NORTHWESTERN UNIVERSITY

UV Photodetectors, Focal Plane Arrays,

and Avalanche Photodiodes

A DISSERTATION

SUBMITTED TO THE GRADUATE SCHOOL IN PARTIAL FULFILLMENT OF THE REQUIREMENTS

for the degree

DOCTOR OF PHILOSOPHY

Field of Electrical and Computer Engineering

By

Ryan McClintock

EVANSTON, ILLINOIS

June 2007

2

© Copyright by Ryan McClintock 2007 All Rights Reserved

3 Abstract

UV Photodetectors, Focal Plane Arrays,

and Avalanche Photodiodes

Ryan McClintock

The study of III-Nitride based optoelectronics devices is a maturing field, but there are still many underdeveloped areas in which to make a contribution of new and original research.

This work specifically targets the goals of realizing high-efficiency back-illuminated solar- blind photodetectors, solar-blind focal plane arrays, and visible- and solar-blind Avalanche photodiodes. Achieving these goals has required systematic development of the material growth and characterization, device modeling and design, device fabrication and processing, and the device testing and qualification. This work describes the research conducted and presents relevant devices results.

The AlGaN material system has a tunable direct bandgap that is ideally suited to detection of ultraviolet light, however this material system suffers from several key issues, making realization of high-efficiency photodetectors difficult: large dislocation densities, low n-type and p-type doping efficiency, and lattice and thermal expansion mismatches leading to cracking of the material. All of these problems are exacerbated by the increased aluminum

4 compositions necessary in back-illuminated and solar-blind devices. Overcoming these obstacles has required extensive development and optimization of the material growth techniques necessary: this includes everything from the growth of the buffer and template, to the growth of the active region.

The broad area devices realized in this work demonstrate a quantum efficiency that is among the highest ever reported for a back-illuminated solar-blind photodetector (responsivity of 157 mA/W at 280nm, external quantum efficiency of 68%). Taking advantage of the back illuminated nature of these detectors, we have successfully developed the technology to hybridize and test a solar-blind focal plane array camera. The initial focal plane array shows good uniformity and reasonable operability, and several images from this first camera are presented. However, in order to improve the performance of these devices to the point where they can effectively compete with photo-multiplier tube technology, it is necessary to develop devices with internal gain. To this end GaN and AlGaN based avalanche photodiodes have been studied, and we report the first realization of a solar-blind back-illuminated avalanche photodiode. The next logical step is to continue this work and realize Geiger mode avalanche photodiodes capable of single photon detection.

5 Acknowledgements

I would like to first thank my research advisor Professor Manijeh Razeghi for her continued support and guidance throughout both my undergraduate and graduate education here at Northwestern University. Without her, and the wonderful research facilities she has provided here at the Center for Quantum Devices, this work would never have been possible.

I would also like to thank my committee members for agreeing to participate in my defense, and would like to acknowledge their support and encouragement.

I would like to acknowledge the federal government for providing three years of support under a National Defense Science and Education Graduate fellowship. As well, I would like to acknowledge Northwestern University for supporting my fourth year with a University

Fellowship. And I would finally like to acknowledge the Richter Trust for supporting me during my dissertation year with a Richter Trust Fellowship.

I would also like to specially thank my close colleagues studying III-Nitrides: Kathryn

Minder, Dr. Patrick Kung, Dr. Jose L. Pau, Alireza Yasan, and Can Bayram Without their assistance and support, much of this work would not have been possible. I would also like to acknowledge all of the other graduate students, both past and present, at the Center for

Quantum Devices.

Finally I would like to thank my parents: Dr. Marty McClintock and Mrs. Nancy

McClintock. To them I am eternally grateful, and I doubt that I will ever fully understand everything they have done for me.

6 i. Table of Contents

Abstract ...... 3 Acknowledgements...... 5 i. Table of Contents...... 6 ii. List of Figures ...... 9 iii. List of Tables...... 19 1. Introduction ...... 20 2. Background...... 23 2.1. The Solar Ultraviolet Spectrum 23 2.2. UV Photodetector Applications 25 2.3. UV Detection Technologies 31 2.4. Historic Development of III-Nitride Based UV Photodetectors 34 2.4.1. Photoconductors 35 2.4.2. Schottky Metal-Semiconductor-Metal Detectors 37 2.4.3. Schottky Barrier Photodiodes 40 2.4.4. Photocathodes 41 2.4.5. Front Illuminated p-i-n Photodiodes 42 2.4.6. Back Illuminated p-i-n Photodiodes 45 2.4.7. Avalanche photodiodes 48 3. Important UV Photodetector Characteristics...... 52 3.1. General Photodetector Parameters 52 3.2. Basic Noise Analysis Theory 54 3.3. Noise Analysis in AlGaN p-i-n Photodiodes 59 3.4. Avalanche Photodiode Parameters 61 3.5. Noise Analysis in Avalanche Photodiodes 65 4. Wide Band-Gap III-Nitride Material Growth...... 69 4.1. Introduction 69 4.2. Metal Organic Chemical Vapor Deposition, an Overview 69 4.3. Growth Nucleation: Low and Intermediate Temperature Buffers 78 4.4. Atomic Layer Epitaxy for Growth of AlN and AlGaN 81 4.4.1. Growth of AlN by ALE 82 4.4.2. Growth of AlGaN / AlN Superlattice Template 85 4.5. Growth and Doping of Wide-Bandgap AlGaN Material 89 4.5.1. N-type Doping of AlGaN 89 4.5.2. P-Type Doping of AlGaN 92 5. Experimental Procedure: Large Area Single Element Detectors ...... 95 5.1. Introduction 95 5.2. Material Growth and Characterization 96 5.3. p-i-n Photodetector Processing Overview 104

7 5.4. Photodetector Measurement and Discussion 105 5.5. Deep UV Back-Illuminated Photodetectors (255 nm) 111 6. Experimental Procedure: Solar-Blind Focal Plane Arrays...... 114 6.1. Introduction 114 6.1.1. Focal Plane Array Technology 115 6.1.2. Historic Development of UV Focal Plane Arrays 117 6.2. Material Growth and Characterization 119 6.3. FPA pixel characteristics 120 6.4. FPA Processing Overview 124 6.5. FPA Images and Discussion 126 6.6. Improvement of Device Performance 129 7. Experimental Procedure: AlGaN Based Avalanche Photodiodes...... 133 7.1. Introduction 133 7.2. Material growth and device processing 135 7.2.1. Material Growth 135 7.2.2. Device Processing 137 7.3. Unbiased device performance 138 7.3.1. Current-voltage characteristics at low bias 138 7.3.2. Photoresponse 141 7.4. Avalanche Mode device operation 142 7.4.1. Current-voltage curves under bias 142 7.4.2. Device Modeling 146 7.4.3. Ruling out other origins for the gain 149 7.5. Conclusion 149 8. Experimental Procedure: GaN Based Avalanche Photodiodes...... 151 8.1. Introduction 151 8.2. Material Growth 152 8.3. Device Processing 155 8.4. Device Results 156 8.4.1. I-V measurements 156 8.4.2. Gain Measurements 162 8.4.3. Origins of the observed multiplication 167 8.4.4. Noise Analysis 168 8.5. Conclusion 170 9. Future Work: Geiger Mode APDs...... 172 9.1. Introduction 172 9.2. Preliminary Device Results 173 10. Appendices ...... 177 10.1. Appendix 1: Material Characterization Techniques 177 10.1.1. Structural Characterization 177 10.1.2. Optical Characterization 189 10.1.3. Electrical Characterization 195 10.2. Appendix 2: Device Testing and Characterization 200 10.2.1. Current Voltage Measurement 200 10.2.2. Responsivity and External Quantum Efficiency 203

8 10.2.3. Bias Dependant Responsivity 209 10.2.4. Gain Measurement 210 10.2.5. Noise Measurement 211 10.2.6. Geiger Mode APD Measurement 213 10.3. Appendix 3: ROIC Specifications and Camera Interfacing 216 10.3.1. ROIC Specifications 216 10.3.2. Camera System Electronics 220 10.3.3. Imaging Optics 223 10.3.4. Imaging Software and Image Correction 227 10.4. Appendix 4: Sample UV FPA movies 231 10.4.1. Movie 1: Moving CQD Logo 231 10.4.2. Movie 2: Dancing Exposed Electric Arc 232 10.4.3. Movie 3: Reflection from a Patterned Mirror 233 10.5. Appendix 5: Development of a Portable Camera System 234 10.5.1. Project Goal: 234 10.5.2. Software Design: 240 10.5.3. Hardware Design: 264 10.6. Appendix 6: Development of a Semiconductor Wafer Cleaning System 271 10.6.1. Systems Overview 273 10.6.2. Control and Software Design 279 11. References ...... 283 12. Curriculum Vita ...... 301 12.1. Education 301 12.2. Honors 301 12.3. Graduate Coursework 301 12.4. List of Publications 302

9 ii. List of Figures

Figure 1. Solar Spectral Irradiance as seen from space (light blue), and from sea level (orange). On earth absorption by atmospheric ozone strongly attenuates wavelengths less than ~290nm thus creating a solar blind window (data taken from ref. 1)...... 24

Figure 2. III-Nitride UV photodetectors have a number of civilian and military applications as illustrated above...... 26

Figure 3. Illustrations of the operation of secure UV-based NLOS communication system in a variety of environments...... 27

Figure 4. UV Absorption and fluorescence of three common biological markers (tyrosine, tryptophan, NADH) illustrating the detector and emitter requirements...... 29

Figure 5. The quaternary AlInGaN material system is indicated in blue. For comparison other viable semiconductor materials are shown in black...... 32

Figure 6. Detective Quantum Efficiencies of materials for potential use in UV imaging applications. In the solar blind region of the spectrum AlGaN has a clear advantage over both photocathodes such as Cs2Te , and UV enhanced CCDs...... 33

Figure 7. Alternatives to AlGaN for solar blind imaging, and their disadvantages ...... 34

Figure 8. Photoconductors exhibitinbg sharp cut-off wavelengths covering the entire AlxGa1-xN compositional range: from GaN (365nm) to AlN (200 nm). The inset shows a simplified photoconductor structure...... 36

Figure 9. Shows the geometry of a typical interdigitated finger MSM device with a length of 150 μm a finger width of 2 μm and a pitch of 10 μm...... 38

Figure 10. Shows a typical spectral response of a Schottky based MSM photodetector...... 39

Figure 11. Shows the typical spectral response of a Schottky barrier photodetector. The corresponding device structure is shown in the inset...... 41

Figure 12. Shows the typical structure of a front illuminated AlxGa1-xN based p-i-n photodiode...... 43

Figure 13. The spectral response of a set of front illuminated AlxGa1-xN p-i-n photodiodes with various Al compositions. The line at the top indicates the theoretical maximum for a p-i-n photodetector corresponding to a quantum efficiency of one...... 45

10 Figure 14. The basic structure of a back illuminated p-i-n photodetector...... 46

Figure 15. Typical photoresponse of a back-illuminated solar blind photodetector showing the difference between the front- and back-illuminated responses...... 47

Figure 16. Mosaic image of micro-plasma luminescence from an un-optimized array of 16 diodes showing the luminescent nature of the process, and the distribution of defects...... 49

Figure 17. GaN based APD showing an avalanche multiplication of nearly 1500 at a breakdown field strength of 2.7 MV/cm...... 50

Figure 18. AlGaN-Based Solar-Blind APD showing a maximum gain of 2000 at an electric field of 7 MV/cm...... 51

Figure 19. The dark current of GaN p-i-n photodetector at different applied biases...... 56

Figure 20. The noise power spectral density of a GaN p-i-n photodetector...... 58

Figure 21. Signal, noise, and dark current contributions to the output of an APD...... 63

Figure 22. Noise power spectral density versus bias around breakdown for a GaN APD...... 66

Figure 23. Aixtron AIX 200/4 MOCVD reactor used for growth of III-Nitride material discussed in this work...... 70

Figure 24. Schematic diagram of the MOCVD reactor used to grow the material discussed in this work. This diagram shows the full gas handling system, the reactor growth chamber, and the vacuum system...... 72

Figure 25. Schematic diagram of the 4 step MOCVD growth process...... 74

Figure 26. Growth efficiency of Al0.4Ga0.6N at different growth pressures. The dashed lines mark the growth efficiency of 1000 mm/mole corresponding to the growth pressure of ~35 mbar...... 76

Figure 27. Photoluminescence for a series of AlGaN:Si layers grown at different growth pressures showing the material quality tradeoff associated with growth at lower pressures...... 77

Figure 28. Crystallography of AlN grown on top of sapphire. The image at the left shows a plan view. This illustrates to the 30 degree rotationof the crystal structure that leads to a 13.2 % effective lattice mismatch. The image at the right shows a sectional view showing the formation of a misfit dislocation every 8 atoms...... 79

Figure 29. High resolution TEM image of the sapphire/ AlN interface showing the generation of misfit dislocations...... 81

11 Figure 30. Diagram of the valve switching used to grow AlN by atomic layer epitaxy...... 83

Figure 31. High resolution x-ray diffraction scans from an ALE grown AlN, showing an extremely narrow systemic (002) and narrow asymetric(105) width owing to the high-quality of the material...... 84

Figure 32. AFM Image of the surface of ALE grown AlN on a low temperature AlN buffer...... 84

Figure 33. Diagram of the valve switching sequence used to grow high aluminum composition AlGaN by atomic layer epitaxial...... 86

Figure 34. AFM image showing the high-quality of the surface of an AlGaN/AlN SL grown on AlN, all grown using the ALE technique...... 87

Figure 35. Macroscopic hexagonal etch pits are formed by the worst of the defects after etching in a soultion of hot KOH or H3PO4...... 88

Figure 36. Activation energy of Mg in AlxGa1-xN as a function of Al mole fraction...... 93

Figure 37. AFM image showing the surface of the high-quality AlGaN/AlN SL template, 3 nm data scale. The RMS roughness for the 5 mm square scan shown above is 1.3 Å...... 97

Figure 38. Optical transmission and absorbance squared of the Al0.5Ga0.5N:Si:In conduction layer grown on a high-quality AlN/AlGaN SL template. The conduction layer shows a sharp cut off owning to the excellent material quality, with the absorption edge occurring at 260 nm...... 100

Figure 39. Schematic cross-section showing the structure of a back-illuminated solar- blind photodetector...... 101

Figure 40. Responsivity vs. wavelength for a Back illuminated p-i-n photodiode, showing a significant negative photoresponse...... 102

Figure 41. I-V curves for several different Cp2Mg flows used to optimize the p-type layers. The structure and measurement geometry are shown in the lower right hand corner...... 103

Figure 42. Optical micrograph of a 1mm x 1mm Solar-blind photodetectors showing the device layout, and the proximity of the n-contact...... 104

Figure 43. I-V curve of a typical back illuminated photodiode in semi-log scale...... 106

Figure 44. Natural log of current versus voltage. The low current regime is modeled and an ideality factor of 2.89 is extracted...... 107

12 Figure 45. Responsivity vs. wavelength for a typical photodiode, showing a peak responsivity of 150 mA/W at a wavelength of 280 nm, and no negative photoresponse. This peak responsivity corresponds to an external quantum efficiency of 68%...... 108

Figure 46. Shows the external quantum efficiency of the device in linear scale...... 109

Figure 47. Schematic diagram of the reflective and transmission losses in our device...... 110

Figure 48. 255 nm deep UV photodetector device structure...... 111

Figure 49. Unbiased responsivity in log scale (left) and the corresponding unbiased external quantum efficiency in linear scale (right)...... 113

Figure 50. Shows a schematic cross-section of a III-Nitride FPA flip-chip bonded to a Si ROIC...... 116

Figure 51. Shows a simplified schematic of a typical CITA unit cell as used in a ROIC...... 116

Figure 52. I-V curve of a single 25 μm × 25 μm FPA pixel shown in linear scale. The inset shows the same data in logarithmic scale...... 121

Figure 53. I-V statistics for the array, showing the good current spreading...... 122

Figure 54. Unbiased responsivity in log scale (left) and the corresponding unbiased external quantum efficiency in linear scale (right) from a representative 25 μm x 25 μm pixel...... 123

Figure 55. Electron micrograph of the focal plane array before bonding to the ROIC: after lithographic patterning, metallization, and deposition of the indium bumps...... 124

Figure 56. 320 × 256 Image of: A) a paper-cutout in front of a Germicidal lamp and B) an electric arc as seen with the solar-blind FPA camera...... 126

Figure 57. Image of a human likeness as seen with the solar-blind FPA camera...... 127

Figure 58. Histogram of Pixel intensity for a flat field of UV illumination. This figure shows the variation in pixel response within the array...... 128

Figure 59. Cross-sectional diagram of the current and proposed FPA designs. The top half shows the current design illustrating how a significant amount of light reaches the ROIC, the bottom half shows the proposed solution that will make the FPA 100% opaque...... 130

Figure 60. Top view of an FPA processed with a dark-field layer. Gold is the top contact, and the maroon area surrounding the top contacts in the black-out layer, it

13 almost overlaps the top contact and blocks out 99% of the area that would otherwise be exposed...... 131

Figure 61. 15 μm tall indium bump after reflow, seen from an oblique angle...... 132

Figure 62. Comparison of exiting UV detection technologies to AlGaN based detectors, showing the gap between the detectivity of PMTs and AlGaN based devices, and the need for APDs (in red). (After references and ) ...... 134

Figure 63. Schematic cross-section showing the device structure of the back-illuminated APD...... 136

Figure 64. Scanning electron micrograph of an APD after processing. The common n- contact(not shown) is far removed from the mesas to avoid air breakdown of the devices...... 138

Figure 65. Comparison of Dark current density between 1mm x 1mm devices and 25 μm x 25 μm devices, showing the disparity in scaling of the dark current...... 139

Figure 66. Left) Current-Voltage curve showing the current density in log scale. Right) Natural log of current and fit to linear region used to extract an ideality factor (n) of 2.8...... 140

Figure 67. Left) Unbiased responsivity shown in log scale. Right) External quantum efficiency shown in linear scale...... 142

Figure 68. Current-voltage behavior as a function of applied reverse bias, both under illumination and in the dark. Errors bars have been added to indicate the variation over 3 consecutive pairs of alternating light-dark measurements of the same diode...... 143

Figure 69. Calculated photocurrent (left axis) and corresponding gain (right axis) from the data of Figure 68. Error bars indicate +/- 1 standard deviation. Breakdown is taken at 40 volts...... 144

Figure 70. Schematic diagram of parallel and sequential ionization. Sequential ionization (left) can lead to Geiger mode operation, however parallel (right) restricts the multiplication to a geometric increase...... 146

Figure 71. Electric Field Profile under various applied reverse biases...... 147

Figure 72. Avalanche gain model of the device. The solid curve shows the experimental data, the dashed curve shows the model...... 148

Figure 73. Atomic force microscopy (AFM) imaging of the surface of a high-quality GaN layer grown on an AlN template...... 153

14 Figure 74. High-resolution x-ray diffraction of GaN on an AlN template layer...... 154

Figure 75. Schematic diagram of GaN APDs on AlN templates (left) and GaN templates (right)...... 155

Figure 76. Schematic diagram of the APD as processed...... 156

Figure 77. Breakdown voltage as a function of intrinsic layer thickness used to extract a critical electric field strength of 2.73 MV/cm...... 158

Figure 78. Model of the electric field profile across the device as a function of the applied reverse bias: experimentally breakdown occurred at 102 V corresponding to 3.2 MV/cm in the model...... 159

Figure 79. Breakdown characteristics of samples A, B, and C are shown. Inset: The experimental breakdown voltages obtained for different thicknesses of the intrinsic layer...... 161

Figure 80. Variation of dark current of a GaN APD biased 2V above breakdown. The standard deviation is less than 6% over more than 60 hours...... 162

Figure 81. Multiplication factors for electrons (Mn) and holes (Mp) obtained from sample A (left) and sample B (right)...... 164

Figure 82. Solid lines: ionization factors obtained for electrons (αn) and holes (βp) from experiment. The dashed lines represent theoretical values for βp and αn, as extracted from ref 180...... 165

Figure 83. Evolution of the spectral response of a GaN APD near breakdown. Spectra are shown on the left, and the evolution at three selected wavelengths are shown on the right...... 166

Figure 84. Evolution of the device breakdown voltage with temperature...... 167

Figure 85. The Spectral Power Density (Sn) of sample A at the onset of breakdown is shown from 91 V to 102 V with 1 V steps. The two narrow spikes at 60 and 120 Hz correspond to line noise...... 169

Figure 86. Left) Spectral power density is plotted as a function of total current for sample A under front- (triangles) and back-illumination (squares). Right) calculated excess noise factors for front- and back-illumination...... 170

Figure 87. (Top) Schematic diagram of the passive quenching circuit used to apply the DC bias and AC excitation pulse to the Geiger mode APDs...... 174

Figure 88. Geiger-mode spectral response of a GaN APD detecting at the 10 photons-per- pulse level...... 175

15 Figure 89. Left) Linear mode photoresponse at 75V reverse bias. Right) Geiger mode photoresponse with 10 Photons/pulse illumination...... 176

Figure 90. A.) Optical Microscope used to investigate broad area surface morphology B.) Representative optical micrograph taken at a magnification of 100x, showing a 50% AlGaN sample with a large number of cracks...... 179

Figure 91. Custom Designed software created for capturing of still images using an optical microscope...... 180

Figure 92. Schematic diagram of a scanning electron microscope showing the electron source, the lenses and scanning coils, the sample under test, and the electron detector (based on ref. )...... 181

Figure 93. Schematic diagram of an atomic force microscope (Based on Digital Instrument Multimode SPM manuals)...... 184

Figure 94. Diffraction from a set of crystal planes according to Bragg’s law...... 187

Figure 95. UV photoluminescence setup showing the three laser excitation sources that can be used to stimulate PL, as well as the sample stage, focusing optics and monochromator...... 190

Figure 96. Custom Labview software developed as part of this work to facilitate the measurement and analysis of routine Photoluminescence...... 192

Figure 97. Ultraviolet transmission measurement system showing the xenon lamp, monochromator, chopper, lens, sample holder, and the UV-enhanced silicon photodetector...... 193

Figure 98. Custom Labview software developed as part of this work to measure UV transmission...... 194

Figure 99. Custom Hall effect measurement electronics assembled to facilitate the measurement of high impedance III-Nitride materials...... 196

Figure 100. Schematic diagram of a typical hall mobility measurement system showing the van der Pauw contact geometry as used for measurement of the resitivity (A) and for the determination of the hall coefficient (B)...... 198

Figure 101. Current-Voltage characterization setup used to measure and record I-V curves. The probe station has probe-tips as small as .5 μm, and is suitable for probing to individual pixels of an unbounded FPA...... 201

Figure 102. Typical IV curve of a single 25μm x 25μm FPA pixel shown in log scale. The linear fit of this data gives an ideality factor of 3.7; slopes for ideality factors of 1 and 2 are displayed for comparison purposes...... 202

16 Figure 103. Measurement setup used to characterize the photoresponse of both broad- area photodetectors and individual text pixels from an FPA...... 205

Figure 104. Photograph of a wafer on the probe-station used for responsivity measurements showing the fiber optical cables (above and below wafer) and the two triaxial probes...... 206

Figure 105. Custom written software used to make UV detector responsivity and external quantum efficiency measurements...... 207

Figure 106. Xenon lamp flux as a function of time since turning on the lamp showing the variation during lamp warm up. The blue bars indicate a +/- 1% variation in xenon lamp power...... 208

Figure 107. Example of permutation of data from gain measurements used to generate gain data and error bars...... 211

Figure 108. Custom software written to facilitate the collection of noise Spectral Power Density as a function of reverse bias for characterization of photodetectors and APDs...... 213

Figure 109. Left) diagram showing the self quenching that the 100kΩ resistor ideally provides to the APD, Right) Schematic diagram of the Geiger mode APD biasing circuit...... 214

Figure 110. Indigo ROIC with 320 x 256 FPA bonded to it. The chip is only ~1 cm x 1cm in size and contains almost all of the electronics necessary to operate as a solar-blind UV imager...... 217

Figure 111. General overview of ROIC specifications and a list of the imaging conditions used to obtain all of the images presented within the context of this work...... 218

Figure 112. Focal plane array camera system used to record images presented in this work. Shown with side panel removed to illustrate the internal electronics necessary to operate the ROIC. NOTE: Although a Dewar is shown, the FPA is only operated at room temperature under an ambient atmosphere...... 222

Figure 113. Transmission spectrum of the 280 nm band-pass filter used in camera head optics...... 224

Figure 114. Schematic diagram of the optics comprising the solar-blind imaging system used to obtain the images presented in this work...... 225

Figure 115. Schematic Diagram of the optics used to record reflective UV images using the solar-blind focal plane array...... 226

17 Figure 116. Schematic diagram of the imaging optics used to record images for emissive UV sources, such as an electric arc...... 227

Figure 117. This figure shows a single frame from the middle of the attached movie #1...... 231

Figure 118. This figure shows the first frame of the attached movie #2...... 232

Figure 119. This figure shows the first frame of the attached movie #3...... 233

Figure 120. The system is small, lightweight, and can be easily packaged for transport to diverse locations in a hard carrying case requiring only LN2 and a laptop computer for operation...... 234

Figure 121. Schematic diagram of the components that make up the portable camera system...... 236

Figure 122. Overview of the electronics package layout ...... 237

Figure 123. Screen capture of the various windows that make up the portable camera system user interface...... 238

Figure 124. Virtex 4 Mini-Module used to run the embedded portion of the software, also shows the EEPROM, RAM, and Ethernet-PHY used by the Virtex-4 FPGA...... 239

Figure 125. Schematic diagram showing the building blocks (and their sub components) that constitute the portable camera system software...... 240

Figure 126. Main program window...... 242

Figure 127. Right-Click pixel processing details as shown on the status bar...... 245

Figure 128. Video scope and camera setup user interface ...... 246

Figure 129. Background subtraction and limited bad-pixel replacement user interface...... 249

Figure 130. Two-point NUC and bad-pixel detection user interface ...... 252

Figure 131. Video capture user interface ...... 253

Figure 132. FPA camera system image transport packet layout...... 254

Figure 133. User interface for low-level control of programmable camera biases...... 255

Figure 134. Control serial word interface for Indigo 9705 & 9809 ROICS...... 256

Figure 135. Windowing serial word interface for Indigo 9705 & 9809 ROICs...... 257

Figure 136. FPA Camera Source interface, showing diagnostic information...... 258

18 Figure 137. Advanced image processing user interface...... 259

Figure 138. Brightness and contrast user interface. Allows adjustment of gamma and colorization...... 261

Figure 139. Cross-sectional view of the camera system electronics showing the front and back planes showing the attachment to the dewar...... 265

Figure 140. Embedded CPU Daughter card and power supply board...... 266

Figure 141. Custom designed analog card with video gain pipeline, programmable offset generator, and two channel 14 bit 20MSPS ADC...... 268

Figure 142. 6 channel programmable bias card...... 269

Figure 143. Clock driver and 2 channel programmable bias card...... 270

Figure 144. Wafer holder (center) and multi-nozzle spray arm (top) showing the Semiconductor wafer cleaning system installed in a chemical hood...... 272

Figure 145. General layout of the semiconductor wafer cleaning system, showing the location of support equipment in the chase, away from the main use interface which occupies minimum space in the hood...... 273

Figure 146. Left) picture of the insulated counter-current heat exchangers where they meet the valves. Right> schematic diagram of the three heat exchanger located behind the hood...... 274

Figure 147. Left) Photo of the solvent storage tanks located within the flammable storage cabinet. Right) Schematic diagram of the solvent pressurization plumbing (green), flexible lines (blue) and tanks (red)...... 275

Figure 148. Left) Photo of bottom of wafer fixture showing water lift pump (top shelf), and waste collection container (bottom shelf), Right) schematic diagram of the waste collection plumbing showing the lift pump draining waste water into a nearby acids sink...... 278

Figure 149. Semiconductor wafer cleaning system electronics and pneumatics control package...... 281

Figure 150. Main control panel for interfacing with the semiconductor wafer cleaning system...... 282

19 iii. List of Tables

Table 1. Dark current and shot noise levels for various reverse bias voltage values...... 57

Table 2. Sources used in MOCVD growth and doping of Al(In)GaN...... 71

Table 3. Comparison of electrical properties of n-type Al0.5Ga0.5N grown with different approaches to the doping of the material...... 90

Table 4. Table of approximate carrier concentrations of the various layers used later in the modeling of this device structure...... 136

Table 5. Table of device structures and limited device characteristics...... 159

Table 6. Serial mode word bits and the functions they control...... 219

Table 7. List of the 14 signals necessary for the operation of an FPA based on the Indigo ROIC ...... 220

Table 8. A list of the available menu functions as well as their descriptions...... 243

20 1. Introduction

The study of III-Nitride based optoelectronics devices is a relatively new and exciting field, and there are still many underdeveloped areas in which to make a contribution of new and original research. This work specifically targets the goals of realizing high efficiency back- illuminated solar-blind photodetectors, solar-blind focal plane arrays, and Avalanche photodiodes. However to achieve these goals has required systematic development of the material growth and characterization, device modeling and design, device fabrication and processing, and the device testing and qualification.

This work has encompasses a time span of over 7 years, has lead to more than 26 journal articles and conference proceedings, and has even been reported in the mainstream press several times. At the start of this research GaN material growth had been well established and high efficiency top-illuminated GaN based photodetectors had already been realized.

Preliminary AlGaN growth techniques were already established at the center for quantum devices, and top-illuminated photodetectors of varying quality had already been demonstrated across the material compositional range. However, little research had really been conducted into the realization of back-illuminated photodetectors (necessary for focal-plane arrays). What back-illuminated solar-blind photodetectors did exist had vastly inferior performance, showing an eternal quantum efficiency of less than 15%. Before the start of this work we had not established any capabilities in either focal plane arrays (ultraviolet or infrared), or avalanche photodiodes.

In order to realize high quantum efficiency back-illuminated solar-blind photodetectors it was necessary to systematically develop a new approach to the growth of high-quality high-

21 aluminum composition materials. This involved complete optimization of every layer involved in the growth of back-illuminated ultraviolet photodetectors: all the way from the buffer layer at the bottom, all the way to the top contact layer. For each layer, all growth- related parameters such as the growth temperature, the growth pressure, reactor gas flows, and the V/III ratio were carefully studied and optimized to maximize the device performance. New buffer layers and novel low V/III ratio growth techniques were developed to realize a new high-quality AlN template layer growth technique that was instrumental in the realization of the high quantum-efficiency back illuminated solar-blind photodetectors, focal-plane arrays, and avalanche photodiodes. With these optimizations, and this new template, we realized devices with an external quantum efficiency of 68%; this is still among the highest values ever reported in the literature, and was a significant achievement at the time.

With the realization of high quantum efficiency back-illuminated photodetectors the work proceed to the next logical step, realization of a solar-blind focal plane array camera. This involved significant work to develop the device processing, control the device uniformity, develop of indium bump deposition and hybridization technology, and to develop testing and characterization techniques for focal-plane arrays. Due to the immense complexity and cost associated with realizing ultraviolet focal-plane array camera much of this work was carried out in conjunction with parallel work to develop infrared camera technology. However in the end this work paid off: we used a commercial read-out integrated circuit to successfully realize the first solar-blind focal plane array fabricated entirely at a university. This camera has been further extended, and as part of this work we have developed an entirely custom portable camera system to electrically operate these focal-plane arrays, and provide imaging.

22 The final stage of this work has been the development of avalanche photodiodes operating in the ultraviolet. This is an entirely original area of research that had been little explored in the past. Our material and processing improvements have allowed us to realize avalanche photodiodes. We were the first to report avalanche multiplication from a back- illuminated solar-blind avalanche photodiodes. These devices had a gain in excess of 1000, which was as high as any GaN based device reported at that time. We have gone on from here to go back and more extensively study GaN based avalanche photodiodes. We were the first to report back-illuminated p-i-n based GaN avalanche photodiodes, and are currently pursuing these for array applications. We have also made significant study of the ionization characteristics of holes and electrons in GaN, resulting in one of the first experimental determinations of the impact ionization coefficients in GaN.

The next logical step for this research is the realization avalanche photodiodes operating in Geiger mode, and capable of single photon detection. Significant strides have already been made in this direction.

23 2. Background

2.1. The Solar Ultraviolet Spectrum

This work focuses exclusively on the ultraviolet (UV) portion of this spectrum. The majority of the solar radiation that reaches the earth lies close to the visible region of the spectrum. The intensity falls off from there slowly into the infrared and more quickly into the ultraviolet with the spectrum resembling that of a typical blackbody source with a temperature of approximately 5800 K. This solar spectral irradiance is shown in Figure 1; wherein the data is taken from reference1.

Ultraviolet light is defined as light having a wavelength less than about 400 nm, but longer than that of soft X-rays. This portion of the spectrum can be divided into four major regions: UV A covering wavelengths in the range of 400 to 315 nm, UV B covering 315 to 280,

UV C, 200 to 280 nm,2 and vacuum UV or far UV, between 200 nm and ~10 nm. AlGaN based III-Nitride photodetectors are well poised to cover a large portion of the UV spectrum:

AlGaN ranges from binary GaN with a direct band gap of 3.4 eV (365 nm), to binary AlN with a band gap of 6.2 eV (200 nm). Latter in this section this versatility is demonstrated for both photoconductors and p-i-n photodiodes in Figure 8 and in Figure 13, respectively.

24

Solar Blind Window: λ= 240 to 290 nm

Figure 1. Solar Spectral Irradiance as seen from space (light blue), and from sea level

(orange). On earth absorption by atmospheric ozone strongly attenuates wavelengths less than

~290nm thus creating a solar blind window (data taken from ref. 1).

In terms of UV detectors, a further distinction in made in the UV spectrum by separating devices based upon the strategic wavelength of 290 nm. Detectors with a cutoff wavelength less than 400 nm are termed visible-blind, due to their insensitivity to visible radiation. This is a desirable UV detector trait in that it reduces interference from visible light.

However as shown above in Figure 1 the solar irradiance reaching the earth is still significant from 400nm down to about 290 nm where is falls off very strongly. Below 290nm almost no light reaches the earth’s surface due to atmospheric absorption by ozone (O3) in the upper atmosphere3. This creates a universal low background window called the solar-blind window; and because of this, UV detectors with a cutoff wavelength less than 290nm are termed solar-

25 blind detectors, and are ideal for terrestrial detection of man-made UV sources. The early III- nitride detector research focused on visible blind detectors, but recently as the growth of high

Al-composition AlGaN material has matured; interest has focused more strongly on detectors operating in the solar blind region of the ultraviolet spectrum.

2.2. UV Photodetector Applications

The development of UV photodetectors has been driven by numerous applications in the defense, commercial, and scientific arenas. These include, for example, covert space-to- space communications, secure non line-of-sight communications, early missile threat detection,

UV spectroscopy, chemical and biological threat detection, flame detection and monitoring, power line monitoring, UV environmental monitoring, and UV astronomy 4, 5 , 6. In the past few years, technological and scientific advances in high Al composition AlGaN and AlN based semiconductor materials have led to a renewed interest in ultraviolet photodetectors, especially solar blind photodetectors (due to the low natural background). Solar-blind detectors allow for a number of unique applications, and the vast majority of the applications listed above take advantage of the solar blind region of the spectrum as illustrated in Figure 2 below.

26

Figure 2. III-Nitride UV photodetectors have a number of civilian and military applications as illustrated above.

