * * 1

2-

CM

F o^

I

c

1

•

1

E

«

1

1 1

*

1 1

2000-

«* " *

^: !5 o

1

* X

1000900800700600- —f8 10

—1-

20

—^ 40

H — — 1 - —1 60 80100

1

200

400

Temperature (K) Figure 5.11. Temperature dependence of Hall mobility for an GaAlN/GaN HEMT heterostructure.

In the temperature range of 300-500 K, the sheet carrier density appeared to be constant as seen in Figure 5.12, while a 50% reduction in the Hall mobility is observed in the same range of temperature. This behavior of the mobility, also observed in SiC material, will lead to a decrease in the current density and output power density of the associated HEMT device. This is the major reason for the decrease in the device current density with the gate periphery, as previously published. aec 603b 1E13

10000

E o

Q

-J 1000

en O

J2

o £

1E12

100 300

350

400 450 Temperature (K)

500

Figure 5.12. Temperature dependence of Hall mobility and sheet carrier density for a GaAlN/GaN HEMT heterostructure.

MOCVD Growth of Group III Nitrides Btructufe hemt

Omega 63.04700 Phi4B.OO 2Theta126.06503 Psi 000

me

81

XO.DO YO.00

1K-

/

100-

\

"

tAptjniiteriL

\

-

Simulation

/ 101-

1

"^^

fM^*^f^^rt#^f^|,,

0.10.010,001-

-,

jiiiiti jiilitii i^wll^llllJ^J •SOflO

4000

-2000

1

li

U[.i,,,lT"iWp'i'l

0

2000

WL^

,

,

4O0S

,—Emu—1—ft~JR—3Kk—jJWWWfc—,

u

6000 ArgbClseconcte)

Figure 5.13. X-ray rocking curves of a GaAlN/GaN HEMT heterostructure (experiment and simulation).

The high-mobility value obtained at low temperatures (/177 K = 3670 cvc^fW s) for the above-mentioned HEMT structure reveals a good quality of the GaAlN/GaN interface, in good agreement with the structural properties checked by HR-XRD and TEM measurements. Figure 5.13 shows a nice fit of the experimental and calculated X-ray rocking curves related to the mentioned GaAlN/GaN HEMT heterostructure. The main characteristics of the HEMT heterostructure extracted from this profile have been the following: thickness of each layer (woaAiN = 30 nm, WoaN = 3 ixm), Al content of the high-bandgap layer (xî = 19%). The observation of fringes on the experimental spectrum, which suggests a good structural quality of the HEMT heterostructure, was confirmed at the atomic scale by TEM (Figure 5.14). The GaAlN/GaN interface is perfectly delineated, as observed in Figure 5.14, and 57 monolayers (5.18 A) can be counted in the whole thickness of the GaAlN layer which fit the expected 300 A already measured using X-ray.

AEC 603

GaAIN

(JiiN

Figure 5.14.

TEM cross section of a GaAlN/GaN HEMT structure.


82 HEMT

AecS04.zfl6

3 11

OmegaG40K00 Phi 4 00 2Theta 128.010)0 Psi 0.03

Aec811.z00

Y0.00

Counls/s

Aec808.zQ0 Aec809.z00

\

llf-

/

J

im-

Aec810.z00 - Aec807.z00

\

10-

'^^Wl^

^ -2906

0

»m

20D0

I^^MM^^

Figure 5.15, X-ray rocking curves of six GaAlN/GaN/Al203 HEMT structures.

A good reproducibility of the growth process is illustrated in Figure 5.15, which shows the X-ray rocking curves of six HEMT structures which have been grown with the same growth parameters and following the same structure design. A nice superimposition of the different figures is observed which fit with an Al content of 22.5%. The good electrical characteristics of such HEMT structures: room temperature mobility higher than 1000 cvc^/W s associated with a high sheet carrier density of 1.4 X 10^^ cm~^, and a low sheet resistance of about 500 H, have been confirmed by the high performance of the corresponding HEMT devices. A successful comparison of our HEMT structures' electrical data (/i, A^^, /?•) with the already pubUshed data is shown in Figure 5.16. 2000

> lioe

I

•

m&^f Ridl»y # t l i

ABU

ATM! \ A •

" ^ S ^ ^ C I U SSBE

IC I f ^ ^ p , ^^^1*^

C

o 1000 E

a.

§00 *

Silicon Carbide Substrates

A NRCAPL,75,19iS.953 '

^

• .

6

7t

DERA data on SiC

THALES

r

8

I

10

12

I

I

14

I

I

I

It

IZJ

I I

I—I

fi

Sh^st Carrier Cons^ *10e12 [em*^

Figure 5.16.

GaAlN/GaN HEMT carrier concentration versus mobility data.

I

I ft

m


83

5,3,1.3 Device Results, Device processing has been performed using a conventional mesa isolated process with chlorine RIE (SiCl3, BCI3) for mesa definition. Either Ti/Al or Ti/Al/Ni/Au metallisations were used for ohmic contact formation and Pt/Au or Ir/Au for the gate metal. The devices were realized with gate lengths (Lg) of 0.5,0.3 and 0.15 juim. Typical ohmic contact resistances of 0.3-0.5 fit mm were obtained after RTA annealing of the metal alloys (850°C, 30 s) while gate contacts with an ideality factor lower than 2 and a Schottky barrier of about 0.8 eV have been achieved. The processed wafers have shown a good homogeneity within an edge exclusion of about 3-5 mm. The mapping of the device-related parameters such as the sheet resistivity of the active layer (Figure 5.17), the maximum saturation current /^ss, the pinch-off voltage Vp (Figure 5.17) and the transconductance g^ are in good agreement with the variation of the basic material properties {/JL, NS, XAI, etc.). For 0.5 juim X 100 |xm devices, we have obtained a typical g^ between 180 and 200 mS and an /^ss maximum of 0.85 A/mm at + 1 V. The bar chart given in Figure 5.18 depicts the main device results (/^ss, ,^m' ^p) of two processed 2-in. diameter GaAlN/GaN HEMT wafers, showing homogeneous data. This indicates stable processing conditions for both epitaxy and processing of the wafers. The small signal microwave performances of 0.5 ixm X 100 jjim devices from the above wafer (AEC837) are shown in Figure 5.19. They show a similar unilateral gain around 25 db at 2 GHz, current-gain and power-gain cut-off frequencies (fi and/jâx)^ respectively, of 20 and 108 GHz. Load-pull measurements performed at 2 GHz on such wafers (AEC837 and AEC842) have shown remarkably high output power density and absolute power level for devices from wafer AEC837. The measurements have been performed on wafer without additional thermal coupling of the devices.

1

2

3

m •If

• •

368

s4;r

m 579

5

& «J

S41

n

m l«

• AEC 837 Figure 5.17.

7

8

9

10

1

2

3

"' 6

7

a

•3

10

Ugs ab/Y 4 770

M'i

-m.

m

saa

m

K>1

~m

a«6 S^

\sn

m

4^ .4?^

S3S

H

4

4f5

533

s^

5M 5?S

^

•

e m &n

4

im

m

H

-m

m

.4J1B

^.m

•AM%

>iSi

S

,

-^-i^;

4J14 4 5S2 4510

-48^

^

4.45B 4&. 4 496

'm

> AEC 837

tt

4JS4

441

Aim

Sheet resistance and pinch-off voltage mapping on a 2-in. GaAlN/GaN HEMT wafer.

84


I WW

AEC 837

AEC 842

Wafers

Figure 5.18.

Summary of device data such as ^^-max, /^^s, Vp from representative GaAlN/GaN 2-in. wafers in bar chart form.

Small devices (0.5 |xm X 100 luim) exhibited at 2 GHz a maximum output power density of 4.2 W/mm with a PAE of 45% as shown in Figure 5.20a. The scaling of output power and power density with device size is given in Figure 5.20b. The significant reduction of power density with increase in the transistor size is due to thermal effects. Nevertheless, the absolute power level of 3.2 W for 1 mm devices on sapphire substrates measured on wafer without efficient coupling to a heat sink is in good agreement with the international state-of-the-art. A summary of the microwave data (output power and PAE) before and after passivation are depicted in bar chart form in Figure 5.21a and b. The passivation increases both

50 ,

•

ft= 19.9 GHz fmax=108GHzJ

•

40 30 J

20 J

0 V g s = - 2 V , Vds = 20V, Lg=0.5Mm -10

^^•t^^^n^m

10

0.1

"W^r^^^

f (GHz) Figure 5.19.

Small signal microwave properties of typical 0.5 |xm X 100 jxm devices from wafer AEC837.


-15 -10

-5

0

400

5

85

600

800

Gate width [\\m)

Figure 5.20.

(a) Pout, PAE and gain versus pin for small devices (0.5 jxm X 100 iJim) at 2 GHz; (b) scaling of output power and power density with device size at 2 GHz.

microwave output power and PAE. However the degree of improvement depends on the individual wafer. High material quality GaAlN/GaN HEMT wafers have shown a minor influence on passivation (AEC837). In contrast, the data from wafers with somewhat lower quality (higher defect density) can be improved significantly by the passivation (AEC842).

B f f l g RF-power before passrâtion ir::..;.l RF-power final measurement

AEC 837 AEC 842 Wafers

^ • M PAE: before passivaljon P ^ ^ PAE: final meaBuremenl

AEC 837

AEC 842

Wafers Figure 5.21. (a) Summary of representative microwave data (output power) at 2 GHz before and after passivation (0.5 |xm X 100 jxm devices); (b) summary of representative microwave data (PAE) at 2 GHz before and after passivation (0.5 jjim X 100 |xm devices).

86


5,3,2 Silicon Carbide Substrate 5,3,2,1 Experimental Procedure, GaAlN/GaN heterostructures with 22% Al content were grown by MOVPE in a single wafer 200RF AIXTRON reactor on semi-insulating on-axis 4H-SiC substrates, using triethylgallium (TEG), trimethylaluminum (TMA), and ammonia (NH3) as group III and group V precursors. The growth has been performed at low pressure (50 mb), high temperature (1160°C), and the growth rate was roughly 0.5 |jLm/h for GaAlN and 1.2 juim/h for GaN. The GaAlN/GaN HEMT structures under study consisted of GaN or GaAlN nucleation layers, followed by a 1 ixm thick insulating GaN buffer layer, then a 27 nm Si doped GaAlN layer with 22% Al content, and finally a 3 nm thick undoped GaN cap layer. The transport properties of such heterostructures were studied as a function of temperature from 100 to 500 K. Hall mobihty about 1100 cmÂ^ s at 300 K has been obtained, associated with a sheet carrier density of 1.2 X 10^^ cm~^. Since HEMT devices are expected to operate at high temperatures, there is considerable interest to study the influence of temperature on electron mobility. Such an influence can explain the limitation in power and efficiency of the HEMT devices at large drain bias. As shown in Figure 5.22, increasing temperature from room temperature to 500 K reduces the electron mobility by a factor of 2.5. Several growth parameters have been investigated, aiming at optimizing the material quality of the GaAlN/GaN HEMT structures, namely, the growth temperature, thickness and composition of the nucleation layer, and the SiC surface preparation. Various characterization techniques such as atomic force microscopy (AFM), TEM, HR-XRD,

1E14

-i

•

1

•

1

•

E o

t.

n 10000

r-

D— Mobility n— Sheet Carrier Density 1E13

'°^.• ° ^ „

^ 1000

n—D—D-HIl—13—[11-^3—D—D^KD;:;a~D—n—D—D-H^

(0

Si

O

"^^-O

1E12 100 150 200 250

o

100 300 350 400 450 500 550

Temperature (K) Figure 5.22.

Sheet carrier density and hall mobiHty of a GaAlN/GaN HEMT structure as a function of temperature.


87

capacitance-voltage (C-V) and Eddy current probe measurements have been used again for this study. 5,3.2,2 Material Property Results, Due to the large lattice mismatch between GaN and SiC (Aa/a ~ 3.5%), the deposition of the first monolayers is a critical point of the MOCVD process of GaN on SiC. Three major steps in the GaAlN/GaN/SiC growth process have been identified and their impact on the physical properties of the HEMT structures clearly demonstrated [11]: • •

nucleation layer growth temperature, nucleation layer composition. substrate surface preparation.

5,3,3 GaN Nucleation Layer Our first approach of the GaAlN/GaN/SiC growth studies has been concerned with the optimization of HEMT structures based on GaN nucleation layers. The growth temperature of the first GaN monolayers has been identified to have a marked influence on the structural properties of the GaAlN/GaN/SiC HEMT structures. At low growth temperatures, HEMT structures with a ''mirror like" surface morphology are obtained, while a three-dimensional growth process is observed at higher growth temperatures. This observation has been confirmed by HR-XRD measurements. As clearly seen in Figure 5.23, the rocking curves related to GaAlN/GaN/SiC HEMT structures grown at two different GaN nucleation layer growth temperatures, namely 985 and 990°C, exhibit very different X-ray signatures. A good crystallographic quality of the HEMT structure

10^ SiC

GaN FWHIVi=105arcsec FWHM=150arcsec Q.

63

64

65

67

Omega/2Theta (°) Figure 5.23. X-ray rocking curves of two GaAlN/GaN HEMT structures, for two different nucleation layer growth temperatures.

88


Figure 5.24.

TEM cross section of a GaN/SiC interface (low growth temperature GaN nucleation layer).

deposited at the lower GaN nucleation layer growth temperature is observed, evidenced by well-delineated peak satellites, up to the second order, of a superlattice structure implemented in the GaN buffer layer, which is indicative of sharp interfaces. TEM analysis revealed that GaN/SiC interfaces grown at low temperatures (980°C) are indeed sharp with steps originating from the substrate misorientation (Figure 5.24), while more ill-defined at higher growth temperatures. 5.3.4

Substrate Surface Preparation

We have observed that the SiC surface preparation is another key point of the GaAlN/GaN growth process on SiC, as illustrated below. AFM measurements performed on as-received substrates have confirmed the poor quality of the surface preparation realized by the substrate suppliers, as seen in Figure 5.25a. A "specific" surface preparation of SiC substrates by Novasic has allowed to obtain a very good surface morphology characterized by an RMS close to 0.7 A and steps at the atomic level (Figure 5.25b).

a) RMS = 6.9 A

b) RMS = 0.7 A

Figure 5.25.

Surface morphology of an "as-received" SiC substrate (a) and after a Novasic surface preparation; (b) 5 |xm X 5 |jLm surface area.


89

Moreover, correlated with this ex situ specific surface preparation by Novasic, we have observed that an in situ anneaUng of the SiC substrate at high temperatures (~ 1000°C) under hydrogen flow, has led to a real breakthrough of the crystalline quality of the GaAlN/GaN epilayers. A reduction of 50% of the GaN rocking curve FWHM (FWHM = 50 arcsec) has been measured as compared to the GaAlN/GaN epilayers grown on as-received substrates (FWHM = 1 0 0 arcsec). The SiC surface preparation also plays an important role on the electrical properties of such GaAlN/GaN HEMT structures. GaAlN/GaN HEMT structures grown on as-received SiC substrates have presented very high sheet resistances {Ru = 4000-15,000 d ) , very far from the targeted value of 500 fl obtained on SiC substrates with the specific surface preparation mentioned above. This evolution of the sheet resistance with the substrate surface preparation may be explained by a strong increase in the piezoelectric effect induced by strain or by the generation of deep traps due to the high density of defects at the GaN/SiC interface. Indeed, due to some surface kinetic aspects, the use of a GaN nucleation layer favors an island growth mode for the first monolayers on SiC. This growth mode would generate more tensile strain and defects in the GaN buffer layer. The substrate surface preparation could play, in this aspect, a catalyst role. This was confirmed by HR-XRD measurements. The GaN buffer lattice parameter (c) was found to be close to 5.158 A for the highly resistive HEMT structures as compared to the 5.1850 A for the unstrained GaN lattice. 5.3,5 GaAlN Nucleation Layers The strong influence of the substrate surface preparation and growth temperature on the growth of the first atomic layers on SiC makes the GaAlN/GaN/SiC growth process clearly non-reproducible using a GaN nucleation layer. In order to improve the reliability of the MOVPE process, we have investigated the growth of GaAlN/GaN/SiC HEMT structures with a GaAlN (Al content = 25%) based nucleation layer. In this case, the growth of the nucleation layer on SiC starts using a step flow mode, leading to a GaN buffer layer also grown in a step flow mode and so less tensilely strained in the HEMT structure. Indeed, the GaN buffer lattice parameter, checked by X-ray, was found to be close to 5.17 A in that case. As a rule, we found a clear improvement of the physical properties of the GaAlN/GaN HEMT structures, as compared to HEMT structures with GaN nucleation layers, with a good reproducibility from run to run. Due to less strain and less defects in the HEMT structures, the residual carrier level in the GaN buffer, checked by C- V measurements, was found to be a decade lower as depicted in Figure 5.26. Mapping of the sheet resistance and pinch-off voltage using C-V and Eddy current probe measurements performed on such GaAlN-based nucleation layer HEMT structures reveal a good homogeneity of these electrical characteristics on 2-in. wafers, as seen in Figure 5.27. Figure 5.28 shows an AFM image of the surface of such a layer and confirms a very good crystaUine quality of the GaAlN/GaN epilayers with a RMS close to 0.3 nm.

90

Optoelectronic Devices: Ill-Nitrides GaN nucleation layer

GaAIN nucleation layer

1.E+20

'R 0.4

Figure 5.26.

0.6 0.8 Depth (|im)

1

2 Depth (|im)

Capacitance-voltage depth profiles of two GaAlN/GaN HEMT wafers with GaN and GaAIN nucleation layers.

53,6 Device Results Devices related to GaAlN/GaN/SiC HEMT structures based on GaN and GaAIN nucleation layers have been fabricated using a conventional mesa isolated process with RIE for mesa definition. Ti/Al/Ni/Au was used for ohmic contact formation and Pt/Au for the gate metal. The devices were realized with gate length (Lg) of 0.5, 0.3 and 0.25 |xm. Under static measurements we found a maximum drain current /^ss around 1 A/mm and a pinch-off voltage of - 5 V for devices with a gate length of 0.5 juim and a GaN nucleation layer. Better static results have been obtained for devices with the same geometry but with GaAIN nucleation layer. As a matter of fact, an /^ss up to 1.5 A/mm has been recorded [12]. Devices related to HEMT wafers with GaN nucleation layers have been measured at 10 GHz using a load-pull system. They exhibited a CW output power in excess of 2.8 W/mm for a gate length of 0.5 luim, as shown in Figure 5.29.

Figure 5.27.

Sheet resistance and V pinch-ofiF mapping of a 2-in. GaAlN/GaN/SiC HEMt wafer.


Figure 5.28.

91

Surface morphology of a 2-in. GaAlN/GaN/SiC HEMT wafer: RMS = 0.3 nm.

At 2 GHz, we have obtained an absolute output power in excess of 6.5 W for 2 mm X 0.3 ^lm devices and a maximum output power density of 3.5 W/mm has been reached for 1 mm X 0.3 (xm devices. The scahng of the absolute output power and power density versus the gate width of the devices is given in Figure 5.30a and b. A comparison of our results with data from different wafer suppliers is made (Figure 5.30a and b).

5

Figure 5.29.

10 15 Input Power (dBm)

CW performance at 10 GHz of a 0.3 mm X 0.5 ^jim device measured under probes.

92

Optoelectronic Devices: Ill-Nitrides (a)

1

Ha

"ififtiaio

1 1

lfirl»

1

Id

„..._

a1 4 4

4

m

._._

_

•

1

•1

1x128

• •1

• C»4M4W m QCMM-OT

V

4

a

1

w-NBS130C—

_ ^

This work T

1

1

2

1

1 imnri

1

.•-•—..-.^

4

1

(b) « 123C80

tnas

Itxlli — * —

44

immb

3 pr

' 2

3

This work 4

Qitipwidiii lifiin) Figure 5.30.

(a) Scaling of the absolute power versus gate width measured at 2 GHz (Lg = 0.3 [xm). (b) Scaling of the power density versus gate width measured at 2 GHz (0.3 jxm).

The shorter gate of the devices (Lg = 0.3 |xm as compared to Lg = 0.5 U | Lm for the former device process) has led to significantly higher cut-off frequencies: /t increased from 20 to 35 GHz whereas /âx is higher than 100 GHz.

93

MOCVD Growth of Group III Nitrides Vds = 30V and VQS = -ZVat 10GHz

-PoLrt[dBm] Gain[dB] ^PAE(%)

10 15 Pabs (dBm) Figure 5.31. Load-pull measurement at 10 GHz of a 0.3 mm X 0.25 ixm GaAlN/GaN/SiC HEMT device.

An improvement of the microwave performances has been observed for devices issued from HEMT wafers with GaAlN nucleation layer. CW on-wafer load-pull measurements have been performed at 10 GHz. For 0.3 mm X 0.25 ixm devices, a power density of 4 W/mm has been obtained with a PAE of 28% as shown in Figure 5.31.

5.4.

CONCLUSION

GaN MESFET structures and GaAlN/GaN HEMT structures have been grown on sapphire and silicon carbide by low-pressure MOVPE. The influence of some critical growth parameters on the physical properties of the device structures has been identified and their optimization has led to high-microwave performance devices such as CW output power up to 4 W/mm at 10 GHz for GaAlN/GaN HEMTs on SiC.

ACKNOWLEDGEMENTS The authors would like to thank C. Brylinki and J.-C. Jacquet for fruitful discussions and a careful reading of the manuscript, B. Grimbert, E. Morvan, M. Laurent, N. Caillas, V. Hoel and J. Wtirfl for device processing, E. Delos, D. Ducatteau, C. Gaquiere and J. Graffeuil for microwave and noise measurements, R. Aubry, N. Sarazin, M. Peschang, D. Lancefield, R. Seitz, E. Pereira, D. Theron and the Novasic Company for material characterization, S. Delage and J.-C. De Jaeger in charge of GaN project at TRT/IEMN common laboratory and D. Pons who launched the GaN program at TRT. The research work presented here has been supported by the French Ministry of Defence (DGA/STTC).

94

Optoelectronic

Devices:

Ill-Nitrides

REFERENCES [1] Shealy, J.R., Kaper, V., Tilak, V., Prunty, T., Smart, J.A., Green, B. & Eastman, L.F. (2002) An AlGaN/GaN high-electron-mobility transistor with an AIN sub-buffer layer. /. Condens. Matter Phys., 14, 3499-3509. [2] Hobgood, H.McD. (2003) Silicon carbide crystal and substrate technology: a survey of recent advances, ICSCRM2003, Proc, Lyon, p. 46. [3] Micovic, M., Nguyen, N.X., Janke, P., Wong, W.S., Hashimoto, P., McCray, L.M. & Nguyen, C. (2000) Electron. Lett., 36, 358. [4] di Forte-Poisson, M.A., Huet, F., Romann, A., Tordjman, M., Trassaert, S., Boudart, B., Theron, D., Seitz, R. & Pereira, E. (1999) LP-MOCVD growth of GaN MESFETs, Proceedings of the European Workshop on Metal Organic Vapor Phase Epitaxy VIII, Prague, p. 77. [5] Leroux, M., Beaumont, B., Grandjean, N., Massies, J. & Gibart, P., (1996) Characterization of near band edge optical transitions in undoped GaN/Sapphire, grown by MVPE, HVPE and GSMBE, Proceedings of MRS Fall Meeting, Boston 1996. [6] Trassaert, S. (2000) Realisation Technologique de transistors aeffet de champ dans les fiUeres InP et GaN pour amplification de puissance Hyperfrequence, Thase de doctorat de I'Universite des sciences et technologies de Lille, 4 Fevrier 2000. [7] Balandin, A., Cai, S., Li, R., Wang, K.L., Ramgopal, V. & Viswanathan, C.R. (1998) Flicker noise in GaN/AlGaN doped channel heterostructure field effect transistors. IEEE Electron. Dev. Lett., 19 (12), 475-477. [8] Balandin, A., Morozov, S.V., Cai, S., Li, R., Wang, K.L., Wijeratne, G. & Viswanathan, C.R. (1999) Low flicker-noise GaN/AlGaN heterostructure field-effect transistors for microwave communications. IEEE Trans. Microwave Theory Tech., 47 (8), 1413-1471. [9] Levinshtein, M.E., Rumyantsev, S.L., Gaska, R., Yang, J.W. & Shur, M.S. (1998) AlGaN/GaN high electron mobility field effect transistors with low 1// noise. Appl. Phys. Lett., 37 (8), 1089-1091. [10] di Forte-Poisson, M.A., Romann, A., Tordjman, M., Dessertenne, B., Cassette, S., Surrugue, M., Frapsauce, N., Adam, D., Delage, S.L., Boudart, B., Gaquiere, C , Vellas, N., Lancefield, D. & di Persio, J. (2001) LP-MOCVD growth of GaAlN on sapphire. Application to HEMT's devices. Proceedings of the European Workshop on Metal Organic Vapor Phase Epitaxy IX, Wrexham, p. 115. [11] di Forte-Poisson, M.-A., Romann, A., Tordjman, M., Magis, M., Di Persio, J., Jacques, Ch. & Vicente, P. (2003) LP-MOCVD growth of GaN on Silicon Carbide. /. Cryst. Growth, 248, 533-538. [12] di Forte-Poisson, M.-A., Magis, M., Tordjman, J., Aubry, R., Peschang, M., Delage, S.L., Di Persio, J., Grimbert, B., Hoel, V., Delos, E., Ducatteau, D. & Gaquiere, C. (2003) LP-MOCVD growth of GaAlN/GaN heterostructures on Silicon Carbide. Application to HEMT's devices. Proceedings of MRS Fall Meeting, Boston.


Chapter 6

Growth of Nitride Quantum Dots Hyun Jin Kim, Soon-Yong Kwon and Euijoon Yoon School of Materials Science and Engineering, Seoul National University, Seoul 151-742, South Korea

6.1. INTRODUCTION For the past few decades, low-dimensional quantum structures, such as quantum wells (QWs), quantum wires (QWRs), and quantum dots (QDs) have been attracting lots of interest due to their potential advantages compared with bulk materials (Figure 6.1). Among these, QDs are expected to be the most promising due to their unique electronic states, such as 5-function-like density of states, three-dimensional (3D) carrier confinement, etc. Due to their unique properties, the semiconductor laser with a QD active layer is expected to have ultra-low threshold current, reduced temperature sensitivity, narrower spectral line width, and high-modulation bandwidth, etc. [1,2]. Furthermore, the semiconductor photodetector with QDs are also expected to have the sensitivity for the normally incident light, enhanced photoexcited carrier lifetime, reduced dark current, and higher electric gain [3]. It was in 1982 that the concept of QDs was proposed for the first time as artificial atoms for semiconductor laser application by Arakawa and Sasaki [4]. Since then, there has been lots of research devoted to the realization of predicted potential advantages of QDs. However, it took about 10 years to realize the fabrication of the practical QD structures. In the 1990s, both selective growth and self-assembled growth technique without the formation of nonradiative defects were well developed. Particularly, the StranskiKrastanow (SK) growth mode was very successful for InGaAs/GaAs systems [5-7]. As a result, lasers, detectors for both inter-band and inter-subband transitions have been successfully demonstrated using the InGaAs/GaAs QDs [8-10]. From the late 1990s this material system has been extended to other material systems. Among those material systems, nitride semiconductors have received the most attention for applications to blue and ultraviolet (UV) light emitting devices, especially, due to the possibility of much bigger impact of QDs in GaN-based LDs [11]. In nitride semiconductors, the larger effective mass of electrons nic or the larger ratio of the effective mass of holes m^ to rric {mjm^), compared to GaAs-based semiconductors.

E-mail address: [email protected] (E. Yoon).

95

96

Optoelectronic Devices: Ill-Nitrides Bulk

m

Quantum Film

KH

Quantum Wire

Quantum Dot

ii> -io>

^LJVJ Energy

Energy

Energy

Energy

Figure 6.1. Density of states in bulk material, quantum wells, quantum wires, and quantum dots.

results in an increase in the threshold current density of QW lasers: the threshold current density /th of GaN-based QW lasers is expected to be ~ 1 kA/cm^, whereas that of GaAsbased QW lasers to be ~ 100 A/cm^. On the other hand, it was suggested that, if QDs are used in the active region and the size of the QDs is small enough that the population of carriers in the higher subband can be ignored, the achievable threshold current /th in both GaAs-based and GaN-based LDs is almost the same, about 100 nA-1 mA, as shown in Figure 6.2. Therefore, with the use of QDs, the threshold current density is reduced by a factor of 100 in GaN-based lasers compared to GaAs-based lasers, suggesting that the impact of QDs is much bigger in GaN-based LDs than in GaAs-LDs. The promising aspect in the applications of the nitride QDs is also confirmed in the emission mechanism of the InGaN active layer. InGaN active layers for blue and green light emitting diodes (LEDs) are known to have excellent optical properties despite the presence of high-density defects. It is widely accepted that their high-luminescence efficiency is due to the carrier localization induced by the presence of compositional fluctuation and/or QD-like features usually observed in InGaN layers of sufficiently high In content [12]. Therefore, the introduction of QDs would also increase the luminescence efficiency in the system of high dislocation density and especially in material systems without compositional fluctuation such as low In content InGaN, AlGaN, etc. Since the late 1990s, several methods to form QDs using nitride semiconductors were suggested. They can be categorized into four methods: (1) using SK growth mode, (2) using "anti-surfactant", (3) using selective epitaxy, and (4) other novel methods. One of the most attractive methods for defect-free QD formation is the SK growth in lattice-mismatched semiconductor systems, widely used in the QD fabrication of other material systems, such as InGaAs/GaAs [5-9], InAs/InP [10,13], InP/GaAs [14,15],

Growth of Nitride Quantum Dots

97

Gan QW Lasers

(5

QD Lasers 4 6 m,, / m_ Figure 6.2.

8

10

Carrier density at transparency condition plotted as a function of the ratio of effective mass of holes to that of electrons for quantum well (QW) lasers and quantum dot (QD) lasers [11].

GaSb/GaAs [16], InSb/InP [17], SiGe/Si [18], ZnCdSe/ZnSe [19], etc. The schematic diagram of the SK growth mode is presented in Figure 6.3. In the SK growth mode, the mismatched epitaxy is initially accommodated by biaxial compression in a layer-bylayer (2D) growth region, traditionally called as "wetting layer". After the deposition of a few monolayers, the strain energy builds up. Then, the evolution to 3D islands becomes more favorable than the continued, strained planar growth. Such islands are referred to as self-assembled QDs (SAQDs).

Volmer-Wfeber mode

•••••••••••••H

••••••••••!

Figure 6.3. Schematic diagram of typical 3D growth mode (Volmer-Weber mode) and the StranskiKrastanow (SK) growth mode. A wetting layer between the substrate and the quantum dots is present in SK growth mode.

98

Optoelectronic Devices: Ill-Nitrides In contents of InGaN (%) 20 40 60 80

100

2 4 6 8 ID Mismatch of InGalSJ on GaN(%) \

4100^

AIN

CD

P

50 ^

s

Ei c

4

(Ai O

0)

c

0)

1^0 O

GaN

Q. (Q

4o

c o o

G)

I 2 QQ

InN 3.1

^

3.2

3.3 3.4 Lattice constant (A)

3.6

3.6

Figure 6.4. Bandgap energies of AIN, GaN, and InN vs their lattice constants. Ticks and labels in top axis and right axis show the variation in lattice mismatch of InGaN grown on GaN and that of GaN grown on AlGaN, respectively.

Nitride semiconductors can be epitaxially grown to form a strained heterostructure, which is indispensable for the SK growth mode. Figure 6.4 shows the lattice constants of AIN, GaN, and InN compared with their bandgap energies. As shown in Figure 6.4, the lattice mismatch between AlGaN and GaN ranges from 0 to 2.4%, and that between InGaN and GaN ranges from 0 to 11.1%, which is sufficiently large to have all growth modes (the strained 2D growth mode, the SK growth mode, and the relaxed 3D growth mode). By the proper combination of QD and substrate materials, the extent of strain in the film and subsequent QD growth behavior can be controlled. It is also worth noting that the large difference in their bandgap energies in Figure 6.4 (from 0.7 eV (InN) to 6.2 eV (AIN)) is also advantageous for the realization of heterostructure with large band-offsets, which would enhance the carrier localization efficiency of QDs. The nitride QDs were also demonstrated by using "anti-surfactant". The pretreatment of the growth surface by anti-surfactant, such as tetraethylsilane (TESi, Si(C2H5)4), were found to result in the 3D growth mode of the subsequent layer, even in the case with little lattice-mismatch. The presence of the wetting layer was not confirmed in the QDs grown by this method. Since this growth mode does not require lattice mismatch, the restriction in the choice of film/substrate combination can be somewhat relaxed.


99

The method using selective epitaxy attempted in the nitride QD formation is similar to the method of patterning QWs, generally used in other III-V compound semiconductors. In this case, at first, pyramidal structures are grown on a patterned substrate using selective epitaxy. Then, QDs are grown on top of those pyramidal structures. However, this method has similar advantages with the method of patterning QWs, such as (1) the possibility of almost arbitrary lateral shape, size, and position realization, (2) general compatibility with the modem ULSI technologies in the patterning process of this method. However, there are some drawbacks, such as defects or damages induced by etching process. Besides the techniques described above, other novel techniques were also attempted to fabricate the nitride QDs, such as the nitridation of metallic droplets [20-22], the surface passivation and/or pretreatment [23-25], the laser ablation [26], the ion implantation [27], the colloidal synthesis [28], etc., by several research groups. In this chapter, various techniques to form nitride QDs are reviewed, and the current status of nitride QD research is discussed. In particular, emphasis is put on three major growth techniques such as growth of strain-induced SK QDs, growth by anti-surfactant, and growth by selective epitaxy. Table 6.1 summarizes the three major techniques for nitride QD formation and material systems explored. Effects of various parameters in each growth techniques and changes in physical properties are reviewed. Finally, novel formation techniques are briefly reviewed.

Table 6.1. Summary of three major techniques to form nitride quantum dots Techniques SAQDs by SK growth mode

Organization

Authors

Growth

Material system

CEA-Grenoble

Daudin et al.

MBE

CNRS

Damilano et al.

MBE

University of Tokyo Seoul National University University of Montpellier II

Tachibana et al.

MOCVD

Kim et al.

MOCVD

GaN/AlN [29-31,37]; InGaN/GaN [35] GaN/AlN/Si [32]; InGaN/GaN [33,34,36] GaN/AlN/6H-SiC [38]; InGaN/GaN [39-41] In-rich InGaN/GaN [42,43]

Briot et al.

MOCVD

InN/GaN [45]

QDs using anti-surfactant

RIKEN

Tanaka et al.

MOCVD; MBE

GaN/AlGaN/6H-SiC [49-53]; InGaN/GaN/6H-SiC [54]

QDs by selective epitaxy

University of Tokyo University of Tokushima

Tachibana et al.

MOCVD

Wang et al.

MOCVD

InGaN/GaN [55]; GaN/AlGaN [56] InGaN/GaN [57,58]

100 6.2.

