Citation
Effects of Dislocations on Electronic Properties of III-Nitride Materials

Material Information

Title:
Effects of Dislocations on Electronic Properties of III-Nitride Materials
Creator:
WEST, ALLEN M. ( Author, Primary )
Copyright Date:
2008

Subjects

Subjects / Keywords:
Carbon ( jstor )
Crystals ( jstor )
Density ( jstor )
Electronics ( jstor )
Gallium ( jstor )
Impurities ( jstor )
Nitrides ( jstor )
Nitrogen ( jstor )
Oxygen ( jstor )
Solubility ( jstor )

Record Information

Source Institution:
University of Florida
Holding Location:
University of Florida
Rights Management:
Copyright Allen M. West. Permission granted to University of Florida to digitize and display this item for non-profit research and educational purposes. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder.
Embargo Date:
5/1/2005
Resource Identifier:
436098662 ( OCLC )

Downloads

This item has the following downloads:

west_a ( .pdf )

west_a_Page_039.txt

west_a_Page_018.txt

west_a_Page_036.txt

west_a_Page_032.txt

west_a_Page_010.txt

west_a_Page_050.txt

west_a_Page_008.txt

west_a_Page_101.txt

west_a_Page_098.txt

west_a_Page_065.txt

west_a_Page_107.txt

west_a_Page_009.txt

west_a_Page_012.txt

west_a_Page_033.txt

west_a_Page_108.txt

west_a_Page_066.txt

west_a_Page_088.txt

west_a_Page_049.txt

west_a_Page_100.txt

west_a_Page_061.txt

west_a_Page_097.txt

west_a_Page_038.txt

west_a_Page_046.txt

west_a_Page_096.txt

west_a_Page_089.txt

west_a_Page_076.txt

west_a_Page_015.txt

west_a_Page_102.txt

west_a_Page_043.txt

west_a_Page_059.txt

west_a_Page_085.txt

west_a_Page_056.txt

west_a_Page_114.txt

west_a_Page_116.txt

west_a_Page_054.txt

west_a_Page_019.txt

west_a_Page_084.txt

west_a_Page_029.txt

west_a_Page_103.txt

west_a_Page_028.txt

west_a_Page_113.txt

west_a_Page_013.txt

west_a_Page_104.txt

west_a_Page_037.txt

west_a_Page_087.txt

west_a_Page_040.txt

west_a_Page_086.txt

west_a_Page_069.txt

west_a_Page_090.txt

west_a_Page_111.txt

west_a_Page_042.txt

west_a_Page_005.txt

west_a_Page_110.txt

west_a_Page_109.txt

west_a_Page_023.txt

west_a_Page_053.txt

west_a_Page_074.txt

west_a_Page_016.txt

west_a_Page_072.txt

west_a_Page_057.txt

west_a_Page_031.txt

west_a_Page_067.txt

west_a_Page_026.txt

west_a_Page_064.txt

west_a_Page_006.txt

west_a_Page_004.txt

west_a_Page_022.txt

west_a_Page_112.txt

west_a_Page_060.txt

west_a_Page_073.txt

west_a_Page_099.txt

west_a_Page_020.txt

west_a_Page_048.txt

west_a_Page_041.txt

west_a_Page_052.txt

west_a_Page_014.txt

west_a_Page_025.txt

west_a_Page_027.txt

west_a_Page_017.txt

west_a_Page_001.txt

west_a_Page_091.txt

west_a_Page_078.txt

west_a_Page_106.txt

west_a_Page_093.txt

west_a_Page_071.txt

west_a_Page_080.txt

west_a_Page_077.txt

west_a_Page_075.txt

west_a_Page_063.txt

west_a_Page_094.txt

west_a_Page_035.txt

west_a_Page_047.txt

west_a_Page_058.txt

west_a_Page_105.txt

west_a_Page_003.txt

west_a_Page_034.txt

west_a_Page_117.txt

west_a_Page_083.txt

west_a_Page_095.txt

west_a_Page_068.txt

west_a_Page_092.txt

west_a_Page_002.txt

west_a_pdf.txt

west_a_Page_079.txt

west_a_Page_045.txt

west_a_Page_082.txt

west_a_Page_062.txt

west_a_Page_055.txt

west_a_Page_021.txt

west_a_Page_081.txt

west_a_Page_051.txt

west_a_Page_030.txt

west_a_Page_070.txt

west_a_Page_007.txt

west_a_Page_011.txt

west_a_Page_044.txt

west_a_Page_115.txt

west_a_Page_024.txt


Full Text












EFFECTS OF DISLOCATIONS ON ELECTRONIC PROPERTIES OF III-NITRIDE
MATERIALS















By

ALLEN M. WEST


A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY

UNIVERSITY OF FLORIDA


2005




























"I hereby dedicate this [dissertation] to ... myself."

(credit to Rodney Dangerfield as Thornton Melon, Back to School)















ACKNOWLEDGMENTS

I would first like to thank Dr. Cammy Abernathy for all of her guidance and

support throughout my tenure in the Materials Science Department. She was the ideal

advisor for me, and was crucial to my success. I thank Dr. Stephen Pearton, Dr. Fan Ren,

and Dr. David Norton as well for serving on my committee. I especially thank

Dr. Crisalle for serving as my external committee member from chemical engineering.

I credit Dr. Crisalle for his work as Graduate Admissions Coordinator for the Chemical

Engineering Department. If not for Dr. Crisalle, I would never have ended up at UF.

I thank Dr. Karen Waldrip for all of her support and guidance during my time at

Sandia. Many of my successes in graduate school were due, in large part, to Karen's

tireless efforts. I thank Dr. Stephen Lee and Dr. Daniel Koleske for their significant

contributions to my research efforts. I further acknowledge Dr. Andrew Allerman for his

thorough feedback and instilling strong sense of attention to details in me.

I acknowledge Dr. Brent Gila for his contributions to the work in this dissertation,

and I thank Dr. Andrea Onstine and Danielle Stodilka for their support and collaboration

for various classes and qualifying exams. I especially thank Dr. Jerry Thaler for being a

reliable liaison in Florida while trying to complete the bureaucratic requirements for

graduation from New Mexico. I gratefully acknowledge Jerry's efforts in this regard.

I also thank Jerry, Dr. Omar Bchir, Jeff Sharp, Brandy Colwell, and Sarah Rich for their

camaraderie throughout my graduate school experience.









I acknowledge Dr. Sam Kozak for the inspiration and advice he provided during

my undergraduate days. I also thank Dr. Ken Van Ness and Dr. Erich Uffelman for

serving as my advisors at Washington and Lee. Both were very important to my

intellectual development. I also thank Dr. Douglas Szajda, Dr. Marcia France, and

Dr. Steven Desjardins. All were superb professors.

Dating back to high school, I must acknowledge Mr. James Morrison for piquing

my interest in science and tolerating me for three years. I also acknowledge Mr. Robert

Bulkeley for doing an outstanding job of advising me throughout high school.

Finally, and most importantly, I thank my family for their undying support

throughout my graduate school experience.
















TABLE OF CONTENTS

page

A C K N O W L E D G M E N T S ......... .................................................................................... iii

LIST OF FIGURES ............. ................... ............ .......... ............... .. viii

ABSTRACT .............. .......................................... xi

CHAPTER

1 IN TRODU CTION .................................... .......... .... ............. .

Background and M otivation ...................................................... .......................
Electronic Properties of III-N itrides ............. ............................................................
Chemical, Structural, and Physical Properties of III-Nitrides......................................2
Suitability of III-Nitrides in Optoelectronic Devices ............................................
Suitability of III-Nitrides in Electronic Devices........................................................3
Progress in III-Nitride Material Development.................. ...........................5
Availability of III-N itride Based D evices................................. ...................... 6
State ent of the P problem ............................................................................. ........ 6
O u tlin e ............................................................................ . 7

2 EXPERIM ENTAL M ETHOD S........................................... ........................... 10

G row th Techniques................. ... ...................... .. ............ .... .............. 10
Metalorganic Chemical Vapor Deposition.................. ......................................10
M olecular B eam Epitaxy ......................................................... ............. 14
G allium N itride Bulk G row th ................................... ........................... .. ........ 15
C haracterization T techniques ........................................................... .....................16
High-Resolution X-Ray Diffraction.....................................................16
Transmission Electron Microscopy ........................... ........ ...............17
Secondary Ion Mass Spectroscopy ........... ................................. ...............18
Superconducting Quantum Interference Device Magnetometry .........................18
Capacitance-Voltage .................... ........................................ .. ............... 19
Inductively-Coupled Plasma-Mass Spectroscopy ............................................19
Scanning Electron M icroscopy....................................... ......................... 20
Electron Dispersive X-Ray Spectroscopy ................................. ............... 20
A uger Electron Spectroscopy ................................... .......................... .. ......... 21




v









3 DETERMINATION OF DISLOCATION DENSITY OF III-NITRIDE FILMS BY
X -R A Y D IF FR A C T IO N ................................................................... ....................23

In tro du ctio n ......................................................... ............. ................ 2 3
Characterization of D islocations ........................................ ....................... 23
Previous W ork ............. .... ............ .................................. ....... 24
Tw ist ............... ......... ....... ..... .... ... ............ ........... 25
Incoherence ............... .......... ..... ........... .......... .......... 26
F o c u s ............................................................................................................... 2 6
E xperim ental M ethods......................................................................... ..................27
M eth od ............................................. ........................................... ..........................2 9
S ou rces of B roadening ............................................................... ....................2 9
M o d el .............. .......... .. ...................................................... ... .3 1
Determination of Convolution......................... ............................. 32
Calculation of Dislocation Density ............... ....... .. ..................................33
Simplified Model .............. .................. .. ... ................ 34
R results and D discussion ........................... ...... ... .. ...... .............. 35
S u m m a ry ........................................................................................................3 7

4 EFFECT OF DISLOCATIONS ON IMPURITY INCORPORATION IN
METALORGANIC CHEMICAL VAPOR DEPOSITION GROWTH...................54

Introduction ...................................................................................... .. ... .... ..54
S ilic o n .......................................................................................5 4
O x y g e n ................................................................5 5
C a rb o n ............................................................................................................ 5 6
Previous W ork ..................................... ....................... 57
F o c u s ............................................................................................................... 5 8
Experim ental M ethods................................................. 58
R esu lts an d D iscu ssion .......................................................................................... 59
S u m m a ry ......................................................................................................6 1

5 GALLIUM NITRIDE BULK CRYSTAL GROWTH BY DISSOLUTION AND
RECRYSTALLIZATION OF GALLIUM NITRIDE POWDER ..............72

Introdu action .............. ................. .................................................................. 72
Previous Work ................ ......... ..................72
P proposed M methods ...............................................................74
Electroplating ................................................. 74
Electrochemical Reduction of Nitrogen Gas .....................................................75
Dissolution and Recrystallization of Gallium Nitride Crystals ...........................75
F o c u s ...................................................................................................................... 7 6
Experim ental M ethods................................................. 76









R results and D discussion ............................... ........................................ .. ..........77
Dissolution of Gallium Nitride in Lithium Chloride ...................................77
Gallium Nitride Recrystallization................... ....... ........................... 78
Summary ......... ........ ............ ............. ............... 80

6 SUMMARY AND FUTURE DIRECTIONS............... .................................97

L IST O F R E F E R E N C E S ............. .............................................................................. 10 1

B IO G R A PH ICA L SK ETCH ........... ..................................................... .....................105
















LIST OF FIGURES


Figure p

1-1 Band gap engineering diagram for III-V materials ..........................................8

2-1 Schematic of metalorganic chemical vapor deposition operation..........................22

3-1 Schem atic of reciprocal lattice points probed ............ ................... ..................39

3-2 Adjacent domains tilted with respect to one other ...........................................40

3-3 Adjacent domains twisted with respect to one another..........................................41

3-4 Cross-section transmission electron microscopy image of a low dislocation density
sam p le ......................................................... .................. 4 2

3-5 Cross-section transmission electron microscopy image of a high dislocation density
sam p le ......................................................... .................. 4 3

3-6 Weak beam transmission electron microscopy image and convergent electron beam
diffraction im age ...................... .................... .. .. ........... .... ....... 44

3-7 Broadening effects demonstrated in reciprocal space.................. ................45

3-8 Plotted data of full width at half maximum vs. reciprocal lattice vector magnitude46

3-9 Fit of full width at half maximum data vs. angle of inclination .............................47

3-10 Mathematical and geometric derivation of the random distribution method for
calculating dislocation density ........................................... .......................... 48

3-11 Calculation of dislocation density based on values for coherence length ..............49

3-12 Comparison of methods of determining dislocation density by XRD ..................50

3-13 Dislocation density determined by transmission electron microscopy vs. dislocation
density determined by x-ray diffraction.... ........... ....................................... 51

3-14 Dependence of dislocation density on growth pressure...........................................52

3-15 Effect of dislocation density on room temperature mobility of gallium nitride high
electron m obility transistors ......................................................... ............. 53









4-1 Impurity profiles for low dislocation density at the top ...........................................62

4-2 Impurity profiles for mid dislocation density at the top ...........................................63

4-3 Impurity profiles for high dislocation density at the top........................................64

4-4 Impurity profiles for high dislocation density at the flat.......................................65

4-5 Impurity profiles for second series of overgrowths for low dislocation density at the
flat.............. .................... ..................................... ......... ...... 6 6

4-6 Impurity profiles for second series of overgrowths for mid dislocation density at the
flat.............. .................... ..................................... ......... ...... 6 7

4-7 Impurity profiles for second series of overgrowths for high dislocation density at
th e fla t .......................................................................... 6 8

4-8 Comparison of impurity profiles for different dislocation density templates ..........69

5-1 Gallium nitride crystal with dimensions of 0.9 mm x 0.6 mm .............................82

5-2 Ideal solubility of gallium nitride ......................................................... ..................83

5-3 Experimentally determined solubility of gallium nitride in lithium chloride
com pared to the ideal solubility curve ........................................ ............... 84

5-4 Experimentally determined solubility determined after employing a 2 |tm filter....85

5-5 Scanning electron microscopy image of a bare gallium nitride surface...................86

5-6. Scanning electron microscopy image of a gallium nitride surface exposed to
gallium nitride-in-lithium chloride melt........... ............................ .............87

5-7 Close-up scanning electron microscopy image of a gallium nitride crystallite .......88

5-8 Scanning electron microscopy image of a treated gallium nitride surface after
exposure to the gallium nitride-in-lithium chloride melt ......................................89

5-9 Scanning electron microscopy image of an untreated gallium nitride surface after
exposure to a gallium nitride-in-lithium chloride melt ........................................90

5-10 Scanning electron microscopy image focusing on needles on the untreated gallium
nitride surface at 750x ........................ .............. .............................91

5-11 Scanning electron microscopy image focusing on needles of the untreated gallium
nitride surface at 1900 x ............................................................................. .... 92

5-12 Election dispersive spectroscopy images for elemental analysis on needle
stru ctu res ............................................................................ 9 3









5-13 Electron dispersive spectroscopy images of a patch on the gallium nitride surface 94

5-14 Scanning electron microscopy image of a deposited gallium nitride crystallite with
a thickness of 5 m ............. .... ...................................................... ......... ...... 95

5-15 Electron dispersive spectroscopy instrument depth calibration ............................96















Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy

EFFECTS OF DISLOCATIONS ON ELECTRONIC PROPERTIES OF III-NITRIDE
MATERIALS

By

Allen M. West

May 2005

Chair: Cammy Abernathy
Major Department: Materials Science and Engineering

By varying metalorganic chemical vapor deposition (MOCVD) growth conditions,

III-nitride samples with dislocation densities ranging over two orders of magnitude were

grown, and the effects of dislocations were studied. Initially, a novel technique to

determine dislocation density by high-resolution x-ray diffraction (HRXRD) was

developed. Results were compared to those obtained by transmission electron

microscopy (TEM) and agreed within a factor of 1.5. This HRXRD-based technique

provided rapid and accurate feedback for use in growth-optimization studies; and became

a basis for future studies of the effects of dislocation on specific phenomena associated

with III-nitride technology. This included a study of the effect of dislocations on

impurity incorporation during MOCVD growth. Dislocations were previously

conjectured to provide energetically favorable sites for impurity incorporation, but

experimental results did not confirm this.









Finally, a novel growth technique was investigated for gallium nitride (GaN) bulk

crystals, based on a simple dissolution/recrystallization process. GaN was successfully

dissolved in a molten LiCl medium, and then recrystallized on a nominally homoepitaxial

MOCVD-grown GaN template. This provides the basis for a potentially viable method to

fabricate high quality GaN bulk crystals for homoepitaxial growth.














CHAPTER 1
INTRODUCTION

Background and Motivation

Gallium nitride (GaN) and III-nitride alloys (Al-, Ga-, In-N) demonstrate

outstanding potential to significantly advance solid-state electronic and optoelectronic

technologies. Currently, solid-state electronics comprise a multi-hundred billion-dollar

industry, and consist primarily of silicon (Si) and gallium arsenide (GaAs)-based devices.

These materials, however, are intrinsically limited for certain applications (namely,

high-power electronics and low-wavelength optoelectronics). Such applications have

far-reaching implications, with significant promise for commercial, military, and national

security purposes. III-nitride-based materials are ideal for these applications because of

their favorable intrinsic electronic, structural, and chemical characteristics. Because of

these properties, GaN and related alloys have recently been subject to intense research;

and scientists have made significant progress in advancing III-nitride technology.

Electronic Properties of III-Nitrides

Electronically, III-nitrides have a large, direct, tunable band gap, which makes

these materials ideally suited for many technologically important electronic and

optoelectronic applications. The band gap for III-nitrides ranges from 0.7 eV for indium

nitride (InN) [1] to 6.2 eV for aluminum nitride (A1N) [2]. Alloys throughout this entire

range are possible to achieve, depending on the composition of group III element (Al, Ga,

In). III-nitride materials' large band gap gives them distinct capabilities for use in

high-power electronic and optoelectronic applications. The tunable nature of the band









gap in III-nitrides provides the possibility for large band offsets, which enables greater

control over carrier confinement, and better performance in some devices, including

HEMTs and multi-quantum well (MQW) LEDs.

Chemical, Structural, and Physical Properties of III-Nitrides

III-nitrides offer distinct advantages because of their chemical, structural, and

physical properties. They are outstandingly hard, corrosion-resistant in most

environments, and remarkably stable at high temperatures. III-nitrides exhibit the

wurtzite crystal structure, which provides a significant spontaneous polarization field

along the c-axis, and a high piezoelectric polarization field [3-5]. Together, these fields

enable greater sheet charge densities in III-nitride heterostructures strained in

compression; thus, allowing greater capabilities for III-nitride based electronic devices.

III-nitride materials are also significantly more thermally conductive than their

III-arsenide counterparts and silicon, allowing for efficient heat removal. This saves

substantial resources by reducing the need for external cooling. Silicon-based devices

burn out at temperatures exceeding 1400C, necessitating the installation of powerful

cooling fans for silicon-based devices, for high-power applications. GaN-based devices

are operable in environments >3000C [3], so the high temperature tolerance and the

efficient heat removal of III-nitrides make this material system ideal for

high-temperature, high power applications.

Suitability of III-Nitrides in Optoelectronic Devices

III-nitrides represent one of few known material systems that allow efficient

emission of radiation in the green, blue, and deep ultraviolet range. Figure 1-1 shows the

band gap dependence on the lattice constant, and the range of achievable wavelengths for









III-nitride emitters. III-nitride LEDs and LDs are exceptionally tolerant of defects, as

they show efficient emission of radiation despite high dislocation densities (p = 10-1011

cm-2).

Because of the unique capabilities of III-nitride optoelectronic devices, they are

ideal for commercial, military, and national-defense applications. Commercially,

monochromatic III-nitride LEDs are used for illumination purposes in traffic signals and

outdoor displays. One monochromatic traffic light saves 800 kWh/year in power,

compared to a typical incandescent white bulb with a monochromatic shutter [6].

III-nitride-based UV VCSELs are also being developed to pump red-green-blue (RGB)

phosphors for significantly more efficient and cost-effective white-lighting purposes.

This could provide energy savings estimated at $100 billion annually. Furthermore, blue

LDs are used commercially low-cost, ultra-high-density optical data storage in digital

versatile disks (DVDs) [7] that can store >1 gigabit/cm2 of information [6]. Militarily,

III-nitride photodetectors are currently being researched for use in solar blind detectors

for missile warning systems and for malignant biological and chemical agent detection

[8,9].

Suitability of III-Nitrides in Electronic Devices

Table 1-1 shows critical electronic properties of several semiconductor materials.

The intrinsic electronic properties of GaN and AlGaN/GaN heterostructures allow

significantly more power output than any other known semiconductor material system.

The large, tunable band gap of III-nitrides makes these materials ideal for use in certain

electronic devices, such as high electron mobility transistors (HEMTs). Breakdown

voltage (Vb) goes as the square of the band-gap energy, giving GaN and AIN-based









electronic devices a distinct advantage over SiC (band gap, Eg = 3.2 eV) in terms of

power switching, enabling a device to achieve large power density at high frequencies

[3,7]. The theoretical maximum power output achievable is directly proportional to the

band gap to the fourth power (Pmax oc Eg4), and the saturation velocity squared (Pmax oc

Vs2). Because of the large band gap and saturation velocity, III-nitride based devices are

ideal for high-power applications.

For a HEMT device, the output power is shown in Equation 1-1, where Imax is the

maximum current possible for a particular device.


out b max (1-1)
8

Higher saturation currents are achievable with higher carrier concentrations, and the

polarization effects of AlGaN/GaN heterostructures provide outstandingly high sheet

densities in a two-dimensional electron gas (2DEG). Furthermore, the polarization fields

in GaN HEMT structures remove the need for external doping to achieve very high sheet

densities. This ability to operate without ionized impurities diminishes the problem of

gate leakage in Schottky diodes [5]. Finally, for operable high-power electronic devices,

heat dissipation is crucial, since such devices are thermally limited. The high thermal

conductivity of GaN enables these HEMT devices to operate more efficiently and more

cost-effectively [3].

These characteristics make III-nitrides ideal for use in high-power devices for

both commercial and military applications. Commercially, GaN-based HEMTs can be

used in cellular communications base stations, telecommunications, hybrid electric cars,

and switches for high-power electric grids. For military purposes, GaN-based HEMTs

are candidates as amplifiers for synthetic aperture radar (SAR), smart weapons, electronic









warfare, and surveillance. They will eventually replace bulky traveling wave tubes,

enabling significantly higher output power, with added benefits of reduced mass, volume,

cost, and cooling requirements.

Progress in III-Nitride Material Development

Performance of III-nitride based devices has improved by leaps and bounds in

recent years due to growth and processing of these materials becoming more mature. As

growth techniques have progressed, microstructural quality of III-nitride films used in

devices has improved; and performance of both electronic [10] and optoelectronic

devices has improved accordingly [11]. Optoelectronically, the first nitride-based blue

LEDs were fabricated in the 1970s, when the external efficiency hardly exceeded 0.1%.

External efficiencies improved with novel growth breakthroughs throughout the 1980s,

and reached 1.5% by 1992. As growth of III-nitride films continued to mature, this value

reached 20% by 2002 [6]. Further strides with III-nitride optoelectronic devices will be

made as microstructural quality of III-nitride films continues to improve. GaN-based

electronic devices have experienced similar improvements as III-nitride technology has

become more advanced. The earliest reported GaN HEMT state-of-the-art was 1.1

W/mm in 1995, and exceeded 10 W/mm by 2001 [10]. Since 2001, GaN-based

technology has progressed rapidly, and the current state-of-the-art GaN HEMT power

density is an astounding 32.2 W/mm [12]. Conversely, GaAs-based HEMTs have

attained merely 1 W/mm for power density, despite the fact that GaAs-based technology

is significantly better studied.









Availability of III-Nitride Based Devices

Considering these substantial advantages and the potential widespread use of

GaN-based devices, commercial devices based on GaN are currently in the research

stage. Comprehensive studies of III-nitride materials are underway, in order to better

exploit these materials' advantages, where considerable growth and processing issues are

being addressed. The most significant of these issues involves the effects of dislocations

on III-nitride thin films. Because there are no commercially available substrates that

allow for homoepitaxial growth of III-nitride thin films, these devices must be grown

heteroepitaxially on sapphire or silicon carbide, at high lattice mismatch. This causes

extremely high dislocation densities (p=10-1011 cm-2).

Statement of the Problem

Currently, there is much disagreement concerning the exact effects of dislocations

on III-nitride electrical properties and device performance. To advance the science of

III-nitrides, and consequently III-nitride based device performance, the precise effects of

dislocations on electronic properties must be determined. In other semiconductor

material systems (such as Si and GaAs), a dislocation density of 104 cm-2 renders a

device inoperable. III-nitride films with dislocation densities of-1011 cm2 have been

used in operable devices with surprisingly high efficiency. Because of this high tolerance

for defects, some researchers have postulated that dislocations are electrically inactive in

III-nitride materials [13]. Others, however, have indicated that dislocations have a

deleterious effect on device performance [14-17], and will significantly affect reliability

and scalability of these devices. Determination of the effects of dislocation density on









electronic properties would greatly aid optimization of III-nitride device performance,

especially as long as these devices are grown heteroepitaxially.

Outline

The remainder of this dissertation is structured as follows: Chapter 2 describes the

growth and characterization techniques used in this work. Chapter 3 illustrates one

technique in depth, based on x-ray diffraction (XRD) to determine dislocation density in

III-nitride thin films. This method develops a theoretical model to determine dislocation

density from XRD results; and allows for rapid, accurate feedback for dislocation density.

This is extremely helpful in optimization studies for metalorganic chemical vapor

deposition (MOCVD) III-nitride growth studies. Chapter 4 describes a study on the

effect of dislocation density on impurity incorporation in GaN films during MOCVD

growth. Impurities can dramatically affect electronic properties of GaN, by forming deep

trap states and acting as scattering centers. Understanding how dislocations affect

impurity incorporation is critical in optimizing MOCVD growth parameters for device

structures. Chapter 5 describes an investigation into a novel method to fabricate III-

nitride bulk crystals. Bulk III-nitride substrates would allow homoepitaxial growth of

III-nitride thin films, which would minimize dislocation density for device structures.

This would greatly enhance performance for devices for which dislocations have a

deleterious effect. Finally, Chapter 6 summarizes significant results and recommends

future work for these projects.











b.U 0.2
AIN

5.0 e ANitrides
4.5 (blue/UV)
4.0 0.3
4.0 Phosphides 0
3 Ga (yellow/red)
M' 3.0 0.4 j
S2.5 p0.5
0,
S1. Arsenides
e 1.5fr (near-IR) b 3
1.0 GaAs 1.0
Antimonides GaSb
(far/mid-IR) InAs
0.0 1 I. I. ..L .. .L L L . .. 5.0
4.2 4.4 4. 4.8 5.0 5.2 5.4 5.6 5.8 6. 6. 6.4 6.
Lattice Constant (Angstroms)



Figure 1-1. Band gap engineering diagram for III-V materials. III-nitrides are ideal
candidates for use as green, blue, and deep-UV light emitters.