Heat sources such as flames, jet engines, or missile plumes emit light throughout the

UV portion of the spectrum corresponding to their black-body temperature. These man-made

UV sources can easily be detected at wavelengths less than 290 nm due to the non-existence of a terrestrial background signature. The military in particular is interested developing ground an air based solar-blind sensors to detect the UV signature of an active missile plume, and provide early warning and potentially allow for missile tracking and ultimately interception.7,8

A s

olar-blind UV based tracking device would be immune to solar interference and thus capable of tracking a signature across the whole horizon. For non-tracking applications focused only on threat detection, utilizing a solar-blind detector greatly simplifies the design of the system, eliminating the need to monitor and discriminate against the background.

27 Another important application of UV photodetectors or special interest to the government is covert UV-based non-line-of-sight (NLOS) communications.9 UV-NLOS communications is a secure means to send data using low-power UV sources, and relies on the strong back-scattering of UV and low natural background. The basic operation of such a system is illustrated in Figure 3 below.10 The emitter is pointed towards the sky and the UV light is scattered back towards the earth; due to the short wavelength UV is scattered more strongly than other wavelengths. This makes the system ideal for use where a direct line-of-sight cannot be established, such as in dense terrain or in an Urban-canyon environment when presence of tall cement and iron buildings would make radio communications difficult.

Urban Canyons

Principle of Operation Rayleigh Scattering

Squad and Device Communications

10 to 250 m

Source Detector λ≤280nm λ≤280nm

Figure 3. Illustrations of the operation of secure UV-based NLOS communication system in a variety of environments.

28 Solar-blind UV has a strong extinction coefficient, due to both high scattering and the high absorption at these wavelengths. This makes eavesdropping on the NLOS communication’s UV signal very difficult from any significant distance, particularly in the forward direction: this is in marked contrast to conventional RF which can travel thousands of miles or more. The UV signal is almost completely extinguished after a distance of approximately 250 meters making this a secure covert means of communication. Both the UV detectors and the UV sources necessary for such a system can be realized in the III-Nitride material system making it an ideal choice for the development of a compact secure portable communication system.

Another significant government application of UV detectors (and sources) is biological agent detection. Biological agents could have devastating effects on public health, as the anthrax scare of 2001 made us all too aware.11 There is a significant lag time between a covert attack and the wide-spread appearance of symptoms which makes the general lack of readily available real time detection systems a significant problem. These agents, such as anthrax, smallpox, Marburg virus, Ebola virus, pneumonic plague, and tularemia can, in principle, be simply manufactured and transported in mass quantities, and can cause high rates of mortality if sufficient mitigation procedures are not enacted in a timely manner in place. This makes the development and dissemination of an effective low cost real time detection system a critical weapon in the defense against a bio-terror attack, allowing authorities to time to warn the population, identify the contaminated areas, and enact quarantine procedures before the exposure overwhelms response capabilities.12 The quicker the detection can be made, the less the spread of biological agents will be allowed to spread limiting the quarantine area necessary,

29 and saving more lives, as well as reducing the expenditure of resources necessary to enact a safe and effective cleanup.

Many of the organic and inorganic compounds that make up a bio-agent have absorption lines and/or florescence lines in the UV region of the spectrum. By utilizing the

AlGaN material systems to realize detectors and emitters tuned to these strategic wavelengths, it is possible to create a spectral fingerprint which can be used to identify the presence of specific biological-agents in real-time.13,14 Figure 4 shows the absorption and emission spectra of the three most important biological markers, and the role that UV detectors and emitters play in their spectroscopic detection.

NEED: NEED: UV Emitters at Photodetectors/filters with cutoff λ~280, 340 nm λ~300~470 nm

274 nm 340 nm 470 nm 278 nm 303 nm 356 nm Tryptophan

A A F Tyrosine F Normalized Intensity (a.u.) Intensity Normalized Normalized Intensity (a.u.) Normalized Intensity

300 400 500 250 300 350 400 450 500 550 Wavelength (nm) Wavelength (nm)

Figure 4. UV Absorption and fluorescence of three common biological markers (tyrosine, tryptophan, NADH) illustrating the detector and emitter requirements.

A more civilian application of UV photodetectors is the monitoring of high voltage electrical transmission equipment. Due to exposure to sunlight, airborne pollution, and the

30 weather, the insulators on high voltage transmission equipment degrade with time. In the case of catastrophic failure the high voltage can ionize the air and cause a flash-over where the high voltage arcs across the insulator shorting the system and tripping protection fuses in the distribution network leading to outages effecting potentially thousands of homes and businesses.15 However prior to this catastrophic failure occurring, the partial ionization of the air around the insulator leads to the emission of high energy photons, many falling in the solar- blind region of the spectrum. Developing a solar-blind camera would allow for monitoring of this faint UV signature, even during board daylight. In addition, the solar-blind UV portion of the spectrum is ideal for early detection of this coronal discharge and can provide superior performance to thermal imaging.16

Another application of special civilian interest is UV astronomy. Many objects in the sky emit in the visible portion of the spectrum, and many important discoveries have resulted from observation at these wavelengths. However, young stars and stellar remnants (white dwarfs) tend to be much hotter emitting substantial quantity of their radiation in the ultraviolet portion of the spectrum. In addition, many of the important atomic resonance lines are in the

UV, or are Doppler shifted into the UV.17 This makes UV astronomy ideal for studying the origins and elemental makeup of the universe. Of particular interest are solar-blind imagers; while there is no ozone layer in space, objects studied in UV astronomy are often 4 to 8 orders of magnitude brighter in the visible than in the UV, and a high degree of visible-blindness is necessary to discriminate effectively.18

Probably the most promising industrial application for UV photodetectors would be UV flame detection, and combustion monitoring.19 Undetected burner extinction can have deadly consequences if the fuel flow is not stopped immediately and combustible gasses are allowed to

31 build up to unsafe levels. In the late 1950’s primitive visible-blind UV detection tubes were invented20, and industrial furnaces and boilers have been fitted with similar devices ever since.

These tubes rely upon the photoelectric effect to create the electrons necessary to ionize a gas and allow for conduction. They have the advantage of visible-blind operation and faster response than anything that had been invented at that time. However the spectral shape is not ideal for flame detection, and the devices are fragile glass tubes that can burn-out with time.21

In addition to monitoring the presence of known flames, UV flame detectors can be integrated into the fire suppression system as a way of reliably detection fire in environments not suitable to the use of traditional smoke and thermal detectors. UV fire detection is particularly well suited to the detection of alcohol or hydrogen fires where the actual flame can not be easily seen by the human eye.22 The development of a low cost reliable detector tuned to the detection and/or monitor of flames would be an ideal application for III-Nitride detectors.

2.3. UV Detection Technologies

For many decades, the detection of UV light has been accomplished using photomultiplier tubes (PMTs). These enjoy a high sensitivity to UV photons while being insensitive or “blind” to photons with wavelengths longer than the detector cutoff wavelength.

However, they are fragile vacuum tube devices that require bulky high-voltage power sources to operate. This inherent complexity also makes them relatively expensive. A solid-state alternative to PMTs is silicon-based photodetectors.23 However, Si-based devices are not as robust as AlGaN based photodetectors, and they have considerable sensitivity to photons in the visible and infrared spectral regions, in addition to the ultraviolet portion of the spectrum. The

32 out of band response is commonly addressed by the use of filters, such as a Woods glass optical filter. However, these filters increase the size and weight of the device, and reduce the overall quantum efficiency of the system. It is only in the second half of the 1990’s that wide bandgap

III-Nitride semiconductors, AlxGa1-xN in particular, have begun to emerge as the most promising material systems for such a device, thanks to their exceptional material properties.

III-Nitrides are uniquely suited to the detection of ultraviolet light. The AlGaN material system covers the bulk of the UV portion of the spectrum, allowing tunable cutoffs from 200 nm (6.2 eV) to 365nm (3.4 eV). The addition of indium increases the range even further allowing III-Nitrides to cover most of the UV, including the strategic solar-blind UV, and the entire visible portion of the spectrum, as shown below in Figure 5. No other semiconductor material system can offer such a wide tunable coverage of the ultraviolet and visible portion of the spectrum.

C 7 AlN Solar Blind 6 BN C 5

4 ZnS GaN ZnO MgSe SiC UV 3 ZnSe AlP CdS ZnTe GaP AlAs 2 CdSe AlSb CdTe IR GaAs InP

Bandgap Energy (eV) 1 InN Direct Bandgap GaSb Indirect Bandgap InAs InSb 0 3.0 3.5 4.0 4.5 5.0 5.5 6.0 6.5 7.0 Lattice Constant (A)

Figure 5. The quaternary AlInGaN material system is indicated in blue. For comparison other viable semiconductor materials are shown in black.

33 In addition to allowing tunable coverage of the UV and visible portions of the spectrum,

III-Nitride based detectors are able to do this with efficiencies that exceed those of other detector options. The theoretical quantum efficiencies for a number of different photocathode devices, and filtered CCDs are shown below in Figure 6 for comparison to AlGaN detectors. In can clearly be seen that AlGaN offers a significantly higher detective quantum efficiency in the longer wave portion of the spectrum, only being bested in the very short wavelengths approaching the soft x-rays.

AlGaN

Figure 6. Detective Quantum Efficiencies of materials for potential use in UV imaging applications. In the solar blind region of the spectrum AlGaN has a clear advantage over both photocathodes such as Cs2Te , and UV enhanced CCDs

The advantages of a III-Nitride based approach to solar blind UV detection are even more pronounced when one considers imaging applications, as summarized below in Figure 7. The additional size and complexity of a photocathode and micro-channel plate approach to UV

34 imaging (the 2D analog of a PMT) make these devices extremely expensive, bulky, and fragile.

Comparing flittered CCDs to AlGaN based detectors, the efficiency is almost ten times larger, making AlGaN devices much more sensitive.

Filtered Sillicon CCDs Photocathode & Microchannel plate

- UV enhanced Silicon CCD - Based on fragile vacuum tube - Solar blindness achieved via a technology. woods filter. - Requires bulky high voltage - Suffer from high out-of-band power supply to opperate. detection, - Intrinsically less efficient than - Intrinsically less efficient than AlGaN photodetectors at most AlGaN photodetectors wavelengths

Figure 7. Alternatives to AlGaN for solar blind imaging, and their disadvantages

2.4. Historic Development of III-Nitride Based UV

Photodetectors

The development of III-Nitride based ultraviolet photodetectors took off in the early

1990’s. The initial research work focused on GaN-based detectors, and was primarily an offshoot from the technological advances resulting from the drive to develop high quality GaN material for blue light emitting diodes and lasers.24 The few photodetectors that were demonstrated at the time were simple GaN photoconductors and Schottky photodiodes. During the second half of the 1990’s wide band gap GaN had established itself as a promising material for UV detectors, and soon AlxGa1-xN was being investigated for development of detectors operating over the entire 200~400 nm range. These earliest photodetectors were primarily photoconductors27,28, Schottky metal-semiconductor-metal detectors 35, 37, 36, 38 , 39, and Schottky barrier photodiodes37,36. By the late 1990’s research began to focus of the realization of p-i-n

35 photodiodes, 53 , 54 , 55 , 56 , 59 , 60, because of a number of intrinsic advantages. As an alternative approach to realizing UV p-i-n detectors, the negative electron affinity of GaN was used to make photocathode detectors.47,,48 49 . (A nice table comparing the performance characteristics of these early detectors can be found the last page of Ref 8).

More recently, research has been geared toward achieving shorter cut-off wavelength p- i-n photodetectors and has especially focused on developing detectors operating in the strategic solar-blind window 57, 58 , 61 , 62. This push for shorter wavelength has required the use of higher

Al-content compounds, and has introduced many technological challenges related to the growth and processing of this wide bandgap material. However, by the early 2000’s, AlxGa1-xN photodetectors had reached a certain level of maturity, and back-illuminated photodetectors117,122 were being investigated for potential solar- and visible-blind UV imaging applications142, 143 , 144 , 145 , 146 , 147 , 148 , 149 , 150.

The latest stage in III-Nitride based detector development has been the push for the realization of avalanche photodiodes (APDs). A number of groups reported Avalanche multiplication in GaN based devices in the late 90’s, but these early reports were attributed to noisy micro-plasma multiplication.65,66 Latter around 2000, groups discovered that they could suppress the micro-plasmas by adopting smaller device designs.67,,,,,,68 69 70 71 72 73 The most significant recent APD results have been the discovery of solar-Blind APDs in 2005.74,75

However GaN APD continue to be of interest in parallel with this.25,26

2.4.1. Photoconductors

Photoconductive detectors were the first III-Nitride based UV to be demonstrated, as they are generally the simplest detectors to grow and fabricate. They consist of a simple slab of

36 semiconductor material with two ohmic contacts. After GaN, the first AlxGa1-xN photoconductive detectors covering the complete range of Al concentrations (0≤x≤1) were rapidly reported.27

By growing single undoped epilayers with thicknesses 0.5~1.5 μm across the entire

AlGaN compositional range, UV photoconductors exhibiting sharp cut-off wavelengths from

365 to 200 nm have been realized as shown in Figure 8. These devices were grown by

MOCVD on basal plane sapphire substrates; the metal contacts were Ti/Au. The peak responsivity for x=0.34 (λcutoff ~ 285nm) is about 0.6 A/W. This was the first proof that AlxGa1-

28 xN materials were suitable for solar-blind detector applications.

AlxGa1-xN

x=0.0 (GaN) x=0.21 x=0.64

x=0.75 x=0.34

AlxGa1-xN

Photoresponse (a.u.) x=1.0 (AlN) x=0.47

(00•1) Al2O3

200 250 300 350 400 Wavelength (nm)

Figure 8. Photoconductors exhibitinbg sharp cut-off wavelengths covering the entire

AlxGa1-xN compositional range: from GaN (365nm) to AlN (200 nm). The inset shows a simplified photoconductor structure.

37 However recently interest in photoconductors has subsided due to the inability to fully resolve the issue of persistent photoconductivity. Persistent photoconductivity in III-Nitrides has been observed in both GaN and AlGaN based photodetectors based on un-doped, and both p-type and n-type photoconductors.29, 30 , 31 , 32 , 33 The increased conduction persists long after the photo-excitation is removed, and the effect can easily last for up to several hours. This makes the responsivity of the photoconductor a complex function of the time the sample has been illuminated, or kept in the dark.

Persistent photoconductivity is commonly attributed to defects and dislocations in the material. It is proposed that charges accumulate at the surfaces of the photoconductor and around bulk dislocations. These space charge regions then modulate the effective conduction cross section of the photoconductor. Both the persistent photoconductivity and the gain dependence on optical power have been modeled by considering this light induced band bending due related to defects.34

2.4.2. Schottky Metal-Semiconductor-Metal Detectors

Schottky metal-semiconductor-metal (MSM) photodetectors are also relatively simple photodetectors to realize; III-Nitride based devices were first realized soon after photoconductors. They generally consist of a single epitaxial layer with two interdigitated

Schottky metal contacts deposited on the surface; this creates two back-to-back rectifying junctions. Electron-hole pairs are generated when photons are absorbed near the depletion regions formed at these Schottky junctions, and this lead to a photo-response. The geometry of a typical interdigitated finger device is shown in Figure 9 below.

38

Figure 9. Shows the geometry of a typical interdigitated finger MSM device with a length of 150 μm a finger width of 2 μm and a pitch of 10 μm.

In the case of GaN and AlxGa1-xN materials, these devices are arguably simpler to fabricate because there is no need to highly dope the material and to achieve ohmic contacts.

MSM detectors exhibit all of the desirable attributes of a practical photodetector, such as: high gain, low dark current, high speed, large bandwidth and high sensitivity. However, these devices require an applied bias to operate, and their performance characteristics are dependant on this applied bias, since it changes the volume of the depletion region.

GaN based Schottky MSM devices have been extensively studied.35,36 A typical spectral response from such a detector is shown in Figure 10. A three order of magnitude sharp cut-off at the band-edge of GaN is observed. The high responsivity obtained is indicative of the presence of internal gain in these devices (similar to photoconductors). There have also been reports of detectors with different layer thicknesses and finger geometry which exhibited external quantum efficiencies of up to about 50 % with applied bias in the range of 5 to 20 V

39 and without internal gain.37 These devices also show very low dark currents, as low as 800 fA at a bias of 10 V.

100

10

1

Responsivity (A/W) 0.1 200 300 400 500 600

Wavelength (nm)

Figure 10. Shows a typical spectral response of a Schottky based MSM photodetector.

Latter research has focused more on realizing shorter wavelengths, in particular solar- blind, MSM detectors. For example, using an Al0.4Ga0.6N epilayer on sapphire, an external quantum efficiency as high as 49 % (responsivity 107 mA/W) at a 90 V bias has been reported for an MSM detector with peak responsivity wavelength of 272 nm.38

Most recently, back-illuminated solar-blind MSM detectors have been investigated in an effort to move the technology toward focal plane arrays in which the epilayer (front) side of the device needs to be connected to the readout circuitry.39 In the case of these MSM detectors, back-illumination has the advantage of avoiding the blockage of incident photons by the interdigitated metal contacts, thus enhancing the quantum efficiency. However, because the

40 depletion region (the active region) of the device is located at the epilayer/metal interface

(front), the incident photons have to first traverse the substrate and most of the AlxGa1-xN epilayer before reaching the active region. This can be partially circumvented by utilizing a heterostructure in which a larger bandgap epilayer (e.g. AlxGa1-xN) is first grown on the substrate before the active AlyGa1-yN layer with x>y.

Schottky MSM detectors hold some promise for the realization of commercial solar- blind detectors. However these devices require application of a bias, which can be significant when high concentrations of Al are used in solar-blind devices. For this reason and others, fabrication of focal plane arrays seems to have exclusively favored the use of p-i-n based structures over these MSM devices.

2.4.3. Schottky Barrier Photodiodes

Schottky barrier photodiodes have received much less attention than other types of

40 photodetectors , solar-blind AlxGa1-xN Shottky photodiodes weren’t demonstrated until the year 200041, mainly because of the very rapid interest and progress in Schottky metal- semiconductor-metal detectors. Schottky barrier photodetectors consist of a layer of semiconductor with two different contacts, one ohmic and one rectifying. Electron-hole pairs are generated when photons are absorbed near the depletion region formed at the Schottky junction. This leads to a photo voltage developing across the two contacts.

The typical spectral response of a Schottky barrier photodetector is shown in Figure 11, with the corresponding device structure shown in the inset. A peak responsivity of 70 mA/W was realized at 272 nm, without applied bias, corresponding to an external quantum efficiency

41 of 32 %. Schottky barrier detectors exhibit good solar-blindness: this example shows a four order of magnitude rejection ratio between the peak response and visible wavelengths.

Spectral Responsivity

-1 Ni/Au 10 3000 Å AlGaN

1 μ m AlGaN:Si -2 Ti/Au 10 5000 Å GaN:Si 2 μ m GaN:Si -3 10 Sapphire Device Structure 10-4 10-5 Responsivity (A/W) 10-6 200 300 400 500 600 Wavelength (nm)

Figure 11. Shows the typical spectral response of a Schottky barrier photodetector. The corresponding device structure is shown in the inset.

2.4.4. Photocathodes

A less traditional approach to realizing UV detectors that has been studied in parallel with the other approaches it the fabrication of photocathodes. Photocathodes rely on the negative electron affinity of GaN42,43 to convert photons into photo-electrons that can then be directly read out, or amplified with an electron multiplication cascade. For imaging applications this would consist of the photocathode and an electron bombardment CCD for direct readout or the insertion of a micro-channel plate for electron signal amplification before readout.

42 AlN is the best candidate for fabricating a photocathode due to its low electron affinity of around 2 eV,44 None of the III-Nitrides natively have a negative electron affinity.45 However, it was discovered that by activating the surface with cesium or oxygen it is possible to reduce the electron affinity by about 2.6 eV giving AlN a true negative electron affinity, and GaN an effective negative electron affinity. 42 ,45 ,46 Despite the advantages of AlN based devices, the first reported III-Nitride photocathode detectors were cesiated p-type GaN-based devices reported around 2002 with a QE of around 30%.47,48 Since then AlGaN and InGaN based devices have been demonstrated with quantum efficiencies in excess of 40%,49 and GaN based devices with quantum efficiencies as high at 71% have been demonstrated.50,51 However photocathodes are analogous to PMTs and thus share many of their disadvantages, such as requiring a vacuum tube and a large bias to extract the photo-electrons and collect the signal.

2.4.5. Front Illuminated p-i-n Photodiodes

Most of the recent research work on UV and especially solar-blind detectors has been focused on realizing p-i-n photodiodes: in fact the rest of the discussions in this work will focus exclusively on p-i-n based photodetectors and focal plane arrays. Figure 12 shows the typical structure of a top illuminated AlxGa1-xN photodiode. Almost all p-i-n photodiodes are oriented such that the n-type layer is closer to the substrate; mainly because the quality of n-type AlxGa1- xN is typically far superior to that of p-type AlxGa1-xN.

43 Ni/Au

2000 Å AlGaN:Mg p-contact 2000 Å AlGaN 1 μm AlGaN:Si Ti/Au

2.5 μm GaN:Si

n-contact Sapphire

Side view Top view

Figure 12. Shows the typical structure of a front illuminated AlxGa1-xN based p-i-n photodiode.

The interest in p-i-n photodiode UV detectors is driven by their intrinsic advantages: (i) a very low dark current due to the large potential barrier, (ii) a high speed of operation, (iii) a high impedance suitable for FPA readout circuitry, (iv) a direct control of the quantum efficiency and speed through the control thickness of the intrinsic layer, and (v) the device can operate under low or zero applied bias. There are two modes of operation for photodiodes: photovoltaic (operation under no bias) and photoconductive (operation under reverse bias).

The mode of operation is chosen based on the desired application; in the photovoltaic mode, the dark current is at the lowest, while in the photoconductive mode the device exhibits a faster response.

A wide range of AlxGa1-xN p-i-n photodiodes have been demonstrated, most of which are front-illuminated.52,,,,,,,53 54 55 56 57 58 59 Devices, which are going to be operated only under front illumination, are generally designed as an AlGaN p-i-n structure on a rather thick (>2 μm)

GaN template layer, on sapphire substrate. The use of this thick GaN layer helps partially isolate the active region from defects that arise at the substrate/epilayer interface. The p-i-n

44 structure can either be a homostructure or a heterostructure. A thin (50~100 Å) p-type GaN layer is sometimes incorporated above the p-i-n structure in order to increase the collection of photo-generated carriers, and improve the contact resistance: this is because p-type GaN is easier to dope than p-type AlGaN, and thus is better suited to lateral current collection. p-GaN also has the advantage of being transparent, whereas a metal contact would block the incident radiation.

The spectral response of a set of front illuminated AlxGa1-xN p-i-n photodiodes with different Al compositions is shown in Figure 13. By tailoring the Al content in the active region from 0 to 70%, the peak responsivity wavelength can be tuned from 364 to 232 nm.

This shows that both solar-blind and visible-blind p-i-n photodiodes can successfully be achieved using AlxGa1-xN materials. Due to the quality of the thick GaN templates, and careful device optimization, the external quantum efficiency of these devices approaches the theoretical maximum for a p-i-n detector.

45

-1 x=0 10 QE=1 x=0.45 10-2 x=0.30 x=0.70 x=0.18 10-3 x=0.15 x=0.05 10-4

Responsivity (A/W) 10-5

10-6 200 300 400 500 600 Wavelength (nm)

Figure 13. The spectral response of a set of front illuminated AlxGa1-xN p-i-n photodiodes with various Al compositions. The line at the top indicates the theoretical maximum for a p-i-n photodetector corresponding to a quantum efficiency of one.

2.4.6. Back Illuminated p-i-n Photodiodes

The concept of back-illuminated p-i-n devices had first been reported in this material system in 1998 for GaN photodiodes,60 for which the purpose then was to enhance the quantum efficiency by avoiding absorption of photons in the top p-type GaN layer. To do so, an n-type

Al0.28Ga0.72N layer was first grown on the sapphire substrate, before depositing the i- and p-

46 type GaN layers. Because this ternary layer has a larger bandgap energy then GaN, it is transparent to the photon wavelengths that were being detected.

Since then, the growing desire for solar-blind focal plane arrays made from AlxGa1-xN materials has driven the development of back-illuminated p-i-n photodiodes 52, 61 , 62 , 63.

Although sapphire is transparent across the entire AlGaN compositional range, no thick GaN layers can be used prior to growing the p-i-n structure because any low Al composition layers will strongly absorb the incident light if their bandgap energy is smaller than that of the photodetector active region. Building off of the first back-illuminated heterostructure GaN p-i- n photodiodes, AlGaN based devices use a similar p-i-n heterostructure, as shown in Figure 14; the bottom layers have a higher Al content and thus larger bandgap than the intrinsic and p-type

AlGaN layers. Indeed, this way, the photons of interest with a wavelength near the desired peak responsivity wavelength are able to reach the depletion region with minimal absorption by the thick bottom n-type AlGaN layer. Such a p-i-n structure therefore acts as a band-pass filter.

Ni /Au

1000 Å AlGaN:Mg

2000 Å AlGaN Ti /Au : AlGaN:Si

Sapphire

Figure 14. The basic structure of a back illuminated p-i-n photodetector

47 Once back illuminated AlGaN photodetectors started to become established, research began to concentrate on the development of back illuminated photodetectors operation in the solar-blind region. For solar-blind photodetectors, the typical Al concentrations in the bottom n-type are in the range 50~60 %, while they are 37~40 % in the i- and p-type layers. A typical spectral response curve for an AlGaN solar-blind detector designed for both front and back- illumination is shown in Figure 15; in this figure the band-pass behavior of the device can be clearly observed when operated under back-illumination External quantum efficiencies of 68

% (responsivity 135 mA/W) have been reported for unbiased photodiodes exhibiting a peak responsivity at 280 nm. Dark currents as low as 5 nA/cm2 at a bias of –10 V and a detectivity between 4.2x1011 and 5.3x1013 cmHz½/W at 0 V have also been reported .

Figure 15. Typical photoresponse of a back-illuminated solar blind photodetector showing the difference between the front- and back-illuminated responses.

48 2.4.7. Avalanche photodiodes

III-Nitride photodetectors have demonstrated high efficiency detection of ultraviolet light covering the spectral range from 365 to 200 nm. However the detectivity obtained by these detectors cannot yet rival the noise performance of commercial photo-multiplier tubes

(PMTs). PMTs are glass vacuum tube devices that use a photocathode to generate free electrons with are multiplied by cascade bombardment of a series of dynodes, typically obtaining internal gain on the order of one-million. This low noise internal gain is responsible for giving PMTs a high signal to noise ratio. For III-Nitride based devices to effectively compete in applications that require this sensitivity, it is necessary to improve the signal to noise ratio. One route to achieving this is the develop avalanche photodiodes (APDs). In semiconductors it is possible to obtain internal gain by taking advantage of avalanche multiplication under high electric fields. Unlike photoconductive gain earlier reported in

AlGaN based devices,64 avalanche gain is in principle capable of lower noise and faster response times thus increasing the sensitivity of these photodetectors.

There have been a few early reports of breakdown in GaN based devices all the way back in 1996. 65,66 However the breakdown of these early devices was attributed to micro- plasma breakdown. Micro-plasma breakdown occurs due to ionization at local concentrations of the electric field, such as occur at defects (i.e. the threading dislocations typical of GaN).

The name arises from the characteristic defect luminescence that can be seen to arise at these localized breakdown regions, as shown in Figure 16 below.

49

Figure 16. Mosaic image of micro-plasma luminescence from an un-optimized array of

16 diodes showing the luminescent nature of the process, and the distribution of defects.

Micro-plasma breakdown is associated with much longer time constants than traditional breakdown. In addition the micro-plasma mechanism is associated with a much higher excess noise, making it highly undesirable. This was an early discouragement to the pursuit of GaN

APDs, but soon groups began reporting micro-plasma free avalanche multiplication. However it was only by drastically shrinking the device size (typically to diode less than 40 μm in diameter), that micro-plasma formation was able to be suppressed. 67,,68 69 It was found that the critical mesa diameter for micro-plasma free multiplication was 50 μm. These devices were only operated in linear mode, and typically exhibited small gains in the range of only 10 to 20, and had breakdown critical field strengths ranging from 3 to 4 MV/cm-1.

The next year reports of Geiger mode operation of GaN APDs were published. 70

Geiger mode is favored over linear mode in low-light situation for its photon counting

50 capabilities. In addition, the statistical nature of photon counting makes the devices relatively insensitive to excess multiplication noise, and dependant only upon the dark count rate and the detective quantum efficiency.

20 18 16 1500 1E-5 1000

A) 14

μ 1E-6 100

12 Gain

1E-7 10 1000 Gain 10

1E-8 1 8 (A) PhotoCurrent 6 1E-9 0.1 0 -10 -20 -30 -40 -50 -60 -70 -80 500 Reverse Bias (V) 4 PhotoCurrent ( 2

0 1020304050607080 Reverse Bias (V)

Figure 17. GaN based APD showing an avalanche multiplication of nearly 1500 at a breakdown field strength of 2.7 MV/cm.

It in only very recently, that interest in GaN APDs has really resumed,71,72 with one group reporting an impressive gain of 1500 at being reported at a breakdown field strength of

2.7 MV/cm-1, as shown in Figure 17 above.73 Recently the first solar-blind AlGaN-based APD was also demonstrated74, with the first Schottky-barrier APD being demonstrated shortly their- after.75

51

Figure 18. AlGaN-Based Solar-Blind APD showing a maximum gain of 2000 at an electric field of 7 MV/cm.

52 3. Important UV Photodetector Characteristics

3.1. General Photodetector Parameters

One of the most important photodetector parameters is responsivity; it’s a measure of the opto-electric conversion of a photodetector. There are two types of responsivity: current responsivity and voltage responsivity, however since all the devices discussed in this word are photodiodes, we will only consider current responsivity. Responsivity (ℜ) is defined as the photocurrent (in amps), per unit of incident optical power (in watts). It depends upon the number of electron hole pairs generated per incident photon (the quantum efficiency (n)), the wavelength of the incident photons, and the gain, if any, of the photodetector (g). The responsivity is thus given by the following equation:

λη =ℜ qg , (1) hc where (h) is Planck’s constant, (c) is the speed of light, and (q) is the electron charge. The quantum efficiency (η) in the above equation is the external quantum efficiency. It is also possible to estimate an internal quantum efficiency (ηi) for the active region by removing the photon losses associated with reflection and incomplete absorption within the photodetector:

−αd ηη i −−= eR )1)(1( . (2)

Where in the above equation (R) is the optical reflectivity, (α) is the absorption coefficient of the photodetector’s active region, and (d) is the effective thickness of the absorbing region.

This formula can be modified similarly to account for absorption outside of the active region, such as in the underlying template layers.

53 Noise equivalent power (NEP) is another parameter commonly used to characterize photodetectors. Essentially, the NEP represents the minimum detectable signal; it is defined as the incident light power necessary to obtain a unity signal to noise ratio in the photodetector signal, and is given by

i 2 n (3) NEP = , ℜ

where (ℜ) is the responsivity of the photodetector, and (in) is the total noise current. The noise current and its constituent elements are discussed in more detail in section 3.2.

NEP is also used in the calculation of detectivity (D*), another common figure of merit for many different types of photodetectors. It characterizes the signal to noise performance of the device, and is defined as:

ℜ opt BWA opt BWA D* = = , (4) 2 NEP in

where (Aopt) is the optical area of the detector. The detectivity of infrared photodetectors is usually limited not by the detector, but rather by the presence of naturally occurring infrared background radiation. However, in the case of UV, the background falls off very quickly with decreasing wavelength, and is essentially non-existent at wavelengths below 280 nm; III-

Nitride UV photodetectors are usually limited by internally generated noise.76

For a more detailed treatment of these basic photodetector parameters see chapter 13 of reference 77, or chapter 14 of reference 78.

54 3.2. Basic Noise Analysis Theory

The noise response of a detector is an important parameter to understand in more detail because it ultimately limits the maximum detectivity. Understanding the basic concept of noise and the common origins and types of noise can be useful in the characterization and development of high quality UV photodetectors; although typically, devices made from wide band gap material are expected to have a low noise level. This section provides a general overview of noise analysis theory.

The total noise power generated in a photodetector is given by:

2 = nn RiP L , (5)

where (RL) is the load resistance and (in) is the noise current. The total noise current representation consists of contributions from many noise sources, but this can be simplified into the three most dominate sources of noise currents are 1/f noise (jitter noise), shot noise

(generation-recombination noise), and Johnson noise (thermal noise):

2 2 2 2 n f1 shot ++= iiii J . (6)

Analysis of the noise power spectrum combined with a basic understanding of the three dominate noise sources can be useful in understanding the operation of UV photodetectors.

Noise analysis can often identify the limiting factor for the operation of a particular detector.

For 1/f noise, the noise power is calculated as the area under the noise power spectral density curve in the bandwidth range of interest, (BW). The 1/f noise contribution is represented as follows:

55 B 1 B 2 f1 = n = n + ∫∫∫ n df)f(Sdf)f(Sdf)f(Si 0 0 1 (7) B 1 SS =+= BWSdf + )1(ln 00 ∫ 0 , 1 f although it is also common to just neglect the noise frequencies below 1 Hz and measure the area under the noise power spectral density graph directly. The 1/f noise has a strong frequency dependence, and tends to dominates at low frequency (less than 1 kHz) in Nitride based devices.

At intermediate frequencies shot noise generally dominates. The shot noise can be estimated by:

2 shot = 2 d BWeIi , (8) where (e) is the electron charge. This is also called generation-recombination noise because it describes the effects of generation and recombination centers created by the impurities or lattice defects in the material. The rates depend on the individual nature of the center, its predominant state of charge carriers, and the position of the level within the band gap.

The spectral density of Johnson noise is constant; it is essentially white noise. It is associated with the finite resistance of the device, (R), and is due to the random thermal motion of charge carriers in the semiconductor crystal, not to be confused with the fluctuation of the total number of charge carriers. The Johnson noise current can be written as

56

2 4kTBW iJ = , (9) Req

where (Req) is the parallel combination of the junction resistance, the external load resistance, and the input resistance of the amplifier. Either Johnson or shot noise is usually dominant at high frequencies due to the inversely proportional qualities of the 1/f to frequency

The noise in the GaN based detectors is traditionally lower than commonly found in other III-V semiconductors due to its significantly higher resistance and wider bandgap. In fact, the noise of these devices can be extremely difficult to measure. The shot noise can be measured directly from the I-V curve, but the Johnson noise can be difficult to measure as the device impedance tends to approaches that of the instrumentation used to measure it, and the

1/f noise cannot be directly measured except at large biases and the actual photovoltaic (zero bias) 1/f noise has to be extrapolated from a series of higher voltage measurements.

10-1 10-3 10-5 10-7 10-9 Dark Current (A) 10-11 10-13

-10 -5 0 5 Bias Voltage (V)

Figure 19. The dark current of GaN p-i-n photodetector at different applied biases.

57 It is possible that the device is limited by shot noise. Shot noise can be calculated from the dark current: the I-V curve, in the absence of illumination, for a typical GaN p-i-n device is shown in Figure 19. Based on this IV curve, the expected shot noise values of the device at various reverse bias levels are calculated in Table 1.

Bias Dark Current Shot noise Voltage (V) (nA) level (A2/Hz) -5 1.13 3.6×10-28 -6 2.63 8.4×10-28 -7 6.15 2.0×10-27 -8 10.57 3.4×10-27 -9 20.94 6.7×10-27 -10 40.00 1.3×10-26 Table 1. Dark current and shot noise levels for various reverse bias voltage values.