Optoelectronic Devices: Ill-Nitrides GROWTH OF STRAIN-INDUCED QUANTUM DOTS

6,2,1 Self-assembled Quantum Dots by MBE One of the most attractive methods to form defect free QDs is the SK growth in latticemismatched systems. The growth of SAQDs by the SK growth mode has been successfully demonstrated using both MBE and MOCVD. The GaN QD growth using SK growth mode was reported for the first time by Daudin et al. [29]. They monitored the changes in growth mode of GaN on AIN by reflection high-energy electron diffraction (RHEED). Their growth of (0001) AIN and GaN with wurtzite structure was carried out by MBE on (0001) sapphire substrates. Atomic nitrogen was produced by an RF plasma source. After nitridation of a sapphire substrate by exposing it to nitrogen plasma, a thin (about 15 ml) AIN buffer was deposited at a substrate temperature T^ of 500°C, followed by the growth of a 2-|ULm-thick GaN buffer and of a 200-nm-thick AIN layer at 650°C. AIN layer was fully relaxed with respect to the GaN buffer, judging from the changes in RHEED streak spacing. Subsequently, GaN epilayer was deposited on AIN and its relaxation was studied by RHEED. The growth behavior of GaN on AIN was rather different depending on substrate temperature, 7^, as shown in Figure 6.5. At 720''C, the Bragg spot intensity increased after deposition of 2 ml, which was attributed to 2D/3D growth mode transition. For further GaN deposition, the Bragg spot intensity remained constant, indicating a persistent 3D growth mode. In contrast, at 620°C, the indication of 2D/3D transition is hardly observed and the Bragg spot intensity decreased rapidly down to the value corresponding to the reflectivity of a smooth GaN surface, suggesting a fast recovery of 2D growth. On the other hand, at intermediate temperature ranges, for example, at 690°C, the increase in the Bragg spot intensity after the deposition of 2 ml GaN was followed by a rapid decrease after deposition of about 6 ml. This rapid decrease in intensity was found to be associated with the transition of streaky RHEED pattern to spotty pattern. By precise control of SK growth mode by RHEED observation, GaN QDs were obtained [29,30]. The image of a smooth AIN base layer prior to GaN deposition is shown in Figure 6.6(a). When about 4 ml GaN layers were deposited on the AIN surface, GaN QDs were formed as shown in Figure 6.6(b). They are typically 10 nm wide, 2 nm high and their density was about 5 X 10^^ cm~^. The size and density of QDs could be further controlled by using growth interruption. Figure 6.6(c) shows GaN QDs obtained at 710°C by depositing 2 ml GaN, followed by exposure to N plasma for 50 s. Clearly, the QD density was reduced (5 X 10^^ cm~^), however, their size became larger (25 nm wide and 5 nm high) than in the previous case, due to the coalescence process occurring under atomic nitrogen flux. Inclined RHEED streaks were observed in this case, indicating that pyramidal QDs with sixfold symmetry and (10-13) facets were formed after the coalescence.

Growth of Nitride Quantum Dots (a)

101

3 :

2.5 [•

•

•

Tg = 720°C

'•

3 1.5 3 0.5 0 ^-0.5

\ f

Pl-^. [[

1

U A ^ /i

-

Time (s)

:

)

2

4

6

8

\

10 •

(0

c 0

Q

10

15

25

Thickness (ML) Figure 6.5.

Change in in-plane lattice parameter (dotted line) and Bragg spot intensity (full line) as a function of the thickness of GaN on AIN [29].

The atmosphere during the growth interruption played an important role in the coalescence process. GaN QD formation at 700°C by depositing 3 ml GaN on AIN surface and subsequent growth interruption under either N plasma or under vacuum for 3 min at 700°C resulted in different surface morphology. Growth interruption under vacuum led to further coalescence process with sparse and larger QDs, as shown in Figure 6.7, presumably due to the increase in the mean-free-path of Ga on N-deficient surface.

102


200 nm

200

nm Figure 6.6. (a) AFM observation of a smooth AIN surface, (b) GaN QDs formed by depositing 4 ml GaN on the AIN surface followed by cooling down under vacuum, (c) GaN QDs formed by depositing 2 ml GaN on the AIN surface followed by exposure to N plasma for 50 s [29].


103

Figure 6.7. 200 nm X 200 nm AFM images for GaN QDs grown on AIN at 700°C and experienced the growth interruption and cooling process (a) under N plasma flux, and (b) under vacuum [30].

The optical properties of the GaN SAQDs could be tailored by controlHng their sizes [31]. PL spectra at 2 K from "small dots" with a typical height of 2.3 nm (8 nm in diameter) and "large dots" with a typical height of 4.1 nm (17 nm in diameter) are compared in Figure 6.8. The PL peak from "small dots" is centered at 3.75 eV, nearly 0.3 eV blue shifted with respect to the GaN bandgap. On the other hand, that from "large dots" in the blue at 2.95 eV, i.e. 0.5 eV below the bulk GaN bandgap energy. This striking QD size effect is attributed to the quantum confined Stark effect caused by the presence of a huge piezoelectric field in the QDs along the c-axis. Estimated piezoelectric field present in these QDs was around 5.5 MV/cm, which is more than one order of magnitude larger than the piezoelectric field found in zinc-blende semiconductors for the same amount of strain. Due to this large value, piezoelectric field effects are dominating for QDs whose heights are larger than only 3 nm [31].

I I I—I I I I I—I I I—|—I—r—i—I I I I—|—r

hT=2K

2.4

Large dots

2.6 2.8

I I I—I—I I I I—I—pi

Small dots

3.0 3.2 3.4 3.6 Photon Energy (eV)

3.8 4.0

Figure 6.8. PL spectra (2 K) of small and large GaN QDs grown on AIN [31].

104


T ^ mm

CD

1000

600

600

400

200

Wavefength (nm) Figure 6.9. Room temperature PL spectra from GaN QDs on A1N/Si(lll). (a), (b), and (c) correspond to a single plane of GaN QD layer with nominal thickness of 7, 10, and 12 ml, respectively, (d) is the PL spectrum from four-layer-stacked QDs, producing white light [32].

Damilano et al. reported the further red shift of PL emission from GaN QD samples of different QD sizes [32]. They could observe the intense room temperature PL emission from blue to even orange, depending on QD size. Corresponding PL spectra of blue, green, and orange QDs are shown in Figure 6.9. They also grew the stacked GaN QD sample containing four planes of QDs of properly chosen sizes to produce white light, as shown in Figure 6.9(d). MBE growth of InGaN QDs by SK growth mode was also reported. The soHd composition of InGaN, consequently the amount of lattice mismatch, controls its growth mode. Grandjean et al. studied the growth of In^Gai_;,N on GaN with RHEED [33]. It is evident that the critical thickness for 2D/3D transition decreased with increasing In composition in InGaN as shown in Figure 6.10, and that InGaN with high In composition would be favorable for the growth of SAQDs. It was found that only 3 ml could be grown layer-by-layer before islanding when In composition was greater than 0.3. When In composition was less than 0.12, the growth mode no longer underwent a strong 2D/3D transition (denoted by open squares), but rather a progressive roughening of the growth surface occurred, most likely due to insufficient driving force for 2D/3D morphological transition. By careful control of the SK growth mode, the successful growth of InGaN SAQDs on GaN by MBE was made by Damilano et al. [34]. The mean size of their typical QDs was


105

15

c10 o 2 Q Q

5h

CM

• . . . . - •

0 Figure 6.10.

10

20 30 40 In composition (%)

50

Critical thickness associated with the 2D/3D growth mode transition at various In composition for the growth strained InGaN on GaN at 600°C [33].

about 35 nm in diameter and 4 nm in height. The QD density was ~ 5 X 10^^ cm~^, which was greater than the dislocation density in the GaN base layer (~5 X 10^ cm~^). The In content in these QDs was 15% and the critical thickness was 4 - 5 ml. Adelmann et al. also reported the growth of InGaN SAQDs on GaN by MBE [35]. When In content of their QDs increased to 35%, the mean QD size was decreased to about 27 nm in diameter and 2.9 nm in height. The QD density was increased to ~ 9 X 10^^cm~^ and the critical thickness was decreased to 2 ml. The changes in QD size affected the PL emission wavelength. Damilano et al. also reported that the PL peak energy from their Ino.2Gao.8N/GaN QDs shifted from 3.03 to 2.51 eV with increasing QD size, as shown in Figure 6.11 [34,36]. Changes in PL emission wavelength were attributed to both a decrease in the carrier quantum confinement energy and an increase in the quantum confined Stark effect. The main parameter which determines the QD energy level for a given In composition is the QD height because of the strong built-in electric field along the c-axis and relatively large QD diameter. Gogneau et al. proposed the modified SK mode to grow GaN QDs on AIN by MBE [37]. The schematic diagram of this technique is presented in Figure 6.12. A GaN layer was grown under Ga-rich conditions with absorbed excess Ga bilayer on the growing surface. The resulting GaN epilayer was 2D and had a flat surface as examined by RHEED (Figure 6.12(a)). After the growth was stopped under Ga flux, the surface remained 2D. The RHEED pattern remains unchanged (Figure 6.12(b)). Under vacuum, however, the Ga film desorbed in a few seconds (Figure 6.12(c)), and the GaN layer transformed into facetted 3D surface. The RHEED pattern evidences this island formation by the presence of additional lines characteristic of facets (Figure 6.12(d)).


106

-,

,

,

^

,

p—,

,

,

J

RcMDm temperature

,

,

,

y-

M\ 3.03 0V

J'2.85 eV

eV

A"'

t\ J.

2.51 eV

2.0

2.2

2.4

2.6

2.8

3.0

3.2

3.4

3.6

Photon energy (eV) Figure 6.11. Room temperature PL spectra of the InGaN/GaN QDs with different sizes. The PL peak energy of 3.03, 2.85, 2.71, and 2.51 eV correspond to the nominal thickness of 1, 1.5, 2, and 3 nm, respectively [36].

(c)

Ga

ftttt

(d)

Figure 6.12. Schematic diagram of the experimental procedure for the growth of GaN SAQDs by a modified SK growth mode. Ga desorption during growth interruption induces the SAQD formation [37].


107

(a)i

Figure 6.13. AFM images (1 |xm X 1 yun) of the high density GaN QDs grown on AIN by MBE at 750°C. The GaN coverages are (a) 2.8 ml, (b) 6 ml, (c) 10 ml, and (d) 13 ml [37].

They found that the formation of GaN QDs by this method was governed by GaN coverage. The resulting GaN QDs grown with different GaN coverages are shown in Figure 6.13. It was observed that the island density increased rapidly from the GaN coverage of 2.8-6 ml before saturation, and followed by a slow decrease. It also appears that the QD density can be controlled over about one order of magnitude

108


(3 X 10 - 2 X 10^^ cm~^ range) for coverages varying from 2.8 to 6 ml. In comparison, the density of GaN QDs grown by SK mode is already 5 X 10^^ cm~^ for a coverage of only 4 ml and saturates at around this value [30,31]- Therefore, the slight modification of growth process in MBE gave us process flexibility to control QD density as well as size.

6,2,2 Self-assembled Quantum Dots by MOCVD The growth of SAQDs using MOCVD was also reported by several groups. Miyamura et al. reported the growth of GaN SAQDs on AIN surface by MOCVD [38]. They grew a 110-nm-thick AIN layer on a (0001) 6H-SiC substrate at 1180°C with an NH3 flow rate of 2.0 slm. After the AIN layer was grown, the growth temperature was reduced to 960-990°C to grow GaN QDs, relatively lower than the typical GaN growth temperature. The growth temperature and V/III ratio were quite important parameters for the formation of GaN QDs. These parameters affected the migration and/or evaporation of Ga atoms on the GaN surface, leading to surface diffusion to form the 3D nanostructures. The flow rate of NH3 must be reduced to 3 seem (corresponding V/III ratio of 30) during the successful growth of GaN QDs to induce SAQD formation.

(a)

3.4 ML

(b)

5.8 ML 5.0 nm

2.5 nm

0.0 nm

9.2 ML

200 nm Figure 6.14.

AFM images (500 nm X 500 nm) of GaN QDs by MOCVD. The GaN coverages are (a) 3.4, (b) 5.8, (c) 6.9, and (d) 9.2 ml, respectively [38].


109

A typical morphology of GaN QDs grown on AIN by MOCVD is shown in Figure 6.14. The growth temperature was 965°C under the V/III ratio of 26. The initial GaN surface exhibited a typical 2D growth mode, as shown in Figure 6.14(a). However, as GaN coverage increased more than 4 ml, 2D/3D growth mode transition began with the formation of QDs by SK mode. The average diameter and height of the QDs were 20 and 2 nm, respectively, by AFM measurement. The density of the QDs could be increased to 5 X 10^^ cm~^ by optimizing the growth temperature and the GaN coverage. PL studies on the GaN QDs revealed the existence of the wetting layer, suggesting that the QDs were surely grown by SK growth mode. Figure 6.15 shows the PL spectra of AINcapped GaN samples at room temperature. The sample of Figure 6.15(a) is grown under GaN deposition of 4 ml, and it may contain the wetting layer and no QDs, as shown in Figure 6.14(a). The samples in Figure 6.15(b) and (c) both contain QDs evolved by increasing the equivalent thickness of GaN deposition more than 4 ml. QDs in Figure 6.15(b) and (c) were of different sizes controlled by V/III ratio and growth temperature. Figure 6.15(b) and (c) shows different intense peaks at around 4.1 and 3.5 eV, respectively. These peaks originate from the QD structure and are consistent with their average sizes. However, these three samples exhibit the same PL peak around 4.7 eV,

1

•

1

•

1.0

^

1 < Q D s (c)

1 • 1 > 1 Q D s (b) W L (a)

'

1

'

1 1-

Peak energy Sample (a): 4.77 eV Sample (b): 4.18 eV Sample (c): 3.59 eV

0.8

M

FWHM Sample (a): 240 meV Sample (b): 480 meV Sample (c): 580 meV

CO

S

0.6

c _i

S

1 /

by A r F excimejr I ^^5cited er(193nm)atRT

0-4

03

E o 2

0.2

0.0

A —

\

1

1

Gan bulk ...

1

2.5

1

1

3.0

1

1

3.5

1

1

4.0

1

1

1

4.5

1

5.0

1

1

5.5

1

i

6.0

Photon Energy [eV] Figure 6.15.

Room temperature PL spectra of GaN QDs grown on AIN by MOCVD [38].

no

Figure 6.16.


AFM images of (a) a GaN buffer layer, and InGaN SAQDs formed on GaN by the deposition of (b) 6.4 ml and (c) 19.1 ml [39].

which is the wetting layer peak, because QD samples grown by SK growth mode would have the wetting layer of almost the same thickness. The growth of InGaN QD on GaN using MOCVD was also reported by Tachibana et al. [39]. They grew 30 nm of a GaN nucleation layer at 500°C and a thick GaN buffer layer at 1075°C on (0001) sapphire substrate. After the growth of the thick GaN buffer layer, growth temperature was reduced to grow InGaN QDs. V/III ratio during InGaN QD growth was kept at about 7000, which was much higher than 30 used in the case of GaN QDs, presumably due to the increased lattice mismatch in InGaN/GaN QDs. The morphologies of the resulting InGaN QD layers with different coverages are shown in Figure 6.16. As InGaN coverage was increased, the density of QDs increased. The In content of the InGaN QDs was estimated as about 25%. The typical InGaN QDs were 19.5 nm wide and 4.5 nm high. The density of typical InGaN QDs was estimated as 6 x 1 0 ^ cm~^, relatively lower than the density of the GaN QDs on AIN addressed above. They further grew a stacked InGaN QD structure and investigated its optical properties [40]. The room temperature PL spectra from the InGaN QD structures with different


111

sjngle-sheet 3-stacked — 10-stacked HO'Cd 20 mW RT

'c CD

c

x10

2.0 Figure 6.17.

2.6 3.0 Energy [eV]

3.5

PL spectra at room temperature from (a) a single-layer InGaN QDs, (b) three-layer stacked InGaN QDs, and (c) 10-layer stacked InGaN QDs [40].

number of stacks are shown in Figure 6.17. Stacking of InGaN QDs resulted in an increase in QD density as well as integrated PL intensity. The PL intensity from the three-layer stacked InGaN QDs is about 40 times higher than that from the single-layer QDs. Moreover, the PL intensity from 10-layer stacked InGaN QDs is about 180 times higher than that from the single-layer QDs. In general, the integrated intensity is proportional to the number of stacks, and it is quite likely that the structural quality of InGaN QDs improved as stacking continued. They also grew a laser structure using this 10-layer stacked InGaN QDs [40,41]. The active layer was sandwiched by the Alo.07Crao.93N cladding layers. The cavity facet was fabricated by reactive ion etching (RIE). The laser structure was characterized at room temperature under optical excitation, and the stimulated emission could be observed as shown in Figure 6.18. Above the threshold excitation power, the width of emission peak

en 'c ID

. RT

^ ro

3^ (0

s

-(c)/=1.8/,,

1 1 Jlk

1 1

x1

nm

1

(a) n o r m a l _ E L — ^ ' ^ ' ^ ' ' - - ^ ^ ^

I

JKO.1

^0 "co

E LU

3) quasi-2D structure step flow mode

•

î-N boiid

AlCiaN surface

<e) 1D OD structure

(d) plan-view

•

# ^

GaN growth area Si"N bonding

A-.

^ ^ ^ J ^ ^Si-N masking lerrilory

(c) surface potential energy têtween Si N bonds

^îiif

A(GaN(0001) surface Figure 6.25. Schematic diagram of the model explaining the structural transitions observed by AFM. (a), (b), and (c) correspond to 0, 2.2 to 4.8, and 8.1 nmol cases, respectively, (d) A plan-view diagram of (c) to illustrate the GaN nucleation sites affected by Si-N bonding, (e) A possible surface potential energy diagram across the region along the line A - B in (d) [50].

GaN nucleation process, resulting in morphological transition. Such an area caused by the nano-scale modification of the surface may act as a nano-scale mask where GaN deposition is suppressed. It may also create a kinetic energy barrier for adatom diffusion. Tanaka et al. thought that GaN deposition would be suppressed at this place (see the masking territory in Figure 6.25(d) and (e)). Figure 6.25(e) illustrates a possible surface potential energy diagram in their model across the region along the line denoted "n" in Figure 6.25(d). The potential gradient (chemical potential) induced by Si-N bonds would lead the migration of adatom to the potential-minima regions, and the deposition of


119

1 1 1 1 1 I I • • I'P I I I I I I I I 1 I

T = 80K , !\

40/l20nm 7/21 nm 3.5/10nm

t

3.40 3,45 3.50 3.55 3.60 3.65 Photon energy (eV) Figure 6.26.

PL spectra of GaN QDs with three different sizes [51].

GaN would occur preferentially at those places. The large amount of holes observed in Figure 6.24(b) and (c) would correspond to the area of nano-scale masking. Therefore, as the masking area increased, the surface structure with both 2D and 3D characteristics would appear, and eventually the whole surface would be composed of 3D structures [50]. They also investigated optical properties of these GaN QDs. Figure 6.26 shows the 80 K PL spectra of the GaN QD samples of different sizes [51]. The different QD sizes are obtained by changing the deposition time of GaN while keeping the substrate temperature constant. The PL emission energy was observed to shift to higher energy with decreasing QD size. This effect was supposed to be a combination of a blue shift from the confinement-induced shift of the electronic levels and a red-shift from the piezoelectric field-induced quantum-confined Stark effect. An LD structure with these GaN QDs was prepared and optical pumping experiment was made [52]. GaN QDs, with an average size of about 10 nm width, 1 - 2 nm height, and a density of about 10^^ cm~^ were used in this experiment. Their LD structure is schematically shown in Figure 6.27. Stimulated emission was clearly observed at 21 K from the GaN QD LD structure under high excitation pump power density, as shown in Figure 6.28. At the pump power density of 0.78 MW/cm^, a sharp peak centered at 3.49 eV appeared. The FWHM was about 10 meV. The emission peak showed a red-shift of about 50 meV from the spontaneous emission peak observed at 77 K. Recently, the GaN QDs formed by this technique was used for the active layer of UV LED [53]. The room temperature UV emission at 360 nm was reported by current injection. This is the first report on the device application by current injection. The growth of InGaN QDs using TESi anti-surfactant was also reported by Hirayama et al. [54] In order to achieve a surface suitable for the growth of InGaN QDs, a "twolayered buffer" was used: 100 nm-thick Alo.12Gao.88N layer (topmost) and 300-nm-thick Alo.24Gao.76N layer deposited at HOOT on a 6H-SiC substrate. Prior to the InGaN

120

Optoelectronic Devices: Ill-Nitrides N| Met

>^k23^%77^ ciliickling Jpyer ^bjO«G%9^N barrieriiiyef OaN quantum doii ^^.i2*^'*o.8s'^ byrri^r layer AÔ^QSO.IWN cladding lnycr

Figure 6.27.

\ [ 3TOO) c^BWimd plane

The schematic diagram of a GaN QD laser structure with AlGaN barrier layers [52].

growth, a small amount of Si anti-surfactant modifying the surface properties was deposited at 1120°C, which is similar to that of the TESi pretreatment for GaN QD growth. Then, InGaN growth was performed at around 800°C. This resulted in a 3D nano-scale island growth. They found that the QD density was controllable in the range of cm by increasing the TESi dose. Similar trend was also reported in the growth of GaN QDs. A typical morphology of the InGaN QDs with an approximate density of 10^ ^ cm~^ obtained by this technique is shown in Figure 6.29. The average size and thickness of QDs are about 10 and 5 nm, respectively. Room temperature PL emissions from InGaN QDs grown with Si anti-surfactant were observed as shown in Figure 6.30 [54]. In this experiment, all InGaN QDs were

1

1

1

1

1

I

1

1

1

1

1 1 1 1 1

h 21K

1 1

1

1

1 J

MWA:m'

!\'' I •

f

•

[

• : 1

B

A

»

1

[

O.TgMWAim^

h r

3.35

Figure 6.28.

1

•

; r J,05MWifcm^

,..

' 1

1

A.^.—

.^4

1

1

1

1

n—"jr •

•

•

•

3.45 3,5 Energy (eV>

0.68MWA m^ '

1

?^^^^^

155

3.6

Emission spectra from GaN QDs with a cavity length of 1 mm under optical power densities of 0.68, 0.78, 1.05, 1.4MW/cm2 [52].


121

4. ^

if

I

1

-''•' lOOnni

Figure 6.29.

A typical A I ^ image of InGaN QDs grown on the AlGaN surface by using anti-surfactant [54].

grown with the same TESi dose. The PL peak energy position was controlled by changing the growth temperature, resulting in changes of In mole fraction of InGaN QDs. For comparison, InGaN QW structure with an estimated thickness of 5 nm was grown without TESi exposure on the AlGaN surface. It was found that the PL emission of the InGaN QD samples was several times stronger than that of the QW structure. This result strongly suggests that the increased luminescence efficiency is caused by the additional carrier confinement. From the calculation with PL peak positions in Figure 6.30, the In contents of InGaN QDs grown at 770, 800, 830, and 850°C were estimated as 52, 38, 25, and 22%, respectively. However, this calculation seemed to be based on the old value of InN bandgap energy of 1.9 eV; accordingly. In composition would be modified.

2.5

M Energy (eV)

Figure 6.30.

Room temperature PL spectra of InGaN QDs grown on AlGaN at various temperatures [54].

122


6.4. GROWTH OF QUANTUM DOTS BY SELECTIVE EPITAXY The aim of this method is to control the position, size and density of the QDs by high-resolution patterning techniques. With the help of the recent improvement of lithography techniques, it is expected that the position control and the uniformity of the QDs would be greatly enhanced by this technique; however, QD density obtainable by this technique is far lower than by either SK growth method or anti-surfactant growth method. Tachibana et al. reported the growth of InGaN QD structures by selective MOCVD on a Si02/GaN/sapphire substrate patterned by conventional photolithography [55]. At first, they grew a 2-|jLm-thick GaN layer on a (0001) sapphire substrate. Then, 40 nm of Si02 was deposited on the GaN layer by sputtering and the Si02 layer was patterned by conventional photolithography and a buffered HF solution. The pattern was grid-like, with 4 |xm period and square openings of 2 jxm side length. Selective growth of GaN was then performed and uniform hexagonal GaN pyramids were realized, as shown in Figure 6.31(b) and (c). Then, the growth of three periods of InGaN/GaN MQWs was

(a) IHQ ogGô 98^^ barriers and capping layer

-,,. InGaN QD

InGaisi QW

(b)

^ A A

^

wP

^^

WW

Ww

A A A A A A A A 2um

(c)

100 nm

Figure 6.31. (a) A schematic of InGaN QDs formed at the apex of hexagonal GaN pyramids, (b) bird's-eyeview SEM micrograph, (c) cross section SEM micrograph [55].


123

followed on the hexagonal GaN pyramids. The schematic diagram of this structure is illustrated in Figure 6.31(a). The lateral size of InGaN QDs formed at the apex of GaN pyramids was no more than 30 nm, estimated from the curvature of the apex of pyramid in Figure 6.31(c). PL spectra were measured at room temperature. With the emission peak at 3.4 eV from the GaN bulk layer, a very broad peak (the FWHM is 290 meV) was also observed at 2.88 eV (430 nm), as shown in Figure 6.32. To identify the regions giving the PL emission at 430 nm, micro-PL intensity images were recorded at room temperature. Figure 6.33(a) shows an image consisting mainly of reflected light. Hexagonal shapes can be seen very clearly. Figure 6.33(b) shows a micro-PL intensity image of PL of wavelength around 430 nm. The same area was observed in Figure 6.33(b) as in Figure 6.33(a). It is evident that the detected light was from the InGaN QD structure on the apex of pyramids, not from the GaN bulk. They also reported the successful growth of GaN/AlGaN QD structures on a Si02/GaN/sapphire substrate patterned using the similar technique with that of InGaN QDs [48,56]. QD density of these dots was less than 10^ cm~^ due to the resolution limitation imposed by photolithography. Wang et al. demonstrated the formation of higher density InGaN QDs by selective growth on Si-patterned GaN epilayer/GaN buffer layer/(0001) sapphire substrate with the improved patterning technique [57,58]. The Si mask was patterned by a partial removal of Si mask by 30keV Ga^ focused ion-beam (FIB) irradiation and subsequent photoassisted wet (PAW) etching of remaining Si mask with a solution of 3KOH:H202. If the window is patterned only by FIB, the damage of the GaN underlayer is inevitable.

He-Cd Laser 5 W/cm3 RT

3.0

GaN Bulk-

3.2

Energy (eV) Figure 6.32. PL spectrum of the selectively grown InGaN QDs at room temperature [55].


124

Figure 6.33.

(a) Optical microscope image of selectively grown InGaN QDs, (b) micro-PL image at a wavelength of 430 nm [55].

mainly due to the highly energetic Ga"^ atoms penetrating the residual Si layer during FIB process. The growth of GaN pyramids and InGaN/GaN MQWs was followed. The schematic diagram of this procedure is shown in Figure 6.34. The typical top-view scanning electron microscope (SEM) image of an InGaN QD sample is shown in Figure 6.35(a). There are 25 QD structures in each luim square area (2.5 X 10^ cm~^) and they looked uniformly arranged. An AFM image of the InGaN QD array in Figure 6.35(b), however, showed that the dimensions of each QD were not completely uniform due to the fluctuations in the window size by this patterning technique.

p

FIB milling

Siiiiask(50iiiii)

siliiiiiii

of patterns PAW Etching with OKOH-l-HiOj)

MOCVD Selective growth of InGaN QDs

GftM Figure 6.34.

p

Schematic diagram of FIB/PAW process used for the fabrication of InGaN/GaN QDs [58].

125


Figure 6.35.

(a) A typical top-view SEM image of selectively grown InGaN QDs. Diameter of the FIB windows was 80 nm; (b) 3D AFM image of the same QDs [58].

The CL spectrum measured at 80 K from the InGaN QD sample is shown in Figure 6.36. In this measurement, the electron beam was focused on an area covering 25 QDs. A clear blue shift of about 240 meV could be observed when compared with the emission from a reference InGaN QW sample. This value corresponds to an InGaN QD with a diameter of ~ 3 nm. The FWHM of the emission peak from QD was 48 meV, whereas that of the emission peak from QW was 84 meV. This result shows that the size fluctuation of QDs was suppressed to a certain extent by their technique.

6.5. NOVEL TECHNIQUES FOR QUANTUM DOT GROWTH Oliver et al. reported the InGaN QD growth on a GaN/sapphire substrate by MOCVD [20]. They had grown two InGaN epilayers under identical conditions, and then annealed

InOsN/GaN ODs T=SOK FWHM =48 nitV

3.29

3.31

3.33

3.37

Energy («V) Figure 6.36.

Cathodoluminescence spectrum measured at 80 K from selectively grown InGaN QDs. The inset shows a PL spectrum obtained for a reference InGaN MQW sample [57].

126

Figure 6.37.


AFM images (1.8 fxin X 1.8 |xm) of InGaN epilayers on GaN after (a) annealing under NH3 for 30 s, (b) annealing under N2 for 30 s [20].

the two samples at 700°C for 30 s after the growth under different conditions: under NH3 flow or under N2 flow. The anneal under N2 flow changed the surface morphology significantly as shown in Figure 6.37(b). Small nanostructures with a density of 5 X 10^^ cm~^, and an average height of 0.93 nm could be found along with many small pits in the wetting layer. They observed very sharp emission peaks with typical linewidth of —700 ixeV at 4.2 K by micro-PL experiments. Their time-resolved PL studies on the sample revealed that excitons have lifetimes around 2 ns at 4.2 K. They suggested that pits in Figure 6.37(b) were created by the decomposition of In-rich regions in InGaN epilayer (formed by spinodal decomposition), which appeared to be unstable in a nitrogen atmosphere. The nanostructures observed in Figure 6.37(b) might be very small In droplets, also formed due to decomposition, since they found that these structures could be removed by the HC1:3H20 etchant. These In droplets would then react with ammonia, before or during the growth of the capping layer, and this would result in the formation of InGaN QDs, possibly by some interdiffusion with the GaN capping layer [20]. Similar methods using the nitridation of metallic droplets were also reported by other research groups. Kawasaki et al. demonstrated the GaN QDs formed on an AlGaN/SiC substrate by MBE using Ga droplets [21]. At first, they deposited the Ga droplets on the substrate surface by supplying Ga at 300°C. Then, the sample was annealed under NH3 gas flow for 10 min in order to nitridate the Ga droplets. Figure 6.38(a) shows the SEM image of QDs obtained by this technique at a nitridation temperature of 600°C. The histogram of the QD diameter in Figure 6.38(b) shows that the GaN QDs formed were between 4 and 15 nm in diameter with an average diameter of approximately 9 nm. The QD density was greater than 3 X 10^^ cm~^. Hu et al. also reported the growth of QDs by MOCVD using the formation of Ga liquid droplet and nitridation process [22]. The fabrication of InGaN QDs by surface pretreatment using TMIn was reported by Zhang et al. [23]. TMIn pre-treatment with only TMIn and NH3 input flow was made just before the growth of InGaN well layer. The typical microstructure of InGaN QDs formed


:••'

12

14

111

16

Figure 6.38. (a) SEM image of GaN QDs formed by Ga droplet formation and subsequent nitridation at 600°C. (b) Histogram of the GaN QDs with mean diameter of 9 nm, and standard deviation of 2 nm [21].

(a)

6.00 nm

*^^Pi^tf'•' ''':^^^0l • Sf^'i?

Figure 6.39. Cross section HRTEM images of InGaN SQW (a) with and (b) without TMIn pretreatment. Arrows point to the low-barrier/well interfaces [23].

128

Optoelectronic Devices: Ill-Nitrides (b)

-I

1

1

1

r-

-»

1

f

1

1-

^6oo InGaN dots (Sample D)

UOO ^

200

2.0 600 Figure 6.40.

2.2

2.4

2.6 2.8 3.0 Energy (eV)

3.2

3.4

(a) The AFM surface morphology of InGaN QDs from by TMIn pretreatment, (b) comparison of PL spectra of InGaN QDs and homogeneous InGaN film [24].

in QW with TMIn pre-treatment was shown in Figure 6.39(a). Obvious contrasts, originating from In-rich regions in the low-In composition matrix, can be clearly seen along the lower-barrier/well interface. The average dimension of the contrast was as small as ~ 4 nm in width and 1.5 nm in height. The microstructure of the SQW without the TMIn treatment is shown in Figure 6.39(b) for comparison. Both the lower- and upperbarrier/well interfaces are quite sharp without obvious contrast, indicating that the growth mode was 2D throughout the growth of SQW structure without TMIn pre-treatment. The surface passivation method was also attempted to form InGaN QDs on GaN by Chen et al. [24,25]. High-temperature grown GaN surface was passivated by air exposure for 24 h and it was reloaded for the growth of a low temperature GaN layer at 550°C. It was found that the surface of the low temperature GaN layer had nano-scale roughness. They suggested that the surface passivation would increase the energy barrier for the hopping of the atoms, resulting in a decrease in the surface diffusion length. Subsequently, the growth of InGaN QDs was followed at 800°C. Figure 6.40(a) shows the image of InGaN QD structure grown on the passivated GaN surface. The average size of InGaN QDs was measured as 80 nm in diameter and 5 nm in height. The density of QDs was estimated as 5X 10'^ cm . PL spectrum of InGaN QDs was shown in Figure 6.40(b), compared with that of homogeneous InGaN thin films grown without surface passivation. 6.6. CONCLUSIONS So far, we have reviewed the various methods to form nitride QDs. They were classified into four methods: (1) using SK growth mode, (2) using "anti-surfactant", (3) using selective


129

epitaxy, and (4) other novel techniques. Growth of GaN/AlGaN, InN/GaN, InGaN/GaN QDs by MBE or MOCVD reported in the Hterature was reviewed. The first paper on the growth of nitride QDs appeared in 1996 and the number of publications and researchers are relatively small compared to III-V QDs. Nitride QD research is still in the developmental stage, however, considering the strong potential of nitride QDs in device applications, especially in LDs, UV-LEDs, etc., the interest from both academia and industry in nitride QDs increases very rapidly and the future of nitride QDs seems very promising.

REFERENCES [1] Bimberg, D., Grundmann, M. & Ledentsov, N.N. (1999) Quantum dot heterostructures, Wiley, New York. [2] Tatebayashi, J., Hatori, N., Kakuma, H., Ebe, H., Sudo, H., Kuramata, A., Nakata, Y., Sugawara, M. & Arakawa, Y. (2003) Electron. Lett., 39, 1130. [3] Phillips, J., Kamath, K., Zhou, X., Chervela, N. & Bhattacharya, P. (1997) Appl. Phys. Lett., 71, 2079. [4] Arakawa, Y. & Sasaki, H. (1982) Appl. Phys. Lett., 40, 939. [5] Xie, Q., Kobayashi, N.P., Ramachandran, T.R., Kalburge, A., Chen, P. & Madhukar, A. (1996) J. Vac. Sci. Technol, B, 14, 2203. [6] Lubyshev, D.I., Gonzalez-Borrer, P.P., Mareg, E., Jr., Petitprez, E., La Scala, N., Jr. & Basmaji, P. (1996) Appl. Phys. Lett., 68, 205. [7] Heinrichsdorff, F., Krost, A., Grundmann, M., Bimberg, D., Bertram, P., Christen, J., Kosogov, A. & Werner, P. (1997) /. Cryst. Growth, 170, 568. [8] Allen, C.Ni., Poole, P.J., Marshall, P., Eraser, J., Raymond, S. & Fafard, S. (2002) Appl. Phys. Lett., 80, 3629. [9] Finkman, E., Maimon, S., Immer, V., Bahir, G., Schacham, S.E., Gauthier-Lafaye, O., Herriot, S., Julien, F.H., Gendry, M. & Brault, J. (2000) Physica E, 1, 139. [10] Borgstrom, M., Bryllert, T., Sass, T., Bustafon, B., Wemersson, L.-E., Seifert, W. & Samuelson, L. (2001) Appl. Phys. Lett., 78, 3232. [11] Arakawa, Y. (2001) Phys. Stat. Sol. (a), 188, 37. [12] Chichibu, S., Azuhata, T., Sota, T. & Nakamura, S. (1996) Appl. Phys. Lett., 69, 4188. [13] Yoon, S., Moon, Y., Lee, T.-W., Yoon, E. & Kim, Y.D. (1999) Appl. Phys. Lett., 74, 2029. [14] Sopanen, M., Lipsanen, H. & Ahopelto, J. (1995) Appl. Phys. Lett., 65, 1662. [15] Marchand, H., Desjardins, P., Guillon, S., Paultre, J.-E., Bougrioua, Z., Yip, Y.-F. & Masut, R.A. (1997) Appl. Phys. Lett., 71, 527. [16] Chidley, E.T.R., Haywood, S.K., Mallard, R.E., Mason, N.J., Nicholas, R.J., Walker, P.J. & Warburton, R.J. (1989) Appl. Phys. Lett., 54, 1241. [17] Ferrer, J.C, Peiro, F., Comet, A., Morante, J.R., Uztmeier, T., Armelles, G. & Briones, F. (1996) Appl. Phys. Lett., 69, 3887. [18] Eaglesham, D.J. & Cerullo, M. (1990) Phys. Rev. Lett., 64, 1943. [19] Lowisch, M., Rabe, M., Stegemann, B., Henneberger, F., Grundmann, M., Turck, V. & Bimberg, D. (1996) Phys. Rev. B, 54, R11074. [20] Oliver, R.A., Briggs, G.A.D., Kappers, M.J., Humphreys, C.J., Yasin, S., Rice, J.H., Smith, J.D. & Taylor, R.A. (2003) Appl. Phys. Lett., 83, 755.