Table 1-1. Electronic properties of semiconductor materials used in electronic device
applications.
Semiconductor Material Si GaAs SiC GaN
Band Gap (eV) 1.1 1.4 3.3 3.4
Saturated Electron Velocity (x107 cm/s) 1 2.1 2 2.7
Breakdown Field (MV/cm) 0.3 0.4 2 3.3
Thermal Conductivity (W/cm-K) 1.5 0.5 4.5 >1.7
Bulk Electron Mobility (cm2/V-s) 1500 8500 700 1200
Heterostructure None AlGaAs/InGaAs None AlGaN/GaN
2DEG Sheet Density (x1012 cm-2) NA <4 NA 20














CHAPTER 2
EXPERIMENTAL METHODS

Chapter 1 showed the technological importance of III-nitride-based technology

and identified the fundamental problem of high dislocation density. Chapter 2 describes

the experimental procedures involved in growth and characterization of III-nitride

materials used in these studies. This includes a description of the two different

film-growth techniques, specifically metalorganic chemical vapor deposition (MOCVD)

and molecular beam epitaxy (MBE), followed by a description of a novel bulk-growth

method for GaN. The characterization techniques used are also explained.

Growth Techniques

Both MOCVD and MBE were used to accomplish different goals for growing

films in this work. Growth of III-nitride materials is relatively new, because of the lack

of III-nitride lattice-matched substrates and the difficulty associated with incorporation of

nitrogen. For MOCVD, ammonia (NH3) is the most prevalent precursor; but its stability

makes it difficult to incorporate nitrogen, and requires high growth temperatures (usually

-10500C). For MBE growth of III-nitrides, the nitrogen source is a reactive nitrogen

plasma, which consists of nitrogen gas and an electron cyclotron resonance (ECR)

plasma source.

Metalorganic Chemical Vapor Deposition

MOCVD offers the following advantages over MBE: high growth rates, high

throughput, ability to grow at high temperatures, external feeds for precursors, high









growth pressures, and lower costs associated with this technique over MBE. Higher

growth rates decrease the time required to perform material optimization studies, and

higher throughput is desirable when fabricating commercial devices, once the

material-optimization growth studies are complete. Growth at higher temperatures

enables more effective cracking of nitrogen precursor to incorporate nitrogen into the

film; which, in turn, results in higher growth rates. The external feed lines that provide

the precursors are advantageous, since they allow the growth chamber to be isolated

when replacing the sources on depletion. This is a significant advantage over MBE, since

replacing the solid sources requires venting the growth chamber to atmosphere. Baking

out the impurities from this exposure can take up to 2 weeks, whereas changing group III

gas bubblers or ammonia sources for MOCVD usually takes less than 1 hour.

The growth pressures associated with MOCVD are relatively high (usually 40 to

500 Torr for III-nitride growth) compared to those of MBE (<10-6 Torr). This removes

the need for ultra high vacuum (UHV) conditions, which require diffusion and cryogenic

pumps. Diffusion pumps are problematic because they are susceptible to back-streaming

oil into the growth chamber, which may incorporate into the film. Cryogenic pumps

require a constant flow of liquid nitrogen (L-N2), which can be expensive, depending on

the time and frequency of growth experiments. Finally, for all of the advantages that

MBE provides for devices in other material systems, MOCVD-grown III-nitride films for

electronic devices have demonstrated superior electronic properties and device

performance over their MBE grown counterparts [18].

The MOCVD growths were performed in an EMCORETM D-125 short jar reactor

with the wafers placed on a silicon carbide-coated graphite rotating platen, containing up









to three wafers. There are several advantages to this type of reactor. First, it allows for a

high degree of aluminum incorporation into films, since the short jar design significantly

reduces the parasitic pre-reactions of the aluminum precursor with ammonia due to

upstream heating. Second, the multi-wafer EMCORETM D-125 allows a greater throughput

of device-quality films once the growth conditions are optimized. Third, this reactor

allows for excellent thermal control and uniformity across the platen, and therefore across

the individual wafers. This is especially useful when growing a film that is temperature

sensitive. Fourth, the rotation rate of the platen can reach 2100 revolutions per minute

(rpm). This allows a high degree of control over the boundary layer thickness, which

could improve growth uniformity and growth rate. Growth uniformity can be monitored

in-situ, since the EMCORETM D-125 contains three view ports where it is possible to

monitor the film growth at different points across the wafers on the platen. Finally, this

reactor provides a load-lock, directly connected to a turbo-molecular pump (TMP)

backed by a dry-scroll rough pump. This allows the reaction chamber to remain under

vacuum during sample exchange, which reduces the amount of contamination during film

growth. Figure 2-1 is a schematic of the MOCVD operation.

The precursors used for the films are trimethyl gallium (TMG) and trimethyl

aluminum (TMA) for gallium and aluminum respectively, and ammonia (NH3) for

nitrogen, with hydrogen (H2) and nitrogen (N2) serving as carrier gases. The precursors

undergo the following reactions:

(CH3)3Ga(g) + NH3(g)4 GaN(s) + 3CH4(g)

(CH3)3Al(g) + NH3(g) + AIN(S) + 3CH4(g)









The solid III-nitride film deposits on the substrate, while the gaseous byproducts are

pulled through an EBARA rough pump, and out through the exhaust. In all likelihood,

the reactions rarely occur as shown above, as the reactions shown above are

oversimplified. Many side reactions and parasitic pre-reactions complicate the MOCVD

film deposition process; but overall, the reactions shown above are a useful

simplification.

The parameters that affect the growth of the film are temperature and pressure, as

well as flows of metalorganics, ammonia, and carrier gases. One challenge posed by the

MOCVD growth technique is determining which parameters) to vary, in order to grow

the film with the desired qualities. Since our goal was to determine the effects of

dislocations on electronic properties and device performance for III-nitride materials, it

was necessary to grow films over a wide range of dislocation densities by MOCVD. The

MOCVD growth parameters that were varied include the type of nucleation layer (A1N

vs. GaN), growth temperature, and growth pressure, in order to grow films of varying

dislocation density. To further expand this study, two different substrates were used:

silicon carbide (SiC) and sapphire (A1203). SiC has a less severe lattice mismatch than

sapphire with III-nitrides (3.5% vs. 16% for GaN and 13.29% vs. 1% for A1N) [19-21];

and therefore, enables experimenters to grow lower-dislocation-density-films. Sapphire

is more widely used, because it is significantly less expensive; insulating sapphire wafers

are $60 each, while device-grade SiC wafers are $5000 each.

For growths on SiC, a 1,400 A high-temperature (11250C) A1N nucleation layer

was first deposited, followed by a GaN buffer layer at 10500C. The pressure for these

growths varied from 100-200 Torr. For growths on sapphire, more variables were









explored. For some films, a low-temperature (6000C) A1N nucleation layer was first

deposited; while others had a low temperature (530C) GaN nucleation layer, depending

on the desired dislocation density. The pressure of the overlying GaN layers varied from

20-500 Torr, and the temperature of this layer varied from 1040-10600C. Another factor

that affected the film microstructure was recovery time. A fast recovery provided a film

with a higher dislocation density, while a longer delay between the growth of the

nucleation layer and the GaN over-layer provided a film with a higher microstructural

quality [22].

By altering the growth conditions as needed, it was possible to grow GaN films

with a very broad range of dislocation densities. The material grown for our study ranged

from 4.21x10 cm-2 1.07x1011 cm-2. This allowed for thorough studies of the effects of

microstructure on electronic and magnetic properties, as well as device performance and

impurity incorporation.

Molecular Beam Epitaxy

Molecular beam epitaxy (MBE) is advantageous when abrupt interfaces and

precise control of layer thickness are required. MBE is also an inherently safer process

when growing III-nitride films, since all precursors used in MOCVD are either toxic or

pyrophilic; whereas none of the precursors used in MBE present such hazards. This

removes the need for strict engineering safety controls and gas sensors that are required

for MOCVD.

For our study, MBE offered a distinct advantage when growing GaN films with

transition metal dopants and when growing oxide films. MBE is better suited for

transition-metal dopant growths, because of the difficulty associated with obtaining an









MOCVD transition-metal source that can incorporate efficiently into a III-nitride film

and that has a high enough vapor pressure to provide a substantial molar flux in the

carrier gas. MBE is also better for oxide growths, because of the lack of oxygen sources

for MOCVD.

For the MBE growths in our study, two series of films were grown: chromium

(Cr)-doped aluminum nitride A1N and scandium oxide (Sc203). For the Cr-doped A1N

films, the group III precursor was solid 99.99999% pure (7N) aluminum metal, the

dopant source was 5N solid Cr, and the nitrogen precursor was 6N nitrogen gas. To

create reactive nitrogen radicals, the nitrogen gas was ionized by a radiofrequency (RF)

plasma unit with a frequency of 13.56 MHz. For the scandium oxide growths, the

precursors were solid 4N scandium metal and 6N oxygen gas, which was ionized at 13.56

MHz. All of the solid sources were evaporated by resistively heating poly-boronitride

(PBN) containers, and the atomic fluxes were controlled by adjusting the shutter

apertures.

Gallium Nitride Bulk Growth

The GaN bulk growth studies were performed in a home-built stainless-steel

container, fitted for a 1" diameter quartz tube. The method to grow GaN bulk crystals

was simply a dissolution/recrystallization process of polycrystalline GaN powder in a

molten lithium chloride (LiC1) solvent. The chamber was resistively heated up to

-1000C to effectively dissolve GaN powder and maximize deposition of GaN

crystallites.

Procedurally, the solubility was tested with respect to temperature. Once it was

determined that GaN dissolved sufficiently well at achievable temperatures, the









experiment was repeated to test the recrystallization of GaN on a nominally

homoepitaxial template of MOCVD-grown GaN on sapphire substrate. Other effects

such as convective fluid dynamics and surface treatment were tested to maximize growth

on the surface.

Characterization Techniques

After the MOCVD, MBE, and bulk growths, samples were removed from their

respective growth chambers; and their properties were assessed using several techniques.

Specifically, the microstructural, magnetic, and electronic characteristics were

determined. All of the techniques used are described in the subsections to follow.

High-Resolution X-Ray Diffraction

High-Resolution X-Ray Diffraction (HRXRD) is a powerful technique that

provides information about a material's microstructure and lattice constant. Our study

used only rocking curves (co-scans) to ascertain the microstructure of the GaN thin films.

The microstructural quality of a thin film is closely related to the width of the diffracted

x-ray intensity, with the full-width at half maximum (FWHM) used as the figure of merit.

Since HRXRD is a diffraction phenomenon, its measurements must be taken with respect

to reciprocal space, where reciprocal lattice points (RLPs) correspond to specific lattice

planes. If it were possible to perform HRXRD analysis on a defect-free crystal of infinite

dimensions, the results would be a series of delta functions at every RLP. Once finite

dimensions and defects are imposed on the crystal, broadening can be observed and

measured by HRXRD. The extent to which broadening is observed is indicative of the

microstructural quality; the higher the measured FWHM, the higher the dislocation

density.









All of the HRXRD measurements were made using a Philips (Almelo, Holland)

X'Pert MRD high-resolution triple-axis x-ray diffractometer with a sealed Cu anode

collimated by a four-crystal Ge(220) monochromator. The Cu anode produced an x-ray

wavelength of k(Cuk~1)=1.5404 A.

Transmission Electron Microscopy

Transmission Electron Microscopy (TEM) is a universally accepted method to

determine dislocation density in materials. TEM operates by emitting a focused electron

beam through a sample, into a series of magnetic lenses, and into a fluorescent screen

detector. It is a robust technique for determining microstructural characteristics of

materials, since it provides high spacial resolution (with a magnification of 106x). Using

this technique, dislocation density is simply computed as the number of dislocations

counted over a known measured area. TEM is a destructive technique, since sample

preparation requires a thickness that is transparent to the electron beam (<3000 A).

The microscopy was performed with a Philips CM20 instrument at 200kV. Some

specimens were prepared in plan-view while others were prepared in cross section. The

samples observed in cross-section were prepared by a focused ion beam; while those

observed in plan-view were prepared by a traditional mechanical polishing, dimple, and

ion mill procedure.









Secondary Ion Mass Spectroscopy

Secondary Ion Mass Spectroscopy (SIMS) is a superior method to determine

impurity concentrations with respect to film depth. It is a destructive technique, but is

known to provide accurate results for impurity profiles. SIMS results are obtained by

accelerating heavy metal ions at a target sample, sputtering off constituent atoms. These

atoms from the target sample become ionized, and are attracted to a charged detector.

The SIMS work in our study was performed in a PHI Quadrupole SIMS

spectrometer (Charles Evans and Associates, Cambridge, MA). The sputtering ion was

cesium (Cs ), with a primary ion energy of 4keV. The cesium angle of incidence was

600, and the instrument was optimized for sensitivity for carbon, oxygen, and silicon.

Detection limits were 1x1016 cm-3 for carbon, 5x1015 cm-3 for oxygen, and 6x1015 cm-3

for silicon.

Superconducting Quantum Interference Device Magnetometry

A Superconducting Quantum Interference Device Magnetometry (SQUID)

Quantum Design Magnetic Properties Measurement System was used to ascertain

magnetic properties of DMS materials. Hysteresis loops from magnetization vs. field

measurements verify that a sample is ferromagnetic. The second technique to determine

magnetic properties was Field-Cooled/Zero Field-Cooled (FC/ZFC) measurement. This

technique involves cooling a sample to 10K under an applied field while the sample's

magnetization is measured (FC), and again while the sample's temperature increases to

room temperature (ZFC).












Capacitance-Voltage

Capacitance-voltage (CV) is a non-destructive technique that provides

information about a film's electrical properties. In our study, a mercury (Hg) contact CV

was used to probe for a buried conductive layer in a nominally insulating GaN film, and

to confirm the existence of a conductive 2DEG in a AlGaN/GaN HEMT structure. The

depletion region of a HEMT structure is modeled as a parallel plate capacitor, where the

mercury droplet acts as a metal Schottky contact, and the sheet density of a 2DEG is

given as


N(x)-C3 (2-1)
qsA' AVJ


where C is the capacitance, q is the carrier charge, ; is the permittivity, A is the area of

the mercury drop, and Vr is the reverse bias voltage. The apparatus used for

experimentation was an MSI Electronic Model 412-2L mercury probe, with a droplet

diameter of 2.03x103 cm-2. Electronic measurements were obtained by a Hewlett

Packard 4284A Precision Inductance-Capacitance-Resistance meter, operating at 1 MHz.

Inductively Coupled Plasma-Mass Spectroscopy

Inductively coupled plasma-mass spectroscopy (ICP-MS) is a powerful technique

to determine the existence and concentrations of trace elements. This technique operates

by injecting an aerosolized liquid sample into an argon plasma at a temperature of 7000

K. When the atoms from the sample collide with the energetic argon ions, the become

charged. The ions are then drawn into a high vacuum chamber with a quadrupole mass

analyzer. The quadrupole filters out atoms of undesired masses by alternating









radiofrequency (RF) and direct current (DC) fields, so that only atoms of the desired

mass can pass into the detector. The detector is calibrated by correlating the intensity of

the signal created by these collisions with the detector to a set of standards with known

concentrations. In this study, a Perkin Elmer (Boston, MA) Elan 6100 instrument was

used to determine the concentration of gallium and lithium in samples of GaN dissolved

in LiCl.

Scanning Electron Microscopy

Scanning electron microscopy (SEM) is a useful technique to determine

morphological information on a surface at extremely high magnifications (10-300,000x).

SEM operates by emitting an electron beam at the sample surface in a vacuum. When

this electron beam hits the surface, secondary electrons are emitted and collected in a

detector, which forms an image, giving precise topographical information. The

instrument used was a JEOL (Peabody, MA) JSM5800 tungsten filament scanning

electron microscope. The electron beam energies ranged from 5-20 keV.

Electron Dispersive X-Ray Spectroscopy

Electron dispersive x-ray spectroscopy (EDS) is used in conjunction with SEM to

provide information about elemental analysis on surface features of a sample. Other than

emitting secondary electrons when exposed to an incident electron beam, a sample will

also emit x-rays. This is because, when inner shell electrons are ejected from an atom,

electrons at high energy shells will fill the inner shell vacancy, which emits a photon.

The energy of the photon emitted is related to the element. EDS measures the energy and

intensity of the emitted radiation, which provides qualitative and quantitative analysis of









surface impurities. The EDS data in this study was obtained using an EDAX Phoenix

system, fitted with a 3,000 A window, allowing for analysis of low-Z elements.

Auger Electron Spectroscopy

Auger Electron Spectroscopy (AES) was employed for surface elemental analysis.

AES operates by emitting an incident electron beam at a sample, which causes surface

atoms to emit electrons of an energy that is characteristic of the atom. This technique

operates in the following manner: an incident electron knocks off an inner shell electron

on the sample surface, which causes an electron of higher energy to fall into this inner

shell. This causes emission of a photon, which is reabsorbed by the surface atom, and

causes an outer shell electron to be emitted. It is this emitted electron that is detected,

and its characteristic energy is indicative of its atom. The Auger instrument used in this

research was a Perkin Elmer 6600, that is capable of providing compositional data for

elemental analysis for samples of 1 atomic %.


































Figure 2-1. Schematic of metalorganic chemical vapor deposition operation. Cold
precursors and carrier gases flow over a heated substrate. Precursor gases
undergo a heterogeneous reaction at the substrate surface, depositing a solid
film, while gaseous byproducts are pumped out through the system exhaust.














CHAPTER 3
DETERMINATION OF DISLOCATION DENSITY OF III-NITRIDE FILMS BY
X-RAY DIFFRACTION

Introduction

There is currently no universally accepted method that allows for rapid and

accurate determination of threading dislocation density in III-nitride thin films. X-ray

diffraction is a promising technique to determine dislocation density because it provides

results quickly and nondestructively, and when modeled correctly, allows for accurate

determination of microstructural parameters of crystalline materials. The goal of this

chapter is to develop a method to determine dislocation density based on x-ray diffraction

that provides accurate results as quickly as possible. Such a method would facilitate the

establishment of a correlation between dislocation density and film properties. This

information, in turn, would allow experimenters to determine how dislocations influence

electrical and optical properties of device layers.

Characterization of Dislocations

Several characterization techniques have been employed to ascertain dislocation

density in III-nitride films, including transmission electron microscopy (TEM), atomic

force microscopy (AFM), cathodoluminescence (CL), and high-resolution x-ray

diffraction (HRXRD). TEM is generally accepted as the most accurate and best

understood method, but the sample preparation is time consuming and destructive, and

some care must be used in how the dislocations are imaged and counted [23]. AFM has

also been used, with the dislocations identified as slight depressions near step









terminations, and these depressions have been correlated to dislocations with a screw

component [24]. Dislocations with a pure edge component are occasionally observed on

terraces; however, the large tip radius usually precludes direct observation. Using CL

imaging, dark spots are observed on the GaN surface and have been correlated to the

dislocation density in TEM samples [25]. In CL, the dislocations that act as non-radiative

recombination are imaged as dark spots while dislocations that radiate may have the same

intensity as the background GaN. X-ray diffraction provides quick results that are

correlated to the microstructure of thin films, and the microstructure is directly related to

dislocation density. Because of the limitations associated with TEM, AFM, and CL,

HRXRD is an attractive alternative to ascertain dislocation density for III-nitride thin

films.

Previous Work

Researchers began using XRD to study microstructures in metals as early as the

1950s [26-29]. By comparing the angular distribution of x-ray intensity for a given

reflection of cold-worked vs. annealed specimens (i.e., those with high and low

dislocation densities, respectively), a correlation emerged between the XRD line width

and the dislocation density. Several models were proposed to calculate dislocation

density based on the degree of diffracted x-ray scattering. These studies provided a

valuable theoretical and experimental foundation for determination of dislocation density

by XRD.

In the 1990s, these classic works were applied to heteroepitaxial GaN-based thin

films [30-34]. These modern works accounted for line width broadening due to strain,

coherence length, rotations of crystallographic domains, and interdependence of these









rotations; however, there were disagreements concerning the relative importance of each

of these broadening effects when estimating threading dislocation density. As a result,

there is currently no unified analytical model from which to determine dislocation density

based on XRD line width results. This work resolves these differences by providing

insight into each of the significant broadening mechanisms, concentrating heavily on

twist and incoherence.

Twist

When ascertaining the crystalline quality of III-nitride films by XRD, it is crucial

to determine the microstructural twist. Threading edge dislocations (b = (11-20)) are

often dominant in heteroepitaxial III-nitride films, and these are directly related to

crystallographic twist. Recognizing this, Heying et al. [35] explicitly demonstrated the

need for asymmetric scans in skew geometry in order to better characterize the

dislocation character of GaN films. Several models were subsequently proposed on how

to best determine twist in GaN films. Metzger et al.[30] and Kang et al.[31] proposed

similar procedures that employ a 4-scan in order to determine twist. However, it is not

possible to obtain accurate results by this method with most XRD systems because the

instrumental out-of-plane broadening of a (-scan is -1.50, which corresponds to a

dislocation density on the order of 1011 cm2, thereby grossly overestimating the edge

component of dislocation density based on twist.

Srikant et al. [32] proposed a model that employs asymmetric rocking curves

(co-scans), and accounts for tilt and twist dependence on both inclination angle (2) and

measured full width at half maximum (FWHM). By treating twist as a free parameter,

several scans at different angles of inclination are required for this method, with twist









being determined by optimizing the fit to the FWHM data. This model was also the basis

of the work of Chierchia et al. [33] and Sun et al. [34] to determine twist. One significant

disadvantage with this method is the necessity of performing so many scans at

asymmetric reflections, lengthening the measurement and analysis time.

Incoherence

Another phenomenon often neglected when determining dislocation density by

XRD is the effect of incoherence broadening on the measured FWHM. The model

proposed by Srikant et al. only considers tilt and twist contributions to broadening [32].

Chierchia et al. discussed broadening due to incoherence for symmetric scans, but did not

explicitly include it when calculating dislocation density [33]. Both Chierchia et al. and

Sun et al. [34] stated that their calculations of twist were overestimations because they

neglected coherence length broadening upon determining twist [33]. The degree to which

incoherence broadening affects the calculation of dislocation density, however, was never

described in detail.

Focus

In this chapter, a theoretical and experimental approach is presented for using

HRXRD to measure threading dislocation densities of III-nitride thin films in a rapid,

non-destructive manner. This method relates HRXRD line widths to dislocation density,

allowing resolution of screw and edge dislocation densities from the total dislocation

content. The model developed in this work unifies and simplifies other proposed models

and accounts for all significant microstructural causes for broadening. This model was

verified by comparing the dislocation density results obtained by HRXRD to those

obtained by TEM on III-nitride films grown by metalorganic chemical vapor deposition









(MOCVD) on both SiC and sapphire substrates. The model was further simplified by

carefully considering the magnitudes of the contributions of each broadening term,

leading to accurate dislocation density determinations from as few as two HRXRD

reflections.

Experimental Methods

To obtain the dislocation density from the HRXRD line widths, the individual

broadening effects from tilt, twist, and lateral coherence length were deconvolved from a

series of symmetric ((0002), (0004), (0006)) and skew-symmetric ((10-11), (20-22))

rocking curves. These reciprocal lattice points are shown in Figure 3-1. Crystallographic

tilt is defined as the out-of-plane rotation of adjacent domains in the crystal, as

demonstrated in Figure 3-2, and could be determined directly from the symmetric series

of scans. The asymmetric reciprocal lattice points were chosen based on the significant

twist component and the relatively high measured intensity associated with them. Twist

is defined as the in-plane rotation of adjacent domains (Figure 3-3).

Measurements were made using a Philips X'Pert MRD high-resolution triple-axis

x-ray diffractometer with a sealed Cu anode collimated by a four-crystal Ge(220)

monochromator. The Cu anode produced an x-ray wavelength of k(Cuk l)=1.5404 A. Of

the ten samples studied by HRXRD, seven were selected for comparison with TEM over

the widest range of dislocation density. The microscopy was performed with a Philips

CM20 instrument at 200kV. One specimen was prepared in plan view while the other six

were prepared in cross section. Three were prepared by focused ion beam, while the

balance was prepared by a traditional mechanical polishing, dimple, and ion mill









procedure. Cross-section TEM (XTEM) images of low and high dislocation density

samples are shown in Figure 3-4 and Figure 3-5, respectively.

To determine dislocation densities of the samples in cross section, dislocations

were imaged and counted as usual, under two-beam bright field, dark field, and weak

beam (Figure 3-6a) conditions and counted as usual. Specimen thicknesses were

determined by convergent beam electron diffraction (CBED) using a (11-20) two-beam

condition (Figure 3-6b). The number of parallel diffraction intensity oscillations

("fringes") observed in the CBED disks is related to the extinction distance g and the

specimen thickness t. Cross sectional thicknesses were assigned based on an average of

the first two linear fits of the fringe spacings according to the formula



S,2 + 1
11 2 7 211 t2

where s, is the fringe spacing of the ith fringe, and n is an integer. For more detailed

information on this technique, see reference 15.

The samples were grown by MOCVD in an EMCORETM D-125 reactor on both

silicon carbide and (0001) sapphire substrates under a variety of growth conditions in

order to vary the dislocation densities; the growth temperature for all films was 10500C,

and the precursors were trimethylgallium (TMG), trimethylaluminum (TMA), and

ammonia (NH3) with hydrogen (H2) and nitrogen (N2) carrier gases. The pressure was

varied from 140 300 Torr. All samples analyzed were either GaN or AlxGa(l-x)N, with

aluminum content as high as 45%.









Method

There are several challenges posed by employing HRXRD to determine threading

dislocation density. These involve the following:

Identifying the significant sources of line width broadening.
Formulating a model that accounts for these broadening effects.
Establishing a procedure using XRD scans to probe the film layers, using the
proper geometry.
Determining the extent to which microstructural defects influence measured
HRXRD line width.
calculating dislocation density from this information.

Computation of dislocation density is important for comparison studies between

materials. Microstructural quality of III-nitride thin films is often reported by quoting

the FWHM values of select reflections. However, a simple comparison of line widths for

particular XRD scans may be insufficient when comparing crystalline quality of thin

films. There are two reasons for this: first, dissimilar XRD systems may exhibit

significantly different intrinsic broadening, which must be accounted for. Second,

research groups may employ any of a variety of reflections to ascertain crystalline quality

of thin films, and these reflections are affected differently by microstructural

imperfections. Since there is no one universally employed XRD reflection, calculating

dislocation density with a microstructural model is the only way to compare different

samples conclusively.

Sources of Broadening

When performing HRXRD rocking curves on single crystal materials, the

potential sources of broadening are due to:

Crystallographic rotations.
Inhomogeneous strain fields.
Curvature of the film.
Intrinsic rocking curve width.









Broadening due to the monochromator.