The noise power density of a typical GaN p-i-n detector, at different voltage biases, is measured using a FFT Spectral Analyzer; the data is shown in Figure 20. However the unbiased noise cannot be directly measured using this setup. More details of the actual noise measurement setup and its limitations are given in Appendix 2: Device Testing and

Characterization.

58 -5 V

-18 -6 V /Hz)

2 10 -7 V -8 V 10-19 -9 V -10 V 10-20

10-21

10-22

-23 Noise Power Density (A 10

10 100 Frequency (Hz)

Figure 20. The noise power spectral density of a GaN p-i-n photodetector.

In this case the noise power spectral density Sn, satisfies the following empirical relationship:

I = SS d 2 (10) n 0 f γ (A /Hz),

where (Id) is the dark current, (f) is the frequency, and (S0) and (γ) are fitting parameters. The best fits usually occur with γ∼1, and the noise parameter S0 should be independent of biasing.

Using this equation, the noise power spectral density shown in Figure 20 is fit for each of the different biases. These fitting parameters can then be extrapolated back to zero bias to extract a photovoltaic 1/f noise.

By comparison of the 1/f and shot noise, the dominate source of detector noise can be determined. The values for shot noise in the table are much less than the minimum 1/f noise shown in the graph, even after extrapolation back to zero. Thus we can safely neglect the shot

59 noise contribution to the noise in these devices. Johnson noise can also be ruled out as an important noise contribution due the fact that Johnson noise is characterized as white noise, however the spectral power density shows a strong frequency dependence (this would not necessarily be the case if we were interested in high-frequency operation). Based upon analysis do the noise spectrum we can determine that 1/f noise dominates in GaN devices; this is the noise behavior would have expected for a typical III-Nitride detector operating at low frequencies.

3.3. Noise Analysis in AlGaN p-i-n Photodiodes

The flicker (1/f) noise is expected to dominate at low to intermediate frequencies, however in AlGaN based devices the 1/f noise tends to be so small that it cannot be measured even under reverse biases of up to 10 volts. Usually shot would be expected to dominate at moderate and higher frequencies; however with small area AlGaN photodiodes, as in the case of a FPAs and APDs, the dark current can be so small that shot noise is still insignificant. In high resistance, high Al composition, small area, photovoltaic devices with low leakage currents Johnson noise then becomes the significant source of noise. With high Al composition photodetectors (specifically solar-blind detectors) the contribution from thermal noise (Johnson noise) easily becomes larger than the background radiation. This means that the detectivity of the photodetector is thermal limited. In this case the detectivity can be expressed as follows:

60 AR D* ℜ= 0 (11) T kT ,

where (R0) is the resistance at zero-bias, (A) is the area of the photodetector, and (ℜ) is the responsivity. This establishes a maximum detectivity if the photodetector frequency and leakage current are such that neither 1/f noise nor shot noise are significant.77, 79

In this case the detectivity can be estimated by measuring the photodetector’s responsivity and zero-bias resistance, the temperature and area should already be known.

Measurement of the responsivity is in principle straightforward; however III-Nitride photodetector resistances are usually so high that they cannot be directly measured. This again introduces the need for extrapolation in order to determine the detectivity. A method of extrapolating the zero-bias resistance has been proposed by Campbell et. al..80

After removal of any instrument induced offsets exponential fits made to both the forward and reverse bias curves. The forward and reverse bias fits can then be combined to form a single equation for the low current behavior of the diode:

∗Vb ∗Vd eceaI −+−= )1()1( , (12) where a and b, and c and d are the exponential fitting parameters from the reverse and the forward bias fits, respectively. The value of Ro can then be estimated by taking the inverse of the derivative at zero:

61 1 dI 1 = , R0 = . (13) R0 dV V =0 + cdab

This method of exponential fitting provides a good estimate of the zero-bias resistance of an AlGaN photodetector when direct measurement is precluded by instrument noise. The stability of this fit is also found to be better than polynomial fits.80

3.4. Avalanche Photodiode Parameters

The operation of avalanche devices is similar to those of common detectors prior to the onset of breakdown. However, at breakdown avalanche multiplication occurs, and the devices behave sufficiently different, and require characterization by a different set of parameters.

The breakdown voltage (Vbr) is defined as the voltage at which the photocurrent begins

to increase sharply with increasing reverse bias; looking at the I-V curve, it resembles the

forward bias turn-on, but typically occurs at voltages around 100 V. It is characterized by an

apparent increase in the quantum efficiency of the device above unity, and hence a term for

gain (g) is introduced as shown in the quantum efficiency calculation in the previous section.

This accounts for the multiplication of carriers within the device.

In APDs the gain (g) is usually replaced with the symbol (M) to stand for the avalanche multiplication to reinforce the origins of the gain. The avalanche gain (M), defined as the

difference between the primary multiplied photocurrent and the multiplied dark current,

normalized by the difference between the primary unmultiplied photocurrent and the

unmultiplied dark current:

62

[]I-I darkdIlluminate multiplied M = Equation 14 []I-I darkdIlluminate unmultiplied

This equation paradoxically requires measuring both the multiplied and unmultiplied currents

simultaneously. In practice the unmultiplied currents are assumed not to vary as a function of

reverse bias. This equation can be further simplified by replacing Iilluminated – Idark with

Iphotocurrent. The photocurrent can be measured by subtracting alternate measurements of the

device in light and dark conditions. The data can then be normalized to the photocurrent just

before breakdown (unmultiplied photocurrent) to obtain multiplication (M) versus reverse bias.

The avalanche gain is primarily dependant upon two material parameters: the electron

and hole impact ionization coefficients αn and βp. The gain can thus be split into electron and

hole terms (Mn and Mp, respectively) and represented in term of the ionization coefficients as:

1 M = , n w ⎡ x ⎤ − αn −− βα pn ')(exp1 dxdx 0 ⎣⎢ ∫∫ 0 ⎦⎥ (15) ⎡ w ⎤ −− βα pn )(exp dx ⎢ ∫0 ⎥ M = ⎣ ⎦ p w ⎡ x ⎤ − αn −− βα pn ')(exp1 dxdx 0 ⎣⎢ ∫∫ 0 ⎦⎥

Where the assumption is made that electrons are injected into the device at x=0 and

holes at x=w, with w being the width of the multiplication and x being the position within the

multiplication region. Unfortunately electron hole pairs are generated throughout the device,

and both αn and βp vary as a function of the electric field which is not constant across the

device, and varies with applied bias making analysis difficult. In general this simplified by

assuming a uniform electric field and constant αn and βp to become:

63

[ − β αnp ] {αnW [ − β αnp )/(1exp)/(1 ]} M n = , (16) − np {}nW []− αβααβ np )/(1exp)/(1

Avalanche multiplication is not inherently completely noise-free. The output signal of

the APD consists of the multiplied signal, the multiplied dark current, the un-multiplied dark

current, the detection noise, and the excess multiplication noise, as shown below in Figure 21

below. Avalanche multiplication will multiply the dark current and any noise in the signal,

however multiplication is a statistical process and depending upon the ratio of the electron and

hole ionization coefficients, there will be an additional amount of noise introduced directly by

the statistical nature of the multiplication process

Figure 21. Signal, noise, and dark current contributions to the output of an APD.

This makes the excess noise factor (F) a parameter of interest. For the worst case

scenario where αn equals βp, the excess noise is just the average multiplication and thus we have F = M. In the case where the ratio βp/αn is not unity the excess multiplication noise factor

can be given as:

64 ⎧⎛ ⎛ β ⎞⎞ ⎫ ⎪⎜ p ⎟ 2 ⎪ MF nn ⎨ ⎜ −−= ⎟ []n − /)1(11 MM n ⎬, ⎜ ⎜ α ⎟⎟ ⎩⎪⎝ ⎝ n ⎠⎠ ⎭⎪ (17)

⎧⎛ ⎛ α ⎞⎞ ⎫ ⎪⎜ ⎜ n ⎟⎟ 2 ⎪ MF pp ⎨ −−= []p − /)1(11 MM p ⎬ . ⎜ ⎜ β ⎟⎟ ⎩⎪⎝ ⎝ p ⎠⎠ ⎭⎪

It can be seen that the excess multiplication noise factor depends upon the multiplication, but is also strongly dependant upon the ratio of the ionization coefficients, which makes study of these coefficients a matter of great interest.

The ionization coefficients can be experimentally extracted from measurements of the gain in front and back illumination. To do this we make the assumption that for a given p-i-n structure, for top illumination we inject primarily electrons into the active region, and that for back illumination we inject primarily holes. This allows us to use the difference in the multiplication factor observed for these two illumination geometries, to approximate the ionization coefficients using the following equation:

1 ⎛⎞M ()1VM− ⎛⎞ ()V β ()EL= ppn p ⎜⎟⎜⎟ WMVpn()− MV ()⎝⎠ MV n () ⎝⎠ (18) . 1 ⎛⎞M ()1VM− ⎛ ()V⎞ α ()EL= nnn n ⎜⎟⎜⎟ WMV⎝⎠np()− MV ()⎝ MV p ()⎠

An excellent reference for a more detailed overview of avalanche photodiodes parameters in G. Stillman and C. Wolfe’s chapter “Avalanche Photodiodes” in volume 12 of

Semiconductor and semimetals.81

65 3.5. Noise Analysis in Avalanche Photodiodes

The main justification for adopting an avalanche photodiode design is to improve the noise performance of the detector system. Avalanche multiplication has the potential to offer low noise amplification of the detected signal, better than would be possible with external electronics, and thus improving the overall detective performance of the UV detection system.

From the standpoint of solely the detector, the shot noise can be expected to increase significantly as the larger reverse biases involved induce significant leakage current through the device. The excess noise factor also introduces significant noise to the APD operation.

However the internal gain can offset these effects in cases where to obtain the same amplification level using external circuitry would require a high gain amplifier that would use a large feedback resistor, and introduce significant thermal noise. Taking advantage of internal avalanche multiplication also has the advantage of not decreasing the bandwidth of the system the same way that a larger feedback resistor would with an external amplifier. The end result of this is that in certain situation where a low light level signal is on interest, an APD with its internal avalanche multiplication can provide an improved signal to noise ratio while maintaining a wide operation bandwidth.

The total dark current in the APD results from an unmultiplied leakage current and from

a multiplied current. The unmultiplied dark current (ID-unmultiplied) is primarily due to surface

leakage, and other parallel leakage paths that do not directly transverse the multiplication

region of the device. The rest of the dark current (ID-multiplied) is subjected to multiplication in

the device, and thus the actual total dark current of the device depends upon the multiplication

factor (M) as follows:

66 (19) = II DD −ummultipli + I Ded −multiplied ⋅ M .

In addition, this current is multiplied by the excess noise factor (F). This yields a new equation for the shot noise in the case of an APD operating in the dark with a gain M and bandwidth

BW:

2 2 (20) shot = (2 Iqi D−ummultipli I Ded −multiplied ⋅+ )BWFM .

However at low frequencies it is still expected that 1/f noise will dominate the operation of the APDs. This has been experimentally confirmed for a number of devices. The power spectral density is shown below in Figure 22 for a GaN APD around the breakdown voltage. It shows that 1/f noise dominates below 300 Hz, but that beyond 300 Hz a strong white noise component can be seen in the signal.

1E-16 91V-102V / 1V step /Hz) 2 1E-17 1E-18 1E-19 1E-20 1E-21

1E-22 1E-23 1E-24 1E-25 1E-26 Spectral Power Density (A Power Density Spectral 100 1000 10000 Frequency (Hz)

Figure 22. Noise power spectral density versus bias around breakdown for a GaN APD.

67 By investigating this white noise region above 300 Hz it is possible to approximate the values of the noise as a function of reverse bias. Correlating this with the multiplication factor as a function of bias it is possible to come up with an experimental determination of the excess noise factor by rearranging equation (20) above to become:

S F = n (21) 2qIM 2 .

In the case where the APD is operated in Geiger mode in a photon counting setup,

concerns such as the excess multiplication noise, and the shot noise of the detection system

become somewhat irrelevant. In Geiger mode the APD is biased well beyond breakdown, the

arrival of a photon(s) generates a free carrier that initiates the avalanche process. This results

in a large current pulse as the device undergoes breakdown. A quenching circuit is then used to

either actively or passively quench this process. In this case the detected signal is represented

as a series of pulses, and the noise of each individual pulse is irrelevant, and only the fact that a

pulse was counted is important. This makes photo detection ideally a binary counting process.

However in a real application pulse height discrimination is used to set a minimum pulse height

threshold to reduce dark counts and to help in correctly counting pulses where two photons

arrive at the same time. This procedure is limited to only very low fluxes of light where the

arrival rate of the photons is significantly longer than the reset time of the quenching circuit.

The other significant limiting factor is the dark current: the dark current must be significantly

low that the number of spontaneously generated pulses remains much smaller than the photon

generated pulses.

In Geiger mode APDs a new term, the dark-count rate (DRC) is introduced. This is

coupled with the definition of the photon detection efficiency (DE%) to give the ability of the

68 detector to detect photons; the Photon detection efficiency is defined in term of the incident photon rate in photons per second (PPS) and the resulting count rate in counts per second (CPS) as:

DE (% −= PPSDCRCPS × %100/) . (22)

An excellent reference for noise characterization of APDs that also goes on to discuss

the noise equivalent power (NEP), and noise equivalent input (NEI) is a white paper of APD

noise released by Voxtel Inc., an electro optical system manufacturer specializing in photon

counters.82

69 4. Wide Band-Gap III-Nitride Material Growth

4.1. Introduction

All of the growth in this work is carried out in a horizontal flow low-pressure metal-

organic chemical vapor deposition (MOCVD) reactor. The reactor is an AIXTRON 200/4-HT

capable of growing at temperatures in excess of 1350 °C, and pressures as low as 10 mBar.

Double side polished basal plane (00.1) sapphire is used as the substrate for all of the growth.

However, because sapphire is not lattice matched to AlN or GaN, it is necessary to use a buffer

layer for nucleation of the growth; this is typically a thin 200 Å low-temperature AlN or GaN

layer.

For top illuminated devices a thick GaN is used as a template layer. However for the

bulk of the back-illuminated photodetectors, FPAs, and APDs discussed in this work, an AlN

or an AlN / {AlN /AlGaN} SL layer is grown using atomic layer epitaxy, and serves as a

transparent template in conjunction with the use of a transparent AlN buffer. The p-i-n

photodetector or avalanche photodiode structure then requires growth of n-, i-, and p-type

(Al)GaN to define the active region. Structures are then capped with a thin highly-doped p-

GaN layer to aid in the formation of ohmic contacts.

4.2. Metal Organic Chemical Vapor Deposition, an Overview

Metal-organic chemical vapor deposition (MOCVD) is used to grow the III-Nitride material within this work. MOCVD is an epitaxial deposition technique that uses both gas sources and organo-metallic sources to grow thin films. The reactor used in the context of this

70 work is a commercial Aixtron 200/4, low-pressure, horizontal flow MOCVD reactor. It was

originally installed in 1994, and at the time was the first Aixtron reactor used for the growth of

III-Nitrides. Part of this work has not only been developing the material, but also working to

develop the reactor itself to support the growth of high quality AlGaN material. The MOCVD

reactor consists four major parts: a gas mixing cabinet, a high-temperature reactor chamber, a

glove box for oxygen-free sample handling, and a sophisticated computer control system. The glove box, reactor chamber, and computer control electronics are shown below in Figure 23.

Figure 23. Aixtron AIX 200/4 MOCVD reactor used for growth of III-Nitride material

discussed in this work.

The gas mixing cabinet handles the sources necessary for the MOCVD growth. Growing

AlGa(In)N requires aluminum, gallium, and nitrogen be fed into the reactor in controlled ratios.

In addition to these basic three sources indium is sometimes added to achieve improved

material quality and enhanced electrical conduction in the material. The doping of the material

is achieved using silicon and magnesium as the n-type and p-type dopants, respectively. The

various elements used in the growth and the sources from which they are derived are shown in

Table 2 below. These sources are divided into the two groups, the hydrides and the organo-

71 metallics: the hydrides include the nitrogen source (ammonia), and the silicon dopant source

(silane), and the organo-metallic include the group-III sources (TMGa, TMAl, & TMIn), as

well as the magnesium dopant source (Cp2Mg).

Element Group Source Name Common Name Source Type N V Ammonia NH3 Hydride Ga III Tri-methyl gallium TMGa Organo-metallic Al III Tri-methyl aluminum TMAl Organo-metallic In III Tri-methyl indium TMIn Organo-metallic Mg (p-type II Bis-cyclopentadienyl- Cp2Mg Organo-metallic dopant) magnesium Si (n-type IV Silane SiH4 Hydride dopant) Table 2. Sources used in MOCVD growth and doping of Al(In)GaN

The gas mixing cabinet handles all of these sources through an array of mass flow controllers (MFCs), valve manifolds, and pressure controllers (PCs). The hydride sources are all compressed gas sources; they are fed to the reactor from external gas cabinets. The flow of these sources in controlled using pneumatic valves and MFCs; in addition, each hydride source can also be supplemented with an additional dilution push flow controlled by a separate MFC.

The organo-metallic sources are either liquids or fine powders through which a carrier gas is passed to obtain a saturated partial pressure of the source; these sources are housed within the gas mixing cabinet. In order to precisely control the partial pressure of the organo-metallic source in the vapor stream, and thus the molar flow, it is necessary to control the pressure and temperature of the organo-metallic sources. This is achieved though the use of pressure controllers (PCs) and temperature-regulated water baths in which the organo-metallic blubbers rest. In addition to the pressure controller, the actual flow is controlled by a MFC, and like the hydrides there is also a dilution push flow provided by a second MFC. Because of the possibility for pre-reactions between the group-III and group-V sources, the hydride and

72 organo-metallic sources are handled by separate manifolds and do not combine until they reach

the reactor chamber. A full schematic diagram of the MOCVD reactor detailing the gas mixing

capabilities is shown below in Figure 24. This shows all of the source and push MFCs as well as the PCs necessary to precisely control the molar flow rates of the various sources as they entering the reactor.

Figure 24. Schematic diagram of the MOCVD reactor used to grow the material discussed in this work. This diagram shows the full gas handling system, the reactor growth chamber, and the vacuum system.

This simplified schematic diagram fails to address the physical lay out of the gas

handling system. In practice it is necessary to minimize the lengths of many of the pipes in

order to support the rapid switching of sources and to allow for the formation of reasonably

73 sharp interfaces, especially when coupled with redundant sources. However the most

significant reason to limit the switching volume is to allow for the complete temporal

separation of Group-III and Group-V sources when growing AlGaN and AlN using the atomic

layer epitaxy method as described latter in this section (4.4 Atomic Layer Epitaxy for Growth of AlN and AlGaN)

All of the pneumatic valves, MFCs, and PCs are controlled by an Allen Bradly PLC 5 industrial computer system. The computer subsystem controls all of the 72 Digital outputs, 144

Digital inputs, as well as the 40 Analog Inputs/Output pairs necessary to operate the reactor.

Recipes can be written and then compiled and loaded into the PLC system in order to control all aspects of the reactor. In addition a personal computer is capable of interfacing with the

PLC and providing real-time feedback with the ability to adjust the reactor during the growth process to account for any changing circumstances. The PLC also handles limited safety interlocking of the reactor to prevent unsafe conditions from arising. In addition a seconds completely hardware based safety interlocking system moderates the PLC and provided exhaustive interlocking of the system.

The actual epitaxial growth takes place within the reactor chamber. The ends of the reactor are made of electro polished 316 stainless steel; everything inside is either made of graphite of quartz. The wafer, typically double side polished sapphire, sits on a rotating graphite suceptor in the center of a rectangular liner tube within the reactor. The suceptor is inductively heated by an RF-coil; everything else in the reactor is water-cooled. The organo- metallics enter the reactor at the top, and the hydrides enter at the bottom. A quartz divider plate keeps the gasses separate and helps to form a laminar flow as the gasses reach the

74 suceptor. Immediately before the hot growth region over the suceptor the divider plate disappears and the two laminar gas streams begin to intermingle.

The actual epitaxial growth process is divided into 4 major steps, as shown below:

1.) The reactants defuse through the boundary layer into the stagnant region and are

adsorbed onto the wafer surface.

2.) The reactants then migrate along the surface until they find a suitable bonding

site where the chemical reaction takes place. The thermodynamic of the system

are controlled so as to strongly favor reaction at the leading edge of a layer thus

leading good quality layer-by-layer epitaxial growth.

3.) The chemical reaction then occurs, releasing the by products

4.) These byproducts then diffuse back through the boundary layer and are swept

away by the laminar flow, and enter the vacuum system.

4. 1.

2. 3.

Figure 25. Schematic diagram of the 4 step MOCVD growth process.

75 The above process represents the ideal case for MOCVD growth of AlxGa1-xN; however in reality the reactions do not always proceed in such a straightforward manner; this is especially true for the growth of high aluminum composition AlGaN layers. The separation of the reactants is not absolute and there is a tendency for them to combine prior to reaching the substrate. This creates gas phase particles of AlN and AlGaN, and removes viable reactants from the gas stream. Since MOCVD growth of III-Nitrides is traditionally performed with an over pressure of ammonia this leads to a reduction of the group three growth efficiency. A preference for the reaction of gallium over aluminum can also lead to non-stoichiometric growth, which can make compositional control difficult. One solution to eliminating the parasitic pre-reactions is to decrease the growth pressure; this decreases the partial pressures of the reactants in the gas stream, and by providing less chance for interaction it leads to reduced parasitic reactions. Another way to reduce the parasitic reactions is to separate the reactants temporally using a pulsed atomic layer epitaxy technique, as discussed later in section 4.4.

Atomic Layer Epitaxy for Growth of AlN and AlGaN.

As an experiment illustrating the relation ship between pressure and parasitic reactions,

several Al0.4Ga0.6N layers were grown at various growth pressures. In order to qualitatively

analyze the patristic reactions, the construct of growth efficiency is used. Growth efficiency is

defined as the growth rate divided by the total group III molar flow. The numeric value of

growth efficiency can be expressed as a function of the reactor pressure; it also includes fitting

constants related to the reactor geometry and the conditions under which growth takes place.

Assuming similar growth condition, growth efficiency can be used to study the parasitic

reactions as a function reactor pressure. The rate equation for growth efficiency can be

expressed with the following reduced expression83:

76

⎛ p 2 ⎞ = EGEG 0 ⎜− )(exp.].[.. ⎟ (23) ⎜ p ⎟ ⎝ 0 ⎠ ,

where G.E. is the growth efficiency, [G.E.]0 is the ideal growth efficiency in the absence of parasitic reactions, p is the reactor pressure, and p0 is a pressure term related to reactor geometry, total gas flow rate, NH3 concentration, and the reaction rate constant.

3200 Model: growth efficiency G.E.=[G.E.] exp{-(p/p )2} 2800 0 0

m/mole)

μ 2400 Chi^2 = 5245.99142 R^2 = 0.99812 2000 [G.E.] 3278.813 ±86.92402 0 1600 p 33.43898 ±1.18142 0 1200 Al Ga N 800 0.4 0.6

400

Growth Efficiency ( 0 0 25 50 75 100 125 Pressure (mbar)

Figure 26. Growth efficiency of Al0.4Ga0.6N at different growth pressures. The dashed

lines mark the growth efficiency of 1000 mm/mole corresponding to the growth pressure of ~35

mbar.

Figure 26 shows the growth efficiency for several Al0.4Ga0.6N layers grown at various growth pressures. Typically, growth efficiencies in excess of a thousand μm/mole indicate minimal parasitic reactions. According to this fit, a growth pressure of less than ~35 mbar satisfies this condition. However reducing the pressure is often detrimental to the material

77 quality, as witnessed by structural and optical characterization. For example, although the

symmetric (00.2) x-ray scan of the Al0.4Ga0.6N layer grown at 10 mbar shows a small value for

the full-width at half-maximum (FWHM), the asymmetric (105) x-ray scan reveals a broad peak thus implying there are a large number of threading dislocations in the epilayer.

Figure 27. Photoluminescence for a series of AlGaN:Si layers grown at different growth

pressures showing the material quality tradeoff associated with growth at lower pressures.

As the AlGaN growth pressure is reduced, the optical quality of the material begins to

degrade, as shown above in Figure 27. The photoluminescence (PL) spectrum exhibits a weak

band-edge related peak and a relatively intensely-broad low-energy defect-related peak at a

pressure of 10 mBar. The poor material quality in this case is believed to be due to the nitrogen

vacancies that arise from the reduced number of active nitrogen adatoms present on the epilayer

surface at low pressures. Increasing the pressure suppresses the defect related luminescence

and increases the band-edge luminescence; at pressure of 50 mBar or higher the defect

luminescence is negligible for moderate aluminum compositions. As a compromise most of the

AlGaN layers employed in the growth of the back-illuminated UV photodetectors, FPAs, and

APDs presented within the context of this work have been grown at 50 mBar, with 25 mBar

78 only being used for the highest aluminum composition detectors. This was one of the critical

advancements in making high efficiency detectors possible, leading to significant advances

over our early photodetectors grown at 10 mBar.

4.3. Growth Nucleation: Low and Intermediate Temperature

Buffers

There are no commercially viable lattice matched substrates for the growth of AlN or

GaN. In the past we have studied both AlN substrates provided by Crystal-IS,84 as well as

freestanding HVPE grown GaN substrates provided by Samsung;85 however neither is available in the qualities or prices necessary for serious research. In the context of this work, the other problem is that neither is really transplant to solar-blind UV light: the former due to its smaller band-gap, and the latter due to the immature material quality. As such, the vast majority of the day-to-day material growth is conducted on sapphire substrates. In fact all of the photodetectors and focal plane array presented have been grown on double side polished basal plane sapphire substrates. With the AlN and GaN substrates only being used for the UV

LED work86 (a parallel work not discussed here).

Sapphire has a bandgap in excess of 8 eV and does not absorb appreciably anywhere

within the cut-off wavelength range obtainable with the AlxGa1-xN material system. It can

readily be obtained in 2” and larger double side polished wafers with excellent quality with. It

is well suited to the development of FPA due to the large uniform wafers available and the lack

of any absorption. However it is not well matched to AlN in terms of lattice or thermal

expansion coefficients. Sapphire has a hexagonal crystal structure with the lattice constants

79 a=4.785Å and C=12.991Å. AlN is also of the hexagonal family but the lattice constants are

quite different at a = 3 .112 Å and c = 4 .979 Å. Strictly speaking this should correspond to a

35% mismatch, however when AlN is grown on top of sapphire it undergoes a 30 rotation, as

shown in Figure 28 below. Taking into account this rotation the effective lattice mismatch can

then be calculated using equation (24) below; this yields a much better but still large value of

13.2 %.

a a − sapphire AlN 3 − 747.2112.3 a =Δ = = %2.13 (24) asapphire 747.2 3

Figure 28. Crystallography of AlN grown on top of sapphire. The image at the left shows

a plan view. This illustrates to the 30 degree rotationof the crystal structure that leads to a

13.2 % effective lattice mismatch. The image at the right shows a sectional view showing the formation of a misfit dislocation every 8 atoms.

This 13.2 % lattice miss match results in the formation of one misfit dislocation for every

9 atoms of sapphire, or every 8 atoms of AlN.87,88 This 9 to 8 mismatch is illustrated in the

idealized cross-sectional view on the right of Figure 28 above. It is virtually impossible to

80 grow AlN of a suitable quality directly on top of sapphire; instead it is usually necessary to

grow a compliant low temperature buffer layer between the sapphire substrate and the epitaxial

film. AlN grown directly on sapphire is full of dislocations and resembles a polycrystalline mosaic of discontinuous crystal grains due to this large mismatch. Traditionally a low-

temperature GaN or AlN layer is generally grown on top of the sapphire substrate to initiate nucleation.89 This is then followed by the high temperature growth of 1 to 2 μm of AlN or

AlGaN, which serves as a template layer, separating the device active region form the highly

dislocated sapphire interface. Using this approach it is possible to achieve acceptable

dislocation densities for a 1 um AlN template layer in the range of 108 to 1010 cm-2. It is possible to achieve even better dislocation densities with GaN of sufficient thickness, however

GaN is not suitable for back illuminated devices, and its use as a template is only as a basis of comparison within the context of this work.

The buffer used in the majority of this work is a 200 Å AlN buffer grown at ~700 °C; by contrast, bulk AlN on top of this buffer is usually grown at the higher temperature of 1300 °C.

Before growth of the buffer the sapphire substrate is first prepared by annealing in the reactor under only H2 gas at 1200 °C for 10 minutes to desorbe any surface impurities and prepare the

surface for growth of the AlN buffer layer90. This layer is deposited is a poly-amorphous film and then latter assumes its full crystalline structure during the subsequent annealing and growth of the template layer on top. This allows the thick AlN to maintain its crystal structure despite the 13.2% difference between the AlN and the sapphire below.

We have experimentally characterized this sapphire to AlN interface using high-

resolution transmission electron microscopy (HR-TEM)91. We have observer the same 8 to 9

ratio as predicated by theory.87,88 The transition from the aluminum oxide to the aluminum

81 nitride phase is shown in Figure 29 below with the interface and atomic arrangement highlighted.

AlN

8

Generation of misfit Interface dislocations through termination of atomic Sapphire 9 planes (00.1)

Figure 29. High resolution TEM image of the sapphire/ AlN interface showing the

generation of misfit dislocations.

In addition to the AlN buffer layer described in this section it is also possible to use a

GaN or even an AlGaN buffer. However the realization of high-quantum efficiency back

illumined solar-blind photodetectors precludes the use of GaN and low aluminum composition

AlxGa1-xN buffers layers due to the unacceptable losses that would occur, even in the ~200 Å thick buffer layer.

4.4. Atomic Layer Epitaxy for Growth of AlN and AlGaN

On top of the nucleation layer, it is then necessary to grow a thickness of AlN or AlGaN before the growth of the active layers can be completed. While the AlN buffer helps significantly, and ultimately allows for the smooth growth of the subsequent layers; the lattice

82 miss-match is still large, and thus a large numbers of dislocations are formed at the interface.

Typically the template is grown ~1 μm thick, though a thicker template should yield better

material with a lower dislocation density. The growth of this layer allows some of the

dislocation to merge together, turn sideways, or terminate altogether. However we have found

that excessively thick template layer growth leads to increased cracking of the material, with

the critical total device thickness being about 2 μm for 280nm photodetectors and decreasing

for shorter wavelength devices.

The template layer consists of AlN, AlGaN, or a superlattice of the two. These layers are

grown using a novel pulsed atomic layer epitaxy (ALE) technique. Pulsed-ALE involves the

temporal separation of the reactants as a means of reducing or eliminating the parasitic

reactions that would normally occur during the growth of high aluminum composition material.

It has been found that be growing the template layer in this manner the growth rate can

approximately be doubled, and the material quality is significantly improved due in part to the

increased surface diffusion that occurs when the material is grown via ALE.92

4.4.1. Growth of AlN by ALE

AlN is used a first part of the template layer for the solar-blind photodetectors, FPAs and

APDs discussed in this work. AlN is chosen because of its transparency to the wavelengths of

interest and because it has the best lattice match to the sapphire substrate and the AlN buffer

layer use to nucleate the growth. This material is grown using the novel ALE technique where

by the aluminum and nitrogen sources are temporally separated within the reactor. First a brief

pulse of only ammonia and a balance of hydrogen and nitrogen gas is sent to the reactor. This

deposits a thin layer of free nitrogen atoms on the surface of the wafer. This flow of ammonia

83 is then interrupted while simultaneously, a flow of tri-methel-aluminum is sent to the reactor.

This then deposits a layer of free aluminum that reacts with the free nitrogen on the surface and

begins to form an AlN layer. Prior to the deposition of the first pulse of aluminum the buffer is

annealed under ammonia as the reactor temperature is ramped from the ~750 °C used for

growth of the buffer the ~1200 °C used for growth of the AlN. After this the aluminum and

nitrogen sources are switched every 4 seconds as shown in Figure 30 below.

TMAl

NH3 Pulse width = 4 sec

Figure 30. Diagram of the valve switching used to grow AlN by atomic layer epitaxy.

This growth technique yields AlN material with excellent structural properties; this forms

the basis on which these high quality photodetectors, FPAs, and APDs have been realized. The

material exhibits typical symmetric (002) X-ray diffraction FWHMs of less than 60 arc-seconds

and typical asymmetric (105) FWHMs of ~ 300 arc-seconds (X-ray diffraction as well as other

semiconductor metrology techniques are discussed in more detail in Appendix 1: Material

Characterization Techniques). The symmetric and asymmetric omega/two-theta scans for a

typical ALE grown AlN layer as shown below in Figure 31. In addition to having good overall crystalline quality the ALE grown AlN material also has an exceptionally smooth surface with well-ordered atomic steps. The AFM of the same sample is also shown below in Figure 32,

and shows a smooth well ordered surface with an RMS roughness of less than 1 Å.

84

ALE AlN (400nm) on LT-AlN bfr. ALE AlN (400nm) on LT-AlN bfr.

1000

10 100

CPS (105) peak

CPS FWHM 336" (002) peak 10 FWHM 59"

1 1 17.00 17.25 17.50 17.75 18.00 75.3 75.4 75.5 75.6 75.7 ω \ 2θ angle (deg) ω \ 2θ angle (deg)

Figure 31. High resolution x-ray diffraction scans from an ALE grown AlN, showing an extremely narrow systemic (002) and narrow asymetric(105) width owing to the high-quality of the material.

Figure 32. AFM Image of the surface of ALE grown AlN on a low temperature AlN buffer.

85

4.4.2. Growth of AlGaN / AlN Superlattice Template

In addition to the use of ALE grown AlN as a template layer, it is also possible to

implement a template consisting of an Al0.87Ga0.13N/AlN (50Å/50Å) superlattice grown on top

of the aforementioned AlN material. This superlattice is also grown by the ALE method;

however the inclusion of gallium in the AlGaN portion of the superlattice requires a slightly

different approach. The superlattice is grown because it is hypothesized that periodic

interfaces can act to partially filter the dislocations and potentially elevate some of the strain

inherent in growth of AlN on sapphire substrates.93

There are several possible approaches to the growth of the necessary AlGaN material

using the ALE technique. The issue of parasitic reactions between the gallium source and the

ammonia is significantly less problematic than the issue of reactions between the aluminum and

nitrogen sources. Instead the major issue becomes the instability of the gallium surface,

especially at the elevated temperatures used to grow AlN by ALE. Three are three possibilities:

1.) Separate the ammonia and keep the group three sources together, 2.) Pulse both the

aluminum and the gallium sources separately, and 3.) Keep the gallium source open all the

time. However, we have found that none of these solutions is ideal. In no case is the AlxGa1- xN composition stoichiometric, the aluminum composition tends to depend strongly on the

growth temperature. As a compromise we have adopted approach #2 as depicted below in

Figure 33. Using this approach we are able to obtain Al0.87Ga0.13N material with excellent

material quality, when the growth temperature is reduced slightly to ~1150 °C.

86

TMAl

NH3

TMGa

Pulse width = 4 sec

Figure 33. Diagram of the valve switching sequence used to grow high aluminum

composition AlGaN by atomic layer epitaxial.

This AlGaN material is then grown as a superlattice by switching between the valve

phasing shown above in Figure 33 and that discussed earlier for AlN, and shown in Figure 30.