130

Optoelectronic

Devices:

Ill-Nitrides

[21] Kawasaki, K., Yamazaki, D., Kinoshita, A., Hirayama, H., Tsutsui, K. & Aoyagi, Y. (2001) AppL Phys. Lett., 79, 2243. [22] Hu, C.-W., Bell, A., Fonce, F.A., Smith, D.A. & Tsong, I.S.T. (2002) Appl. Phys. Lett., 81, 3236. [23] Zhang, J., Hao, M., Li, P. & Chua, S.J. (2002) Appl. Phys. Lett., 80, 485. [24] Chen, Z., Lu, D., Yuan, H., Han, P., Liu, X., Li, Y., Wang, X., Lu, Y. & Wang, Z. (2002) J. Cryst. Growth, 235, 188. [25] Huang, J.S., Chen, Z., Luo, X.D., Xu, Z.Y. & Ge, W.K. (2004) /. Cryst. Growth, 260, 13. [26] Goodwin, T.J., Leppert, V.J. & Risbud, S.H. (1997) Appl. Phys. Lett., 70, 3122. [27] Borsella, E., Garcia, M.A., Mattei, G., Maurizio, C. & Mazzoldi, P. (2001) J. Appl. Phys., 90, 4467. [28] Mii, O.I., Ahrenkiel, S.P., Bertram, D. & Nozik, A.J. (1999) Appl. Phys. Lett., 75, 478. [29] Daudin, B., Widmann, F., Feuillet, G., Samson, Y., Arlery, M. & Rouviere, J.L. (1997) Phys. Rev. B, 56, R7069. [30] Widmann, F., Daudin, B., Feuillet, G., Samson, Y., Rouviere, J.L. & Pelekanos, N. (1998) J. Appl. Phys., 83, 7618. [31] Widmann, F., Simon, J., Daudin, B., Feulliet, G., Rouviere, J.L., Pelekanos, N.T. & Fishman, G. (1998) Phys. Rev. B, 58, R15989. [32] Damilano, B., Grandjean, N., Semond, F., Massies, J. & Leroux, M. (1999) Appl. Phys. Lett., 75, 962. [33] Grandjean, N. & Massies, J. (1998) Appl Phys. Lett., 72, 1078. [34] Damilano, B., Grandjean, N., Dalmasso, S. & Massies, J. (1999) Appl. Phys. Lett., IS, 3751. [35] Adelmann, C , Simon, J., Feuillet, G., Pelekanos, N.T., Daudin, B. & Fishman, G. (2000) A/?/?/. Phys. Lett., 76, 1570. [36] Damilano, B., Vezian, S., Grandjean, N. & Massies, J. (1999) Jpn. J. Appl. Phys., 38, L1357. [37] Gogneau, N., Jalabert, D., Monroy, E., Shibata, T., Tanaka, M. & Daudin, B. (2003) /. Appl. Phys., 94, 2254. [38] Miyamura, M., Tachibana, K. & Arakawa, Y. (2002) Appl. Phys. Lett., 80, 3937. [39] Tachibana, K., Someya, T. & Arakawa, Y. (1999) Appl Phys. Lett., 74, 383. [40] Tachibana, K., Someya, T. & Arakawa, Y. (1999) Phys. Stat. Sol (a), 176, 629. [41] Tachibana, K., Someya, T., Arakawa, Y., Werner, R. & Forchel, A. (1999) Appl Phys. Lett., 75, 2605. [42] Kim, H.J., Na, H., Kwon, S.-Y., Kim, Y.-W. & Yoon, E. (2003) The Fifth International Conference on Nitride Semiconductors (ICNS-5), May 25-30, Nara, Japan. [43] Kim, H.J., Na, H., Kwon, S.-Y., Seo, H.-C, Kim, H.J., Shin, Y., Lee, K.-H., Kim, D.H., Oh, H.J., Yoon, S., Sone, C , Park, Y. & Yoon, E. (2004) J. Cryst. Growth, 269, 95. [44] Kim, H.J., Na, H., Kwon, S.-Y., Seo, H.-C, Kim, H.J., Shin, Y., Lee, G.-H., Kim, Y.-W., Yoon, S., Oh, H.J., Sone, C , Park, Y., Cho, Y.-H., Sun, Y. & Yoon, E. (2003) Phys. Stat. Sol (c), 0, 2834. [45] Briot, O., Maleyre, B. & Ruffenach, S. (2003) Appl Phys. Lett., 83, 2919. [46] Tersoff, J., Teichert, C. & Lagally, M.G. (1996) Phys. Rev. Lett., 76, 1675. [47] Louviere, J.L., Simon, J., Pelekanos, N., Daudin, B. & Feuillet, G. (1999) Appl Phys. Lett., 15, 2632. [48] Tachibana, K., Someya, T., Ishida, S. & Arakawa, Y. (2001) Phys. Stat. Sol (b), 228, 187. [49] Tanaka, S., Iwai, S. & Aoyagi, Y. (1996) Appl Phys. Lett., 69, 4096. [50] Tanaka, S., Suemune, L, Ramvall, P. & Aoyagi, Y. (1999) Phys. Stat. Sol (b), 216, 431. [51] Ramvall, P., Tanaka, S., Nomura, S., Riblet, P. & Aoyagi, Y. (1999) Appl Phys. Lett., 73,1104.


131

[52] Tanaka, S., Hirayama, H., Aoyagi, Y., Narukawa, Y., Kawakami, Y., Fujita, S. & Fujita, S. (1997) AppL Phys. Lett., 71, 1299. [53] Tanaka, S., Lee, J.S., Ramvall, P. & Okagawa, H. (2003) Jpn. J. Appl Phys., 42, L885. [54] Hirayama, H., Tanaka, S., Ramvall, P. & Aoyagi, Y. (1998) Appl. Phys. Lett., 72, 1736. [55] Tachibana, K., Someya, T., Ishida, S. & Arakawa, Y. (2000) Appl. Phys. Lett., 76, 3212. [56] Tachibana, K., Someya, T., Ishida, S. & Arakawa, Y. (2002) /. Cryst. Growth, 237-239, 1312-1315. [57] Wang, J., Nozaki, M., Lachab, M., Ishikawa, Y., Fareed, Q., Wang, T., Hao, M. & Sakai, S. (1999) Appl. Phys. Lett., 75, 950. [58] Lachab, M., Nozaki, M., Wang, J., Ishikawa, Y., Fareed, Q., Wang, T., Nishikawa, T., Nishino, K. & Sakai, S. (2000) /. Appl. Phys., 87, 1374.


Chapter 7

AIN Epitaxial Layers for UV Photonics H.X. Jiang and J.Y. Lin Department of Physics, Kansas State University, Manhattan, KS 66506-2601, USA

7.1.

INTRODUCTION

Ill-nitride wide bandgap semiconductors have been widely recognized as technologically important materials. Photonic devices based on Ill-nitrides offer many benefits including UV/blue/green emission (allowing chem-bio-agents detection and higher optical storage density), large band offsets of InN/GaN/AlN heterostructures (allowing novel quantum well device design), and inherently high emission efficiencies. These unique features may allow the creation of optoelectronic and photonic devices with unprecedented properties and functions. The research activities on Al;,Gai_;,N (3.4 < E^ < 6.2 eV) with high AIN mole fractions and devices operating in the ultraviolet (UV) spectral regions are still in their embryonic state. Achieving device quality Al-rich AlGaN with high conductivities and high quantum efficiencies remains as one of the foremost challenges for the nitride community. AIN and Al-rich AlGaN alloys, covering wavelengths from 300 to 200 nm, are ideal materials for the development of chip-scale UV light sources/sensors, because AlGaN is the only ultra-wide-bandgap semiconductor system in which the bandgap can be easily engineered through the use of alloying and heterostructure design. Efficient solid-state UV light sources/sensors are crucial in many fields of research and development. For instance, protein fluorescence is generally excited by UV light; monitoring changes of intrinsic fluorescence in a protein can provide important information on its structural changes [1]. Thus, the availability of chip-scale UV light sources is expected to open up new opportunities for medical research and health care. Solid-state UV light sources also have applications in water purification, equipment/personnel decontamination, and white light generation [2]. There is an urgent need to develop new approaches to further improve material quality with reduced dislocation density and unintentional impurities and improved surface morphologies in Al-rich AlGaN alloys, which would enhance the doping efficiency and device performance. AIN is an end point of the AlGaN alloy system. A full understanding of the AlGaN alloy system (particularly Al-rich AlGaN alloys) could not be achieved before the binary AIN

E-mail address: [email protected] (H.X. Jiang).

133

134


material is well understood. Moreover, AIN is unique due to the fact that no other semiconductors possess such a large direct bandgap as well as the ability of bandgap engineering through the use of heterostructures. In spite of the recognition of the importance of AIN, many of its fundamental optical emission properties are not well known in the past due to the lack of high-quality materials as well as technical difficulties involved for the deep UV (down to 200 nm) optical measurements. Rapid progress has been made recently in the epitaxial growth and understanding of the basic physical properties, as well as the device applications of AIN epilayers. This chapter aims at providing a brief summary on these recent advances with emphasis on the fundamental optical properties, basic impurity parameters, and conductivity control of AIN and AlGaN epitaxial films with high Al contents. In Section 7.2, we discuss epitaxial growth and characterization techniques to identify high quality films; such as atomic force microscopy (AFM) for studying the surface morphology and SIMS for probing unintentional impurities such as oxygen. In Section 7.3, we present recent advances in understanding the fundamental optical properties of AIN. The detailed band structure near the F-point of wurtzite (WZ) AIN is presented. The unique band structure of AIN, i.e. the negative crystal-field splitting, affects profoundly the optical properties of AlGaN alloys, in particular of Al-rich AlGaN alloys. One of the immediate consequences is that the dominant band-edge emission in GaN (AIN) is with polarization of E L c {E\\c). Accordingly, the emission intensity in Al-rich Al^cGai-^^N alloys decrease with increasing x for epilayers grown on the c-plane sapphire. The recombination dynamics of the bound exciton (I2) and free exciton (FX) transitions in AIN epilayers were probed by deep UV time-resolved photoluminescence (PL). The PL decay lifetimes were found to be around 80 ps for the bound exciton and 50 ps for the free exciton at 10 K in AIN epilayers, which are slightly shorter than those in GaN. This is a direct consequence of the large energy bandgap of AIN. The extrapolated radiative decay lifetimes in AIN epilayers increases with temperature according to T^'^ between 100 and 200 K and are affected by the free-exciton dissociation at temperatures above 200 K, following the same trend as in GaN. From the low-temperature (10 K) emission spectra, the temperature dependence of the recombination lifetime, and the PL emission intensity activation energy, the binding energies of the donor bound excitons and free excitons in AIN were deduced to be around 16 and 80 meV, respectively. The observed large freeexciton binding energy implies that excitons in AIN are extremely robust entities that would survive well above the room temperature. Impurity transitions involving nitrogen vacancies generated during ion implantation and Al vacancies and/or complexes in as-grown layers have been studied. The results indicate that VAI and/or V A I - O N complexes are deep acceptors with an energy level of 2.59 eV above the valence band of AIN, which is directly correlated with the reduced conductivities in Al-rich AlGaN and AIN and is thus detrimental to optoelectronic devices using AIN and AlGaN epilayers. The experimentally determined nitrogen vacancy energy level is around

AIN Epitaxial Layers for UV Photonics

135

260 meV. As a consequence of the large activation energy (0.26 eV) as well as high formation energy, VN in AIN cannot contribute significantly to the n-type conductivity. With recent advances in epitaxial growth, conductive n-type Al-rich Al^^Gai-^cN alloys with high Al contents (x > 0.7) have been obtained. Section 7.4 summarizes recent advances in the conductivity control of AlGaN alloys of high Al contents and AIN and understanding of impurity parameters in these materials. A room temperature n-type resistivity as low as 0.0075 H cm with an electron concentration of 3.3 X 10^^ cm~^ and mobility of 25 crn^fV s has been obtained for Alo.7Gao.3N. The resistivity was observed to increase by almost one order of magnitude as the Al content was increased by about 8%, due to the deepening of the Si donor level with increasing x. Transport measurements have indicated that n-type conduction in pure AIN can be achieved. It was found that heavy doping is needed to bring down the donor activation energy and achieve higher conductivities in Al-rich AlGaN alloys. Mg acceptor ionization energy in Al^^Grai-jcN alloys as a function of x has also been measured, from which a binding energy of 0.51 eV for Mg acceptor in AIN was determined. Although Mg acceptors are an effective mass state in this ultra-large-bandgap semiconductor, as a consequence of this large acceptor binding energy of 0.51 eV, only a very small fraction (about 10~^) of Mg dopants can be activated at room temperature in Mg-doped AIN, implying that it is extremely difficult to achieve p-type AIN by Mg doping. Section 7.5 discusses applications of AIN epitaxial layers. These include the insertion of AIN epilayers in UV and deep UV emitters, serving as active layers or dislocation filters. Since a high quality AIN epilayer is UV transparent all the way down to 200 nm and can be grown with superior surface morphology over an AlGaN alloy, it is an ideal template for the subsequent UV photonic device structure growth. Applications of AIN epilayers for other types of active device applications such as for surface acoustic wave (SAW) and electron emission devices are also discussed. In Section 7.6, we present concluding remarks with focuses on future prospects and remaining challenges.

7.2. EPITAXIAL GROWTH AND CHARACTERIZATION

AIN epilayers have been grown by MOCVD, MBE, RF reactive sputtering, and pulsed laser ablation on different substrates (sapphire, SiC, Si, diamond, and AIN bulk crystals) [3-42]. In general, it is much more difficult to obtain high-quality AIN epilayers than GaN epilayers. This is because the growth of high-quality AIN requires much higher temperature and lower V/III ratio. Moreover, achieving conductive Al;cGai_;,N alloys with high Al contents is very challenging due to several well-known mechanisms: (i) an increase in the ionization energy of the dopants and (ii) an enhanced compensation of native defects (cation vacancy and cation vacancy-oxygen complex) with an increase in

136


the alloy composition [43,44]. For MOCVD growth of AIN and Al-rich AlGaN alloys, the metal organic sources used are typically trimethylgallium (TMGa) for Ga and trimethylaluminum (TMAl) for Al. Mg and Si are the common choice as acceptor and donor impurities, respectively. For Mg-doping of AlGaN and AIN, bis-cyclopentadienylmagnesium (Cp2Mg) can be transported into the growth chamber with ammonia during growth. The gas sources are blue ammonia (NH3) for nitrogen and Saline (SiH4) for Si doping and the doping levels can be controlled by the flow rates. There are several methods to confirm the targeted Al contents in Al;fGai_;^-N alloys, including energy dispersive X-ray (EDX) microanalysis. X-ray diffraction (XRD) measurement, secondary ion mass spectroscopy (SIMS) measurement, and optical measurements such as the PL spectral peak position of the band-edge transition. For high Al content Al^cGrai-^cN alloys, the targeted Si- and Mg-dopant concentrations can be verified by SIMS measurements (by Charles Evans & Associates). Determining the quality of the Ill-nitride materials is key to improving the manufacturing process. Unlike many other Ill-nitride compounds, however, measuring the optical and electrical properties of AIN has been a challenge because the material is a good insulator, which renders some standard characterization methods, such as Hall measurement, useless. On the other hand, XRD measurements can provide information about crystalline quality; it provides, however, very little information about electrical and optical qualities of AIN epilayers. As shown in Figure 7.1, the full width at half maximum (FWHM) of the XRD rocking curve of the (0002) reflection peak of AIN epilayers grown on sapphire by the authors' group can be as narrow as 50 arcsec, which is smaller than that 80000 -

KSU-A897 70000 - AIN/sapphire

(002) Rocking Curve

60000 50000 B

40000 -

c

5

FWHM = 50"

30000 20000 -

1 1 J v_

100000-1000018.0

1

1

18.2

1

1

1

18.4

1

18.6

1

1

18.8

1

19.0

e (deg) Figure 7.1. XRD rocking curve of the (0002) reflection peak of AIN epilayers grown on sapphire by the authors' group at Kansas State University and the XRD measurement was performed by Air Force Research Laboratory, Sensors Directorate, Hanscom AFB (M.L. Nakarmi, Ph.D. Thesis, in preparation).


137

of the same reflection peak in GaN epilayers grown on sapphire and is also comparable or even smaller than the best value reported for AIN grown on 6H-SiC (68 arcsec) [12]. However, the value is two times larger than that reported for thin AIN single crystalline platelets (25 arcsec) [36] and five times larger than that of AIN bulk crystals [27]. It is established now that the out-of-plane structural features in Ill-nitrides, e.g. the symmetric (0002) XRD peak in AIN, cannot be used to correlate the material quality and the optical and electrical properties [13,14]. This was attributed to the fact that in Ill-nitrides the edge dislocations only distort the asymmetric planes and induce no significant distortion on the symmetric planes. Thus, there is only a strong correspondence between the in-plane structural order and the electrical and optical properties of Ill-nitrides. For optoelectronic device applications based on AIN and high Al content AlGaN alloys, it is essential to characterize their optical and optoelectronic properties directly. It has been clearly demonstrated in the past that the optical characterization techniques, especially time-resolved optical studies, which provide the temporal characteristics of carrier recombination, together with spectra information are indeed a powerful method for determining the optical processes or transition mechanisms involved because different optical transitions have different dynamical behaviors. The dynamics of injected carriers involved in optical processes is determined by the sample crystalline quality, purity, alloy composition, and quantum well interface properties in different materials and device structures. More importantly, the dynamics of various optical transitions can provide important information regarding excitation and energy transformation processes and recombination lifetimes of injected carriers, which are strongly correlated with quantities such as the quantum efficiency and optical gain. Recently, a deep UV laser spectroscopy system has been developed in author's laboratory (Figure 7.2) which allows one to make picosecond time-resolved PL measurements of AIN and thus, to effectively characterize, and thereby improve the material's quality. The deep UV picosecond time-resolved laser spectroscopy system consists of a frequency quadrupled 100 fs Ti:sapphire laser with an excitation photon energy set around 6.28 eV (with a 76 MHz repetition rate and a 3 mW average power), a monochromator (1.3 m), and a streak camera with a detection capability ranging from 185-800 nm and a time resolution of 2ps. Because only high-quality semiconductor materials emit predominantly exciton PL, the identification of excitonic spectra reveals the optical quality of the sample. The effectiveness of PL spectroscopy measurement is illustrated in Figure 7.3(a), where the room temperature PL spectra for two selective AIN epilayers (KSU A-767 and KSU A-1080) grown under different growth conditions are shown, which contain different oxygen impurity concentrations. The oxygen and carbon impurity profiles as measured by SIMS are shown in Figure 7.3(b) for sample KSU A-1080. Comparing the PL spectra of the two samples, it can be seen that the optical quality or


138 MCP-PMT Single Photon Counting ~30ps

Monochromator (186-800 nm)

Monochromator (800-1700 nm)

H >^ ^ ^

MCP-PMT Single Photon Counting ~30 ps

InGaAs Detector

195 nm (10 mW) 200 fs

Figure 7.2. The femto-second deep UV time-resolved photoluminescence measurement system at Kansas State University. The system is integrated with a near-field scanning optical microscopy and is specifically designed for high Al content AlGaN and AlInGaN alloys, covering the band-band emission in pure AIN. Capabilities include: 2 ps time resolution, 50 nm spatial resolution (SNOM/AFM), 190 nm < A < 1700 nm.

(b)10^^

(a) 8 6-1

4J

AIN (KSU A-767) ©2=1x1020 cm-3

A

300 K

5.98 eV

1

2

£ 8

Jw AIN (KSU A-1080) O2=2x10i7cm-3

I 5.98 eV:

6 4 2

(x10)

0 4 E(eV)

5

^10^ E E o 6 ô^^ o O

J\ 10^' 0.0

0.5 1.0 Depth (|um)

1.5

Figure 7.3. (a) Room temperature PL spectra measured for selective AIN epilayers grown under different conditions. The oxygen concentrations are also indicated, (b) Oxygen and carbon impurity profiles in sample KSU-A1080, as probed by SIMS (performed by Charles Evans & Associate) (M.L. Nakarmi, Ph.D Thesis, in preparation).


139

the intensity ratio of the band-edge to the deep-level impurity transitions depends strongly on the growth conditions. Furthermore, Figure 7.3(a) shows that the PL emission intensity ratio of the band-edge transition line near 5.97(±0.01) eV to the impurity transition line near 3.40 eV is directly correlated with the oxygen impurity concentration. In the better-optimized AIN epilayer (KSU A-1080) with a lower oxygen impurity concentration, the emission intensity of the impurity transition is two orders of magnitude lower than that of the band-edge transition at room temperature, indicating a significant improvement in material quality. Figure 7.4 compares the PL spectra of AIN and GaN epilayers covering a broad spectral range from 2.2 to 6.2 eV for AIN and 1.8 to 3.6 eV for GaN [34]. It is interesting to note that although the 10 K band-edge emission intensity is about one order of magnitude lower in AIN than in GaN, the room temperature emission intensities are comparable for both compounds. This impHes that it is possible to obtain AIN that displays less thermal quenching and fewer problems resulting from impurities, dislocations and nonradiative recombination channels than GaN, particularly at elevated temperatures. This points to the great potential of AIN for many device applications, because it is already well known that the detrimental effect of dislocations/impurities in GaN is much less severe than in other III-V and II-VI semiconductors. For optimized epilayers grown by MOCVD, the surface morphology of AIN is comparable to those of GaN. This is illustrated in Figure 7.5, where scanning electron (b) 0.4

(a) 12T=10K AIN, KSUA-742

^
800°C usually lead to stronger apparent bowing (Z?>+1.3eV); while growths initiated using low-temperature buffers on sapphire, followed by high-temperature growth, lead to weaker bowing (^ < +1.3 eV) [60]. The degree of polarization (P) is defined by P = (Ij_ - I\\)/(Ij_ + /||), where /^ and /|| are the integrated PL intensities for the polarization components of £^ -L c and E\\c, respectively. Figure 7.11(a) plots P as a function of x. P decreases almost linearly with increasing jc, and P = 0 3.tx = 0.25. The representative band structures near the F-point of Al;,Gai_^N alloys are depicted in Figure 7.11(b) for (a) x = 0, (b) x = 0.25, and (c) x= 1. The conduction bands have Fy symmetry in both AIN and GaN. Compared with the band structure of GaN, the most significant difference in AIN is the negative crystalfield splitting AQF (-219 meV) compared with a positive value (+38 meV) in GaN, as

(a) 1.0

0.0

0.2

0.4

0.6

0.8

1.0

Al content (x)

Figure 7.11. (a) The degree of polarization P vs. x in Al^^Gai-^cN alloys, (b) The band structure of wurtzite Al;,Gai_;,N near the F-point for jc = 0, 0.25, and jc = 1 (after Ref. [58]).

148


discussed in Section 7.3.1. Because of this large negative ZICF in AIN, the order of the valence bands in AIN is different from that in GaN. The top valence band has T^ (Fy) symmetry in GaN (AIN) because of the positive (negative) ZicF- Therefore, light emission due to the recombination between the conduction band electrons and the holes in the top valence band is polarized with £'||c in AIN, which is opposite to that in GaN (£• _L c). This unique band structure of AIN affects acutely the optical properties of AlGaN alloys, in particular of Al-rich AlGaN alloys. When Al content is increased from X = 0 to 0.25, the valence band with T^ symmetry evolves as the lowest valence band (C-band) in GaN to the topmost valence band (A-band) in Al;,Gai_;,N alloys {x > 0.25). At X = 0.25, three valence bands become degenerated at the F-point and the degree of polarization P becomes zero. Thus, the experimentally observed x dependence of the polarization degree shown in Figure 7.11(a) is a direct consequence of the band structure evolution with x. Figure 7.12(a) shows the FWHM of PL emission spectra of Al;,-Gai_;^;N alloys with different x measured at 10 K. The variation of the FWHM with x follows the general trend of the previous theoretical prediction [62-64], i.e. it increases as a function of x, but decreases as x further increases from jc = 0.7 to 1. Figure 7.12(b) shows the variation of the integrated PL emission intensity of Al;,Gai_;,N alloys with the Al content, Ipi vs. x, measured at 10 K for both polarization orientations of £" .L c and E\\c. The emission intensity of the £ ± c component decreases with increasing x, while /pL of the E\\c component is almost independent of x except for GaN. It is well documented that, the overall PL emission intensity decreases with increasing x in Al^Gai_;^N alloys [51,63]. (a)

100^

T=10K AI,Gai.,N:

80

0)

E, 60 J •

1 40S 20 0-

(b)

•

•

•

• •

"•

1 I

'

•

1

11

8

.-^ d "1 E

•

6 4.

X

• E_Lc • X

0)

X •

X

X

•

•

X E//C X

; J

2 •

c

n. U T 0.0

1

0.2

04^

0.6

0.8

1

Al content (x) Figure 7.12. (a) The full width at half maxima (FWHM) of PL vs. x in A^Gai-^^N alloys at 10 K. (b) PL emission intensities for E L c and E\\c vs. x in AljcGai__^N alloys measured at 10 K (after Ref. [58]).


149

The efficiency of UV light emitting diodes (LEDs) using Al^^Gai-jcN alloys as active layers is also known to decrease with increasing emission wavelength. The results shown in Figure 7.12(b) suggest that the unique optical property of AlGaN alloys is partly responsible for the low emission efficiency in AlGaN alloys and related UV LEDs, i.e. the emission intensity of the E\\c component decreases with increasing x. The fact that the emission intensity of the E\\c component is almost independent of x seems to preclude the dislocations and nonradiative centers being the dominant cause for the reduced emission efficiency in Al;^Gai __^N alloys with increasing x. In terms of implications of this unique polarization property on device applications, the most significant effects include (i) the power output of LEDs decreases and (ii) the light guiding for edge-emitting laser diodes (LDs) with a predominantly TM mode enhances with an increase of the Al content if AlGaN alloys are being used as active layers, due to the negative crystal-field induced unique polarization property of AlGaN alloys. It is well known in conventional LEDs that the extraction efficiency is only about 5% from each side due to internal reflection. The light can only escape from the top and bottom surfaces when it is within a cone of about 6^ ~ 20°, where 9^ is the critical angle of total internal reflection. For all types of existing LEDs from near infrared to blue color, emitted photons in this cone are able to escape since the polarization of emitted light is mainly perpendicular to the crystal axis within this cone (E 1 c) [65]. However, for UV LEDs using Al^Gai-;,N as active layers (x > 0.25), the most dominant emission will be photons with polarization parallel to the c-axis iE\\c), which implies that UV photons can no longer be extracted easily from the escaping cone. Figure 7.13(a) illustrates schematically this situation for UV LEDs using Al;cGai_;cN alloys as active layers, which shows that only photons polarized perpendicular to the c-axis (E 1 c) can be extracted from the escaping cone (^c ~ 20°). However, the E ± c polarization component is prohibited in the escape cone. We thus emphasize that finding methods for enhancing the light extraction is more critical in UV LEDs with Al;cCrai_;cN active layers than in blue/green LEDs with InGaN/GaN active layers. Techniques for extracting light of transverse propagation such as |x-LEDs [66], photonic crystals [67,68] and other methods are not only recommended, but also necessary for future high-power short wavelength nitride UV emitters. For edge-emitting LDs based on AlGaN alloys, since light cannot leak out from the top and bottom layers due to their unique polarization property, the guiding effect of light is thus enhanced. Figure 7.13(b) shows the schematic diagram of LDs with Al^^Gai-j^N alloys as active layers. The transverse electric (TE) mode is usually the dominant laser emission in all other semiconductor LDs, where the electric field of the mode is parallel to the layer interfaces. However, for LDs with Al_^Gai_;cN as active layers (x > 0.25), the transverse magnetic (TM) mode should be the dominant laser emission, in which the magnetic field is parallel to the layer interfaces.

150

Optoelectronic Devices: Ill-Nitrides (a)

AixGa^,xN(x>oj2g^ LED with AIGaN as active layer

(b)

I / / /

/

AlyGai.yN(y>x) AlxGai.xN(x;>0^ AlyGai.yN(y>x) LD with AIGaN as active layer

Figure 7.13. (a) Schematic diagram of UV LEDs with Al^Gai -j,N alloys as active layers. The light escape cone is about dc ^ 20°, within which any photons extracted are nearly polarized perpendicular to the c-axis. (b) Schematic diagram of LDs with Al;,Gai _^N alloys as active layers. In contrast with other semiconductor LDs with TE being the dominant mode, the TM mode is expected to be the dominant mode in AIGaN UV LDs (after Ref. [58]).

7.3.3 Exciton Recombination Dynamics in AIN The recombination dynamics associated with the fundamental optical transitions in AIN are unknown due to the lack of high quality materials as well as technical difficulties involved for the deep UV (down to 200 nm) time-resolved PL measurements in the past. The availability of AIN epilayers with high optical quality as well as the deep UV picosecond time-resolved PL system opens the possibility to probe the recombination dynamics associated with fundamental optical transitions in AIN. Recently, deep UV picosecond time-resolved PL spectroscopy has been employed to study the recombination dynamics of the donor-bound exciton (I2) and free exciton (FX) transitions [53]. Figure 7.14(a) shows the low-temperature (10 K) PL spectrum for an AIN epilayer, in which the dominant emission line at 6.015 eV is due to the neutral donor-bound-exciton recombination (D^X) or I2. A second emission line at the higher energy side around 6.031 eV is also clearly resolved, which is attributed to free-exciton transition (FX). The I2 and free exciton emission peak positions shown in Figure 7.14(a) are 2 meV different from the PL spectra shown in Figure 7.6(a). This is quite common for Ill-nitride epilayers as

AIN Epitaxial Layers for UV Photonics (a)

(b)

1.5

AIN epilayer KSU1767

151 | i/"\

0.015-co

-

> CD

-

'(/)

E

0

a:

•

0.010-

•

0.005 0 000 - ' 1 1 .0 1.5

•

'1

•

-

1 — 1 — 1 — 1' — 1 — ' — 1 — ' — 1 — ' —

2.0 2.5

3.0 3.5

Nsi(10^9cm-3)

4.0 4.

2 3 Nsi(10^9cm-3)

Figure 7.22. (a) Room temperature resistivity as a function of the Si dopant concentration, A^si • (t>) The Si donor activation energy in Alo.7Gao.3N as a function of dopant concentration, A^si- Filled circles are measured values and solid line is the least-squares fit with the bandgap renormalization effect of Eq. (7.7). The inset is an Arrehenius plot of resistivity of n-Alo.7Gao.3N with A^si = 3.5 X 10^^ cm~^. Filled circles are the measurement result; solid line is the least squares fit of data with Eq. (7.1). The fitted activation energy is 23 meV (after Ref. [115]).

suggested for GaN [121,122]. The resistivity results shown in Figure 7.22(b) suggest that heavy doping is needed to achieve highly conductive Al;cGai_;cN alloys with high x for device applications. Another interesting parameter one could obtain from Eq. (7.7) is the maximum Si-doping level required to bring EQ down to zero (metallic behavior). Using the fitting results, this doping level can be extrapolated to be about 2 X 10^^ cm~^. However, the resistivity data shown in Figure 7.22 indicate that heavy doping could induce degradation of material quality and impurity compensation. Thus, an optimal doping scheme has to be employed in obtaining highly conductive Al;cGai-;cN alloys with high x. The attainment of highly conductive Al^^Gai-^^N alloys with high Al contents made the measurement of the Si donor parameters in Al^^Gai-^^-N alloys as a function of the Al content in high x end possible [116,117]. Figure 7.23(a) presents the room temperature Hall measurement results of n-Al^cGai-^^N (x > 0.7), showing the Al content (x) dependent resistivity, electron concentration, and electron mobility, respectively. This set of samples were grown on sapphire substrates by MOCVD with a thickness of about 1 iJim. A 0.5 (xm AIN epilayer was first deposited on (0001) sapphire substrate with a low temperature buffer, followed by the growth of Si-doped Al;,Gai_;,N epilayer [117]. The Si dopant concentration (Nsi) was 3.5 X 10^^ cm~^ in all samples as guided by the result shown in Figure 7.22(a). The dopant concentration, A^si^ for selective samples was


164

^) 1

(a) E o

UU- -

1

I

I

1

'

1

•

1

'

1

•

1

J

75-

a

•

> b^ 5 0 O UJ

• 25-

•

^ o

0.03-

^

0.02

Q.

r

0.01- V

(3 4

0- — ' — 1 — ^

85

90

Al Content (%)

95

100

•

n-AI^Gai

A

-

x= 0.77

8 12 16 1000/T(K-^)

-

T—'—1—'—1—'—1—'—1—•—

65 70 75 80 85 90 95 100 Al Content (%)

Figure 7.23. (a) Room temperature Hall measurement results of n-Al;,Gai -J>i (x > 0.7). The top, middle and bottom sections of (a) show the resistivity, electron concentration, and electron mobihty as functions of the Al content (x), respectively; lines are guide to the eye. For all samples, the silicon dopant concentration was 3.5 X 10^^ cm~^. (b) Si donor activation energy as a function of the Al content, x. The inset shows the fitting result for Alo.77Gao.23N, and an activation energy of 41 meV was obtained (after Ref. [117]).

determined by SIMS measurements. The resistivity increases with the Al content (x) rapidly and the dependence can be described by the following empirical equation: p(Al^Gai_^N) = p(AlN) X IQ-^^-^^^^-^^

(7.8)

from which one can deduce that the resistivity of n-Al;,.Gai_;^N (x > 0.7) increases by about one order of magnitude when Al content, jc, is increased by about 8%. This rapid increase in resistivity with x is predominantly due to an increase in donor ionization energy. The temperature-dependent resistivity results for n-Al;,Gai_;,N (x > 0.7) in the temperature range from 70 to 620 K have been measured. At the fixed Si dopant concentration of3.5X lO^^cm"^, a strong temperature dependence is seen for samples with high Al content (x > 0.8). Weaker temperature dependence is observed for samples with lower Al contents (x < 0.8), which indicates the trend of moving towards the metallic behavior due to heavy Si doping. The Al content dependence of the Si donor ionization energy (£"0) is depicted in Figure 7.23(b), which clearly shows that EQ increases linearly with an increase in the Al content. The inset of Figure 7.23(b) illustrates an Arrhenius plot of the resistivity data for Alo.77Gao.23N, where a value of 41 meV for the Si donor activation energy (£"0) was obtained in this alloy sample. This deepening of £"0 is due to the fact that with an increase


165

in the Al content in AlGaN alloys, the bandgap and electron effective mass increase, while the dielectric constant and bandgap renormalization effect decrease. For pure AIN, an estimated value ofEo of about 120 meV is obtained by extrapolating the experimental data shown in Figure 7.23. Taniyasu et al. estimated that the donor ionization energy in heavily doped AIN is about 85 meV [116]. It is believed that, for optimized materials, the predominant cause for the rapid increase in resistivity of Al-rich Al^^Gai-^^N alloys with x is the deepening of the Si donor energy level. These recent experimental results indicate that we can achieve measurable n-type conductivity in pure AIN. The effect of persistent photoconductivity has been observed in Si-doped AIN grown by MBE, suggesting that Si may undergo a DX-like metastability [111]. However, more recent results seem to suggest that silicon impurities act more like an effective mass state [116,117], in agreement with a previous theoretical prediction [102]. These discrepancies may be accounted for by the fact that the concentrations of oxygen impurities and Al vacancies (VAI) or VAI complexes have been significantly reduced in highly conductive AlGaN alloys grown recendy by MOCVD and MBE. By further optimizing the growth condition and hence minimizing the effects of impurity compensation, it is possible to further improve the conductivity of pure AIN by heavy doping. 7,4,2 Mg Acceptors in AlGaN and AIN Understating the electrical properties of Mg-doped AlxGai-xN alloy is critical for further improving the p-type conductivity and hence the performance of optoelectronic devices based on these materials. Several groups have reported MOCVD growth and electrical properties of Mg-doped Al;cGai_;cN alloys with low Al contents [123-126]. The authors' group has achieved p-type conduction in Al;cGai-;cN epilayers for x up to 0.27 [127]. Mg doping in AIN has also been attempted. However, all as-grown and post growth annealed Mg-doped layers were highly resistive, although SIMS measurements revealed that Mgdopant concentration was about 7 X 10^^ cm~^ in Mg-doped AIN epilayers [127]. Figure 7.24(a) shows PL spectra measured at different temperatures for Mg-doped AIN epilayers. The low temperature (10 K) PL spectrum of an undoped AIN epilayer grown under identical conditions is also included for comparison. The band-edge transition at 6.06 eV is the dominant transition in undoped AIN epilayers, which is at a slightly higher energy position than those shown in Figures 7.6 and 7.14. Two broad emission lines with much lower emission intensities related with deep-level impurities at about 2.94 and 4.40 eV are also observable in undoped AIN at 10 K. Mg-doped AIN exhibits two dominant emission lines at 4.70 and 5.54 eV at 10 K, which are absent in undoped AIN and are thus attributed to Mg impurity related transitions. The spectral peak positions of these emission lines suggest that they are either band-to-impurity or DAP transitions. However, the slow component of the recombination lifetimes of both emission lines measured at 10 K are about 1 |JLS (not shown), which precludes the possibility for the band-to-impurity

166

Optoelectronic Devices: Ill-Nitrides (a) AIN: Mg

4

^

3

KSU-A1081

KSUA-1081 •

T=300K

(b)

....JxlO)

AIN: Mg Ep=5.54 eV Eo=60 eV

E _a)

(x10) _l

1 .-. ^ ^ - v i(x10) '

1 J fitted by Ln[l3jlo]=-Ln[1+ce-ô/kT]

0.00 2.0T

^.5.54 eV^

0.02

0.04

0.06

0.08

0.10

.