For the purposes of this study, broadening effects due to film curvature, intrinsic rocking

curve width and the monochromator are neglected, since all of these effects are less than

12 arc sec. [36]

Figure 3-7 demonstrates specific broadening effects in reciprocal space and how

the diffractometer elucidates each of these broadening effects. Figure 3-7(a) is a view of

the film in cross-section with the relevant broadening effects at an asymmetric reciprocal

lattice point (RLP) in skew geometry. Khkl is the reciprocal lattice vector between the

origin of reciprocal space and the RLP being probed. Its direction is related to the angle

between the surface normal of the crystal and the normal of the set of diffracting planes,

while its length is related to the d-spacing of the set of planes associated with the RLP.

Rotations of crystallographic domains broaden the RLP transverse to Khkl, while

inhomogeneous strain broadens in the direction parallel to Khkl. Broadening due to

incoherence occurs both parallel as well as transverse to Khkl.

Specific broadening effects due to crystallographic rotations and incoherence are

demonstrated in Figure 3-7(a). Pure domain tilt and twist are observed about the Ky and

Kz axis, respectively. The effects of tilt and twist measured by the diffractometer are the

projections of the rotations about Ky and Kz, respectively, on the RLP. The extent to

which each of these rotational broadening contributions affects the measured line width

of a rocking curve (co-scan) is geometrically related by Xhkl, the angle between Kz and the

RLP being probed. Figure 3-7(a) also demonstrates broadening due to incoherence, as

domain thickness (h) and lateral coherence length (L) are shown explicitly. Because

threading dislocations propagate along the growth plane normal during III-nitride growth,









the domain thickness is assumed to be equal to the film thickness. The lateral coherence

length is defined as the average distance between threading dislocations.

The tilt, twist, and lateral coherence length all affect broadening of an asymmetric

RLP when a rocking curve is performed in skew geometry. Since the film thicknesses in

this study were >8000 A, vertical coherence length had an insignificant effect on

measured broadening. Furthermore, inhomogeneous strain broadening was neglected

because all scans employed in this study were triple axis rocking curves, which diminish

the observed broadening effects in the direction parallel to Khkl. Therefore, the

microstructural imperfections that significantly contributed to our measured values for

FWHM were projections of rotations about both Ky and Kz (tilt and twist) on the probed

RLP and lateral incoherence (27T/L).

Figure 3-7(b) shows broadening effects of the skew-symmetric RLP in plan-view.

The measured broadening (Fhkl), is the measured FWHM and broadening due to tilt, twist

and incoherence are indicated explicitly. These microstructural broadening effects were

included in the proposed model that was used to calculate dislocation density.

Model

The following reciprocal space model describes the (hkl) dependence of the

convolved peak-width contributions from the combined effects of tilt (Frlt), twist (Ftwst)

and coherence length (L) broadening:


Frh (Fr tt COS Xhki ) + (rtF sin hkI) + 2; (3-1)
| K hki L









The geometric dependence of tilt and twist on measured line width is apparent, and the

observed broadening due to incoherence is inversely proportional to the magnitude of the

reciprocal lattice vector. The exponent, n, denotes the type of convolution of the model.

Determination of Convolution

One major discrepancy among previous works is the type of convolution, i.e., the

value of n is used in the model (Equation 3-1). The choice of n is derived from the

best-fit function to the intensity distribution [32]. For example, for a Lorentzian

convolution, n=1 and for a Gaussian convolution, n=2. For symmetric scans, both

Metzger et al. [30] and Chierchia et al. [33] assumed a Lorentzian convolution, where tilt

and symmetric coherence length were determined from a Hall-Williamson plot.

The results obtained in this study indicated that the lateral coherence length value

was unrealistically high when assuming a Lorentzian convolution, and a more plausible

value for coherence length was obtained when a Gaussian convolution was used. For

example, a film with a screw dislocation density of 7.7x108 cm-2 measured from TEM

gave the following results: assuming a Lorentzian convolution, the lateral coherence

length was -31,000 A, indicating a dislocation density, p -1x107 cm-2. For the same

sample, assuming a Gaussian convolution, the lateral coherence length was 5,000 A,

indicating a dislocation density, p -4x10 cm-2. Similar results were reported by

Chierchia et al.[33]. Therefore, a Gaussian convolution (n=2) was used in our analyses

with Eq. 1, which is similar to the models of Hordon and Averbach [26] and Ayers [36].

Figure 3-8 shows FWHM data vs. the magnitude of the reciprocal lattice vector

(I/lhkl) for a typical sample assuming a Gaussian convolution. Figure 3-9 shows the

dependence of the FWIHM on angle of inclination (X) assuming a Gaussian convolution









for one sample of GaN on sapphire. The Gaussian assumption of the model fits the

FWHM vs. data well, and explicitly demonstrates that setting the exponent, n=2,

provides accurate results.

Calculation of Dislocation Density

With the proposed model, two series of scans at different angles were required

to deconvolve each individual broadening effect on the experimentally determined

FWHM. This deconvolution procedure involves fitting the FWHM data to the model, as

shown in Figure 3-8, and is described in detail in the Results and Discussion section.

From the experimentally determined values for tilt, twist, and coherence length (both

symmetric and asymmetric), two independent methods were used to calculate dislocation

density.

The first method uses the tilt and twist values with the classic formulation of

Dunn and Koch [27], where the density of dislocations, p, is given by p=F2/4.36b2

Figure 3-9 shows the mathematical and geometrical derivation upon which this method is

based. Here, F is the calculated rotational broadening effect (tilt or twist) and b is the

applicable Burger's vector. With this formula, it is possible to calculate both the screw

and edge dislocation density with the deconvolved values for tilt and twist, respectively,

using the appropriate Burger's vector. It is important to note that while the coherence

length contribution was not used explicitly in calculating dislocation density by this

method, its broadening effect was accounted for in Equation 3-1.

The second method employed to compute dislocation density was that proposed

by Hordon and Averbach [26]. This method assumes that the measured value for

coherence length equals the root mean square spacing of randomly spaced threads, where









the dislocation density, p=1/L2, as shown in Figure 3-10. In order to compute the total

dislocation density, the value of coherence length was L(lo-11), the coherence length due to

all dislocations (i.e. the coherence length as determined by asymmetric scans in skew

geometry, since these are sensitive to both tilt and twist). In order to calculate screw

dislocation density, the symmetric coherence length, L(ooo0) was input for the value of L.

The edge dislocation density could be resolved by taking the difference between the total

threading dislocation density and screw dislocation density.

Simplified Model

It is possible to simplify Eq. 1 by neglecting lateral coherence length broadening

in order to determine dislocation density more rapidly:

2l = (tt COS zhk )2 + (Fr t sin hl )2 (3-2)

This modification makes it possible to determine the relative tilt and twist of the crystal

using only one symmetric scan (e.g., (0004)) and one asymmetric scan (e.g., (10-11)) in

skew geometry. Eq. 2 provides an upper bound for rotational broadening effects and a

concomitant overestimation of dislocation density. However, Eq. 2 is useful for

providing more rapid results to ascertain relative crystalline quality of III-nitride thin

films. For example, when characterizing an 8000 A thick GaN film with a dislocation
9 -2
density -109 cm-2, this two-scan procedure can provide results within an hour, as opposed

to >3 hours for the full five-scan procedure required when incoherence broadening is

included in the analysis. This is a considerable improvement when rapid results are

desired for growth optimization studies.

Comparing dislocation density results determined from the full five-scan

procedure to results obtained using this condensed model, the abridged procedure









provided results to within 15%. In all samples studied, the spacing of dislocations was

assumed to be random rather than piled-up. A piled-up distribution of dislocations may

affect the incoherence broadening term more significantly and must be accounted for, as

the value for coherence length may decrease significantly for such a distribution.

Furthermore, when the dislocations are piled up, the calculation of dislocation density

must be performed using a different method, the details of which are described in

Reference 8.

Results and Discussion

For each sample, the individual broadening effects were deconvolved using

Equation 3-1 with the measured FWHM data. Figure 3-8 demonstrates a fit of the model

to the FWHM data. From the fit of the model to the data, the magnitude of each

broadening contribution was determined. The three symmetric scans were employed to

determine tilt and symmetric coherence length, as twist does not affect these data. Once

this value for tilt was established, twist and asymmetric coherence length were next

determined by fitting the model to the FWHM values of the skew-symmetric reflections.

It is important to note that coherence length for a symmetric family of planes may be

significantly different than that of an asymmetric family of planes.

Two independent methods were used to calculate dislocation density using these

values for tilt, twist, and coherence length (both symmetric and asymmetric). These

included the formulation of Dunn and Koch [27] (p=F2/4.36b2) and that proposed by

Hordon and Averbach [26] (p=1/L2). Figure 3-12 shows the agreement between the two

methods within a factor of 3. However, the formulation of Dunn and Koch was more

reliable when compared to TEM results. This was expected since the coherence length









broadening effect was slight compared to those of crystallographic rotations, and

therefore, had a greater intrinsic error.

Figure 3-13 shows dislocation densities determined by HRXRD, using the

formulation of Dunn and Koch compared to those determined by TEM. The values for

dislocation density ranged from 5.4x10 cm-2 to 7.9x109 cm-2. All of the values for

dislocation density determined by HRXRD were obtained using Eq. 1 and the full

five-scan procedure, for which incoherence broadening was accounted. The values for

dislocation density determined by HRXRD agreed with those determined by TEM within

a factor of 1.5 for all seven samples.

With confidence that the XRD-based method described in this chapter is accurate

with respect to TEM, it served as a basis for all subsequent growth optimization studies

described throughout this dissertation with respect to dislocation density. The first such

optimization study observed the effect of growth pressure on dislocation density. Figure

3-14 demonstrates these results, and exhibits that increasing the growth pressure causes a

decrease in dislocation density. Performing this study on so many samples using TEM

would be impractical because of the time and cost associated with TEM sample

preparation. Such studies can only be performed by a non-destructive, rapid technique

such as XRD.

Ultimately, III-nitride researchers seek to conclusively determine the effect of

dislocations on device performance. Using the XRD-based method described above to

determine dislocation density, the effect of dislocations on HEMT performance was

ascertained with respect to electronic properties. For the purposes of this experiment, the

HEMT performance assumed to be proportional to the sheet density-mobility product (ns









x [t), widely used as the figure of merit for GaN HEMT device performance [38]. In a

series of experiments, electronic properties obtained by Hall measurements were

compared to dislocation density. These results are shown in Figure 3-15. While

dislocations appear to inhibit room temperature mobility for a given range of sheet

density, no trend could be concluded. The reason is that the three samples of highest

dislocation density were grown early in the HEMT growth optimization process, and

other room temperature scattering mechanisms are believed to have had a stronger effect

on these samples compared to those that exhibit higher mobility. These scattering

mechanisms include interface roughness and alloy disorder. As the HEMT growths were

optimized, higher quality interfaces with fewer variations in alloy disorder from the

AlGaN barrier layer were most likely the cause for the improved mobility. Further study

is necessary to conclusively determine the precise effects of dislocations on room

temperature mobility.

Summary

This chapter describes a method to determine dislocation density from measured

HRXRD line widths. A geometrically derived reciprocal space-based model was

developed that allows for the determination of individual microstructural broadening

effects in III-nitride thin films. This model unifies previous works that employ x-ray

diffraction to calculate dislocation density, but demonstrates distinct advantages in terms

of simplicity, accuracy, and time efficiency. Realizing the microstructural tilt, twist, and

coherence length, dislocation density was calculated by two separate, independent

methods, which agreed within a factor of 3. The model was simplified to neglect

broadening due to incoherence, which provided an upper bound for dislocation density









from only two scans. Finally, the validity of this model was tested by comparing

dislocation density results to those obtained by TEM, and agreement within a factor of

1.5 was found. These results show that the HRXRD-based method to determine

dislocation density presented here is accurate, as demonstrated by the agreement of our

XRD-based values for dislocation density with TEM, and time-efficient, as results were

obtained within an hour. Thus, this HRXRD-based method to determine dislocation

density may provide researchers with a valuable tool for optimizing growth conditions for

heteroepitaxial III-nitride films. The method demonstrated in this chapter serves as a

basis for the following chapters of this dissertation, since rapid and accurate feedback are

required to perform intensive surveys on the effects of dislocations on electronic

properties and device performance of GaN thin films.














(0006)




(20-22)

(0002) (22
X(o01)
(10-11)


KII

Figure 3-1. Schematic of reciprocal lattice points probed. Procedurally, the reciprocal
lattice points probed were symmetric (0001) and asymmetric (10-11). Together,
these data could be used to determine the tilt, twist, and coherence length in
the crystal.






40










Figure 3-2. Adjacent domains tilted with respect to one other. The axis of rotation is
orthogonal to the c-axis.






41












Figure 3-3. Adjacent domains twisted with respect to one another. The axis of rotation is
orthogonal to the plane of the domain, parallel to the c-axis.






42








Figure 3-4. Cross-section transmission electron microscopy image of a low dislocation
density sample. The dislocation density for this sample is 2.1x109 cm-2







43










Figure 3-5. Cross-section transmission electron microscopy image of a high dislocation
density sample. The calculated dislocation density for this sample is 6.8x109
c -2
cm



















































Figure 3-6. Weak beam transmission electron microscopy image and convergent electron
beam diffraction image. A) A sample weak beam TEM image of a film with a
dislocation density of 2.59x109 cm-2. B) Convergent beam electron
diffraction image showing the oscillations in diffracted intensity. Disks are
the (000) and (11-20) reflections.


: : Y: : 777 7:: 7 V P I, Ac:::::: =


" ':I'"~~~~'~:':~.'_'':'~'"' 7 "T: :;


..... .....
.. ..
.........











Ewald
sphere \ s


\ I Ir I
/ ~'d K etector-
aperture
/ integration

2u r 1

S / bbroadening due to
(b) K rotations about K, and K,
(D) K \K

Domain
S twist broadening due to
rotations about K,
4 inmanin

'-- .....--
J Li





((hkl) planes/ domain tilt


Figure 3-7. Broadening effects demonstrated in reciprocal space. In order to accurately
determine twist, asymmetric rocking curves in skew geometry are necessary.
In (a), the film is shown in cross section. The crystallographic rotations affect
measured broadening proportional to their projections of their respective axes
onto the reciprocal lattice point being probed. Broadening due to incoherence
is proportional to 27T/L. In (b), the measured broadening effects are shown in
plan view. These include tilt, twist, and incoherence.







46



-Symmetric data
3 0.290
0.290 -m -Skew-symmetric data

S0.250

0.210

0.170
2 4 6 8
IKI (1/A)


Figure 3-8. Plotted data of full width at half maximum vs. reciprocal lattice vector
magnitude. As |Khkl increases, incoherence broadening diminishes. The
curvature of the plot as KIhkl approaches 0 indicates the coherence length
contribution on the measured FWHM.











0.15

0.13


' 0.11

M 0.09

LL
0.07

0.05


0 10 20 30 40 50 60 70 80 90
Chi (degrees)


Figure 3-9. Fit of full width at half maximum data vs. angle of inclination. The data
shown are for one sample of GaN on sapphire at every observable reciprocal
lattice point. The fit of the data assumes a Gaussian convolution without
accounting for broadening due to incoherence. The data fit well, and
demonstrate that a Gaussian convolution is acceptable. Most of the data lie
above the fit due to the neglecting of incoherence broadening.






48


-2-T In 2 b l

L

Figure 3-10. Mathematical and geometric derivation of the random distribution method
for calculating dislocation density. The coherence length, L, and the
rotational broadening effect, F, are indicated explicitly. The opposite side of
this right triangle was derived to be a constant multiplied by the appropriate
Burger's vector.







49





LL
LL
L


Figure 3-11. Calculation of dislocation density based on values for coherence length.
Since coherence length is defined as the average distance between
dislocations, dislocation density is simply calculated as p=/L2.













: 9.9





0 0 9.1

0
-j
8.7
8.7 9.1 9.5 9.9
Log Dislocation Density (Dunn/Koch)


Figure 3-12. Comparison of methods of determining dislocation density by XRD. The
agreement shown here demonstrates internal consistency from the XRD-based
methods to determine dislocation density.













c 9.0E+09
E



o 6.0E+09

o
o
3.0E+09


I-
0.0E+0
O.OE+00 3.0E+09 6.0E+09 9.0E+09
XRD Dislocation Density (cm"2)


Figure 3-13. Dislocation density determined by transmission electron microscopy vs.
dislocation density determined by x-ray diffraction. All values determined by
XRD agreed with TEM within a factor of 1.5.











3.0E+09
OGaN 150torr
E GaN 10Otorr
2.5E+09
: C

o 2.0E+09 -
o 1

8 1.5E+09 -


1.0E+09
50 100 150 200
Pressure (Torr)


Figure 3-14. Dependence of dislocation density on growth pressure. Higher growth
pressure causes a lower dislocation density. Dislocation density results were
obtained by x-ray diffraction.











2000
S*ns = 5.2 E+12
1600 ns = 9 E+12
Ans = 2.2 E+13
S1200 ns = 1.3 E+13
E 1200 -
0

800
O

400
A
0
0.00E+00 2.00E+09 4.00E+09 6.00E+09
Dislocation Density (cm-2)

Figure 3-15. Effect of dislocation density on room temperature mobility of gallium
nitride high electron mobility transistors. No clear effect is evident.














CHAPTER 4
EFFECT OF DISLOCATIONS ON IMPURITY INCORPORATION IN METAL
ORGANIC CHEMICAL VAPOR DEPOSITION GROWTH

Introduction

Aside from high dislocation density, MOCVD grown GaN and III-nitride films

contain extremely high concentrations of impurities. Due to the harsh MOCVD

conditions required for growth of III-nitride materials, typical GaN films have impurity

concentrations >1016 cm-3, and sometimes as high as 1020 cm-3 [39]. These impurities

have a profound effect on electronic properties of III-nitride thin films and make it

difficult to optimize MOCVD growth conditions. Impurities have also been implicated in

providing mid-gap states and inhibiting device performance. The most prominent of

these impurities are silicon, oxygen, and carbon.

Silicon

Silicon is a common impurity in GaN and III-nitride thin films. It is used as an

n-type dopant, as it acts as a shallow donor in a III-nitride lattice when it is a

substitutional impurity on a gallium lattice site (SiGa). The energy differential of the

silicon donor state to the conduction band minimum (CBM) has been widely reported,

although there is some disagreement. Researchers have reported values of 22 meV [40],

30.18 meV [41], 30.8 meV [42], 31.7 meV [43], and 42 meV [44]. Despite the variation

in these reported results, all of these values are indicative of a shallow donor in GaN and

AlGaN films, and is the dopant of choice when n-type conductivity is desired. The









source of silicon in unintentionally doped (UID) III-nitride thin films is widely believed

to be from decomposition of silicon carbide (SiC) substrates [45] and susceptors.

Oxygen

Oxygen is another impurity that has been implicated as an n-type dopant when it

incorporates substitutionally on a nitrogen lattice site (ON). Many researchers have

implicated oxygen as being responsible for the significant background n-type

conductivity of UID GaN films. Several research groups have reported that ON donor

level states are between 32-34 meV from the CBM [40-42] while other groups have

reported lower values. Joshkin et al. reported a value of 23.5 meV [46], while others

have reported values between 2-10 meV [44,47]. Despite the variance in these values,

oxygen is widely considered a shallow donor. Many researchers believe that oxygen

impurities act as n-type dopants, simply because of its extra valence electron relative to

nitrogen, while others have proposed a more complicated mechanism for oxygen

providing the background n-type conductivity of UID GaN. For example, Oila et al. have

proposed that oxygen impurities promote the existence of gallium vacancies (VGa), as

oxygen-gallium vacancy complexes (ON-VGa) are energetically stable in the GaN lattice.

According to Oila et al., it is the gallium vacancies that provide the background n-type

conductivity [48]. Overall, however, many researchers attribute the background

conductivity of GaN to the presence of oxygen in the lattice, despite the disagreements in

the proposed mechanisms.

The sources of oxygen include residual oxygen gas in the ammonia source [47]

and atmospheric contamination of the growth chamber. Aside from these sources of

contamination, the sapphire (A1203) substrate [45] provides a significant source of









oxygen. When sapphire is subjected to the high temperatures employed during MOCVD

growth of in a hydrogen ambient, the sapphire surface is etched, and the oxygen atoms

incorporate into the GaN film. Performing electrical measurements, researchers have

found that the GaN film is significantly n-type within -0.4 |jm of the sapphire interface,

due to a buried conductive layer (BCL) [49-53]. This BCL is undesirable when

attempting to grow semi-insulating GaN films, and researchers have attempted to

compensate these residual carriers by intentionally incorporating deep acceptors, like

carbon and iron [49].

Carbon

Carbon is another common impurity in MOCVD grown GaN films. Carbon is an

amphoteric dopant, as its electronic properties depend on whether it incorporates

substitutionally on a gallium site or on the nitrogen site [54,55]. If carbon incorporates

on a gallium site, it acts as a shallow donor, with an ionization energy of -0.2 eV.

Conversely, if carbon incorporates on a nitrogen site, it acts as a shallow acceptor with

values of ionization energy reported as -0.3 eV [55] and 34 meV [42]. If carbon

incorporates interstitially, its energy level is mid-gap, and its donor/acceptor properties

depend on the Fermi level. If carbon incorporates interstitially into n-type GaN, it will

act as a deep acceptor, but if carbon incorporates into p-type GaN, its energy state will be

indicative of a deep donor [55]. Carbon is generally considered a deep acceptor, and

when GaN films must be insulating, MOCVD researchers often intentionally adjust

growth conditions to favor carbon incorporation in order to compensate the background

n-type conductivity of UID GaN. However, growers try to minimize carbon









incorporation near active regions, as it acts as a scattering center, and has been implicated

in gate leakage and dispersion in GaN-based field effect transistors (FETs).

The sources of carbon in MOCVD growth of III-nitrides are the metalorganic

group III precursors, and, to a lesser extent, SiC substrates and susceptors. Carbon is a

common impurity in any MOCVD growth process, and its incorporation can dramatically

affect the film's properties, but growth conditions can be adjusted to control the extent to

which carbon incorporates.

Previous Work

Growth experimenters have demonstrated that MOCVD growth conditions have a

significant effect on impurity incorporation in GaN films. Specifically, the effects of

temperature, pressure, and flows of TMG, ammonia, and hydrogen on silicon and carbon

incorporation were studied in-depth [39]. A correlation between resistivity, dislocation

density, and carbon concentration in GaN films has also been established, indicating that

threading edge dislocations and/or carbon act as compensating centers [56].

Furthermore, threading edge dislocations have been shown to provide energetically

favorable sites for impurity incorporation, due to the miscoordinated atoms along

extended line defects [57,58] in other material systems. Studies on growth of III-nitrides

have only investigated the effects of growth conditions on impurity incorporation on

templates with nominally identical microstructures. What is lacking to date is the effect

of dislocations on impurity incorporation in MOCVD-grown III-nitride films [39,56].

Both Koleske et al. and Wickenden et al. have indicated that threading edge dislocations

facilitate carbon incorporation, but conclusive studies to confirm this have not been

performed.









Focus

The experimentation in this chapter aims to determine the effects of dislocation

density on impurity incorporation during MOCVD growth. Specifically, silicon, oxygen,

and carbon incorporation are observed with respect to growth conditions on templates of

variable dislocation densities. By subjecting templates with vastly different dislocation

densities to identical growth conditions, the effect of dislocation density on impurity

incorporation could be determined.

Experimental Methods

Templates were grown by MOCVD on sapphire substrates, and dislocation

density was intentionally altered to obtain two series of GaN films with the largest range

of dislocation density possible. Specifically, the growth conditions that were altered were

growth pressure, nucleation layer, and recovery time. Template dislocation densities

ranged from 4.59x10 cm-2 8.76x1010 cm-2 for the first growth series, as shown in Table

4-1, and 4.46x108 cm-2 -7.6x1010 cm-2 for the second growth series, as shown in Table

4-2. As expected, conditions that minimized dislocation density were higher growth

pressures, GaN nucleation layers, and delayed recoveries. Higher growth pressures cause

larger grain sizes during the initial stages of high temperature growth, which cause fewer

dislocations upon coalescence [59]. GaN nucleation layers provide lattice-matched

templates upon which to grow high temperature GaN. Delayed recovery is believed to

cause dislocations to propagate laterally rather than thread along the c-axis to the surface.

The characterization performed in this experimentation included XRD to

determine dislocation density, using the method described in Chapter 3, and SIMS to

determine impurity incorporation. Two series of growth runs were performed on









templates of dislocation density varying over two orders of magnitude, while the only

varied growth parameter was pressure in the overgrowth. For both overgrowth runs,

variable dislocation density templates were grown at 10500C, while pressure was varied

from 20-500 Torr, and silicon was incorporated as a marker when changing pressure.

Impurity incorporation was measured for all three samples from both runs.

Results and Discussion

The first series of SIMS data is shown in Figures 4-1 to 4-4. In all samples,

oxygen and silicon were at the detection limit of the instrument. The carbon profiles

changed in all cases, presumably due to changes in growth pressure, but this could not be

confirmed due to the fact that the silane did not incorporate. Another observation was

that the radial position of the wafer had a distinct effect on impurity incorporation. The

profile shown in Figure 4-3 was taken from the same wafer as that of Figure 4-4, but the

SIMS results from Figure 4-3 were obtained near the top of the wafer, while those from

Figure 4-4 were obtained near the flat. This demonstrates that impurity profiles are more

pronounced near the flat part of the wafer.

From the results of the first series of runs, two adjustments were made: 1) all

SIMS analysis was performed near the flat part of the wafer to obtain more distinct

profiles, and 2) the silane source was confirmed to be online by Hall measurements of

previous samples to ensure intentionally doped n-type conductivity. The SIMS results

for this second series of runs are shown in Figures 4-5 to 4-7. Once again, oxygen

impurities were below detection limit, which demonstrates that oxygen does not

incorporate significantly into GaN films grown in this reactor. Silicon was also at the

detection limit, except when it was intentionally incorporated to indicate changes in









pressure. Carbon profiles, however, changed significantly with changes in pressure for

all samples. As expected, carbon incorporated inversely with pressure; the maximum

carbon incorporation occurred when the growth pressure was 20 Torr, and the minimum

incorporation occurred at 500 Torr for all samples. In order to determine the effect of

dislocation density on carbon incorporation, average values of carbon concentration were

determined for each sample at each pressure and compared. These results are shown in

Figure 4-8, and show that there is no clear trend confirming that dislocation density

promotes carbon incorporation. For all growth pressures, the template with a dislocation

density of 5.59x109 cm-2 had the highest carbon concentration.

Although dislocation density did not have an observable effect on impurity

incorporation, pressure had a significant effect. Figure 4-8 indicates that carbon

incorporation is, to some extent, a pressure-dependent process. At higher pressures, there

are greater molecular interactions between carbon and hydride gases, which promote

carbon removal from the film during MOCVD growth. Threading dislocations, however,

do not appear to provide significantly energetically favorable sites for impurity

incorporation.

In order to confirm the finding that dislocations do not promote incorporation of

impurities (specifically carbon) during MOCVD growth, two series of experiments are

proposed. First, a similar series of experiments could be performed varying parameter of

growth temperature. It is possible that the carbon removal mechanism at high pressures

is significantly different than that of higher temperature, so dislocations may have an

observable effect on carbon incorporation by altering temperature. Second, a similar

series of experiments could be performed with broader range of dislocation density









templates. This could be achieved by growing templates employing unconventional

methods to minimize dislocation density, including epitaxial lateral overgrowth (ELO) or

cantilever epitaxy (CE). These techniques essentially filter dislocations and have been

used to grow GaN films with dislocation densities below 107 cm-2. This would provide a

broader range of dislocation densities for the templates for the purpose of this

experimentation.