The number of pulses is chosen to yield a (50Å/50Å) superlattice of Al0.87Ga0.13N/AlN. This

yields improved material quality for the template layer; and because the Al0.87Ga0.13N is

transparent to wavelengths longer than 230 nm, it makes an ideal template for the development

of high-quality back-illuminated photodetector, FPAs, and APDs.

87

12-period Al0.87Ga0.13N/AlN SL on AlN RMS roughness ~ 1.3Å

Figure 34. AFM image showing the high-quality of the surface of an AlGaN/AlN SL

grown on AlN, all grown using the ALE technique.

Atomic force microscopy of this template, as shown above in Figure 34, reveals a flat

surface with well-ordered atomic steps and an RMS roughness of only 1.3 Å for a 5 μm × 5 μm

scan size. X-ray diffraction reveals narrow full width at half maximums (FWHMs) for both the

AlN and AlGaN/AlN superlattice: 61 and 62 arc seconds respectively for the symmetric (00.2)

peaks; and, 260 and 270 arc seconds respectively for the asymmetric (10.5) peaks.

Despite the narrow x-ray diffraction, and the well ordered surface by AFM, AlN and

AlGaN/AlN SLs are far from defect free. In the AFM above there are about 50 to 100 small black dislocations visible. Considering the area of the scan, this yield a dislocation density in

the range of 1×108 to 1×109cm-1. This is no better than the values typically obtained for GaN.

88 It is only through extensive optimization of the growth conditions, careful tuning of the buffer

layer, and timing of the pulsed ALE growth that results these good have been achieved and the

performance difference between front and back illuminated devices has been reduced.

However, in addition to the defects seen by AFM, a small number of these defects are

catastrophic defects that severely degrade devices such that their performance is intolerable.

By etching the AlN in a solution of hot KOH or H3PO4, the worst of these defects form large hexagonal pits that can be seen to cover the surface of the wafer.

Figure 35. Macroscopic hexagonal etch pits are formed by the worst of the defects after etching in a soultion of hot KOH or H3PO4.

These fatal defects are an especially large problem for the development of APDs where large reverse biases require high-quality material with low leakage currents. These defects can act as micro-plasma recombination centers under sufficient reverse bias increasing avalanche

multiplication noise and degrading device performance or in the worst cases causing the device

89 leakage current to exceed the power dissipation rating of the device before the onset of

breakdown, making avalanche gain impossible to realize. To this end, a large part of this work has been working to refine the quality of these template layers, with a temporary solution being to test only devices smaller than the mean distance between these catastrophic dislocations.

4.5. Growth and Doping of Wide-Bandgap AlGaN Material

The growth of p-i-n photodetectors and FPAs requires more than just excellent quality

material; it is also imperative that the material can be effectively doped both p- and n-type.

However, the doping of nitrides is has always been an issue since their original discovery. The

problem of doping is further exacerbated by the high aluminum compositions necessary for the

growth of solar-blind devices, typically 40 to 50% aluminum. We have been able to achieve

excellent results for the n-type doping of Al0.5Ga0.5N using a combination of silicon and indium; however doping of the material is still a limiting factor for the devices.

4.5.1. N-type Doping of AlGaN

The most common dopant for the growth of n-type III-Nitrides in silicon. However, even

without silicon, as-deposited (Al)GaN is weakly n-type due to defect related compensation of

the material. The incorporation of oxygen, the most common unintentional donor, can lead the

un-intentional n-type conductivity.94 This makes the n-type doping of (Al)GaN significantly

easier than p-type doping, however with increasing aluminum compositions the ability of

effectively dope the material diminishes.95 Impurities, dislocations, and native defects can form

acceptor-like compensation sites leading to reduced n-type conductivity. For an aluminum

90 composition greater than ~40% oxygen can become a DX center and begin to behave like a

deep acceptor.96,97 Carbon impurities have also been suggested to behave like acceptors.98

Cation vacancies are another acceptor-like compensating sources whose formation energy decreases with increasing Al composition.99 In addition to all of these cases, dislocations may also introduce acceptor-like centers through dangling bonds along the dislocation line.100

The addition of indium to the growth of the conduction layer helps improve the

conductivity of the Si doped material, but can have a negative impact on the morphology.

Without indium at first, by careful optimization of the SiH4 flow, and the use of the high- quality dislocation filtering superlattice template described above, Al0.5Ga0.5N:Si had a with a

tolerable carrier concentration of n ~ -1×1018 cm-3 and a mobility of μ ~ 40 cm2/V·s (values

obtained via Hall Mobility Measurement, as described in Appendix 1: Material

Characterization Techniques). The addition of ~25 μmol/min of TMIn yields a carrier

concentration of n ~ -5×1018 cm-3 and mobility of μ ~ 60 cm2/V·s, corresponding to an

electrical conductivity ~5 times higher than conventional singly-doped AlGaN of the same

101 aluminum composition. We are able to consistently obtain Al0.5Ga0.5N:Si-In with a sheet resistivity of 0.02 Ω.cm as shown below in Table 3.

Material Growth Resistivity Carrier Concentration Temperature (Ω.cm) (cm-3) (ºC) 18 Al0.5Ga0.5N:Si 1050 1 1×10 18 Al0.5Ga0.5N:Si-In 1050 0.02 5×10 19 Al0.5Ga0.5N:Si-In 900 0.007 2.2×10

Table 3. Comparison of electrical properties of n-type Al0.5Ga0.5N grown with different approaches to the doping of the material.

91 There are several hypothesizes as to how the addition of indium acts to enhance the conductivity of n-type AlGaN layers. A large part of the enhancement may be due to the improved material quality; the addition of indium into ternary AlGaN layers results in the reduction of defect density102, as long as the temperature is still sufficiently high. This

improvement in the structural quality reduces the dislocation-induced compensation, and can

lead to more effective doping of the material. Indium may also fill some of the incorporation-

defects that hinder the doping of high Al-content AlGaN layers, such as DX centers and cation

vacancies. Specifically, indium may occupy the cation vacancies that inhibit the acceptor formation.

A number of groups have also reported similar success doping high Al content AlGaN

using Si-In co-doping 103, 104. Even better results can be obtained if the temperature is reduced significantly in order to facilitate more indium incorporation and thus further improve the conductivity of the material. This growth of AlGaN:Si-In at a much reduced growth temperature can yield good results, as shown above in Table 3. However two problems can

easily arise with low temperature growth: first the low temperature growth results in poorer

morphology which then adversely affects the quality of the subsequent epilayer growth; and

secondly the resistivity of the low-temperature AlGaN:Si-In layer, while initially quite low, has

a tendency to increase after the subsequent high-temperature device re-growth, possibly due to

desorbtion of a large part of the indium from the co-doped material. There is a tradeoff

involved but we have found the optimum temperature to be a relatively high temperature of

~1100 ºC for growth of the conduction layer.

Indium also serves several other beneficial purposes, the most important to growth of

solar blind photodetectors is that it helps further relieve some of the strain in the layer 105, and

92 thus allows for growth of a thicker crack-free Al0.5Ga0.5N:Si-In conduction layer, while still realizing a completely crack-free device structure. The ability to grow thicker means that, with the inclusion of indium, not only is the material resistivity reduced, but the over all sheet resistivity is also reduced.

4.5.2. P-Type Doping of AlGaN

The effective p-type doping of III-Nitrides is one of the largest limiting factors in

obtaining working devices. Despite the fact that GaN has been researched for over 30 years106 it wasn’t until 1989 that the first truly p-type material was obtained via low-energy electron beam irradiation (LEEBI).107 The most common p-type dopant for the growth of III-Nitrides is

magnesium; however, as deposited (Al)GaN:Mg results in heavily compensated material that is

still n-type. The LEEBI was necessary to de-compensate the material and activate the

magnesium, thus allowing the material to become p-type. Several years latter, after intense

study, a breakthrough was made when it was discovered that the annealing of (Al)GaN:Mg

under nitrogen also yields p-type material.108 This led to an understanding of the role hydrogen

plays in the passivation of the magnesium.109,110 The thermal annealing of (Al)GaN:Mg post

growth is necessary to break these H-Mg bonds. This creates free H2 gas and leaves activated

magnesium acceptors in the material.

93

550 5 500 450 400 350

5 300 1 4 250 3 1. CQD 1 2. Tanaka et al, APL 65(1994), p.593 200 2 3. Kozodoy et al., APL 74(1999), p.3683 4. Suzuki et al., JCG 189/190(1998), p. 511 150 5. Nam et al., APL 83(2003), p.878 Mg Activation Energy (meV) 1 0 20406080100 Al Composition (%)

Figure 36. Activation energy of Mg in AlxGa1-xN as a function of Al mole fraction

Despite the fact that we now have a good understanding of the Mg doping mechanism in

III-Nitrides, the issues of effectively growing and doping p-type material have still not fully

been resolved. Not only it the Mg-H activation process not 100% efficient, but Magnesium is

also somewhat of a deep-acceptor: the activation energy of Mg in GaN:Mg is 125 meV and

increases up to 510 meV for AlN:Mg as shown in Figure 36.111,112 This means that a room temperature, even if the H-Mg bonds have all been broken, that statistically speaking only a small fraction on the Mg ions will be thermally activated. In addition, compensation by native defects is increased for higher aluminum compositions. Nitrogen vacancies and cation interstitials both have reduced formation energies for higher aluminum compositions, and inhibit the ability to effective dope the material.

For the material discussed in the work, we typically anneal our p-type layers in a rapid

thermal annealing system under a flowing purge of dry N2 gas. The thermal annealing

94 temperature is typically 1000 °C and the duration is 30 seconds. Typically p-type GaN can be

achieved with a carrier concentration, mobility, and bulk resistivity of 1x1018 cm-3, 7 cm2/V·s,

113 and 0.8 Ω·cm, respectively. However AlxGa1-xN:Mg for x greater than about 0.2 is usually

too highly resistive to be measurable via hall mobility measurement (as limited by our initial

system as described in Appendix 1: Material Characterization Techniques). This has made

optimization of p-type AlGaN a particularly difficult challenge. One rout to overcoming this

limitation is the use of p-type AlGaN/GaN superlattices. Using this approach we have obtained

Al0.26Ga0.74N/GaN:Mg superlattices with even better properties than those of bulk GaN.

However we have since abandoned that approach in favor of bulk AlGaN:Mg growth when

working on devices operating in the solar-blind region, and thus requiring AlxGa1-xN with x >

0.36.

95 5. Experimental Procedure: Large Area Single Element

Detectors

5.1. Introduction

In the proceedings sections we have briefly discussed the common types of III-Nitride photodetectors, provided some basic background on the operational and discussed the basics of the material growth. In this section we discuss the first main thrust of this work: back illuminated solar-blind p-i-n photodetectors, and look in depth at the device structure and its growth, the processing steps and procedures, and provide an overview the basic characteristics of these devices.

Back illuminated detectors operating in the solar blind region are of special interest, however they are among the most difficult to grow and fabricate. The AlGaN material system has a wide direct bandgap and is ideally suited to detection of UV light in the solar blind range

(λ < 285 nm), however this wavelength region requires Al compositions of around 50%. The

AlGaN material system suffers from several key problems: large dislocation densities, low doping n-type and p-type efficiency (i.e. conductivity), and lattice and thermal expansion mismatches leading to cracking of the material. All of these problems are exacerbated by the increased aluminum compositions necessary in back-illuminated solar-blind photodetectors.

However with careful control and tuning of the growth, and the introduction of some novel techniques, it has been possible to overcome some of these limitations and realize high performance back-illuminated solar-blind photo detectors.

96 5.2. Material Growth and Characterization

Growth is carried out in a horizontal flow low-pressure metal-organic chemical vapor deposition reactor (AIXTRON 200/4-HT). Double side polished basal plane (00.1) sapphire is chosen because it is the only well-established substrate that is UV transparent at the wavelengths of interest; it needs to be double side polished to minimize reflection at substrate since the light will be entering the device through this back surface. The nucleation of the growth begins with the deposition of a thin 200 Å low-temperature AlN buffer layer. It is also possible to use an AlGaN buffer with a sufficiently high Al composition to avoid absorption, however even a thin GaN buffer will absorb an appreciable quality of the incident light and significantly reduce the efficiency of the photodetector. On top of this nucleation layer, 350 nm of high-quality AlN is grown at a temperature of ~1100 °C. This is followed by a 30-

114 period Al0.87Ga0.13N/AlN (50Å/50Å) dislocation-filtering strain-relief superlattice. Atomic force microscopy of this template, as shown in Figure 37, reveals a flat surface with well- ordered atomic steps and an RMS roughness of only 1.3 Å for a 5 μm × 5 μm scan size. X-ray diffraction reveals narrow full width at half maximums (FWHMs) for both the AlN and

AlGaN/AlN superlattice: 61 and 62 arc seconds respectively for the symmetric (00.2) peaks; and, 260 and 270 arc seconds respectively for the asymmetric (10.5) peaks. Similar

AlGaN/AlN superlattices have been instrumental in the realization of back emission UV light emitting diodes operating in the same solar-blind range.115,116

97

Figure 37. AFM image showing the surface of the high-quality AlGaN/AlN SL template,

3 nm data scale. The RMS roughness for the 5 mm square scan shown above is 1.3 Å.

Before transitioning into the device structure, a second high-temperature AlN layer is grown, 50 nm thick. Lateral conduction is achieved through the use of an 800 nm Al0.5Ga0.5N silicon-indium co-doped layer. In order to maximize the lateral current conduction, and thus collection of photo-generated carriers, the conduction layer should ideally be as thick as possible. The limiting factor of conduction layer thickness is generally cracking of subsequent growth, and depends on material quality; thus the emphasis put on high-quality of the buffer and template layers.

The addition of indium to the growth of the conduction layer helps improve the conductivity of the Si doped material, but can have a negative impact on the morphology due to the reduced temperatures required to obtain indium incorporation. Preliminary Al0.5Ga0.5N:Si

98 layers were grown on the high-quality dislocation filtering superlattice template described

above without the used of indium. By carefully optimizing the SiH4 flow these layers had a

tolerable carrier concentration of n ~ -1×1018 cm-3 and a mobility of μ ~ 40 cm2/V·s. The

addition of ~25 μmol/min of TMIn yields an improved carrier concentration of n ~ -5×1018 cm-

3 and mobility of μ ~ 60 cm2/V·s, corresponding to an electrical conductivity ~5 times higher than conventional singly-doped AlGaN of the same aluminum composition.117

A number of groups have also reported similar success doping high Al content AlGaN

using Si-In co-doping 118, 119. In order to facilitate indium incorporation and improve the conductivity of the material, they tend to grow the AlGaN:Si-In at a much reduced growth temperature. This can yield good results for the conduction of the layer, however two problems can easily arise with low temperature growth: first the low temperature growth results in poorer morphology which then adversely affects the quality of the subsequent epilayer’s growth; and secondly the resistivity of the low-temperature AlGaN:Si-In layer, while initially low, has a tendency to increase after the subsequent high-temperature device regrowth, possibly due to desorbtion of a large part of the indium from the co-doped material. There is a tradeoff involved but we have found the optimum temperature to be a relatively high temperature of

~1100 ºC for growth of the conduction layer.

Indium also serves several other beneficial purposes, the most important to growth of

solar blind photodetectors is that it helps further relieve some of the strain in the layer 120, and

thus allows for growth of a thicker crack-free Al0.5Ga0.5N:Si-In conduction layer, while still realizing a completely crack-free device structure.

Before the conduction layer and template can be used to grow an active region, it is

necessary to confirm the UV transparency at the wavelengths of interest. UV transmission

99 measurement involves using a xenon light source, a monochromater, and a calibrated UV

enhanced Si photodetector as described in detail in Appendix 1: Material Characterization

Techniques. Optical transmission of a typical conduction layer is shown in Figure 38; it shows

a sharp cutoff, and well-defined fringes owing to the high-quality of the material and smooth

surface of material grown as described above. In addition the absorbance squared can be

calculated; this data is also plotted in Figure 38. By performing a linear fit an absorption edge of 260nm is calculated. The template will absorb light with a wavelength shorter than this, and we will see a second cut-off in the detector response. Thus creating the traditional band pass like response spectra typical of back illuminated photodetectors.

100

5 Absorbance^2 4 Linear Fit 3 Absorbtion Cutoff

2 4.76 eV 1 260 nm

Absorbance^2 (a.u.) 0 3.5 4.0 4.5 5.0 5.5 Energy (eV)

100 80 % Transmision 60

40 20 0

Optical Transmission (%) Optical Transmission 250 300 350 400 Wavelength (nm)

Figure 38. Optical transmission and absorbance squared of the Al0.5Ga0.5N:Si:In

conduction layer grown on a high-quality AlN/AlGaN SL template. The conduction layer

shows a sharp cut off owning to the excellent material quality, with the absorption edge

occurring at 260 nm.

On top of this highly conductive transparent layer the p-i-n photodiode structure consisting of 100 nm n-type Al0.45Ga0.55N:Si followed by 200 nm un-doped Al0.36Ga0.64N and

50 nm p-type Al0.36Ga0.64N:Mg is grown. In order to help in the formation of ohmic contacts, it is common to cap the structure with a thin highly doped p-GaN layer. The complete structure

101 is shown in Figure 39 below. Inspection of the material after growth reveals a smooth surface with no signs of cracking.

Figure 39. Schematic cross-section showing the structure of a back-illuminated solar-

blind photodetector.

P-type doping is an on going problem in III-Nitrides, and the doping of the p-GaN and

p-AlGaN are especially important to the operation of the photodetector. If the material is not

sufficiently well doped the formation of a p-type ohmic contact becomes difficult. Non-ohmic

p-type contacts can lead to a significant negative photoresponse arising out to 364 nm from

Schottky barrier photodetector like behavior of the contact. This behavior has been reported in

the literature, and is attributed to the non-ohmic nature of p-type contacts 117, 121. The presence

of a negative photoresponse subtracts from the overall efficiency of the device and decreases

the device’s degree of solar-blindness. A typical photoresponse for a solar-blind back

illuminated photodetector exhibiting a negative photoresponse is shown in Figure 40. In order

102 to maximize the out-of-band rejection ratio, the photoresponse from the Schottky needs to be

minimized.

Figure 40. Responsivity vs. wavelength for a Back illuminated p-i-n photodiode, showing

a significant negative photoresponse.

It was found that this effect arises in our devices due to slightly over doping the p-

AlGaN.122 Optimization of is difficult because direct attempts to characterize the

Al0.36Ga0.64N:Mg layer by Hall effect measurements reveal that the material is highly-resistive

(measurement limited). Thus, in order to optimize the material it is proposed to use the transfer length method (TLM) to study the I-V characteristics, specific contact resistivities, and sheet resistances as a function of doping.

103 To examine the quality of the p-type material a series of test structure are grown. These

begin with the same high-quality superlattice template grown on sapphire, as previously

described. Then 100 nm of Al0.36Ga0.64N:Mg, and in order to help in the formation of ohmic contacts, this layer is then capped with the same highly doped thin p-GaN layer, as in the photodetector structure. This test structure is shown below in the inset of Figure 41. The

contact resistivities can then be extrapolated from TLM measurements, and the insight

provided can be used to optimize the AlGaN:Mg. It was found that a slight reduction in the

doping level resulted in a much more ohmic contact (optimized-doping). Further reduction did

not improve the I-V characteristics, and would have a detrimental effect on the p-i-n junction.

600 Various Cp Mg flows 2 Over-doped 400 Optimized-doping Under-doped

A) 200 μ 0

-200 Ni/Au Ni/Au (30Å/30Å) 15 μm (30Å/30Å) Current ( p-GaN:Mg (50 nm)

-400 p-Al 0.36 Ga0.64N:Mg (100 nm)

-600 AlGaN /AlN SL Template

-10 -5 0 5 10 Voltage (v)

Figure 41. I-V curves for several different Cp2Mg flows used to optimize the p-type

layers. The structure and measurement geometry are shown in the lower right hand corner.

104 5.3. p-i-n Photodetector Processing Overview

In order to fabricate the material into diodes, the samples are first annealed at 1000 ºC for 30 seconds, under N2 ambient, to activate the magnesium in the p-type layers. Then 30 Å

Ni / 30 Å Au is deposited on top of the mesa, and annealed under ambient air at 500 ºC for 10 minutes, to form p-type ohmic contacts. The devices are lithographically patterned into 1 mm ×

1 mm square mesas and etched using electron cyclotron resonance (ECR) etching. A 300 Å Ti

/ 1800 Å Al ring contact is then deposited around the mesa, and is then annealed under nitrogen to form an ohmic n-type contact to the Al0.5Ga0.5N:Si-In conduction layer. Finally a 400 Å Ti /

1200 Å Au bonding pad is deposited on top of the mesa to facilitate wire bonding of the device.

An optical micrograph of a processed and wire-boned device is shown below in Figure 42.

Figure 42. Optical micrograph of a 1mm x 1mm Solar-blind photodetectors showing the device layout, and the proximity of the n-contact.

Additional post processing of the device is also worth considering, although none is

implemented for these devices. The electrical properties of the device can be improved by

105 passivation of the surface. However, there has been less investigation of passivation of high

Al-composition p-i-n photodetectors; the majority of the work focuses on Schottky barrier

123, 124 photodetectors and MSMs. Both Si3N4 and SiO2 have been implemented, and their

performance characteristics have been compared.125 In surveying the literature the overwhelming preference seems to be for SiO2 passivation. SiO2 also has the advantage of having much higher transmission in the solar blind region, thus a SiO2 passivated device can optimally be used for both front and back illumination.

Additionally an anti reflection coating can be applied to the back of the sapphire

substrate to reduce reflection at the surface and thus improve the external quantum efficiency of

the device. However, to date very few groups seem to have actually implemented anti-

reflection coating on their devices, and the majority of publications make a point of saying

“without anti reflection coating.” In the case presented here, no post processing of the device

was implemented.

5.4. Photodetector Measurement and Discussion

The processed devices are mounted to a copper block with a hole in the center to

facilitate handling, provide pads for wire bonding of the device, and to allow illumination of

the sample for back-illuminated responsivity measurement. Working with back illuminated

devices can become a challenge because of the increased complexity involved in packing and

handling these devices. A novel fiber-coupled measurement setup was adopted to help

facilitate back-illuminated device testing. The full details of device characterization

methodologies are provided in Appendix 2: Device Testing and Characterization. The current-

106 voltage (I-V) parameters of the device are measured using a low noise probe station and a

semiconductor parameter analyzer. A typical I-V curve is shown in Figure 43.

10-1 10-2 10-3 1x10-4 1x10-5

10-6

Current (A) 10-7 10-8 10-9

-202468 Voltage (V)

Figure 43. I-V curve of a typical back illuminated photodiode in semi-log scale.

From the IV curve it is possible to determine the device turn-on and estimate the series

resistance by plotting I(dV/dI) vs. current and taking a linear fit in the high current regime as

discussed in Reference 126. In this case a turn on of 6.8 V, and a series resistance of 202 Ω are estimated. The leakage is determined by applying 5 volts of reverse bias to a dark junction, and measuring the current; in this case the leakage current density was 230 μA/cm2.

In addition, the ideality factor can be extracted in order to gain insight into the

conduction mechanisms operating in the photodetectors. Figure 44 shows the natural log of the current, in the low current regime for a typical device. Under low current injection, the effects

107 of series resistance become negligible. The ideality factor is calculated by fitting the linear part of the curve at low current with the following equation:

qV CI ×= exp( ) , (25) f nkT this yields an ideality factor of n = 2.89 for this diode.

-6 I-V Data -8 Linear Fit -10 -12 -14

-16 Ideality Factor -18 η = 2.89

ln (Current) -20 -22 -24 -26 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 Voltage (V)

Figure 44. Natural log of current versus voltage. The low current regime is modeled and an ideality factor of 2.89 is extracted.

This equation arises from the combination of the equations for diffusion current and recombination current, the two currents that usually dominate in the diode current. The ideality factor is expected to have a value between 1 and 2. Whereas if n is closer to 1, then diffusion current dominates, however if n is closer to 2, recombination current dominates. However, a

108 value of n = 2.89 falls well outside of this expected range, suggesting that an additional process makes a significant contribution to conduction. The most probable explanation for the non- ideality of this device is the weak Schottky-like nature of the p contact creating a second diode in series with the p-i-n junction. On unoptimized diodes, multiple ideality factors can actually be extracted as each of the junctions turns on as the voltage increases.

Figure 45. Responsivity vs. wavelength for a typical photodiode, showing a peak responsivity of 150 mA/W at a wavelength of 280 nm, and no negative photoresponse. This peak responsivity corresponds to an external quantum efficiency of 68%.

The photoresponse is measured using a high intensity xenon lamp and monochromator, calibrated with a reference calibrated UV-enhanced silicon photodetector, such as Newport model number 1779. The photoresponse under back illumination for a typical device is shown in Figure 45. The spectral response drops four orders of magnitude from the peak into the near-

109 UV region, and no negative photoresponse occurs: The photoresponse curve never changes

sign, at any point, in contrast to the photoresponse of the un-optimized diode that was shown in

Figure 40. This device shows an unbiased peak responsivity of 150 mA/W at 280 nm, with a

FWHM of 12 nm; this responsivity corresponds to a record-high-value of 68% for the external quantum efficiency of the device (shown in Figure 46). Under a –5V bias (measurement setup

limited) the responsivity increases to 169 mA/W, corresponding to an external quantum

efficiency of 74%.122 Analyzing the noise performance of the device as discussed in section 3.3

1 Noise Analysis in AlGaN p-i-n Photodiodes, we get a detectivity of ~ 102.2 10 ⋅× 2 WHzcm )/( .

Figure 46. Shows the external quantum efficiency of the device in linear scale.

Based upon these direct measurements of the external quantum efficiency of the

photodetector, we can extrapolate an internal quantum efficiency by taking into account reflective losses at the substrate and template layer interfaces, and incomplete absorption in the

200 nm thick active region of the device. Theory based upon the published indexes of

110 refraction for sapphire and AlN would predict that 8.46% and 0.58% of the light would be

reflected by those interfaces, respectively.127,128 However from the transmission experiments discussed above, we find that the actual value is closer to 12%. Based upon the 200 nm

129 absorption region and the published absorption constants for Al0.36Ga0.44N, the expected

absorption in the active region is only 92%. Combining all of these correction factors together,

as summarized in Figure 47 below, it becomes possible to estimate the internal quantum

efficiency of the device. We expect the internal quantum efficiency to be ~81% in unbiased

operation, and to increase to nearly 88% under a moderate 5 volts bias.

Figure 47. Schematic diagram of the reflective and transmission losses in our device.

In summary, we over viewed the growth and processing of high efficiency AlGaN-based back-illuminated solar-blind p-i-n photodetectors. High-quality AlN and AlGaN/AlN superlattice layers and a high-lateral-conductivity n-type AlGaN:Si-In co-doped layer are

discussed as a means of improving the device. In addition optimization of the p-AlGaN layer

in order to eliminate the out-of-band negative photoresponse is discussed.

111 5.5. Deep UV Back-Illuminated Photodetectors (255 nm)

As an extension to the work on back-illuminated solar blind photodetectors, there have been efforts to pursue shorter wavelength devices operating at 255nm. The progression from

280 nm to 255 nm is technically non-trivial, however it is still interesting to consider the realization of deep UV photodetectors. As the aluminum composition is increased, parasitic reactions between the aluminum precursor (trimethylaluminum) and ammonia (NH3) lead to reduced growth rates and lower material quality. In addition, adequate n-type doping becomes increasingly difficult to achieve. This, combined with the well-known problem of p-type doping, makes the realization of these very short wavelength back-illuminated p-i-n photodetectors challenging. However we have successfully demonstrated 255 nm back- illuminated photodetectors with a peak responsivity of 95 mA/W at 255 nm.

Figure 48. 255 nm deep UV photodetector device structure.

112 The basic device structure is the same as discussed previously for the 280 nm photodetectors, however the aluminum compositions have all increased significantly (shown in

Figure 48). The conduction layer is now Al0.67Ga0.33N:Si-In, and the aluminum composition in

the absorber region has increased to be Al0.57Ga0.43N. In order to accommodate these elevated aluminum compositions a slight modification of the growth is necessary: the growth pressure is decreased from 50 mbar (used for the 280 nm photodetectors) to the much lower pressure of 25 mbar. Reducing the pressure helps suppress parasitic pre-reactions and yields more manageable growth rates. However as the pressure is decreased the material quality is degraded due to the increased occurrence of nitrogen vacancies.130 We have found that 25 mbar

is an unfortunate compromise required for shorter wavelength devices. The same Si-In co-

doping scheme was also found to be effective at these reduced pressures and aluminum

compositions: Hall effect measurement of the Al0.67Ga0.33N:Si-In layer revealed the resistivity,

mobility, and carrier concentration of this layer to be 0.05 Ω.cm, 26 cm2/V.s, and -4.7×1018 cm-

3, respectively.

113

100 -1 10 Un-Biased Un-Biased 80 External Q.E. -2 Peak Responsivity 10 95.7 ma/W 46.5% @ 255 nm @ 255 nm 60 10-3

1x10-4 40

-5 1x10 20 Responsivity (A/W) Responsivity -6 10 0 200 225 250 275 300 325 350 375 400 425 450 200 225 250 275 300 325 350 375 400 425 450

Wavelength (%) Efficiency Quantum External Wavelength

Figure 49. Unbiased responsivity in log scale (left) and the corresponding unbiased external quantum efficiency in linear scale (right).

The photoresponse under back illumination is shown in Figure 49 above. This device shows an unbiased peak responsivity of 95.7 mA/W at 255 nm with a FWHM of ~7 nm, which corresponds to a value of 46.5% for the external quantum efficiency of the device. The absolute response drops three orders of magnitude from the peak into the near-UV region.

However, two slight humps in the near-visible response can be seen; these are due to the material quality and difficulty in realizing good ohmic contacts to a back-illuminated p-i-n photodetectors with such high aluminum compositions. None the less, this represents a significant achievement in back-illuminated deep-UV photodetection.

114 6. Experimental Procedure: Solar-Blind Focal Plane

Arrays

6.1. Introduction

Because of the low natural background in the solar blind region (λ < 290 nm), photodetectors and focal plane array (FPA) imagers operating in this range allow for a number of unique applications as described previously in section 2.2 UV Photodetector Applications.

Currently solar-blind imaging is performed with either a photocathode and micro-channel plate combination, or a UV-enhanced silicon photodiode with a band-pass filter. Neither of these options is ideal: the photocathode and micro-channel plate combination is a fragile vacuum tube device requiring a high-voltage power supply, while the silicon-photodiode is not intrinsically solar-blind and suffers from increased size and complexity, and decreased efficiency due to the requirement of optical filtering. In the past few years, technological and scientific advances in high Al composition AlGaN and AlN based semiconductor materials have led to the development of visible blind p-i-n photodiode FPA cameras131,132 ,133 ,134 and a

renewed interest in the development of intrinsically solar-blind FPA cameras. The first solar-

blind FPA camera was reported by BAE systems in 2000; however the paper did not provide

any actual images from the camera135. The first images from a solar-blind FPA camera were not published until 2002, but even then the quality was lacking and they were unable to provide a full frame 320 × 256 imaging.136

In this work, we report the successful synthesis, fabrication, and testing of an AlGaN- based back-illuminated solar-blind ultraviolet focal plane array, fully realized within our

115 research laboratories. The array consists of 320 x 256 individual p-i-n photodiodes bonded to

an Indigo 9809 read-out integrated circuit (ROIC).137 This ROIC uses a capacitive trans- impedance input amplifier (CTIA) circuit to interface with the pixels in order to record the low background image. This ROIC also contains all the row and column select logic necessary to use four clocks and one analogue output to interface with and external camera driving system.

The details of the ROIC are provided in Appendix 3: ROIC Specifications and Camera

Interfacing. The camera system used for general FPA testing is a CamIRa infrared imaging

system manufactured by SE-IR138. However, as an extension of this work the electrical engineering and design necessary to build and entirely custom portable camera system was undertaken, the details of that project are provided in Appendix 5: Development of a Portable

Camera System.

6.1.1. Focal Plane Array Technology

There is a considerable increase in complexity in going from the single pixel large area photodetectors to developing 2-dimentional cameras based on focal plane arrays (FPAs). The many thousands of pixels in a focal plane arrays require the use of a sophisticated multiplexer to interface them to image creation electronics. Typically this multiplexer is manifested as a separate silicon application specific integrated circuit. In a very small arrays the pixels can be individually wire bonded to an a neighboring image processing circuit; however in larger arrays it becomes necessary to use flip chip technology to hybridize the FPA to a Si read out integrated circuit (ROIC), shown in Figure 50. The ROIC contains a sample and hold unit cell for each individual pixel: in the case of detection low background UV signals with high impedance III-Nitride photovoltaic pixels, this is traditionally a CTIA (capacitive trans-

116 impedance input amplifier) unit cell. This cell uses a trans-impedance amplifier with capacitive

feedback to integrate the photocurrent 139: a simplified schematic of a CTIA unit cell is shown

in Figure 51. The ROIC also contains multiplexer electronics to sequentially readout the various rows and columns of the array. Depending on the application, the ROIC may contain more sophisticated imaging processing electronics: this varies from the simplest ROICs that just output the raw pixel data, to more complex designs that allow programmatic control of gain and integration, and output NTSC compatible video.140

Figure 50. Shows a schematic cross-section of a III-Nitride FPA flip-chip bonded to a Si

ROIC.

Figure 51. Shows a simplified schematic of a typical CITA unit cell as used in a ROIC

117 Hybridization with a ROIC requires that the photodetector be illuminated through the

substrate so that the epitaxial side can be indium bump bonded directly to the ROIC. As

discussed earlier, back illumination introduces a number of complications for III-Nitride

photodetectors; most of these problems arise from the need for high Al content in the template

and conduction layers in order to provide UV transparency. The photodetector pixels are

generally on the order of 30 × 30 μm2 with only 5 μm between adjacent pixels, and there are

81,920 of these pixels in a quarter-NTSC format array, all spanning an area of less than 1 cm2.

Fabrication of focal plane arrays requires a consistently high level of uniformity over a

large area. It’s also necessary to obtain good current spreading to allow current collection from

the center of the array, which may be up to 5 mm from the n-contact. Hybridization also

introduces its own host of problems: it requires additional processing to passivate the detectors,

open a window, and deposit Indium bumps on each of the mesas as well as the ROIC itself. In

addition a specialized flip-chip bonding tool is necessary to allow proper registration of the

FPA and the ROIC, and to form a mechanically and electrically robust bond across the entirety

of the array. Despite these difficulties, a number of groups have been successful in reporting

GaN and AlGaN based focal plane arrays.141, 142 , 144 , 145 , 146 , 147 , 148 , 149 , 150

6.1.2. Historic Development of UV Focal Plane Arrays

Linear arrays of III-Nitride photodetectors were being reported as early as 1996; an example from this era is the linear array of 16 photoconductors on GaN using MSM structures reported by Huang et. al..141 By 1997, the first two-dimensional arrays were being reported.

However the array dimensions were still too small to be of practical application for UV imaging; for example, a simple 8x8 GaN Schottky photodiode array was barely able to record

118 the image of two parallel lines.142 The first practical 256 x 256 UV imaging camera was demonstrated in 1999 by the NASA/Goddard Space Flight Center.143 This camera consisted of

an array 65,532 30 μm x 30 μm photoconductive GaN UV detector elements In bump boned to

a Si read-out intergraded circuit. The FPA assembly was integrated with driver electronics and

a computer so as to create a complete imaging system. This camera was capable of taking

pictures with complex two-dimensional geometry. However, the GaN-based devices were not

solar-blind.