.

•

.

•

0.12

Ep=4.70 eV Eo=0.50 eV

•••^.^

\ / ^ 6.02 eV 4.70 e V N - / \ I

(180 150 K, respectively. The solid lines are the least squares fit of data with the thermal activation behavior described by Eq. (7.1). The fitted activation energy £"0 (= 60 meV) for the 5.54 eV emission line again suggests that the 5.54 eV emission line is a DAP transition involving a Mg acceptor and a shallow donor (with an ionization energy of 60 meV). Although the ionization energy of the shallow donor involved in the optical transition is in agreement with a value estimated for substitutional shallow donors in AIN by the effective mass theory by taking an electron effective mass of m* = 0.33mo [129,130], the chemical nature of the donor impurity involved is not clear. Possible candidates are Si and O impurities. The monotonic decrease of the 5.54 eV emission intensity with increasing temperature shown in Figure 7.24(a) is due to the thermal dissociation of the shallow donors involved. By neglecting the Coulomb interaction between the ionized donors and acceptors {-e^/sr with r being the distance between the ionized donor and acceptor and s being the static dielectric constant), a value of Ep, = 0.52 eV for the Mg acceptor binding energy in AIN is deduced from E;, ^ E^ - 5.54 eV - 0.06 eV with £„ = 6.12 eV at 10 K [53].


167

The 4.70 eV emission line was assigned to a DAP transition involving a deep-level donor and Mg acceptor. Due to the competing recombination channel at 5.54 eV at low temperatures, the thermal activation process of the 4.70 eV emission line is more complicated than that of the 5.54 eV line. However, the Mg impurity level Ep^ can also be obtained simply from the Arrhenius plot of the 4.70 eV PL intensity for T > 150 K, above which the 5.54 eV emission hne is no longer present. Based on the PL spectral peak position of the 4.70 eV line, it can be concluded that the energy level of the donor involved is deeper than that of the Mg impurities. Thus, the thermal activation behavior of the 4.70 eV hne is a direct measure of the activation of Mg impurities. A value of EQ = 0.50 eV is obtained from Figure 7.24(b) for the 4.70 eV emission hne, which corroborates quite well with the Mg acceptor binding energy of 0.52 eV deduced from the Arrhenius plot of the 5.54 eV hne. Therefore, the Mg acceptor level in AIN is about 0.51(± 0.01) eV above the valence band. Again by neglecting the Coulomb interaction between the ionized donors and acceptors, the binding energy of the deep-level donor (E^) involved in the 4.70 eV transition can be calculated from E^^ ^ E^ - 4 JO eV - 0.5 eV « 0.90 eV. Based on the results shown in Figure 7.24, the energy diagram for the impurity levels in AIN can be constructed, as shown in Figure 7.25. The upper limit of Mg^ level in AIN is estimated to be about 0.97 eV if we assume that the Mg^ level is lined up within the bandgaps of GaN and AIN near the interface of a AlN/GaN heterojunction [131,132]. The conduction band offset parameter is assumed to be 70% for the GaN/AIN heterostructure and EA is 0.17 eV for Mg acceptors in GaN [125,133,134]. The lower limit of the Mg^ energy level in AIN is determined by the effective mass theory with values scattered

0.90 eV

T

1.81 eV

\ shallow donor 0.06-0.12 eV

CB

deep donorT 4.70 eV 5_54ev6.11eV 3.50 eV 0.17 eV • ^ -

0.80 eV (30 %) GaN

0.97 eV

MgO (calculated)

_

I Mg°(exp). 0.51 eV 1 AIN

0.421 eVt O.aOieV '

•

*_

VB

Figure 7.25. Energy diagram showing the impurity levels, including Mg acceptor level, in AIN. The upper limit of Mg acceptor level in AIN is estimated to be about 0.97 eV by lining up its energy levels within the bandgaps of GaN and AIN near the interface of an AlN/GaN heterojunction. The shaded region shows Ej^ predicted by the effective mass theory with uncertainty due to scattered values of the reported hole effective masses (after Ref. [127]).

168


between 0.42 and 0.80 eV, as indicated by the shaded region in Figure 7.25, due to the uncertainty in the hole effective mass in AIN [135,136]. The experimental value of £A = 0.51 eV is close to the lower limit values obtained by the effective mass theory. By using the experimentally determined value of Ep^ = 0.51 eV and the high frequency dielectric constant of s(oo) = 4.6 [136,137], an average hole effective mass m^ of about 0.8mo in AIN is estimated, which agrees well with several calculation results [130,138]. The value of ml in GaN is determined to be about 0.46mo by using m}} = 2.03mo and m^ = 0.33mo [139]. The larger ml in AIN suggests a smaller hole mobility in AIN than in GaN. Figure 7.26 presents the determined Mg acceptor ionization energy in AIN and in Al;^Gai_;,N as a function of the Al content x. The dotted line is the linear extrapolation of data from Al^^Gai-^^N alloy with low Al content. The solid line is to guide the eyes by including the value for Mg-doped AIN. It is clear that the activation energy increases with the bandgap, as predicted by the effective mass theory. With the value of Mg acceptor ionization energy being determined for AIN, the issue regarding the possibility of achieving p-type AIN with Mg doping can be addressesed. As a consequence of the large value of Ej^ = 0.51 eV, only a very small fraction (e~^^^^^ = g-0.51/0.025 _ 2Q-9^ of the Mg dopants can contribute free holes for conduction at room

(a) 1.0

(b) 10^ AlxGa-i.xN: Mg

10

AlxGa-i.xN: Mg guide to the eyes

16

D Hal-effect data

0.8

E 10'

EA=0.17+0.5X

A 0.6

PL data guide to the eyes

10^ p=NAe-Â/^T N^=^ 020cm-3

>

10

(D< LJJ

10

10"-^ o 10°

a

|ih=1 cnr» A/s |Xh=10cmÂ/s guide to the eyes„„.-

10^ 10^

0.0 0.0

0.2

0.4 0.6 0.8 Al content (x)

1.0

10^ 0.0

0.2

0.4 0.6 Al content (x)

0.8

1.0

Figure 7.26. (a) Mg acceptor activation energy in Mg-doped Al^Gai-;cN as a function of the Al content (x). Closed squares are data from AI^GSLI-XN of low Al contest as determined from Hall measurements, whereas the filled triangle is the data for AIN obtained from Figure 7.24(b). The dotted line is the linear extrapolation of data from Al;,Gai_;,N of low Al contest to x = 1 and the solid line is a guide to the eyes, (b) Calculated hole concentration (p) and resistivity (p) of Mg-doped Al^^Gai -Ji?L (a),

(10.5)

where /?L is the load resistance and /„ is the noise current. The total noise current representation consists of contributions from many noise sources, but these can be simplified into the three most dominate sources of noise currents: 1// noise (jitter noise), shot noise (generation-recombination noise), and Johnson noise (thermal noise): < 0 = 53%) Mg-doped AlGaN by using an alternating gas flow growth technique in metal-organic vapor phase epitaxy (MOVPE). We used these techniques to fabricate 310-nm-band deep UV light-emitting diodes (LEDs) with quaternary InAlGaN active regions. We achieved sub-milliwatt output power under RT pulsed operation for 308-314 nm LEDs. We also demonstrated a high output power of 7.4 mW from a 352 nm quaternary InAlGaN-based LED fabricated on a GaN substrate under RT CW operation.

11.1. INTRODUCTION AlGaN or InAlGaN alloys are attracting much attention as candidate materials for reahzing deep-ultraviolet (UV) LEDs or LDs. Figure 11.1 shows the relationship between the direct transition bandgap energy and the lattice constant of the wurtzite (WZ) InAlGaN material system and the lasing wavelengths of gas lasers. The direct transition emission of AlGaN can be adjusted between 6.2 eV (AIN) and 3.4 eV (GaN). The wide emission range

E-mail address: [email protected] (H. Hirayama).

285

286

Optoelectronic Devices: Ill-Nitrides i93nm Wavelength ArF

257nnn • Ar-SHG

325nm He-Cd

200nm

6.0

248nm KrF

308nm xeCI

300nm r

400nm

Excimer Lasers 500nm

rr

Developing Area for UV-LEDs orLDs

5.0

ir-

> 4.0

iS 3.0

GaN

Q.

(0 U> "D C

CD 2.0

700nm I

Wavelength of Ga^Lasers 1.5jum

1.0K

InN

3.0 ^4.0 Lattice Const. (A) Figure 11.1.

Relationship between the direct transition energy bandgap and the lattice constant of wurtzite (WZ) InAlGaN material and the lasing wavelengths of gas lasers.

in the UV of (In)AlGaN covers the lasing wavelengths of various gas or solid-state UV lasers, including XeCl (308 nm) or KrF (248 nm) excimer lasers, N2 lasers (337 nm) and He-Cd (325 nm) or Ar-SHG (257 nm) lasers, as shown in Figure 11.1. Figure 11.2 illustrates some applications for high-efficiency semiconductor UV light sources with wavelengths between 250 and 350 nm. There are many appHcations for high-efficiency UV LEDs or LDs, i.e. long-lifetime white lighting, high-density optical storage, medical fields or biochemical processes and the purification of the environment. They are also useful for household air cleaners or sterilization systems, automobile exhaust purifiers, UV sensing systems, etc. Fluorescent lamps could be replaced by long-lifetime white lamps excited with UV LED arrays if low-cost, efficient UV LEDs could be achieved. To obtain high absorption efficiency of the phosphor used in white lighting, a wavelength below 350 nm is required. Efficient UV lights have also recently attracted considerable attention for the purification of the environment, i.e. the purification of river water, industrial waste water or atmospheric gases. It has been pointed out that an efficient 320-340 nm UV fight source is required for the photocatalytic decomposition of refractory pollutants (dioxin, PCB or NO^, gas, etc.) with titanium oxide (Ti02).

Quaternary InAlGaN-based UV LEDs

287

Applications of UV Light

White Lighting hOOOLEDs : 100W

|f

#4 White

High-Efficiency 30-40% '-°"9 J-ifetime:

HigH)ensity Memory DeepUV Laser Diode |(250-300nm)|

DVD Recorder using UV-Laser

Take place of fluorescent lamp U V L E D Array

U V D V D Disl
-Hrgh Densitv

J { Purified l^rf' Water)

Industrial Pollutants

Dioxin, PCB, U V L E D A r r a y etc. ."^Br-.Uunm

Another Applications:

(Water and Air Purfficafion) Rivem. Oceans. Lakes The atmosphere

• Automobile exiiaust • UV Sensors (NOx gas ) Purifier • Chemical or Biochemical Industries • Household Air Cleaner

Figure 11.2. Applications of UV LEDs or LDs with emission wavelengths of 250-350 nm.

In order to obtain high-intensity UV emission at a shorter wavelength than can be realized by InGaN, the development of AlGaN or quaternary InAlGaN-based UV emitters is necessary. Research into AlGaN-based UV LEDs was initiated by several research groups between 1996-1998. UV LEDs with wavelengths between 330 and 355 nm [1-3] have been reported using AlGaN-based structures. However, their output power was still low, due to a number of technical problems. Recently, high-power near-UV LEDs at a wavelength of 365 nm have been realized by Nichia, who use InGaN in the emitting regions [4]. External quantum efficiencies (EQE) as high as 30% have already been achieved for near-UV LEDs using InGaN. On the other hand, it is still difficult to obtain high EQE for wavelengths below 360 nm. There are several reports of 300-365 nm UV LEDs using AlGaN QWs [5-8] or InAlGaN QWs [9-12]. The maximum EQE for a 351 nm LED using an AlGaN QW fabricated on a GaN substrate was reported to be approximately 1% [5]. Recently, deep-UV LEDs with wavelengths shorter than 300 nm have been developed, i.e. 250 nm[13],267 nm[14],278 nm[15],280 nm[16], 285 nm [17] and 292 nm [18] using AlGaN-based QWs. The values of EQE obtained from 250-350 nm LEDs are still below 0.1% under CW operation. The realization of high-efficiency deep UV LEDs with wavelengths below 360 nm is still challenging because of some major problems in producing AlGaN-based UV emitters, i.e. the difficulty in obtaining efficient UV emission from AlGaN QWs at room

288


temperature (RT) and the difficulty of injecting holes through high-Al-content AlGaN. A reduction in threading dislocation density (TDD) is necessary in order to obtain efficient UV emission from AlGaN-based structures. However, reducing the TDD of AIN or highAl-content AlGaN is still difficult to achieve. The use of quaternary InAlGaN emitting layers is considered to be quite effective for the realization of high-efficiency UV LEDs or LDs using group III nitrides due to the efficient emission they exhibit at wavelengths shorter than 360 nm, which is attributed to Insegregation effects [19,20]. We proposed the use of the emission from a localized electronhole pair in the In-segregation region in quaternary InAlGaN for the purpose of obtaining RT bright UV emission [21]. It has been reported that the quantum-dot-like region [22] formed by In-segregation in InGaN QWs is very effective for the suppression of nonradiative recombination and that an InGaN QW exhibits efficient emission at RT [23,24]. A similar effect to that obtained in InGaN QWs is expected for quaternary InAlGaN. Due to this effect, quaternary InAlGaN is very promising for use as the active layer of 300350 nm-band LDs or LEDs. We have demonstrated that the intensity of the 320-340 nm UV photoluminescence (PL) from quaternary InAlGaN is as strong as that of the 430 nm blue emission from InGaN at RT [19]. We have already demonstrated high-efficiency UV emission from quaternary QWs of InviAlyiGai_;,i_^,iN/In;,2A1^2Gai_;,2->^2N in the wavelength range between 300 and 350 nm at RT [20] and efficient 340 nm-band InAlGaN-based LEDs on SiC substrates [9,25]. The most important advantage of the use of quaternary InAlGaN is that high-efficiency deep-UV emission is obtained, even for a layer grown on buffer with high TDD. The emission intensity of AlGaN epitaxial layers is very sensitive to TDD. Therefore, UV devices using AlGaN-based active regions need to be fabricated on low TDD buffer-layers or substrates. On the other hand, quaternary InAlGaN UV LEDs show high-efficiency operation, even when fabricated on high TDD buffer layers grown on sapphire substrates. The use of InAlGaN quaternary as the emitting region is considered to be particularly useful for deep-UV LEDs, since reducing the TDD is still difficult for high Al-content AlGaN. In this chapter, we describe techniques for the realization of high-efficiency 300350 nm-band UV LEDs. In Section 11.2, we describe the growth and optical properties of high-Al-content Al;,Gai_;,N for application to deep-UV emitters. The growth conditions for high-quality Al_;,Gai_;cN are investigated and a single peak spectrum from high-Alcontent Al_;^Gai-;,N (Al content up to 80%) is observed emitting from near the band-edge. Intense UV emission around 230-280 nm is demonstrated from Al^^Gai-^^N/Al^Gap^N multi-quantum wells (MQWs) with wide-bandgap AlGaN barriers. The intensity of the deep-UV emission from AlGaN-based MQWs is shown to be as high as that of the blue emission from InGaN-based QWs at 77 K. However, the 200 nm-band UV emission from AlGaN-based QWs at RT is much weaker than the blue emission from InGaN. In Section 11.3, we describe the growth and the optical properties of quaternary In;Âl^Gai_;,_^N to obtain intense UV emission at RT by using the In-segregation effect.


289

We reveal that efficient emission is obtained from lnx\P^yiGdii-x\-y\^l^^x2^^y2^^\-x2-y:^ QWs in the wavelength range between 290 and 380 nm at RT. The UV emission from quaternary InAlGaN is shown to be as strong as the blue emission from InGaN at RT. For the realization of deep-UV LEDs, it is also necessary to achieve high-Al-content p-type AlGaN. In Section 11.4, we describe an alternating gas flow growth technique for producing high-Al-content p-type AlGaN. We reveal that the use of an alternating gas flow is quite effective for obtaining high-Al-content AlGaN with high crystalline quality. We observed hole conductivity in Mg-doped AlGaN with Al compositions as high as 53% that was grown using the alternating gas flow method. In Section 11.5, we describe UV LEDs with InAlGaN emitting regions. We demonstrated sub-milliwatt output power under RT pulsed operation from 308-314 nm quaternary InAlGaN-based LEDs fabricated on sapphire substrates. We also fabricated quaternary InAlGaN QW LEDs on GaN substrates in order to eliminate the effects of TDD. We demonstrate an output power of 7.4 mW at a wavelength of 352 nm under RT CW operation. We also demonstrate the highest EQE ever obtained for 350 nm-band UV LEDs with top emission geometry. From these results, the advantage of using the quaternary InAlGaN for 300-350 nm-band UV-emitting devices is confirmed.

11.2. GROWTH AND OPTICAL PROPERTIES OF Al^Gai_^N In this section, we describe how we investigated the basic growth conditions and optical properties of AlGaN as a first step in developing quaternary InAlGaN with efficient UV emission at RT. We demonstrate efficient UV PL at wavelengths ranging from 230-280 nm from Al^Gai-^NCAlNyAl^^Gai^N MQWs grown on SiC by MOVPE [26,27]. We systematically investigated the PL intensity of AlGaN-based MQWs with widebandgap AlGaN barriers as functions of both the QW thickness and the Al content of the barriers [28,29]. The efficiency of the deep-UV emission from AlGaN-based QWs is shown to be as high as that of the blue emission from InGaN-based QWs at 77 K. The samples were grown at 76 Torr on the Si-face of an on-axis 6H-SiC (0001) substrate, using a conventional horizontal-type MOVPE system. Ammonia (NH3), trimethylaluminum (TMAl) and trimethylgallium (TMGa) were used as precursors, with H2 as the carrier gas. N2 gas was supplied independently by a separate line in order to control the gas flow. Typical gas flows were 2 standard liters per minute (SLM), 2 and 0.5 SLM for NH3, H2 and N2, respectively. The molar fluxes of TMGa and TMAl for the growth of Al;,Gai_;,N (x = 0.11-1) were 38-3.8 and 2.6-45 |ULmol/min, respectively. Under these conditions, the growth rates of Alo.11Gao.89N, Alo.40Gao.60N and AIN were approximately 2.4, 1.0 and 0.4 luim/h, respectively. The substrate temperature during the growth of the Al;,Gai_;,N, as measured using a thermocouple located at the substrate susceptor, was approximately 1140,1170 and 1200°C for Al contents (JC) of 10-40,40-80

290


and 80-100%, respectively. All of the samples were undoped. The molar fraction x of Al in the Al^^-Gai-^^N alloy was determined by four-crystal X-ray diffraction (XRD) measurements. We investigated the optical properties of the Al;,Gai_;cN alloy over the entire AIN compositional range, i.e. GaN to AIN. Figure 11.3 shows the spectra and full-width at half maximum (FWHM) of the PL emission observed from the Al;cGai-;cN alloy over the entire AIN compositional range at 77 K. We observed a single-peaked PL spectrum from near the band-edge over the entire Al compositional range. The yellow emission around 500-550 nm was negligible, even for high-Al-content AlGaN. The phonon-replica peaks seen at the low-energy side of each spectrum for Al contents of 0.11-0.53 confirms the high crystalline quality of the AlGaN. Typical values of FWHM of the PL spectrum at 77 K were approximately 20, 65 and 100 meV for Al contents of 10-20, 35-60 and 70-95%, respectively. We observed PL emission from AIN (208 nm) from near the band edge. Figure 11.4 shows the PL spectra of high-Al-content Al;,Gai_;cN (where x is greater than 80%) measured at 77 K. We observed that the PL emission of AIN is significantly

AI,Gai.,N

0.9^ 0.81' 0.69'

AIN 0.88 0.73

0.4^ ^ J

0.55

^-^^

^-..'

. J nft ^ ' ^

0.24 0.14 0.08 QgN

c

(U

200

220

240

260 280 300 320 Wavelength (nm)

150

I 100 X

50

0

200

Excited with ArF Excimer laser(193nm) 1

1

220

240

260 280 300 320 Wavelength (nm)

340

360

1

1

380

Excited with Ar-SHG laser (257nm)

340

360

380

Figure 11.3. Spectra and full-width at half maximum (FWHM) of the PL emission observed from the Al;(.Gai_^N alloy over the entire AIN compositional range measured at 77 K.


291

10^ AI^Ga 1-xN

'c

(Al:88%) I

(Al:83%)

(Al:93%)

CO

c J)

AIN

A

V 10^ 200

^

210

Excited with ArF Excimer laser (193nm) Measured at 77K 1 220

1 230

240

Wavelength (nm) Figure 11.4.

PL spectra of high-Al-content Al^Gai_;,N alloy (JC > 0.83) measured at 77 K.

enhanced by incorporating 5-10% of GaN into AIN. We hypothesize that the Al crystalline quality is greatly improved by introducing a small amount of Ga, which we characterized by XRD measurements. We then fabricated four series of AlGaN MQW samples, consisting of Al^Gai-^^N barriers with different Al contents. Table 11.1 summarizes the structure and the thickness of the buffer, barrier and well layers and the PL peak wavelength range of each series of AlGaN MQWs. Figure 11.5 shows an example of the schematic layer structure of an AIN/ AlGaN five-layer MQW sample (sample series (a) in Table 11.1). In order to achieve a flat surface suitable for the growth of AlGaN QWs, an approximately 250-400-nm-thick

Table 11.1. Structure and thickness of the buffer, barrier and quantum-well layers, displayed with the PL peakwavelength range of each series of AlGaN MQWs Sample series

Structure

Buffer

Barrier

(thickness, nm)

(thickness, nm)

Well (thickness, nm)

Peak wavelength (nm)

(a)

5-MQW

AIN (250)

AIN (5)

Alo.i8Gao.82 N (1.2-3.3)

229-285

(b)

on 6 H - S i C 5-MQW

Alo.8Gao.2N (250)

Alo.8Gao.2N (5)

Alo.i8Gao.82 N (1.3-3.3)

238-288

(c)


Alo.7Gao.3N (300)

Alo.7Gao.3N (5)

Alo.12Gao.88 N (1.4-2.7)

255-303

(d)


Alo.53Gao.47N (400)

Alo.53Gao.47N (5)

Alo.iiGao.89 N (1.4-3.4)

272-343

on 6 H - S i C

292

Optoelectronic Devices: Ill-Nitrides AIN(AIGaN) Capping Layer(20nnn) | \ AIGaNWell(1.2-3.3nm) /AIN(AIGaN) Barrier (5nm) 5-Layer MQW AIN(AIGaN) Buffer Layer (300nm)

Figure 11.5. Schematic layer structure of the fabricated AlN/Alo.i8Gao.82N MQW sample.

AlN(AlGaN) buffer layer was deposited, followed by a very thin AIN layer on a SiC wafer. We confirmed a step-flow-grown surface by atomic force microscopy (AFM) for each series of samples. The TDD of the AlGaN buffer was approximately 1 X 10^^ cm~^ [30]. As the next step, a five-layer MQW consisting of 1.2-3.4 nm-thick Al;,-Gai_;,N wells (jc = 0.11-0.18) and 5-nm-thick Al.Gap^N barriers (> = 0.53-1) and a 10-nm-thick Al3;Gat^,N cap {y = 0.53-1) were grown. The well and barrier thicknesses were simply estimated from the growth rate of the bulk AlGaN or AIN. Figure 11.6 shows PL spectra of (a) AlN/Alo.i8Crao.82N, (b) Alo.80Gao.20N/Alo.i8Gao.82N (c) Alo.70Gao.30N/Alo.12Gao.88N and (d) Alo.53Gao.47N/Alo.12Gao.88N five-layer MQWs for various well thicknesses. The samples were excited at 77 K using a Xe-lamp light source (215 nm) for sample series (a) and (b), a Xe-lamp light source (227 nm) for sample series (c) and an Ar-SHG laser (257 nm) for sample series (d). The excitation power densities with the Xe-lamp source and the Ar-SHG laser were approximately 20 W/cm^ and 5 kW/cm^, respectively. We obtained a single-peaked intense PL emission from every MQW. No yellow emission was observed from any of the samples. The most efficient emission was obtained at wavelengths of 234, 245, 255 and 282 nm for sample series (a), (b), (c) and (d), respectively. The optimum value for the well thickness was approximately 1.5 nm for each series of samples. The PL spectra of bulk Alo.80Gao.20N and bulk Alo.53Gao.47N (thickness approximately 400 nm) are also shown in Figure 11.6(b) and (d), respectively, as references. The PL intensity of the MQWs was 20-30 times larger than that of the bulk AlGaN, due to a quantized confinement effect. A quantized energy level shift can be clearly observed, as seen in Figure 11.6(a)-(d). The PL peak energies for different well thicknesses agree well with the calculated levels of the quantized energies, taking into account the piezoelectric field applied in the QW regions. The values of the piezoelectric field in the well regions, as estimated from the positions of the measured quantized levels, are more than 2 MV/cm, as already pointed out by Bemardini et al. [31]. The PL intensity depends heavily on the well thickness. A rapid reduction in the PL intensity with increasing well thickness was seen for each series of samples. The reason for this is considered to be a reduction of the radiative

[

1 C5 CM

d

CO

(0

\- z

—

2CNJ o

c

E

>

.'S O CD ^

1 ^ ^

^

1 £

1 l\

/ J

s

0)

^

1

2

(0

1 00 00

^

CD n (

1

cr

CNJ \

f^"^^—^-

T3 0 E

d

^ >

|v. CM

c

E

"~1_

0

1

1

1

•

CO

o CO

CO

A o

CN CO

o

o CO

o o

CN

00

CD CN

O CN

o

CN CN

O CD CO

o

O CN CN

CN

o

CD CN

o

00 CM

C)

^ v i^ oo -^^rl c\i

CN CO

o

o -sr £ J CO

1

^

^ •- / o E ;s \^] C CO 1^

1

iT:

E

U X LU X CD

°\.

Q

1 1 J _L_ (siiun qjB) Ai!SU8iu|

csi

E c c 0

>

i

c

o

I o < ^ -d

c?

00

:z;

oSH ^

^ ^

c ? |

S S2

^5

293

%K

S

«^-i

z ^ '^

5

d

o

00

:z;

^^ ^ c

c?

> i5 ^

(D CD

SI

^\1 ?

(siiun qjB) Aj!SU9iu|

_P_P

^

O z CO zCO CO

d d > (D TO O d

O CD ^

c

E

<
o H CM ^ S v

CO (1)

II 0)

(simn qjB) Ajisuaûi

1 ^

CO

-x^^^>^

o

L<
1.7 A for 7 mm diameter

E-mail address: [email protected] (S.J. Pearton).

323

324


devices and low forward turn-on voltages (Vp ~ 18 V). The reverse breakdown voltages in these structures are still limited by avalanche breakdown at defects and/or surfaces. The rapid progress in improving both defect density and purity of these free-standing substrates makes them the most promising approach for achieving both high V^ and on-state currents [11,12], in comparison to methods such as metal organic chemical vapor deposition (MOVCD) of lightly doped stand-off layers [5,13-16]. A simple model for avalanche breakdown in GaN resulting from impact ionization produces the relation [3] VB - 1.98X10^Â^"^^ where A^ is the doping concentration in the GaN. Currently, all GaN rectifiers show performance limited by the presence of defects and by breakdown initiated in the depletion region near the electrode comers. In SiC rectifiers, a wide variety of edge termination methods have been employed to smooth out the electric field distribution around the rectifying contact periphery, including mesas [17], high-resistivity layers created by ion implantation [18,19], field plates [20,21] and guard rings [22]. The situation is far less developed for GaN, with just a few reports of combined guard rings/ field-plate termination [8,9,23]. While SiC has been workhorse in the research area of high power devices, GaN has been dominant in the commercialization of light emitting devices and limited in the application of high power rectifiers due to the lack of freestanding substrates. The recent success of growing GaN freestanding wafers by hydride vapor phase epitaxy (HVPE) technology has geared up the power devices applications of GaN, especially high power rectifiers. In this chapter, the status of GaN high power rectifiers will be presented.

12.2. BACKGROUND In this review section, the theoretical calculations are given for reverse breakdown voltage and on-state resistance. Most parameters for GaN are extracted from epilayer GaN, and not from bulk GaN due to the lack of data. 12,2,1 Temperature Dependence ofBandgap (i) GaN EJQW)

= 3.475 - 9.39 X 10"^ 3.396 + 9 .39xlO"'^f

r + 772 300^ 300 + 772

r + 772

Design and Fabrication of GaN High Power Rectifiers

325

3.55 3.50 >- • • •

3-45

•

•

GaN

^

Eg = 3.396-^.5x10"^(T~-300)

3.40

^

|

/

I3.35 CD

D)

1

3.30

CG CD

3.25

I

4H-SiC

1 1

3.20 3.15 3.10 1

_i

100

200

1

_J

1

300

1

1

400

1

1—J

500

600

Temperature (K) Figure 12.1. Temperature dependence of GaN and 6H-SiC bandgap as a function of temperature. Dotted lines represent the linear fit mainly between 200 and 400 K [1].

For a linear fit between 200 and 400 K, £g(eV) = 3.463 - 5.3 X 10"^(r - 300), as shown in Figure 12.1. (ii) 6H-SiC £ (eV) = 3.19 - 3.3 X lO'^^Cr - 300) 72.2.2 Effective Density of States The effective density-of-states is derived by the following equation for most semiconductors, and determined by the effective masses of carriers and temperature. .3/2

NciT) = 2

h^

)

* \ 3/2 ,

2.50945 X 10 19

NîT) = 2 ( ^ ^ ^ ) ' " = 2.50945 X lO'^

Uo/

V300^

Uo/ v3oo;

N^{T) = 2.3 X 10'^(r/300)3'^ ml = 0.20TOO for wurztite GaN; N^{T) = 4.6 X lO^' (^/300)3'^ ml = 1.50mo for wurztite GaN [2].

326

GaN 6H-SiC

Optoelectronic Devices: Ill-Nitrides N, (cm~^)

A^v (cm 0

2.3 X 10^^ 1.66 X 10'^

4.6 X 10' 3.29 X 10^'

12,2,3 Intrinsic Carrier Concentration The intrinsic carrier concentration n^ is described by the following equations, where A^c is the density of states in the conduction band, A^^ the density of states in the valence band, and £'g is temperature-dependent bandgap. î(r)-V^VcAêxp(-^) n;(T)= l.QSxlO^^r^^êxp^

20488.6

f

)

for GaN

n^(300) = 2.25 X 10~^^ cm"^ for GaN, n,(300) = 1.6 X 10"^ cm"^ for 6H-SiC. This small intrinsic carrier concentration at room temperature for wide bandgap semiconductors (Figure 12.2) causes the numerical underflow errors when calculating minority carrier concentration due to np =

n[.

3.0 3.5 Temperature (1000/K)

4.0

Figure 12.2. Intrinsic carrier concentration in SiC and GaN as a function of temperature [3].


327

12,2,4 Incomplete Ionization of Impurity Atoms Acceptor dopants for GaN are not fully ionized even at high temperatures. This incomplete ionization of impurities can be expressed by the following equation using Fermi-Dirac statistics with impurity ionization levels and degeneracy factors for the conduction and valence bands. N^ =

Nl =

E, - E,, = kTlni^^^,

£fn -E, = ^ ^ l n ( ^ )

= 0.016 for Si, AÂ = 0.175 for Mg. In Medici™, the incomplete ionization of impurities is selected by specifying the FERMIDIRAC and INCOMPLE parameters on the MODELS statement. And the band degeneracy is given by the impurity dependent parameter GB. The donor and acceptor impurity activation energies are incorporated by the parameter EBO in the IMPURITY statement. Medici can provide for the doping and temperature dependence of the impurity activation energies. For very high doping concentrations such as more than 1.0 X 10^^ cm~^, the transition from incomplete ionization to complete ionization happens. Medici will take the complete ionization if the parameter HIGH.DOP is specified on the MODELS statement. The complete ionization will be assumed above for impurity concentration greater than HDT.MAX. A£D

12,2,5 Mobility Model (1) Analytical mobility model. (Temperature and concentration dependent mobility model)

,

Carriers Electrons Holes

Mrnaxl of\f\ ]

'V300

^n

A^max (Cm^/V S)

Mmin ( c m ^ A ^ S)

A^ref ( c m - ' )

a

P

7

1000 170

55 30

2 X 10^^ 3 X 10^^

-2.0 -5.0

-3.8 -3.7

1.0 2.0

328


1000 ^•~»-,^

*\

o

— , , . s^ tiectron \

Hole 100 :

\\

! \

\

- s'.

I "~~~1

i

l\.________ i 1—1—1 1 1 m

1E15

1

I

1 1 1 1 iii

1E16

1

1E17

1

1 1 1 1

III

I

l_L 1 1

1E18

1.LI

1E19

1

1

1 1 1

III

1E20

Carrier concentration (cm~^) Figure 12.3. Low-field mobility of electrons and holes as a function of doping concentration in GaN at room temperature [4],

(2) Field dependent mobility model. Mn(^) =

M'n i/A

[-m']

BET AN = 2.0 BETAP - 1.0 for GaN 2.7 X 10^

Figure 12.3 is the low field mobility at 300 K as a function of doping concentration. 12.2,6 Generation and Recombination (1) Shockley-Read-Hall (SRH) lifetime. The SRH recombination-generation rate RSRK is given by ^SRH

pn — n{ rJn + n{) + Tîp + n,)


329

where the Hfetimes T^ and Tp of electrons and holes, respectively, are dependent upon the doping level, as described by the Scharfetter relation

1+ The lifetimes for GaN are observed in the order of 1 -100 ns. The following relation is often used for Si and is also used for GaN. '^nO = 5TpO

(2) Auger recombination Âu = (^pP + C^n){np - nl) Auger recombination constants Q = 1 X 10"^^, Cp = 1 X 10"^^ cm^/s for GaN. Carriers

T„,p (s)

iVf"

7ns

Q p (cm'/s)

Electrons Holes

1 X 10"^ 1 X 10"^

5 X 10^^ 5 X 10^^

1 1

1 X 10~^° 1 X 10"^^

12,2,7 Reverse Breakdown Voltage The phenomenon of reverse breakdown is explained by avalanche multiplication, which involves impact ionization between host atoms and high-energy carriers. When a highenergy hole or electron under high electric field impacts an electron in the valence band, it will produce a new electron-hole pair (EHP). This newly generated EHP will cause other collisions and multiply carriers very rapidly. Avalanche breakdown is defined to occur in theory when ap exp

(an - a^)dx dx > 1

a^ — aQ exp

( ^ )

where Wj^ is the depletion width, a^ and a^ are the ionization rates of electrons and holes. Oguzman [32] and Kolnik [33] calculated the hole-initiated and electron-initiated ionization rate of electron and hole, respectively, for both wurtzite and zinc blende GaN. Figure 12.4 represents the calculated impact ionization coefficients as a function of inverse electric field for electrons and holes in wurtzite GaN and 6H-SiC [34-36]. GaN = 8.85xl0êxp[-^:^^]

(cm-^) [2]

330


10Ê-' E 0)

CD N

GaN electron & hole • Power-law fit for GaN 6H-SiC electron 6H-SiC hole • Power-law fit for 6H-SiC

Q.