Summary

The effect of threading dislocation density on impurity incorporation was

investigated in this chapter. Oxygen and silicon were at the SIMS detection limit in all

cases (except when silane was intentionally flowed into the growth chamber to indicate

pressure changes), indicating that an insignificant amount of oxygen and silicon

incorporated into the GaN films. Carbon incorporation depended heavily on pressure,

and higher growth pressures caused significantly less carbon to incorporate into the films.

Dislocation density, however, had no observable effect on carbon incorporation. This

result indicates that threading dislocations do not provide significantly energetically

favorable sites for impurity incorporation in MOCVD grown GaN films. The carbon

incorporation in this experiment was pressure-driven by a mechanism dependent on

molecular interactions.











1E+21 1E+08


1E+20 Ga-> 1E+07

1 1E+06
u 1E+19
E 0-
O C
S1E+05 O
Z 1E+18
O C
D1 E+04 (
< n0
1E+17 -
Z 1E+03 3
U O
w O O
Z 1E+16
S1E+02

1E+15
1E+01


1E+14 1E+00
0 2 4 6 8 10
DEPTH (microns)


Figure 4-1. Impurity profiles for low dislocation density at the top. The impurities under
study were silicon, oxygen, and carbon for the first series of overgrowth
studies. The dislocation density of the template is 4.59x 10 cm-2. The SIMS
sample was taken from the top part of the wafer.











1E+21 1E+08



1E+20 Ga-> 1 1E+07


1E+06
S1E+19
E -0
o 1E+05 c
1E+18 C o
Z (D
O
-1 E+04 D

1E+17
Z
LU 1E+03 o
o 00
01E+16
U.' 1E+02


1E+15
1E+15 E+01


1 E+14 1E+00
0 2 4 6 8 10
DEPTH (microns)


Figure 4-2. Impurity profiles for mid dislocation density at the top. The impurities under
study were silicon, oxygen, and carbon for the first series of overgrowth
studies. The dislocation density of the template is 6.66x109 cm-2. The SIMS
sample was taken from the top part of the wafer.










1E+21 1E+08


1E+20 Ga-> 1E+07

1+ 1E+06
E 1E+19 "0
o0 C
1E+05 8
Z 1E+18 00
O
F- 1E+04 (
1E+17
S1E+03
U O 0
O 1E+16 O
Z 1E+02
O
1E+15 1E+01


1E+14 1E+00
0 2 4 6 8 10
DEPTH (microns)


Figure 4-3. Impurity profiles for high dislocation density at the top. The impurities under
study were silicon, oxygen, and carbon for the first series of overgrowth
studies. The dislocation density of the template is 8.76x1010 cm-2. The SIMS
sample was taken from the top part of the wafer.










1E+21 1E+08


1E+20 Ga-> 1E+07

S1E+06
E 1E+19 1E
0 2 -
o 1E+05 0
Z 1E+18
0
DP 1E+04 (
S1E+17 0U)
1E+03 0
uJ o
U 1E+16 C)
Z 1E+02
O
0
C) 1E+15 1E+01


1E+14 1E+00
0 2 4 6 8 10
DEPTH (microns)


Figure 4-4. Impurity profiles for high dislocation density at the flat. The impurities
under study were silicon, oxygen, and carbon for the first series of overgrowth
studies. The dislocation density of the template is 8.76x1010 cm-2. The SIMS
sample was taken from the flat part of the wafer.






























Figure 4-5. Impurity profiles for second series of overgrowths for low dislocation density
at the flat. The impurities under study were silicon, oxygen, and carbon. The
dislocation density of the template is 4.46x10 cm-2












1.00E+19 --r ..... .... ...

1 n: ,, ,n II "

E 1.00E+18
C

1.00E+17



S1.00E+16



1.00E+15
0 1 2 3 4 5
Depth (|tm)


Figure 4-6. Impurity profiles for second series of overgrowths for mid dislocation
density at the flat. The impurities under study were silicon, oxygen, and
carbon. The dislocation density of the template is 5.59x109 cm2.























0 1 2 3 4 5 6
Depth (u)m)


Figure 4-7. Impurity profiles for second series of overgrowths for high dislocation
density at the flat. The impurities under study were silicon, oxygen, and
carbon. The dislocation density of the template is 7.6x1010 cm2.











Carbon Incorporation vs. Pressure
1.00E+19
High dislocation density template
E Low dislocation density template
Si ,lid dislocation density template

c
.0 1.00E+18 -


o -

C
O
S 1.OOE+17 -

0



1.00E+16
0 100 200 300 400 500 600
Pressure (Torr)


Figure 4-8. Comparison of impurity profiles for different dislocation density templates.
Impurity profiles for silicon, oxygen, and carbon for variable dislocation
density templates during MOCVD growth with variable pressure are shown.
The low dislocation density template is 4.46x10 cm-2, the mid dislocation
density template is 5.59x109 cm-2, and the high dislocation density template is
7.6x1010 cm-2. The SIMS samples were all taken from the flat part of the
wafer.






70


Table 4-1. Growth conditions for the templates for the first series of SIMS results.
Conditions that promote lower dislocation density are higher pressure, delayed
recovery, and GaN nucleation layers. Dislocation densities were intentionally
altered to provide templates with as large of a range for dislocation density
possible.
Dislocation density Growth pressure Nucleation Recovery
Sample (cm-2) (Torr) Layer Time
1 4.59x108 500 GaN Delayed
2 6.66x109 50 GaN Rapid
3 8.76x1010 70 A1N Rapid






71


Table 4-2. Growth conditions for the templates for the second series of SIMS results.
Dislocation densities were intentionally altered to provide templates with as
large of a range for dislocation density possible.
Dislocation density Growth pressure Nucleation Recovery
Sample (cm-2) (Torr) Layer Time
1 4.46x108 500 GaN Delayed
2 5.59x109 50 GaN Rapid
3 7.6x1010 70 A1N Rapid














CHAPTER 5
GALLIUM NITRIDE BULK CRYSTAL GROWTH BY DISSOLUTION AND
RECRYSTALLIZATION OF GALLIUM NITRIDE POWDER

Introduction

Poor material quality is widely considered to be the most significant factor

limiting GaN-based device performance, and this is primarily due to a lack of availability

of lattice-matched substrates. Ideally, GaN film growth would be performed on single

crystal GaN bulk substrates, but the intrinsic properties of this material makes fabrication

of bulk crystals difficult. In other semiconductor material systems, such as silicon and

gallium arsenide, bulk crystals have been fabricated from melts, where boules of high

crystalline quality material have been achieved with dimensions as large as 12-inches x

12 feet. Similar methods are not feasible for GaN because the GaN dissociates prior to

melting at atmospheric pressure. Because of this, no commercially viable bulk growth

method for GaN has been developed to date, despite the potential benefits of low defect

density GaN bulk crystals.

Previous Work

Researchers have demonstrated several methods to successfully fabricate GaN

bulk crystals. The two most prominent of these are an ultra high nitrogen pressure

method, and an ammonothermal technique. Both approaches offer a distinct set of

advantages and disadvantages relative to the other.

The ultra high nitrogen pressure technique entails subjecting a pool of liquid

gallium metal to extremely harsh conditions in a nitrogen gas (N2) environment [60], with









temperatures of 16000C and nitrogen overpressures of -45,000 atm. GaN crystals are

formed by dissolving the nitrogen gas into the gallium liquid, where N2 dissociates into

atomic nitrogen, and bonds with the gallium. Using this method, researchers have

successfully fabricated crystals of GaN that are 1 cm2 x 100 |tm, with extremely low

dislocation densities (-100 cm-2). The drawbacks to using this approach include slow

growth kinetics, high impurity concentrations, the inability to grow boules of material,

and the high pressure and high temperature requirements. The GaN crystals are grown as

a crust on the liquid gallium surface, and a 1 cm2 x 100 |tm crystal is formed after -1

month under the extreme conditions required. Furthermore, these crystals contain -1018

cm-3 oxygen impurities, which may have an undesirable effect on the electronic

properties of the GaN crystal. Finally, the temperature and pressure requirements add

significant cost to this method.

The ammonthermal technique has been demonstrated to successfully grow GaN

bulk crystals by dissolving GaN feedstock into liquid ammonia, and precipitating single

crystal GaN upon supersaturation [61]. This method demonstrates significant

improvement over the high nitrogen pressure-driven process described above with respect

to growth rate (-0.5 mm/week). Furthermore, large area boules are possible to extract

from a seed crystal. The disadvantages of this method include higher dislocation

densities (-106 cm-2) and high impurity incorporation (-1017-18 cm-3). Moreover, this

process also requires harsh conditions, with temperatures of 550C and pressures of

-4000 atm in order to dissolve GaN into a liquid ammonia medium.

Although both methods of GaN bulk growth described above were significant

breakthroughs, neither are commercially viable processes to fabricate GaN crystals. This









is primarily due to the high-pressure requirement, which prevents both processes from

being scalable and manufacturable. If a method to fabricate GaN bulk crystals at

atmospheric pressure were possible, this would provide a distinct advantage over other

methods currently employed.

Proposed Methods

In order to develop an atmospheric pressure process to form GaN crystals, three

series of experiments have been proposed. The first involves electroplating of GaN using

a gallium metal electrode and a reactive nitride ion (N3-) from a lithium nitride (Li3N)

precursor. A second possible technique comprises an electrochemical reduction of

nitrogen gas (N2) to form nitride ions (N3-) that react with gallium to form GaN. A third

potential method is a simple dissolution/recrystallization process of GaN powder in a

liquid medium, where the formation of GaN single crystal would be driven by a thermal

gradient.

Electroplating

Electroplating of GaN from gallium and nitride precursors could prove to be a

promising method to form high quality GaN crystals. The difficulty, however, was

determining a suitable host environment in which a nitride ion could exist without

reacting quickly and explosively. In an unrelated series of experiments, Goto et al.

determined that a molten alkali-halide host environment could successfully dissolve

stable nitride ions [62]. Using this, the possibility of forming GaN crystals from gallium

and nitride precursors was investigated. Figure 5-1 shows a GaN crystal of macroscopic

dimensions (0.9 mm x 0.6 mm) that was successfully synthesized using this method.









This result was encouraging, but the electroplating method suffered from four significant

drawbacks:

Electroplating does not provide material with optimal crystalline quality.
The electroplating process would be very difficult to scale, in order to develop a
controllable, manufacturable process.
The Li3N precursor is extremely expensive, making this process less cost-
effective.
The buildup of Li+ ions with GaN deposition complicates the electrochemistry
[63].

In order to make this process more cost effective and controllable, the electrochemical

reduction of nitrogen gas to nitride ions was investigated (1/2 N2 + 3 e- N3-).

Electrochemical Reduction of Nitrogen Gas

Goto and Ito investigated the possibility of electrochemically reducing nitrogen

gas into nitride ions in a molten alkali-halide liquid medium [63]. Using this result, the

possibility of fabricating GaN crystals from electrochemically oxidized gallium with

electrochemically reduced nitrogen gas was investigated, and GaN crystals were

successfully formed. Using this method, bulk GaN crystals could possibly deposit on a

seed crystal, and be extracted as a boule. While this process offers several advantages,

the major disadvantage is the difficulty of controlling the fluid dynamics to form the

crystals. Because nitride ions are so reactive, they must be separated from gallium ions

until they reach the desired growth surface, which would be difficult to model.

Dissolution and Recrystallization of Gallium Nitride Crystals

The ideal method to grow bulk GaN crystals would be based on a simple

dissolution/recrystallization process of GaN in a liquid medium. Using such a method, it

would be possible to control the rate of crystal growth by controlling the temperature









gradient and the fluid mechanics in the growth vessel. This process could eventually be

used to form large GaN crystals, that could be extracted as a boule from the melt.

Focus

In this chapter, a novel bulk growth method of GaN crystals is introduced, based

on a simple dissolution/recrystallization process in a molten alkali-halide salt. If GaN

can dissolve sufficiently in such a medium, then researchers could develop a

temperature-dependent, atmospheric pressure process to fabricate GaN bulk crystals.

Such a method could prove to be a scalable and manufacturable process that could lead to

production of commercially available GaN bulk crystals and wafers for homoepitaxial

GaN growth. Specifically, this chapter examines whether a molten halide medium is

capable of dissolving GaN, and if so, whether it is possible to recrystallize GaN on a

nominally homoepitaxial GaN template.

Experimental Methods

All of the experimentation in this chapter was performed in a home-built stainless

steel heater vessel fitted for a 1" quartz tube in a glove box with a nitrogen gas ambient.

To determine the temperature-dependent solubility of GaN, aliquots of GaN in molten

lithium chloride (LiC1) were taken at various temperatures with a quartz rod. The

solubility was determined by inductive coupled plasma-mass spectrometry (ICP-MS),

where the ratio of the gallium to lithium concentration was calculated. Recrystallization

studies were performed in quartz tubes in the same stainless steel chamber, where the

recrystallization surface was MOCVD-grown GaN on sapphire. Deposited GaN

crystallites were observed by scanning electron microscopy (SEM) and elemental









analysis on these crystallites was performed by energy dispersive spectroscopy (EDS) in

order to determine the extent to which GaN recrystallized on the wafer surface.

Results and Discussion

In order to grow GaN crystals at a fast enough rate to be commercially viable, it is

necessary to dissolve a relatively high concentration of GaN into solution. If the

solubility is found to be sufficiently high, it would be possible to grow GaN bulk crystals

at a reasonable rate upon supersaturation. The basis for the experimentation

demonstrated in this chapter is the ideal solubility curve for GaN, shown in Figure 5-2.

This curve was calculated by the following equation [64]:


Inx = T -1 (5-1)


Under diffusion-limited conditions, a solubility of -0.1% for GaN could provide a crystal

growth rate of -0.5 mm/h. This corresponds to a temperature of 1230C on the ideal

solubility curve, and since this process occurs at atmospheric pressure, these conditions

are mild enough for a commercially viable method of GaN crystal growth. If the selected

solvent has favorable interactions with GaN, then experimentally, the data should lie

above this curve, corresponding to higher solubilities, and consequently, potentially

higher growth rates. The first series of experiments determined the temperature

dependent solubility of GaN powder in LiC1.

Dissolution of Gallium Nitride in Lithium Chloride

The temperature dependent solubility of GaN in LiCl was experimentally

determined by performing ICP-MS analysis of aliquots of solution at various

temperatures. These temperatures ranged from 750C 10000C, as shown in Figure 5-3.

The experimental data indicated that, in most cases, more GaN dissolved than was









predicted by the ideal solubility curve. The scatter in the data was most likely due to the

fact that the ICP-MS is an inorganic, water-based technique, and since GaN is insoluble

in water, some of the crystallites may have precipitated out, and segregated. In order to

address this, a 2 |tm filter was used to filter samples for analysis prior to injection into the

ICP-MS. These results, shown in Figure 5-4, represent a lower bound for solubility of

GaN in LiC1, but they were still significantly above the ideal solubility curve. These data

indicate that 10-50 ppm of GaN were soluble in LiCl at temperatures ranging from 800C

- 950C, corresponding to a growth rate of -50-100 [tm/h. These values were sufficient

to commence experimentation of recrystallization of GaN on a GaN template.

Gallium Nitride Recrystallization

In order to determine whether GaN could recrystallize out of solution, a series of

experiments was performed, where GaN powder in LiCl was heated to -9500C, and then,

became supersaturated upon cool-down. This supersaturated GaN would presumably

deposit on a lattice-matched template. SEM images of a surface not exposed to the melt

(Figure 5-5) were obtained and compared to one that was (Figure 5-6). Note the distinct

differences in surface morphology. Upon further investigation, individual features were

observed at higher magnifications, as shown in Figure 5-7. This feature appears to

exhibit a hexagonal shape, indicative of the wurtzite crystal structure.

To further promote GaN recrystallization, another experiment was run with

stirring at the maximum temperature (950C), prior to cool-down. The result of this

experiment was that more surface features appeared. From this finding, a third

experiment was performed, with 1) half of the wafer surface treated with nitric acid

(HNO3) at 500C, and 2) constant agitation during cool-down. The surface treatment was









performed in order to etch the native oxide off the growth surface, and the agitation was

performed in order to provide forced convection, and greater recrystallization. This

provided two interesting results. First, more features were formed on the wafer surface.

Second, the features that formed on the treated surface tended to be different from those

on the untreated surface. On the treated surface, there was a high density of patchy

features, shown in Figure 5-8, whereas, the untreated surface had a high density of

needle-like features, shown in Figures 5-9 to 5-11.

In order to determine the chemical composition of the two observed features,

elemental analysis was performed by EDS. The EDS results for a needle structure is

shown in Figure 5-12, and strongly indicate that this structure is comprised of etched

quartz, as evidenced by the heavy silicon and oxygen concentrations as well as the

gallium void indicated. Elemental analysis of a patch structure is shown in Figure 5-13,

and is indicative of GaN; the gallium and nitrogen concentrations are high, and silicon

and oxygen are practically nonexistent. Another interesting finding is that silicon oxide

(SiOx) preferentially deposited on the untreated surface, which presumably contained the

native oxide, whereas GaN preferentially deposited on the treated surface.

A cross-sectional view of the patch demonstrated in the SEM image in Figure

5-13 is shown in Figure 5-14. From the scale shown, this GaN crystallite is -5 |tm in

thickness, which indicates that the growth rate in this experiment was -2.5 [tm/h; this is

significantly higher than any other bulk growth technique. Furthermore, the depth of

field of the EDS instrument is shown in Figure 5-15 for the highest power used in this

study. Since the feature probed is -5 |tm, and the depth of the instrument is -1600 nm,

this further supports that the crystallite is that of pure GaN.









While the experimentation described in this chapter is relatively unrefined, and far

from optimized, it demonstrates that a simple dissolution/recrystallization process is a

potentially viability one for the fabrication of GaN bulk crystals from a molten

alkali-halide medium. The results were encouraging, and could provide a first step in the

development of a novel method for growing GaN crystals. Significant improvements

could be made to the process, including solvent optimization, greater purity of reagents,

greater control over the fluid dynamics, and using a true seed crystal on which to

crystallize GaN from the molten solvent. Eventually, using this method, researchers

could potentially develop a commercially feasible, rapid growth rate process to fabricate

high crystalline quality GaN boules for homoepitaxial GaN growth.

Summary

An ideal solubility curve was calculated for GaN with respect to temperature,

based on intrinsic thermodynamic parameters. The solubility of GaN in LiCl was

experimentally determined, and demonstrated that GaN was more soluble than the ideal

curve indicated. From these results, a GaN template was placed in a LiCl medium

containing GaN powder, which was heated to -10000C, and then cooled. Deposition was

observed on the GaN template surface by SEM. Further experimentation demonstrated

that GaN crystallites deposited at a higher rate after pretreatment of the GaN template

surface in an acid solution, and with constant fluid agitation during cool-down.

Elemental analysis determined that patchy structures as thick as 5 |tm were, in fact, GaN.

These crystallites formed after merely 2 hours, indicating an extraordinarily high growth

rate, relative to other GaN crystal growth techniques.






81


Potential improvements to this experimentation include

* More sophisticated equipment to allow greater control over thermal gradients and
fluid dynamics.
* Use of higher purity reagents.
* More comprehensive studies optimizing GaN solubility.
Use of a true seed crystal upon which to deposit dissolved GaN.






82















Figure 5-1. Gallium nitride crystal with dimensions of 0.9 mm x 0.6 mm. The bulk
growth technique to form this crystal was electrodeposition.












Ideal Solubility of GaN


0.25

0.2

$ 0.15

0.1
o
S0.05

0


1000 1100 1200 1300 1400

Temperature (C)


Figure 5-2. Ideal solubility of gallium nitride. The data for this curve were determined
by Equation 5-1.


.


r;


I


I












-120
E
0100
> 80

S60
o
S40 -
z
S20
0
750


Solubility of GaN in LiCI


800 850 900 950 1000
Temperature (oC)


1050


Figure 5-3. Experimentally determined solubility of gallium nitride in lithium chloride
compared to the ideal solubility curve. These data points were obtained
without the use of a filter.











Solubility of GaN in LiCI


40

35

- 30 -
E
. 25

S20

| 15
O
0) 10

5

0-
650


Experimentally determined solubility determined after employing a 2 |tm
filter. These data represent a lower bound for solubility of gallium nitride in
lithium chloride.


750 850 950 1050
Temperature (oC)


Figure 5-4.





























Figure 5-5. Scanning electron microscopy image of a bare gallium nitride surface. The
morphology is smooth.























S,- .











Figure 5-6. Scanning electron microscopy image of a gallium nitride surface exposed to
gallium nitride-in-lithium chloride melt. The morphology is noticeably
rougher than the bare, untreated surface.





























Figure 5-7. Close-up scanning electron microscopy image of a gallium nitride crystallite.
Note the hexagonal structure.




Full Text

PAGE 1

EFFECTS OF DISLOCATIONS ON ELECT RONIC PROPERTIES OF III-NITRIDE MATERIALS By ALLEN M. WEST A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLOR IDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2005

PAGE 2

I hereby dedicate this [dissertation] to . myself. (credit to Rodney Dangerfield as Thornton Melon, Back to School)

PAGE 3

ACKNOWLEDGMENTS I would first like to thank Dr. Cammy Abernathy for all of her guidance and support throughout my tenure in the Materials Science Department. She was the ideal advisor for me, and was crucial to my success. I thank Dr. Stephen Pearton, Dr. Fan Ren, and Dr. David Norton as well for serving on my committee. I especially thank Dr. Crisalle for serving as my external comm ittee member from chemical engineering. I credit Dr. Crisalle for his work as Gradua te Admissions Coordinator for the Chemical Engineering Department. If not for Dr. Crisalle, I would never have ended up at UF. I thank Dr. Karen Waldrip for all of her support and guidance during my time at Sandia. Many of my successes in graduate school were due, in large part, to Karens tireless efforts. I thank Dr. Stephen Lee and Dr. Daniel Koleske for their significant contributions to my research efforts. I further acknowledge Dr. Andrew Allerman for his thorough feedback and instilling strong sense of attention to details in me. I acknowledge Dr. Brent Gila for his contributions to the work in this dissertation, and I thank Dr. Andrea Onstine and Danielle Stodilka for their support and collaboration for various classes and qualifying exams. I especially thank Dr. Jerry Thaler for being a reliable liaison in Florida while trying to complete the bureaucratic requirements for graduation from New Mexico. I gratefully acknowledge Jerrys efforts in this regard. I also thank Jerry, Dr. Omar Bchir, Jeff Shar p, Brandy Colwell, and Sarah Rich for their camaraderie throughout my graduate school experience. iii

PAGE 4

I acknowledge Dr. Sam Kozak for the inspiration and advice he provided during my undergraduate days. I also thank Dr. Ken Van Ness and Dr. Erich Uffelman for serving as my advisors at Washington and Lee. Both were very important to my intellectual development. I also thank Dr. Douglas Szajda, Dr. Marcia France, and Dr. Steven Desjardins. All were superb professors. Dating back to high school, I must acknowledge Mr. James Morrison for piquing my interest in science and tolerating me for three years. I also acknowledge Mr. Robert Bulkeley for doing an outstanding job of advising me throughout high school. Finally, and most importantly, I thank my family for their undying support throughout my graduate school experience. iv

PAGE 5

TABLE OF CONTENTS page ACKNOWLEDGMENTS .................................................................................................iii LIST OF FIGURES .........................................................................................................viii ABSTRACT.......................................................................................................................xi CHAPTER 1 INTRODUCTION........................................................................................................1 Background and Motivation.........................................................................................1 Electronic Properties of III-Nitrides.............................................................................1 Chemical, Structural, and Physical Properties of III-Nitrides......................................2 Suitability of III-Nitrides in Optoelectronic Devices...................................................2 Suitability of III-Nitrides in Electronic Devices...........................................................3 Progress in III-Nitride Material Development..............................................................5 Availability of III-Nitride Based Devices.....................................................................6 Statement of the Problem..............................................................................................6 Outline..........................................................................................................................7 2 EXPERIMENTAL METHODS..................................................................................10 Growth Techniques.....................................................................................................10 Metalorganic Chemical Vapor Deposition..........................................................10 Molecular Beam Epitaxy.....................................................................................14 Gallium Nitride Bulk Growth..............................................................................15 Characterization Techniques......................................................................................16 High-Resolution X-Ray Diffraction....................................................................16 Transmission Electron Microscopy.....................................................................17 Secondary Ion Mass Spectroscopy......................................................................18 Superconducting Quantum Interference Device Magnetometry.........................18 Capacitance-Voltage............................................................................................19 Inductively-Coupled Plasma-Mass Spectroscopy...............................................19 Scanning Electron Microscopy............................................................................20 Electron Dispersive X-Ray Spectroscopy...........................................................20 Auger Electron Spectroscopy..............................................................................21 v

PAGE 6

3 DETERMINATION OF DISLOCATION DENSITY OF III-NITRIDE FILMS BY X-RAY DIFFRACTION............................................................................................23 Introduction.................................................................................................................23 Characterization of Dislocations.........................................................................23 Previous Work.....................................................................................................24 Twist....................................................................................................................25 Incoherence..........................................................................................................26 Focus....................................................................................................................26 Experimental Methods................................................................................................27 Method........................................................................................................................29 Sources of Broadening........................................................................................29 Model...................................................................................................................31 Determination of Convolution.............................................................................32 Calculation of Dislocation Density.....................................................................33 Simplified Model.................................................................................................34 Results and Discussion...............................................................................................35 Summary.....................................................................................................................37 4 EFFECT OF DISLOCATIONS ON IMPURITY INCORPORATION IN METALORGANIC CHEMICAL VAPOR DEPOSITION GROWTH.....................54 Introduction.................................................................................................................54 Silicon..................................................................................................................54 Oxygen................................................................................................................55 Carbon.................................................................................................................56 Previous Work.....................................................................................................57 Focus....................................................................................................................58 Experimental Methods................................................................................................58 Results and Discussion...............................................................................................59 Summary.....................................................................................................................61 5 GALLIUM NITRIDE BULK CRYSTAL GROWTH BY DISSOLUTION AND RECRYSTALLIZATION OF GALLIUM NITRIDE POWDER..............................72 Introduction.................................................................................................................72 Previous Work.....................................................................................................72 Proposed Methods......................................................................................................74 Electroplating......................................................................................................74 Electrochemical Reduction of Nitrogen Gas.......................................................75 Dissolution and Recrystallization of Gallium Nitride Crystals...........................75 Focus...........................................................................................................................76 Experimental Methods................................................................................................76 vi

PAGE 7

Results and Discussion...............................................................................................77 Dissolution of Gallium Nitride in Lithium Chloride...........................................77 Gallium Nitride Recrystallization........................................................................78 Summary.....................................................................................................................80 6 SUMMARY AND FUTURE DIRECTIONS.............................................................97 LIST OF REFERENCES.................................................................................................101 BIOGRAPHICAL SKETCH...........................................................................................105 vii