Building a camera based off of photoconductive detector elements is easier than using

photovoltaic elements; a photoconductive array can take advantage of the potentially high

photoconductive gain to realize a much larger responsivity than is possible with p-i-n photovoltaic detector elements. A GaN p-i-n phtotodetector, assuming 100% external quantum efficiency, has a maximum theoretical responsivity of 0.270 A/W, whereas the photoconductive pixels in the array described above had an estimated responsivity of 690 A/W . However, photovoltaic detectors generally have the advantage of much faster response times than photoconductors; for example, the photoconductive array had an estimated response time to 0.2 to 1.0 mS, whereas a similar p-i-n photodiode array has as estimated response time of 1 to 100 nS.144 This allows for much faster frame rates, limited only by the integration time necessary to obtain an adequate signal. An additional advantage is that unlike photoconductors, photodiodes can operate without an applied bias; operation of theses photodetectors without bias significantly reduces the noise levels, and allows for much higher detectivities.

The first successful demonstration of a UV camera based on GaN/AlGaN hertrostructure p-i-n photodetector elements was also reported in 1999.145 The pixels of this

array had an estimated responsivity of 0.2 A/W from 365 to 320 nm. This visible-blind camera

119 was capable of imaging a simple alphanumeric scene composed of a brass gobo backlit by a

Hg(Ar) pencil lamp. In a second publication an attempt was made to image human

likenesses,146 however the 32x32 resolution of this first camera limits the usefulness. Since

then, other groups have reported 32 x 32 arrays of p-i-n photodetectors;147 in addition larger

128 x 128 and 320 x 256 148 visible blind p-i-n based cameras have been demonstrated.

Successes in the development of GaN based UV focal plane arrays, and single pixel

AlGaN photodetectors, especially back-illuminated devices, have recently lead to the

realization of focal plane arrays based on AlGaN absorption layers.149, 150 However, to date,

very few truly solar blind cameras have been realized. Reference 150, in addition to reporting several longer wavelength cameras, also reports that in 2001 BAE systems successfully processed a wafer provide by Cree Inc.151 to create a 265 to 285 nm Solar blind camera based on an Al0.44Ga0.56N absorption layer. However, they only provide statistics on the uniformity and pixel operation, and they don’t publish any images from this first reported AlGaN based solar-blind camera. A year latter North Carolina State University reported a solar-blind camera and published a severely cropped image of several geometric shapes; however, the quality of the image shown reveals that there is a great deal of improvement necessary before III-Nitride solar blind cameras can be commercialized.

6.2. Material Growth and Characterization

The back-illuminated AlGaN p-i-n photodetectors used in the fabrication of this array are grown by low-pressure metal-organic chemical vapor deposition in an Aixtron 200/4 horizontal flow reactor on c-plane sapphire substrates. First a low-temperature AlN buffer is deposited,

120 then a ~ 1 μm thick high-temperature AlN template layer follows. The n-type lateral

conduction layer consists of Al0.5Ga0.5N:Si-In grown at an intermediate temperature. The

active region consists of a simple p-i-n junction with a 200 nm thick Al0.32Ga0.68N intrinsic

region. The structure is then capped with a p-GaN contact layer to assist in the formation of

ohmic contacts. This structure is similar to that of the broad-area back-illuminated single

element detectors described in the previous chapter that possessed a 68% external quantum

efficiency at 0 volts applied bias.152,153

6.3. FPA pixel characteristics

The processed FPA consists of an array of 320 x 256 discrete 25 μm x 25 μm pixels on a

30 μm period with a single common n-contact ring around the periphery of the array. Before bonding the FPA to the ROIC, an extensive study of the electrical properties of the individual pixels of the array was conducted. The I-V curve of a representative pixel from the middle of the array is shown in Figure 52. The device shows a sharp turn-on at 4.7 volts with a series resistance of 4.3 KΩ. The I-V curve in log scale shows the high quality of the p-n junction, with a reverse bias leakage current that falls below the limitations of our measurement setup.

By fitting the I-V curve154 an ideality factor of 3.6 is estimated.

121

1.2 1x10-3

-4 1.01x10 1x10-5 1x10-6

-7 0.81x10

-8

1x10 (A)Abs(I) 0.61x10-9 -10 1x10 1x10-11 0.4 -20246810 V (Volts) η = 3.6 Current (mA) V = 4.7 V 0.2 on R = 4256 Ω 0.0 s -4-3-2-101234567891011 Voltage (V)

Figure 52. I-V curve of a single 25 μm × 25 μm FPA pixel shown in linear scale. The

inset shows the same data in logarithmic scale.

Collecting I-V curves form a single central row of the array reveals the trends shown in

Figure 53 for the turn-on voltage, series resistance, and ideality factor. The ideality factor remains relatively constant across the array whereas the turn on voltage decreases and the series resistance increases slightly as the distance from the N-contact ring is increased. After the first

~20 pixels the increase in resistance levels off dramatically with the series resistance only increasing ~20% from there to the center of the array. This shows the excellent lateral conduction of the co-doped n-AlGaN conduction layer, despite its high Al composition.

122

60% V on 45% η R 30% s

15%

0%

-15%

Percent Change -30%

-45% 0 20 40 60 80 100 120 140 160 FPA Column

Figure 53. I-V statistics for the array, showing the good current spreading.

The photoresponse of the device was measured using a high intensity xenon arc lamp and monochromator as described in full detail in Appendix 2: Device Testing and Characterization.

The photoresponse under back illumination is shown in Figure 67 below. This device shows an

unbiased peak responsivity of 93.2 mA/W at 278 nm with a FWHM of ~10 nm, which

corresponds to a value of 42% for the external quantum efficiency of the device. The absolute

response drops three orders of magnitude from the peak into the near-UV region. No

contribution from the p-GaN can be seen in the spectral response curve.

123

100 10-1 Un-Biased Un-Biased Peak Responsivity 80 External Q.E. 42% @ 278.25 nm 10-2 93.2 ma/W @ 278 nm 60

-3 m Efficiency (%) 10 40

1x10-4 20 Responsivity (A/W) -5 1x10 0 200 225 250 275 300 325 350 375 400 Quantu External 200 225 250 275 300 325 350 375 400 Wavelength (nm) Wavelength (nm)

Figure 54. Unbiased responsivity in log scale (left) and the corresponding unbiased external quantum efficiency in linear scale (right) from a representative 25 μm x 25 μm pixel.

Unfortunately, the performance of these 25 μm x 25 μm detectors is slightly inferior to

the previously reported large area single element detectors based on the same material as

discussed earlier in this work. Those 1 mm x 1mm devices exhibited a record external

quantum high efficiency of 68%; by contrast, the same material fabricated into 25 μm x 25 μm

detectors only exhibits an external quantum efficiency of 42%: a 40% reduction in the device

performance. We expect the smaller devices to have on average fewer defects making the

performance far better, and we see this in the reverse leakage current of the devices. However

the smaller devices show a much larger series resistance: 4 kΩ for the pixels versus 200 Ω for

the large area devices. This larger series resistance is good for reducing the leakage current but

may hinder efferent carrier collection. In addition, when working with 25 μm x 25 μm

detectors the magnitude of the errors can be significant adding uncertainty to the 42%. For

example, even if the calibration aperture accurate to +/- ½ μm that still accounts for an 8%

124 variation in the area; and, if the sidewalls are not perfectly vertical the actual device area could

easily be 10% smaller than expected.

6.4. FPA Processing Overview

After analysis of the electrical characteristics of the FPA, the array needs to be hybridized

to the ROIC so that the imaging properties of the array can be investigated. However, before the array can be hybridized to the ROIC, further processing is necessary to prepare the FPA and

ROIC for bonding. To protect the array during flip-chip bonding SiO2 is first deposited on the surface and then selectively opened above each of the pixels. Then, using a novel lithographic process for lift-off of thick metal films155, indium bumps are deposited on the pixels of both the

ROIC and the FPA. The completed structure of the FPA, after deposition of the indium bumps,

is shown in Figure 55. The thus-prepared FPA and the ROIC are then bonded using a visible-

flip chip aligner, and then the indium bumps are reflowed to finalize the bonding. The finished

solar-blind imager is then loaded into a leadless chip carrier for testing in the camera system.

Figure 55. Electron micrograph of the focal plane array before bonding to the ROIC: after lithographic patterning, metallization, and deposition of the indium bumps.

125 The leadless chip carrier holds the array in the camera head. Immediately in front of the array is an aperture to block stray light. This is followed by a 280 nm band pass filter, and then a single 32 mm focal length UVGFS optic is used to collect the light and form the image on the

FPA. The complete specification of the camera system and imaging optics are provided in

Appendix 3: ROIC Specifications and Camera Interfacing

Ideally the band pass filter in this setup would be unnecessary; however, despite the III-

Nitride photodiodes being solar-blind, the Si based unit cells of the ROIC are not solar, or even visible blind. Without the filter a sufficient quantity of visible light would be incident upon the

ROIC. Since the III-Nitride FPA is transparent except for the small 15 μm × 15 μm p-type

contact pads, which block only 25% of the light. Attempts to overcome this limitation are

discussed at the end of this chapter in the section 6.6 Improvement of Device Performance.

126 6.5. FPA Images and Discussion

A

B

Figure 56. 320 × 256 Image of: A) a paper-cutout in front of a Germicidal lamp and B) an electric arc as seen with the solar-blind FPA camera.

127

Figure 57. Image of a human likeness as seen with the solar-blind FPA camera.

Because 280 nm radiation is not easily found in nature, the camera requires the construction of a man-made scene to image. A simple scene can be generated by using a florescent type short-wave UV lamp with a shadow mask placed in front of it to form shapes for the camera to image. Figure 56A shows an image of the letters “CQD”. It is also possible

to image arcing and coronal discharge, such as in diagnostics of high voltage equipment and is

one potential application of solar-blind imaging. Figure 56B shows the solar-blind image of the high-frequency electric arc generated by a small fly-back transformer. It is also possible to image the reflection of UV light off of a patterned reflective surface; aluminum is one of the few metals with good reflection in the solar blind region. Figure 57 shows a solar-blind image after reflection of a UV light off of a polished aluminum plate that has been selectively patterned with the image of a human likeness. (Full details of the imaging geometries used are provided in the Imaging Optics section of Appendix 3: ROIC Specifications and Camera

128 Interfacing ) Movies of taken from the above three images are also available in Appendix 4:

Sample UV FPA movies.

In all cases, the camera is operating with the detectors unbiased at a frame rate of 34 Hz; this corresponds to a frame integration time of ~29 ms. Using a 10 ff capacitor in the CTIA unit cells, this corresponds to a gain of ~1012 V/A. No bad pixel replacement has been performed on the image; the only image correction performed is a simple background subtraction. The images show good uniformity and contrast; the major defects include a limited number of dead pixels randomly scattered about the array, and one central scan line that does not operate. A histogram showing the good uniformity of the pixel response under moderate flat-field illumination is shown in Figure 58. Improvements in the processing and handling of the FPA should improve pixel yields and eliminate the presence of bad scan lines.

8000

7000

6000

5000

4000

3000 Counts 2000

1000

0 5.10 5.15 5.20 5.25 5.30 5.35 5.40 Voltage

Figure 58. Histogram of Pixel intensity for a flat field of UV illumination. This figure shows the variation in pixel response within the array.

129 6.6. Improvement of Device Performance

The ultimate goal of this thrust of the work is the realization of solar-blind focal plane

arrays. And to a large degree this has been achieved. It is impressive that we have been able to

realize an AlGaN-based back-illuminated solar-blind ultraviolet focal plane array operating at a

wavelength of 280 nm. Particularly unique is that this FPA camera was realized completely

within our university research lab: everything from the growth and processing of the p-i-n

photodiode array to the hybridization to the commercial 320×256 read-out integrated circuit.

However this is only a preliminary FPA, and for widespread adoption of the technology to

occur, further refinements will be necessary.

A great deal of time was spent optimizing the material though the growth and

fabrication of broad area devices before the first FPA was ever fabricated. The III-Nitride

material used in the fabrication of these FPAs is strong and consistently of good quality,

producing devices with external quantum efficiencies ranging from about 50 to 68 %.

However the first FPA test pixels only showed a 42% external quantum efficacy and this

indicates there may still be some room for improvement in the material and device processing.

In addition to the material and general processing, two major issues were discovered with the

fabrication of the current FPAs: the presence of a visible response due to the light incident

though the FPA onto the ROIC, and the slow response speed of the current FPAs as

demonstrated in the movies in Appendix 4: Sample UV FPA movies.

The fabrication of a UV focal plane arrays using commercial silicon-based ROICs has the undesirable side effect of visible light passing through the FPA and causing a visible response directly in the silicon of the ROIC itself. There are three strategies that can be used to

130 solve this problem and maintain the expected solar-blindness of the FPA: i) the ROIC can be

designed with specific shielding taken into consideration, ii) a thin metal shielding layer can be

inserted on the focal plane array (FPA) side to block transmission of the undesirable light

(Black out layer, Figure 14), or iii) an opaque underfill material can be injected between the

FPA and the ROIC to block this light while simultaneously improving the mechanical stability

of the assembly.

Modifying the ROIC is difficult and costly so instead we have chosen to focus on the

other two options. Research is currently underway to redesign the FPA so that it is completely

opaque and does not allow any visible light to reach the ROIC. This should allow the removal

of the 280 nm band pass filter and thus yield an 80% improvement in the external quantum

efficiency of the camera system. It will also reduce the complexity and weight of the camera

thus making it more desirable for solar-blind imaging applications.

Figure 59. Cross-sectional diagram of the current and proposed FPA designs. The top half shows the current design illustrating how a significant amount of light reaches the ROIC, the bottom half shows the proposed solution that will make the FPA 100% opaque.

131 We have already tested preliminary structures that mask out the ROIC completely. It involves imbedding a metal layer inside of the SiOx passivation layer as illustrated in Figure 59 above. We have successfully embedded opaque metal layers between SiOx passivation layers to

realize nearly opaque arrays as shown in Figure 60 below. However the added processing steps

increase the complexity of the array fabrication and need to be further refined before further

FPAs can be tested.

Figure 60. Top view of an FPA processed with a dark-field layer. Gold is the top contact, and the maroon area surrounding the top contacts in the black-out layer, it almost overlaps the

top contact and blocks out 99% of the area that would otherwise be exposed.

The far simpler option is to underfill the FPA with an opaque underfill material.

Underfill is typically an epoxy based polymer with a low viscosity that wets the void between the ROIC and the FPA and is drawn into the space via capillary action. There are to major categories of underfill: filled and unfilled. Unfilled underfills are usually transparent and are not well suited to the blocking of visible light. However, many commercial underfills are

132 loaded with a filler to help obtain a thermal expansion match to the silicon ROIC; in most cases this is a silica material or other inorganic compound with a particle size around 5 μm. This gives the material an opaque black or white color when cured. However the filler material has a finite size of the same magnitude as the FPA-ROIC spacing, and this makes the capillary filling technique problematic. In order to effectively underfill FPAs with a filled underfill it is necessary to increase the bump height. In the FPA reported above the bump height was on the order of 2 to 5 μm, making it unsuitable for underfilling. However using novel muti-layer photoresist techniques, and post deposition re-flow steps, we have realized indium bumps as tall as 15 μm, as shown in Figure 61 below. If deposited on both the ROIC and the FPA side, this will give a finished bonding height in the range of 20 μm. That should be more than sufficient for underfilling with an opaque filled underfill.

Figure 61. 15 μm tall indium bump after reflow, seen from an oblique angle.

133 7. Experimental Procedure: AlGaN Based Avalanche

Photodiodes

7.1. Introduction

In the previous sections we have discussed high quantum efficiency solar-blind photodetectors and focal plane arrays. In this section we take this research and extend it to the special case of realizing back-illuminated solar-blind avalanche photodiodes (APDs), and look in depth at the device structure and its growth, the processing steps and procedures, and provide an overview the basic characteristics of these devices.

Back illuminated detectors operating in the solar blind region are of special interest, however they do not necessarily provide the sensitivity required for all applications, as described in section 2.2 UV Photodetector Applications. Photomultiplier tube (PMT) based

UV detectors can still provide superior sensitivity to even the highest efficiency AlGaN based photodetectors. PMTs obtain their high sensitivity by taking advantage of internal gain

(typically ~106), however these detectors are bulky, fragile glass tubes that require large biases

(typically 1000 V) to operate effectively.156 Thus, it is highly desirable to have a smaller

semiconductor-based photodetector capable of realizing this level of sensitivity.157 The current

state of the art in AlGaN based detectors is illustrated below in Figure 62 relative to the other competing technologies.

134

Figure 62. Comparison of exiting UV detection technologies to AlGaN based detectors, showing the gap between the detectivity of PMTs and AlGaN based devices, and the need for

APDs (in red). (After references 158 and 159)

In semiconductors it is possible to obtain internal gain by taking advantage of avalanche multiplication under high electric fields. Unlike photoconductive gain earlier reported in

AlGaN based devices,160 avalanche gain is in principle capable of lower noise and faster response times thus increasing the sensitivity of these photodetectors. However prior to this work, there have been few reports discussing III-Nitride based APDs. All existing reports have focused on binary GaN detectors with a cutoff at 364 nm.161,,,162 163 164 The only discussion of true solar blind AlGaN-based APDs has been purely theoretical.165,,166 167 However we have

135 been able to realize solar-blind back-illuminated APDs with linear more avalanche gains in

excess of 1000.168,169

7.2. Material growth and device processing

7.2.1. Material Growth

The material growth and device structure are similar to that discussed in section 4 Wide

Band-Gap III-Nitride Material Growth. The devices were grown in an AIXTRON 200/4-HT

horizontal flow low-pressure metal-organic chemical vapor deposition (MOCVD) reactor.

Basal plane (00.1) sapphire was used as the substrate; it was double-side-polished prior to

growth in order to allow for the realization of back-illuminated photodetectors. Growth on the

sapphire substrate was nucleated with a thin 200 Å low-temperature AlN buffer layer. On top of this a 1 μm template consisting of high quality AlN was grown by atomic layer epitaxy170 at a temperature of ~1300°C.

On top of this template, an n-type lateral conduction layer consisting of Al0.5Ga0.5N:Si-In

was grown. The silicon-indium co-doping of this layer yields a carrier concentration of n ~ -

2×1018 cm-3 and mobility of μ ~ 60 cm2/V·s. This layer is several times more conductive than conventional singly-doped AlGaN:Si of the same aluminum composition. The use of indium also helps partially relax some of the inherent strain, and allows for the growth of a thicker crack-free conduction layer.

136 Ti/Au Thin Ni/Au p-GaN:Mg (500 Å)

p-Al0.36Ga0.64N (500Å) i-Al0.36Ga0.64N (2000Å)

Ti/Al Al0.5Ga0.5N:Si (650nm) Contact layer

AlN (850nm) LT- AlN buffer Sapphire (0001) For Back-Illumination For Back-Illumination

Figure 63. Schematic cross-section showing the device structure of the back-illuminated APD.

Table 4. Table of approximate carrier concentrations of the various layers used later in the modeling of this device structure.

The p-i-n active region consists of 100 nm singly-doped n-type Al0.45Ga0.55N:Si followed by a 125 nm undoped Al0.36Ga0.64N absorber region and a 50 nm p-type Al0.36Ga0.64N:Mg. This

is then followed by highly doped p-type GaN:Mg in order to take advantage of the p-GaN/p-

137 AlGaN heterostructure and help in the formation of ohmic p-type contacts.171 A schematic

diagram of the complete device structure is shown in Figure 63 above; approximate carrier

concentrations are given in the accompanying table. Inspection of the material after growth

reveals a smooth crack-free surface.

7.2.2. Device Processing

The samples are first annealed at 1000 ºC for 30 seconds, under dry N2, to activate the

magnesium in the p-type layers. Then thin layers of 30 Å Ni followed by 30 Å Au are

deposited, and annealed under ambient air at 500 ºC for 10 minutes, in order to form ohmic

contacts to the p-type material on top of the device. The material is then patterned into 25 μm ×

25μm square photodetectors using an electron cyclotron resonance (ECR) dry etching system.

A common n-type contact consisting of 300 Å Ti / 1800 Å Al is then deposited and annealed

under nitrogen to form an ohmic n-type contact to the Al0.5Ga0.5N:Si-In conduction layer. This

contact and the mesas are then covered with a final 400 Å Ti / 300 Å Pt / 1200 Å Au layer in

order to facilitate bonding after the annealing. The devices are then covered with 100 nm of

SiO2 to help protect the mesas and prevent premature breakdown of the devices. Windows were opened via wet etching. Three diodes out of the array is shown below in Figure 64.

138

Figure 64. Scanning electron micrograph of an APD after processing. The common n-

contact(not shown) is far removed from the mesas to avoid air breakdown of the devices.

7.3. Unbiased device performance

7.3.1. Current-voltage characteristics at low bias

First, the current-voltage (I-V) characteristics of these devices are investigated in the dark

under small applied bias. These small 625 μm2 devices have very low dark currents: up to about 10 volts the measurement is limited by the ~10 fA noise floor of the curve tracer used to record these I-V curves. In terms of current density this corresponds to the 1.6×10-8 A·cm-2

noise floor observed below in Figure 66 and on the right of Figure 65.

139

Figure 65. Comparison of Dark current density between 1mm x 1mm devices and 25 μm x 25 μm devices, showing the disparity in scaling of the dark current.

This dark current density is in sharp contrast to that of larger 1 mm2 devices fabricated

from the same material which typically showed much larger dark current densities, up to 1×10-3

A·cm-2, as show above in Figure 65. This scaling of the leakage current more strongly than just with the perimeter or area of the devices indicates that defects may play a significant role in the leakage current. This correlates with observations in GaN APDs that the devices require diameters smaller than about 50 μm in order to avoid defect related gain; it is expected that

AlGaN based solar-blind APDs may require devices even smaller to avoid defect related gain.

This makes prospects for realizing large area solar-blind APDs difficult without eliminating the

catastrophic defects that are almost guaranteed to occur in larger devices. One potential option

is to create a large fused array of small detectors, and simply allow the occasional small

detector with a catastrophic defect to burn out leaving behind a large effective area device.

140

2x104 ) 2 2x102

2x100

2x10-2

2x10-4

2x10-6

2x10Current Density (A/cm -8

-10-8-6-4-20 2 4 6 810 V (Volts)

-5 IV data -10 Linear Fit

-15 η = 2.70166 -20

ln (I) (A) (I) ln -25

-30

-35 -10 -5 0 5 10 Aplied Bias (V)

Figure 66. Left) Current-Voltage curve showing the current density in log scale. Right)

Natural log of current and fit to linear region used to extract an ideality factor (n) of 2.8.

The ideal diode equation, shown below, is used to fit the forward bias data.

qV CI ×= exp( ) Equation 26 f nkT

This fitting allows the extraction an ideality factor (n), which can provide insight on the

conduction mechanisms operating in these photodetectors. The ideality diode equation arises

141 from the combination of the equations for diffusion current and recombination current: the two

currents that usually dominate the diode current. The ideality factor is expected to have a value

between 1 and 2. Whereas if n is closer to 1, then diffusion current dominates and if n is closer to 2 then recombination current dominates. The linear fit shown above in Figure 66 yields an ideality factor of n = 2.7 for this diode. This falls slightly outside of this expected range, but is closer to 2 than most previous devices owing to the high quality of this p-n junction, and minimal contribution from other conduction processes. In addition to the ideality factor a series resistance of 14 kΩ can be extracted; this value is so large primarily due to the small area of the devices, and the large lateral distance to the common n-contact.

7.3.2. Photoresponse

The unbiased photoresponse of these devices was measured using a high intensity xenon arc lamp attached to a monochromator as described in Appendix 2: Device Testing and

Characterization. After calibration of the system, the device was illuminated from the back,

through the substrate using the fiber-optic sample stage and probe-station. The resulting

photoresponse is shown in Figure 67.

This device shows an unbiased peak responsivity of 103 mA/W at 272 nm with a FWHM of ~10 nm, which corresponds to a value of 48% for the external quantum efficiency. The absolute response drops three orders of magnitude from the peak into the near-UV region. No contribution from the p-GaN can be seen in the spectral response curve. This confirms that the photoresponse is from the AlGaN p-i-n structure and not a Schottky junction between the p-

GaN and the metal contact.

142

55 Un-Biased Un-Biased 50 100 Ext. Quantum Eficiency Peak Responsivity 45 103 mA/W 40 48% @ 271 nm 10 @ 272 nm 35 30

25 1 20 15 0.1 10

Responsivity (mA/W) Responsivity 5 0 200 250 300 350 400 450

200 250 300 350 400 450 (%) Quantum Efficiency External Wavelength (nm) Wavelength (nm)

Figure 67. Left) Unbiased responsivity shown in log scale. Right) External quantum

efficiency shown in linear scale.

7.4. Avalanche Mode device operation

7.4.1. Current-voltage curves under bias

The I-V curves of these devices were then recorded under reverse biases up of 170 volts.

After about 20 volts the dark current comes out of the noise floor and begins to increase. The same measurement under illumination with broadband white light from the xenon lamp shows a large photocurrent that increases moderately with bias up to about 40 volts and then begins to undergo gain, and rapidly increases with further increasing reverse bias. Both the dark and illuminated I-V curves are plotted in log scale below in Figure 68. These measurements are

generally non-destructive, and the same device is consistent from measurement to

measurement. The figure below includes error bars from 6 alternating measurements in the

143 illuminated and dark conditions. It can be seen that at all voltages there is a clear separation

between the dark current and photocurrent.

1x10-3 1x10-4 Dark Current 1x10-5 White Light Illumination 1x10-6 1x10-7 1x10-8 1x10-9 1x10-10 -11 1x10Current (A) 1x10-12 1x10-13 1x10-14 0 20 40 60 80 100 120 140 160 Reverse Bias (V)

Figure 68. Current-voltage behavior as a function of applied reverse bias, both under illumination and in the dark. Errors bars have been added to indicate the variation over 3 consecutive pairs of alternating light-dark measurements of the same diode.

The avalanche gain (M) is defined as the difference between the primary multiplied current and the multiplied dark current, normalized by the difference between the primary unmultiplied current and the unmultiplied dark current. The gain can be calculated from Figure

68 above using Equation 27 below.

dIlluminate dark (V)I-(V)I M = Equation 27 dIlluminate dark (0)I-(0)I

144 The unmultiplied currents are taken from the flatter portion of the curves prior to the onset of avalanche gain. The calculated photocurrent and the corresponding gain (M) are shown below in Figure 69.

1x10-5 Peak Gain: 1000 10-6 1485 @ 170V

100 -7

10 Gain

10 10-8

1 -9

PhotoCurrent (A) PhotoCurrent 10

0.1 10-10 0 20 40 60 80 100 120 140 160 Reverse Bias (V)

Figure 69. Calculated photocurrent (left axis) and corresponding gain (right axis) from

the data of Figure 68. Error bars indicate +/- 1 standard deviation. Breakdown is taken at

40 volts.

These devices show a soft avalanche breakdown starting at ~40 volts. The gain then increases exponentially until saturating at a gain of greater than ~1500 at a reverse bias of ~170 volts. These devices never develop a sharp Geiger mode breakdown. This is in contrast to the results typical of GaN APDs where the gain tends to holds off until larger voltages and then rapidly increases exhibiting a hard Geiger mode breakdown. Results of Geiger mode APD are discussed in more detain in chapter 0

145

FutureX Work: Geiger Mode APDs.X

This lack of a sharp breakdown can be explained partially in terms of the ionization coefficients. In GaN APDs the ionization coefficients for holes and electrons are believed to be

172 173 similar at the breakdown field strengthTPD DPT and we have confirmed this experimentallyTPD DPT. This means that both carriers contribute to the gain and help to support the hyper-exponential Geiger mode breakdown of these devices shortly after the onset of gain. However, this is not necessarily the case for all materials, and if the ionization coefficients differ significantly, only a single carrier may support gain at the electric fields present within the device. If this is the case in AlGaN APDs then the gain is expected to increase geometrically and would never show a Geiger mode breakdown. This may be the cause of the soft avalanche breakdown and absence of a hard ‘breakdown voltage’ in our devices. A schematic diagram is shown below of sequential breakdown, typical of Geiger mode, and parallel breakdown, expected for these devices. Even though parallel breakdown precludes the use of these devices a single photon counters, it can be beneficial to the noise performance of the devices operating in linear mode.

A significant improvement to the device performance can be realized by implementing a separate amplification and multiplication (SAM) structure that would ensure that only the gain

dominating carrier is injected in the multiplication region. As shown in FigureX 70 X below there is significantly less temporal dispersion of the avalanche pulse in the parallel case, and the orderly parallel progression reduces the influence of noise resulting from the statistical nature of avalanche breakdown.

146

Figure 70. Schematic diagram of parallel and sequential ionization. Sequential

ionization (left) can lead to Geiger mode operation, however parallel (right) restricts the

multiplication to a geometric increase.

7.4.2. Device Modeling

174 Using a 1D finite element modelTPD DPT, we investigated the electric field build-up in the multiplication region. The doping concentrations used in the model were given with the

structure in TableX 4.X These doping concentrations are estimates inferred from Hall measurements of independent layers used for calibration of the structure; however the actual active carrier concentrations in the device as grown may differ slightly. This model also neglects the piezoelectric fields due the strained layers. Plots of the evolution of the modeled

electric field profile are shown below in FigureX 71.X

147

200 V 190 V 6.5 180 V 6.0 170 V 160 V 5.5 150 V 140 V 5.0 130 V 120 V 4.5 110 V 4.0 100 V 90 V 3.5 80 V 3.0 70 V 60 V 2.5 50 V 2.0 40 V 1.5 30 V 20 V Electric Field (MV/cm) 1.0 10 V 0.5 0 V 0.0 0 50 100 150 200 250 300 350 400 Position (nm)

Figure 71. Electric Field Profile under various applied reverse biases.

From this model it is possible to estimate the electric field at the onset of gain to be only

~2 MV·cm-1 which increases to ~6 MV·cm-1 at 170V where the maximum gain is reached.

However, it is expected that the peak electric field may be significantly larger due to inhomogeneity of the doping, and ionized defects. This model only considers the average electric field in the plane of the device. This model also reveals that the electric field is not well confined to the gain region. This is due to the difficulty of achieving highly doped

AlGaN. This means that the peak electric field remains rather small in spite of the large applied reverse bias. It also means that electric field is high enough to support gain outside of

148 the intended multiplication region. The effective multiplication region is closer to 175 nm

wide rather than the designed 125 nm.

Physically, avalanche gain is a function of the ionization coefficient (α) and multiplication width (W). This behavior can be fitted using an empirical equation that models

175 the ionization coefficient with two fitting parameters α0 and C.

⎧ ⎧− C ⎫⎫ WM α exp)exp( ⎨W α 0 ⋅⋅=⋅= exp⎨ ⎬⎬. Equation 28 ⎩ ⎩ F ⎭⎭

The multiplication region width is chosen as 175 nm rather than the designed 125 nm in response to the electric field profile model above. The electric field (F) is determined from the model as well. This model provides a reasonable fit to the avalanche gain data with the fitting

8 parameters C = 5.9 MV/cm, αo = 1.2 × 10 . Figure 72 below shows good agreement of the

avalanche gain model fit (red) and experimental data (black).

Avalanche Gain Gain Model Fit 1000

100

10 Model: Avalanche Gain Gain (M) Gain W 175 nm α 1.20833 E8 o C 5.919537 E6 1

1234567 Electric Field (MV/cm)

Figure 72. Avalanche gain model of the device. The solid curve shows the

experimental data, the dashed curve shows the model.

149 7.4.3. Ruling out other origins for the gain

The diodes fit well with an avalanche gain model; however it is worthwhile to rule out

other possible origins of the gain observed in these devices. There are three main gain

mechanisms that could be responsible for the large gain observed in these devices: avalanche

multiplication, Zener tunneling, and photoconductive gain.

In order to rule out Zener tunneling the temperature dependence of the dark current gain

in these devices is investigated. The temperature-dependent dark current shows a large positive

temperature coefficient, and a thermal activation energy of ~240 meV can be extracted from

this evolution. This strong temperature dependence observed rules out the temperature-

insensitive process of Zener tunneling as the dominant gain mechanism. Photoconductive gain can also be ruled out based on the shape and magnitude of the gain observed. If photoconductive gain were the dominant gain mechanism, then the gain would be expected to

176 increase linearly with voltage.TPD DPT Instead, in FigureX 69,X it can be seen to increase exponentially.

This allows us to rule out Zener tunneling and photoconductive gain, and together with the avalanche gain model fit, it allows us to attribute the gain to avalanche multiplication of carriers inside the device.

7.5. Conclusion

We have demonstrated the first AlGaN based solar-blind avalanche photodiodes, with a gain of around 1500 at 170 volts of reverse bias. This gain has been attributed to avalanche multiplication of the photogenerated carriers within the device, and shows good agreement with that model. Modeling of the electric field build up indicates that the maximum gain occurs at a

150 peak electric field strength of 6.0 MV/cm, with the onset of gain occurring at 2.0 MV/cm.

These devices do not show a Geiger mode breakdown and will not be suitable for photon counting operation. However, these devices show the potential of AlGaN based APDs, and with further improvement of the material, and sufficient device optimization, it may be possible to achieve devices with sensitivities approaching those of commercial PMTs.

151 8. Experimental Procedure: GaN Based Avalanche

Photodiodes

8.1. Introduction

The back-illuminated solar-blind APDs of the previous section are a significant accomplishment; however, to better understand avalanche multiplication it is worthwhile to go back and study GaN APDs in more detail. The high-quality AlN templates developed as part of this work also lead themselves well to the development of back-illuminated GaN APDs, an unpursued endeavor that will be required for the eventual realization of APD arrays. Back illumination also allows for more in-depth study of the device operation since structure can be designed to allow either front or back illumination of the device. GaN APDs also sever as the logical starting point for the realization of Geiger mode APDs, as will be explored in more detail in the next chapter.

Since 2000, there have been several groups reporting the development and characterization of front-illuminated GaN and AlGaN APDs 177, 178, 179 In these devices, front- illumination yields multiplication primarily dominated by electron-initiated ionization, since high quality III-Nitride growth typically requires that p-layers are grown at the top of p-n and p-i-n structures. However, theory has predicted a significantly higher hole impact-ionization coefficient,180 and hence higher multiplication factors and lower excess noise are expected for back-illuminated GaN p-i-n diodes, contrary to most other III-V semiconductors. Back- illumination also allows easier integration and packaging layouts through flip-chip technology as required for the production of APD arrays.181 However, although a few works have assessed

152 back-illumination in AlGaN APDs,182, 183 none have done so in GaN APDs. The main issues for UV back-illuminated GaN APDs are the limitation of absorption in the GaN below the active region, and the problem of the interface quality between the GaN device structure and the transparent template layer. In this chapter, we report on the growth, fabrication, and characterization of back-illuminated GaN APDs on thick AlN templates. Comparison of the performance of these same devices under front- and back- illumination allows us to reach a better understanding of carrier multiplication in this material and to determine experimentally both electron and hole ionization coefficients. In addition, all the reported structures were grown on more conventional GaN templates for comparison to standard front-illumination.

Multiplication noise characteristics at the onset of the avalanche breakdown are also discussed.