E 10^ ^

- J

10° 2.5x10-'^

I

S.OxlO-"^

GaN

""-.^

I

\

S.SxIO"^

4.0x10"^

4.5x10-"^

S.OxlO""^

Inverse electric field (V/cm)"'' Figure 12.4. Calculated impact ionization coefficients as a function of inverse electric field for electrons and holes in wurtzite GaN and 4H-SiC [32,33].

6H-SiC an = 1.66 X lOêxp

/

-

1.273 X 10^ \

1 4^10

ap = 5.18 X 10^ exp( - —^

(cm-')

) (cm"^) [34]

To calculate the integrals of impact ionization without the aid of computer is timeconsuming and almost impossible. Thus, a power-law approximation can make the calculation easier in Fulop's form [36]; a^^^ = AE^ GaN (Fulop's form) a^^^ = A£" = 9.1 X m~^'E' [36] 6H-SiC (Fulop's form) a^^^ = AE"" = 4.55 X 10"^^£^ [37]


331

Simplified breakdown condition is expressed by the following ionization integral. rV

Jo

aeff dx = 1

Therefore, the depletion layer width at breakdown for GaN is

The critical electric field (Figure 12.5) can be calculated by ID Poisson's equation (dE/dx = qN^/sso) and obtained through numerical substitutions. d^V _ _ d ^ _ dx^ dx ,1/8

'

U

/

o

680 /

qN^ SSQ \l/7

X'^

\AWj

If Poisson's equation is solved with the voltage and electric field relationship, the breakdown voltage for non-punchthrough junction case is given by (Figure 12.6) ^Vpp = 2.87 X lO^Â^B ^^"^

7x10^

[ 6x10^ h

E

GaN^/^

5x10® TJ

^ a> o 4x10® o Q) h o

^

iC=

s

3x10®

o 2x10®

^,.^^\\-^\Z

f ^ t-"^

1x10® 1E15

^

L

1E16

1

J.

1

l-,_J-„U,iJL.

1E17

1E18

Doping concentration (cm"^) Figure 12.5. The critical electric field for 6H-SiC and wurtzite GaN as a function of doping concentration.

332


F 10^

F ^^^^"^^'^^-^^^' --..^^^^GaN 10^

f

I

eHJsic ^^^^^^^^^^^^^^

^

^

^

102

10^ 1E15

1

—

1

1

1

1

1

•

I

1

1E16

1

1E17

•

1

i

1

l_LJ

1E18

Doping concentration (cm~^) Figure 12.6. The reverse breakdown voltage of non-punch through junction for 6H-SiC and wurtzite GaN as a function of doping concentration.

Each the depletion layer width, the critical electric field, and the breakdown voltage for 6H-SiC is given by

»'-(i)'"(^r=^-^x'»'"•"

In the case of punchthrough junction diode, the breakdown voltage is given by BVpT = E^WTTT PT

-

2eeo

Figure 12.7 is a plot of theoretical breakdown voltage of GaN punchthrough diode as a function of doping concentration and drift region thickness. It can be seen that 3 fim epilayer with doping concentration of 10^^ cm~^ gives more than 900 V of breakdown voltage. The actual experimental value of breakdown voltage is far from these theoretical predictions. The material imperfection such as the vertically threading dislocations leads to premature breakdown. Therefore, the edge termination technique should be developed for GaN to prevent the early breakdown, and the crystal quality should be advanced to improve the GaN device performance.


1E15

1E16

333

1E17

Doping concentration (cm"'^) Figure 12.7. The reverse breakdown voltage of punch through junction for GaN as a function of doping concentration and drift region thickness.

12,2,8 On-state Resistance The specific on-state resistance of unipolar diode is a sum of the drift region resistance, the contact resistance and the substrate resistance. R.diode

^drift + ^siib + ^(

The specific on-state resistance of drift region is given by R

= r

^

_

^D

where /x is the low-field mobility {î = 1000 cm^/Vs for GaN), N^ is the doping concentration of drift region and W^ is the drift region thickness. The on-state resistance of drift region for GaN can be expressed by the reverse breakdown voltage and given by R^^ = 2.4 X 10" ^^^y^-^ (Ct cm^) as shown in Figures 12.8 and 12.9. For 6H-SiC R^^ = 1.45 X lO'^^^y^-^ ( a cm^)

334

Optoelectronic Devices: Ill-Nitrides 0.030

—-—— ———-

0.025 £ ^

r

0.020

i

c

I

0.015

/

GaN

(D

S 0.010 - - - -

-—

•

-

—

-

—

^^^

"

c

o 0.005

-..-.••^•-n,

„ „ — •

1

'—

2000

4000

6000

8000

10000

Reverse breakdown voltage (V) Figure 12.8. The specific on-state resistance for GaN Schottky diode as a function of breakdown voltage.

1 r : £ ?

/

• ,. ,.ÂIGaN-UF AIGaN-UF GaN-UF/

Si /^ 0.1 IT

a

AIGaN-UF. GaN-UF /^

'-

c

•

CO

/ GaN-Calte(Dh /

\

3 (A C

o o

1E-3

'

/

1

102

1.

A / /

GaN-UF

""7

1E-4

/

6H-SiCy^ / GaN

GaN-UT

^/^ , _ ^ ' ' GaN-UF

/

GaN-UF/

/

/•

2? 0 . 0 1 ^

1

/

GaN Schottky rectifiers

1

I I — I I I ,

/

I

n

10^

„ 1

1

il

1

In, 1

1

10"^

Reverse breakdown voltage (V) Figure 12.9. The specific on-state resistance for Si, 6H-SiC, and GaN diode as a function of breakdown voltage.


335

Hybrid Electric Vehicle - Puris Regenerative brake equipment

-^^r?'""'"'"*'^^^^^^"-.'

Niclcel-metai-hycfride battery

Engine MMa0^'

^S îiiiiiBittil

]§Wf^^] 1

Fuel Tanl

= ^g,GaN(l - X)-^ ^g,AlNA^ - Â 50 [xm provided the carrier concentration is < 5 X 10^^ cm~^, so that both high forward currents and high reverse breakdown voltages should be possible with this approach. 12.7.

CONCLUSIONS

In summary, the size and geometry dependence of GaN Schottky rectifiers on both sapphire and quasi-bulk substrates has been investigated. The reverse breakdown voltage increases dramatically as contact size is decreased and is also much larger for vertically depleting devices. The low on-state resistances produce high figure-of-merits for the rectifiers and show their potential for applications involving high power electronic control systems [24-38]. REFERENCES [1] [2] [3] [4] [5] [6] [7] [8]

[9] [10] [11] [12]

Heydt, G.T. & Skromme, B.J. (1998) Mater. Res. Soc. Symp. Proc, 483, 3. Brown, E.R. (1998) Solid State Electrochem., 43, 1918. Trivedi, M. & Shenai, K. (1999) /. Appl. Phys., 85, 6889. Pearton, S.J., Ren, F., Zhang, A.P. & Lee, K.P. (2000) Mater. Sci. Eng., R30, 55. Bandic, Z.Z., Bridger, D.M., Piquette, E.G., McGill, T.C., Vaudo, R.P., Phanse, V.M. & Redwing, J.M. (1999) Appl. Phys. Lett., 74, 1266. Zhang, A.P., Johnson, J.W., Ren, F., Han, J., Polyakov, A.J., Smirnov, N.B., Govorkov, A.V., Redwing, J.M., Lee, K.P. & Pearton, S.J. (2001) Appl Phys. Lett., 78, 823. Zhang, A.P., Dang, G., Ren, F., Han, J., Polyakov, A.Y., Smirnov, N.B., Govorkov, A.V., Redwing, J.M., Cao, X.A. & Pearton, S.J. (2000) Appl. Phys. Lett., 76, 1767. Johnson, J.W., LaRoche, J.R., Ren, F., Gila, B.P., Overberg, M.E., Abemathy, C.R., Chyi, J.I., Chuo, C.C., Nee, T.E., Lee, CM., Lee, K.P., Park, S.S., Park, J.I. & Pearton, S.J. (2001) Solid State Electrochem., 45, 405. Johnson, J.W., Zhang, A.P., Luo, W.B., Ren, F., Pearton, S.J., Park, S.S., Park, Y.J. & Chyi, J.I. (2002) IEEE Trans. Electron. Devices, 49, 32. Johnson, J.W., Luo, B., Ren, F., Palmer, D., Pearton, S.J., Park, S.S. & Park, Y.J. (2002) Solid State Electron., 46, 911. Park, S.S., Park, l.W. & Choh, S.H. (2000) Jpn. J. Appl. Phys., 39, LI 141. Morkoc, H. (2001) Mater Sci. Eng., R33, 135.

350

Optoelectronic

Devices:

Ill-Nitrides

[13] Zhang, A.P., Dang, G.T., Ren, F., Cho, H., Lee, K.P., Pearton, SJ., Chti, J.-L, Nee, T.E. & Chuo, C.C. (1999) Solid State Electron., 44, 619. [14] Ren, F., Zhang, A.P., Dang, G., Cao, X.A., Cho, H., Pearton, SJ., Chyi, J.I., Lee, CM. & Chuo, C.C. (1999) Solid State Electron., 44, 619. [15] Zhu, T.G., Lambert, D.J., Shelton, B.S., Wong, M.M., Chowdhurg, V. & Dupuis, R.D. (2000) Appl. Phys. Lett., 11, 2918. [16] Shelton, B.S., Zhu, T.G., Lambert, D.J. & Dupis, R.D. (2001) IEEE Trans. Electron. Devices, 48, 1498. [17] Neudeck, P.G., Larkin, D.J., Rowell, A. & Matus, L.G. (1994) Appl. Phys. Lett., 64, 1386. [18] Morisette, D.T., Cooper, J.A., Jr., Melloch, M.R., Doing, G.M., Shenoy, P.M., Zakari, M. & Gladish, J. (2001) Trans. IEEE Electron. Devices, 48, 349. [19] Itoh, A., Kimoto, T. & Matsunami, H. (1996) Electron. IEEE Devices Lett., 17, 139. [20] Saxena, V., Su, J.N. & Steckl, A.J. (1999) IEEE Trans. Electron. Devices, 46, 456. [21] Tarplee, M.C., Madangarli, V.P., Zhang, Q. & Sudarshan, T.S. (2001) IEEE Trans. Electron. Devices, 48, 2659. [22] Dyakonova, N.V., Ivanov, P.A., Koglov, V.A., Levinshtein, M.E., Palmour, J.W., Pumyantsev, S.L. & Singh, R. (1999) IEEE Trans. Electron. Devices, 46, 2188. [23] Zhang, A.P., Dang, G., Cao, X.A., Cho, H., Ren, F., Han, J., Chyi, J.-L, Lee, CM., Nee, T.E., Chuo, C.C, Chi, G.C, Chu, S.N.G., Wilson, R.G. & Pearton, S.J. (2000) MRS Internet Nitride J. Semicond. Res., 551, W11.67. [24] Mehandru, R., Gila, B.P., Kim, J., Johnson, J.W., Lee, K.P., Luo, B., Onstine, A.H., Abemathy, C.R., Pearton, S.J. & Ren, F. (2002) Electrochem. Solid State Lett., 5, G51. [25] Hong, M., Ng, H.M., Kwo, J., Kortran, A.R., Baillargeon, J.N., Chu, S.N.G., Mannaerts, J.P., Cho, A.Y., Ren, F., Abemathy, CR. & Pearton, S.J., (2000) Presented at the 197th Electrochemical Society Meeting, May 2000, Toronto, Canada. [26] Luo, B., Johnson, J.W., Kim, J., Mehandru, R., Ren, F., Gila, B., Onstine, A.H., Abemathy, C.R., Pearton, S.J., Baca, A.G., Briggs, R.D., Shul, R.J., Monier, C & Han, J. (2002) Appl. Phys. Lett., 80,1661. [27] Brezeane, G., Fernandez, J., Millan, J., Rebello, J., Badila, M. & Dilimot, G. (1998) Mater Sci. Forum, 264-268, 941. [28] Teisseyre, H., Perlin, P., Suski, T., Grzegory, I., Porowski, S., Jun, J., Pietraszko, A. & Moustakas, T.D. (1994) J. Appl. Phys., 76, 2429. [29] Levinshtein, M.E., Rumyantsev, S.L. & Shur, M.S. (2001) Properties of Advanced Semiconductor Materials, Wiley Interscience, New York. [30] Kolessar, R., Nee, H.-P. (2001) APEC 2001, Sixteenth Annual IEEE, vol. 2. [31] Mnatsakanov, T.T., Dmitriev, M., Bathova, L., Osinsty, A. & De Huth, L. (2003) Solid State Electron., 47, 2498. [32] Oguzman, I.H., Bellotti, E., Brennan, K.F., Kolnik, J., Wang, R. & Ruden, P.P. (1997) J. Appl. Phys., 81, 12. [33] Ruff, M., Mitlehner, H. & Helbig, R. (1994) IEEE Trans. Electron. Devices, 41, 1040. [34] Trivedi, M. & Shenai, K. (1999) J. Appl. Phys., 85, 6889. [35] Kolnik, J., Oguzman, H., Brennan, K.F., Wang, R. & Ruden, P.P. (1997) /. Appl. Phys., 81, 2. [36] Fulop, W. (1967) Solid State Electron., 10, 39. [37] He, J., Wang, Y., Zhang, X., Xi, X., Chan, M., Huang, R. & Hu, C (2002) IEEE Trans. Electron. Devices, 49, 933. [38] Ren, F. & Zolper, J.C (2003) Wide Energy Bandgap Electronic Device, World Scientific, Singapore, p. 125.


Chapter 13

GaN Negative Dififerential Resistance Components with Terahertz Operation Capability: From Fundamentals to Devices Dimitris Pavlidis Department of Electrical Engineering and Computer Science, The University of Michigan, 1301 Beal Ave., Ann Arbor, MI 48109-2122, USA and Department of High-Frequency Electronics, Technische Universitdt Darmstadt, Merckstaê 25, D-64823 Darmstadt, Germany

13.1.

INTRODUCTION

Negative differential resistance (NDR) devices are solid-state components based on the NDR principle, providing a solution for low-cost, low noise, and high power microwave radiation sources. Such devices are extensively used in intrusion alarms, microwave test instruments, and automotive collision avoidance systems. Currently, these devices are fabricated on GaAs and InP, which are III-V, direct bandgap semiconductor materials exhibiting NDR. Although the devices based on these materials have proven to be reliable, they are limited in power and operation frequencies due to fundamental material features related to the bandstructure and transport properties. For higher frequency and high power applications like military radar, high-speed communications, and possibly biological agent detection, it is necessary to explore other semiconductor materials. One such material that is being investigated for these applications of NDR devices is GaN. GaN is a wide-bandgap semiconductor material whose velocity-field characteristics are expected to present NDR. The advantages of using GaN over GaAs and InP, in terms of increased power handling capability and higher operation frequency, are directly related to its larger intervalley energy gap and higher peak and saturation velocities. Although in GaN, further confirmation is needed regarding the presence of a transferred-electron effect (which causes NDR) and the processing technology of GaN is in its early stages, it has a lot of potential and thus is worth investigating.

E-mail address: [email protected] (D. Pavlidis).

351

352 13.2.

Optoelectronic Devices: Ill-Nitrides PRINCIPLES OF OPERATION OF NDR DEVICES

NDR devices operate on the principles of the Gunn effect although, as explained later other mechanisms may also play a role in the presence of NDR in GaN-based devices. The generalized Gunn effect describes the mechanism of electron transfer leading to a negative differential conductivity (resistance) in a homogeneous, bulk semiconductor material [1]. These devices are unipolar, two terminal devices. The semiconductor materials that exhibit the Gunn effect must be direct bandgap materials that have more than one valley in the conduction band and the effective mass and the density of states in the upper valley(s) must be higher than those in the main valley. The typical velocity-field characteristics of a bulk material exhibiting the Gunn effect are shown in Figure 13.1a. The main valley of the conduction band is where the electrons initially reside with little or no external electric field applied as shown in Figure 13.1b. If the external field is increased to a value above a threshold field. Figure 13.1c, many of the electrons acquire enough energy to be scattered ("transferred") into the upper satellite valley. Since the effective mass in the upper satellite valley is larger than in the main valley, the mobility, and the average drift velocity of the electrons is reduced. The mobility is given by er m*

f^

(13.1)

where r is the relaxation time and m* is the effective mass in a semiconductor material. The differential mobility, defined as, /JL^ = dv/dE, becomes negative when the electric field is above the threshold value as this can be seen in Figure 13.1a. This leads to the NDR.

(b)

ik

(a) ^pk

/^J\.

^saf /

1

^th

J,

• £

£<e,y

^'sat

(c) e

-• k

(d) £

\J -> k

-• k Cfh < e < e«;

e >£^,

Figure 13.1. (a) Generalized velocity-field characteristics of a transferred electron device, (b)-(d) Simplified energy-band diagram for a direct two-valley semiconductor showing electron transfer.

GaN Negative Differential Resistance Components

353

As the field is continually increased beyond the saturation field. Figure 13.Id, the drift velocity of the electrons saturates. To see how microwave frequencies can be produced from a bulk semiconductor material exhibiting the transferred electron effect as described above, further analysis is needed in the region of negative differential mobility, which is between the threshold field and the velocity saturation field in Figure 13.1a. A sample of bulk semiconductor material biased in the negative differential mobility region, (under uniform doping concentration and uniform electric field) is thermodynamically unstable. In an attempt to establish a steady state, an energetically more favorable state is reached if, instead of having a homogeneous distribution of charge over the sample length, the charges split up into space charge regions [1]. Any existing space charge inhomogeneity, Q{x,t), traveling at a velocity, v, will follow an exponential law given in Eq. (13.2) that can be derived from Maxwell's equations [2]: Q(x, t) = Q(x - vt, 0)exp^- - J

(13.2)

T is the energy relaxation time given by

where N^ is the doping concentration and fi^ is the differential mobility. From Eq. (13.2) it is clear that at low electric fields, when fi^ > 0, the charge inhomogeneity decays with r = r^, the dielectric relaxation time. At electric fields above the threshold value, when in^ < 0, the charge inhomogeneity can grow. The growth factor is given by

(V)(T)

S,V

where / is the device length and v is the electron velocity. Equivalently, NDI > T T ^

(13.5)

If the growth factor given by Eq. (13.5) exceeds the threshold level, then the charge inhomogeneity reaches a significant level and grows into a dipole domain. From Eq. (13.5) it is clear that the design of NDR devices lies in choosing the device length, /, and the doping concentration, A^^. One would want to choose / as small as possible to minimize the electron transit time from cathode to anode and Nj^ as large as possible to satisfy Eq. (13.5). However, there is a lower limit in choosing the device length, called the "dead space layer". The dead space layer arises, due to the fact that the scattering of electrons into the upper satellite valley occurs over a finite distance. It is well known that using a method of hot electron injection can reduce the dead-space layer and enhance the performance of

354


NDR devices [3]. Furthermore, to avoid the formation of static domains at the anode, Nj) should not exceed the critical doping concentration given by A^CRiT =

(13.6)

^xFj^/q

In Eq. (13.6), Fj^ is the threshold electric field. Eqs. (13.5) and (13.6) are central to the design of NDR devices. From the above discussion, since the charge is inhomogeneous in the negative differential mobility region, the electric field distribution over the sample length is also inhomogeneous. A typical electric field distribution in a NDR device is shown in Figure 13.2a. It can be seen from Figure 13.2a that there is a region of high electric field surrounded by a region of low electric field. The corresponding charge distribution is shown in Figure 13.2b. In Figure 13.2 the accumulation of charge in the region where the electric field is increasing is a result of significant decrease in mobility caused by scattering of electrons into the upper valley. In the region of the sample where the electric field is decreasing, the electrons accumulated in the upper valley lose energy and scatter back into the lower valley of the conduction band. When this happens, a depletion region is created. The accumulation and the depletion regions together form the dipole domain. The dipole domain forms near the cathode and propagates along the length of the device to the anode as shown in Figure 13.3. When the domain reaches the anode it collapses and a new domain is formed at the cathode. The process of domain build-up, propagation, and extinction is repeated.

(a) ^ 4 eh

->x

(b) A/A

-•X

Figure 13.2. (a) Electric field profile for a dipole domain, (b) Carrier concentration for a dipole domain.

355


Figure 13.3. Schematic representation of electric field and charge inhomogeneity in a NDR device.

In this way, transit-time current oscillations are produced in the external circuit which have a fundamental frequency given by

/ = - = T

L

VD

L

(13.7)

where, v^ is the electron drift velocity and L is the length of the active region of the device. The fundamental frequencies of these devices can be well within the microwave region for velocities on the order of 10^ cm/s and sample lengths in the micron range making them suitable for microwave signal sources.

13.3. TRANSPORT PROPERTIES OF GaN GaN is becoming the material of choice for high temperature and high-speed semiconductor devices due to the very large bandgap and large peak electron velocity in the material. It is important to understand the basic material properties of GaN in order to evaluate the advantages of using GaN in the NDR signal generators. The study of the fundamental properties of bulk GaN material through simulations indicates that this material exhibits a transferred-electron effect. Many simulations have been done using Monte Carlo techniques to study in detail the basic transport properties in the GaN material. It is well known that GaN crystallizes in both zinc-blende and wurtzite (the commonly used phase for experimental device development) structures with slightly different material properties and substantially different band structures. Studies of the material properties of both wurtzite and zinc-blende GaN have been done using numerical simulations. Kolnik et al. [4] have done calculations of the drift velocity versus electric field dependencies based on the ensemble Monte Carlo simulations. These results are shown in Figure 13.4. These simulation results show that GaN exhibits negative differential mobility. Various possibilities are suggested for the nature of the NDR effect, including the electron intervalley transfer and the inflection of the central valley. Although further confirmation is needed on the nature of the NDR, Gunn domain instability is expected according to


356

r~ 2h

4

V

^"^^^

f l y

5' ••-zinc blende QaN o-wuitzita GaN • ^ M

0

100

200

300

400

500

Electric field [kV/cm] Figure 13.4. Calculated steady state electron drift velocity in bulk GaN as a function of applied electric field along the (100) direction in the zinc-blende phase and within the basal plane along the (1010) direction in the wurtzite phase. (After Ref. [4]).

the simulated velocity-field characteristics, making this material very promising in applications based on the Gunn effect. Other simulations of GaN properties such as the electron transit time have also been done using Monte Carlo techniques. The electron transit time, the time it takes for electrons to travel across the active region of the device, is fundamental in determining the frequency of operation of the device. The result of the Monte Carlo simulation done by Foutz et al. [5] comparing the electron transit time, for GaN and GaAs, is shown in Figure 13.5. From Figure 13.5, it is obvious that the electron transit time in GaN is much lower than in GaAs. This is due to the fact that the velocity of electrons is higher in GaN than in GaAs. Furthermore, characterization of GaN grown by plasma-assisted molecular beam epitaxy on (0001) sapphire was done by Moustakas et al. [6]. Transport properties for a large number of GaN samples, grown with a different nitridation and GaN buffer conditions, were investigated by Hall effect measurements [6]. The obtained dependence of electron mobility as a function of free carrier concentration at room temperature is shown in Figure 13.6. In Figure 13.6 the bell shape of the curves is believed to be due to scattering by charged threading dislocations combined with ionized impurity scattering. At very high carrier concentrations, the dislocations are screened out and impurity scattering becomes the dominant mechanism. Moustakas et al. [6] also studied the vertical and lateral transport in the GaN films grown by plasma assisted MBE on sapphire. They found that the vertical


357

6-1

GaN (Zincblende) 0.0

—I 0.2

1 1 0.4 0.6 Distance (micron)

r— 0.8

1.0

Figure 13.5. Electron transit time as a function of distance. The field strengths chosen minimize the transit time across 1 |xm. The applied fields are 5 kV/cm for GaAs and 150 kV/cm for wurtzite GaN and 120 kV/cm for zincblende GaN. (After Ref. [5]).

1000

iiiii

1 1 1 iiiiii

1 Ill

1 N„,,=5x10^cm-^:

' 0,.

I

> ^l

2 N„„=2x10^»cm-^:

T N
200.0 h -L 0.0

100.0

J_

_L

J-

200.0

300.0

400.0

500.0

Temperature (K) Figure 13.8. Mobility of zinc-blende GaN for various temperatures. (After Ref. [7]).

360


the temperature is increased, the low field mobility is decreased significantly. The decrease in low field mobility at high temperatures can be explained by considering the dominant scattering mechanisms. From Figure 13.7 it is clear that polar optical phonon scattering is dominant at low fields and high temperatures resulting in the degradation of mobility. In Figure 13.9 the simulated electron drift velocity is plotted as a function of electric field with temperature as a parameter. While the peak velocity increases significantly with decreasing temperature the saturation velocity is less dependent on temperature. Based on the above results it is expected that diode operation will be limited more for higher doping designs and higher temperature of operation. When a high electric field F > Fj^, is applied to bulk GaN, electrons experience a negative differential mobility /X-NDR [8]- Under these conditions, a non-uniformity of electron concentration would grow at a rate 1/TDDR. TDDR is known as the differential dielectric relaxation time and is calculated using expression (13.8) TDDR = ^

^

(13.8)

where A^ is the electron concentration, s is the dielectric constant, and ÛNDR is the peak negative differential mobility. It is recognized that domain growth lasts for at least 3 X TQDR [9] and, thus, the operation frequency of NDR devices can be limited by the active layer doping. The dependence of frequency capabilities on A^ for GaN and GaAs was calculated using the material parameters of Table 13.1. The material parameters used are summarized again in Table 13.2 together with the TNDR values and the parameters for Zb GaN. The negative differential mobility in GaAs is larger than in GaN and, therefore, for low-doped devices, growth of electron domains occurs faster in GaAs than in GaN.

3.0

'

' —•'

1

'

r

I «

P

'•^

1

1

1

GaN 2.0

> Q

'

-t- ^ -•-

r r 0.0

W

0.0

,

1

100.0

1

T=100K TS300K T=500K

_ L ,

200.0

1

1

300.0

1

400.0

J

500.0

Electric field (kV/cm) Figure 13.9. Drift velocity as a function of electric field for undoped, zinc-blende GaN for various temperatures. (After Ref. [7]).


361

Table 13.2. Semiconductor material parameters of GaAs and GaN Material

FxH

(KV/cm) GaAs WzGaN ZbGaN

3.5 150 80

M-NDR

^SAT

^PEAK

(MV/cm)

(cm/s)

(cm/s)

(V/cm^/s)

(V/cm^/s)

0.4 2 1.2

0.6 X 10^ 2X10^ 1.7X10^

1.5 X 10^^ 2.9 X 10^ 3.5 X 10^

8000 280 730

-2500 -50 -220

%DR ( p s )

9.4 1.4 0.25

However, as A^ is increased, TDDR is reduced, and the frequency capability improves until it reaches the NDR relaxation frequency/NDR discussed in Section 13.2. Since/N^R^ < f§^ the frequency capability of GaN-based devices improves for higher A^ without being limited by /NDR as in case of GaAs. This leads to GaN NDR operation that exceeds the GaAs limit of 105 GHz for GaN doping levels above 5 X 10^^ cm~l (A^ X L) criteria for the possibility of Gunn domain instability are based on the fact that the domain growth rate 1/TDDR should be higher than the transit frequency /x = VPEAK^Â (N and L indicate the electron density and length of the semiconductor in which the Gunn effect may be present): {N^L^) > (NDo - ^^^^55AK

(13.9)

where Np^ is the doping and L^ is the thickness of the active layer and the factor 3 accounts for the domain growth time as explained earlier. The critical values of (NL) product for GaN and GaAs were calculated using Eq. (13.9) and the material parameters of Table 13.1, and the results are summarized in Table 13.3. The results show that, due to a higher peak velocity and a smaller negative mobility, (NL)Q for GaN is 10-100 times larger than for GaAs. However, if AÂ exceeds the critical doping concentration A^CRIT^ static domains can be formed inside the active layer [9]. Formation of parasitic static domains results in a decrease of output power and may lead to an early breakdown. Values of critical doping concentration A^CRIT calculated using Eq. (13.10) for cases of GaN and GaAs are also listed in Table 13.3. A^CRiT = ^ ^ (13.10) q Due to the large difference in threshold electricfields,A^CRIT in GaN is much higher than in GaAs and, thus, the active region in GaN diodes can be doped significantly higher Table 13.3. (NL)o products and critical doping levels for GaN and GaAs Material (NL)o (cm-2)

GaAs

ZbGaN

WzGaN

0.1 X 10^^ 3.4 X 10^^

2.5 X 10^^ 1.2X10^^

8.2 X 10^^ 4.3 X 10^^

362

Optoelectronic Devices: Ill-Nitrides Cathode GaN 0.1 urn

Active layer GaN n-type 10^ ^cm-^

Anode 0.1 M^m n-type lO^ôn"^

Bias

Figure 13.10. The schematics of the GaN NDR diode oscillator.

(^ 10^^ cm ^) than in GaAs designs (- 10 cm ). The latter is a very important result in terms of feasibility of GaN-based NDR diodes since the availability of low-doped GaN material (A^^ < 5 X 10^^ cm~^) is still limited. Higher doping of active layers in GaN NDR diodes also leads to reduction of TD^R in this material, helping to increase its frequency capability. A typical GaN NDR diode designed to operate at ~ 100 GHz had an n-type active layer with thickness L^ of 3 luim and doping A^^ of 1 X 10^^ cm~^. The active layer was sandwiched between anode and cathode layers and their corresponding ohmic contacts. Both contact layers were 0.1 |xm-thick and doped at 1 X 10^^ cm~^. The diameter of the diode D was selected to be 50 |xm. A final three-dimensional model of GaN NDR diode oscillator is shown in Figure 13.10 together with the bias supply and a parallel LCR circuit used to represent the resonant cavity.

13.4. GaN NDR DEVICE SIMULATION Based on the theoretical calculations for the design of NDR devices given by Eqs. (13.5)(13.7), a GaN NDR oscillator was designed. A schematic of the device design is presented in Figure 13.10 connected to an LCR resonant cavity. The radius of the diode is 10 [xm with a 3 |jLm long active layer. The active layer is n-type GaN, doped at 1 X 10^^ cm~^. The ohmic contacts to the diode are made using 0.1 |JLm thick, highly doped n^ GaN material with a doping level of 1 X 10^^ cm~^. Based on the material and design parameters, the device with an active layer length of 3 luim is expected to operate at 100 GHz and the device with an active layer of 5 juim is expected to work at 60 GHz. The values of the resistor (/?), inductor (L), and capacitor (C) values are chosen so that the natural frequency of the resonant cavity is at the operating frequency of the diode. The structure of Figure 13.10 was simulated using the device simulator MEDICL The models used for the simulations include the analytical mobility model which takes into account the concentration and temperature dependent empirical mobility, field dependent mobility model, energy balance model, electron/hole energy relaxation time, impact


363

ionization model and negative differential mobility model. Since MEDICI did not internally have the data for GaN at the time the simulations were performed, and further confidence regarding material parameters was felt to be necessary with in-house performed experiments, a material data file had to be created using the material parameters for Wz GaN selected from Ref. [4] and from various pubHcations [5-6] on GaN material studies including the field dependent mobility model based on the v-F characteristics calculated by Monte Carlo simulations by Kolnik et al. [4]. Fitting of device characteristics as reported in publications or developed in-house has helped in adjusting the material values to more realistic conditions. Comparisons of simulated performance with experimental characteristics of GaN-based MESFETs and PIN diodes were made to enable validation of the selected parameters. Further details on the adopted approach are presented elsewhere [10]. A low-field electron mobility of juin = 280 and 60 CTCI'N S were assumed for wurtzite (Wz) GaN doped at A^ = 5 X 10^^ and 1 X 10^^ cm~^ respectively [11]. The value of electron lifetime Tn = 7 ns and hole lifetime Tp = 0.1 ns used in the simulations was based on the experimental data measured by an electron-beam-induced current method [12]. Coefficients for calculating impact-ionization rates in GaN were obtained by fitting to the theoretical predictions presented in Ref. [13] and verified by comparing simulation results with experimental breakdown voltages reported for GaN PIN diodes [14]. Models for field dependence of electron mobility in GaN were based on the v-F characteristics calculated by Monte-Carlo simulations [4,15]. Velocity-field characteristics, evaluated in these studies, demonstrated a bulk NDR effect in the high-field region due to the intervalley transfer. However, the threshold field for intervalley transfer and consequent appearance of NDR in GaN was much larger than in conventional semiconductors such as GaAs. An increase of the threshold field is caused by a larger separation between the satellite and central valleys in Wz GaN where AE is —2.1 eV compared to AJE" ~ 0.3 eV for GaAs. The analytical expression (Eq. (13.11)) for v - F characteristics in GaAs [16] was used in the simulations

v(F) = fjiF

^ ^ ^^™/

(13.11)

UTH/ This was fitted to the results of Ref. [15] in order to obtain the v - F dependence of GaN, which manifests a higher peak velocity VPEAK (^ X 10^ versus 1.5 X 10^ cm/s), increased saturation velocity VSAT (2 X 10^ versus 0.6 X 10^ cm/s), and much larger threshold field F^H (150 versus 3.5 KV/cm) compared with GaAs. At the same time, the GaN low-field mobility of 280 cmÂ^ s and peak negative differential mobility /XNDR = max(-dv/dF) of 50 cmÂ^ s are lower than the GaAs values of 8000 and 2500 cm^fV s, respectively.

364


According to recent studies of GaN band structure, the T-valley inflection point, at which the group electron velocity is maximal, was found to be located below the lowest satellite valley in both Zb (zinc-blende) [17] and Wz GaN [18]. Although further studies are necessary to confirm this, it is possible that the reduction of electron drift velocity in the r valley caused by carriers becoming energetic enough to approach the inflection point can lead to bulk NDR. This contrasts other semiconductors, where intervalley transfer or impact ionization are initiated at a lower field than the inflection-point NDR [19]. Quantum devices employing superlattice structures to reduce the threshold field of inflection-based NDR mechanisms were first proposed by Esaki [20] and their feasibility has since been confirmed [21]. The reported v-F characteristics of Zb GaN calculated using Monte Carlo simulations were based on a band structure containing the /"-valley inflection point, and the results indicated that NDR was indeed caused primarily by the dispersion of the electron drift velocity in the Tvalley [17]. The inflection-based NDR manifested a threshold field Fjn of 80 KV/cm and peak velocity VPEAK of 3.8 X 10^ cm/s compared with FXH = 110 KV/cm and VpEAK = 2.7 X 10^ cm/s calculated in [4] for intervalley-transfer-based NDR. However, by far a more important consequence of the inflection-based NDR is the elimination of the intervalley-transfer relaxation time from the time required for NDR formation and, thus, a possibility of significantly increased frequency capability for GaN inflection-based NDR diodes. Frequency-independent v-F characteristics can be used to describe electron transport in the presence of time-varying electric field as long as the frequency of operation/ is much lower than the NDR relaxation frequency/NDR defined by expression (13.12) /

150kV/cm(r-A) 150kV/cm(r-M) 175kV/cm(r-M)

2

•t 1 i

0

1

1

0 1.

2

0

1 Tinne (ps) J

2 1

3

Time (ps) Figure 13.20. Full band Monte Carlo calculation of the temporal evolution of the electron velocity in the cdirection for four representative electric fields. Inset: comparison of calculated transient electron velocity for cdirection {F-A) and basal plane (F-M) transport [After [30]].

378

Optoelectronic Devices: III-Nitrides

underscored by the fact that multivalley analytical Monte Carlo studies of GaN predict velocity overshoot and NDR primarily due to intervalley scattering. Since the steady-state velocity for transport in the c-direction decreases at fields larger than 180 kV/cm, the NDR calculated for the 180-300 kV/cm range is primarily associated not with intervalley transfer, but with nonparabolicity in the F valley. These results may have a dramatic effect on the design of high-power, high-frequency electronics and avalanche photodiodes, NDR devices. Based on the previous discussion, lateral high-frequency electronic devices such as HFETs could benefit by operating with transport in the c-direction, for which both the transient and steady-state electron velocity are higher, while vertical devices such as avalanche photodiodes may improve with transport in the basal plane, for which intervalley transfer leading to impact ionization is more likely and the existence of a negative effective mass does not preclude diode breakdown. GaN NDR devices are for the above-mentioned reasons explored in both the c- and (2-direction since traditional NDR by intervalley transfer may be possible in the (2-direction while c-direction NDR operation may or may not be feasible by nonparabolicity effects depending on the energy of states available and the current associated with such a mechanism.