PAGE 8

LIST OF FIGURES Figure page 1-1 Band gap engineering diagram for III-V materials....................................................8 2-1 Schematic of metalorganic chemical vapor deposition operation............................22 3-1 Schematic of reciprocal lattice points probed..........................................................39 3-2 Adjacent domains tilted with respect to one other...................................................40 3-3 Adjacent domains twisted with respect to one another............................................41 3-4 Cross-section transmission electron microscopy image of a low dislocation density sample.......................................................................................................................42 3-5 Cross-section transmission electron microscopy image of a high dislocation density sample.......................................................................................................................43 3-6 Weak beam transmission electron microscopy image and convergent electron beam diffraction image......................................................................................................44 3-7 Broadening effects demonstrated in reciprocal space..............................................45 3-8 Plotted data of full width at half maximum vs. reciprocal lattice vector magnitude46 3-9 Fit of full width at half maximum data vs. angle of inclination...............................47 3-10 Mathematical and geometric derivation of the random distribution method for calculating dislocation density.................................................................................48 3-11 Calculation of dislocation density based on values for coherence length................49 3-12 Comparison of methods of determining dislocation density by XRD.....................50 3-13 Dislocation density determined by transmission electron microscopy vs. dislocation density determined by x-ray diffraction...................................................................51 3-14 Dependence of dislocation density on growth pressure...........................................52 3-15 Effect of dislocation density on room temperature mobility of gallium nitride high electron mobility transistors.....................................................................................53 viii

PAGE 9

4-1 Impurity profiles for low dislocation density at the top...........................................62 4-2 Impurity profiles for mid dislocation density at the top...........................................63 4-3 Impurity profiles for high dislocation density at the top..........................................64 4-4 Impurity profiles for high dislocation density at the flat..........................................65 4-5 Impurity profiles for second series of overgrowths for low dislocation density at the flat.............................................................................................................................66 4-6 Impurity profiles for second series of overgrowths for mid dislocation density at the flat.............................................................................................................................67 4-7 Impurity profiles for second series of overgrowths for high dislocation density at the flat.......................................................................................................................68 4-8 Comparison of impurity profiles for different dislocation density templates..........69 5-1 Gallium nitride crystal with dimensions of 0.9 mm 0.6 mm................................82 5-2 Ideal solubility of gallium nitride.............................................................................83 5-3 Experimentally determined solubility of gallium nitride in lithium chloride compared to the ideal solubility curve.....................................................................84 5-4 Experimentally determined solubility determined after employing a 2 m filter....85 5-5 Scanning electron microscopy image of a bare gallium nitride surface...................86 5-6. Scanning electron microscopy image of a gallium nitride surface exposed to gallium nitride-in-lithium chloride melt...................................................................87 5-7 Close-up scanning electron microscopy image of a gallium nitride crystallite.......88 5-8 Scanning electron microscopy image of a treated gallium nitride surface after exposure to the gallium nitride-in-lithium chloride melt.........................................89 5-9 Scanning electron microscopy image of an untreated gallium nitride surface after exposure to a gallium nitride-in-lithium chloride melt............................................90 5-10 Scanning electron microscopy image focusing on needles on the untreated gallium nitride surface at 750..............................................................................................91 5-11 Scanning electron microscopy image focusing on needles of the untreated gallium nitride surface at 1900............................................................................................92 5-12 Election dispersive spectroscopy images for elemental analysis on needle structures..................................................................................................................93 ix

PAGE 10

5-13 Electron dispersive spectroscopy images of a patch on the gallium nitride surface94 5-14 Scanning electron microscopy image of a deposited gallium nitride crystallite with a thickness of 5 m...................................................................................................95 5-15 Electron dispersive spectroscopy instrument depth calibration...............................96 x

PAGE 11

Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy EFFECTS OF DISLOCATIONS ON ELECTRONIC PROPERTIES OF III-NITRIDE MATERIALS By Allen M. West May 2005 Chair: Cammy Abernathy Major Department: Materials Science and Engineering By varying metalorganic chemical vapor deposition (MOCVD) growth conditions, III-nitride samples with dislocation densities ranging over two orders of magnitude were grown, and the effects of dislocations were studied. Initially, a novel technique to determine dislocation density by high-resolution x-ray diffraction (HRXRD) was developed. Results were compared to those obtained by transmission electron microscopy (TEM) and agreed within a factor of 1.5. This HRXRD-based technique provided rapid and accurate feedback for use in growth-optimization studies; and became a basis for future studies of the effects of dislocation on specific phenomena associated with III-nitride technology. This included a study of the effect of dislocations on impurity incorporation during MOCVD growth. Dislocations were previously conjectured to provide energetically favorable sites for impurity incorporation, but experimental results did not confirm this. xi

PAGE 12

Finally, a novel growth technique was investigated for gallium nitride (GaN) bulk crystals, based on a simple dissolution/recrystallization process. GaN was successfully dissolved in a molten LiCl medium, and then recrystallized on a nominally homoepitaxial MOCVD-grown GaN template. This provides the basis for a potentially viable method to fabricate high quality GaN bulk crystals for homoepitaxial growth. xii

PAGE 13

CHAPTER 1 INTRODUCTION Background and Motivation Gallium nitride (GaN) and III-nitride alloys (Al-, Ga-, In-N) demonstrate outstanding potential to significantly advance solid-state electronic and optoelectronic technologies. Currently, solid-state electronics comprise a multi-hundred billion-dollar industry, and consist primarily of silicon (Si) and gallium arsenide (GaAs)-based devices. These materials, however, are intrinsically limited for certain applications (namely, high-power electronics and low-wavelength optoelectronics). Such applications have far-reaching implications, with significant promise for commercial, military, and national security purposes. III-nitride-based materials are ideal for these applications because of their favorable intrinsic electronic, structural, and chemical characteristics. Because of these properties, GaN and related alloys have recently been subject to intense research; and scientists have made significant progress in advancing III-nitride technology. Electronic Properties of III-Nitrides Electronically, III-nitrides have a large, direct, tunable band gap, which makes these materials ideally suited for many technologically important electronic and optoelectronic applications. The band gap for III-nitrides ranges from 0.7 eV for indium nitride (InN) [1] to 6.2 eV for aluminum nitride (AlN) [2]. Alloys throughout this entire range are possible to achieve, depending on the composition of group III element (Al, Ga, In). III-nitride materials large band gap gives them distinct capabilities for use in high-power electronic and optoelectronic applications. The tunable nature of the band 1

PAGE 14

2 gap in III-nitrides provides the possibility for large band offsets, which enables greater control over carrier confinement, and better performance in some devices, including HEMTs and multi-quantum well (MQW) LEDs. Chemical, Structural, and Physical Properties of III-Nitrides III-nitrides offer distinct advantages because of their chemical, structural, and physical properties. They are outstandingly hard, corrosion-resistant in most environments, and remarkably stable at high temperatures. III-nitrides exhibit the wurtzite crystal structure, which provides a significant spontaneous polarization field along the c-axis, and a high piezoelectric polarization field [3]. Together, these fields enable greater sheet charge densities in III-nitride heterostructures strained in compression; thus, allowing greater capabilities for III-nitride based electronic devices. III-nitride materials are also significantly more thermally conductive than their III-arsenide counterparts and silicon, allowing for efficient heat removal. This saves substantial resources by reducing the need for external cooling. Silicon-based devices burn out at temperatures exceeding 140C, necessitating the installation of powerful cooling fans for silicon-based devices, for high-power applications. GaN-based devices are operable in environments >300C [3], so the high temperature tolerance and the efficient heat removal of III-nitrides make this material system ideal for high-temperature, high power applications. Suitability of III-Nitrides in Optoelectronic Devices III-nitrides represent one of few known material systems that allow efficient emission of radiation in the green, blue, and deep ultraviolet range. Figure 1-1 shows the band gap dependence on the lattice constant, and the range of achievable wavelengths for

PAGE 15

3 III-nitride emitters. III-nitride LEDs and LDs are exceptionally tolerant of defects, as they show efficient emission of radiation despite high dislocation densities ( = 10 8 11 cm -2 ). Because of the unique capabilities of III-nitride optoelectronic devices, they are ideal for commercial, military, and national-defense applications. Commercially, monochromatic III-nitride LEDs are used for illumination purposes in traffic signals and outdoor displays. One monochromatic traffic light saves 800 kWh/year in power, compared to a typical incandescent white bulb with a monochromatic shutter [6]. III-nitride-based UV VCSELs are also being developed to pump red-green-blue (RGB) phosphors for significantly more efficient and cost-effective white-lighting purposes. This could provide energy savings estimated at $100 billion annually. Furthermore, blue LDs are used commercially low-cost, ultra-high-density optical data storage in digital versatile disks (DVDs) [7] that can store >1 gigabit/cm 2 of information [6]. Militarily, III-nitride photodetectors are currently being researched for use in solar blind detectors for missile warning systems and for malignant biological and chemical agent detection [8,9]. Suitability of III-Nitrides in Electronic Devices Table 1-1 shows critical electronic properties of several semiconductor materials. The intrinsic electronic properties of GaN and AlGaN/GaN heterostructures allow significantly more power output than any other known semiconductor material system. The large, tunable band gap of III-nitrides makes these materials ideal for use in certain electronic devices, such as high electron mobility transistors (HEMTs). Breakdown voltage (V b ) goes as the square of the band-gap energy, giving GaN and AlN-based

PAGE 16

4 electronic devices a distinct advantage over SiC (band gap, E g = 3.2 eV) in terms of power switching, enabling a device to achieve large power density at high frequencies [3,7]. The theoretical maximum power output achievable is directly proportional to the band gap to the fourth power (P max E g 4 ), and the saturation velocity squared (P max v s 2 ). Because of the large band gap and saturation velocity, III-nitride based devices are ideal for high-power applications. For a HEMT device, the output power is shown in Equation 1-1, where I max is the maximum current possible for a particular device. 8maxIVPbout (1-1) Higher saturation currents are achievable with higher carrier concentrations, and the polarization effects of AlGaN/GaN heterostructures provide outstandingly high sheet densities in a two-dimensional electron gas (2DEG). Furthermore, the polarization fields in GaN HEMT structures remove the need for external doping to achieve very high sheet densities. This ability to operate without ionized impurities diminishes the problem of gate leakage in Schottky diodes [5]. Finally, for operable high-power electronic devices, heat dissipation is crucial, since such devices are thermally limited. The high thermal conductivity of GaN enables these HEMT devices to operate more efficiently and more cost-effectively [3]. These characteristics make III-nitrides ideal for use in high-power devices for both commercial and military applications. Commercially, GaN-based HEMTs can be used in cellular communications base stations, telecommunications, hybrid electric cars, and switches for high-power electric grids. For military purposes, GaN-based HEMTs are candidates as amplifiers for synthetic aperture radar (SAR), smart weapons, electronic

PAGE 17

5 warfare, and surveillance. They will eventually replace bulky traveling wave tubes, enabling significantly higher output power, with added benefits of reduced mass, volume, cost, and cooling requirements. Progress in III-Nitride Material Development Performance of III-nitride based devices has improved by leaps and bounds in recent years due to growth and processing of these materials becoming more mature. As growth techniques have progressed, microstructural quality of III-nitride films used in devices has improved; and performance of both electronic [10] and optoelectronic devices has improved accordingly [11]. Optoelectronically, the first nitride-based blue LEDs were fabricated in the 1970s, when the external efficiency hardly exceeded 0.1%. External efficiencies improved with novel growth breakthroughs throughout the 1980s, and reached 1.5% by 1992. As growth of III-nitride films continued to mature, this value reached 20% by 2002 [6]. Further strides with III-nitride optoelectronic devices will be made as microstructural quality of III-nitride films continues to improve. GaN-based electronic devices have experienced similar improvements as III-nitride technology has become more advanced. The earliest reported GaN HEMT state-of-the-art was 1.1 W/mm in 1995, and exceeded 10 W/mm by 2001 [10]. Since 2001, GaN-based technology has progressed rapidly, and the current state-of-the-art GaN HEMT power density is an astounding 32.2 W/mm [12]. Conversely, GaAs-based HEMTs have attained merely 1 W/mm for power density, despite the fact that GaAs-based technology is significantly better studied.

PAGE 18

6 Availability of III-Nitride Based Devices Considering these substantial advantages and the potential widespread use of GaN-based devices, commercial devices based on GaN are currently in the research stage. Comprehensive studies of IIInitride materials are underway, in order to better exploit these materials advantages, where considerable growth and processing issues are being addressed. The most significant of these issues involves the effects of dislocations on III-nitride thin films. Because there are no commercially available substrates that allow for homoepitaxial growth of IIInitride thin films, these devices must be grown heteroepitaxially on sapphire or silicon carbide, at high lattice mismatch. This causes extremely high dislocation densities (=10 8 11 cm -2 ). Statement of the Problem Currently, there is much disagreement concerning the exact effects of dislocations on III-nitride electrical properties and device performance. To advance the science of IIInitrides, and consequently IIInitride based device performance, the precise effects of dislocations on electronic properties must be determined. In other semiconductor material systems (such as Si and GaAs), a dislocation density of ~10 4 cm -2 renders a device inoperable. IIInitride films with dislocation densities of ~10 11 cm -2 have been used in operable devices with surprisingly high efficiency. Because of this high tolerance for defects, some researchers have postulated that dislocations are electrically inactive in III-nitride materials [13]. Others, however, have indicated that dislocations have a deleterious effect on device performance [14], and will significantly affect reliability and scalability of these devices. Determination of the effects of dislocation density on

PAGE 19

7 electronic properties would greatly aid optimization of IIInitride device performance, especially as long as these devices are grown heteroepitaxially. Outline The remainder of this dissertation is structured as follows: Chapter 2 describes the growth and characterization techniques used in this work. Chapter 3 illustrates one technique in depth, based on x-ray diffraction (XRD) to determine dislocation density in IIInitride thin films. This method develops a theoretical model to determine dislocation density from XRD results; and allows for rapid, accurate feedback for dislocation density. This is extremely helpful in optimization studies for metalorganic chemical vapor deposition (MOCVD) IIInitride growth studies. Chapter 4 describes a study on the effect of dislocation density on impurity incorporation in GaN films during MOCVD growth. Impurities can dramatically affect electronic properties of GaN, by forming deep trap states and acting as scattering centers. Understanding how dislocations affect impurity incorporation is critical in optimizing MOCVD growth parameters for device structures. Chapter 5 describes an investigation into a novel method to fabricate IIInitride bulk crystals. Bulk IIInitride substrates would allow homoepitaxial growth of IIInitride thin films, which would minimize dislocation density for device structures. This would greatly enhance performance for devices for which dislocations have a deleterious effect. Finally, Chapter 6 summarizes significant results and recommends future work for these projects.

PAGE 20

8 0.2 0.4 0.3 5.2 4.8 5.0 4.6 6.6 6.4 Energy Gap (eV) Lattice Constant (Angstroms) 6.2 6.0 5.8 5.6 5.4 Wavelength (m ) 0.5 5.0 2.0 1.0 0.0 4.2 4.4 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 5.5 6.0 Arsenides (near-IR) Antimonides (far/mid-IR) Phosphides (yellow/red) InSb GaSb InAs InP AlAs GaAs GaP AlSb AlN GaN InN Nitrides (blue/UV) Figure 1-1. Band gap engineering diagram for III-V materials. III-nitrides are ideal candidates for use as green, blue, and deep-UV light emitters.

PAGE 21

9 Table 1-1. Electronic properties of semiconductor materials used in electronic device applications. Semiconductor Material Si GaAs SiC GaN Band Gap (eV) 1.1 1.4 3.3 3.4 Saturated Electron Velocity (10 7 cm/s) 1 2.1 2 2.7 Breakdown Field (MV/cm) 0.3 0.4 2 3.3 Thermal Conductivity (W/cm-K) 1.5 0.5 4.5 >1.7 Bulk Electron Mobility (cm 2 /V-s) 1500 8500 700 1200 Heterostructure None AlGaAs/InGaAs None AlGaN/GaN 2DEG Sheet Density (10 12 cm -2 ) NA <4 NA 20

PAGE 22

CHAPTER 2 EXPERIMENTAL METHODS Chapter 1 showed the technological importance of III-nitride-based technology and identified the fundamental problem of high dislocation density. Chapter 2 describes the experimental procedures involved in growth and characterization of III-nitride materials used in these studies. This includes a description of the two different film-growth techniques, specifically metalorganic chemical vapor deposition (MOCVD) and molecular beam epitaxy (MBE), followed by a description of a novel bulk-growth method for GaN. The characterization techniques used are also explained. Growth Techniques Both MOCVD and MBE were used to accomplish different goals for growing films in this work. Growth of III-nitride materials is relatively new, because of the lack of III-nitride lattice-matched substrates and the difficulty associated with incorporation of nitrogen. For MOCVD, ammonia (NH 3 ) is the most prevalent precursor; but its stability makes it difficult to incorporate nitrogen, and requires high growth temperatures (usually ~1050C). For MBE growth of III-nitrides, the nitrogen source is a reactive nitrogen plasma, which consists of nitrogen gas and an electron cyclotron resonance (ECR) plasma source. Metalorganic Chemical Vapor Deposition MOCVD offers the following advantages over MBE: high growth rates, high throughput, ability to grow at high temperatures, external feeds for precursors, high 10

PAGE 23

11 growth pressures, and lower costs associated with this technique over MBE. Higher growth rates decrease the time required to perform material optimization studies, and higher throughput is desirable when fabricating commercial devices, once the material-optimization growth studies are complete. Growth at higher temperatures enables more effective cracking of nitrogen precursor to incorporate nitrogen into the film; which, in turn, results in higher growth rates. The external feed lines that provide the precursors are advantageous, since they allow the growth chamber to be isolated when replacing the sources on depletion. This is a significant advantage over MBE, since replacing the solid sources requires venting the growth chamber to atmosphere. Baking out the impurities from this exposure can take up to 2 weeks, whereas changing group III gas bubblers or ammonia sources for MOCVD usually takes less than 1 hour. The growth pressures associated with MOCVD are relatively high (usually 40 to 500 Torr for III-nitride growth) compared to those of MBE (<10 -6 Torr). This removes the need for ultra high vacuum (UHV) conditions, which require diffusion and cryogenic pumps. Diffusion pumps are problematic because they are susceptible to back-streaming oil into the growth chamber, which may incorporate into the film. Cryogenic pumps require a constant flow of liquid nitrogen (L-N 2 ), which can be expensive, depending on the time and frequency of growth experiments. Finally, for all of the advantages that MBE provides for devices in other material systems, MOCVD-grown III-nitride films for electronic devices have demonstrated superior electronic properties and device performance over their MBE grown counterparts [18]. The MOCVD growths were performed in an EMCORE D-125 short jar reactor with the wafers placed on a silicon carbide-coated graphite rotating platen, containing up

PAGE 24

12 to three wafers. There are several advantages to this type of reactor. First, it allows for a high degree of aluminum incorporation into films, since the short jar design significantly reduces the parasitic pre-reactions of the aluminum precursor with ammonia due to upstream heating. Second, the multi-wafer EMCORE D-125 allows a greater throughput of device-quality films once the growth conditions are optimized. Third, this reactor allows for excellent thermal control and uniformity across the platen, and therefore across the individual wafers. This is especially useful when growing a film that is temperature sensitive. Fourth, the rotation rate of the platen can reach 2100 revolutions per minute (rpm). This allows a high degree of control over the boundary layer thickness, which could improve growth uniformity and growth rate. Growth uniformity can be monitored in-situ, since the EMCORE D-125 contains three view ports where it is possible to monitor the film growth at different points across the wafers on the platen. Finally, this reactor provides a load-lock, directly connected to a turbo-molecular pump (TMP) backed by a dry-scroll rough pump. This allows the reaction chamber to remain under vacuum during sample exchange, which reduces the amount of contamination during film growth. Figure 2-1 is a schematic of the MOCVD operation. The precursors used for the films are trimethyl gallium (TMG) and trimethyl aluminum (TMA) for gallium and aluminum respectively, and ammonia (NH 3 ) for nitrogen, with hydrogen (H 2 ) and nitrogen (N 2 ) serving as carrier gases. The precursors undergo the following reactions: (CH 3 ) 3 Ga (g) + NH 3(g) GaN (s) + 3CH 4(g) (CH 3 ) 3 Al (g) + NH 3(g) AlN (s) + 3CH 4(g)

PAGE 25

13 The solid III-nitride film deposits on the substrate, while the gaseous byproducts are pulled through an EBARA rough pump, and out through the exhaust. In all likelihood, the reactions rarely occur as shown above, as the reactions shown above are oversimplified. Many side reactions and parasitic pre-reactions complicate the MOCVD film deposition process; but overall, the reactions shown above are a useful simplification. The parameters that affect the growth of the film are temperature and pressure, as well as flows of metalorganics, ammonia, and carrier gases. One challenge posed by the MOCVD growth technique is determining which parameter(s) to vary, in order to grow the film with the desired qualities. Since our goal was to determine the effects of dislocations on electronic properties and device performance for III-nitride materials, it was necessary to grow films over a wide range of dislocation densities by MOCVD. The MOCVD growth parameters that were varied include the type of nucleation layer (AlN vs. GaN), growth temperature, and growth pressure, in order to grow films of varying dislocation density. To further expand this study, two different substrates were used: silicon carbide (SiC) and sapphire (Al 2 O 3 ). SiC has a less severe lattice mismatch than sapphire with III-nitrides (3.5% vs. 16% for GaN and 13.29% vs. 1% for AlN) [19]; and therefore, enables experimenters to grow lower-dislocation-density-films. Sapphire is more widely used, because it is significantly less expensive; insulating sapphire wafers are $60 each, while device-grade SiC wafers are $5000 each. For growths on SiC, a 1,400 high-temperature (1125C) AlN nucleation layer was first deposited, followed by a GaN buffer layer at 1050C. The pressure for these growths varied from 100 Torr. For growths on sapphire, more variables were

PAGE 26

14 explored. For some films, a low-temperature (600C) AlN nucleation layer was first deposited; while others had a low temperature (530C) GaN nucleation layer, depending on the desired dislocation density. The pressure of the overlying GaN layers varied from 20 Torr, and the temperature of this layer varied from 1040C. Another factor that affected the film microstructure was recovery time. A fast recovery provided a film with a higher dislocation density, while a longer delay between the growth of the nucleation layer and the GaN over-layer provided a film with a higher microstructural quality [22]. By altering the growth conditions as needed, it was possible to grow GaN films with a very broad range of dislocation densities. The material grown for our study ranged from 4.2110 8 cm -2 1.0710 11 cm -2 This allowed for thorough studies of the effects of microstructure on electronic and magnetic properties, as well as device performance and impurity incorporation. Molecular Beam Epitaxy Molecular beam epitaxy (MBE) is advantageous when abrupt interfaces and precise control of layer thickness are required. MBE is also an inherently safer process when growing III-nitride films, since all precursors used in MOCVD are either toxic or pyrophilic; whereas none of the precursors used in MBE present such hazards. This removes the need for strict engineering safety controls and gas sensors that are required for MOCVD. For our study, MBE offered a distinct advantage when growing GaN films with transition metal dopants and when growing oxide films. MBE is better suited for transition-metal dopant growths, because of the difficulty associated with obtaining an

PAGE 27

15 MOCVD transition-metal source that can incorporate efficiently into a III-nitride film and that has a high enough vapor pressure to provide a substantial molar flux in the carrier gas. MBE is also better for oxide growths, because of the lack of oxygen sources for MOCVD. For the MBE growths in our study, two series of films were grown: chromium (Cr)-doped aluminum nitride AlN and scandium oxide (Sc 2 O 3 ). For the Cr-doped AlN films, the group III precursor was solid 99.99999% pure (7N) aluminum metal, the dopant source was 5N solid Cr, and the nitrogen precursor was 6N nitrogen gas. To create reactive nitrogen radicals, the nitrogen gas was ionized by a radiofrequency (RF) plasma unit with a frequency of 13.56 MHz. For the scandium oxide growths, the precursors were solid 4N scandium metal and 6N oxygen gas, which was ionized at 13.56 MHz. All of the solid sources were evaporated by resistively heating poly-boronitride (PBN) containers, and the atomic fluxes were controlled by adjusting the shutter apertures. Gallium Nitride Bulk Growth The GaN bulk growth studies were performed in a home-built stainless-steel container, fitted for a 1 diameter quartz tube. The method to grow GaN bulk crystals was simply a dissolution/recrystallization process of polycrystalline GaN powder in a molten lithium chloride (LiCl) solvent. The chamber was resistively heated up to ~1000C to effectively dissolve GaN powder and maximize deposition of GaN crystallites. Procedurally, the solubility was tested with respect to temperature. Once it was determined that GaN dissolved sufficiently well at achievable temperatures, the

PAGE 28

16 experiment was repeated to test the recrystallization of GaN on a nominally homoepitaxial template of MOCVD-grown GaN on sapphire substrate. Other effects such as convective fluid dynamics and surface treatment were tested to maximize growth on the surface. Characterization Techniques After the MOCVD, MBE, and bulk growths, samples were removed from their respective growth chambers; and their properties were assessed using several techniques. Specifically, the microstructural, magnetic, and electronic characteristics were determined. All of the techniques used are described in the subsections to follow. High-Resolution X-Ray Diffraction High-Resolution X-Ray Diffraction (HRXRD) is a powerful technique that provides information about a materials microstructure and lattice constant. Our study used only rocking curves (-scans) to ascertain the microstructure of the GaN thin films. The microstructural quality of a thin film is closely related to the width of the diffracted x-ray intensity, with the full-width at half maximum (FWHM) used as the figure of merit. Since HRXRD is a diffraction phenomenon, its measurements must be taken with respect to reciprocal space, where reciprocal lattice points (RLPs) correspond to specific lattice planes. If it were possible to perform HRXRD analysis on a defect-free crystal of infinite dimensions, the results would be a series of delta functions at every RLP. Once finite dimensions and defects are imposed on the crystal, broadening can be observed and measured by HRXRD. The extent to which broadening is observed is indicative of the microstructural quality; the higher the measured FWHM, the higher the dislocation density.

PAGE 29

17 All of the HRXRD measurements were made using a Philips (Almelo, Holland) XPert MRD high-resolution triple-axis x-ray diffractometer with a sealed Cu anode collimated by a four-crystal Ge(220) monochromator. The Cu anode produced an x-ray wavelength of Cu k )=1.5404 Transmission Electron Microscopy Transmission Electron Microscopy (TEM) is a universally accepted method to determine dislocation density in materials. TEM operates by emitting a focused electron beam through a sample, into a series of magnetic lenses, and into a fluorescent screen detector. It is a robust technique for determining microstructural characteristics of materials, since it provides high spacial resolution (with a magnification of 10 6 ). Using this technique, dislocation density is simply computed as the number of dislocations counted over a known measured area. TEM is a destructive technique, since sample preparation requires a thickness that is transparent to the electron beam (<3000 ). The microscopy was performed with a Philips CM20 instrument at 200kV. Some specimens were prepared in plan-view while others were prepared in cross section. The samples observed in cross-section were prepared by a focused ion beam; while those observed in plan-view were prepared by a traditional mechanical polishing, dimple, and ion mill procedure.