8.2. Material Growth

p-i-n samples were grown in an Aixtron 200/4-HT metal-organic chemical vapor deposition (MOCVD) reactor on double-sided polished sapphire substrates. The AlN template layer growth is similar to that discussed in chapter 4Wide Band-Gap III-Nitride Material

Growth. It consists of 500 nm of high-quality AlN grown by atomic layer epitaxy atop a 20 nm

low-temperature AlN nucleation layer. To realize GaN based devices on this template, it was

necessary to developing the growth of high-quality GaN directly atop AlN. However, for

comparison, samples were also grown on 1.5 μm thick GaN templates grown atop low

temperature GaN buffer layers.

153

Figure 73. Atomic force microscopy (AFM) imaging of the surface of a high-quality

GaN layer grown on an AlN template.

To develop the growth of high-quality GaN on AlN templates, a series of thick test samples were grown and characterized. These test showed GaN with a smooth surface and very few observable dislocations, as shown by AFM in Figure 73 above. The RMS roughness

of this surface was only 1.29 Å for a 5 μm x 5 μm scan area. X-ray diffraction revealed a

narrow (00.2) symmetric peak with a line-width of ~82 arc seconds as shown in Figure 74 .

Room temperature PL of these layers also exhibited a narrow peak at 361 nm with a line-width

of ~40 meV and no observable yellow defect related luminescence. Hall mobility measurement

of undoped GaN layers yielded a mobility of 450~500 cm2/V·s with an unintentional n-type carrier concentration of 1016 cm-3. Intentionally doped GaN layers yielded carrier

concentrations of 1~2×1018 cm-3 for both n- and p-type layers.

154

GaN 10k

1k AlN Al O 2 3 CPS 100

10

1 17 18 19 20 ω/2θ (degrees)

Figure 74. High-resolution x-ray diffraction of GaN on an AlN template layer.

These calibrated GaN epilayers were used as building blocks for the back-illuminated

GaN p-i-n avalanche photodiodes reported here. The p- and n-GaN layers consist of Mg- and

Si-doped GaN, respectively, while the intrinsic region consists of un-intentionally doped GaN.

In this study a variety of different device structures were grown with layer thicknesses ranging from 100 to 640 nm for the p-type region and from 100 to 400 nm for the n-type region; the thickness of the intrinsic layer was varied from zero to 200 nm. Typical device structures for the devices grown on GaN and AlN template are shown below in Figure 75.

155

Figure 75. Schematic diagram of GaN APDs on AlN templates (left) and GaN templates

(right).

8.3. Device Processing

The samples are first annealed at 1000 ºC for 30 seconds, under dry N2, to activate the magnesium in the p-type layers. Then thin layers of 30 Å Ni followed by 30 Å Au are deposited, and annealed under ambient air at 500 ºC for 10 minutes, in order to form ohmic contacts to the p-type material on top of the device. The material is then patterned into 25 μm ×

25μm square photodetectors using an electron cyclotron resonance (ECR) dry etching system.

A common n-type contact consisting of 400 Å Ti / 1200 Å Au is then deposited to form an ohmic n-type contact to bottom n-GaN conduction layer. This contact and the mesas are then covered with a final 400 Å Ti / 1200 Å Au layer in order to facilitate probing the device. The devices are then covered with 300 nm of SiO2 to help protect the mesas and prevent premature

156 breakdown of the devices. 10 µm x 10 µm windows were opened on top of the devices using wet etching. A schematic diagram of the processed device is show below in Figure 76.

Figure 76. Schematic diagram of the APD as processed.

8.4. Device Results

8.4.1. I-V measurements

Current-voltage measurements were made of the p-i-n diodes after fabrication. The series resistance of the diodes was calculated from the I-V characteristics under forward bias.

Values between 465 and 3150 ohms were found for samples grown on AlN templates while the same exact structures grown on an GaN:Si/GaN template exhibited values between 29 and 190 ohms. The series resistance in all cases is dominated by the thickness of the n-GaN layer, increasing linearly with the inverse of it. This indicates that the conduction along the n-

GaN/AlN interface is not a significant contribution for back-illuminated avalanche photodiodes. The large disparity between the series resistance of the devices grown on GaN and AlN is due to the large difference in effective n-type layer thicknesses due to the extra conduction of the GaN:Si/GaN template. The devices on AlN template had between 100 and

157 600 nm of conductive GaN underneath, while the same device on GaN had 2.1 to 2.6 μm of

highly doped n-GaN. In most cases, it would be desirable to have as small a series resistance

as possible, as long as the leakage current of the device does not increase. However, there is a

significant trade off between the thickness of the underlying GaN used to decrease the series resistance, and the back-illuminated performance of the device; since this conduction layer is highly absorbing and does not contribute to the photoresponse high efficiency detection

requires a minimum of thickness. Another option that could be pursued is a heterojunction

APD that uses a wider bandgap material for the n-type layer, similar to what was done in the case of AlGaN APDs.

The breakdown characteristics were observed to vary as function of the different layer

thicknesses. By measuring the breakdown voltage as a function of the i-region thickness, it is

possible to experimentally extract a value for the critical electric field strength in these devices.

Figure 77 below shows Vbd as a function of i-region thickness for all the samples under study.

From fitting the data with a simple abrupt junction model (assuming the entire potential is

dropped across the intrinsic region), a critical electric field of 2.73 MV/cm is obtained; this is

close to the previous values reported in the literature.184

158

Breakdown = 2.73 MV/cm 100

75

50 Breakdown Voltage (V) Voltage Breakdown

0 50 100 150 200 i-GaN thickness (nm)

Figure 77. Breakdown voltage as a function of intrinsic layer thickness used to extract a

critical electric field strength of 2.73 MV/cm.

In addition to the above experimental determination, the electric build up was modeled in

a device with a 200 nm thick intrinsic region. This was done using the same 1D finite element

185 model of the device as discussed in the previous section.TPD DPT The same device was then grown

and tested, showing a breakdown voltage at 102 V. From the model, as shown below in FigureX

78 X this corresponds to an average electric field in the multiplication region of ~3.2 MV/cm.

This is ~20% larger than found by experiment. This discrepancy is most likely due to the

model overestimating the confinement of the electric filed. In reality the interfaces are less

demarcated and the multiplication region may have a higher background doping levels due to a

faint memory effect from growing the underlying n-GaN layer.

159

6.0 200 V 5.5 190 V 180 V 5.0 170 V 4.5 160 V 150 V 4.0 140 V 130 V 3.5 120 V 3.0 110 V 100 V

2.5 90 V 80 V 2.0 70 V 1.5 60 V 50 V 1.0 45 V 0.5 40 V

Electric Field (|MV/cm|) Field Electric 30 V 0.0 20 V 10 V 0 50 100 150 200 250 300 350 400 0 V Depth (nm)

Figure 78. Model of the electric field profile across the device as a function of the applied reverse bias: experimentally breakdown occurred at 102 V corresponding to 3.2

MV/cm in the model.

The remainder of this discussion will focus on the following three samples: samples A, B,

and C, as shown in TableX 5 X below. Samples A and B have different layer thicknesses. Sample

C has a structure identical to sample B and was grown in the same growth, however sample B

was grown on an AlN template and sample C was grown on an GaN:Si/GaN template.

n-GaN i-GaN p-GaN VbdB VonB B n RsB (nm) (nm) (nm) (V) (V) (Ω) A 100 200 100 102 6.76 2.99 2849 B 400 150 285 78 6.04 3.23 522 C 400+2000 150 285 78 3.69 2.68 59.7 Table 5. Table of device structures and limited device characteristics.

Illustrative reverse bias current-voltage (I-V) characteristics are shown below in FigureX

79 X for samples A, B, and C. For samples A and B, the breakdown voltage (VbdB )B reduced from

186 102 V to 78, as a consequence of the narrower intrinsic region.TPD DPT ThisP is consistent with the

160

trend observed in FigureX 77;X other experiments rule out the effect of the thickness of the p-type or n-type layers on the breakdown voltage. Samples C & B shown the same breakdown voltage, as expected since they have the same device structure (what little difference there is, can be attributed to a ~1 V offset in the breakdown induced by the difference in series resistance). This shows the minimal effect of the p- and n-region thicknesses on the breakdown voltage. The main differences in the I-V characteristics of samples B and C are the steepness of the breakdown and the leakage current before the onset of breakdown. Sample C has a breakdown that is significantly steeper than that of sample B. This is due primarily to the increased series resistance of sample B. This effect is much more pronounced in sample A where the breakdown in noticeable soft. By comparing the slope for all three samples, we can extrapolate out the series resistance and come up with a differential resistance for the avalanche breakdown process of ~2.15 kΩ.

161

1x10-2 1x10-3 1x10-4 1x10-5 1x10-6 1x10-7 -8

1x10 BAC 1x10-9 1x10-10 10-11 Dark current (A) 10-12 10-13 10-14 0 20406080100 Reverse Bias (V)

Figure 79. Breakdown characteristics of samples A, B, and C are shown. Inset: The

experimental breakdown voltages obtained for different thicknesses of the intrinsic layer.

The other main difference between the I-V characteristics of the samples is the leakage current at the onset of breakdown. Sample B and C are very similar. We would expect sample

C to have the higher leakage current due to its lower series resistance, however the quality of the thick GaN template is slightly better than that of the thinner AlN template resulting in a lower overall leakage current for the sample on the GaN template. This observation illustrates the tradeoff involved with designing devices for back illumination, and indicates that there is still room for improvement of the AlN template layer quality. However contrary to this, sample A has by far the lowest leakage current even though it has a significantly higher breakdown voltage. This is due to the significantly larger series resistance of this device.

162 Having a high series resistance will be a handicap for the fabrication of devices operating in

Geiger-mode where a steep breakdown in required; but in linear mode, this can help to keep moderate levels of dark current above breakdown and provide for higher stability of the device.

Dark current of sample A is plotted below in FigureX 80 X as a function of time; after more than 60

h of continuous operation 2 V above the breakdown voltage the standard deviation of the

current is only 6%.

100

10

Dark Current (pA)

1 0 6 12 18 24 30 36 42 48 54 60 66 Elapsed Time (hours)

Figure 80. Variation of dark current of a GaN APD biased 2V above breakdown. The standard deviation is less than 6% over more than 60 hours.

8.4.2. Gain Measurements

Photocurrent measurements of samples A and B were performed under back- and front-

2 illumination with a frequency-doubled argon ion laser at 244nm (optical power=102 W/cmP ).P

The measurement setup is described in full detail in Appendix 10.2.4X X GainX Measurement.X Due

163 to the need to probe the device it is not possible to ensure that the same incident power is used in front illumination. However, the shadowing effects of the metal contacts and probes are only a minor source of error for the gain of these devices.

Being able to measure the devices in both configurations allows us to selectively inject primarily holes or electrons into the active region to selectively initiate the multiplication process. Multiplication factors were calculated from these measurements, and are shown below

in FigureX 81.X As observed, back-illumination provides significantly higher gain and a lower breakdown voltage than front-illumination. This case is reproduced for both samples A and B: in fact, the same result was obtained in all the p-i-n samples tested, regardless of the structure.

Ionization events start at voltages above 50V for holes and 70V for electrons in sample A, and

35V and 65V, respectively in sample B. This agrees with theory’s prediction of a higher hole

impact-ionization coefficient in GaN. X180 This also illustrated a significant performance advantage that can help motivates the decision to design devices for a back illumination geometry. Impact ionization coefficients also play a role in the noise performance of APDs as will be discussed in the next section.

164

Sample A Sample B

M (Back) M (Back) 100 p 100 p M (Front) M (Front) n n

10 10

Gain Gain

1 1

0 20406080100 0 20406080100 Reverse Bias (V) Reverse Bias (V)

Figure 81. Multiplication factors for electrons (MnB B) and holes (MpB B) obtained from sample

A (left) and sample B (right).

Based upon the multiplication curves shown above in FigureX 81 X and assuming that back- illumination corresponds to solely hole initiated ionization, and front illumination corresponds to solely electron initiated ionization, it is possible to extract experimental values of the electron and hole impact ionization coefficients in GaN. Ionization coefficients vary as a function of the electric field; the electric field varies across the device, and as a function of the

reverse bias as shown by the model in FigureX 78.X To determine the Ionization coefficients, the

electric field is assumed constant across the multiplication region, and directly proportional to

the reveres bias. The multiplication width is assumed to be the same as the i-region thickness

(W) giving the following simplified relation for electric field: E=V/W. The ionization

coefficients for holes (βpB )B and electrons (αnB )B can then be calculated from the multiplication

187 factors for electrons (MnB )B and holes (MpB )B TPD DPT as:T

165 1 ⎛⎞M ()1VMV− ⎛⎞ () β ()ELn= pp p ⎜⎟⎜⎟ WMV⎝⎠pn()− MV ()⎝⎠ MV n () (29) 1 ⎛⎞⎛⎞M ()1VMV− () α ()ELn= ⎜⎟⎜⎟nn n WMV⎜⎟⎜⎟()− MV () MV () ⎝⎠⎝⎠np p.

The resulting ionization coefficients are then plotted in FigureX 82 X by below. This shows a large gap between the electron and hole ionization coefficients with reasonable agreement at low electric fields. However we have found that the electron and hole ionization coefficients

follow the same trend as a function of electric field, contrary to what the theory had proposed. X180

The origins of this discrepancy are still under investigation.

) -1 5 β (theory) 10 p β α (theory) p n 104

α 103 n

102

1

Ionization coefficient (cm 10 0.20 0.25 0.30 0.35 0.40 0.45 0.50 0.55 1/E (x10-6 cm/V)

Figure 82. Solid lines: ionization factors obtained for electrons (αnB B) and holes (βpB B) from

experiment. The dashed lines represent theoretical values for βpB B and αnB ,B as extracted from ref

180X .X

166 In addition to recording the gain as a function of the applied reverse bias, we have also

investigated the spectral response of these devices, as shown in FigureX 83 X below. In unbiased operation the devices have a peak responsivity of 82 mA/W at 361 nm, corresponding to an external quantum efficiency of only ~30%. Depending upon the thickness of the bottom n-GaN layer between most of the light is absorbed outside of the depletion region: even in the best case scenario of 100 nm of n-GaN, only 70% of the light actually reaches the intrinsic region.

Under moderate reverse bias the signal increases slowly at all wavelengths less than 360 nm: this may be due primarily to an expansion of the depletion region with increasing reverse bias.

Upon reaching the breakdown voltage at ~65 V the response can be seen to increase more rapidly. For illustrative purposes, the progression of the response at three different

wavelengths is shown on the right of FigureX 83.X

3 10

75 V 102 65 V 55 V 45 V 1 361 nm 10 35 V 100 25 V

15 V 0 10 5 V 325 nm 0 V

-1 10 Responsivity (mA/W) Responsivity (mA/W) Responsivity 10 275 nm 250 275 300 325 350 375 400 425 450 -55 1525354555657585 Wavelength (nm) Reverse Bias (V)

Figure 83. Evolution of the spectral response of a GaN APD near breakdown. Spectra

are shown on the left, and the evolution at three selected wavelengths are shown on the right.

167

8.4.3. Origins of the observed multiplication

The diodes follow an avalanche gain model, however it is worthwhile to rule out other possible origins of the gain as we did with the AlGaN based devices. There are three main gain mechanisms that could be responsible for the large gain observed in these devices: avalanche multiplication, Zener tunneling, and photoconductive gain.

In order to rule out Zener tunneling the temperature dependence of the breakdown voltage in these devices is investigated. The breakdown voltage obtained at room temperature

was in good agreement with the 2.73 MV/cm critical electric field reported earlier [192X ].X We

see a clear linear increase of the breakdown voltage with respect to temperature as expected for

impact ionization where hot carriers are scattered by thermal phonons. As shown in FigureX 84 X below, a positive thermal coefficient of 8.7 mV/K can be extracted. This value is on the same

188 order as reported in the literature.TPD DPT This strong temperature dependence observed rules out the temperature-insensitive process of Zener tunneling as the dominant gain mechanism.

10-2 Increasing T 77.5 o o Linear Fit 23 C to 290 C 77.0 -3 8.7 mV/K 10 76.5 1x10-4

76.0

-5 1x10 75.5 Current (A) 10-6 75.0 Break Down (V) Voltage -7 74.5 10 0 50 100 150 200 250 300 66 68 70 72 74 76 78 80 ο Voltage (V) Temperature ( C)

Figure 84. Evolution of the device breakdown voltage with temperature.

168 Photoconductive gain can also be ruled out based on the shape and magnitude of the gain

observed. If photoconductive gain were the dominant gain mechanism, then the gain would be

189 expected to increase linearly with voltage.TPD DPT Instead, in FigureX 84,X it can be seen that for all

temperatures, the gain increase hyper-exponentially. This allows us to rule out Zener tunneling

and photoconductive gain, and attribute the gain to avalanche multiplication of carriers inside

the device.

8.4.4. Noise Analysis

Noise measurements of samples A and B were performed in darkness and under front- and back-illumination. For illuminated measurements, the sample was illuminated with 340

2 nm filtered light from the Xe lamp (11 W/mP ).P The signal from the detector was amplified with

7 a low noise trans-impedance amplifier with 1×10P P V/A before analysis using a FFT spectrum

analyzer with a 100 KHz bandwidth as described in section 10.2.5X X NoiseX Measurement.X The

-27 2 noise floor of this instrumentation was in the 10P P AP /HzP range. Results for sample A under

back-illumination are shown in FigureX 85.X It is noticeable that at low frequencies 1/f noise

dominates the noise characteristic. Spectral Power Density (SnB )B for 1/f noise followed the

2 γ -9 relationship SsIfn = 0 , with average s0B B and γ values of 1.4x10P P and ≤0.33 ± 0.05,

respectively for sample A. In Sample B, γ=1.14 ± 0.09 and s0B wasB at least three orders of

magnitude higher which suggests that the thin n-GaN plays a significant role in noise reduction.

The γ value obtained for sample A is among the lowest values ever published in GaN. This

190 agrees with previous reports of a strong reduction of 1/f with thin n-GaN layers.TPD DPT

169 Sample A

-16 Back-illuminated @ 340 nm

1x10/Hz) 2 -17 10 102V -- 1V per step 10-18 1x10-19 -20

1x10 1x10-21 10-22 1x10-23 -24 1x10 91V -25

Spectral Power Density (A Spectral Power Density 10 10 100 1000 10000 Frequency (Hz)

Figure 85. The Spectral Power Density (Sn) of sample A at the onset of breakdown is

shown from 91 V to 102 V with 1 V steps. The two narrow spikes at 60 and 120 Hz correspond

to line noise.

At the onset of avalanche breakdown, a white noise contribution becomes dominant in

the medium frequency range (1 KHz-30 KHz). To identify the origin of that noise, SnB wasB

plotted against current for front- and back-illumination (FigureX 86).X The rapid increase of the noise level with current suggests that it is likely to be excess multiplication noise. The excess

2 noise factor (F) was calculated from F = Sn 2qIM , where q is the electron charge, I is the

current under front- or back-illumination and M is the multiplication factor (inset of FigureX 86).X

The results show that the excess multiplication noise factor is significantly lower under back-

illumination, confirming again the higher impact ionization coefficient for holes in GaN.

170

1x10-19 /Hz) 2 Front- 4 Front- 1x10-20 10

1x10-21 Back- 103 10-22

-23 1x10 102 Back- 1x10-24 101 10-25 Excess Noise Factor Excess Noise -26 10 0 -9 -8 -7 -6 10 Spectral Power Density (A Spectral Power 10 10 10 10 1 10 100 Total Current (A) Gain

Figure 86. Left) Spectral power density is plotted as a function of total current for sample

A under front- (triangles) and back-illumination (squares). Right) calculated excess noise factors for front- and back-illumination.

8.5. Conclusion

In summary, we have realized the first back-illuminated GaN based APDs. These devices have demonstrated avalanche multiplication with gains ranging from at little as 10 to more than

1500. Because of the back-illuminated design of these devices we were able to measure under both front and back-illumination to extract information on the ionization of electrons and holes in GaN. We observed a strong preference for hole impact-ionization, and were able to extract experimental value for the electron and hole ionization coefficients in GaN. These numbers agree fairly well with the predictions made in the literature.

We also investigated the noise characteristics of back-illuminated GaN APDs as compared to those of front-illuminated APDs. This indicates that the hole-initiated

171 multiplication (from back illuminating the device) yields gain and noise characteristics with superior performance. The same device showed almost an order of magnitude more gain in back illumination, and an excess noise factor at the same gain that was more than three orders of magnitude less. This work lays the foundation for the development of Geiger mode APDs, as discussed in the next chapter.

172 9. Future Work: Geiger Mode APDs

9.1. Introduction

(Al)GaN has demonstrated itself as a promising semiconductor material system for the

development of ultraviolet detectors, and has been capable of realizing a variety of detectors to date. Most interesting is the recent research into avalanche photodiodes. APDs are highly

desired for their increased sensitivity, and they are growingly establishing themselves as viable

candidates for reliable UV detection where there is a need for low light level detection in the

UV. However, the next step for this research is the realization of single-photon counting capabilities in the UV through the use of Geiger mode APDs. Geiger mode is favored over linear mode for its photon counting capabilities in low light situations. Geiger mode offers much higher sensitivity, and has the advantage of the avoiding the issue of excess multiplication noise by taking a statistical approach to photon detection. These devise are

dependant only upon the dark count rate and the detection efficiency, as outlined in section 3.4X X

AvalancheX Photodiode Parameters.X

The first Geiger mode APDs based upon GaN were reported in 2001 by S. Verghese et.

191 al.,TP D DPT P but there have not been any reports since then. P ThoseP preliminary devices employed a

GaN p-i-n structure grown on 7 μm thick HVPE grown GaN template layer. They had a 250 nm multiplication region, and their device showed a breakdown voltage of around 80 Volts. In linear mode, the devices had an external quantum efficiency of ~25%, and a maximum gain of only around 10. Using a passive quenching circuit and a 10 nS pulsed excitation strategy, they were able to realize Geiger mode operation of their devices, with a maximum photon detection

173 efficiency of 13.5%, and a dark count rate of 400,000 per second. This allowed them to

measure photon fluxes of 20 photons per pulse.

Along the same lines as that research, we have already begun to develop GaN APDs

based upon our linear mode GaN APD results. Our novel AlN template based GaN APD

structure allows for operation of our device under either front or back illumination. Theory

and experiment indicate that hole-initiated multiplication in GaN yields gain and noise

characteristics with superior performances; and back illuminating standard p-i-n GaN diodes

192,193 devices would favor such hole-initiated multiplication.TPD DPT TD DTP

As such, our devices have demonstrated a significantly higher external quantum

efficiency, and much larger gain. Designing APDs for back-illumination is also critical for the

future realization of large scale APD arrays. Typical unbiased external quantum efficiencies

are 40%. Gain values between 5 and 20 are typically obtained at breakdown, and biasing these

devices beyond breakdown yields linear mode gain in excess of 1,500.

9.2. Preliminary Device Results

To test the preliminary Geiger mode operation of our devices, we employed a passive quenching circuit and a pulsed mode excitation similar to that discussed above. The APDs are

DC biased through a ~100 KΩ resistor at 74-75V (just below breakdown) and taken into

breakdown with a capacitively-coupled 10-ns pulse of ~7 V at a repetition rate of 10 KHz. The

photon source used to illuminate the sample is the same Xenon lamp coupled to the

monochromator, and the optical fiber is sued to deliver the light to the back of the wafer. The

basic layout of the biasing and quenching circuit is shown below in FigureX 87.X

174

50 nF APD 100 kΩ + VDC 5 kΩ - Oscilloscope

Figure 87. (Top) Schematic diagram of the passive quenching circuit used to apply the

DC bias and AC excitation pulse to the Geiger mode APDs.

In operation the DC bias brings the device close to avalanche, but it is current limited via the ~100 kΩ resistor. The 50 nF capacitor prevents the DC from bleeding back into the pulse

driver, but allows the ~7V pulse to be superimposed upon the DC bias, bringing the APD into

breakdown. However, in the absence of photons reaching the device, there is a statistical

probability that that a carrier will not be present in the multiplication region to initiate the

avalanche process. If there are carriers, then the device undergoes avalanche breakdown and a

large current is conducted through the device. At the end of the pulse the voltage bias is

reduced bringing the device back out of breakdown and quenching the multiplication process.

The ~100 kΩ resistor serves to limit the device current to safe levels, and causes the detector

bias to decrease in proportion to the avalanche multiplication. This effect can self-quench the

device for very short pulses, but acts more to prevent the device from latching-up after

breakdown, ensuring that the reduction in bias after the pulse is sufficient the quench the

device. A 50Ω to 20 kΩ resistor is placed at the device output to convert the current pulse to a

voltage pulse suitable for measurement.

175 Using this setup, we have successfully demonstrated single photon counting capabilities from diodes with active areas up to 14,000 µm2. This represents a significant improvement over previously reported results. The size of device that can be used for single photon detection is limited the dark current of the larger area devise. For larger devices, the dark

current dominates the current transient and masks the photocurrent pulse. FigureX 88 X below

shows typical pulses for a number of wavelengths when the device is illuminated with a fluence

of 10 photons per pulse.

0.6 230nm 310nm 350nm 0.4

0.2 410nm Photocurrent(mA) 0.0

-20-100 10203040506070 Time (ns)

Figure 88. Geiger-mode spectral response of a GaN APD detecting at the 10 photons- per-pulse level.

The effective Geiger mode gain of these devices in was around 1×106. The response under a 10-photon/pulse flux was measured at different wavelengths finding a flat response

176 above the bandgap. No pulse was detected for wavelengths higher than 400 nm thanks to the

visible-blind nature of the devices. This is comparable to the linear mode behavior of these

devices, as shown below in FigureX 89.X

Photoresponse Photoresponse Geiger Mode Linear Mode 10 ph/pulse 75V Reverse Bias Photovoltage (A.U.) Responsivity (A.U.) Responsivity

260 280 300 320 340 360 380 400 260 280 300 320 340 360 380 400 Wavlength (nm) Wavelength (nm)

Figure 89. Left) Linear mode photoresponse at 75V reverse bias. Right) Geiger mode

photoresponse with 10 Photons/pulse illumination.

Single photon detection efficiencies (SPDE) of 4.2% were achieved at a dark count

probability (DCP) of 0.026. For significantly smaller devices (25 μm x 25 μm), we have obtained a SPDE of 28% at a DCP of 0.004. This represents a significant improvement over the previously published results, but there is still ample room for this work to continue.

177 10. Appendices

10.1. Appendix 1: Material Characterization Techniques

A major part of this work was the systematic development of the material growth technology necessary to produce the high-quality high aluminum composition AlGaN necessary for Solar-Blind photodetectors, focal plane arrays, and Avalanche photodiodes. To that end, it is often necessary to study the basic structural, optical, and electrical characteristics of the material. This appendix seeks to provide a brief overview of the material characterization technologies that were used in the completion of this research.

10.1.1. Structural Characterization

The growth of III-Nitrides requires that a great deal of attention be paid to the structural properties of the material. The material system has been around since the 1960’s but the material growth still lacks maturity especially for higher aluminum composition AlGaN layers such as those discussed in this work. There are still no widely available lattice matched substrates for the growth of nitrides and as such the material often has dislocation densities in

7 10 -2 the 10P P to 10P P cmP P range. A major task in the development of Solar-blind photodetectors and

focal plane arrays is the optimization of the structural quality of high aluminum composition

material. It order to carry our this research the four most commonly used structural

characterization techniques are optical microscopy, scanning electron microscopy, atomic force

microscopy, and x-ray diffraction.

178 10.1.1.1. Optical Microscopy

Optical microscopy involves imaging the sample with white light in order to see the

general quality of the surface, and to look for contamination and especially to look for the

presence of cracks on the surface of the wafer. This is the first material characterization;

typically performed on the wafer immediately after it is removed from the reactor. It provides

a low-resolution overview of the entire wafer surface. It is the only structural characterization

that is preformed on the entire wafer; all of the other characterization methods provide a higher

resolution but are only able to sample a portion of the wafer.

The sample is first blown with dry high-purity N2 to remove any loose dust or flakes

that may have fallen on top of it after the growth. It is then placed on the stage of an optical

microscope, illuminated from the top, and observed with 40x magnification. FigureX 90AX shows

an illustrated image of an optical microscope as it is used to observe the sample. The sample

focused and then briefly scanned for gross defects and signs of growth abnormalities such as

visible texture or signs of contamination. Areas of interest are observed at higher resolutions,

up to 400x. An example of a 100X optical micrograph is shown in FigureX 90B.X In addition any comments on the morphology of the sample are also recorder. This information is placed in a central repository with other wafer characterization for future reference.

179

Figure 90. A.) Optical Microscope used to investigate broad area surface morphology

B.) Representative optical micrograph taken at a magnification of 100x, showing a 50% AlGaN

sample with a large number of cracks.

As part of this work a computer controlled image recording facility was setup. Prior to

this a camera and thermal video printer were used to record representative image of all wafers.

This system consists of a windows 2000 computer and a USB color video camera. Custom

software was written to handle the collection and sorting of the images. The main motivation

for writing custom software was to eliminate the need for a keyboard due to space and

cleanliness concerns, and to facilitate the easy sorting of images from the many groups that

share this facility. This software, shown below in FigureX 91.X It consists of a preview window,

a menu bar with two sets of buttons for each of the groups, and a thumbnail of the last saved

image. Clicking one of the menu buttons prompts the user for a file name to save the image.

Each button keeps track of its own file history to make sorting pictures by group easier. When the image is saved, a series of 16 images are first captured and then averaged together to get a low noise 32bit color image from the native I420 12bit per pixel video data.

180

Figure 91. Custom Designed software created for capturing of still images using an optical microscope.

10.1.1.2. Scanning Electron Microscopy

Scanning electron microscopy is another valuable tool in the semiconductor

characterization arsenal. It provides structural characterization at magnifications from 50x all

the way to 500,000x. This resolution is too high to realistically allow examination of the entire

sample, however it provides the necessary resolving power to detect surface roughness in

between what is resolvable with visible light, and the atomic scale that AFM reveals. In our

facility we use a Hitachi S 4500 field emission scanning electron microscope.

The scanning electron microscope works by generating free electrons via field emission.

A voltage of ~5 KV is applied to an atomically sharp tip which causes quantum mechanical tunneling of electrons due bending of the potential barrier. This is in contrast to the more common thermionic emission of electrons from a tungsten hairpin filament, and it has the advantage of providing for a much smaller ultimate focused beam-spot size and thus allows

181 resolution of smaller features than would otherwise be possible. The electrons extracted from

this field-emission source are then accelerated towards the sample to form a beam with a

current of 10 μA and a potential of 4 KeV. Before the beam reaches the sample a series of

condenser lenses and apertures are used to shape the beam. A set of stigmators and scanning

coils correct for aberrations and raster the beam, respectively. The final lens, called the

objective lens, creates an electron spot on the surface with a diameter on the order of a few

angstroms. This spot is rapidly scanned over the sample, and the secondary electrons emitted

from the interaction of this beam with the material are collected using a small applied-bias. An

image of the surface is constructed by mapping this detector signal as the intensity at the

corresponding beam position and showing the image on the display screen. A schematic

diagram of the scanning electron microscope is shown in FigureX 92.X

Electron Gun Aperture

Condenser Lenses Scan Generator Aperture Scan Coils Objective Lens

Amp Display CRT Secondary Specimen Electron Detector

Vacuum Pumps

Figure 92. Schematic diagram of a scanning electron microscope showing the electron source, the lenses and scanning coils, the sample under test, and the electron detector (based

194 on ref. TPD DPT).

182 Typically the surface of the sample is focused at a magnification of 100,000x; this is a factor of two higher than the magnification at which images will be recorded and thus allows the quality of focus obtained to be better than that possible if the microscope is merely focused at 50,000x. The focusing of a scanning electron microscope involves first moving the sample to the correct working distance, typically 12 mm, and then converging the electron beam so as to obtain a reasonable image of the sample. The electron beam is then centered in the column by applying a small AC signal to the lens and then adjusting the biases until the image remains centered; this helps reduce the effects of spherical and chromatic aberrations within the lens system. The sample surface is then focused as well as possible, after the stigmators are adjusted to compensate for any astigmatism in the lenses. After that, the focus is adjusted a final time and the magnification is decreased back to 50,000x before recording images of the sample surface.

Typical a 50,000x magnification image of the surface is recorded for every wafer. The

sample is then rotated and moved back into position so that its cross-section can be observed.

Because of the use of electrons in imaging the SEM provides a degree of material contrast and

under ideal condition it is possible to make out the individual layers that make up a growth.

This capability is used to measure the thicknesses of the various layers grown. The ability to

differentiate between different layers depends upon the difference in the aluminum composition

as well as the average aluminum composition; for high resistance materials, such as high Al composition AlGaN, the electrons can build up locally in the material thus creating an electric field that repels the electron beam and make imaging difficult if not impossible.

183 10.1.1.3. Atomic Force Microscopy

In our facility we use a Digital Instruments Nanoscope III AFM that is capable of

performing a number of different scanning probe microscopy techniques. The most common

of these techniques is atomic force microscopy (AFM). It is the highest resolution surface

characterization technique we typically employ. The field of view is limited to a box on the

order of a few microns to as small as 500 nm on edge. The surface within this region can be

probed with near atomic resolution. This technique cannot quite resolve the individual atoms, but it is able to clearly make out the atomic steps on the sample surface. The shape and intersection of these steps can be used to judge the quality of the crystal lattice. It is also

possible to directly see the termination of certain types of dislocations and defects on the surface; in addition, acid pre-treatment of the surface can be employed to selectively etch the dislocations and enlarge them so they can be more easily detected. By counting the dislocation terminations per unit of area it is possible to determine an estimate of the dislocation density

8 10 (usually in the range of 10P P to 10P P dislocations per cm for AlGaN).

Atomic force microscopy obtains an image of the surface by scanning an atomically sharp tip over the surface of the sample. The tip is typically made of a silicon and by using selective etching the actual point of the tip that interacts with the sample surface can have a radius of curvature of approximately 5 nm, depending upon tip wear. The tip sits at the end of small silicon cantilever that acts like a spring and allows the tip to follow the surface topology.

The sample is held rigidly, and the tip and spring arm are mounted on the end of a precision X-

Y-Z piezoelectric actuator column. Stepped motors are used to being the tip and sample into close proximity, and then by manipulating the piezoelectric column the tip can be brought into contact with the surface. This piezoelectric column is also able to raster the tip over the surface

184 of the sample. As the tip is scanned a laser beam is reflected off of the back of the tip and onto

a 4-quadrant photodetector that is used to measure the actual tip deflection. A schematic

diagram of an atomic force microscope is shown in FigureX 93.X This is the most straightforward form of atomic force microscopy, however it is more common to operate with the tip in tapping mode. In this mode the tip is driven so that it oscillates at a frequency of several hundred kilohertz. This causes the tip to repeatedly make and break contact with the surface of the sample. This has the advantage of drastically reducing the duration of tip contact with the surface and thus eliminates lateral effects as the tip is scanned over the surface of the sample.

It also has a tendency to extend the useful life of the tip by reducing wear, and can make the imaging less sensitive to surface contamination when imaging a non-pristine surface such as occurs when imaging in ambient air. In tapping mode the electronics operate a feedback loop that attempts to adjust the magnitude of the driving function to keep this amplitude of the oscillations constant. An image of the surface topology can be reconstructed from this amplitude data.

Laser Diode 4-Quadrant Detector

Cantilever Probe-Tip

Sample

Figure 93. Schematic diagram of an atomic force microscope (Based on Digital

Instrument Multimode SPM manuals).