13.6. FABRICATION As it was previously discussed, theoretical and experimental studies show that the dislocations in the film significantly limit the lateral mobility in GaN films, whereas the vertical mobility was hardly affected. Based on this consideration, NDR devices should be designed and fabricated as vertical structures. However, other considerations related to basal plane transport features may dictate a lateral GaN NDR diode design. A schematic of a vertical NDR structure is shown in Figure 13.21. As shown in Figure 13.21, the device structure is quite simple consisting of a thin layer of highly doped n^ GaN on top for making the anode contact via an airbridge, followed by

air-bridge

Buffer layer

Substrate (Si, SiC, or Sapphire) Figure 13.21.

A schematic of a vertical NDR structure.


379

a thick layer of n~ GaN beneath for the active layer and beneath this is another layer of n^ GaN for making the cathode contact. The GaN material is grown on a substrate of silicon, silicon-carbide or sapphire with a buffer layer technology to lattice match the GaN material to the substrate. The general technology used to fabricate the above structure mainly involves anisotropic etching of the material using a series of masks and can be accomplished in about 5-6 masks, which is relatively inexpensive compared to other device fabrication. Etching of GaN material can be done using a dry etching technique such as reactive ion etching (RIE) or inductively coupled plasma etching (ICP) or a wet etching technique such as photo-electrochemical etching (PEC). The best etching method, in terms of etch rate, selectivity and post etch surface quality (which is a determining factor for the contact resistance), is still being investigated since the processing technology for GaN is still under development. There are two kinds of technologies that can be used to fabricate vertical NDR devices, the on-wafer and stand-alone technologies, which are selected based on the type of testing and application that follows. A schematic of the on-wafer technology is shown in Fi gure 13.22. The basic fabrication steps for the on-wafer technology starts with the isolation etch in order to electrically separate each device on the wafer. The next step is the mesa etch which defines the placement of the ohmic contacts. Of course, following the mesa etch is the actual deposition of the metal for the anode and cathode ohmic

NDR Diode Technology Top n+ layer

Substrate

Active ri-layer ^ Isolation etch: Discrete device| Bottom separation ' ri+ layer Anode metal

> Mesa etch: Diodes structure definition Cathode metal

^ Metal Deposition: Ohmic contacts

^ Au-Plating: Air Bridges

On-wafer NDR diode

Figure 13.22. On-wafer technology for GaN NDR device.

380

Optoelectronic Devices: Ill-Nitrides Beam leads Anode metaj Cathode metal

Active region Figure 13.23. Stand-alone technology for GaN NDR device.

contacts. Finally the transmission lines, i.e. coplanar waveguide lines and airbridges are defined and electroplated for on-wafer, high-frequency testing. For mounting the devices on an integrated heat sink and testing in a microwave cavity, it is necessary to use the stand-alone technology. Like the on-wafer technology, the standalone technology also requires the isolation and mesa etch as the first two steps. Next it is necessary to create a via hole from the back side and electroplate the anode and cathode contacts. Finally, the substrate is completely removed and individual devices are released. The stand-alone technology is shown by a schematic in Figure 13.23. In the solid-state electronics laboratory at the University of Michigan, GaN NDR devices have been fabricated using the on-wafer technology. The fabricated devices have an active layer, which is 30 |jLm in diameter and has a thickness of 4 jjim. An SEM picture of such a device is shown in Figure 13.24. A device with such a large diameter and relatively thick active layer, as in Figure 13.24, is capable of high current values leading to improved power performance and also facilitates high-frequency testing. On the other hand, such large devices impose several

-

"'^'-^^^^^r:::,-

Au-plated air-bridge

Isolation

Diode mesa

Air-bridge pillar Anode metal

Cathode pad i-plated cathode metal Figure 13.24. SEM photo of a GaN NDR device fabricated at the University of Michigan using the on-wafer technology. The device has an active layer with a diameter of 30 [xm and thickness of 4 iJim.

381

GaN Negative Differential Resistance Components Isolation Au-plated air-bridge

Air-bridge pillar Au-plated cathode metal Figure 13.25.

SEM photo of a GaN NDR device fabricated at the University of Michigan. The device has an active layer with a diameter of 10 juim and thickness of 0.5 |JLm.

difficulties. The large amount of current flowing in the device, combined with the thick active layer and low thermal conductivity of GaN, all contribute to self-heating effects, which severely limit the operation of the devices. To reduce thermal effects, smaller geometry diodes were designed and fabricated. The diameter was reduced from 30-50 to 8-20 fxm and the thickness of the active region was reduced from 4 - 6 to 0.5 ixm. An SEM photo of such a device is shown in Figure 13.25. | Lm In the ideal case, the expected power dissipation would be 25 times smaller in a 10 U diameter device compared to a device with 50 jxm diameter. The experimentally observed power dissipation for devices with active layer thickness of 0.5 ixm and doping of 1 X lO^^cm""^ is about 4 - 5 W for devices with a diameter of 50 iJim ( 7 = 15 V, / = 330 mA) and about 1-2 W for devices with 10 iJim diameter (V = 15 V, / = 130 mA). Besides reducing the size of the device it is also necessary to package the device in an integrated heat sink in order to overcome self-heating effects. For on-wafer technology the integrated heat sink becomes the substrate.

Device'

Heat flow

Heat sink

Heat sink with high thermal conductance Figure 13.26.

Device Heat sink

^

Heat flow

h Heat sink with low thermal conductance

Schematic showing the heat flow for a device mounted on a good heat sink and one mounted on a poor heat sink.

382

Optoelectronic Devices: Ill-Nitrides Table 13.6. Thermal conductivity of possible substrate for GaN Substrate

Thermal conductivity (W/cm K)

Sapphire Silicon Silicon carbide

0.32 1.48 3

Figure 13.26 shows a schematic of heat flow from a device mounted on a good heat sink and one mounted on a poor heat sink while Table 13.6 compares the thermal conductivity of possible substrates for GaN. It is obvious that the stand-alone technology is expected to offer definite advantages. Thermal simulations have been performed based on various designs and technology approaches and are currently implemented in device demonstrations. It is obvious from Table 13.6 that the best substrate for thermal dissipation is silicon carbide. However, since SiC is quite expensive, silicon substrates are a good alternative. Moreover, silicon substrates can be easily thinned down to about 100 |JLm by using a lapping tool to further improve heat dissipation. Substrates of this type have already been employed for GaN NDR devices at the author's laboratory and further work is in progress. For stand-alone devices, heat dissipation can be facilitated by the combination of very thin electroplating of the anode/cathode contacts and then mounting these devices on diamond heat sinks.

13.7. TESTING The general testing set-up involves a high-power pulse generator to apply the DC bias to the device in the NDR region of the current-voltage characteristics, which can be

Q

QE) n

spectrum analyzer |j

1 n

\i ,

1

Out

Pulse Gen erator ^

Current Drobe

r1llh

r7r\ I \'

CaVI (y

, . , - •

Biasing terminal

Output terminal

Figure 13.27. Block diagram of the testing set-up used for a device mounted in a microwave cavity.


383

ouu -

250-

^•••••••%>

_ 200-

300 K) [5-8]. Second, it would be a major advantage if there were already an existing technology base for the material in other applications. Most of the works in the past have focused on (Ga,Mn)As and (In,Mn)As. There are indeed major markets for their host materials in infrared light-emitting diodes (LEDs) and lasers and high-speed digital electronics (GaAs) and magnetic sensors (InAs). In samples carefully grown single-phase by Molecular Beam Epitaxy (MBE), the highest Curie temperatures reported are -- 110 K for (Ga,Mn)As and ~ 35 K for (In,Mn)As as one of the most effective methods for investigating spin-polarized transport is by monitoring the polarized electroluminescence output from a quantum-well (QW) LED into which the spin current is injected. Quantum selection rules relating the initial carrier spin polarization and the subsequent polarized optical output can provide a quantitative measure of the injection efficiency. There are a number of essential requirements for achieving practical spintronic devices in addition to the efficient electrical injection of spin-polarized carriers. These include the ability to transport the carriers with high transmission efficiency within the host semiconductor or conducting oxide, the ability to detect or collect the spin-polarized carriers and to be able to control the transport through external means such as biasing of a gate contact on a transistor structure. We focus on a particular and emerging aspect of spintronics, namely, recent developments in achieving practical magnetic ordering temperatures in technologically useful semiconductors [5-10]. While the progress in synthesizing and controlling the magnetic properties of Ill-arsenide semiconductors has been astounding, the reported Curie temperatures are too low to have significant practical impact. Other materials for which room temperature ferromagnetism has been reported include (Cd,Mn)GeP2 [6], (Zn,Mn)GeP2 [7], ZnSnAs2 [8], (Zn,Co)0 [9] and (Co,Ti)02 [10]. Some of these chalcopyrites and wide bandgap oxides have interesting optical

Ferromagnetism in GaN and Related Materials

389

properties, but they lack a technology and experience base as large as that of most semiconductors. The key breakthrough that focused attention on wide bandgap semiconductors as being the most promising for achieving practical ordering temperatures was the theoretical work of Dietl et al. [11]. They predicted that cubic GaN doped with ~ 5 at.% of Mn and containing a high concentration of holes (3.5 X 10^^ cm~^) should exhibit a Curie temperature exceeding room temperature. In the period following the appearance of this work, there has been tremendous progress on both the realization of high-quality (Ga,Mn)N epitaxial layers and on the theory of ferromagnetism in these so-called dilute magnetic semiconductors (DMS). The term DMS refers to the fact that some fraction of the atoms in a non-magnetic semiconductor like GaN is replaced by magnetic ions. A key, unanswered question is whether the resulting material is indeed an alloy of (Ga,Mn)N or whether it remains as GaN with clusters, precipitates or second phases that are responsible for the observed magnetic properties [12]. Table 14.1 summarized recent work in realizing room temperature ferromagnetism.

Table 14.1. Compilation of semiconductors showing room temperature ferromagnetism Material

Bandgap of host (eV)

Cdi-;cMn;,GeP2 (Ga,Mn)N (Ga,Mn)N (Ga,Mn)N (Ga,Cr)N

1.72 3.4 3.4 3.4 3.4

(ZnO):Co

3.1-3.6

(Al,Cr)N

6.2

(Ga,Mn)P:C

2.2

(Zni_,Mn,)GeP2

1.83-2.8

(ZnMn)GeP2 ZnSnAs2 ZnSiGeN2 SiC

300 K. In material known to have second phases, the magnetization versus temperature behavior shows either a spin-glass type transition or cusps corresponding to the presence of multiple phases. While a number of groups have reported room temperature ferromagnetism in GaN doped with Mn or Cr [31,32,34,35], there has been Uttle investigation of the electrical and optical properties of the material. The carrier-induced ferromagnetism model requires hole-induced interactions between the spins of the substitutional transition metal ions. However, the few reports of the position of Mn in the GaN bandgap find it to be very deep, ^v + 1-4 eV, where it would be an ineffective acceptor dopant [45-47]. In addition, most of the data reported for GaMnN indicate it is either insulating or n-type. The resistivity of GaMnN grown by MBE has been examined using both Schottky diode and transmission line method (TLM) measurements and optical absorption spectra from GaMnN films as a function of Mn concentration. Features at £'c — 1.9 eV are found in the absorption spectra, corresponding to transitions from Mn to the conduction band. However, this state does not control the Fermi level position in the GaMnN and the material remains high-resistivity n-type with a thermal activation energy of ~ 0.1 eV. The results have interesting indications for both the existing theoretical models for DMSs and for the potential technological uses of GaMnN. Figure 14.7 (top) shows the low bias / - V characteristics at 25°C from the GaMnN/GaN device structure. The results are consistent with the GaMnN having a high resistance and the characteristic is approximately linear over this voltage range. The resistivity of this layer can be obtained from the relation p = RA/W where R is the total measured resistance of the GaMnN layer of thickness W between the top contact of area A and the ohmic contacts. For W = 0.3 luum, this translates to p= 3.1 X 10^ H cm. Even at higher biases, the I-V characteristic is dominated by the high resistance of the GaMnN (Figure 14.7, bottom). From TLM ohmic metal patterns placed directly on the GaMnN layer, we were able to measure the temperature dependence of the sheet resistivity. Figure 14.8 shows an Arrhenius plot of this data, corrected for the temperature dependence of the mobility contribution to the sheet resistivity. Over a broad range of temperature, the sheet resistivity varies as Ps = Pso Qxp(-EJkT)


399

6.0x10-^ Schottky contact 80 |jm Current-Voltage

4.0x10"^ 2.0x10-^ h 0.0

-2.0x10-^

Resistivity=3.7x10^ Qcm

-4.0x10-^ -6.0x10-^

-10

0 Bias (V)

10

3.0x10-5 2.5x10-5 2.0x10-5 .-v

=^"'^^ where V is the applied voltage, R^ the series resistance, e is the electronic charge and n is the ideality factor. The extracted values in terms of saturation current density /§ were 4.28 X 10"^ A/cm^ n=l.2l at 298 K, 6.60 X 10"^ A/cm^ n = 1.17 at 323 K, 1.08 X 10"^ A/cm^ n = 1.16 at 348 K, and 8.42 X 10"^ A/cm^, at 373 K. In the latter case, we could not determine an accurate value of n. The saturation current can also be represented as T exp

h=AA' 10-

10-

I

^

'

I

'

I

'

\

r

kT

) ,-o-o,-o-o-q

y

o

Figure 14.18. Forward I-V-T

^

/^

100 |im diameter

characteristics from 100 jxin diameter Pt contacts on n-(Ga,Mn)N.


411

where A** is the Richardson's constant for (Ga,Mn)N, A the diode area and (/)B is the Schottky barrier height. The theoretical value of A** for n-GaN is 24 A/cm^ ¥? [15], but is not known for (Ga,Mn)N because the electron effective mass has not been determined. From the / - V - T characteristics, we extracted a value for A** of 91.2 AJcvc? K^, but there is a large uncertainty (~ ± 60%) in this due to narrow range of measuremental temperatures from which we had to extrapolate to obtain the estimated Richardson's constant. For comparison, a similar analysis for Pt/n-GaN Schottky diodes reported values of 64.7-73.2 for A**. From the measured I-V characteristics at each temperature we were able to extract the (/)B values, as shown in Figure 14.19. The barrier height at 25°C was 0.82 ± 0.04 eV, which compares to a value of 1.08 eV for Pt on n-GaN determined from I-V characteristics [14,16]. We have not yet determined the bandgap of (Ga,Mn)N but preliminary data suggest it is smaller than GaN and this is consistent with the smaller barrier height for Pt on (Ga,Mn)N. It should also be noted that in the early stage of developing a new materials system it is common to have a wide range of values reported for barrier heights due to the presence of surface defects, interfacial layers and material inhomogeneities. The barrier height was also extracted from the activation energy plots in the measured I-V-T data and the saturation current at each temperature. Figure 14.20 shows the plot of ln(/s/Ar^) versus inverse temperature, yielding a barrier height of 0.91 ± 0.06 eV which is consistent with the value at 25°C derived from the forward I-V characteristics. 1

u.»u

0.85

'

1

1

'

-\

_ •

CD CQ

1

r 11

0.75

'

'

^

1

0.80

CD

1

~~

— - 1'

11

-

-

0.70 Pt/n-(Ga, Mn)N n=3.5x10^^cm-3

1

0.65

1

300

,

1

1

320

340

,

1

,

360

1

380

Temperature (K) Figure 14.19. Barrier height extracted from forward I-V characteristics at different temperatures.

Optoelectronic Devices: Ill-Nitride

412

Figure 14.21 shows the reverse I-V characteristics from the diodes as a function of temperature. The reverse leakage depends on both bias and temperature. From a moderately doped sample of the type studied here, we would expect thermionic emission to be the dominant leakage current mechanism. According to image-force barrier height lowering, this leakage current density, / L can be written as -/s exp

m

where A(/)B is the image-force barrier height lowering, given by eE^

. 1/2

V 4178^

where E^, is the electric field strength at the metal/semiconductor interface and ss is the permittivity. The experimental dependence of 7L ^^ bias and temperature is stronger than predicted from this last equation. The large bandgap of (Ga,Mn)N makes the intrinsic carrier concentration in a depletion region very small, suggesting that contributions to the reverse leakage from generation in the depletion region are small. Therefore, the additional leakage must originate in contributions from other mechanisms such as thermionic field emission or surface leakage.

-20

Pt/n-(Ga, Mn)N 0.91+-0.06eV

-30 0.0026

JL

0.0028

0.0030

0.0032

0.0034

1/T(K-^) Figure 14.20. Arrhenius plot of IsÂT^ used to extract barrier height of Pt on n-(Ga,Mn)N.


413

0.00

^q(j(jijtjmmMummMm -0.02

-0.04

o

-0.06

-0.08

-5

-4

-3

z -2

jP

100 |am diameter 75°C 50°C 25°C -1

0

Bias (V) Figure 14.21. Reverse I-V-T

characteristics from 100 ixm diameter Pt contacts on n-(Ga,Mn)N.

In summary Pt contacts on n-(Ga,Mn)N show rectifying behavior with a barrier height at 25°C of 0.8-0.9 eV, depending on which analysis method is employed. This is a useful first step in making gated spin FET structures based on this DMS.

14.5. AIN AIN plays an important role in many areas of solid-state devices [52-67], including thin film phosphors, nitride-based MIS heterostructure transistors, thin-film gas sensors, acoustic wave resonators, UV LEDs, distributed Bragg reflectors, heat spreaders and heterojunction diodes. AIN may also be promising in the emerging field of spintronics, due to its predicted high Curie temperature {TQ) when doped with particular transition metals. Room temperature ferromagnetism has been reported for Cr-doped AIN thin films deposited by reactive sputtering [52] or MBE [53]. Ion implantation provides a versatile and convenient method for introducing transition metals into semiconductors for examination of their effects on the structural and magnetic properties of the resulting material [68]. AIN is an ideal host in this regard, since Kucheyev et al. [69] reported that single crystal epilayers of AIN grown on sapphire substrates did not become amorphous even at LN2 temperatures for high doses of keV heavy ions such as Au. In addition, very high quality AIN on sapphire has recently been reported by several groups [70,71], providing well-characterized material in which to examine the properties of transition

414


metals. The fabrication of ferromagnetic AIN would create a wider range of possible all semiconductor spin-dependent devices. For example, ferromagnetic AlMnN could be used as a magnetic barrier in a tunnel junction where it would serve as a spin filter. The predicted Curie temperature of AlMnN is greater than 300 K and recently a Curie temperature of more than 340 K has been observed for AlN:Cr [52,53]. Growth of the films was carried out by gas-source MBE. Solid A1(7N) and Mn(7N) sources were heated in standard effusion cells. Gaseous nitrogen was supplied by an Oxford RF plasma head. All films were grown on (0001) oriented sapphire substrates, indium mounted to Mo blocks. AlMnN and AIN films were grown at a temperature of 780°C, as indicated by the substrate heater thermocouple. Sapphire substrates were first nitridated for 30min at a substrate temperature of 1000°C under 1.1 SCCM nitrogen (chamber pressure = 1.9 X 10~^ Torr). Nucleation at 575°C for 10 min and a 30 min buffer layer at 950T followed nitridation, both under 1.1 SCCM nitrogen. Both AIN and AlMnN films were grown with a substrate temperature of 780°C and an Al effusion cell temperature of 1150°C. The Mn cell temperature was varied from 635 to 658°C. The growth rate of the AIN was 0.2 |xm/h and the growth rate of the AlMnN films was 0.16 ixm/h. In situ reflection high energy electron diffraction (RHEED) was used to monitor films during growth. AIN demonstrated 2D growth and AlMnN films demonstrated 2D/3D growth. AlMnN grown with an Mn cell temperature of 635°C was found to be single phase. The AlMnN with Ty^^ = 658°C formed AlMn as detected by powder XRD. A Mn cell temperature of 650°C was found to be the upper limit of single-phase AlMnN under previously mentioned growth conditions. For comparison, a layer of Mn4N was also grown on sapphire. The lattice constant was found to decrease as the Mn cell temperature increased for single-phase material. A similar pattern was observed for single-phase GaMnN films grown in the same system under different conditions. GaN implanted with Mn has been reported to exhibit substitutional or near substitutional incorporation. It is expected that the incorporation of interstitial Mn should either increase or have no effect on the lattice constant. The observation of a decrease in the lattice constant of the AlMnN films suggests that the Mn occupies a substitutional site. This is further confirmed by Hall analysis, which showed pure AIN to be highly resistive as expected and material containing an AlMn second phase to be highly conductive n-type. In contrast, single-phase AlMnN was found to be p-type. If Mn incorporates substitutionally, one would expect by analogy with its behavior in other III-V materials that it would behave as a deep acceptor. The observation of p-type behavior fits this explanation. Magnetic remanence and coercivity indicating hysteresis was observed in ternary AlMnN films at 10,100, and 300 K. Measurements over 300 K were not possible due to limitations in the magnetometer. Saturation magnetization was found to decrease at 300 K compared to 100 K for AlMnN grown at Tîn = 650°C. The values of temperature dependent saturation magnetization of AlMnN are shown in Figure 14.22. This data indicates an approaching TQ

Ferromagnetism in GaN and Related Materials 1.0x10"^ n

1

'

r

n

'

1

'

1

'

1

415 '

r

8.0x10-^H

6.0x10-^H CD N

OJ C CD

4.0x10-® -AIMnN, TMn = 650

c

o (D

^

2.0x10^

"m C O

0.050

100

150

200

"T"

T"

250

300

Temperature (K) Figure 14.22. Saturation magnetization versus temperature for single-phase AlMnN grown at 650°C. Saturation magnetization was extracted from SQUID hysteresis loops.

for this material. However, hysteresis persisted to 300 K, demonstrating soft ferromagnetism at room temperature. Figure 14.23 depicts the field dependence of magnetization for AlMnN and AINfilms.Pure AIN grown under the same conditions as AlMnN demonstrated paramagnetic behavior. This indicates that ferromagnetism arises with the addition of Mn. The diamagnetic background due to the sapphire substrate was subtracted from the raw data and the subsequent corrected data was used for analysis. The magnetization was not normalized to the Mn concentration due to the difficulty in calculation of the precise amount of Mn in the films. Perpendicular measurements were found to have lower values for saturation magnetization in AlMnN films. The hysteresis observed in the AlMnN is believed to arise from the inclusion of Mn into the AIN lattice. Clusters of second phases, undetectable by methods mentioned above, are not thought to be the cause of ferromagnetism observed at 300 K. This is supported by magnetic analysis of material containing the most likely cluster phases, AlMn and Mn4N. Magnetization as a function of temperature for AIN, single-phase AlMnN, AlMnN with an AlMn phase present, and Mn4N show substantially different behavior, as shown in Figure 14.24. The reason for the low T paramagnetic behavior seen in AIN and AlMnN films is still unknown. The M versus T of Mn4N clearly indicates ferromagnetic behavior, and the formation of clusters has been proposed as the cause of hysteresis in some ferromagnetic III-V materials. However, the formation of Mn4N clusters does not

416

Optoelectronic Devices: Ill-Nitride 2.0x10-5

1

1.5x10-5

1

•

I

-

I

'

1

'

1

AIMnN,TMn=650

TT

Il

1.0x10-5

E

5.0x10-®

0

c o

-

!

1 0.0

r

CD N

•"••

1

-5.0x10-® 1

^

-1.0x10-5

ji

-1.5x10-5 •

-2.0x10-5

J

-1000

-500

0 H(Oe)

500

.

1

1000

Figure 14.23. Magnetization versus applied field for single-phase AlMnN.

influence the magnetization above 250 K, since clearly the magnetization drops to zero at that temperature. Also, the magnetization versus temperature indicates that the formation of AlMn clusters is not the cause of the ferromagnetism observed, evidenced by the order of magnitude difference between the values of magnetization over 150 K. Hence, the incorporation of Mn into the AIN lattice forming the ferromagnetic ternary AlMnN is most likely the reason for the observed hysteresis. In conclusion, room temperature ferromagnetism has been observed in AlMnN grown by gas-source MBE. The lattice constant decreased with increasing Mn cell temperature for single-phase material, indicating constant site occupation, probably substitutional. Hysteresis in M versus H at room temperature was observed in single-phase material and the magnetization as a function of temperature suggests ferromagnetism caused by AlMnN, not clusters. Ferromagnetism has also been observed in transition metal-implanted films. Implantation of Cr~^, Co"^ or Mn"^ ions was carried out at an energy of 250 keV (corresponding to a projected range of ~ 1500 A in each case) and a fixed dose of 3 X 10^^ cm~^. As a rough guide, the peak transition metal concentrations, located at the projected range, are ~ 3 at.% in the AIN. After implantation, the samples were annealed at 950°C, 2 min under flowing N2. PL measurements were carried out with a quadrupled Ti: sapphire laser as an excitation source together with a streak camera, providing an excitation power of ~ 3 mW at 196 nm [72].


— ' — 1 — ' — I — ' — I — ' — I —

1.5x10"

1.0x10-

i ' 5.0x10-^1

All

—•— single phase AIMnN ~ê- AIMnN/AIMn -B-AIN - Mn,N

— AIMnN »-AIMnN/AIMn »-AIN

v;v-vvA \/X/\ A \ A

• ^ } A ^

0.0 50

100

150

200

Temperature (K)

250

/ 300

50

100

150

200

250

300

Temperature (K)

Figure 14.24. Magnetization versus temperature for AIN, AIMnN, Mn4N and AIMnN/AlMn. The magnetic signal is determined by subtracting the zero field-cooled trace from the field-cooled curve.

PL spectra taken at 10 K of the AIN implanted with Cr, Mn or Co after annealing at 950°C for 2 min looked basically identical in each case, even without annealing. The unimplanted AIN showed strong band-edge emission at ~ 6.05 eV and two broad emission bands related with deep level impurities at —3.0 and 4.40 eV each of which had peak intensity of ~ 1% of the band-edge emission intensity. The Cr-, Mn- and Co-implanted AIN showed an absence of band-edge emission, which suggests that the point defect recombination centers created during implantation are stable against annealing at 950°C. Well-defined hysteresis was present in the Co-implanted AIN, with a coercive field of ~ 160 Oe at 300 K and 230 Oe at 10 K. The diamagnetic contributions from the substrate have been subtracted out of the data. At 300 K, the saturation moment, MQ = glJi^S, where g is the degeneracy factor, [x^, the Bohr magnetron and 5, the total number of spins, was calculated to be ~ 0.65/xg for Cr. This value is lower than the theoretical value of 3/>IB expected for a half-filled d-band of Cr, if all of the Cr ions were participating in the ferromagnetic signal. Disorder effects due to implantation-induced change may contribute to creating a distribution of exchange couplings that favor antiferromagnetism and reduce the effective magnetism. Similar data is shown in Figure 14.25 for the Co-implanted AIN. Once again there is hysteresis present at 300 K, with a coercive field of ~ 1750 Oe at 300 K and 240 Oe at 100 K and a calculated saturation moment of 0.52/XB for Co. The FC and ZFC magnetization versus temperature are shown at the bottom of the figure. In this case the differences extend to —100 K. Figure 14.26 (top) shows magnetization versus field at 100 K for Mn-implanted AIN. This was the highest temperature for which clear hysteresis could be obtained. The coercive field was —220 Oe at both 100 and 10 K. The FC and ZFC phases are almost coincident at an applied field of 500 Oe, as shown at the bottom of the figure and consistent

418


with lower overall magnitude of the magnetization. The calculated saturation moment was for Mn at 100 K, compared to the theoretical value of four. The main 6-26 XRD peaks of the Cr or Mn-implanted samples after 950°C annealing correspond to the expected A1N(0002) and (0004) lines and AI2O3 (0002),(0006) and (0012) substrate peaks and the broad peak at 2^ = 20° is due to short-range disorder from the implantation process and was not observed on the as-grown films. No peaks due to the half-metallic ferromagnetic Cr02 phase were detected in the Cr-implanted sample and other potential second phases which could form, such as Cr, CrN [73-75], Cr2N and Al^^-Cr^;) were not detected and in any case are not ferromagnetic at the temperatures used in these experiments. Similarly, in the case of Co implantation, metaUic Co has a Curie temperature of 1382 K and CO;,N phases are all Pauli ferromagnetic. Finally, for Mn implantation, metallic Mn is antiferromagnetic while Mn;cN is ferromagnetic with a Curie O.IT^LB

AIN:Co 3%

1.0x10-^

1^ 5.0x10-6-1

..!!'S .\\

300 K

ii

0.0 c -5.0x10-®-^ -1.0x10-5-1

iiU''^

-1000

1000 H (Gauss)

-1.0x10-5 h AIN:Co 3% 500 Oe

-1.1x10-5 i? E (U

-1.2x101-5 ^T J^^^-^^.^-^^ -1.3x10-5 50

100

150

200

250^ 300

Temperature (K) Figure 14.25. Magnetization (300 K) as a function of field (top) and FC and ZFC magnetization as a function of temperature (bottom) for AIN implanted with 3 X 10^^ cm"^ Co^ and annealed at 950°C, 2 min.


419

temperature of 745 K. Thus, secondary ferromagnetic phases are not responsible for the observed magnetic properties. The origin of the observed ferromagnetism is not likely to be carrier-mediated due to the insulating nature of the AIN. Wu et al. [53] suggested that substitutional Al^^Cri-^cN random alloys would have Curie temperatures over 600 K, as estimated from a multicomponent mean-field theory in which the ferromagnetism occurs in a midgap defect band. Another possible mechanism for the observed magnetic properties is that the Mn is not randomly distributed on Al sites but is present as atomic scale clusters. Some mean field theories suggest that Mn clustering can significantly influence TQ as a result of the localization of spin polarized holes near regions of higher Mn concentration. There is also some support for this assertion from local spin density approximation calculations, which predict that it is energetically favorable for the formation of magnetic ion dimers 6x10-6 4x10-6

f\\H Mn 3% implanted 100 K

0

3

E

2x10-6

c

o

0

N

c

D)

^ -2x10-6 J

CD

-4x10-6-1

if

111.::-

-6x10-6 -1000

-500

500

1000

H (Gauss)

-1.5x10-^ IAIN Mn 3% implanted -1.6x10-^

500 Oe

E S

-1.7x10-^

-1.8x10-^ .

-1.9x10-5

•

'

50

100

•

'

•

150

200

I

I

250

L I

i_

300

Tennperature (K)

Figure 14.26. Magnetization (100 K) as a function of field (top) and FC and ZFC magnetization as a function of temperature (bottom) for AIN implanted with 3 X 10^^ cm"^ Mn+ and annealed at 950°C, 2 min.

420


and trimers at second nearest-neighbor sites which are ferromagnetic. The percolation network-Hke model for ferromagnetism in low carrier concentration systems suggested by several groups is another potential mechanism. High doses (3 X 10^^ cm~^) of ion-implanted Co"^, Cr"^ or Mn"^ ions into AIN epilayers on AI2O3 substrates severely degrades the band-edge luminescence, which is not recovered by annealing up to 950°C. In each case the implanted AIN shows ferromagnetic ordering as evidenced by the presence of hysteresis in M versus H loops. The hysteresis persists up to > 3 0 0 K in the case of Cr^ or Co"^ implantation and 100 K for Mn"^ implantation. Less than ~ 20% of the implanted ions contribute to the magnetization, but this might be increased by the use of much higher annealing temperatures. Simple twoterminal resistivity measurements show that the implanted AIN remains insulating (> 10^ n cm) and thus conventional carrier-mediated ferromagnetism is not a likely mechanism for the observed magnetic properties. Implantation provides a versatile method of introducing different transition metal dopants into AIN for examination of their effect on the structural and magnetic properties.

14.6. AlGaN There is also an interest in the use of transition metal-doped AlGaN for possible applications in spintronic devices such as polarized light emitter or spin transistors. The latter exploits quantum interference effects provided electrons with a particular spin can be injected into the channel of the device and a gate bias can be applied to cause splitting of spin-up and spin-down states. A key requirement for spin-based semiconductor devices is the achievement of ferromagnetism, preferably above room temperature. The properties of implanted transition metals in AlGaN is of particular relevance for realization of polarized light emitters or spin transistors since it could serve as the cladding layer in the former and the wide bandgap part of the heterostructure in the latter. 0-26 XRD scans from the n-AlGaN before and after Mn"^, Co"^ or Cr"^ implantation and annealing at 1000°C did not show any observable differences. The highest intensity peaks in all spectra correspond to the expected AlGaN (0 0 0 2) and (0 0 0 4) lines and AI2O3 (0 0 0 2), (0 0 0 6) and (0 0 0 12) substrate peaks. We did not observe any peaks due to second phases that could exhibit ferromagnetism. For example, in the Mn-implanted material, Mn^^N is ferromagnetic with a Curie temperature of 745 K and GaMn is also ferromagnetic with a Curie temperature near 300 K (metallic Mn is antiferromagnetic). In the Co"^-implanted AlGaN, metallic Co has a Curie temperature of 1382 K and Co^^N phases are all Pauli ferromagnetic. Finally, in the Cr'^-implanted AlGaN, CrO is a halfmetallic, while Cr, CrN, Cr2N, Al;,Cr_y and Ga^^Cr^ are not ferromagnetic [36]. However, in such thin layers, it could be possible for small quantities of second phases to be present and remain undetectable by XRD.


All

Well-defined hysteresis at 300 K was observed for the Co-implanted Alo.38Grao.62N, as shown at the top of Figure 14.27. The coercive field was ~ 85 Oe at 300 K and ~ 75 Oe at 10 K. The saturation magnetization was ~ 0.4 emu/cm^ or ~ 0.76^lB calculated saturation moment. This is slightly higher than the value reported for Co~^ implantation into pure AIN under similar conditions (0.52/XB). which is consistent with the higher vacancy concentrations expected to be created in AlGaN due to its lower bond strength. 0.6

1

0.4 I1

£ 0)

'

•

1

T

'

1

'

1

i

n-AIGaN:Co

T = 300K

0.2 I-

'

1

1 1

0.0

-: • • -0.2 CD

-0.4

-0.6

I

.

I

.

-500

-1000

-0.34 -0.35 -0.36 -0.37 -0.38-1 -0.39 H -0.40 -0.41 -0.42 -0.43 H -0.44 -0.45 -0.46 -0.47

1

.

0 H(Oe)

1

•

500

1 1

1000

- Zero field cooled - Field cooled H = 250Oe n-AIGaN:Co

^

50

•|i^-o'5. ^ ^ ^

100 150 200 Temperature (K)

250

300

Figure 14.27. Magnetization (300 K) as a function of field (top) and field-cooled (FC) and zero field-cooled (ZFC) magnetization versus temperature for AlGaN implanted with 3 X 10^^ Co^ annealed at 1000°C for 2 min.