PAGE 30

18 Secondary Ion Mass Spectroscopy Secondary Ion Mass Spectroscopy (SIMS) is a superior method to determine impurity concentrations with respect to film depth. It is a destructive technique, but is known to provide accurate results for impurity profiles. SIMS results are obtained by accelerating heavy metal ions at a target sample, sputtering off constituent atoms. These atoms from the target sample become ionized, and are attracted to a charged detector. The SIMS work in our study was performed in a PHI Quadrupole SIMS spectrometer (Charles Evans and Associates, Cambridge, MA). The sputtering ion was cesium (Cs + ), with a primary ion energy of 4keV. The cesium angle of incidence was 60, and the instrument was optimized for sensitivity for carbon, oxygen, and silicon. Detection limits were 110 16 cm -3 for carbon, 510 15 cm -3 for oxygen, and 610 15 cm -3 for silicon. Superconducting Quantum Interference Device Magnetometry A Superconducting Quantum Interference Device Magnetometry (SQUID) Quantum Design Magnetic Properties Measurement System was used to ascertain magnetic properties of DMS materials. Hysteresis loops from magnetization vs. field measurements verify that a sample is ferromagnetic. The second technique to determine magnetic properties was Field-Cooled/Zero Field-Cooled (FC/ZFC) measurement. This technique involves cooling a sample to 10K under an applied field while the samples magnetization is measured (FC), and again while the samples temperature increases to room temperature (ZFC).

PAGE 31

19 Capacitance-Voltage Capacitance-voltage (CV) is a non-destructive technique that provides information about a films electrical properties. In our study, a mercury (Hg) contact CV was used to probe for a buried conductive layer in a nominally insulating GaN film, and to confirm the existence of a conductive 2DEG in a AlGaN/GaN HEMT structure. The depletion region of a HEMT structure is modeled as a parallel plate capacitor, where the mercury droplet acts as a metal Schottky contact, and the sheet density of a 2DEG is given as 123)(rVCAqCxN (2-1) where C is the capacitance, q is the carrier charge, is the permittivity, A is the area of the mercury drop, and V r is the reverse bias voltage. The apparatus used for experimentation was an MSI Electronic Model 412-2L mercury probe, with a droplet diameter of 2.0310 -3 cm -2 Electronic measurements were obtained by a Hewlett Packard 4284A Precision Inductance-Capacitance-Resistance meter, operating at 1 MHz. Inductively Coupled Plasma-Mass Spectroscopy Inductively coupled plasma-mass spectroscopy (ICP-MS) is a powerful technique to determine the existence and concentrations of trace elements. This technique operates by injecting an aerosolized liquid sample into an argon plasma at a temperature of 7000 K. When the atoms from the sample collide with the energetic argon ions, the become charged. The ions are then drawn into a high vacuum chamber with a quadrupole mass analyzer. The quadrupole filters out atoms of undesired masses by alternating

PAGE 32

20 radiofrequency (RF) and direct current (DC) fields, so that only atoms of the desired mass can pass into the detector. The detector is calibrated by correlating the intensity of the signal created by these collisions with the detector to a set of standards with known concentrations. In this study, a Perkin Elmer (Boston, MA) Elan 6100 instrument was used to determine the concentration of gallium and lithium in samples of GaN dissolved in LiCl. Scanning Electron Microscopy Scanning electron microscopy (SEM) is a useful technique to determine morphological information on a surface at extremely high magnifications (10,000). SEM operates by emitting an electron beam at the sample surface in a vacuum. When this electron beam hits the surface, secondary electrons are emitted and collected in a detector, which forms an image, giving precise topographical information. The instrument used was a JEOL (Peabody, MA) JSM5800 tungsten filament scanning electron microscope. The electron beam energies ranged from 5 keV. Electron Dispersive X-Ray Spectroscopy Electron dispersive x-ray spectroscopy (EDS) is used in conjunction with SEM to provide information about elemental analysis on surface features of a sample. Other than emitting secondary electrons when exposed to an incident electron beam, a sample will also emit x-rays. This is because, when inner shell electrons are ejected from an atom, electrons at high energy shells will fill the inner shell vacancy, which emits a photon. The energy of the photon emitted is related to the element. EDS measures the energy and intensity of the emitted radiation, which provides qualitative and quantitative analysis of

PAGE 33

21 surface impurities. The EDS data in this study was obtained using an EDAX Phoenix system, fitted with a 3,000 window, allowing for analysis of low-Z elements. Auger Electron Spectroscopy Auger Electron Spectroscopy (AES) was employed for surface elemental analysis. AES operates by emitting an incident electron beam at a sample, which causes surface atoms to emit electrons of an energy that is characteristic of the atom. This technique operates in the following manner: an incident electron knocks off an inner shell electron on the sample surface, which causes an electron of higher energy to fall into this inner shell. This causes emission of a photon, which is reabsorbed by the surface atom, and causes an outer shell electron to be emitted. It is this emitted electron that is detected, and its characteristic energy is indicative of its atom. The Auger instrument used in this research was a Perkin Elmer 6600, that is capable of providing compositional data for elemental analysis for samples of 1 atomic %.

PAGE 34

22 Cold Source Gases Reactants Reaction Z one Gaseous Products Substrate Heat Flux Figure 2-1. Schematic of metalorganic chemical vapor deposition operation. Cold precursors and carrier gases flow over a heated substrate. Precursor gases undergo a heterogeneous reaction at the substrate surface, depositing a solid film, while gaseous byproducts are pumped out through the system exhaust.

PAGE 35

CHAPTER 3 DETERMINATION OF DISLOCATION DENSITY OF III-NITRIDE FILMS BY XRAY DIFFRACTION Introduction There is currently no universally accepted method that allows for rapid and accurate determination of threading dislocation density in III-nitride thin films. X-ray diffraction is a promising technique to determine dislocation density because it provides results quickly and nondestructively, and when modeled correctly, allows for accurate determination of microstructural parameters of crystalline materials. The goal of this chapter is to develop a method to determine dislocation density based on x-ray diffraction that provides accurate results as quickly as possible. Such a method would facilitate the establishment of a correlation between dislocation density and film properties. This information, in turn, would allow experimenters to determine how dislocations influence electrical and optical properties of device layers. Characterization of Dislocations Several characterization techniques have been employed to ascertain dislocation density in III-nitride films, including transmission electron microscopy (TEM), atomic force microscopy (AFM), cathodoluminescence (CL), and high-resolution x-ray diffraction (HRXRD). TEM is generally accepted as the most accurate and best understood method, but the sample preparation is time consuming and destructive, and some care must be used in how the dislocations are imaged and counted [23]. AFM has also been used, with the dislocations identified as slight depressions near step 23

PAGE 36

24 terminations, and these depressions have been correlated to dislocations with a screw component [24]. Dislocations with a pure edge component are occasionally observed on terraces; however, the large tip radius usually precludes direct observation. Using CL imaging, dark spots are observed on the GaN surface and have been correlated to the dislocation density in TEM samples [25]. In CL, the dislocations that act as non-radiative recombination are imaged as dark spots while dislocations that radiate may have the same intensity as the background GaN. X-ray diffraction provides quick results that are correlated to the microstructure of thin films, and the microstructure is directly related to dislocation density. Because of the limitations associated with TEM, AFM, and CL, HRXRD is an attractive alternative to ascertain dislocation density for III-nitride thin films. Previous Work Researchers began using XRD to study microstructures in metals as early as the 1950s [26]. By comparing the angular distribution of x-ray intensity for a given reflection of cold-worked vs. annealed specimens (i.e., those with high and low dislocation densities, respectively), a correlation emerged between the XRD line width and the dislocation density. Several models were proposed to calculate dislocation density based on the degree of diffracted x-ray scattering. These studies provided a valuable theoretical and experimental foundation for determination of dislocation density by XRD. In the 1990s, these classic works were applied to heteroepitaxial GaN-based thin films [30]. These modern works accounted for line width broadening due to strain, coherence length, rotations of crystallographic domains, and interdependence of these

PAGE 37

25 rotations; however, there were disagreements concerning the relative importance of each of these broadening effects when estimating threading dislocation density. As a result, there is currently no unified analytical model from which to determine dislocation density based on XRD line width results. This work resolves these differences by providing insight into each of the significant broadening mechanisms, concentrating heavily on twist and incoherence. Twist When ascertaining the crystalline quality of III-nitride films by XRD, it is crucial to determine the microstructural twist. Threading edge dislocations (b = 11-20) are often dominant in heteroepitaxial III-nitride films, and these are directly related to crystallographic twist. Recognizing this, Heying et al. [35] explicitly demonstrated the need for asymmetric scans in skew geometry in order to better characterize the dislocation character of GaN films. Several models were subsequently proposed on how to best determine twist in GaN films. Metzger et al.[30] and Kang et al.[31] proposed similar procedures that employ a -scan in order to determine twist. However, it is not possible to obtain accurate results by this method with most XRD systems because the instrumental out-of-plane broadening of a scan is ~1.5, which corresponds to a dislocation density on the order of 10 11 cm -2 thereby grossly overestimating the edge component of dislocation density based on twist. Srikant et al. [32] proposed a model that employs asymmetric rocking curves (scans), and accounts for tilt and twist dependence on both inclination angle () and measured full width at half maximum (FWHM). By treating twist as a free parameter, several scans at different angles of inclination are required for this method, with twist

PAGE 38

26 being determined by optimizing the fit to the FWHM data. This model was also the basis of the work of Chierchia et al. [33] and Sun et al. [34] to determine twist. One significant disadvantage with this method is the necessity of performing so many scans at asymmetric reflections, lengthening the measurement and analysis time. Incoherence Another phenomenon often neglected when determining dislocation density by XRD is the effect of incoherence broadening on the measured FWHM. The model proposed by Srikant et al. only considers tilt and twist contributions to broadening [32]. Chierchia et al. discussed broadening due to incoherence for symmetric scans, but did not explicitly include it when calculating dislocation density [33]. Both Chierchia et al. and Sun et al. [34] stated that their calculations of twist were overestimations because they neglected coherence length broadening upon determining twist [33]. The degree to which incoherence broadening affects the calculation of dislocation density, however, was never described in detail. Focus In this chapter, a theoretical and experimental approach is presented for using HRXRD to measure threading dislocation densities of III-nitride thin films in a rapid, non-destructive manner. This method relates HRXRD line widths to dislocation density, allowing resolution of screw and edge dislocation densities from the total dislocation content. The model developed in this work unifies and simplifies other proposed models and accounts for all significant microstructural causes for broadening. This model was verified by comparing the dislocation density results obtained by HRXRD to those obtained by TEM on III-nitride films grown by metalorganic chemical vapor deposition

PAGE 39

27 (MOCVD) on both SiC and sapphire substrates. The model was further simplified by carefully considering the magnitudes of the contributions of each broadening term, leading to accurate dislocation density determinations from as few as two HRXRD reflections. Experimental Methods To obtain the dislocation density from the HRXRD line widths, the individual broadening effects from tilt, twist, and lateral coherence length were deconvolved from a series of symmetric ((0002), (0004), (0006)) and skew-symmetric ((10-11), (20-22)) rocking curves. These reciprocal lattice points are shown in Figure 3-1. Crystallographic tilt is defined as the out-of-plane rotation of adjacent domains in the crystal, as demonstrated in Figure 3-2, and could be determined directly from the symmetric series of scans. The asymmetric reciprocal lattice points were chosen based on the significant twist component and the relatively high measured intensity associated with them. Twist is defined as the in-plane rotation of adjacent domains (Figure 3-3). Measurements were made using a Philips XPert MRD high-resolution triple-axis x-ray diffractometer with a sealed Cu anode collimated by a four-crystal Ge(220) monochromator. The Cu anode produced an x-ray wavelength of Cu k )=1.5404 Of the ten samples studied by HRXRD, seven were selected for comparison with TEM over the widest range of dislocation density. The microscopy was performed with a Philips CM20 instrument at 200kV. One specimen was prepared in plan view while the other six were prepared in cross section. Three were prepared by focused ion beam, while the balance was prepared by a traditional mechanical polishing, dimple, and ion mill

PAGE 40

28 procedure. Cross-section TEM (XTEM) images of low and high dislocation density samples are shown in Figure 3-4 and Figure 3-5, respectively. To determine dislocation densities of the samples in cross section, dislocations were imaged and counted as usual, under two-beam bright field, dark field, and weak beam (Figure 3-6a) conditions and counted as usual. Specimen thicknesses were determined by convergent beam electron diffraction (CBED) using a (11-20) two-beam condition (Figure 3-6b). The number of parallel diffraction intensity oscillations (fringes) observed in the CBED disks is related to the extinction distance g and the specimen thickness t. Cross sectional thicknesses were assigned based on an average of the first two linear fits of the fringe spacings according to the formula si2nk2+ 1 g2nk2=1 t2 where s i is the fringe spacing of the i th fringe, and n k is an integer. For more detailed information on this technique, see reference 15. The samples were grown by MOCVD in an EMCORE D-125 reactor on both silicon carbide and (0001) sapphire substrates under a variety of growth conditions in order to vary the dislocation densities; the growth temperature for all films was 1050C, and the precursors were trimethylgallium (TMG), trimethylaluminum (TMA), and ammonia (NH 3 ) with hydrogen (H 2 ) and nitrogen (N 2 ) carrier gases. The pressure was varied from 140 300 Torr. All samples analyzed were either GaN or Al x Ga (1-x) N, with aluminum content as high as 45%.

PAGE 41

29 Method There are several challenges posed by employing HRXRD to determine threading dislocation density. These involve the following: Identifying the significant sources of line width broadening. Formulating a model that accounts for these broadening effects. Establishing a procedure using XRD scans to probe the film layers, using the proper geometry. Determining the extent to which microstructural defects influence measured HRXRD line width. calculating dislocation density from this information. Computation of dislocation density is important for comparison studies between materials. Microstructural quality of III-nitride thin films is often reported by quoting the FWHM values of select reflections. However, a simple comparison of line widths for particular XRD scans may be insufficient when comparing crystalline quality of thin films. There are two reasons for this: first, dissimilar XRD systems may exhibit significantly different intrinsic broadening, which must be accounted for. Second, research groups may employ any of a variety of reflections to ascertain crystalline quality of thin films, and these reflections are affected differently by microstructural imperfections. Since there is no one universally employed XRD reflection, calculating dislocation density with a microstructural model is the only way to compare different samples conclusively. Sources of Broadening When performing HRXRD rocking curves on single crystal materials, the potential sources of broadening are due to: Crystallographic rotations. Inhomogeneous strain fields. Curvature of the film. Intrinsic rocking curve width.

PAGE 42

30 Broadening due to the monochromator. For the purposes of this study, broadening effects due to film curvature, intrinsic rocking curve width and the monochromator are neglected, since all of these effects are less than 12 arc sec. [36] Figure 3-7 demonstrates specific broadening effects in reciprocal space and how the diffractometer elucidates each of these broadening effects. Figure 3-7(a) is a view of the film in cross-section with the relevant broadening effects at an asymmetric reciprocal lattice point (RLP) in skew geometry. K hkl is the reciprocal lattice vector between the origin of reciprocal space and the RLP being probed. Its direction is related to the angle between the surface normal of the crystal and the normal of the set of diffracting planes, while its length is related to the d-spacing of the set of planes associated with the RLP. Rotations of crystallographic domains broaden the RLP transverse to K hkl while inhomogeneous strain broadens in the direction parallel to K hkl Broadening due to incoherence occurs both parallel as well as transverse to K hkl Specific broadening effects due to crystallographic rotations and incoherence are demonstrated in Figure 3-7(a). Pure domain tilt and twist are observed about the K y and K z axis, respectively. The effects of tilt and twist measured by the diffractometer are the projections of the rotations about K y and K z respectively, on the RLP. The extent to which each of these rotational broadening contributions affects the measured line width of a rocking curve (-scan) is geometrically related by hkl the angle between K z and the RLP being probed. Figure 3-7(a) also demonstrates broadening due to incoherence, as domain thickness (h) and lateral coherence length (L) are shown explicitly. Because threading dislocations propagate along the growth plane normal during III-nitride growth,

PAGE 43

31 the domain thickness is assumed to be equal to the film thickness. The lateral coherence length is defined as the average distance between threading dislocations. The tilt, twist, and lateral coherence length all affect broadening of an asymmetric RLP when a rocking curve is performed in skew geometry. Since the film thicknesses in this study were 8000 vertical coherence length had an insignificant effect on measured broadening. Furthermore, inhomogeneous strain broadening was neglected because all scans employed in this study were triple axis rocking curves, which diminish the observed broadening effects in the direction parallel to K hkl Therefore, the microstructural imperfections that significantly contributed to our measured values for FWHM were projections of rotations about both K y and K z (tilt and twist) on the probed RLP and lateral incoherence (2/L). Figure 3-7(b) shows broadening effects of the skew-symmetric RLP in plan-view. The measured broadening ( hkl ), is the measured FWHM and broadening due to tilt, twist and incoherence are indicated explicitly. These microstructural broadening effects were included in the proposed model that was used to calculate dislocation density. Model The following reciprocal space model describes the (hkl) dependence of the convolved peak-width contributions from the combined effects of tilt ( tilt ), twist ( twist ) and coherence length (L) broadening: nhklnhkltwistnhkltiltnhklLK||2sincos (3-1)

PAGE 44

32 The geometric dependence of tilt and twist on measured line width is apparent, and the observed broadening due to incoherence is inversely proportional to the magnitude of the reciprocal lattice vector. The exponent, n, denotes the type of convolution of the model. Determination of Convolution One major discrepancy among previous works is the type of convolution, i.e., the value of n is used in the model (Equation 3-1). The choice of n is derived from the bestfit function to the intensity distribution [32]. For example, for a Lorentzian convolution, n=1 and for a Gaussian convolution, n=2. For symmetric scans, both Metzger et al. [30] and Chierchia et al. [33] assumed a Lorentzian convolution, where tilt and symmetric coherence length were determined from a Hall-Williamson plot. The results obtained in this study indicated that the lateral coherence length value was unrealistically high when assuming a Lorentzian convolution, and a more plausible value for coherence length was obtained when a Gaussian convolution was used. For example, a film with a screw dislocation density of 7.710 8 cm -2 measured from TEM gave the following results: assuming a Lorentzian convolution, the lateral coherence length was ~31,000 indicating a dislocation density, ~110 7 cm -2 For the same sample, assuming a Gaussian convolution, the lateral coherence length was 5,000 indicating a dislocation density, ~410 8 cm -2 Similar results were reported by Chierchia et al.[33]. Therefore, a Gaussian convolution (n=2) was used in our analyses with Eq. 1, which is similar to the models of Hordon and Averbach [26] and Ayers [36]. Figure 3-8 shows FWHM data vs. the magnitude of the reciprocal lattice vector (|K| hkl ) for a typical sample assuming a Gaussian convolution. Figure 3-9 shows the dependence of the FWHM on angle of inclination () assuming a Gaussian convolution

PAGE 45

33 for one sample of GaN on sapphire. The Gaussian assumption of the model fits the FWHM vs. data well, and explicitly demonstrates that setting the exponent, n=2, provides accurate results. Calculation of Dislocation Density With the proposed model, two series of scans at different angles were required to deconvolve each individual broadening effect on the experimentally determined FWHM. This deconvolution procedure involves fitting the FWHM data to the model, as shown in Figure 3-8, and is described in detail in the Results and Discussion section. From the experimentally determined values for tilt, twist, and coherence length (both symmetric and asymmetric), two independent methods were used to calculate dislocation density. The first method uses the tilt and twist values with the classic formulation of Dunn and Koch [27], where the density of dislocations, is given by 2 /4.36b 2 Figure 3-9 shows the mathematical and geometrical derivation upon which this method is based. Here, is the calculated rotational broadening effect (tilt or twist) and b is the applicable Burgers vector. With this formula, it is possible to calculate both the screw and edge dislocation density with the deconvolved values for tilt and twist, respectively, using the appropriate Burgers vector. It is important to note that while the coherence length contribution was not used explicitly in calculating dislocation density by this method, its broadening effect was accounted for in Equation 3-1. The second method employed to compute dislocation density was that proposed by Hordon and Averbach [26]. This method assumes that the measured value for coherence length equals the root mean square spacing of randomly spaced threads, where

PAGE 46

34 the dislocation density, =1/L 2 as shown in Figure 3-10. In order to compute the total dislocation density, the value of coherence length was L (l0-ll) the coherence length due to all dislocations (i.e. the coherence length as determined by asymmetric scans in skew geometry, since these are sensitive to both tilt and twist). In order to calculate screw dislocation density, the symmetric coherence length, L (000l) was input for the value of L. The edge dislocation density could be resolved by taking the difference between the total threading dislocation density and screw dislocation density. Simplified Model It is possible to simplify Eq. 1 by neglecting lateral coherence length broadening in order to determine dislocation density more rapidly: 222sincoshkltwisthkltilthkl (3-2) This modification makes it possible to determine the relative tilt and twist of the crystal using only one symmetric scan (e.g., (0004)) and one asymmetric scan (e.g., (10-11)) in skew geometry. Eq. 2 provides an upper bound for rotational broadening effects and a concomitant overestimation of dislocation density. However, Eq. 2 is useful for providing more rapid results to ascertain relative crystalline quality of III-nitride thin films. For example, when characterizing an 8000 thick GaN film with a dislocation density ~10 9 cm -2 this two-scan procedure can provide results within an hour, as opposed to >3 hours for the full five-scan procedure required when incoherence broadening is included in the analysis. This is a considerable improvement when rapid results are desired for growth optimization studies. Comparing dislocation density results determined from the full five-scan procedure to results obtained using this condensed model, the abridged procedure

PAGE 47

35 provided results to within 15%. In all samples studied, the spacing of dislocations was assumed to be random rather than piled-up. A piled-up distribution of dislocations may affect the incoherence broadening term more significantly and must be accounted for, as the value for coherence length may decrease significantly for such a distribution. Furthermore, when the dislocations are piled up, the calculation of dislocation density must be performed using a different method, the details of which are described in Reference 8. Results and Discussion For each sample, the individual broadening effects were deconvolved using Equation 3-1 with the measured FWHM data. Figure 3-8 demonstrates a fit of the model to the FWHM data. From the fit of the model to the data, the magnitude of each broadening contribution was determined. The three symmetric scans were employed to determine tilt and symmetric coherence length, as twist does not affect these data. Once this value for tilt was established, twist and asymmetric coherence length were next determined by fitting the model to the FWHM values of the skew-symmetric reflections. It is important to note that coherence length for a symmetric family of planes may be significantly different than that of an asymmetric family of planes. Two independent methods were used to calculate dislocation density using these values for tilt, twist, and coherence length (both symmetric and asymmetric). These included the formulation of Dunn and Koch [27] ( 2 /4.36b 2 ) and that proposed by Hordon and Averbach [26] (=1/L 2 ). Figure 3-12 shows the agreement between the two methods within a factor of 3. However, the formulation of Dunn and Koch was more reliable when compared to TEM results. This was expected since the coherence length

PAGE 48

36 broadening effect was slight compared to those of crystallographic rotations, and therefore, had a greater intrinsic error. Figure 3-13 shows dislocation densities determined by HRXRD, using the formulation of Dunn and Koch compared to those determined by TEM. The values for dislocation density ranged from 5.410 8 cm -2 to 7.910 9 cm -2 All of the values for dislocation density determined by HRXRD were obtained using Eq. 1 and the full fivescan procedure, for which incoherence broadening was accounted. The values for dislocation density determined by HRXRD agreed with those determined by TEM within a factor of 1.5 for all seven samples. With confidence that the XRD-based method described in this chapter is accurate with respect to TEM, it served as a basis for all subsequent growth optimization studies described throughout this dissertation with respect to dislocation density. The first such optimization study observed the effect of growth pressure on dislocation density. Figure 3-14 demonstrates these results, and exhibits that increasing the growth pressure causes a decrease in dislocation density. Performing this study on so many samples using TEM would be impractical because of the time and cost associated with TEM sample preparation. Such studies can only be performed by a non-destructive, rapid technique such as XRD. Ultimately, III-nitride researchers seek to conclusively determine the effect of dislocations on device performance. Using the XRD-based method described above to determine dislocation density, the effect of dislocations on HEMT performance was ascertained with respect to electronic properties. For the purposes of this experiment, the HEMT performance assumed to be proportional to the sheet densitymobility product (n s

PAGE 49

37 ), widely used as the figure of merit for GaN HEMT device performance [38]. In a series of experiments, electronic properties obtained by Hall measurements were compared to dislocation density. These results are shown in Figure 3-15. While dislocations appear to inhibit room temperature mobility for a given range of sheet density, no trend could be concluded. The reason is that the three samples of highest dislocation density were grown early in the HEMT growth optimization process, and other room temperature scattering mechanisms are believed to have had a stronger effect on these samples compared to those that exhibit higher mobility. These scattering mechanisms include interface roughness and alloy disorder. As the HEMT growths were optimized, higher quality interfaces with fewer variations in alloy disorder from the AlGaN barrier layer were most likely the cause for the improved mobility. Further study is necessary to conclusively determine the precise effects of dislocations on room temperature mobility. Summary This chapter describes a method to determine dislocation density from measured HRXRD line widths. A geometrically derived reciprocal space-based model was developed that allows for the determination of individual microstructural broadening effects in III-nitride thin films. This model unifies previous works that employ x-ray diffraction to calculate dislocation density, but demonstrates distinct advantages in terms of simplicity, accuracy, and time efficiency. Realizing the microstructural tilt, twist, and coherence length, dislocation density was calculated by two separate, independent methods, which agreed within a factor of 3. The model was simplified to neglect broadening due to incoherence, which provided an upper bound for dislocation density

PAGE 50

38 from only two scans. Finally, the validity of this model was tested by comparing dislocation density results to those obtained by TEM, and agreement within a factor of 1.5 was found. These results show that the HRXRD-based method to determine dislocation density presented here is accurate, as demonstrated by the agreement of our XRD-based values for dislocation density with TEM, and time-efficient, as results were obtained within an hour. Thus, this HRXRD-based method to determine dislocation density may provide researchers with a valuable tool for optimizing growth conditions for heteroepitaxial III-nitride films. The method demonstrated in this chapter serves as a basis for the following chapters of this dissertation, since rapid and accurate feedback are required to perform intensive surveys on the effects of dislocations on electronic properties and device performance of GaN thin films.

PAGE 51

39 K (l0l) (0002) (0004) (0006) (10-11) (20-22) K II Figure 3-1. Schematic of reciprocal lattice points probed. Procedurally, the reciprocal lattice points probed were symmetric (000l) and asymmetric (l0-ll). Together, these data could be used to determine the tilt, twist, and coherence length in the crystal.

PAGE 52

40 Figure 3-2. Adjacent domains tilted with respect to one other. The axis of rotation is orthogonal to the c-axis.

PAGE 53

41 Figure 3-3. Adjacent domains twisted with respect to one another. The axis of rotation is orthogonal to the plane of the domain, parallel to the c-axis.