185 AFM is not typically performed on all samples due to the complexity of the procedure and the time associated with rastering the tip over the surface to obtain an image. In addition

the surface morphology of p-GaN, such as the top contact layer on most device structures, has a

characteristic texture that makes imaging of the surface of devices difficult. However AFM has

proven to be an invaluable tool for the characterization and optimization of the constituent

layers. Typically scans are performed in tapping mode using a silicon tip with an oscillation

frequency of ~300 KHz. The tip is scanned over an area of 5 μm x 5μm with a raster rate of 1

Hz. Typical AFM images have vertical data scales ranging from 3 to 10 nm depending upon

the surface roughness. The recorded height data is post processed by flattening to correct for

sample non-planarity. The root mean squared (RMS) roughness is then extracted from the

image to provide a quantitative measure of the surface smoothness.

In addition to using AFM to image as grown samples and determine the RMS roughness,

AFM is also used to estimate the dislocation density of samples. As grown the terminations of

threading dislocations can barely be resolved by the AFM as one or two pixel black dots on

image of the surface. In order to more accurately count the number of dislocations terminating on the sample surface and thus estimate the dislocation density, the sample is etched in hot acid

to selectively enlarge the dislocations. Hot acids, such as phosphoric acid (H3B POB 4B )B or sulfuric

acid (H2B SOB 4B ),B and molten potassium hydroxide (KOH), have been shown to create etch pits at defect sites of GaN, and to lesser extent AlGaN. The sample is typically immersed in hot acid

~200 °C in increments of 15 seconds until the surface begins to shows etch pits, but before the etch pit become so large that they begin to merge together. It is then scanned with the AFM, and the number of etch-pit present in the scan area is counted this number is then adjusted for the scan area to get a dislocation density

186 10.1.1.4. High resolution X-Ray Diffraction

X-Ray diffraction (XRD) is a crystallographic characterization technique that provided

information of the quality and structure of the material. Unlike the other structural

characterization techniques discussed above X-ray diffraction is not strictly a surface technique,

and is capable of probing a larger three-dimensional volume of the sample. The x-rays used in

XRD are of a sufficiently high-energy to have a wavelength smaller than the lattice spacing in

the sample, X-rays. For semiconductor research Cu Kα radiation is the most common x-ray

source; it has an energy of 8047 eV which correspond to a wavelength of only 1.541 Å (for

comparison AlGaN has a c-axis lattice constant of about 5 Å). As shown in FigureX 94 X the

incident X-rays scatter off of individual atoms, but because the atoms exist in an order crystal

structure, these scattered x-rays interfere with each other and create a diffraction pattern.

Constructive interference occurs when the Bragg condition is met:

nλ = 2d sin(θ) . (30)

Where n is the order of the diffraction, λ is the wavelength of the x-rays (1.541 Å), d is

the lattice spacing of the atomic planes that the diffraction arises from, and θ is the angle at which the diffraction occurs. The diffraction pattern is recorded as a function of θ and the location of peaks reveals information about the lattice spacing, and the dispersion of these peaks reveals information about the degree of order of the crystal lattice.

187

Figure 94. Diffraction from a set of crystal planes according to Bragg’s law.

X-ray diffraction experiments are performed using a Phillips MPD1880/HR high- resolution x-ray diffractometer. The x-ray source used is a 1.8 kW sealed copper-anode x-ray tube. This tube consists of a tungsten filament and a copper anode; a small current is applied to the filament and electrons are thermally emitted from the filament with a current of 40 mA.

The entire tube is evacuated, and a potential of 40 kV is applied between the filament and the copper anode. This accelerates the electrons which then impact the copper target with sufficient energy to displace inner core electrons and cause the copper target to fluoresce giving off characteristic x-ray. The X-ray output of the tube is then passes through a pair of germanium channel cut crystals that eliminate all but the Kα characteristic x-rays and eliminate any other x-rays that arise from the sudden deceleration of the x-rays. The beam that exits from the channel cut crystals is narrowly defined both angularly and spectrally. This beam is then incident upon the sample, which is held on the goniometer stage. The goniometer also holds an

188 x-ray detector on a separate arm. The sample and detector are scanned in concert such that the

angle of incidence and the angle of the detector are the same with respect to a normal to the

crystal plane under observation. The diffraction intensity is recorded as a function of the sample stage position to create an XRD scan.

Typically X-ray diffraction is used to determine the aluminum composition of a sample and to help judge the crystallographic quality of the material. From a simple symmetric x-ray diffraction scan of the (002) diffraction peak it is possible to determine the c-plane lattice constant of the material. (This value is usually measured relative to some other layer in the structure such as binary GaN or AlN, or in the more general case the substrate, the absolute value is then extracted from the known layer to minimize the effects of offset that arise in the measurement machine.). The ternary composition can be determined if the lattice constants of

both binary materials (EAlNB B and EGaNB B) are know, as well as the bowing parameter (B), by

applying Vegard’s law:

E AlGaN = xEAlN + (1− x)EGaN − Bx(1− x) . (31)

This same technique also works other ternaries such as InxB GaB 1-xB NB and InxB AlB 1-xB N.B

In addition to the measuring diffraction from crystal planes parallel to the sample surface,

such as the (002) planes described above, it is also sometime desirable to record the diffraction patter from planes that are off access. In the III-Nitrides material system the dominant dislocations are threading dislocations that propagate in the direction of growth (typically along the c-axis); however the spacing of the c-planes is relatively insensitive to these dislocations.

By imaging a plane tilted 25 to 45 degrees with respect to the sample it becomes possible to get a more realistic idea of the dislocation structure. For AlN and high aluminum composition

189 AlGaN layers a (105) scan is typically performed; this plane is tilted 20 degrees with respect to the sample surface and is kinematicaly favored for diffraction. In GaN and low aluminum composition AlGaN layers a (204) scan is more often chosen. This plane is tilted 43 degrees with respect to the sample surface, and although the intensity is not as great as it would be from the (105), in GaN the intensity is still sufficient to get a good signal.

10.1.2. Optical Characterization

In addition to the structural characterization of the material it is often necessary to probe the material optically. Optical characterization can be used to determine the band-gap of the material, detect defect levels, and to gain insight into the quality of the material. The two most common optical characterization techniques are photoluminescence and optical transmission.

Photoluminescence is a simple measurement that uses a laser to optical pump the material; the bandgap can then be determined from the fluorescence and the material quality can be judged from the intensity and full width at half max of this photoluminescence. Optical transmission uses a broad-band white source and a monochromator to measure the absorption as a function of wavelength; the band gap can then be extracted from the absorption edge, and the quality can be judged from the pendellosung oscillations and the rate of fall off at the band-edge.

10.1.2.1. Photoluminescence measurement

Photoluminescence measurement uses one of three excitation sources depending upon the wavelength region of interest: in the near visible region a CW HeCd laser with an output power of 25 mW is used, in the solar-blind region a CW frequency doubled argon ion laser

(FreD) with an output power of 50 mW at 244 nm is used, and finally in the deep UV a 193 nm

190 pulsed excimer laser with a repetition rate of 200 Hz and a energy of 3-5 mJ is used. This set

of excitation sources encompasses the entire range from AlN to InGaN. The monochromator

has a holographic grating with 1200 grooves per inch and a blaze angle of 300 nm. The lenses

used are all CaF so as to be compatible with the entire spectral range. The PMT is a

Hamamatsu R928 photomultiplier tube which is biased at 300 V and has a relatively flat

5 response of ~5x10P P A/W from 200 to 400 nm. The entire optical setup is shown in FigureX 95 X below.

Figure 95. UV photoluminescence setup showing the three laser excitation sources that can be used to stimulate PL, as well as the sample stage, focusing optics and monochromator.

The photo-excited luminescence is collect from the sample using the first lens to collect and collimates the light; the second lens then focuses the light into 350 μm entrance slit of the monochromator. The PMT measures the light coming out of the monochromator, and a lock-in amplifier in concert with an optical chopper on the laser beam create a synchronous detection

191 scheme to help reduce the signal to noise ration of this signal. The monochromator and lock-in

amplifier are computer controlled so as to facilitate automatic data acquisition from the system.

Photoluminescence is by far the most common characterization after optical microscopy and SEM. Almost all samples undergo photoluminescence measurement immediately after growth to determine the aluminum composition and to help judge the material quality of the

layer. The PL intensity is strongly dependant upon doping of the layer; due to screening of

defect states the PL of Si doped layers is often several orders of magnitude grater than those of

un-doped layer. However when making comparisons between similar layer the PL intensity is

a valuable metric. In addition, the width of the peaks are also studied. A broader peak tends to

indicate more inhomogeneity whereas a narrower peak is indicative of a high quality layer, of

the presence of a quantum well. The third useful feature of PL is the ability to resolve smaller

defect related peaks that may come from impurities within the material or means on non-band-

to-band radiate recombination. The dynamic range of the PL setup covers several orders of

magnitude and by looking at the response below the band-gap one can detect features indicative

of the material quality. These can then be used to make relative comparisons among different

layers in order to determine the optimum material design.

In order to support this research we developed a fully automated custom solution to

facilitate measurement and analysis of the photoluminescence. This consists of a series of

custom Labview VIs that interface the laser, monochromator, PMT, lock-in amplifier, and

safety shutter. This software records the photoluminescence of up to 16 different samples. It

then implements a peak-find algorithm to find the PL-peaks, assign peak values, and determine

the FWHM of each peak. This is thin automatically exported to Micorcal Origin for printing

with a minimum of operator effort.

192

Figure 96. Custom Labview software developed as part of this work to facilitate the measurement and analysis of routine Photoluminescence.

10.1.2.2. Optical Transmission

Optical transmission allows the calculation of the transmission and absorbance of a layer

as a function of wavelength. At its simplest this can be used to test the back-side cut-off of a

back illuminated photodetector by measuring the cutoff of the template and conduction layer.

It can also be used to test the photodetector active region to determine the absorption

wavelength and efficiency within the i-type region. In addition the quality of a layer can be

inferred from the sharpness of the cut-off and the presence of well-ordered pendellosung

oscillations owing to the goof morphology and sharp interfaces between the layers. However

the added complexity of measuring transmission makes PL a better-suited characterization

technique for general use.

193

Figure 97. Ultraviolet transmission measurement system showing the xenon lamp, monochromator, chopper, lens, sample holder, and the UV-enhanced silicon photodetector.

The measurement setup requires that a calibration baseline be recorded before the sample itself can be scanned. This involves placing a piece of sapphire over a calibrated aperture on

the sample holder and recording the power of the xenon lamp as a function of wavelength using

a UV-enhanced silicon photodiode hooked up to a current amplifier which is fed to a lock-in

amplifier for signal acquisition. The optical portion of this measurement setup is shown in

FigureX 97 X above. After the background has been recorded, the bare piece of sapphire is replaced with a piece of sapphire with the layer in question grown on it; the same aperture is used and the geometry of the system is not disturbed. The sample is then scanned, and the transmission is derived from this scan data normalized to the background scan. The absorbance squared as a function of energy can then be calculated by converting the wavelength to energy and using the following formula for absorbance squared:

194

2 ⎛ T % ⎞ ⎛ T % ⎞ A = log⎜ ⎟ ⋅log⎜ ⎟ . (32) ⎝100 ⎠ ⎝100 ⎠

To facilitate the recording of baselines, and calculation of UV-transmission and absorbance squared, custom Labview software was developed to facilitate this work. Prior to the development of this software, UV-T had to be done off-site making it underused. The software controls the lock-in Amplifier, trans-impedance amplifier, and spectrometer. It first records the power distribution of the Xenon lamp, and then allows for scanning samples against that calibration to calculate the percent transmission and the absorbance.

Figure 98. Custom Labview software developed as part of this work to measure UV transmission.

195 10.1.3. Electrical Characterization

Hall mobility measurement is the main electrical characterization tool used to study III-

Nitride material. Ideally it is possible to do more in depth study using poloran C-V profiling or admittance spectroscopy, however these are specialized measurement techniques that fall out side of the normal scope of material characterization. Polaron is not really feasible due to difficult in wet etching the material. Admittance spectroscopy, and the more generalized deep- level transient spectroscopy are employed on a limited basis, but only hall mobility measurement will be discussed within the context of this appendix.

10.1.3.1. Hall Mobility Measurement

Hall mobility measurement is a powerful electrical characterization technique that allows determination of the majority carrier type and concentration, the mobility of those carriers, and the resistivity of the sample. However some preliminary sample perpetration is necessary in order to complete this measurement. This measurement assumes a homogeneous sample with a constant thickness; in order to meet this condition the edges of the sample need to be removed. This eliminates any effects that would arise from the reduced thickness around the edges and the less-high material quality typically present there. It is also necessary to significantly reduce the total area of the sample when preparing it for hall measurement; this especially true for measurement of the mobility of high aluminum composition AlGaN layers.

Since doping becomes proportionally harder as the aluminum composition is increased these layers even when of a high-quality do not have very low reistivities. If the sample is not sufficiently small the voltage necessary to force a current through the sample quickly exceed the capacity of the machine thus making measurement impossible. Typically sample are

196 removed of all edges and cut into squares approximately 2 mm on edge; this is about the

smallest sample it is practical to work with. For GaN and other low resitivity material the

sample can be as large as 6 mm on edge.

To help support this work on wide-bandgap AlGaN semiconductors, a custom hall

system was implemented to replace an older commercial BioRad system. The system was

specifically designed to minimize leakage current, maximize the input impedance of the

instrumentation, and reduce the capacitance of the system. The original system had a

maximum range of 6V at 1 μA with a maximum practical sample resistance of 10 to 100 kΩ.

The new system replaced the DAC and leaky FET relays of the old system with High input

impedance (>100 GΩ) electrometers, and a low leakage reed relay based switch matrix. The

current source was also replaced with a precision current source capable of sourcing current for

100 mA to 1 pA. This doubles the range to the system allowing samples up to 1 GΩ to be

measured accurately. A schematic diagram of the hall system is shown below in FigureX 99.X

Figure 99. Custom Hall effect measurement electronics assembled to facilitate the measurement of high impedance III-Nitride materials.

To measure Hall, four ohmic metal contacts are placed around the periphery of the

sample in order the apply currents and measure voltages. These contacts are generally made

197 with a eutectic alloy of gallium and indium which is liquid at room temperature while being

relatively safe to handle. This technique has the advantage allowing nearly instant

measurement of the sample, however for AlxB GaB 1-xB N:SiB with x > 0.6 and un-doped or p-type material the quality of this contact is insufficient to allow successful measurement. In these cases the contacts are usually formed by lithographic pattering, electron-beam evaporation, and subsequent thermal annealing; the edge effects are then removed by etching the sample all the way down to the sapphire to electrically isolate the region under test.

Using the van der Pauw technique the resistivity can be determined using 4 contacts placed around the periphery of an arbitrarily shaped sample assuming a minimum of inhomogeneity within the sample. Current is applied to two adjacent contacts and the resulting

voltage is measured across the opposite two contacts as shown in FigureX 100A.X Several measurements are made for the different possible permutations in order to determine a symmetry correction factor (Q), and a geometry correction factor (F). Assuming the current is held constant for all measurements:

V43 ⋅ I14 V43 Q = = , (33) I12 ⋅V23 V23

and the geometry correction factor (F) can be estimated as:

2 F =1− 0.34657A − 0.09236A , (34)

Where

198

V43 V23 − 2 I I ⎡Q −1⎤ A = 12 14 = . (35) V V ⎢Q +1⎥ 43 + 23 ⎣ ⎦ I12 I14

These two factors are then used in the van der Pauw equation to yield a value for the

sheet resistivity (ρsB ):B

p π ⎡V43 V23 ⎤ ρ s = = ⋅ ⎢ + ⎥ ⋅ F ⋅Q (Ω.cm). (36) t 2⋅ln(2) ⎣ I12 I14 ⎦

I B V43 I12 VH

14 2 3

(A) (B)

Figure 100. Schematic diagram of a typical hall mobility measurement system showing

the van der Pauw contact geometry as used for measurement of the resitivity (A) and for the

determination of the hall coefficient (B).

In order to determine the hall coefficient and thus determine the mobility and carrier

concentration, current is applied across two opposite contact on the sample in the presence of a

magnetic field, and the resulting hall voltage is measured across the other two contacts as

shown in FigureX 100B.X The sheet hall coefficient (RHsB ),B and if the thickness is known, the hall

coefficient (RHB ),B can then be determined from the measured hall voltage (VHB ),B the current (I), and the magnetic field (B):

199

RH VH 2 R = = (mP /C)P . (37) Hs t I ⋅ B

However in order to improve the consistency of the measurements, the hall voltage (VHB )B used is actually an average of all the possible permutations of magnetic and electric field directions: this helps cancel out any offsets within the measurement system. Once the sheet

hall coefficient (RHsB )B has been determined, the sheet carrier concentration (NsB )B can then be determined

-2 N s = 1/ q ⋅ RHs (cmP P) ; (38) or in the case where the thickness of the sample is known, the bulk carrier concentration (N) can be determined:

-3 N =1/ q ⋅t ⋅ RHs (cmP P). (39)

Finally, the hall mobility (μHB )B can then be determined from the sheet hall coefficient (RHsB )B and

the sheet resistivity (ρsB )B as follows:

μ H = RHs / ρ s . (40)

Typically hall is performed at room temperature on all n-type samples. Reisitivities,

motilities, and carrier concentrations are recorded for each sample. Occasionally samples will

be so resistive that reliable results cannot be obtained either due to poor ohmic contacts or over

voltage on the hall machine. In these cases the sample is marked at highly resistive; however

on a limited basis some of these samples are further characterized by formal metallization of

contacts.

200 10.2. Appendix 2: Device Testing and Characterization

This section provides an overview of the discreet deice measurement procedures used to

characterize both broad area devices and FPA test pixels. This includes: the current voltage

measurement to determine the leakage current, turn on voltage, series resistance, and ideality

factor; the measurement of the device responsivity and determination of the external quantum efficiency of the photodetectors; as well avalanche gain and noise characteristics of APDs operating in both linear and Geiger mode.

10.2.1. Current Voltage Measurement

Current-voltage profiles of the devices are taken on wafer using a low noise probe station

in concert with a semiconductor parameter analyzer as shown in FigureX 101 X below. The probe

station is PM 5 manufactured by Karl Suss; it is equipped with a pair of special low noise

triaxial probe arms with tungsten probes. There are two types of probes used: a full shank

probe with a tip radius of ~5 μm for general use, and a reduced shank probe with an ultra fine

tip less than 1 μm in radius that is used for probing pixel size detectors and APDs. The

semiconductor parameter analyzer used to record the I-V curves is a HP model 4155A; is uses

low noise triaxial source measure units to record the IV curve. Using standard integration

-10 times the system is routinely able to obtain a noise floor of ~ 1x10P P A. This measurement is

used to directly determine the turn on voltage, and reverse-bias leakage of the device. This data

can then be further used to extract values for the series resistance and ideality factor of the

device.

201

Figure 101. Current-Voltage characterization setup used to measure and record I-V curves. The probe station has probe-tips as small as .5 μm, and is suitable for probing to individual pixels of an unbounded FPA.

The series resistance is determined through modeling of the IV curve in the high current regime. Under low current injection, the effect of series resistance is negligible; thus, the ideality factor of the diode can be calculated. The ideality factor can be helpful in understanding the conduction mechanisms operating in the device.

202

-6 log ( I ) -8 Linear Fit -10 η = 1 -12 η = 2 -14

-16 η = 3.71606

ln (I) (A) -18 -20 -22 I ~ exp(qV/(nkT)) -24 -20246810 Aplied Bias (V)

Figure 102. Typical IV curve of a single 25μm x 25μm FPA pixel shown in log scale.

The linear fit of this data gives an ideality factor of 3.7; slopes for ideality factors of 1 and 2

are displayed for comparison purposes.

FigureX 102 X shows the low current regime of the IV curve in log scale; the ideality factor

is calculated by fitting the linear part of the curve with equation (41)X ,X this yields an ideality

factor of n = 3.7 for this diode.

qV I = C × exp( ) . (41) f nkT

In equation (41)X ,X IfB B is the forward current, V is voltage, n is the ideality factor, T is

temperature, q is the electron charge, and k is Boltzmann’s constant. Equation (41)X X arises from

the combination of the equations for diffusion current (equation (42)X )X and recombination

current (equation (43)X ),X the two currents that usually dominate the diode current:

203 2 D p ni qV I diff = q exp( ) , (42) τ p N D kT

qW qV I = σν N n exp( ) . (43) rec 2 th t i 2kT

In equations (42)X X and (43)X ,X IdiffB B is the diffusion current, DpB B is the hole diffusion

coefficient, τpB B is the hole lifetime, niB B is the intrinsic carrier concentration, NDB B is the donor

concentration, IrecB B is the recombination current, W is the depletion region width, σ is the

conductivity, ν th is the thermal velocity, and NtB B is the density of traps. Combining equation

(42)X X and equation (43)X X into a single empirical equation yields Equation (41)X ,X where n, the ideality factor, has a value between 1 and 2. If n is closer to 1, then diffusion current

dominates, however if n is closer to 2, recombination current dominates. The inset of FigureX

102 X displays the corresponding slopes for the cases of n = 1 (diffusion) and n = 2

(recombination). However, our value of n = 3.7 falls slightly out of this expected range,

suggesting that an additional process makes a significant contribution to conduction. This non-

ideal behavior has been attributed to tunneling conduction in III-nitrides. For InGaN/AlGaN

heterostructures, it has been proposed that holes tunnel into empty acceptor impurity bands and

195 vacant valence band tails.TPD DPT Moreover, for 285 nm UV LEDs, it has been speculated that high-

196 energy electrons from the active layer are able to tunnel into the p-type barrier layers.TPD D PT

10.2.2. Responsivity and External Quantum Efficiency

The responsivity and external quantum efficiency of the UV photodetector are experimentally determined by illuminating the photodetector with a known flux at a given

204 wavelength, and recording the resulting photocurrent. The measurement is performed using a

75 watt xenon arc lamp as the optical source. The wavelength is controlled using a 0.5 meter monochromator: the entrance slit of the monochromator is set at 1 mm and the exit slit is defined by the 100 μm diameter fiber optic, the grating is a holographic grating with 2400 groves per inch blazed for operation in the UV. This easily allows control of the wavelength between 200 and 450 nm, and easily allows resolution of the rather broad response of a typical photodetector.

A fiber optic cable couples the light from the monochromator to a custom fabricated sample stage or to a floating fiber optic probe. In either case the fiber used is a special solarization resistant UV-grade fiber with a total length of 1 meter. Using computer controlled beam pickoff mirrors the light from the xenon lamp can be directed to either the fiber-optic probe or the fiber mounted in the bottom of the custom stage. It is also possible to divert the entire power of the xenon lamp away from either fiber. In practice this allows for very easy automatic sequential measurement of the dark current, top-illumination photocurrent, and the back-illumination photocurrent. A schematic of the measurement setup is shown below.

205

Figure 103. Measurement setup used to characterize the photoresponse of both broad-area photodetectors and individual text pixels from an FPA.

The light is chopped at the entrance to the monochromator using a custom-fabricated chopper housing to couple the lamp housing to the monochromator. Upon exiting the monochromatic the light is chopped at 200 Hz with a 50% duty cycle using a chopper wheel.

This allows synchronous detector of the photocurrent and thus reduced the noise floor and improves the consistency of the measurements. The photo current is detected using a Kiethly model 428 Transimpedance amplifier hooked up to the voltage input of EG&G model 5209

lock-in amplifier. The sample stage is shown below in FigureX 104.X

206

Figure 104. Photograph of a wafer on the probe-station used for responsivity measurements showing the fiber optical cables (above and below wafer) and the two triaxial probes.

The entire system is controlled by a personal computer with custom instrumentation software that was written in Labview specifically to support this work. The layout of the interface is

shown below in FigureX 105.X The interface manages control of the monochromator, current amplifier, and lock-in amplifier. It allows for calibrating the power of the xenon lamp and then recording a series of photodetector responsivity and quantum efficiencies.

207

Figure 105. Custom written software used to make UV detector responsivity and external quantum efficiency measurements.

Experimentally the power of the Xenon lamp is first recorded using equation (44)X X below.

A calibrated UV enhanced silicon photodetector is used to collect the light. A precision

aperture the almost the same size as the photodetector to be tested is used to control the area of

the silicon detector. In equation (44)X X below G is the gain of the current amplifier, AsiB B is the area

of the aperture placed on the silicon photodetector, IdetB B is the recorded signal obtained by the

lock-in amplifier, and RsiB B is the known responsivity of the silicon photodetector at the current wavelength.

208

I det PXe = . (44) Asi *G *ℜsi

However before the spectral power of the xenon lamp can be recorded it is necessary to

allow the lamp to completely stabilize. Since there is only one detector in the system at any

given time, the xenon lamp must be calibrated before the scan is begun, and then assumed

constant as the detector is switched and the scan is begun. This assumption cannot be

accurately made until the lamp has warmed up for approximately an hour and a half. The

output flux from the xenon lamp at a wavelength of 280 nm is shown in FigureX 106 X as a function of time since the lamp was switched on. The blue bands indicate a +/- 1% stability in the lamp flux. The lamp enters this region at approximately 45 minutes, but 90 minutes is chosen as the standard time in order to ensure consistency of the measurements.

) 4.0 -2 3.5 Xenon Lamp Power 3.0 +/- 1% 2.5

2.0 1.5 1.0

280 nm UV Flux (W m 0.5 0.0 0 102030405060708090100 Elapsed Time (minutes)

Figure 106. Xenon lamp flux as a function of time since turning on the lamp showing the

variation during lamp warm up. The blue bars indicate a +/- 1% variation in xenon lamp

power.

209 The calibrated UV-enhanced silicon photodetector can then be replaced with the

unknown III-Nitride detector. The same scan can then be preformed and IdetB B recorded. From

the previously recorded power of the xenon lamp (PXeB ),B the known current gain (G), and the known area of the photodetector (A), the responsivity (R) can then be derived from equation

(45)X X below. The external quantum efficiency (η) can then be determined from the responsivity

(R), wavelength (λ), and the constants h (Plank’s constant) and c (speed of light) as in equation

(46)X X below:

I ℜ = det , (45) PXe *G * A

hc *ℜ η = . (46) λ

10.2.3. Bias Dependant Responsivity

The measurement of bias dependant responsivity is similar to the generic measurement of the photodetector responsivity, however using the built in biasing capabilities of the current amplifier, or an external voltage source, a small reverse bias is applied to the device. This adds to the build in electric field and helps better sweep out carriers from the active region thus improving the quantum efficiency and thus the responsivity of the photodetector. However the trade off is that more noise is introduced into the device. This measurement is preformed to provide an idea how the operation of the device will improve with the application of bias. It is also done to show that the device is a true p-i-n photodetector, and that it does not show signs of photoconductive gain: a phenomenon usually related to bad device contacts that can lead to the presence of a much slower photodetector response.

210 10.2.4. Gain Measurement

Gain measurements are performed on the same probe station as the responsivity measurements. However instead of using an optical chopper and lock-in to record the difference between the photocurrent and the dark current, the dark current and light current are measured separately as a function of applied reverse bias. This allows for separate recording of the dark current and photocurrent, with the photocurrent being the difference between the light and dark currents.

To record the gain, the device is scanned from 0 to up to 200 VDC using the HP 4155C curve tracer, and the current is recorded at each point. The monochromator is then used to illuminate the sample with the xenon lamp from either the top or bottom. The scan is repeated, this time under illumination. This alternation of light-dark measurements is repeated a total of

3 times to give a matrix of 9 different permutations of the photocurrent. To ensure statistical significance even though the measurements are collected at different times, the mean and variance are calculated on the photocurrents from these nine permutations. This entire process is fully automated to ensure accurate correlation between the dark and light measurements, with

custom Labview software managing all of the instrumentation. As an option to allow

measurements up to 1000 VDC (especially important for AlGaN APDs) the measurements can

also be performed with a Keithley 2410 source meter.

211

1E-3 Gain01-01 AVE GAIN-02-01 GAIN03-01 GAIN02-01 1E-4 GAIN02-02 10 1E-4 Gain02-03 GAIN03-01 Gain

GAIN03-02 GAIN03-03

PhotoCurrent (A) 1E-5 1E-5 (A) PhotoCurrent 1

0 -20 -40 -60 -80 -100 0 -20 -40 -60 -80 -100 Reverse Bias (V) Reverse Bias (V)

Figure 107. Example of permutation of data from gain measurements used to generate gain data and error bars.

This gives a curve plotting the photocurrent versus reverse bias with error bars showing the variance. This data is then normalized to a chosen point where gain = 1 to get a plot of the avalanche multiplication versus bias. Exemplary plots of the nine permutations and the

resulting gain and variance are shown above in FigureX 107.X Similar calculations can be made

to determine the dark current of the device as a function of reverse bias, and the error

associated with that measurement.

10.2.5. Noise Measurement

Noise measurements of the photodetectors and APDs are performed using the same basic

measurement probe station setup as that used for responsivity and gain measurement. This

allows for recording noise in the dark, or under front or back illumination. The device bias is

supplied by the Keithley model 2410 high voltage source meter, though for biases smaller than

5 Volts, the current amplifier can directly supply the bias. The detector noise current is

212 amplified by a Keithley model 485 programmable current amplifier. This is a low noise wide

bandwidth amplifier that converts the input noise current to an output noise voltage with a

7 transfer characteristic of 10P P V/A. The output of this is then fed into a SRS Model 760 FFT

spectrometer with a bandwidth of 100 kHz and a dynamic range of 90 dB. This allows

recording the noise spectral power density spectrum as a function of applied bias. This data can

then be used to extract the excess noise factor as described in section 3.5X X NoiseX Analysis in

Avalanche Photodiodes.X

In order to streamline and automate the collection of noise spectra versus bias, a custom

Labview application was developed as part of this work to control the FFT, Current amplifier,

and various bias supplies. The resolution and bandwidth of the FFT spectrometer is dependant

upon the sampling time, with typical spectrum taking on the order of a minute to acquire. The software allows for unintended collection of data, automatically plotting the evolution of the noise spectral power density and facilitating saving of multiple scans in a single file. The front

panel of this custom software is shown below in FigureX 108.X

213

Figure 108. Custom software written to facilitate the collection of noise Spectral Power

Density as a function of reverse bias for characterization of photodetectors and APDs.

10.2.6. Geiger Mode APD Measurement

Geiger mode APD measurements are discussed within the context of future work, but a great deal of effort has already been invested in this area, and tentative measurement facilities and procedures are being developed to assist this work. At this time all Geiger mode APD measurements are being conducted in a gated pulse setup with passive resistive quenching. In this setup the APD is biased through a load resistor at just below breakdown with an external

DC bias. A small pulsed AC bias is capacitively coupled into the device. This pushes the device over the breakdown voltage, and if photons are present during the pulse they can then trigger an avalanche breakdown in the device result in a photon count. Once it breaks down, the

214 device will latch up and continue to conduct until the AC pulse is completed, thus quenching

the APD. The circuit used to achieve this mode of operation is shown below in.

Figure 109. Left) diagram showing the self quenching that the 100kΩ resistor ideally provides to the APD, Right) Schematic diagram of the Geiger mode APD biasing circuit.

In our preliminary setup the current pulse is read out using a sense resistor to generate a voltage suitable for counting. The voltage pulses can be counted by recording a histogram using a Tektronix model TDS 754D oscilloscope’s data analysis features. This has the advantage of not only allowing counting of pulses, but of also allowing direct observation of the pulse waveform shape. For more accurate pulse counting a Stanford Research model

SR400 dual channel gated photon counter can be used. This allows setting very accurate discriminator thresholds for the pulse height and tuning this value to maximize the photon to dark count ratio. This also has the advantage ob being able to operate in a gated mode, which can make it immune to extra pulsing and ringing that are prone to occur for short pulse widths.

It also allows for counting pulses at up to 200 MHz which ensure that all photon detection events are counted.

The input flux of photons can be established by inserting neutral density in front of the fiber optic cable, or by narrowing the input slit of the monochromator to decrease throughput.

215 The photon flux can then be calibrated by making measurements with the calibrated silicon

detector, and correcting for the photon energy and the pulse width to convert the incident power

into photons per pulse number that can be used in Geiger mode APD experiments as follows: Pλ N = *W (47) qhc ,

Where N is the number of photons per pulse, P is the incident power in W or J/s, λ is the incident wavelength, and W is the pulse width in seconds.

216 10.3. Appendix 3: ROIC Specifications and Camera Interfacing

This section provides an overview of the commercial read out integrated circuit (ROIC)

used to fabricate these focal plane arrays (FPAs). It also provides some details on the camera

system used to interface with these ROIC. The electronic interface of the camera to the ROIC

is discussed as well as the imaging electronics used and their capability for image correction.

Finally the optics used in the camera are discussed briefly in order to provide a better

understanding of the imaging.

10.3.1. ROIC Specifications

The ROIC used in obtaining these results is a commercially available ROIC manufactured by Indigo Systems. It is designed primarily for P-on-N HgCdTe and InGaAs

detectors operating in the infrared, however it has a capacitive transimpedance input amplifier

(CTIA) type input units. This makes this ROIC especially well suited for interfacing to high-

impedance photodetectors such as those based upon the wide-bandgap III-Nitrides. The other

advantage of this ROIC is that it is designed to operate at room temperature unlike many other

ROIC’s designed for infrared that only obtain rated specifications at cryogenic temperatures.

An image of an example ROIC with a solar-blind FPA bonded to it is shown in FigureX 110 X below.

217

Figure 110. Indigo ROIC with 320 x 256 FPA bonded to it. The chip is only ~1 cm x

1cm in size and contains almost all of the electronics necessary to operate as a solar-blind UV imager.

In addition to these basic operating parameters this ROIC also has a number of features that make it especially convenient for the development of solar-blind imaging. The ROIC contains covers an array of 320 x 256 pixels for a total of 81,920 picture elements. This is approximately quarter of a standard NTSC video frame which means that each pixel can produce a standard video signal with a minimum of additional circuitry by doubling the lines and pixels to create a full frame. In addition to standard video frame rates of 30 Hz this ROIC can also be used for the rapid acquisition of images be supporting frame rates up to 350 Hz with proper electronics. The gain, skimming level, bandwidth limits, integration time, readout order, and power dissipation are all programmatically controllable thought the use of a serial word. This makes the array extremely versatile and allows reconfiguration on the fly to obtain

218 optimum imaging conditions of a given photodetector. In addition, there are a number of advanced features such as windowing, multiple frame readout, and anti-bloom control that are included on the ROIC, but that have not yet been implemented in the context of this work. The final big feature is the reliability, each of the individual ROICs are fully factory qualified before delivery, and thus guaranteed to have no dead pixels and to operate fully within specifications.

Imaging Conditions Indigo ICS9809 • 320 x 256 array 320 x 256 Active pixels, no windowing • 30 μm pitch Single Output mode • Adjustable Gain Frame Rate = 34.19 Hz • Adjustable Integration Integration time = 29.03 msec • NTSC compatible output Cint = 10 ff • Power dissipation < 90mW => Gain = ~ 1 x 10 12 V/A • Low noise CITA unit cell Operation in low power mode

Unbiased operation (Vdetcom = Vpos) Output skimming enabled, = 4.3 V Background subtraction performed No bad pixel replacement No other image processing

Figure 111. General overview of ROIC specifications and a list of the imaging conditions used to obtain all of the images presented within the context of this work.

Much of the versatility of this ROIC comes though the use of a serial word to set many of

the ROIC parameters. This requires a minimum of external connection and simplified the

camera implementation. Using the serial word the array can be tuned to by adjusting such

features as the readout mode, integration and gain control, the biasing of the different stages of

the ROIC, the switching of bandwidth limiting capacitors, and skimming. The specific bit of

219

the serial word as shown in TableX 6 X below along with the settings used to obtain all of the images within the context of this work.