422


The bottom part of Figure 14.27 shows the temperature dependence of FC and ZFC magnetization for the Co'^-implanted AlGaN. The fact that these have different values out to ~ 230 K is a further indication of the presence of ferromagnetism in the material. In both epitaxial and ion-implanted transition metal-doped semiconductors, we have found the general result that the hysteresis can be detected to higher temperatures than the difference in FC and ZFC magnetization. As mentioned earlier, the samples exhibited low carrier densities ( < 3 X lO^^cm"^ from Hall measurements) after implantation and annealing, and therefore, carrier-mediated ferromagnetism by free electrons is not expected to be operative. In addition, the Co ionization level is expected to be deep in the AlGaN bandgap, so that there will be no significant contribution to the carrier density from the substitutional fraction of these atoms. More recent percolation network models for ferromagnetism in DMSs suggest that localized carriers may mediate the interaction between magnetic ions in low carrier density systems. The Mn-implanted p-type AlGaN also showed a well-defined hysteresis loop at 300 K, with a coercivity of ~ 6 0 O e (Figure 14.28, top). The saturation moment, MQ = SIHBS where g is the degeneracy factor, /xg the Bohr magneton and S the total number of spins was calculated to be ~ 0.57/XB- The theoretical value would be four if all of the implanted Mn was participating towards the ferromagnetism, so the lower experimental value indicates that only a fraction of the Mn is substitutional and magnetically active. The saturation moment for AlGaN is significantly larger than the value of 0.17/^3 reported for Mn implantation into pure AIN. The temperature dependence of FC and ZFC magnetization is shown at the bottom of Figure 14.28. The ferromagnetism is very weak above ~ 125 K, but is detectable through the hysteresis. By sharp contrast to the case of Mn implanted into p-AlGaN, when we performed the same implants into n-AlGaN, the resulting differences in FC and ZFC magnetization were very weak and hysteresis loops even at 10 K did not show clear evidence of ferromagnetism. The differences from the p-type material may result from the higher AIN mole fraction in the n-type AlGaN, which makes it harder for the implanted ions to become substitutional upon annealing. An alternative explanation is that holes are more efficient at ferromagnetic coupling between the Mn spins than are electrons. This has been reported previously for both n- and p-type GaAs and GaP doped with Mn. We also did not observe any clear evidence for ferromagnetism in the Cr-implanted n-AlGaN. This is a clear difference from the case of Cr-implanted AIN, where hysteresis was reported at 300 K. In conventional DMSs such as (Ga,Mn)As, the magnetization is given by [11] M = ^,^,SNoX,,Bs[

k^^T + T^,)

J

in the mean-field approach, where S is the localized spin, NQ is the concentration of cation sites, Xgff is the effective spin concentration, B^ is the Brillouin function and FQ(M) is the hole contribution to the free-energy functional F (which depends on the magnetization of


423

0.4 0.3 p-AIGaN:Mn 0.2

T = 300 K

0.1

^hii

0.0

• • -0.1 -0.2

iiii*'

-0.3 -0.4

-1000

-500

0

500

1000

H(Oe)

E o 0 c

-0.085 4 -0.090 4 -0.095 4 -0.100 4 -0.105 4 -0.110 4 -0.1154 -0.120 4 -0.125 4 -0.130 4 -0.135 4 -0.140 4 -0.145 4 -0.150 4 -0.155 4 -0.160 4

Zero field cooled Field cooled H = 250 Oe

50

100 150 200 Temperature (K)

250

300

Figure 14.28. Magnetization (300 K) as a function of field (top) and field-cooled (FC) and zero field-cooled (ZFC) magnetization versus temperature for AlGaN implanted with 3 X 10^^ Mn^ annealed at 1000°C for 2 min.

the localized spin). The validity of this model depends on having a high carrier concentration in the magnetic semiconductor and experimentally we do not observe this in the AlGaN and correspondingly we do not observe a Brillouin-like dependence of magnetization on temperature. In pure AIN, Mn produces an absorption line at ~ 1.5 eV from the valence band, suggesting the Mn^^^^"^ acceptor level is deep in the gap and makes

424


the realization of carrier-mediated ferromagnetism unlikely. The mean field models have also shown that Mn clustering can enhance the Curie temperature through localization of localized carriers at these clustered regions. Some calculations suggest it is energetically favorable to form ferromagnetic transition metal ion dimers and trimers at second nearestneighbor sites. Distant pairs would be weakly ferromagnetic or antiferromagnetic. These predictions suggest that the ferromagnetism will be a very strong function of the synthesis conditions used for the magnetic semiconductor. They also suggest that non-equilibrium methods such as ion implantation possess inherent advantages in trying to maximize the Curie temperature because of their ability to achieve solid solubilities for dopants well above those possible with equilibrium synthesis methods.

14.7. POTENTIAL DEVICE APPLICATIONS

Previous articles have discussed some spintronic device concepts such as spin junction diodes and solar cells, optical isolators and electrically controlled ferromagnets [76-80]. The realization of LEDs with a degree of polarized output has been used to measure spin injection efficiency in heterostructures. While the expected advantages of spin-based devices include non-volatility, higher integration densities, lower power operation and higher switching speeds, there are many factors still to consider in whether any of these can be realized. These factors include whether the signal sizes due to spin effects are large enough at room temperature to justify the extra development work needed to make spintronic devices and whether the expected added functionality possible will materialize. Among such devices the simplest seems to be the concept of an LED with one of the contact layers made ferromagnetic by incorporation of transition metal impurities, a socalled spin-LED [77-79]. Such a device should allow to modulate the polarization of the light emitted by the spin-LED by application of an external magnetic field. The most straightforward approach to achieve this goal would be to implant Mn into the top contact p-GaN layer of the standard GaN/InGaN LED. The electrical and luminescent properties of such devices show that they do produce electroluminescence, but due to the difficulty in annealing out all radiation defects the series resistance and the turn-on voltage of such spin-LEDs are very much higher than for ordinary LEDs and the electroluminescence intensity EL is lower. GaMnN layers produced by MBE have a lower density of defects and may be better suited for spin-LEDs. One of the problems with the latter approach is that the MBE-grown GaMnN films with high Curie temperature have n-type conductivity. Therefore, to incorporate such layers into the GaN-based LEDs one has to reverse the usual order of layers and grow an LED structure with the contact n-layer up. It was shown that problems such as higher series resistance due to lower lateral conductivity of p-GaN compared to n-GaN are inherent to these inverted diodes. Also it was shown that it is more


425

difficult to attain a high quahty of the GaN/InGaN multi-quantum-well (MQW) active region when growing it on top of a very heavily Mg-doped p-GaN layer. Incorporation of Mn into the top contact layer also produced a relatively high resistivity of the GaMnN and poorer quality of the ohmic contact. In addition, the ferromagnetism in GaMnN is found to be unstable against the type of high temperature (900°C) anneals used to minimize contact resistance. The reference n-LED structure studied consisted of 2-|ULm-thick undoped semiinsulating GaN, 2-|xm-thick p-GaN(Mg), five undoped QWs of InGaN (—40% In, 3 nm), separated by Si-doped n-GaN barriers (10 nm each) and about 170-nm-thick top n-GaN contact layer. All the layers but the top n-GaN layer were grown by MOCVD in the regime similar to the one used to fabricate standard p-LED structures. The n-type layer was grown by MBE at 700°C. The spin-LED structure differed from the n-LED structure by the structure of the top n-layer which consisted of 20 nm of n-GaN(Si) and 150 nm of GaMnN with the Mn concentration close to 3%. The GaMnN layer was grown at 700°C. All structures were subjected to rapid thermal anneal in nitrogen to activate the Mg acceptors. However, the spin-LED structure was given only a 750°C anneal since the degree of magnetic ordering in the GaMnN MBE-grown layer was greatly reduced upon annealing at temperatures exceeding 800°C. A schematic of the LED structure is shown in Figure 14.29, along with an electroluminescence spectrum taken at 300 K. Room temperature I-V characteristics of the reference structure and the spin-LED structure are compared in Figure 14.30. The spin-LED structure shows a reduced current in both forward and reverse directions. For comparison we also present the I-V characteristic measured on the similarly grown spin-LED structure for which the contacts annealing temperature was 900°C. Obviously, the reduced contact annealing temperature leads to further decrease of the current, because the contact resistance to the p-type layer was higher. This is confirmed by I-V measurements in the dark and with illumination at 85 K. In the test structure the open-circuit voltage deduced from these measurements was about - 0.5 V and had the right sign. For the spin-LED structure studied in this paper the sign of the open circuit voltage upon illumination was positive and the open-circuit voltage was about 1 V, i.e. opposite to normal, most likely due to the parasitic Schottky contact with the lower p-type contact layer. With increasing temperature the situation improved and the sign of the open-circuit voltage became negative, i.e. normal, at temperatures close to 160 K, but the impact of the parasitic Schottky diode was felt at all temperatures in the decreased current and decreased capacitance. For the reference diode, the temperature dependence was slight in both directions showing an activation energy of about 0.17 eV and ideality factor at temperatures higher than room temperature close to 2.5. This was explained by thermally assisted tunneling via the active MQW region. The onset of measurable electroluminescence signal in the test structure was close to 4 V. For the spin-LED structure corresponding activation energies

426

Optoelectronic Devices: Ill-Nitride ffW^^^^MM^W

GaN : Si 100 GaMnN

100 nm

GaN : Si 20 nm GaN : Si 10 nm InGaNM 3 nm GaN i

B I ^^^^^^^ GaN

V c y / ^ ^

lOnm

Mg 2|Lim

GaN i

2|im

Sapphire

420

430

440

450 460 470 Wavelength (nm)

480

490

Figure 14.29. Schematic of led structure (top) and measured 300 K EL spectrum (bottom).

as deduced from Arrhenius plots were close to 0.5-0.6 eV. The ideality factor in the forward direction, if deduced formally from I-V curves, was very high, close to 10. For higher forward voltages between approximately 5 and 10 V the temperature dependence of the forward current was small. Measurable EL signal could be detected at forward voltages close to 15 V and at these forward voltages the temperature dependence of the current again became relatively strong and showed an activation energy of 0.27 eV. At low forward biases the / - V dependences were of the form / - V " with a= 1.4. At higher biases the slope increases to 2.5. At voltages near 2 V the current jumps to another a = 2.5 region that prevails up to about 10 V when the current starts to increase sharply and shows an exponent value of about 8.4, but actually increases exponentially. The magnitude of the jump becomes lower with increasing temperature and the temperature dependence of the current at the second a = 2.5 region is very slight. Illumination of the diode in the rested state with UV light of the deuterium lamp led to a very strong increase of the current and the current remained considerably higher than the dark value after the light was switched off. Illumination of the diode with extrinsic light of either GaAs LED arrays (peak photon energy of 1.4 eV) or AlGaAs LED arrays (peak photon energy of


- 1 6 - 1 4 - 1 2 -10 - 8

-6

-4

All

-2

Voltage (V) Figure 14.30. Room temperature / - V characteristics of control LED (curve 1), of the old spin-LED amiealed at 900°C to form contacts (curve 2) and of the new spin-LED annealed at 750°C to form contacts (curve 3).

1.9 eV) led to much lower than intrinsic UV light increase of photocurrent in the reverse direction and for low forward voltages. Both spin-LEDs showed lower capacitance on the low-frequency plateau indicating that some high-resistivity Schottky diode is switched in series with the normal Schottky diode. This is also confirmed by C-V measurements that, for control sample, show a normal voltage intercept of 1.9 V and a concentration of IXlO^^cm"^ while both spin-LEDs show an un-physically high intercept values exceeding the bandgap of GaN and very high apparent concentrations on the order of 10^^ cm~^. The PICTS spectrum measured on our spin-LED structure at 0.5 V with the deuterium UV lamp light source showed three major features. The first feature was a negative-sign peak (relative to the ordinary sign of the peaks in PICTS) near 120 K with apparent activation energy of 0.16 eV. At these temperatures the sign of photocurrent is "wrong" due to the parasitic Schottky diode with the lower p-GaN contact layer which explains the wrong sign of the peak and allows us to unambiguously relate it to the lower p-type layer. The second feature was a normal-sign shoulder near 320 K with apparent activation energy of 0.8 eV. Finally, at higher temperatures one could see a steady increase in signal due to some very deep trap whose peak position (>1.4eV) could not be measured even with the longest time windows used (current relaxation curves measured for 100 s at each temperature point up to 400 K). Room temperature MCL and EL spectra of the new spin-LED both showed an intense line peaked near 2.7 eV is dominant, which comes from the GaN/InGaN MQW active region. At 90 K the MCL spectra showed, in addition to the intense 2.7 eV band, two bands peaked near 2.5 and 2.95 eV. The position of the 2.7 and 2.95 eV bands shifted to lower

428


energies with increased excitation intensity. The 2.95 eV band is due to some Mn-related centers with a level near 0.5 eV below the bottom of conduction band as discussed in some detail in Refs. [20,21,25]. The 2.5 eV band is thought to be due to luminescence from InGaN QWs with higher In contents formed due to partial separation of the solid solution. The parasitic diode formed with the lower p-GaN layer is detrimental to the device performance. Two approaches could be tried to alleviate the problem. The first is to check if donor co-doping of the GaMnN layer could be achieved without compromising the high Curie temperature of the material. If successful, this procedure should improve the led turn-on voltage. The second approach would be to try to incorporate one more n-GaN layer on top of the GaMnN film. This should improve the ohmic contact to the n-type portion of the structure and improve electron injection into the GaMnN film and thus into the active GaN/InGaN MQW region of the structure. We have shown that measurable electroluminescence signals could be obtained from spin-LED structures with the n-GaMnN layer on top even when the contact annealing temperature is maintained below 800°C in order not to destroy the magnetic ordering in the GaMnN film. With this low process temperature, however, a parasitic Schottky diode with the lower p-GaN contact layer is formed and manifests itself in reduced forward current at high voltages, reduced capacitance of the structure, higher threshold voltage for obtaining a measurable EL signal. In addition, the current at low forward voltages and at reverse voltage is limited by another parasitic junction which we believe to be the i-GaMnN/n-GaN junction in the top contact layer. The forward current at low voltages shows signs of the trap-filling regime and from the activation energy of the current in this region the main traps believed to be located in the GaMnN film have the levels near 0.5-0.6 eV from the bottom of conduction band and could be the same Mn-related centers observed previously in Mn-implanted films and in GaMnN films grown by MBE. Deep level spectra measurements in these spin-LEDs reveal the presence of 0.16 eV Mg acceptors and 0.4 eV hole traps in the p-GaN region and of 0.55 and 0.8 eV traps that are located most likely in the GaMnN layer.

14.8. ISSUES TO BE RESOLVED

The mean-field models consider the ferromagnetism to be mediated by delocalized or weakly localized holes in the p-type materials. The magnetic Mn ion provides a localized spin and acts as an acceptor in most III-V semiconductors so that it can also provide holes. In these models, the TQ is proportional to the density of Mn ions and the hole density. Many aspects of the experimental data can be explained by the basic mean-field model. However, ferromagnetism has been observed in samples that have very low hole concentrations, in insulating material and more recently in n-type material.


429

More work is also needed to establish the energy levels of the Mn, whether there are more effective magnetic dopant atoms and how the magnetic properties are influenced by carrier density and type. Even basic measurements such as how the bandgap changes with Mn concentration in GaN have not been performed. The control of spin injection and manipulation of spin transport by external means such as voltage from a gate contact or magnetic fields from adjacent current lines or ferromagnetic contacts is at the heart of whether spintronics can be exploited in device structures and these areas are still in their infancy.

ACKNOWLEDGEMENTS The work at UF was partially supported by NSF-DMR 0101438 and by the US Army Research Office under grants nos. ARO DAAD 19-01-1-0710 and DAAD 19-02-1-0420, while the work at SNU was partially supported by KOSEF and Samsung Electronics Endowment through CSCMR and by the Seoul National University Research Foundation. The authors are very grateful to their collaborators A.F. Hebard, D.P. Norton, S.N.G. Chu, J.S. Lee, and Z.G. Khim.

REFERENCES [1] Ohno, H. (2000) /. Vac. ScL Technol, B18, 2039. [2] Wolf, S.A., Awschalom, D.D., Buhrman, R.A., Daughton, J.M., von Molnar, S., Roukes, M.L., Chtchelkanova, A.Y. & Treger, D.M. (2001) Science, 294, 1488. [3] Ohno, H., Matsukura, F. & Ohno, Y. (2002) JSAP Int., 5, 4. [4] Awschalom, D.D. & Kikkawa, J.M. (2000) Science, 287, 473. [5] Cho, S., Choi, S., Cha, G.B., Hong, S.C, Kim, Y., Zhao, Y.-J., Freeman, A.J., Ketterson, J.B., Kim, B.J., Kim, Y.C. & Choi, B.C. (2002) Phys. Rev. Lett., 88 257203-1. [6] Medvedkin, G.A., Ishibashi, T., Nishi, T. & Hiyata, K. (2000) Jpn. J. Appl. Phys., 39, L949. [7] Medvedkin, G.A., Hirose, K., Ishibashi, T., Nishi, T., Voevodin, V.G. & Sato, K. (2002) /. Cryst. Growth, 236, 609. [8] Choi, S., Cha, G.B., Hong, S.C, Cho, S., Kim, Y., Ketterson, J.B., Jeong, S.-Y. & Yi, G.C. (2002) Solid State Commun., 122, 165. [9] Ueda, K., Tahata, H. & Kawai, T. (2001) Appl Phys. Lett., 79, 988. [10] Chambers, S.A. (2002) Mater Today, 34-39. [11] Dietl, T., Ohno, H., Matsukura, F., Cibert, J. & Ferrand, D. (2000) Science, 287, 1019. [12] Van Schilfgaarde, M. & Myrasov, O.N. (2001) Phys. Rev. B, 63, 233205. [13] Dietl, T., Ohno, H. & Matsukura, F. (2001) Phys. Rev. B, 63, 195205. [14] Dietl, T. (2001) /. Appl. Phys., 89, 7437. [15] Jungwirth, T., Atkinson, W.A., Lee, B. & MacDonald, A.H. (1999) Phys. Rev. B, 59, 9818. [16] Berciu, M. & Bhatt, R.N. (2001) Phys. Rev. Lett., 87, 108203.

430

Optoelectronic

Devices:

Ill-Nitride

[17] Bhatt, R.N., Berciu, M., Kennett, M.D. & Wan, X. (2002) J. Superconduct.: Incorporating Novel Magnet., 15, 71. [18 Litvinov, V.I. & Dugaev, V.A. (2001) Phys. Rev. Lett., 86, 5593. [19 Konig, J., Lin, H.H. & MacDonald, A.H. (2001) Phys. Rev. Lett., 84, 5628. [20 Schliemann, J., Konig, J. & MacDonald, A.H. (2001) Phys. Rev. B, 64, 165201. [21 Reed, M.L., Ritums, M.K., Stadelmaier, H.H., Reed, M.J., Parker, C.A., Bedair, S.M. & El-Masry, N.A. (2001) Mater Lett., 51, 500. [22: Reed, M.L., El-Masry, N.A., Stadelmaier, H., Ritums, M.E., Reed, N.J., Parker, C.A., Roberts, J.C. & Bedair, S.M. (2001) A/?/?/. Phys. Lett., 79, 3473. [23 Theodoropoulou, N., Hebard, A.F., Overberg, M.E., Abemathy, C.R., Pearton, S.J., Chu, S.N.G. & Wilson, R.G. (2001) Appl. Phys. Lett., 78, 3475. [24: Overberg, M.E., Abemathy, C.R., Pearton, S.J., Theodoropoulou, N.A., McCarthy, K.T. & Hebard, A.F. (2001) Appl. Phys. Lett., 79, 1312. [25 Sonoda, S., Shimizu, S., Sasaki, T., Yamamoto, Y. & Hori, H. (2002) /. Cryst. Growth, 237-239, 1358. [26: Kim, K.H., Lee, K.J., Kim, D.J., Kim, H.J., Ihm, Y.E., Djayaprawira, D., Takahashi, M., Kim, C.S., Kim, C.G. & You, S.H. (2003) Appl. Phys. Lett., 82, 1775. [27: So, Y.L., Kioseoglou, G., Kim, S., Huang, S., Koo, Y.H., Kuwarbara, S., Owa, S., Kondo, T. & Munekata, H. (2001) Appl. Phys. Lett., 79, 3926. [28: Kuwabara, S., Kondo, T., Chikyou, T., Ahmet, P. & Munekata, H. (2001) Jpn. J. Appl. Phys., 40, L724. [29 Thaler, G.T., Overberg, M.E., Gila, B., Frazier, R., Abemathy, C.R., Pearton, S.J., Lee, J.S., Lee, S.Y., Park, Y.D., Khim, Z.G., Kim, J. & Ren, F. (2002) Appl. Phys. Lett., 80, 3964. [3o: Hori, H., Sonoda, S., Sasaki, T., Yamamoto, Y., Shimizu, S., Suga, K. & Kindo, K. (2002) Physica B, 324, 142. [31 Pearton, S.J., Abemathy, C.R., Norton, D.P., Hebard, A.F., Park, Y.D., Boatner, L.A. & Budai, J.D. (2003) Mat. Sci. Eng., R40, 137. [32: Pearton, S.J., Abemathy, C.R., Overberg, M.E., Theodoropoulou, N., Hebard, A.F., Park, Y.D., Ren, F., Kim, J. & Boatner, L.A. (2003) /. Appl. Phys., 93, 1. [33 Dhar, S., Brandt, O., Trampert, A., Daweritz, L., Friedland, K.J., Ploog, K.H., Keller, J., Beschoten, B. & Guntherhold, G. (2003) Appl. Phys. Lett., 82, 2077. [34 Park, M.C., Huh, K.S., Hyong, J.M., Lee, J.M., Chung, J.Y., Lee, K.I., Han, S.H. & Lee, W.Y. (2002) Solid State Commun., 124, 11. [35 Lee, J.S., Lim, J.D., Kim, G., Park, Y.D., Pearton, S.J. & Chu, S.N.G. (2003) J. Appl. Phys., 93, 4512. [36: Ando, K. (2003) Appl. Phys. Lett., 82, 100. [37 Baik, J.M., Kim, J.K., Yang, H.W., Shon, Y., Kang, T.W. & Lee, J.L. (2002) Phys. Stat. Sol. (b), 234, 943. [38 Sarder, K., Raju, A.R., Basal, B., Venkataraman, V. & Rao, C.N.R. (2003) Solid State Commun., 125, 55. [39: Baik, J.M., Yang, H.W., Kim, J.K. & Lee, J.-L. (2003) Appl. Phys. Lett., 82, 583. [40: Shon, Y., Kwon, Y.H., Yuldashev, Sh.U., Park, Y.S., Fu, D.J., Kim, D.Y., Kim, H.S. & Kang, T.W. (2003) J. Appl. Phys., 93, 1546. [41 Dietl, T., Ohno, H. & Matsukura, F. (2001) Phys. Rev. B, 63, 195205. [42 Dugaev, V.K., Litvinov, V.I., Bames, J. & Viera, M. (2003) Phys. Rev. B, 67, 033201. [43 Schliemann, J. (2003) Phys. Rev. B, 67, 045202. [44: Rao, B.K. & Jena, P. (2002) Phys. Rev. Lett., 89, 185504.

Ferromagnetism

in GaN and Related Materials

431

[45] Korotkov, R.Y., Gregie, J.M. & Wessels, B.W. (2002) Appl Phys. Lett, 80, 1731. [46] Graf, T., Gjukic, M., Brandt, M.S., Stutzmann, M. & Ambacher, O. (2002) Appl. Phys. Lett., 81, 5159. [47] Polyakov, A.Y., Govorkov, A.V., Smimov, N.B., Pashkova, N.Y., Thaler, G.T., Overberg, M.E., Frazier, R., Abemathy, C.R., Pearton, S.J., Kim, J. & Ren, F. (2002) /. Appl. Phys., 92, 4989. [48] Theodoropoulou, N.A., Hebard, A.F., Chu, S.N.G., Overberg, M.E., Abemathy, C.R., Pearton, S.J., Wilson, R.G. & Zavada, J.M. (2001) Appl. Phys. Lett., 79, 3452. [49] Pearton, S.J., Overberg, M.E., Thaler, G., Abemathy, C.R., Theodoropoulou, N., Hebard, A.F., Chu, S.N.G., Wilson, R.G., Zavada, J.M., Polyakov, A.Y., Osinsky, A. & Park, Y.D. (2002) /. Vac. Sci. TechnoL, A20, 583. [50] Akinaga, H., Nemeth, S., De Boeck, J., Nistor, L., Bender, H., Borghs, G., Ofuchi, H. & Oshima, M. (2000) Appl. Phys. Lett., 77, 4377. [51] Hashimoto, M., Zhou, Y.Z., Kanamura, M. & Asahi, H. (2002) Solid State Commun., 122, 37. [52] Yang, S.G., Pakhomov, A.B., Hung, S.T. & Wong, C.Y. (2002) Appl. Phys. Lett., 81, 2418. [53] Wu, S.Y., Liu, H.X., Gu, L., Singh, R.K., Budd, L., Schilfgaaarde, M., McCartney, M.R., Smith, D.J., & Newman, N., to be published. [54] Liu, C , Alves, E., Ramos, A.R., da Silva, M.F., Soares, J.C, Matsutani, T. & Kiuchi, M. (2002) Nucl. Inst. Meth. Phys. B, 191, 544. [55] Ohno, H., Shen, S., Matsukura, F., Oiwa, A., Endo, A., Katsumoto, S. & lye, Y. (1996) Appl. Phys. Lett., 69, 363. [56] Janotti, L., Wei, S. & Bellaiche, L. (2003) Appl. Phys. Lett., 82, 766. [57] Martin, A.L., Spalding, CM., Dimitrova, E.I., Van Patten, P.G., Caldwell, M.C., Kordesch, M.E. & Richardson, H.H. (2001) /. Vac. Sci. TechnoL, A19, 1894. [58] Lu, F., Carius, R., Alam, A., Heuken, M. & Buchal, Ch. (2002) /. Appl. Phys., 92, 2457. [59] Cho, D.-H., Shimizu, M., Ide, T., Ookita, H. & Koumwa, H. (2002) Jpn. J. Appl. Phys., 41, 4481. [60] Hu, X., Deng, J., Pala, N., Gaska, R., Shur, M.S., Chen, C.Q., Yang, J., Simin, G., Khan, M.A., Rojo, J.C. & Schwalker, Z.J. (2003) Appl Phys. Lett., 82, 1299. [61] Serina, F., Ng, K.Y.S., Huang, C , Amer, G.W., Romni, L. & Naik, R. {20Q2) Appl. Phys. Lett., 79, 3350. [62] Lee, S.-H., Lee, J.-K. & Yoon, K.H. (2003) /. Vac. Sci. TechnoL, A21, 1. [63] Takagaki, Y., Santos, P., Wiebicke, E., Brandt, O., Schmerr, J.D. & Ploog, K. (2002) AppL Phys. Lett., 81, 2538. [64] Kipshidze, G., Kuryatkov, V., Zhu, K., Vorizov, B., Holtz, M., Nikishin, S. & Temldn, H. (2003) /. AppL Phys., 93, 1363. [65] Nishida, T., Kobayashi, N. & Ban, T. (2003) AppL Phys. Lett., 82, 1. [66] Gaska, R., Chen, C , Yang, J., Kookstis, E., Kahn, M.A., Tamulaitis, G., Yilmog, I., Shur, M.S., Rojo, J.C. & Schowalter, L.J. (2002) AppL Phys. Lett., 81, 4658. [67] Yang, S.G., Pakhomov, A.B., Hung, S.T. & Wong, C.Y. (2002) AppL Phys. Lett., 81, 2418. [68] Theodoropoulou, N., Hebard, A.F., Overberg, M.E., Abemathy, C.R., Pearton, S.J., Chu, S.N.G. & Wilson, R.G. (2003) Phys. Rev. Lett., 89, 107203. [69] Kucheyev, S.O., WiUiams, J.S., Zou, J., Jaegadish, C , Pophristic, M., Guo, S., Ferguson, LT. & Manasreh, M.O. (2002) /. AppL Phys., 92, 3554. [70] Li, J., Nam, K.B., Nakarmi, M.C., Lin, J.Y. & Jiang, H.X. (2002) AppL Phys. Lett., 81, 3365. [71] Hashimoto, M., Zhou, Y.-K., Kanamura, M. & Asahi, H. (2002) Solid State Commun., 122, 37. [72] Nam, K.B., Li, J., Kim, K.H., Lin, J.Y. & Jiang, H.X. (2001) AppL Phys. Lett., 78, 3690.

432

Optoelectronic

Devices:

Ill-Nitride

[73] Ji, Y., Stijkers, GJ., Yang, F.Y., Chien, C.C., Byers, J.M., Angvelovch, A., Xiao, G. & Gupta, A. (2001) Phys. Rev. Lett., 86, 5585. [74] Inumam, K., Okamoto, H. & Yamanka, S.J. (2002) /. Cryst. Growth, 237-239, 2050. [75] Suzuki, K., Kancho, T., Yoshida, H., Morita, H. & Fujimori, H. (1995) /. Alloys Compounds, 224, 232. [76] Ohno, Y., Young, D.K., Beschoten, B., Matsukura, P., Ohno, H. & Awschalom, D.D. (1999) Nature, 402, 790. [77] Jonker, B.T., Park, Y.D., Bennet, B.R., Cheong, H.D., Kioseoglou, G. & Petrou, A. (2000) Phys. Rev. B, 62, 8180. [78] Park, Y.D., Jonker, B.T., Bennet, B.R., Itzkos, G., Funs, M., Kioseoglou, G. & Petrou, A. (2000) Appl. Phys. Lett., 77, 3989. [79] Jonker, B.T., Hanbicki, A.T., Park, Y.D., Itskos, G., Furis, M., Kioseoglou, G. & Petrou, A. (2001) Appl. Phys. Lett., 79, 3098. [80] Pearton, S.J., Overberg, M.E., Abemathy, C , Theodoropoulou, N., Hebard, A.F., Chu, S.N.G., Osinsky, A., Zuflyigin, V., Zhu, L.D., Polyakov, A.Y. & Wilson, R.G. (2002) /. Appl. Phys., 92, 2047. [81] Frazier, R.M., Thaler, G., Abemathy, C.R. & Pearton, S.J. (2004) Appl. Phys. Lett., 84, 2578.

Optoelectronic Devices: Ill-Nitride M. Razeghi and M. Henini (Eds.) © 2004 Elsevier Ltd. All rights reserved.

Chapter 15

Phonons and Electron-Phonon Interactions in Ill-nitride Bulk and Dimensionally Confined Semiconductors and Their Device Implications Michael Stroscio^'^'"^ and Mitra Dutta*'*' ^Department of Bioengineering, University of Illinois at Chicago,Chicago, IL 60607, USA Department of Electrical and Computer Engineering, University of Illinois at Chicago, Chicago, IL 60607, USA ^Department of Physics, University of Illinois at Chicago, Chicago, IL 60607, USA

Fundamental properties of phonons in III-V nitrides are examined with a view toward understanding processes important in the operation of III-V nitride devices. Firstly, confined, interface and propagating modes in wurtzite quantum wells are described in terms of Loudon's model for uniaxial semiconductors and the dielectric continuum model. Basic properties of the phonon modes and carrier-phonon interactions are considered on the basis of this treatment of dimensionally confined phonons in both bulk and lowdimensional wurtzite structures. A key feature of these phonon modes is their enhanced dispersion and its origin from the non-isotropic nature of the wurtzites. As will be discussed, this dispersion has important consequences for phonon propagation and phonon energy spectra. The experimental results of the phonons in the wurtzites are summarized. An analysis of Raman linewidths measured for AIN and GaN wurtzites is made to estimate phonon lifetimes. The second-order phonon decay process of combined point-defect scattering and anharmonic decay is examined as a means of estimating line broadening associated with the decay of phonons in III-V nitrides of wurtzite structure containing point defects. Finally, the importance of understanding the role of interface-phonon interactions and surface-phonon interactions in the case of quantum dot lasers is discussed with the recent use of phonon-assisted transitions to improve the performance of semiconductor lasers by enhancing the population inversion. There has been much interest recently on compound semiconductors formed of group III elements and nitrogen. They have significant potential for optoelectronic and electronic devices due to the interesting properties of this system, including the relatively large bandgap, the large piezoelectric field, the relatively strong electron-phonon

E-mail address: [email protected] (M. Dutta).

433

434


coupling and the relatively high polar-optical phonon energy [1-10]. This chapter concentrates on the phonons in this system of bulk materials as well as the lower dimensional structures; moreover, this chapter focuses on the understanding of how these phonons behave in these structures and their impact in device applications. This group of compounds called nitrides may be either in cubic or hexagonal structure. There has been much investigation of the cubic crystal systems, the archetypical case being the well-studied gallium arsenide-based compounds, and their vibrational modes in the bulk as well as in the lower dimensional systems are better understood both with respect to their basic properties as well as in the manner in which they impact device properties. The behavior of the phonons in the cubic nitrides is similar to those in these systems except for the values of their frequencies. The binary nitride compounds, GaN, AIN and InN can be grown both as wurtzites as well as in zinc blende structures [11-15]. The same is true of the ternary alloys of these materials. However, they are more common and are highly stable in the wurtzite structure [11] and most of the discussion in this chapter will dwell in the materials in this structure. However in some thin films, polymorphs are possible and Raman scattering to identify the phonons has been used as a method to determine the structure as well as the purity of the materials.

15.1.

PHONON MODES IN WURTZITE NITRIDE STRUCTURES

Wurtzite crystals have a more complicated unit-cell structure compared to zinc blende ones, as shown in Figure 15.1. They have four atoms per unit cell and are of lower

Figure 15.1. Unit cell of wurtzite structure.

Phonons and Electron-Phonon Interactions in Ill-nitride Bulk

435

Table 15.1. Phonon frequencies in cm~ ^ of wurtzite GaN modes from Raman scattering experiments Reference

E\

El

Ai(TO)

[21] [22] [23] [24] [25] [26] [27]

145

568 568 569 568 570 569 567

533 533 532 533

144 143 144 142

533 530

Ai(LO)

738 735

î(TO)

î(LO)

559 559 560 559

726

561 558

743

symmetry than the zinc blende crystals. As a result, they have nine optical and three acoustic phonon modes. The space group that the wurtzite crystals belong to is C^^ and the zone centered optical modes are Aj + IBi + £"1 + 2E2 from group theoretical considerations. The Ai.Ei and the two E2 modes are Raman active while the Ex modes are not observed in Raman scattering or are silent [16]. The Aj and the Ex modes are polar as there is a long-range electrostatic field in the crystal. This polar field splits the A^ and the El mode into the longitudinal and transverse components, creating Ai(LO,TO) and £'i(LO,TO) modes. Due to the optical anisotropy of uniaxial crystals, the long-wavelength lattice vibrations can be classified according to mutual orientation between the c-axis, the phonon wave vector q, the electric field E, and the polarization P [17,18]. This divides the lattice vibrations into two groups of phonons: the ordinary and extraordinary. For the ordinary phonon, E and P are both perpendicular to q and the c-axis, dispersionless, while the extraordinary phonon frequencies are dependent upon the angle between the phonon wave vector and the c-axis, 6^. Two of the optical extraordinary branches correspond to the A^ and Ex modes at the F point. For an arbitrary ^^, extraordinary phonons exhibit mixed polarization. The zinc blende structure is much simpler [19,20], with two atoms per unit cell, and belongs to the T2d space group. There is only one Raman active mode of F2 representation which due to the polar nature splits into the TO and LO modes. The Raman frequencies of the phonon modes of GaN are listed below (Table 15.1). The frequencies of the phonons are close in the two structures and can be attributed to the fact that the dispersion curve of the phonons in the wurtzite structure in the [0001] can be obtained by folding the zinc blende phonon dispersion along the [111] direction as one would expect from the increased number of atoms in the unit cell. The growth of the nitride materials has been an ongoing research effort for a while and crystalline forms of the materials have been obtained as needles, platelets as well as more recently in epitaxial layers. The first identification of the phonons in GaN that were Raman active in the wurtzite form were done by IVIanchon et al. [21], Lemos et al. [22], Bums et al. [23], IVlurugkar et al. [24], Cingolani et al. [25] and Azuhata et al. [26].

436

Optoelectronic Devices: Ill-Nitride Table 15.2. Phonon frequencies in cm~^ of zinc blende GaN modes from Raman scattering experiments Reference

TO

LO

[12] [14] [15] [28]

555

737 730 740 740

554 555

The measurements have been much more extensive in the wurtzite material compared to the zinc blende materials as good quality materials in this latter form are not easily obtained. However, measurements of phonons by Raman scattering have been reported by Miyoshi et al. [14] on zinc blende GaN grown in GaAs by metallorganic vapor phase epitaxy. Phonons of zinc blende, wurtzite and mixed phase GaN films grown by molecular beam epitaxy on GaAs substrates were studied by Giehler et al. [15]. Here the differences in the growth conditions were studied and the structural properties were analyzed by X-ray scattering and electron diffraction as well. It was shown here that samples grown under Ga-rich conditions grew predominantly in the wurtzite structure and those grown under the N-rich condition were in the zinc blende condition. Table 15.2 lists the frequencies for the phonons observed under different configurations in Raman scattering. Phonons in the two different structural conditions were also reported by Tabata et al. [12]. Here the GaN samples were grown on GaAs substrates via molecular beam epitaxy. Again the Ga to N flux ratio was used to obtain the different growth conditions. Aluminum nitride AIN and indium nitride InN also are like GaN in that in the stable state they have a wurtzite structure as well as a zinc blende structure. The Raman selection modes and the selection rules for the phonon modes are thus similar (Table 15.3).