PAGE 54

42 g 11-2-2 0.5 m Figure 3-4. Cross-section transmission electron microscopy image of a low dislocation density sample. The dislocation density for this sample is 2.110 9 cm -2

PAGE 55

43 g 11-2-2 0.5 m Figure 3-5. Cross-section transmission electron microscopy image of a high dislocation density sample. The calculated dislocation density for this sample is 6.810 9 cm -2

PAGE 56

44 A B Figure 3-6. Weak beam transmission electron microscopy image and convergent electron beam diffraction image. A) A sample weak beam TEM image of a film with a dislocation density of 2.5910 9 cm -2 B) Convergent beam electron diffraction image showing the oscillations in diffracted intensity. Disks are the (000) and (11-20) reflections.

PAGE 57

45 Figure 3-7. Broadening effects demonstrated in reciprocal space. In order to accurately determine twist, asymmetric rocking curves in skew geometry are necessary. In (a), the film is shown in cross section. The crystallographic rotations affect measured broadening proportional to their projections of their respective axes onto the reciprocal lattice point being probed. Broadening due to incoherence is proportional to 2/L. In (b), the measured broadening effects are shown in plan view. These include tilt, twist, and incoherence.

PAGE 58

46 0.1700.2100.2500.2902468|K| (1/)FWHM (degrees) Symmetric data Skew-symmetric data Figure 3-8. Plotted data of full width at half maximum vs. reciprocal lattice vector magnitude. As |K| hkl increases, incoherence broadening diminishes. The curvature of the plot as |K| hkl approaches 0 indicates the coherence length contribution on the measured FWHM.

PAGE 59

47 0.050.070.090.110.130.150102030405060708090Chi (degrees)FWHM (degrees) Figure 3-9. Fit of full width at half maximum data vs. angle of inclination. The data shown are for one sample of GaN on sapphire at every observable reciprocal lattice point. The fit of the data assumes a Gaussian convolution without accounting for broadening due to incoherence. The data fit well, and demonstrate that a Gaussian convolution is acceptable. Most of the data lie above the fit due to the neglecting of incoherence broadening.

PAGE 60

48 L ||2ln2b Figure 3-10. Mathematical and geometric derivation of the random distribution method for calculating dislocation density. The coherence length, L, and the rotational broadening effect, are indicated explicitly. The opposite side of this right triangle was derived to be a constant multiplied by the appropriate Burgers vector.

PAGE 61

49 L L L L Figure 3-11. Calculation of dislocation density based on values for coherence length. Since coherence length is defined as the average distance between dislocations, dislocation density is simply calculated as =1/L 2

PAGE 62

50 8.79.19.59.98.79.19.59.9Log Dislocation Density (Dunn/Koch)Log Dislocation Density (Hordon/Averbach) Figure 3-12. Comparison of methods of determining dislocation density by XRD. The agreement shown here demonstrates internal consistency from the XRD-based methods to determine dislocation density.

PAGE 63

51 0.0E+003.0E+096.0E+099.0E+090.0E+003.0E+096.0E+099.0E+09XRD Dislocation Density (cm-2)TEM Dislocation Density (cm-2) Figure 3-13. Dislocation density determined by transmission electron microscopy vs. dislocation density determined by x-ray diffraction. All values determined by XRD agreed with TEM within a factor of 1.5.

PAGE 64

52 1.0E+091.5E+092.0E+092.5E+093.0E+0950100150200Pressure (Torr)Dislocation Density (cm-2) GaN 150torr GaN 100torr Figure 3-14. Dependence of dislocation density on growth pressure. Higher growth pressure causes a lower dislocation density. Dislocation density results were obtained by x-ray diffraction.

PAGE 65

53 04008001200160020000.00E+002.00E+094.00E+096.00E+09Dislocation Density (cm-2)Mobility (cm2/V-s) ns = 5.2 E+12 ns = 9 E+12 ns = 2.2 E+13 ns = 1.3 E+13 Figure 3-15. Effect of dislocation density on room temperature mobility of gallium nitride high electron mobility transistors. No clear effect is evident.

PAGE 66

CHAPTER 4 EFFECT OF DISLOCATIONS ON IMPURITY INCORPORATION IN METAL ORGANIC CHEMICAL VAPOR DEPOSITION GROWTH Introduction Aside from high dislocation density, MOCVD grown GaN and III-nitride films contain extremely high concentrations of impurities. Due to the harsh MOCVD conditions required for growth of III-nitride materials, typical GaN films have impurity concentrations 10 16 cm -3 and sometimes as high as 10 20 cm -3 [39]. These impurities have a profound effect on electronic properties of III-nitride thin films and make it difficult to optimize MOCVD growth conditions. Impurities have also been implicated in providing mid-gap states and inhibiting device performance. The most prominent of these impurities are silicon, oxygen, and carbon. Silicon Silicon is a common impurity in GaN and III-nitride thin films. It is used as an ntype dopant, as it acts as a shallow donor in a III-nitride lattice when it is a substitutional impurity on a gallium lattice site (Si Ga ). The energy differential of the silicon donor state to the conduction band minimum (CBM) has been widely reported, although there is some disagreement. Researchers have reported values of 22 meV [40], 30.18 meV [41], 30.8 meV [42], 31.7 meV [43], and 42 meV [44]. Despite the variation in these reported results, all of these values are indicative of a shallow donor in GaN and AlGaN films, and is the dopant of choice when n-type conductivity is desired. The 54

PAGE 67

55 source of silicon in unintentionally doped (UID) III-nitride thin films is widely believed to be from decomposition of silicon carbide (SiC) substrates [45] and susceptors. Oxygen Oxygen is another impurity that has been implicated as an n-type dopant when it incorporates substitutionally on a nitrogen lattice site (O N ). Many researchers have implicated oxygen as being responsible for the significant background n-type conductivity of UID GaN films. Several research groups have reported that O N donor level states are between 32-34 meV from the CBM [40-42] while other groups have reported lower values. Joshkin et al. reported a value of 23.5 meV [46], while others have reported values between 2-10 meV [44,47]. Despite the variance in these values, oxygen is widely considered a shallow donor. Many researchers believe that oxygen impurities act as n-type dopants, simply because of its extra valence electron relative to nitrogen, while others have proposed a more complicated mechanism for oxygen providing the background n-type conductivity of UID GaN. For example, Oila et al. have proposed that oxygen impurities promote the existence of gallium vacancies (V Ga ), as oxygen-gallium vacancy complexes (O N -V Ga ) are energetically stable in the GaN lattice. According to Oila et al., it is the gallium vacancies that provide the background n-type conductivity [48]. Overall, however, many researchers attribute the background conductivity of GaN to the presence of oxygen in the lattice, despite the disagreements in the proposed mechanisms. The sources of oxygen include residual oxygen gas in the ammonia source [47] and atmospheric contamination of the growth chamber. Aside from these sources of contamination, the sapphire (Al 2 O 3 ) substrate [45] provides a significant source of

PAGE 68

56 oxygen. When sapphire is subjected to the high temperatures employed during MOCVD growth of in a hydrogen ambient, the sapphire surface is etched, and the oxygen atoms incorporate into the GaN film. Performing electrical measurements, researchers have found that the GaN film is significantly n-type within ~0.4 m of the sapphire interface, due to a buried conductive layer (BCL) [49]. This BCL is undesirable when attempting to grow semi-insulating GaN films, and researchers have attempted to compensate these residual carriers by intentionally incorporating deep acceptors, like carbon and iron [49]. Carbon Carbon is another common impurity in MOCVD grown GaN films. Carbon is an amphoteric dopant, as its electronic properties depend on whether it incorporates substitutionally on a gallium site or on the nitrogen site [54,55]. If carbon incorporates on a gallium site, it acts as a shallow donor, with an ionization energy of ~0.2 eV. Conversely, if carbon incorporates on a nitrogen site, it acts as a shallow acceptor with values of ionization energy reported as ~0.3 eV [55] and 34 meV [42]. If carbon incorporates interstitially, its energy level is mid-gap, and its donor/acceptor properties depend on the Fermi level. If carbon incorporates interstitially into n-type GaN, it will act as a deep acceptor, but if carbon incorporates into p-type GaN, its energy state will be indicative of a deep donor [55]. Carbon is generally considered a deep acceptor, and when GaN films must be insulating, MOCVD researchers often intentionally adjust growth conditions to favor carbon incorporation in order to compensate the background n-type conductivity of UID GaN. However, growers try to minimize carbon

PAGE 69

57 incorporation near active regions, as it acts as a scattering center, and has been implicated in gate leakage and dispersion in GaN-based field effect transistors (FETs). The sources of carbon in MOCVD growth of III-nitrides are the metalorganic group III precursors, and, to a lesser extent, SiC substrates and susceptors. Carbon is a common impurity in any MOCVD growth process, and its incorporation can dramatically affect the films properties, but growth conditions can be adjusted to control the extent to which carbon incorporates. Previous Work Growth experimenters have demonstrated that MOCVD growth conditions have a significant effect on impurity incorporation in GaN films. Specifically, the effects of temperature, pressure, and flows of TMG, ammonia, and hydrogen on silicon and carbon incorporation were studied in-depth [39]. A correlation between resistivity, dislocation density, and carbon concentration in GaN films has also been established, indicating that threading edge dislocations and/or carbon act as compensating centers [56]. Furthermore, threading edge dislocations have been shown to provide energetically favorable sites for impurity incorporation, due to the miscoordinated atoms along extended line defects [57,58] in other material systems. Studies on growth of III-nitrides have only investigated the effects of growth conditions on impurity incorporation on templates with nominally identical microstructures. What is lacking to date is the effect of dislocations on impurity incorporation in MOCVD-grown III-nitride films [39,56]. Both Koleske et al. and Wickenden et al. have indicated that threading edge dislocations facilitate carbon incorporation, but conclusive studies to confirm this have not been performed.

PAGE 70

58 Focus The experimentation in this chapter aims to determine the effects of dislocation density on impurity incorporation during MOCVD growth. Specifically, silicon, oxygen, and carbon incorporation are observed with respect to growth conditions on templates of variable dislocation densities. By subjecting templates with vastly different dislocation densities to identical growth conditions, the effect of dislocation density on impurity incorporation could be determined. Experimental Methods Templates were grown by MOCVD on sapphire substrates, and dislocation density was intentionally altered to obtain two series of GaN films with the largest range of dislocation density possible. Specifically, the growth conditions that were altered were growth pressure, nucleation layer, and recovery time. Template dislocation densities ranged from 4.5910 8 cm -2 8.7610 10 cm -2 for the first growth series, as shown in Table 4-1, and 4.4610 8 cm -2 7.610 10 cm -2 for the second growth series, as shown in Table 4-2. As expected, conditions that minimized dislocation density were higher growth pressures, GaN nucleation layers, and delayed recoveries. Higher growth pressures cause larger grain sizes during the initial stages of high temperature growth, which cause fewer dislocations upon coalescence [59]. GaN nucleation layers provide lattice-matched templates upon which to grow high temperature GaN. Delayed recovery is believed to cause dislocations to propagate laterally rather than thread along the c-axis to the surface. The characterization performed in this experimentation included XRD to determine dislocation density, using the method described in Chapter 3, and SIMS to determine impurity incorporation. Two series of growth runs were performed on

PAGE 71

59 templates of dislocation density varying over two orders of magnitude, while the only varied growth parameter was pressure in the overgrowth. For both overgrowth runs, variable dislocation density templates were grown at 1050C, while pressure was varied from 20 Torr, and silicon was incorporated as a marker when changing pressure. Impurity incorporation was measured for all three samples from both runs. Results and Discussion The first series of SIMS data is shown in Figures 4-1 to 4-4. In all samples, oxygen and silicon were at the detection limit of the instrument. The carbon profiles changed in all cases, presumably due to changes in growth pressure, but this could not be confirmed due to the fact that the silane did not incorporate. Another observation was that the radial position of the wafer had a distinct effect on impurity incorporation. The profile shown in Figure 4-3 was taken from the same wafer as that of Figure 4-4, but the SIMS results from Figure 4-3 were obtained near the top of the wafer, while those from Figure 4-4 were obtained near the flat. This demonstrates that impurity profiles are more pronounced near the flat part of the wafer. From the results of the first series of runs, two adjustments were made: 1) all SIMS analysis was performed near the flat part of the wafer to obtain more distinct profiles, and 2) the silane source was confirmed to be online by Hall measurements of previous samples to ensure intentionally doped n-type conductivity. The SIMS results for this second series of runs are shown in Figures 4-5 to 4-7. Once again, oxygen impurities were below detection limit, which demonstrates that oxygen does not incorporate significantly into GaN films grown in this reactor. Silicon was also at the detection limit, except when it was intentionally incorporated to indicate changes in

PAGE 72

60 pressure. Carbon profiles, however, changed significantly with changes in pressure for all samples. As expected, carbon incorporated inversely with pressure; the maximum carbon incorporation occurred when the growth pressure was 20 Torr, and the minimum incorporation occurred at 500 Torr for all samples. In order to determine the effect of dislocation density on carbon incorporation, average values of carbon concentration were determined for each sample at each pressure and compared. These results are shown in Figure 4-8, and show that there is no clear trend confirming that dislocation density promotes carbon incorporation. For all growth pressures, the template with a dislocation density of 5.5910 9 cm -2 had the highest carbon concentration. Although dislocation density did not have an observable effect on impurity incorporation, pressure had a significant effect. Figure 4-8 indicates that carbon incorporation is, to some extent, a pressure-dependent process. At higher pressures, there are greater molecular interactions between carbon and hydride gases, which promote carbon removal from the film during MOCVD growth. Threading dislocations, however, do not appear to provide significantly energetically favorable sites for impurity incorporation. In order to confirm the finding that dislocations do not promote incorporation of impurities (specifically carbon) during MOCVD growth, two series of experiments are proposed. First, a similar series of experiments could be performed varying parameter of growth temperature. It is possible that the carbon removal mechanism at high pressures is significantly different than that of higher temperature, so dislocations may have an observable effect on carbon incorporation by altering temperature. Second, a similar series of experiments could be performed with broader range of dislocation density

PAGE 73

61 templates. This could be achieved by growing templates employing unconventional methods to minimize dislocation density, including epitaxial lateral overgrowth (ELO) or cantilever epitaxy (CE). These techniques essentially filter dislocations and have been used to grow GaN films with dislocation densities below 10 7 cm -2 This would provide a broader range of dislocation densities for the templates for the purpose of this experimentation. Summary The effect of threading dislocation density on impurity incorporation was investigated in this chapter. Oxygen and silicon were at the SIMS detection limit in all cases (except when silane was intentionally flowed into the growth chamber to indicate pressure changes), indicating that an insignificant amount of oxygen and silicon incorporated into the GaN films. Carbon incorporation depended heavily on pressure, and higher growth pressures caused significantly less carbon to incorporate into the films. Dislocation density, however, had no observable effect on carbon incorporation. This result indicates that threading dislocations do not provide significantly energetically favorable sites for impurity incorporation in MOCVD grown GaN films. The carbon incorporation in this experiment was pressure-driven by a mechanism dependent on molecular interactions.

PAGE 74

62 1E+141E+151E+161E+171E+181E+191E+201E+210246810DEPTH (microns)CONCENTRATION (atoms/cc)1E+001E+011E+021E+031E+041E+051E+061E+071E+08Counts Per Second Ga-> CO Si Figure 4-1. Impurity profiles for low dislocation density at the top. The impurities under study were silicon, oxygen, and carbon for the first series of overgrowth studies. The dislocation density of the template is 4.5910 8 cm -2 The SIMS sample was taken from the top part of the wafer.

PAGE 75

63 1E+141E+151E+161E+171E+181E+191E+201E+210246810DEPTH (microns)CONCENTRATION (atoms/cc)1E+001E+011E+021E+031E+041E+051E+061E+071E+08Counts Per Second OGa-> C Si Figure 4-2. Impurity profiles for mid dislocation density at the top. The impurities under study were silicon, oxygen, and carbon for the first series of overgrowth studies. The dislocation density of the template is 6.6610 9 cm -2 The SIMS sample was taken from the top part of the wafer.

PAGE 76

64 1E+141E+151E+161E+171E+181E+191E+201E+210246810DEPTH (microns)CONCENTRATION (atoms/cc)1E+001E+011E+021E+031E+041E+051E+061E+071E+08Counts Per Second OGa-> C Si Figure 4-3. Impurity profiles for high dislocation density at the top. The impurities under study were silicon, oxygen, and carbon for the first series of overgrowth studies. The dislocation density of the template is 8.7610 10 cm -2 The SIMS sample was taken from the top part of the wafer.

PAGE 77

65 1E+141E+151E+161E+171E+181E+191E+201E+210246810DEPTH (microns)CONCENTRATION (atoms/cc)1E+001E+011E+021E+031E+041E+051E+061E+071E+08Counts Per Second OGa-> C Si Figure 4-4. Impurity profiles for high dislocation density at the flat. The impurities under study were silicon, oxygen, and carbon for the first series of overgrowth studies. The dislocation density of the template is 8.7610 10 cm -2 The SIMS sample was taken from the flat part of the wafer.

PAGE 78

66 Impurity Depth Profile, = 4.5x108 cm-21.00E+151.00E+161.00E+171.00E+181.00E+1902468Depth (m)Concentration (cm-3) Carbon Silicon Oxygen Figure 4-5. Impurity profiles for second series of overgrowths for low dislocation density at the flat. The impurities under study were silicon, oxygen, and carbon. The dislocation density of the template is 4.4610 8 cm -2

PAGE 79

67 Impurity Depth Profile, = 5.6x109 cm-2 1.00E+151.00E+161.00E+171.00E+181.00E+19012345Depth (m)Concentration (cm-3) Carbon Silicon Oxygen Figure 4-6. Impurity profiles for second series of overgrowths for mid dislocation density at the flat. The impurities under study were silicon, oxygen, and carbon. The dislocation density of the template is 5.5910 9 cm -2

PAGE 80

68 Impurity Depth Profile, = 7.6x1010 cm-21.00E+151.00E+161.00E+171.00E+181.00E+19012345Depth (m)Concentration (cm-3) 6 Carbon Silicon Oxygen Figure 4-7. Impurity profiles for second series of overgrowths for high dislocation density at the flat. The impurities under study were silicon, oxygen, and carbon. The dislocation density of the template is 7.610 10 cm -2

PAGE 81

69 Carbon Incorporation vs. Pressure1.00E+161.00E+171.00E+181.00E+190100200300400500600Pressure (Torr)Carbon Concentration (cm-3) High dislocation density template Low dislocation density template Mid dislocation density template Figure 4-8. Comparison of impurity profiles for different dislocation density templates. Impurity profiles for silicon, oxygen, and carbon for variable dislocation density templates during MOCVD growth with variable pressure are shown. The low dislocation density template is 4.4610 8 cm -2 the mid dislocation density template is 5.5910 9 cm -2 and the high dislocation density template is 7.610 10 cm -2 The SIMS samples were all taken from the flat part of the wafer.

PAGE 82

70 Table 4-1. Growth conditions for the templates for the first series of SIMS results. Conditions that promote lower dislocation density are higher pressure, delayed recovery, and GaN nucleation layers. Dislocation densities were intentionally altered to provide templates with as large of a range for dislocation density possible. Sample Dislocation density (cm -2 ) Growth pressure (Torr) Nucleation Layer Recovery Time 1 4.5910 8 500 GaN Delayed 2 6.6610 9 50 GaN Rapid 3 8.7610 10 70 AlN Rapid

PAGE 83

71 Table 4-2. Growth conditions for the templates for the second series of SIMS results. Dislocation densities were intentionally altered to provide templates with as large of a range for dislocation density possible. Sample Dislocation density (cm -2 ) Growth pressure (Torr) Nucleation Layer Recovery Time 1 4.4610 8 500 GaN Delayed 2 5.5910 9 50 GaN Rapid 3 7.610 10 70 AlN Rapid

PAGE 84

CHAPTER 5 GALLIUM NITRIDE BULK CRYSTAL GROWTH BY DISSOLUTION AND RECRYSTALLIZATION OF GALLIUM NITRIDE POWDER Introduction Poor material quality is widely considered to be the most significant factor limiting GaN-based device performance, and this is primarily due to a lack of availability of lattice-matched substrates. Ideally, GaN film growth would be performed on single crystal GaN bulk substrates, but the intrinsic properties of this material makes fabrication of bulk crystals difficult. In other semiconductor material systems, such as silicon and gallium arsenide, bulk crystals have been fabricated from melts, where boules of high crystalline quality material have been achieved with dimensions as large as 12-inches 12 feet. Similar methods are not feasible for GaN because the GaN dissociates prior to melting at atmospheric pressure. Because of this, no commercially viable bulk growth method for GaN has been developed to date, despite the potential benefits of low defect density GaN bulk crystals. Previous Work Researchers have demonstrated several methods to successfully fabricate GaN bulk crystals. The two most prominent of these are an ultra high nitrogen pressure method, and an ammonothermal technique. Both approaches offer a distinct set of advantages and disadvantages relative to the other. The ultra high nitrogen pressure technique entails subjecting a pool of liquid gallium metal to extremely harsh conditions in a nitrogen gas (N 2 ) environment [60], with 72

PAGE 85

73 temperatures of ~1600C and nitrogen overpressures of ~45,000 atm. GaN crystals are formed by dissolving the nitrogen gas into the gallium liquid, where N 2 dissociates into atomic nitrogen, and bonds with the gallium. Using this method, researchers have successfully fabricated crystals of GaN that are 1 cm 2 100 m, with extremely low dislocation densities (~100 cm -2 ). The drawbacks to using this approach include slow growth kinetics, high impurity concentrations, the inability to grow boules of material, and the high pressure and high temperature requirements. The GaN crystals are grown as a crust on the liquid gallium surface, and a 1 cm 2 100 m crystal is formed after ~1 month under the extreme conditions required. Furthermore, these crystals contain ~10 18 cm -3 oxygen impurities, which may have an undesirable effect on the electronic properties of the GaN crystal. Finally, the temperature and pressure requirements add significant cost to this method. The ammonthermal technique has been demonstrated to successfully grow GaN bulk crystals by dissolving GaN feedstock into liquid ammonia, and precipitating single crystal GaN upon supersaturation [61]. This method demonstrates significant improvement over the high nitrogen pressure-driven process described above with respect to growth rate (~0.5 mm/week). Furthermore, large area boules are possible to extract from a seed crystal. The disadvantages of this method include higher dislocation densities (~10 6 cm -2 ) and high impurity incorporation (~10 17-18 cm -3 ). Moreover, this process also requires harsh conditions, with temperatures of ~550C and pressures of ~4000 atm in order to dissolve GaN into a liquid ammonia medium. Although both methods of GaN bulk growth described above were significant breakthroughs, neither are commercially viable processes to fabricate GaN crystals. This

PAGE 86

74 is primarily due to the high-pressure requirement, which prevents both processes from being scalable and manufacturable. If a method to fabricate GaN bulk crystals at atmospheric pressure were possible, this would provide a distinct advantage over other methods currently employed. Proposed Methods In order to develop an atmospheric pressure process to form GaN crystals, three series of experiments have been proposed. The first involves electroplating of GaN using a gallium metal electrode and a reactive nitride ion (N 3) from a lithium nitride (Li 3 N) precursor. A second possible technique comprises an electrochemical reduction of nitrogen gas (N 2 ) to form nitride ions (N 3) that react with gallium to form GaN. A third potential method is a simple dissolution/recrystallization process of GaN powder in a liquid medium, where the formation of GaN single crystal would be driven by a thermal gradient. Electroplating Electroplating of GaN from gallium and nitride precursors could prove to be a promising method to form high quality GaN crystals. The difficulty, however, was determining a suitable host environment in which a nitride ion could exist without reacting quickly and explosively. In an unrelated series of experiments, Goto et al. determined that a molten alkali-halide host environment could successfully dissolve stable nitride ions [62]. Using this, the possibility of forming GaN crystals from gallium and nitride precursors was investigated. Figure 5-1 shows a GaN crystal of macroscopic dimensions (0.9 mm 0.6 mm) that was successfully synthesized using this method.