Table 15.3. Phonon frequencies in cm ^ of wurtzite AIN and InN modes from Raman scattering experiments Reference

E\

El

Ai(TO)

Ai(LO)

î(TO)

Wurtzite AIN [29] [30] [27] [31]

241 252 246 249

660 660 655 657

607 614 608 610

893 890

673 668

Wurtzite InN [32] [33]

495 491

î(LO)

924 916 913

596 590


437

Some information exists on the stresses in the nitride layers obtained from growing on different nitrides by looking at the shifts in the frequency of the phonon modes [34-36].

15.2. DIELECTRIC CONTINUUM AND LOUDON'S MODEL

In uniaxial materials, Loudon [17,18] showed that the polar phonon characteristics may be affected via two interaction mechanisms: that due to the long-range electrostatic field and the other due to short-range field which has the anisotropy of the force constants. The phonon dynamics that results is due to the dominance of the two mechanisms. If the long-range electrostatic field is the dominant mechanism, the interaction of the polar phonons with the long-range electrostatic field may result in a significant frequency separation between the group of the transverse optical (TO) phonons and the longitudinal optical (LO) phonons and the TO phonons and the LO phonons are separately grouped in a fairly narrow frequency range. As shown by Loudon, the dielectric continuum model [37] may be used to discuss the phonons in a wurtzite structure. This can be done by treating the dielectric constant as having one value parallel to the c-axis and another value, perpendicular to the c-axis. It was shown by Loudon that the phonons in the wurtzite structure can be accurately described by the dielectric continuum model. Nusimovici and Birman [38] provided detailed results later on the lattice dynamics of CdS which has a wurtizite structure. The application of Loudon's model to describe the carrier-polar-optical-phonon interaction as well as the corresponding electron-optical-phonon scattering rates in wurtzite crystals will be discussed further, both for the bulk material and the lower dimensional structure. At ambient temperatures, scattering by polar optical phonons is the main mechanism influencing electron transport [39]. Phonon interactions play very important roles in intersubband laser devices, especially in optical-phonon-assisted intersubband transitions [37,40,41]; thus, a complete and thorough understanding of electron-phonon interaction mechanisms and rates is essential. The dielectric continuum model and the uniaxial model of Loudon were used by Lee et al. [42] to formulate a theory of confined optical phonons in wurtzite heterostructure systems. The Frohlich interaction Hamiltonian was derived for bulk wurtzite, and the subsequent scattering rates were calculated. Loudon's model [17,18] describing the macroscopic equations of uniaxial polar crystals by introducing one dielectric constant associated with the direction parallel to the c-axis, e^, and another dielectric constant associated with the direction perpendicular to the c-axis, s^ h useful in allowing extension to the uniaxial case. For phonon mode displacements, it is convenient to separate the displacements parallel

438

Optoelectronic Devices: Ill-Nitride "T

800

LO-Like 600 TO-Like 400

200 0

30

60

90

Angle 6,^ (deg) Figure 15.2. Angular variation of phonon frequencies in bulk GaN.

to the c-axis, denoted u^, and those perpendicular to it, denoted Uj_ [42]. The phonon frequency for ordinary phonons has a trivial solution (o= cj^ and E(r) = 0, while the phonon frequencies for extraordinary phonons can be obtained by 8j_(a>)sin^(^) + 8,(cu)cos^(^) = 0.

(15.1)

where s^((o)

«2-

- «iz. -0,2

0,20,2-

-^r

(15.2) (15.3)

For the wurtzite materials, the solutions become ^ o = ^zL cos^(^) + WIL sin^(^).

(15.4)

ftô = ^z sin^(^) + (ol cos^(^).

(15.5)

These are predominantly longitudinal and transverse modes, respectively. Figure 15.2 shows the angular variation of the optical-phonon frequencies for GaN. The LO-like mode is almost flat thus indicating that the mode is weakly dependent upon the direction and the TO-like mode exhibits anisotropy. Recently, the angular variation for ZnO, CdS and CdSe have also been calculated [43]. The work of Shapiro and Axe [44] provided a detailed classification of the phonons in the wurtzite structures. In particular, Shapiro and Axe showed that the frequencies of


439

the phonons change with the change of the angle between the propagation direction and the c-axis. In directions that are not the principal axis of the crystal the phonons are of mixed symmetry and are termed quasi-LO or quasi-TO modes. These modes have been studied in Raman scattering by Filippidis et al. [45] and by Bergman et al. [46].

15.3. TERNARY ALLOYS Mixed crystals of the form ABi_^Q are classified into two main groups according to the behavior for the ^ ~ 0 optical phonons. There are two possible situations for zinc blende materials which have been well-studied and referred to as one mode and two-mode behavior [47]. In general, if the frequencies of the modes of the binary materials, AB and AC differ by a large amount then the mixed crystal follows the two mode behavior. If the mixed crystal follows the one-mode behavior then the frequencies of the two binaries that make up the ternary alloy have a much closer values. In addition, there is a third class where the ternary alloy has a one-mode behavior in one frequency range and a twomode behavior in another frequency range. There is a lack of consensus in the early experimental results where the work of Hayashi et al. [48] showed that in the composition range of 0 < jc < 0.15 the AlGaN alloy system exhibits a one-mode behavior. In another study, by Behr et al.[49] found that while the Ai(LO) phonon behaved in a one-mode fashion the E2 phonon showed no change due to alloying. The work of Cros et al. [50] indicated that the E2 mode was a two-mode type while the Al was one-mode. Demangeot et al. [51] found similar results in that Ai(LO), Ai(TO) and the Ei (TO) were all one-mode although infrared measurements of Wisniewski [52] et al. indicated conflicting results for the £'i(TO) modes. This underlines the lack of understanding of the vibrational properties of the alloys which are possibly tied to the difficulties in obtaining consistent material quality under differing growers and growth conditions. The polar phonons in three III-N ternary alloys have been studied theoretically [53] by using the modified random-element isodisplacement (MREI) model. The anisotropy wurtzite structure and the additional modes were added to previous models for the zinc blende structures. It was demonstrated that the polar modes of the ternary alloys GaAlN and GaInN follow a one-mode behavior. These results agree well with the experiments of Ref. [53] except for the £î(TO) case. The one-mode behavior can be understood qualitatively in view of the fact that the nitrogen mass is significantly less than the mass of the other ions. Indeed, the nitrogen ions oscillate with one frequency with respect to planes of mixed group III elements. Demangeot et al. [51] have analyzed Raman data on Gai_Âl_^N based on a generalized dielectric model for coupled LO modes and have obtained support for the apparent one-mode behavior of the polar LO phonons. The MREI model [53] also demonstrates that the Aj and Ex optical

440


phonons in Ga;cAli-;cN and In^^-Gai-^^N exhibit one-mode behavior unUke Ga; ^ '

b diffuse scattering

+«(^'^te)V

,e'^*

(16.15)

Bragg peaks

The terms marked as diffuse scattering are not ^-dependent (though they would be if we had not assumed the statistical independence of Eq. (16.14)) but is A^MC times smaller than the Bragg peaks term. The latter term clearly identifies the order by peaking at the reciprocal lattice vectors G. Figure 16.7, that was actually calculated for the MC sample of InGaN, exemplifies the identification of order. If there is an ordered phase, at lower temperatures {S(k)S(ky) peaks at the reciprocal lattice vector of the ordered configuration, as it will be seen in Section 16.4. As the temperature is raised, the peak decreases and disappears above the transition temperature.

16.3. PHASE SEPARATION IN THE TERNARY ALLOYS In this section we discuss the phase diagrams obtained for the BGaN, BAIN, AlGaN, InGaN, and InAlN alloys, as calculated by combining the first-principles total energy calculations with the GQCA to disorder and compositional fluctuations.

Phase Separation and Ordering

465

16.3.1 BGaN and BAIN In order to exemplify the procedure carried out for obtaining the alloy temperaturecomposition phase diagrams, we show in Figure 16.2 the calculated mixing free energies AF from which we derive the T X x curve for the B;,Ali_;,N alloy. Note that the shape of the curves of both, AF X x and 7 X x is asymmetric. This is due to the fact that the cluster energies, Ej do not vary linearly with the cluster kind, 7, leading to asymmetric dependence of the mixing free energy. As the temperature increases, the typical two-minima behavior of the mixing free energy changes to a single-minimum behavior, as can be seen in Figure 16.2(a). The solid (dashed) line in Figure 16.2(b) corresponds to the binodal (spinodal) curve. Below the critical temperature at T^ ~ 9500 K, the two-minima behavior of AF reflects in the occurrence of a miscibility gap for x in the interval Xi < x < X2. The points x = Xi and x = X2 correspond to those at which the common tangent line touches

(a) 0.05

-

^^-"^000

^

^ ^s._^

0.00

8oog__

CD Q.

5

-0.05

LL

—'

(D 6000

^

•

\ ^

i^rT

8000

1

if.

^ -^ ' ^ ^—

1 i

•

X2 i

X2

"^""^^ \ ^! ^ -H \ \ \

X / / ^ _/ / / / / ^

\

2 8. 4000 J / E / / (D 1 ^ 12000 1 /

\ \

\

[/

n

1 \\

/

:

0.0

1

\

1

0.2

0.4

•

0.6

•

.

0.8

•

\

1.0

Boron content x Figure 16.2. Mixing free energy AF for three different temperatures (a) T X x phase diagram (b), as functions of the B composition, for the B^^-Ali-^N alloy. Solid line: binodal curve; dashed line: spinodal curve.

466


the AF curve (which defines the binodal curve). The two composition values, JC = jcj and X = X2» correspond to the inflection points in the AF curve (which defines the spinodal curve). ¥oYXx < x < xliOxx!2< x < X2 the alloy is metastable against local decomposition because the value of AF for any x in these regions is lower than the average value of AF with any two compositions in the neighborhood of x. This increase in AF acts as a temporary energy barrier against alloy decomposition into its final equilibrium concentrations Xi and X2. For an alloy with concentration between Xj and JC2 there is no such decomposition barrier and the alloy is inherently unstable [46]. Figure 16.3 depicts the temperature versus composition phase diagrams for the B;cCrai _;^N and B^Âli -^. The behavior observed for both alloys is very similar, with very high critical temperatures, T^ ~ 9500 K, which result in a very large miscibility gap. Recent calculations carried out within the framework of the strictly regular solution model combined with a valence force field approach lead to the critical temperatures 11,400 and 8330 K for wurtzite phase BGaN and BAIN, respectively [43]. These results lead to the conclusion that it is very difficult to grow a nitride alloy based on boron, at least in the unstrained case. For typical growth temperatures for BGaN (—lOOOK) and BAIN (~ 1300 K), we observe phase separation for a wide range of composition. For these temperatures the phase diagrams shown in Figure 16.3(a) and (b) indicate that there is a spinodal decomposition in the interval 0.028 < jc < 0.995 for B;,Gai_;cN, and in the interval 11000

-

•

9000 -

/ / /

8000

g 0)

3

7000 6000 -

1 1

/ f

1

/

/

/

\

-

\

\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \

L

1

1

1

1

1

1

0.4 0.6 0.8 Composition x

/

/

\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ ' \ \ \ \

/

I

1 1

1 /

1 1 1 1 1 1 1

/

/

- , 1

0.2

/ / /

/

1 1 /

:

1

X

1 1

1

0" 0.0

/

1 / j /

-

1 1 1

K

1000

/

/

1 1

1 1 1 1

1 1 1 1 1

2000

j

\

1

/ /

/ / / / / / / / / / / / / / 1 / 1 /

"

1 1 1

f

-

/ /

]

\ \ \ \

/ \

4000 3000

\

/

/ / / / / / / / / / / /

g. 5000 E H

BxGai_,N

BxAli_xN

10000

1 1 1

1

1

1.0

0.0

1

0.2

1

1

0.4

1

1

1

0.6

1

0.8

1

1.0

Composition x

Figure 16.3. Phase diagrams T X jc for the BÂli _;,N and BJ3ai -J>i alloys. The horizontal dashed lines around T ^ 1000-1300 K delimit the range of growth temperatures.


467

0.037 < X < 0.949 for BjÂli-^^N. Such a picture is consistent with the experimental findings, explaining why successful growth of these boron-related alloys is achieved only for very small B content, x ^ 0.03 (0.08) for BGaN (BAIN) at growth temperatures [16,65,66]. 16.3.2 AlGaN, InGaN, and InAlN Figure 16.4 depicts the temperature-composition phase diagrams for the AlGaN, InGaN, and InAlN alloys. More in detail they show the spinodal and binodal curves calculated within the GQCA and the ab initio total energy method. We observe that a very low critical temperature, Tc — 70 K, is obtained for the Al^^Gai-^^N alloy, thus for growth temperatures no miscibility gap exists for AlGaN. In contrast to AlGaN, for the In^^-Gai-^^N alloy there is a large miscibility gap at typical growth temperatures, ~ 1000 K, therefore, indicating that unstrained InGaN presents phase separation effects, or spinodal decomposition, with an In-rich phase of x ~ 0.8 (and an In-poor phase of JC ~ 0.2). The phase separation, which should occur at growth temperatures in In^^Gai -j^, is driven by the internal strain due to the mixing of the two lattice-mismatched components InN and GaN. Similar features are obtained when MC simulations are used. The predicted phase separation for the InGaN alloys is in very good agreement with experimental data [67]. Figure 16.5 depicts the calculated phase diagram for In^^Gai-^cN together with the data (filled circles in the figure) for the In-rich phase compositions,

1600

T = 70 K

ln^Gai_^N V1295K

1400 1200 £ (D

0 Q.

E 0

1000 800 h 600 400 200 NTNJ 0 j^S 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 Composition x Composition x Composition x

Figure 16.4. Phase diagrams, T X jc, for the Al;cGai -^^N, In;cGai -^^N, and In^Âli - ^ alloys. The values obtained from the calculations for the critical temperature, T^, are also shown.

468

Optoelectronic Devices: Ill-Nitride 1500 ln^Gai_^N 1250 -

1000

2

1

750

/ / / / / / 1 1 1 1 1 11 /

1

\

\ \

1

/ • / . / / /

\

\\

• \

f

\ \ \

Q.

E

\ \ \ \ \

0

500 -

250

1

0.0

1

0.2

1

I

0.4

I

I

.

0.6

J

0.8

1

1.0

Indium content x Figure 16.5. Calculated TXx phase diagram for unstrained c-In^^-Gai-^N. The horizontal dotted lines give the range of the growth temperatures. The filled circles (experimental results as extracted from Ref. [67]) denote the Indium contents of three different samples, x = 0.07, 0.19, and 0.33. The dashed rectangle denotes the observed In-rich phase, jc = 0.8 ± 0.08 [33,34].

as obtained from X-ray diffraction and resonant Raman spectroscopy measurements performed on c-InGaN thick layers grown on GaAs (001) substrates by IVIBE withx = 0.07, 0.19, and 0.33 [67]. An In-rich phase with x ~ 0.8 is observed in the three samples as obtained from the calculated phase diagram. We should point out that a range of X = 0.72 - 0.88 (shown by the dashed rectangle in Figure 16.5) has been obtained from micro-Raman spectroscopy measurements. We observe for the In;,Ali_;,N alloy the same behavior as for In;,Gai_;,N with a Httle higher critical temperature, T^ = 1485 K. The result depicted in Figure 16.4 indicates that for T ~ 1000 K there is a phase separation for a wide range of composition, for In content between 15 and 79%. This result is in good agreement with experimental findings that show a tendency of phase separation for jc > 0.17 [68]. There are few theoretical works in


469

the literature which analyze the thermodynamic stability in the InAlN alloy, and all of them use the strictly regular solution model, with only one alloy configuration [39,43,44,69] or some but not all configurations [70]. They found a range for the critical temperature which varies from 1474 to 3400 K. The phase diagram shown in Figure 16.4 is much more asymmetric than those obtained through the regular solution model, and we obtain a larger range of miscibility for higher In molar fractions x. The phase boundaries are also found to be steeper for x —• 0 and jv: —^ 1 than for ideal alloys.

16.4.

ORDERED PHASES IN InGaN AND AlGaN ALLOYS

In this section, we discuss the tendency of the InGaN and AlGaN alloys to order, and the influence of the coherently epitaxial growth on them. We would like to know whether, during the growth, the individual components would tend on a microscopic scale to attract or repeal each other, so that there is a short-range order. We would also like to know whether the individual components will tend on a macroscopic scale to cluster into ordered phases of particular stoichiometrics. We have already seen in the section before, that the InGaN alloy has a tendency to "unmix" and the AlGaN alloy to form a solid solution. Besides these behaviors it was already reported the existence of ordered phases in both alloys. Therefore, in this section we analyze the possibility of ordering formation in both alloys for (i) the bulk phase, in other words, when they are unstrained; (ii) the case when the alloys are coherently grown on substrate with different lattice constant, then, in this case, they will be "externally" strained. If the alloy is incoherent with the substrate, then it is free to adopt the in-plane lattice constant that minimizes its free energy. If the alloy is coherent with the substrate, then it must adopt the in-plane lattice constant of the substrate; the resulting elastic strain energy can increase its overall free energy significantly. We will find that the thermodynamic properties of InGaN depend greatly on whether the alloy is coherent or incoherent with the substrate. In fact, such coherence constraints greatly suppress the tendency for alloys to separate into their pure-component "endpoint" phases, and at the same time greatly enhance their tendency to form ordered compounds at certain stoichiometric compositions. These tendencies can be understood from a CE method together with ab initio calculations and MC simulations, as described in Section 16.2.2. As reported, the basic structure of the method involves the calculation of the total energies by a first-principles method, E^p(a), of some configurations, these are then used via Eq. (16.10) to obtain the interaction energies Jfj, which can be used to perform a ground state search and phase diagram calculations. Schematically, first-principles calculations => {EYp(a)} => [Jfi] => ground state search => thermodynamics. In the sequence, we describe each step more in detail and the respective results obtained for InGaN and AlGaN alloys.

470


16.4.1 First Principles Calculations We start by performing first-principles calculations according to Section 16.2.4 in order to obtain the total energies of the set of configurations to be used in the CE (Eq. (16.10)). The set of configurations includes all with 2 and 4 cations per unit cell, and 7I and yl (MoPt2), and the superlattice [3,3] along the (001) direction. The configurations are named as in Ref. [71] to which, for an alloy as A^Bi-xC, we add a fraction n/m, like 1/2 or 3/4, saying that among the m sites of the unit cell n are occupied by an A atom and m — nby 3. B atom. In particular, the name of the configuration "40" (chalcopyrite) was changed to DO222/4 because it has the same unit vectors as D022In the case where the alloys are "externally" strained, we consider that they were grown on a buffer of GaN, as it is usual experimentally. Then, in order to determine the total energy of each configuration under the macroscopic strain produced by the pseudomorphic growth, the lattice parameter in the plane (001), a\\, was held fixed and equal to that of GaN. In this case, most of the configurations may be oriented in two ways, one mostly oriented in the (001) direction, the other mostly oriented perpendicularly. In particular, the configurations Lli, V, and LI2 do not split due to the lattice uniaxial deformation. The configuration total energies were calculated as functions of the lattice parameter c along the (001) direction and minimized with respect to the atomic positions within the unit cell. 16.4.2 The {Jfi} and the Ground State The obtained total energies are then used to determine the interaction energies JfjS. The quality of the CE is determined by comparing the energies for the configurations determined by the CE with the energies determined by a direct calculation (first-principles calculations). If necessary, one can repeat this procedure by adding extra Jfjs until the difference between the predicted energies for {a} and the direct calculated energies are smaller than some prescribed tolerance. In the particular case of InGaN and AlGaN, respectively, we found that the interaction energies parameters {/o,0' -^0,1' -^0,2» J 1,1^ J2,0^ J2,2' -^3,0' -^3,1' A O ' A 2 ' -^2,2» ^ , 0 ' ^ , 2 ' ^2,0^ ^ 2 , 2 } ^^^

{-^0,0' JQ,\^ J0,2^ J\,\^ J\,2^ ^2,1^ J2,2^

-^3,1' JA,2^ ^2,2. ^ , 0 ' ^ , 1 ' ^,2^ ^2,0' ^2,2} ^^c scts able to predict the energies of all the important configurations with tolerable errors. The set of the obtained interaction energies /y^/S is then used in the novel CE (Eq. (16.10)) to predict the energies of new configurations a. This was made for a large set of configurations, a total of 5868. Then, by using the same procedure as in Ref. [49] to identify those structures which minimize the energy expression at each x, we obtained the ground state line (GSL). The GSL is made of straight line pieces in the plane AE" versus x such that any configuration has energy greater or equal to the two phase mixtures corresponding to the straight Hne pieces. A^" is the standard definition of the alloy excess energy taken as the difference between the alloy energy and the mixture energy of the binaries, AE" = E{&) - [xEp^ + (1 - Jc)£'GaN]. where A = In or Al.


All

We first address the results for the unstrained or fully relaxed alloys. It was found for both AlGaN and InGaN alloys that for all alloy compositions there are no stable ordered structures. Then we investigated the relative stability of ordered and disordered phases, considering the coherent growth case, for which a\\ = aoaN- The resulting GSL for AlGaN and InGaN are shown in Figure 16.6(a) and (b), respectively. The results shown in Figure 16.6 comprise interesting features: (i) there are ordered structures with AE < 0, which means that at T = OK these structures are more stable than phase separation; (ii) the biaxial strain induced by the GaN buffer suppresses the phase separation process and acts as a driving force for the ordering formation; (iii) the GSLs are very different in (a)

J^ - 1 4 (b) 0 LJJ

30 meV has been reported for 6H-SiC by Choyke and Patrick from photoluminescence measurements [125]. Qteish et al. [126] obtained A^f = 0.12 eV for 2H-SiC using a supercell pseudopotential approach to calculate the spontaneous polarization. Their VBO between unrelaxed 3 C - and 2H-SiC is similar to the fullpotential results. In Figure 17.5 we summarize the most important band-structure quantities of 3C-, 2H-, 4 H - , and 6H-SiC by showing the bandgap energies, the VBO


492

1

1.2 (0

• • ^

0.9

ID c (75

•^

0.6

3 O CD O O

£

0.3

1

/

/

o

'

r /

3

^

1

3C-SiC 4H-SiC 6H-SiC

y /

•• •

1

/

1

• /

1

1

3.0 Energy (eV)

2.5

1

3.5

Figure 17.5. Measured photoacoustic (PA) signals as a function of energy for 3C - , and 4H-, and 6 H - SiC. The spectra of the samples were normaUzed to the spectra of a highly absorbing film [122].

(with respect to the VBO of 3C-SiC), the energy differences between the two lowest conduction-band minima, and the valence-band split-off bands. The dielectric function s((o) = Si((o)-\-is2(co) of the semiconductors describe the response of the material due to a change in the charge distribution. The dielectric function is thus an important property for describing the screening of the semiconductor near dopants, defects, and other structural perturbations of the crystal. For instance, the static dielectric constant is used to determine the correction to the LDA bandgap energy [Eq. (17.1)] and in the calculation of the polaron masses (Eq. (17.6), see the next section). Within the linear response theory the dielectric function in the long wavelength limit (q = 0) is calculated directly from the electronic structure via the joint density-of-states and the optical matrix elements according to (Figure 17.6) sf\o)) = ^ 2 2

X

) = 8(0)]. The dotted lines in (b) represent the parabolic approximation (see text) with m\\ = 1.83mo for the conduction band in 6H-SiC.

dispersion with m\\ = 1.83mo. At concentrations larger than about —10 cm~ the electron gas concentration is so high that the band filling exceed the parabolic region of the double-well minimum in 6H-SiC. 17,4,1 Metal-Non-metal Transition Above a certain critical doping concentration A^^^ the donors (in n-type materials) or acceptors (in p-type materials) spontaneously become ionized due to strong impurityimpurity interaction. This metallic phase is of great technological importance for designing for instance p-i-n-diodes. We calculated this critical concentration from three different models [78]. The first two models (model No. 1 and No. 2) are based on the Mott picture [139] of overlapping impurity electrons, assuming hydrogen-like wave functions. The third model (model No. 3) is comparing the total energy of the nonmetallic weakly interacting impurity electrons, and the total energy of the metallic strongly interacting impurity system. 17,4,1.1 The Original Mott Model (Model No, 1), In the Mott model of impurity systems [139], the doping-induced MNM transition occurs at the critical impurity

508


concentration A^c given by

-(f)' al = "

e^

(17.22a)

(17.22b)

28(0)ED,A

where the effective Bohr radius a^ is calculated from the ionization energy E^ ^ of a single donor electron (or acceptor hole) since the impurity electron wave function is assumed to be associated with only one conduction or valence band, that is, the many valley effects are neglected. 17.4,1.2 The Mott-Hubbard Model (Model No. 2). Through the use of a MottHubbard tight-binding Hamiltonian, the impurity density-of-states associated with it present two sub-bands that overlap with increasing concentration. This would occur at an impurity concentration for which [141-143] AW = 1.15, (17.23) U where AW is the unperturbed impurity band width in units of £'D,A' ^^^ ^ is the intra-impurity Coulomb interaction energy, also known as the Hubbard-(7, given by U = 0.96ED [144,145]. Such a scenario is well known as the Mott-Hubbard picture for the MNM transition. AW is related to the hopping integral energy T, between adjacent sites / and 7, as [143,144] AW = 21(7)1,

(17.24)

where (T) is defined as the average hopping energy [143,144] {T)= {T(R)P(R)dR

(17.25)

P(R) describes the distribution of the donors. Moreover, T(R) and U are given by [146] T(R) = j ilJi(r)H, (A/(r - R,)dr,

(17.26a)

U = { |î(ri)PliA2(r2)l'^7Tn^ fdridr^. (17.26b) J e(0)lri - r2l Hi is the one-particle Hamiltonian in the effective-theory, including the kinetic energy operator and the Coulomb interaction of the positively charged donor ion electron. 0)(r - Rj) is the simple hydrogenic wave function for the donor ground state at the randomly located site R,. Moreover, we have used a random like Poisson distribution P(R) of the donors with

Sic and III-N Heavily Doped

509

the probability that the nearest donor neighbor lies at a distance (in units of an). P ( / ? ) =3- ^^ / 1 + -R^ ^

Rl\

Y,

(17.27a)

RU) '

where

«„, = ( M - ) - " .

(„.27b,

Eqs. (17.23)-(17.27) show that the values of U and the donor (or acceptor) impurity concentration A^^ âre both related to the ionization energy ^'DA- Using these equations, we have calculated the values of the critical concentration A/^. 17.4,1.3 The Total Energy Approach (Model No. 3). The third model follows a method expounded by Semelius and Berggren [74], where the total energy E^^ of the localized donor electrons in the non-metallic phase is determined and compared with the total energy E^^ of the electron gas in the metallic phase. For low donor concentrations, the total energy of the localized electrons is lower than the corresponding energy of the electron gas, and thus the non-metallic phase is favored. For high donor concentrations the situation is reversed, unless the ionization energy very large. The critical concentration for the transition is obtained as the concentration at which the total energies of the two phases are equal, i.e. Ef^ = E^^. The total energy of the metallic phase is calculated from Eqs. (17.13)-(17.18). The total energy for the non-metallic phase is directly related to the dielectric function of the localized electrons associated with the donors. Leroux Hugon and Ghazali [147], have derived the dielectric function of the donor electrons as a function of impurity concentration, where a hydrogenic wave function was presumed. When the concentration is increased, the donor electrons screen the Coulomb potential of the impurities more strongly, whereby the dielectric function is increased. The change in the dielectric function will modify the ionization energy of the electrons, and thereby, also the total energy. The total energy (expressed in energy per electrons) of the donor electrons in the non-metallic phase is obtained as [5,74]. ,3/2-

z^NM ^tot

In Figure 17.10 we show the total energy per electron in n-type Si (doped with P) and n-type SiC polytypes (doped with N) as functions of donor concentrations for the metallic phase and for the non-metallic phase. It is clear that at low donor concentrations the weakly interacting non-metallic phase is energetically favored, but at concentration above cm the ionized metallic phase is energetically favored. In SiC, the main contribution to this spontaneous ionization is coming from the electron-ion interaction.


510

(b)

(a)

1

u

' '

1

^tot

\^tot

> 0

1 "

25

^D(P)=\: 50 _ 4 6 m e V ^ ^ \ NM

'

ED(N)

75

-

\

2H -SiCy

> ^

50

C

o "o

Si / /

(D

^tot HD(N)=

52meV

HD(N)=

81 m e v \

4H- SiC -

100

\

= 54 meV 3C-SiC

0

-

10^^ 10^^ 10^^ Ionized donor concentration (cm~3)

-

6H- -SiC^

10^ 10' 10^ Ionized donor concentration (cm~3)

Figure 17.10. The total energies per particle of the metallic ^^^t (solid lines) and the non-metallic E^^ (dashed line) phases in n-type (a) 3C-SiC and Si, and in (b) 2H-, 4H-, and 6C-SiC as functions of donor concentration A^^. The dashed lines represents the estimate of the total energy of the electron in the non-metalUc phase of Si:P, 3C-, 2H-, 4H-, and 6C-SiC:N, where the donor binding energies are the low concentrations values.

The critical concentration A^^ is obtained from the intersection point E{ E^f The main ^tot difference between the total energy of the metallic phases of 2H-, 4H-, and 6H-SiC is the number of conduction-band minima which affect the band-filling and the Fermi energy of the electron gas [5,6]. For very high doping concentrations, the kinetic energy of the electron gas dominates. At an impurity concentration of 10^^ cm~^ the total energy per electron in 4H-SiC is about 30% lower than in 2H-SiC and most of this difference arises from the kinetic energy. The calculated critical concentrations for the different n- and p-type SiC materials are presented in Table 17.5. Note that the extended Mott-Hubbard model gives normally slightly smaller A^c than Mott's original model, but all the three computational methods give the same order of A^^ • Moreover, the present results for p-type Si agree with earlier published calculations of Nubile and Ferreira da Silva [143]. The resulting critical concentrations for p-type SiC poly types are high; for Ga and B (with large ionization energies) N^ is near the upper limit to the dopant concentrations of interest in devices. In Mott's model, the critical concentration is proportional to £'D,A' The same relation holds also, with fairly good accuracy, in the Mott-Hubbard model. For the total-energy calculation, however, this is no longer true [5,6], especially for large ionization energies. For sufficiently large ionization energy E^^ does not become equal to f^^ at any concentration, but for these large ionization energies (about 0.35 eV in p-type SiC) one can question the accuracy of using hydrogenic wave functions together with the effective mass approximation.

Sic and III-N Heavily Doped

511

Table 17.5. Critical concentrations for the MNM transition of n- and p-type SiC, calculated using three different methods (see the text) Polytype

Dopant

Si

P B

n-type p-type

3C-SiC

N Al Ga

n-type p-type p-type

4H-SiC

N Al B

n-type p-type p-type

Ga

p-type

N

n-type

Al

p-type

Sc B Ga

p-type p-type p-type

6H-SiC

Critical concentration Nc (cm

^D,A (meV) [2]

45.6 46 54 257 344 52 and 92 191 234-326 267 81, 137.6, and 142,4 239 249 240 320 333

')

Mott

Mott-Hubbard

Total energy

Expt.

7.0X10^^ 7.2X10^^

5.7 X 10^^ 5.9 X 10^^

3.8 X 10^^ 4.3 X 10^^

3.5 X 10^^^^^ 5.0 X 10^^^^^ 4.06 X 10^^^"^

6.0 X 10^^ 6.5 X 10^^ 1.6X10^^

4.9 X 10^^ 4.7 X 10^° 1.2X10^^

3.5 X 10^^ 2.3 X 10^^ 7.9 X 10^^

5.5 X 10^^ 2.7 X 10^° (0.50-1.4) X10^° 7.5 X 10^^

5.6X10^^ 2.1 X 10^° (0.39-1.0) XIO^^ 5.8 X 10^°

5.6X10^^ 8.8 X 10^^ (0.22-1.0) XlO^^ 4.3 X 10^^

2.1 X 10^^

2.8 X 10^^

2.6 X 10^^

5.3 6.0 5.4 1.3 1.4

4.2 X 10^° 4.7 X 10^° 4.3 X 10^° 1.0 X 10^^ 1.2X10^^

2.6 X 10^^ 3.2 X 10^^ 2.7 X 10^^ 8.9 X 10^^ 1.0X10^^

X X X X X

10^° 10^° 10^° 10^^ 10^^

For comparison we also present the calculated and measured values of n-type Si:P and p-type Si:B. Measured critical concentrations are from (a) Refs. [148,149], (b) Ref. [150], and (c) Ref. [151].

Experimental investigations of the critical concentration for n- and p-type SiC polytypes are lacking. We, therefore, compare our corresponding calculations of n-type Si:P and p-type Si:B with measurements. Our calculations of the critical concentration of Si:P (Nc = 3.8 X 10^^ cm~^) is in good agreement with capacitance measurements performed by Castner et al. [149] (A^^ = 3.5 X 10^^ cm~^). Corresponding calculation of Si:B [Nc = (4.3-7.2) X 10^^ cm~^] is in good agreement with Hall measurements by Kubiak et al. [150] (Nc = 5.0 X 10^^ cm"^), and with resistivity measurements by Dai et al. [151] (Nc = 4.06 X 10^^ cm~^). Mott's original model seems to give shghtly too large Nc for both n- and p-type materials, except for n-type 4 H - and 6H-SiC polytypes, when two and three different ionization energies, respectively, are involved [152]. 17,4.2 Reduced and Optical Bandgap Energies For doping concentrations above the MNM critical concentration, the free electron (or hole) gas associated with the ionized donors (acceptors) will screen the crystal host electrons. This screening results in an energy shift in the single-particle energies.

512


described by the real part of the self-energy Rc[hY,j (k, E^(k)/h)]. The energy shifts of the valence-band maximum A^î and of the conduction-band minimum AEî will cause a narrowing of the fundamental bandgap energy. This is the doping-induced BGN. We calculate the single-particle self-energy from the total energy E^^ of the interacting (or perturbed) system consisting of the electron gas and the ionized impurities: E^^ = {'^\Hi I ^ . The self-energy represents the interaction in the total system and the resulting excitation energies include, therefore, not only the single electron excitations but also the collective excitations. However, in the present work only the single-electron excitation energies are relevant. We use the Rayleigh-Schrodinger approximation which relates the total energy to the single-particle energies via [76]:

17/k)

^

^-^

^

The Rayleigh-Schrodinger model states that for a many-particle system with n electrons, the energy of the single-electron state can approximately be taken to be the whole change in the total energy of the many-particle system when an electron in the corresponding state is added or removed. The exchange part of the self-energy is obtained from Eqs. (17.7)(17.18) and (17.27) as [5,6,73,74]

VM>f + "' 2

^

Va.-[^(k')-

Optoelectronic Devices: III Nitrides

Optoelectronic Devices

Semiconductor optoelectronic devices

Optoelectronic Devices and Properties

Optoelectronic Devices: Design, Modeling, and Simulation

Optoelectronic Devices: Design, Modeling, and Simulation

Fundamentals of Guided-Wave Optoelectronic Devices

Optoelectronic Devices: Advanced Simulation and Analysis

Optoelectronic Devices: Advanced Simulation and Analysis

Optoelectronic Devices: Advanced Simulation and Analysis

Rare-Earth Doped III-Nitrides for Optoelectronic and Spintronic Applications (Topics in Applied Physics)

Fundamentals of Guided-Wave Optoelectronic Devices

III-nitride devices and nanoengineering

Optoelectronic Sensors

Optoelectronic Sensors

Semiconductor Optoelectronic Devices: Introduction to Physics and Simulation

Semiconductor Nanostructures for Optoelectronic Devices: Processing, Characterization and Applications

Physics and Applications of Dilute Nitrides (Optoelectronic Properities of Semiconductors and Superlattices, 21)

Devices

Semiconductor Nanostructures for Optoelectronic Applications

Semiconductor Nanostructures for Optoelectronic Applications

Handbook of Refractory Carbides and Nitrides

Iii

Semiconductor Nanostructures for Optoelectronic Applications (Artech House Semiconductor Materials and Devices Library)

Physics of Optoelectronic Devices (Wiley Series in Pure and Applied Optics)

Chemistry of Transition Metal Carbides and Nitrides

Infernal Devices

Magnificent Devices

Nanoscale Devices

ULSI Devices

Photonic Devices

Optoelectronic Devices: III Nitrides