PAGE 87

75 This result was encouraging, but the electroplating method suffered from four significant drawbacks: Electroplating does not provide material with optimal crystalline quality. The electroplating process would be very difficult to scale, in order to develop a controllable, manufacturable process. The Li 3 N precursor is extremely expensive, making this process less cost-effective. The buildup of Li + ions with GaN deposition complicates the electrochemistry [63]. In order to make this process more cost effective and controllable, the electrochemical reduction of nitrogen gas to nitride ions was investigated (1/2 N 2 + 3 e N 3). Electrochemical Reduction of Nitrogen Gas Goto and Ito investigated the possibility of electrochemically reducing nitrogen gas into nitride ions in a molten alkali-halide liquid medium [63]. Using this result, the possibility of fabricating GaN crystals from electrochemically oxidized gallium with electrochemically reduced nitrogen gas was investigated, and GaN crystals were successfully formed. Using this method, bulk GaN crystals could possibly deposit on a seed crystal, and be extracted as a boule. While this process offers several advantages, the major disadvantage is the difficulty of controlling the fluid dynamics to form the crystals. Because nitride ions are so reactive, they must be separated from gallium ions until they reach the desired growth surface, which would be difficult to model. Dissolution and Recrystallization of Gallium Nitride Crystals The ideal method to grow bulk GaN crystals would be based on a simple dissolution/recrystallization process of GaN in a liquid medium. Using such a method, it would be possible to control the rate of crystal growth by controlling the temperature

PAGE 88

76 gradient and the fluid mechanics in the growth vessel. This process could eventually be used to form large GaN crystals, that could be extracted as a boule from the melt. Focus In this chapter, a novel bulk growth method of GaN crystals is introduced, based on a simple dissolution/recrystallization process in a molten alkali-halide salt. If GaN can dissolve sufficiently in such a medium, then researchers could develop a temperaturedependent, atmospheric pressure process to fabricate GaN bulk crystals. Such a method could prove to be a scalable and manufacturable process that could lead to production of commercially available GaN bulk crystals and wafers for homoepitaxial GaN growth. Specifically, this chapter examines whether a molten halide medium is capable of dissolving GaN, and if so, whether it is possible to recrystallize GaN on a nominally homoepitaxial GaN template. Experimental Methods All of the experimentation in this chapter was performed in a home-built stainless steel heater vessel fitted for a 1 quartz tube in a glove box with a nitrogen gas ambient. To determine the temperature-dependent solubility of GaN, aliquots of GaN in molten lithium chloride (LiCl) were taken at various temperatures with a quartz rod. The solubility was determined by inductive coupled plasma-mass spectrometry (ICP-MS), where the ratio of the gallium to lithium concentration was calculated. Recrystallization studies were performed in quartz tubes in the same stainless steel chamber, where the recrystallization surface was MOCVD-grown GaN on sapphire. Deposited GaN crystallites were observed by scanning electron microscopy (SEM) and elemental

PAGE 89

77 analysis on these crystallites was performed by energy dispersive spectroscopy (EDS) in order to determine the extent to which GaN recrystallized on the wafer surface. Results and Discussion In order to grow GaN crystals at a fast enough rate to be commercially viable, it is necessary to dissolve a relatively high concentration of GaN into solution. If the solubility is found to be sufficiently high, it would be possible to grow GaN bulk crystals at a reasonable rate upon supersaturation. The basis for the experimentation demonstrated in this chapter is the ideal solubility curve for GaN, shown in Figure 5-2. This curve was calculated by the following equation [64]: 1lnTTRSxmFnn (5-1) Under diffusion-limited conditions, a solubility of ~0.1% for GaN could provide a crystal growth rate of ~0.5 mm/h. This corresponds to a temperature of ~1230C on the ideal solubility curve, and since this process occurs at atmospheric pressure, these conditions are mild enough for a commercially viable method of GaN crystal growth. If the selected solvent has favorable interactions with GaN, then experimentally, the data should lie above this curve, corresponding to higher solubilities, and consequently, potentially higher growth rates. The first series of experiments determined the temperature dependent solubility of GaN powder in LiCl. Dissolution of Gallium Nitride in Lithium Chloride The temperature dependent solubility of GaN in LiCl was experimentally determined by performing ICP-MS analysis of aliquots of solution at various temperatures. These temperatures ranged from 750C 1000C, as shown in Figure 5-3. The experimental data indicated that, in most cases, more GaN dissolved than was

PAGE 90

78 predicted by the ideal solubility curve. The scatter in the data was most likely due to the fact that the ICP-MS is an inorganic, water-based technique, and since GaN is insoluble in water, some of the crystallites may have precipitated out, and segregated. In order to address this, a 2 m filter was used to filter samples for analysis prior to injection into the ICP-MS. These results, shown in Figure 5-4, represent a lower bound for solubility of GaN in LiCl, but they were still significantly above the ideal solubility curve. These data indicate that 10-50 ppm of GaN were soluble in LiCl at temperatures ranging from 800C 950C, corresponding to a growth rate of ~50-100 m/h. These values were sufficient to commence experimentation of recrystallization of GaN on a GaN template. Gallium Nitride Recrystallization In order to determine whether GaN could recrystallize out of solution, a series of experiments was performed, where GaN powder in LiCl was heated to ~950C, and then, became supersaturated upon cool-down. This supersaturated GaN would presumably deposit on a lattice-matched template. SEM images of a surface not exposed to the melt (Figure 5-5) were obtained and compared to one that was (Figure 5-6). Note the distinct differences in surface morphology. Upon further investigation, individual features were observed at higher magnifications, as shown in Figure 5-7. This feature appears to exhibit a hexagonal shape, indicative of the wurtzite crystal structure. To further promote GaN recrystallization, another experiment was run with stirring at the maximum temperature (950C), prior to cool-down. The result of this experiment was that more surface features appeared. From this finding, a third experiment was performed, with 1) half of the wafer surface treated with nitric acid (HNO 3 ) at 50C, and 2) constant agitation during cool-down. The surface treatment was

PAGE 91

79 performed in order to etch the native oxide off the growth surface, and the agitation was performed in order to provide forced convection, and greater recrystallization. This provided two interesting results. First, more features were formed on the wafer surface. Second, the features that formed on the treated surface tended to be different from those on the untreated surface. On the treated surface, there was a high density of patchy features, shown in Figure 5-8, whereas, the untreated surface had a high density of needle-like features, shown in Figures 5-9 to 5-11. In order to determine the chemical composition of the two observed features, elemental analysis was performed by EDS. The EDS results for a needle structure is shown in Figure 5-12, and strongly indicate that this structure is comprised of etched quartz, as evidenced by the heavy silicon and oxygen concentrations as well as the gallium void indicated. Elemental analysis of a patch structure is shown in Figure 5-13, and is indicative of GaN; the gallium and nitrogen concentrations are high, and silicon and oxygen are practically nonexistent. Another interesting finding is that silicon oxide (SiO x ) preferentially deposited on the untreated surface, which presumably contained the native oxide, whereas GaN preferentially deposited on the treated surface. A cross-sectional view of the patch demonstrated in the SEM image in Figure 5-13 is shown in Figure 5-14. From the scale shown, this GaN crystallite is ~5 m in thickness, which indicates that the growth rate in this experiment was ~2.5 m/h; this is significantly higher than any other bulk growth technique. Furthermore, the depth of field of the EDS instrument is shown in Figure 5-15 for the highest power used in this study. Since the feature probed is ~5 m, and the depth of the instrument is ~1600 nm, this further supports that the crystallite is that of pure GaN.

PAGE 92

80 While the experimentation described in this chapter is relatively unrefined, and far from optimized, it demonstrates that a simple dissolution/recrystallization process is a potentially viability one for the fabrication of GaN bulk crystals from a molten alkalihalide medium. The results were encouraging, and could provide a first step in the development of a novel method for growing GaN crystals. Significant improvements could be made to the process, including solvent optimization, greater purity of reagents, greater control over the fluid dynamics, and using a true seed crystal on which to crystallize GaN from the molten solvent. Eventually, using this method, researchers could potentially develop a commercially feasible, rapid growth rate process to fabricate high crystalline quality GaN boules for homoepitaxial GaN growth. Summary An ideal solubility curve was calculated for GaN with respect to temperature, based on intrinsic thermodynamic parameters. The solubility of GaN in LiCl was experimentally determined, and demonstrated that GaN was more soluble than the ideal curve indicated. From these results, a GaN template was placed in a LiCl medium containing GaN powder, which was heated to ~1000C, and then cooled. Deposition was observed on the GaN template surface by SEM. Further experimentation demonstrated that GaN crystallites deposited at a higher rate after pretreatment of the GaN template surface in an acid solution, and with constant fluid agitation during cool-down. Elemental analysis determined that patchy structures as thick as 5 m were, in fact, GaN. These crystallites formed after merely 2 hours, indicating an extraordinarily high growth rate, relative to other GaN crystal growth techniques.

PAGE 93

81 Potential improvements to this experimentation include More sophisticated equipment to allow greater control over thermal gradients and fluid dynamics. Use of higher purity reagents. More comprehensive studies optimizing GaN solubility. Use of a true seed crystal upon which to deposit dissolved GaN.

PAGE 94

82 Figure 5-1. Gallium nitride crystal with dimensions of 0.9 mm 0.6 mm. The bulk growth technique to form this crystal was electrodeposition.

PAGE 95

83 Ideal Solubility of GaN00.050.10.150.20.2510001100120013001400Temperature (oC)Solubility (%) Figure 5-2. Ideal solubility of gallium nitride. The data for this curve were determined by Equation 5-1.

PAGE 96

84 Solubility of GaN in LiCl02040608010012075080085090095010001050Temperature (oC)GaN Solubility (ppm) measured ideal Figure 5-3. Experimentally determined solubility of gallium nitride in lithium chloride compared to the ideal solubility curv e. These data points were obtained without the use of a filter.

PAGE 97

85 Solubility of GaN in LiCl05101520253035406507508509501050Temperature (oC)Solubility (ppm) Ideal Solubility Experimental Data Figure 5-4. Experimentally determined solubility determined after employing a 2 m filter. These data represent a lower bound for solubility of gallium nitride in lithium chloride.

PAGE 98

86 Figure 5-5. Scanning electron microscopy image of a bare gallium nitride surface. The morphology is smooth.

PAGE 99

87 Figure 5-6. Scanning electron microscopy image of a gallium nitride surface exposed to gallium nitride-in-lithium chloride melt. The morphology is noticeably rougher than the bare, untreated surface.

PAGE 100

88 Figure 5-7. Close-up scanning electron microscopy image of a gallium nitride crystallite. Note the hexagonal structure.

PAGE 101

89 Figure 5-8. Scanning electron microscopy image of a treated gallium nitride surface after exposure to the gallium nitr ide-in-lithium chloride melt. Note the prevalence of patchy structures.

PAGE 102

90 Figure 5-9. Scanning electron microscopy image of an untreated gallium nitride surface after exposure to a gallium nitride-in-lithium chloride melt. Note the prevalence of needle-like structures.

PAGE 103

91 Figure 5-10. Scanning electron microscopy image focusing on needles on the untreated gallium nitride surface at 750. Note th e prevalence of needle structures on this part of the wafer.

PAGE 104

92 Figure 5-11. Scanning electron microscopy image focusing on needles of the untreated gallium nitride surface at 1900. The stru ctures appear to have a hexagonal structure and exhibits columnar growth.

PAGE 105

93 Needle Figure 5-12. Election dispersive spectroscopy images for elemental analysis on needle structures. Note the high con centrations of silicon and oxygen.

PAGE 106

94 Figure 5-13. Electron dispersive spectroscopy images of a patch on the gallium nitride surface. Note the high concentrations of gallium and nitrogen, and low concentrations of silicon and oxygen.

PAGE 107

95 Figure 5-14. Scanning electron microscopy image of a deposited gallium nitride crystallite with a thickness of 5 m. This crystallite was formed after approximately 2 hours.

PAGE 108

96 Figure 5-15. Electron dispersive spectroscopy instrument depth calibration. This represents a Monte Carlo simulation for the highest voltage used (15keV).

PAGE 109

CHAPTER 6 SUMMARY AND FUTURE DIRECTIONS In this dissertation, specific effects of dislocations in III-nitride materials are investigated. Samples of a wide range of dislocation density were grown by MOCVD to determine their specific effects. In order to enable run-to-run feedback for growth optimization studies, a method to determine dislocation density based on HRXRD results was developed. This method allows researchers to obtain results for dislocation density within one hour, and was shown to be accurate by its agreement with TEM results within a factor of 1.5 for all seven samples analyzed. This XRD-based method provided the foundation for subsequent research investigating the effects of dislocations on electronic properties and impurity incorporation of GaN films during MOCVD growth. Further study of this technique should be performed to investigate its accuracy with respect to dislocation type. Screw dislocations are considered to have a different effect on electronic properties than edge dislocations, and determining the density of dislocations with a screw component could be essential for further growth optimization studies. The method shown in Chapter 3 is accurate for dislocation density with an edge component, but has not been shown to be accurate with respect to dislocations with a screw component. In the MOCVD films grown, the edge-type dislocation density was greater than the screw-type dislocation density by approximately an order of magnitude in all cases. Inaccuracies for screw dislocation density would be suppressed, since edge dislocation density is significantly higher. For more conclusive results, a similar series of experiments should be performed studying the accuracy of the technique compared to 97

PAGE 110

98 TEM results for dislocations with a screw component. Furthermore, the technique should be investigated for samples of relatively low dislocation densities ( 10 7 cm -2 ). As growth optimization and III-nitride bulk crystal development continue to progress, samples of lower dislocation density will be grown, and this techniques accuracy should be confirmed for higher quality III-nitride crystals. Furthermore, in samples grown by cantilever epitaxy (CE) and epitaxial lateral overgrowth (ELO), the distribution of dislocations may be piled-up rather than random, so the method to calculated dislocation density proposed in Chapter 3 may need to be modified to account for this. Finally, the effect of film thickness on XRD line width must be ascertained. Currently, the model does not account for broadening due to strain. For samples of low thickness (<4000 ), strain broadening has a noticeable effect on measured XRD line width, providing anomalously high results for dislocation density when using the model in Chapter 3. Determining the thickness where strain broadening must be accounted for would be useful, and including strain broadening into the model for thin samples could provide benefits for growth optimization studies. Therefore, further studies are required to include broadening due to strain for thinner films. For the growth studies investigating impurity incorporation into GaN films, more experimentation is necessary to conclusively determine whether threading edge dislocations provide energetically favorable sites for carbon incorporation during MOCVD growth. One possible series of experiments involves the measurement of carbon incorporation on variable dislocation density templates as temperature is varied. It has been shown that subjecting templates to lower growth temperatures causes more carbon to incorporate. Repeating the experimentation described in Chapter 4 with respect

PAGE 111

99 to temperature could provide more insight on the effect of threading edge dislocations on impurity incorporation. Impurities may incorporate by a different mechanism with respect to growth temperature, as opposed to growth pressure. The proposed mechanism with respect to pressure is one that involves molecular interactions between carbon on the growth surface and the hydride gases in the ambient. The carbon removal mechanism was proposed to be one that involves carbon reacting with hydrides to form methyl radicals, which are subsequently removed from the growth surface and pumped out through the exhaust. Varying temperature while pressure is held constant could provide insight on whether threading edge dislocations provide energetically favorable sites for carbon incorporation. Furthermore, samples over a broader range of dislocation density would be desired for this study. It is possible that the effect of threading edge dislocations (if any) is different for lower dislocation density samples. Growing on CE samples, where dislocation density, ~10 6 cm -2 could provide further insight on potential mechanisms for impurity incorporation. Where dislocations demonstrate a deleterious effect on IIInitride films and IIInitride based devices, III-nitride bulk crystals are desired. In order to further investigate the possibility of depositing GaN bulk crystals, the research from Chapter 5 should be continued to optimize the dissolution/recrystallization process demonstrated. This includes performing solubility studies with a wide variety of alkali-halide salts in order to determine which best dissolves GaN. Once this is determined, deposition optimization studies should be performed. This includes gaining better control over the fluid dynamics and thermal gradients, in order to maximize the rate of GaN crystal formation. More sophisticated equipment will be required for this, such as a multistage

PAGE 112

100 heater and higher purity reagents. Moreover, similar experimentation could be performed to ascertain the possibility of fabricating AlN, InN, bulk crystals as well as alloys of bulk III-nitride crystals. If no solvent is deemed to be suitable for a commercially viable process to extract bulk GaN (or bulk AlN, InN, or III-nitride alloy) crystals, the solubility results could still prove to be beneficial, even if this dissolution/ recrystallization method is not ultimately used. The method based on electrochemical generation of precursors could be employed, where knowing the solubility would be beneficial since it would indicate the extent to which liquid phase pre-reactions of Ga 3+ and N 3ions must be minimized. Using such a method, the growth rate of a GaN bulk crystal would be controlled by fluid dynamics, but developing a process where pre-reactions are minimized could be difficult, due to the highly reactive nature of N 3ions. Finally, if this electrochemical generation of precursors method provides growth rates that are too low to be a commercially viable process, then electro-deposition of GaN films could be further optimized. While this technique for fabricating high quality bulk crystals demonstrates several difficulties, a process could be developed where single crystal GaN wafers could be formed and extracted from solution by employing fluid dynamics to ensure uniform N 3coverage over a liquid gallium electrode. The reactive N 3ions would react with Ga 3+ ions at the gallium liquid interface, forming a single crystal crust of GaN that could be extracted. Using such a method would remove the need to cleave a boule of bulk GaN, as such a process would produce individual GaN wafers. Optimization of the electro-deposition process could then be scaled to produce wafers of desired dimensions.

PAGE 113

LIST OF REFERENCES 1. W. Walukiewicz, S.X. Li, J. Wu, K.M. Yu, J.W.Ager III, E.E. Haller, H. Lu, W.J. Schaff, J. Cryst. Growth, 269, p. 119, (2004). 2. K.B. Nam, J. Li, K.H. Kim, J.Y. Lin, H.X. Jiang, Appl. Phys. Lttr., 78, (23), p. 3690, (2001). 3. L.F. Eastman, U.K. Mishra, IEEE Spectrum, p. 28, (2001). 4. O. Ambacher, B. Foutz, J. Smart, J.R. Shealy, N.G. Weimann, K. Chu, M. Murphy, A.J. Sierakowski, W.J. Schaff, L.F. Eastman, R. Dimitrov, A. Mitchell, M. Stutzmann, J. Appl. Phys., 87, (1), p. 334, (2000). 5. L.F. Eastman, V. Tilak, V. Kaper, J. Smart, R. Thomson, B. Green, J.R. Shealy, T. Prunty, Phys. Stat. Sol., 194, (2), p. 433, (2002). 6. I. Akasaki, J. Cryst. Growth, 237-239, p. 905, (2002). 7. S.J. Pearton, C.R. Abernathy, M.E. Overberg, G.T. Thaler, A.H. Onstine, B.P. Gila, F. Ren, B. Lou, J. Kim, Mat. Today, p. 24, (2002). 8. M. Razeghi, Proc. IEEE, 90, (6), p. 1006, (2002). 9. M.M. Wong, U. Chowdhury, C.J. Collins, B. Yang, J.C. Denyszyn, K.S. Kim, J.C. Campbell, R.D. Dupuis, Phys. Stat. Sol., 188, (1) p. 333, (2001). 10. S. Keller, Y. F. Wu, G. Parish, N. Ziang, J.J. Xu, B.P. Keller, S.P. DenBaars, U.K. Mishra, IEEE Trans. Elec. Dev., 48, (3), p. 552, (2001). 11. I. Akasaki, J. Cryst. Growth, 221, p. 231, (2000). 12. Y.F. Wu, A. Saxler, M. Moore, R.P. Smith, S. Sheppard, P.M. Chavarkar, T. Wisleder, U.K. Mishra, P. Parikh, IEEE Elec. Dev. Lett, 25, (3), p. 117, (2004). 13. J. Elsner, R. Jones, P. Sitch, V. Porezag, M. Elstner, T. Frauenheim, M. Heggie, S. Oborg, P. Briddon, Phys. Rev. Lett. 79, p. 3672, (1997). 14. V. Kirchner, M. Fehrer, S. Figge, H. Heinke, S. Einfeldt, D. Hommel, H. Selke, P. Ryder, Phys, Stat. Sol. 216, p. 659, (1999). 15. D. Look, J. Sizelove, Phys. Rev. Lett. 82, p. 1237, (1999). 101

PAGE 114

102 16. P. Hansen, Y. Strausser, A. Erickson, E. Tarsa, P. Kozodoy, E. Brazel, J. Ibbetson, U. Mishra, V. Narayanamurti, S. DenBaars, J. Speck, Appl. Phys. Lett. 72, p. 2247, (1998). 17. H. Amano, S. Kamiyama, I. Akasaki, Proc. IEEE, 90, (6), p. 1015, (2002). 18. R. Dimitrov, M. Murphy, J. Smart, W. Schaff, J.R. Shealy, L.F. Eastman, O. Ambacher, M. Stutzmann, J. Appl. Phys., 87, (7), p. 3375, (2000). 19. J. Kozlowski, R. Paszkiewicz, R. Korbutowicz, M. Tlaczala, Physica B, 308-310, p. 114, (2001). 20. C.J. Sun, P. Kung, A. Sexler, H. Ohsato, K. Haritos, M. Razeghi, J. Appl. Phys., 75, (8), p. 3964, (1994). 21. N. Onojima, J. Suda, H. Matsunami, J. Cryst, Growth, 237-239, p. 1012, (2002). 22. D.D. Koleske, A.J. Fischer, A.A. Allerman, C.C. Mitchell, K.C. Cross, S.R. Kurtz, J.J. Figiel, K.W. Fullmer, W.G. Breiland, Appl. Phys. Lett., 81, (11), p. 1940, (2002). 23. D. Follstaedt, N. Missert, D. Koleske, C. Mitchell, K. Cross, Appl. Phys. Lett. 83, p. 4797, (2003). 24. D. Kapolnek, X. H. Wu, B. Heying, S. Keller, B. P. Keller, U. K. Mishra, S. P. DenBaars, and J. S. Speck, Appl. Phys. Lett. 67, (11) p.1541, (1995). 25. T. Sugahara, H. Sato, M. Hao, Y. Naoi, S. Kurai, S. Tottori, K. Yamashita, K. Nishino, L. T. Romano, S. Sakai, Jpn. J. Appl. Phys. 37, (4A) p. L398, (1998). 26. M. Hordon, B. Averbach, Acta Metall. 9, (3), p. 237, (1961). 27. C. Dunn, E. Koch, Acta. Metall. 5, (10), p. 548, (1957). 28. G. Williamson, W. Hall, Acta Metall. 1, (1), p. 22, (1953). 29. P. Gay, P. Hirsch, A. Kelly, Acta Metall. 1, (3), p. 315, (1953). 30. T. Metzger, R. Hopler, E. Born, O. Ambacher, M. Stutzmann, R. Stommer, M. Schuster, H. Gobel, S. Christiansen, M. Albrecht, H. Strunk, Phil. Mag. 77, (4), p. 1013, (1998). 31. H. Kang, N. Spencer, D. Nicol, Z. Feng, I. Ferguson, S. Guo, M. Pophristic, B. Peres, Mat. Res. Soc. Symp. Proc., 743, p. 405, (2003). 32. V. Srikant, J. Speck, D. Clarke, J. Appl. Phys. 82, (9), p.4286, (1997). 33. R. Chierchia, T. Bottcher, H. Heinke, S. Einfeldt, S. Figge, D. Hormel, J. Appl. Phys. 93, (11), p. 8918, (2003).

PAGE 115

103 34. Y. Sun, O. Brandt, T. Liu, A. Trampert, K. Ploog, J. Blasing, A. Krost, Appl. Phys. Lett. 81, (26), p. 4928, (2002). 35. B. Heying, X. Wu, S. Keller, Y. Li, D. Kapolnek, B. Keller, S. DenBaars, J. Speck, Appl. Phys. Lett. 68, (5), p. 643, (1996). 36. J. Ayers, J. Cryst. Growth 135, (135), p. 71, (1994). 37. B. Williams, C. Carter, Transmission Electron Microscopy, Plenum Press, New York, 1996, pp. 301-323. 38. N. Zhang, V. Mehrota, S. Chandrasekaran, B. Moran, L. Shen, U. Mishra, E. Etzkorn, D. Clarke, IEEE Pwr. Elec. Spec. Conf., p. 233, (2003). 39. D.D. Koleske, A.E. Wickenden, R.L. Henry, M.E. Twigg, J. Cryst. Growth, 242, p. 55, (2002). 40. W. Gotz, N.M. Johnson, C. Chen, H. Liu, C. Kuo, W. Imler, Appl. Phys. Lett., 68, (22), p. 3144, (1996). 41. W.J. Moore, J.A. Freitas Jr., G.C.B. Braga, R.J. Molnar, S.K. Lee, K.Y. Lee, I.J. Song, Appl. Phys. Lett., 79, (16), p. 2570, (2001). 42. H. Wang, A.B. Chen, J. Appl. Phys., 87, (11), p. 7859, (2000). 43. D. Volm, K. Oettinger, T. Streibl, D. Kovalev, M. Ben-Chorin, J. Diener, B.K. Meyer, J. Majewski, L. Eckey, A. Hoffman, H. Amano, I. Akasaki, K. Hiramatsu, T. Detchprohm, Phys. Rev. B, 53, (24), p. 16543, (1996). 44. X. Xu, H. Liu, C. Shi, Y. Zhao, S. Fung, C.D. Beling, J. Appl. Phys., 90, (12), p. 6130, (2001). 45. J.E. Van Nostrand, J. Solomon, A. Saxler, Q.H. Xie, D.C. Reynolds, D.C. Look, J. Appl. Phys., 87, (12), p. 8766 (2000). 46. V.A. Joshkin, C.A. Parker, S.M. Bedair, J.F. Muth, I.K. Shmagin, R.M. Kolbas, E.L. Piner, R.J. Molnar, J. Appl. Phys., 86, (1), p. 291, (1999). 47. M.A. di Forte-Poisson, F. Huet, A. Romann, M. Tordjman, D. Lancefield, E. Pereira, J. Di Persio, B. Pecz, J. Cryst. Growth, 195, p. 314, (1998). 48. J. Oila, V. Ranki, J. Kivioja, K. Saarinen, P. Hautojarvi, J. Likonen, J.M. Baranowski, K. Pakula, T. Suski, M. Leszczynski, I. Grzegory, Phys. Rev. B, 63, p. 045205-1, (2001). 49. S. Heikman, S. Keller, S.P. DenBaars, U.K. Mishra, Appl. Phys. Lett., 81, (3), p. 439, (2002).

PAGE 116

104 50. J.W.P. Hsu, D.V. Lang, S. Richter, R.N. Kleiman, A.M. Sergent, R.J. Molnar, Appl. Phys. Lett., 77, (18), p. 2873, (2000). 51. M.E. Twigg, D.D. Koleske, A.E. Wickenden, R.L. Henry, S.C. Binari, Appl. Phys. Lett., 79, (26), p. 4322, (2001). 52. D.C. Look, R.J. Molnar, Appl. Phys. Lett., 70, (25), p. 3377, (1997). 53. M.G. Cheong, K.S. Kim, C.S. Oh, N.W. Namgung, G.M. Yang, C.H. Hong, K.Y. Lim, E.K. Suh, K.S. Nahm, H.J. Lee, D.H. Lim, A. Yoshikawa, Appl. Phys. Lett., 77, (16), p. 2557 (2000). 54. C.H. Seager, A.F. Wright, J. Yu, W. Gotz, J. Appl. Phys., 92, (11) p. 6553, (2002). 55. A.F. Wright, J. Appl. Phys., 92, (5), p. 2575, (2002). 56. A.E. Wickenden, D.D. Koleske, R.L. Henry, M.E. Twigg, M. Fatemi, J. Cryst. Growth, 260, p. 54, (2004). 57. R.J. Graham, K.V. Ravi, Appl. Phys. Lett., 60, (11), p. 1310, (1992). 58. R.A. Brown, O. Kononchuk, G.A. Rozgonyi, S. Koveshnikov, A.P. Knights, P.J. Simpson, F. Gonzalez, J. Appl. Phys., 84, (5), p. 2459, (1998). 59. A.E. Wickenden, D.D. Koleske, R.L. Henry, R.J. Gorman, M.E. Twigg, M. Fatemi, J.A. Freitas, W.J. Moore, J. Elec. Mat., 29, (1), p.21, (1999). 60. M. Leszczynski, I. Grzegory, H. Teisseyre, T. Suski, M. Bockowski, J. Jun, J.M. Baranowski, S. Porowski, J. Domagala, J. Cryst. Growth, 169, p.235, (1996). 61. M.J. Callahan, B.G. Wang, L.O. Bouthillette, S.Q. Wang, J.W. Kolis, D.F. Bliss, Mat. Res. Soc. Symp. Proc., 798, p. 263, (2003). 62. T. Goto, M. Tada, Y. Ito, Electrochim. Acta, 39, (8-9), p. 1107, (1994). 63. T. Goto, Y. Ito, Electrochim. Acta, 43, (21-22), p. 3379, (1998). 64. M.J. Gentilcore, Chem. Eng. Proc. Mag. 100, (3), p. 38, (2004).

PAGE 117

BIOGRAPHICAL SKETCH Allen M. West was born on January 26, 1975 in Baltimore, MD. After graduating from the Gilman School in 1994, he attended college at Washington & Lee University in Lexington, VA, where he earned a Bachelor of Science in chemistry-engineering. After graduating, he commenced graduate studies in chemical engineering at Florida State University in Tallahassee, FL. After one year, he transferred to the University of Florida, where he studied under Dr. Oscar Crisalle, and earned a Master of Engineering in chemical engineering. Upon graduation, he joined the Materials Science and Engineering Department, under Dr. Cammy Abernathy. Dr. Abernathy arranged an opportunity for him at Sandia National Laboratories in the spring of 2001, where he completed his graduate research in III-nitride semiconductor materials in 2004. 105