B00: 1 Start Bit of Command word B01: 0 Indicates Command word B02: 0 ITR (Integrate Then Read) mode 0 = Integrate while read B03: 0 GC (Gain control) Cint = 210 or 10fF 0 = 10 fF B04: 0 PW0 (Power Control) B05: 0 PW1 (Power Control) 0,0 = Lowest current B06: 0 I0 (Current Control) B07: 0 I1 (Current Control) B08: 0 I2 (Current Control) 0,0,0 = Lowest Current B09: 0 AP0 (CTIA Bias Adjust) B10: 0 AP1 (CTIA Bias Adjust) B11: 0 AP2 (CTIA Bias Adjust) 0,0,0 = Lowest Current B12: 0 BW0 (Band Width Limit) 0 = No BW limit B13: 0 BW1 (Band Width Limit) 0 = No BW Limit B14: 0 IMRO (Integrate Multiple Readout) 0 = Single readout B15: 0 NDRO (non-destructive Readout) 0 = Destructive readout B16: 0 TS0 (Test Mode Select) B17: 0 TS1 (Test Mode Select) B18: 0 TS2 (Test Mode Select) B19: 0 TS3 (Test Mode Select) B20: 0 TS4 (Test Mode Select) B21: 0 TS5 (Test Mode Select) B22: 0 TS6 (Test Mode Select) B23: 0 TS7 (Test Mode Select) 0,0,0,0,0,0,0,0 = none B24: 0 RO0 (Readout Mode) B25: 0 RO1 (Readout Mode) B26: 0 RO2 (Readout Mode) 0,0,0 = Normal readout direction B27: 0 OM0 (Output Mode) B28: 0 OM1 (Output Mode) 0,0 = Single output B29: 0 RE (Reference Output) 0 = Reference disabled B30: 0 RST N/A B31: 1 OE_EN (Skimming) 1 = Skimming Enabled Table 6. Serial mode word bits and the functions they control.

220 10.3.2. Camera System Electronics

This ROIC requires only 14 signal connections in order to operate. This includes four clocks, one analog output, and 9 bias voltages: two variable DC biases, four fixed positive

voltages and three ground references. The interface specifications are provided below in TableX

7 X below.

9 DC BIASES VDETCOM 4.5 to 5.5 V Detector Common Vos 0 to 5.5 V Offset Skimming Voltage VPOS_CORE 5.5V Analog Positive Supply VPOS 5.5V Analog Positive Supply VPOSOUT 5.5V Output Positive Supply VPD 5.5V Digital Positive Supply VNEG 0.0V Analog Negative VNEGOUT 0.0V Output Negative Supply VND 0.0V Digital Negative 4 CLOCKS CLK VPD to VND Master Clock LSYNC VPD to VND Line Sync FSYNC VPD to VND Frame Sync (& Integration control) DATA VPD to VND Serial Mode Word 1 OUTPUT OUTA > 2.0 V swing Analog pixel data output

Table 7. List of the 14 signals necessary for the operation of an FPA based on the Indigo

ROIC

The camera system used in this research is CamIRa™ system manufactured by the SE-IR

Corporation of Goleta, California. The camera system contains all of the bias supplies, clock

drivers, and analog conditioning circuitry necessary to provide the DC voltages necessary to

operate the ROIC in a variety of different modes. The system and the ROIC are extremely

versatile, however only the necessary parts of the configuration that were used to interface the

ROIC as used within the context of this work will be discussed.

221 A pair of four-channel programmable bias cards provides the 9 DC bias signals necessary. One of the cards provides the four independent fixed 5.5 Volt rails necessary to power the different digital and analog stages of the ROIC. The second card provides the two variable voltages necessary to control the detector bias and offset skimming. Using the provided software these voltages can be arbitrarily varied in order to obtain the best operating conditions for the FPA.

The four clock signals necessary to operate the ROIC are generated by a programmable pattern generator card resident as a PCI card in the main camera control computer. The clock signals are then buffered and converted to the correct full swing signals at the camera head, immediately before they enter the ROIC. This setup is capable of producing 5.5 volt rail to rail pulses with a slew rate of approximately 1ns/1V. Since pixels are clocked in on both the falling and rising edge of the clock, this allows for a pixel rate of up to 400 MHz. The timing waveform can be arbitrarily controlled though the use of a graphical programming environment. In addition the serial word (set by the signal DATA) and the integration time can be set on the fly while the camera is operating thus allowing for real-time optimization of the image. The programmable timing pattern generator resident in the PC also provided the supplementary timing signals that are necessary for the frame grabber to operate.

The analog output of the ROIC is converted to a digital signal at the camera head by an analog to digital converter (ADC) card within the camera head. The camera head is shown in

FigureX 112 X with the side panel removed showing all of the bias clock, and ADC cards. The

ADC card also contains gain and image offset correction electronics to allow adjustment of the

ROIC output signal before conversion. The corrected signal is then converted by the card using

a 10 mega-sample per second (MSPS) ADC with a resolution of 14 bits per sample. The digital

222 data is then fed over a high-frequency RS 422 link the frame grabber where the digital signals

are manipulated in real-time to form the image. The frame grabber is a Coreco F/64 resident in

the control computer it shares a high speed date link with a digital signal processing (DSP)

board also resident in the computer. The signal corrections performed by the DSP board will be

discussed in Section 10.3.4X .X This pair of boards is capable capturing and processing full frame

800x600 video at a rate of 20 MSPS. However in this case the slightly smaller array used allows for higher possible frame rates, limited to 120 Hz by the choice to only use a single analog output; however, the ROIC can support up to 4 outputs, which makes frame rates up to

240 Hz possible with this system.

Figure 112. Focal plane array camera system used to record images presented in this

work. Shown with side panel removed to illustrate the internal electronics necessary to operate

223 the ROIC. NOTE: Although a Dewar is shown, the FPA is only operated at room temperature

under an ambient atmosphere.

10.3.3. Imaging Optics

The imaging optics used in this work are relatively simple, consisting only a single

element lens, a band-pass filter, and an aperture to shield the FPA from stray light. The camera

optics are included in FigureX 114 X thru FigureX 116 X latter in this section. The lens used has a 50.8 mm diameter and a 32 mm focal length; it is constructed of UV grade fused silica and is transparent from 200 nm to 2 um. It is used to collect light from in front of the camera and form an image on the FPA. In order to reduce the aberrations associated with this lens an aperture is used. It blocks the light from the outer edges of the lens and only allows the center of the lens to contribute to image formation. It also is constructed as part of the camera housing in such a way as to completely eliminate any stray light from reaching the FPA. Inside of this aperture is placed a 280 nm Band pass filter. This filter is not ideally necessary, but we found that despite the FPA being truly solar-blind, the visible light incident though the transparent substrate, on to the ROIC was detected. Since the response of the ROIC is fairly broad covering much of the visible on into the near infrared, and the FPA is only sensitive to a very narrow spectral band, the UV signal at 280 nm can be washed out by the broad ambient light signal thus creating problems with the imaging. The band-pass filter has a peak transmission of 17% at 280 nm and a visible rejection ration of ~ 3 orders of magnitude as

shown in FigureX 113 X below.

224

280 nm Band-Pass Filter 100

-1 17.8 % transmission 10 @ 280 nm

10-2

10-3

Transmission 10-4

10-5 260 280 300 320 340 360 380 400 X Axis Title

Figure 113. Transmission spectrum of the 280 nm band-pass filter used in camera head optics.

Using these optics images are collected from scenes falling into one of three categories:

either transmissive, reflective, or emissive scenes. The majority of the images presented within

the context of this report fall in the category of transmissive. This is the easiest type of scene to

image because the signal to noise ratio is the lowest since a bright 280 nm UVB source is

images by the camera though a selective aperture mask as shown in FigureX 114.X In this case a broadband florescent UV lamp is used as the source for 280 nm radiation. This type of fluorescent lamp lacks any phosphors and is made of a special glass to allow full transmission of the UV light. Although the dominant spectral line is the mercury line, the lamp also emits at a variety of other wavelengths covering much of the short-wave UV, including 280 nm. In this case the aperture mask consists of a patterned piece of paper; however, most objects are very good at absorbing UV and almost anything can be placed in from of the lamp.

225 280 nm Band-pass Test Image filter Aperture Plate

UV-FPA (280 nm) ~40% AlGaN UV Source (1) Xenon Lamp, UV Optic (2) Germ Lamp. f=3.2 mm Aperture

Figure 114. Schematic diagram of the optics comprising the solar-blind imaging system used to obtain the images presented in this work.

The other imaging geometry used is reflective. In this mode, the 280 nm UV radiation is reflected off of a patterned mirror. Few metals have good reflectivity at wavelengths as short as 280 nm however aluminum is satisfactory, and as such a polished aluminum plane is used as the mirror in this setup. The pattern used consists of thin layer of xerographic toner that has been bonded to the surface using a toner transfer process. This type of imaging setup has the advantage that relatively complicated scenes can easily be rendered on a computer and printed on special paper coated with a release agent. This print out can then be applied to the surface of the mirror and bonded using a combination of heat and pressure. Wetting the paper activates the release agent and leaves a selectively reflective mirror that can then be imaged with the

solar-blind FPA camera. The imaging geometry used for reflective imaging is shown in FigureX

115 X below.

226 Pattered Image UV Optic f=32 mm

UV-FPA (280 nm) ~40% AlGaN

UV Source

Figure 115. Schematic Diagram of the optics used to record reflective UV images using the solar-blind focal plane array.

The final imaging mode is used for objects that posses inherent 280 nm emission. In

emissive imaging the geometry is relatively straightforward, with the emissive object being

simply placed in front of the camera and directly recorded as shown in FigureX 116 X below. Out of all the modes discussed so far, this is the mostly likely to be used in a real application of a

UV camera. However it was not the preferred mode for this preliminary FPA camera due to the 80% loss in efficiency due to the filter used, and the relative immaturity of the technology.

By refining the processing, the ROIC can be masked, thus eliminating the need for a filter; and with further research, the imaging properties of the camera will improve. This should allow higher-quality imaging of lower flux sources such as coronal discharge and accelerated flames.

227

UV-FPA 30kV (280 nm) ~40% AlGaN

Figure 116. Schematic diagram of the imaging optics used to record images for emissive

UV sources, such as an electric arc.

10.3.4. Imaging Software and Image Correction

The imaging software used to control the camera is also provided by SE-IR. It contains all of the necessary controls to allow for setting of the various voltage biases necessary, it handles programming of the pattern generator to generate all of the necessary clocks, and it controls the image acquisition and digital signal processing necessary to operate the ROIC.

With the exception of the timing file needing to be pre-compiled prior to initializing the FPA, all of the setting can be changed on the fly without the need to turn off the camera; this includes the serial word through the use of a dynamically created patch to the compiled timing file.

The bias cards are resident in the camera head, but are controlled by the computer. The hardware allows for setting the biases to any values between – 10 volts and +10 volts; however, the software is aware that certain voltages may easily destroy the ROIC. As such the software give the use the ability to set up limits on the voltages that can be applied and then at run time limits the values to this range so as to reduce the change that the ROIC may inadvertently be damaged during optimization.

228 The pattern generator is programmed in a custom graphical programming environment.

The actual timing is broken down into a series of blocks that are then repeated at run time to form the full timing signal. This simplifies the programming since a single frame can consist of 100,000 or more clock cycles. Instead it is merely necessary to set up a single repeat of the timing unit for each of the main regions: line scan interval, horizontal blanking interval, and the vertical blanking interval. The software can then be programmed to repeat these units as necessary to generate the full timing patter. An additional benefit of this block-based structure is that signals such as FSYNC that control the integration time can then be arbitrarily controlled by adjusting the line where the begin integration block is placed. This allows for efficient control of the array and facilitates optimization to obtain the best images possible for a given

FPA.

The final responsibility of the control software is management of the digital signal acquisition and processing hardware resident in the PC. The analog output of the ROIC is converted to a digital signal at the camera head by the analog to digital converter (ADC) cards.

This digital data stream is then over a high-frequency RS 422 link to the frame grabber and

DSP boards where the digital signals are acquired and manipulated in real-time to form the image. The software manages the digital signal processing of the array data: this includes the ability to perform background subtraction, 2 point non-uniformity correction, bad pixel replacement, and output mapping with an arbitrary look up table.

The 14 bit video data is processed by the DSP board using 16 bit math operation to preserve the accuracy of the image. The first operation is a 2 point linear correction performed on each pixel using the equation Y = A * (X +B) with separate coefficients for each pixel. This operation takes care of the image subtraction via the B coefficient and can account for non-

229 uniformity using the A coefficient to scale the signal on a per-pixel basis. After the image

undergone this 2-point correction the pixel data is sent to a pixel re-ordering stage that allows

any pixel to be arbitrarily remapped to any other location on the array. This is makes it

possible to substitute bad pixels with their nearest working neighbor pixel; however for the

preliminary images within the context of this work no bad pixel replacement was preformed.

The final stage of the pixel pipeline is the look up table (LUT). The LUT contains a 65 K x 16-

bit table of values that are used to perform various data equalization and clipping operations.

The LUT is actually two separate tables, designed such that correlated double sampling can be used to update one table while the other is in use, and they can then be switched between frames. This would allow improved signal to noise ratios; however, the ROIC used here does not directly support this advanced feature and as such only a static LUT was used to obtain the images within the context of this work.

The corrected data stream is then fed to the frame grabber where it is converted to a video

signal and feed to the display monitor. The Coreco F/64-DSP16 framed grabber uses a 16-bit

digital interface to receive the image data from the DSP board. It then adds video overlays

such as the scale bar and any title text or other image annotation. The output signal is a VGA

compatible video signal that is sent to a standard variable frequency computer monitor for display. In addition to preparing the video signal for display the frame grabber also has a frame

buffer that can be used to capture still frames of video.

Still frames can be captured from the video screen and saved on the computer in the

tagged image file format (TIFF). The all of the still images from the FPA presented within the

context of this work have been generated in this way. However, the frame grabber is not

capable of capturing live video signals directly to video. As such, the movies presented here

230 not captured directly from the FPA and do not represent full frame FPA video. To capture

movies the VGA output signal from the frame-grabber is fed to a TV View Gold Scan

Converter which generates a standard composite video stream from the VGA video signal. A

Dazzle Digital Video Creator-USB video capture device is then used to captures this derived video stream. However do to bandwidth limitation of the universal serial bus (USB) the video stream is limited to only capturing video at a resolution of 312 x 240 pixels, 30 Hz interlaced.

This represents a 4 to 1 compression of the composite video frame and represents degradation

in the image quality present in the movies shown in AppendixX 4: Sample UV FPA movies.X

In addition to capturing still frames as TIFFs the frame-grabber can also capture the corrected FPA pixel data as an array of 16-bit values. This allows for further off-line image processing and can alternately be used to derive image statistics from the FPA image. The

pixel histograms presented in section 6.5X X FPAX Images and Discussion,X are created by processing this data.

231 10.4. Appendix 4: Sample UV FPA movies

10.4.1. Movie 1: Moving CQD Logo

Figure 117. This figure shows a single frame from the middle of the attached movie #1.

File Name: movies\1-CQDHT Logo.mpgTH

Length: 01:28 Format: MPEG-2 Video Size: 312 x 240 File Size: 32.7 MB

Description: This movie shows a paper cutout in front of a Germicidal lamp. The lamp and cut are moved in front of the camera a various speeds to show the uniformity and response characteristics of the array.

232

10.4.2. Movie 2: Dancing Exposed Electric Arc

Figure 118. This figure shows the first frame of the attached movie #2.

File Name: movies\2-DancingHT Arc.mpgTH

Length: 00:12 Format: MPEG-2 Video Size: 312 x 240 File Size: 4.6 MB

Description: This movie shows a low current high-frequency 30 kV electric arc as it jumps between two needles, one stationary and the other moved throughout the movie.

233

10.4.3. Movie 3: Reflection from a Patterned Mirror

Figure 119. This figure shows the first frame of the attached movie #3.

File Name: movies\3-MirrorHT reflection.mpgTH

Length: 00:54 Format: MPEG-2 Video Size: 312 x 240 File Size: 19.6 MB

Description: This movie shows an aluminum mirror to which a patterned black shadow mask has been applied; the image is of Professor Manijeh Razeghi at Northwestern University. This movie shows the patterned mirror selectively reflecting the light from the germicidal lamp, as it is move in next to the camera.

234

10.5. Appendix 5: Development of a Portable Camera System

10.5.1. Project Goal:

The goal of this project was to create a low cost portable camera system capable of

operating our infrared and ultraviolet focal plane array (FPA) imagers. The camera head

weights about 2 Kg, and is less than 12”x6”x4” in size (including the Dewar) making it easy to

mount on a tripod. The bulk of the image processing and display is done on a laptop computer

with only two small wires, power and serial data, connected to the camera head. The entire

system readily fits in a hard case and can easily be transported from site to site to allow for easy

demonstration.

Figure 120. The system is small, lightweight, and can be easily packaged for transport to diverse locations in a hard carrying case requiring only LN2 and a laptop computer for operation.

This is in contrast to the commercial system that the Center for Quantum Devices uses

for lab bench testing and characterization of FPAs. The commercial system weights more than

50 Kg, with a large camera head and requires 5U of rack mount space to hold the power

235 supplies and image processing electronics. In principle this system can be moved, but the

camera head has 4 separate connections, two of which are 32 pin parallel cables making the

system very bulk and difficult to move. In contrast to this, there also exist extremely small

specialized cameras that forgo all versatility for size and weight. These compact systems tend

to employ welded Dewars, and are tied to a specific ROIC. There is a complete lack of mid-

sized system that maintains a degree of versatility while still possessing a small form factor

suitable for portability. The goal of this project was to fill that gap creating a portable system with an interchangeable dewar that is capable or operating all of the ROIC and FPA

combinations that are present at the CQD.

10.5.1.1. Systems Overview:

The system consists of a pour-fill LN2 dewar, electronics package, power supply, and a laptop computer. The dewar houses the FPA, regulates the ROIC temperature, provides an optical interface for the FPA, and contains a series of connectors to facilitate electrical connection to the ROIC. The electronics package is designed to screw onto the back of the dewar and contains a set of 4 of 5 cards responsible for providing biases and clocks to the

ROIC, digitizing the analog output of the ROIC, and marshaling data between the laptop computer and the camera head. These cards are interconnected by a front and back plane, with the front plane also serving as the dewar interface.

236

Figure 121. Schematic diagram of the components that make up the portable camera system

The power supply is a small “brick type” supply similar to that used for a laptop

computer that takes 110 VAC and converts it to +/- 12 VDC and +5 VDC. All other voltages

necessary to operate the camera are generated from these supplies within the camera head. As

an option a rechargeable battery is provided that can alternatively be used to operate the system

for up to 24 hours on a single charge.

For the laptop, any modern laptop with at least windows 2000, a direct-X capable video

card, and a 100 Mbps Ethernet adapter is sufficient for TV quality frame rates. Faster hardware

will allow the system to operate at higher frame rates. The laptop can either be operated from

line power if available or from the internal battery. The level of CPU utilization for TV quality

frame rates is sufficient to allow for moderate battery life.

10.5.1.2. Hardware:

The hardware consists of custom front and back planes with 4 interchangeable custom

th daughter cards. The 5P ,P top, card serves as the control card. It contains a sub-daughter card

237 with a programmable FPGA core and an Ethernet interface. The control card is responsible for

generating the timing necessarily to drive the ROIC. A separate card is then used to buffer

these clocks and adjust the voltage level programmatically. The control card also controls the 8

available DC bias generated by the other cards. Control is by means of commands sent over

the Ethernet bus to the control card. In addition the control card reads out the ADC values, packages them and sends them through the same Ethernet interface to the laptop computer for processing and image display.

Figure 122. Overview of the electronics package layout

10.5.1.3. User Interface:

Since the system is designed to use a laptop computer for image processing the bulk of

the user interface is implemented in software. This allows for the development of a rich

interactive environment for control of the camera system. The UI design conforms to the

common conventions of windows application development. The main program window

consists primarily of the video display with a title bar for and border for manipulation of the

window. A menu bar from which other windows can be spawned provides access to the

various interfaces of the camera system. A status bar at the bottom of the window provides

feedback on the status of the camera and provides context related help as the user navigates the

238 software. The main software is able to spawn additional floating windows to control the

camera biases, ADC conversion, image processing, and brightness and contrast among others.

These windows can be simultaneously open and manipulated to provide the level of the control

that a particular user deems necessary at any given time.

Figure 123. Screen capture of the various windows that make up the portable camera system user interface.

10.5.1.4. Embedded Software:

The embedded software operates on a VIrtex-4 FPGA situated in the control card. The

code is a mixture of VHDL for definition of the logic units and C-code for operation of the

embedded power-PC microcontrollers. A simplified state machine implemented in VHDL is

sued to generate the timing pulses necessary to output the clocks required by the ROIC.

Programmable counters are used to allow for dynamic control of some of the timing while the

camera is operating. Another state machine is used to marshal data from the ADCs into a series

of buffers. The power-PC microcontroller then handles sending of this data to the imbedded

239 Ethernet controller where it is transmitted to the laptop computer for processing. In addition the power-PC microcontroller received packets from the Ethernet controller and interprets commands for the camera head. An internal I2C bus is used to control the various other elements of the camera system hardware. In addition limited read-back capabilities are implemented to allow for querying the status of the camera head. During normal operation all commands are transferred over the bus to the laptop computer where the user interface is actually implemented. However a series of diagnostic LEDs and a reset switch are also provided by the embedded software as a limited back-up user interface to ensure that the camera head is operating and responding to commands from the main user interface.

Figure 124. Virtex 4 Mini-Module used to run the embedded portion of the software, also

shows the EEPROM, RAM, and Ethernet-PHY used by the Virtex-4 FPGA.

10.5.1.5. DirectX Software:

The software consists of C++ code written using the Microsoft Direct Show API to provide a low-level means of accessing the video rendering hardware available on the computer. Low level access to the Ethernet card is provided by an open source packet driver called win-pcap to avoid the overhead associated with a full TCP/IP stack implementation.

The software is broken down into a number of sub-components, written as independent

240 DirectShow filters. There is an input module that handles communication with the camera, controlling the sending of commands and the generation of the raw frame from the network stream. A second module handles the image correction and applies per-pixel offset and scaling, and allows for arbitrary bad-pixel replacement. A third module controls the conversion of the

16 bit image data into a 32-bit color image for display on the computer monitor; it does this by implementing a 16bit look-up table and writing directly to the video card’s memory for maximum performance. All three of these modules are controlled by the main application that handle inter-module communication, and presents the user with interfaces for controlling the various modules in a seamless fashion.

Figure 125. Schematic diagram showing the building blocks (and their sub components) that constitute the portable camera system software.

10.5.2. Software Design:

The software is broken down into five main components: The MainX Application,X network

based camera interface unit, the image correction and processing unit, the brightness and

contrast look-up table, the stock video renderer, and the overarching application that controls

these components and presents the used interface to the user. With the exception of the

Microsoft Windows Direct X API, the packet driver used to access the network at the raw

241 layer, and the stock video renderer used to present the video frames, all of the software

necessary for the camera was custom written as part of this work.

10.5.2.1. Main Application

Installation of the portable camera system software places a shortcut on the user’s desktop that can be used to start the main application. Most settings relevant to the operation of the camera and the previous state of the cameras are automatically stored in an INI file camsys.ini placed in the same folder as the executable.

Upon program startup the application will display the location of the INI file used in the status bar. The window will default to the previous location on the screen, but will automatically size to 2x the size of the video data selected in the INI file. At startup none of the ancillary windows will initially be open. If the camera is running and sending valid data, then the program will immediately start displaying video, otherwise the program will default to a bank screen which it will render at the standby frame rate of ~10 FPS.

The main window follows standard windows graphical user interface standards. It consist of a title bar at the top which can be used to move the window, as well as minimize, maximize, and close button in the upper right hand corner. Immediately below the title bar is the menu. At the bottom of the window is a status bar.

242

Figure 126. Main program window

The menu contains the bulk of the user-interface for this window in the form of a number of drop-down menus from which the user can select functions. In addition, most of the more common functions have keyboard equivalents listed on the menu. Upon highlighting a command in the menu, a description of that command’s operation is shown in the status bar below.

243

File Exit (Ctrl + Q) Exits the program Save Image (Ctrl + S) Prompts User for a File Name and Type, then Grabs and Saves and Image Record Video Configure a codec and start recording video to disk Save NUC Save the current NonUniformity correction to file Load NUC Loades NonUniformity correction from file, and enables image correction Save 16bit Frame (Raw) Saves the raw16-bit grayscale frame Save 32bit Frame (Raw) Saves the raw colorized 32-bit RGBA frame Save LUT (Raw) Saves current LUT used to colorize the 16-bit data and adjust brightness and contrast

Image Processing Brightness and Contrast (Ctrl + C) Adjust Brightness, Contrast, Gamma, and Colorization. 1-Point Correction (Ctrl + 1) Preformes 1-Point Background correction and limited guided pixel replacement. 2-Point NUC (Ctrl + 2) Preformes 2-Point NUC and Histogram Guided pixel replacement BackGround Subtraction Take the current corrected frame as the new background subtraction Clear BG Subtraction Turns off backgroung subtraction Image Normalization Takes the currently visable frame and uses it to normalize the data Clear Normalization Turns off the frame normalization Cear Pixel Replacement Turns off the bad pixel replacement Adv. Image Processing (Ctrl + P) Control Image Correction Parameters

Camera System Camera Setup (Ctrl + V) Show VideoScope, set ADC offsets, adjust select biases. Turn Camera Off (Ctrl + X) turns off the camera head, and powers down the ROIC Turn Camera On Turns on the camera head, and starts receiving data Network Setup (Ctrl + N) Configure network streaming, Control camera Image Rendering (Ctrl + R) Control the video renderer (Microsoft's VMR7)

Window Fullscreen (Ctrl + F) Switch to fullscreen display, ESC to exit. Size to Video Sizes window so 1 pixel = 1 FPA Pixels Size to 2x Video Sizes window so 4 pixel quartets represent each 1 FPA Pixel Freeze (space) Freeze Video Display, (Repeats last frame so you can adjust processing of still frames) Start (space) Resumes feeding in new frames from the network Stop Stops recieving and processing frames

Help About About the Portable Camera System software interface DebugBreak Does nothing (Throws an exception in the message loop of a debug build)

Table 8. A list of the available menu functions as well as their descriptions.

The status bar indicates the current status of the camera system. The left side shows the current frame rate at which the camera is running. The right side of the status bar indicates the status of the software, and camera system. The right most icon indicates weather the camera system has been turned on by the software or not. It does not sense the actual state of the camera system, but only what the software has requested the camera system to do. Actual operational status of the camera should be confirmed by inspecting the light on top of the

244 camera itself. The left icon indicates the rendering status of the software: either running, paused, or stopped. In run mode the camera continues to deliver new frames as they become available from the network. In stop more the camera ignores incoming frames, and does not render any video, and thus it is not possible to adjust the image shown. In pause mode the last frame received from the network is continuously sent to the camera. This allows for image processing to occur on a static frame of video. This means that the brightness and contrast can be changed, that bad-pixel replacement can be adjusted, and that frames can be saved to disk.

The central part of the status bar is used to indicate various informational messages to the user. At startup it displays the INI file location used. When using the menu, it displays menu help strings that can be used by a novice user to determine the operation of various menu commands before they are executed. The third and most important function is the provide per- pixel analysis of the image.

The central video window displays the 32-bit colorized video coming from the camera. It tries to maintain the aspect ratio of the video frame despite the size of the application window.

This is achieved by adding black bars to the video as appropriate. The main window can be sized to the video by using the window menu. Clicking on the video in the main window causes the pixel value at the center of the crosshair to be displayed in the status bar. The behavior depends upon which mouse button is clicked. Clicking the left mouse button cause a message to appear in the status bar indicating the horizontal and vertical location of the pixel as well as the corrected 16 bit hexadecimal value of the currently selected pixel. This is the raw grayscale value before the application of brightness and contrast adjustments, and colorization.

As an additional aid to assist the user, the selected pixel is displayed as a red dot for the frame from which the data was extracted.

245

Figure 127. Right-Click pixel processing details as shown on the status bar.

Right clicking on the video window causes detailed pixel processing information to be

displayed in the status bar. The text is intended to show the approximate algorithm used to arrive at the pixel value displayed on screen. The fist two numbers in parenthesis are the x and y locations of the selected pixel, respectively. The number before the colon added to this is the bad pixel replacement offset applied; this is in 1D linear coordinates rather than 2D x-y coordinates. In the case where the bad pixel offset is non-zero the following number are relatively meaningless since they are replaced during final construction of the frame. In the first set of parenthesis after the colon, the raw ADC value is displayed followed by the pixel offset applied to that pixel. This is followed by the floating-point representation of the gain correction applied to the sum of those numbers. The resulting corrected 16-bit pixel intensity is displayed after the equals sign. This is followed by the hexadecimal representation of that decimal value in parenthesis: this is the only number shown in hexadecimal; all other numbers displayed are base-10.

The camera setup window is the heart of the camera system control, and contains the most commonly used setting in one central location. It is opened by selecting the [camera setup] item under [camera system] menu, or alternately by pressing the (Ctrl + V) key combination, where “V” stands for Video scope.

This dialog includes a large central display of all of the pixel intensities in video-scope form (pixel intensity vs. pixel number). To the left the 4 available ADC offset can be controlled with the sliders. At the upper right the ROIC and ADC timing can be adjusted. At

246 the middle left three of the most common ROIC biases can be fine-tuned through the use of sliders. At the bottom left are controls for turning on and off the camera system.

Figure 128. Video scope and camera setup user interface

The Video scope shows the intensity of every pixel in the focal plane array as a scatter plot of intensity versus linear pixel position. This graph is updated in real time concurrent with the display of video in the main program’s video window. The data is displayed for 16 bit data before brightness and contrast adjustments are applied.

The display starts automatically when the setup window is opened, but can be stopped by pressing the stop button. The display can then be restarted by depressing the adjacent start button. The video scope can either display a 14 bit range or a 16 bit range, i.e. the vertical range can be adjusted by a factor of 4 according to the video signal. Raw uncorrected ADC data is only 14 bits while corrected data is usually 16 bit.

247 If any image correction is in effect at the time the window is opened the correct

checkbox will be checked. Un-checking this box will save any image correction currently in effect and remove it from the image processing pipeline so that the raw ADC signal can be seen. This is especially useful to ensure that ADC clipping does not occur on the raw signal.

The image processing can then be re-applied by again checking the correct checkbox. Care should be exercised with this checkbox since it is not continuously updated, and the application of image correction via an alternate means while the setup window is open may not be correctly reflected by this check box.

At the bottom between the close button and the correct checkbox a status display is updated periodically that shows then number of pixels that come within ~5% of the upper and lower clipping limits specified in the brightness and contrast window. This text displays the decimal number of pixels that fall within this range as well as the percent of the image that this corresponds to.

ADC Offset allows setting the ADC offsets. The camera system can have up to 4

different ADC cards, and each ADC card has an independent offset adjustment. The four

vertical sliders can be used to independently control the ADC offset applied to each ADC.

However in practice it is only necessary to adjust all four biases in sync after bringing the four

video signals into initial agreement. To assist in this task, there are two checkboxes at the bottom of this section.

The lock ratio check box stores the current relative position of the ADC offset, any

adjustment of one of the ADC offsets will be correspondingly applied to the other three. This

causes all four slider to effectively move in unison. In the event that one of the sliders reaches

248 the end of its range it will cease to move, but upon returning the slider group to the center it

will again maintain the same pre-recorded relative offset.

The colorize checkbox is a handy tool that highlights the ADC values from each of the 4

ADC channels in a distinct color on the video scope display. This can be extremely useful for bringing the 4 ADC channels into initial agreement before applying the lock ratio setting.

It should be noted that in single ADC setups the colorization will be meaningless and

only one of the sliders will be used. In this case it is best practices to leave on the lock-ratio

setting, and not use the ADC channel colorization.

The Clocks section is used to control the ROIC timing. The first option frame rate

contains only one option 300 FPS, and like the goggles, it does nothing. It is intended for

future functionality. The other two options control the integration time, and the ADC offset,

respectively. The integration time is controlled in units of scan line pairs, and for a 320 x 256

array ranges from 0 to 128. This can be interfered as 0/128 to 128/128 in terms of fraction of

frame time spent integrating the image. The ADC offset controls a programmable clock delay

present on the ADC clock output, and can be used to fine tune the point at which the ADC

samples the analog pixel data out.

The Bias fine control allows for three of the most important ROIC biases to be

controlled directly from the setup window the first part of the name of each as well as the

current value is displayed at the top of each slider. The extents of the range of the slider are

displayed at the end of the slider. These values are determined by the upper and lower limits

set in the main biases window. The choice of biases shown on the setup window is specific to

the ROIC and can only be controlled through INI file by adjusting the BIAS01, BIAS02, and

BIAS03 keys in the SEUP section of the INI file.

249 The 1-Point Image Correction window is one of the two main image correction interfaces available. It is opened by selecting the [1-Point Correction] item under [Image Processing] menu, or alternately by pressing the (Ctrl + 1) key combination, where the number “1” stands for 1-point correction.

Typically a 1-point correction is performed by pressing the Capture button, waiting for

frames to capture, and then pressing the Apply button to enable the correction. If the Capture button is initially disabled, then the Clear button must first be pressed since the camera will not

allow the user to perform a 1-point correction while image correction is already operating on

the image.

Figure 129. Background subtraction and limited bad-pixel replacement user interface

250 Upon successful capture of frames the check box to the right of the capture button will briefly illuminate and the data will automatically be analyzed. For each pixel the values from all of the frames will be averaged to obtain the mean pixel intensity. This value will then by

compared against the target specified in the edit box and the offset necessary to make the mean

pixel intensity equal to the target value will be calculated. A histogram of number of pixels

versus offset value will then be displayed on upper graph with a log-linear scale. At the same

time a running variance will be calculated for each pixel, and a similar histogram of number of

pixels versus variance value will then be displayed on the lower graph with a log-linear scale.

Bad pixel replacement can then be performed on the image using the displayed data.

The user is presented with two red bars on each of the graphs, pixels between the two bars are marked as good, and any pixels that fall outside of this region are marked as bad. The red bars can be moved by clicking on them and dragging them, alternately the graph can be double clicked at a point of interest and the nearest limit will immediately snap to the cursor position.

If the user wishes to ignore one of the limits the user can double click on either side of the graph and the respective maker will move to that extent. This provides an intuitive visual manner of establishing the bad-pixel map. Text is displayed showing the number of bad-pixels and the breakdown of that number.

In normal operation the user needs to press the Calculate and then the Apply button to store the bad-pixel map and enable bad pixel replacement. However when computing resources are available, the Real Time check box can be enabled. It will perform the above calculate-apply sequence automatically in real-time as either of the markers is move on the graphs above. This provided a much more intuitive experience as the user can actually see the bad-pixels disappearing as each of the histogram buckets falls outside of the selected region.

251 The algorithm used to determine bad pixels uses a nearest-neighbor clockwise search for a non-bad pixel. In the event that a nearest neighbor non-bad pixel cannot be found within 32 attempts a random nearest neighbor is chosen from among the 32 possibilities ensuring that large areas will be filled in as that pixel is corrected, eventually leading to a valid pixel at the edge of the bad region. This complex (potentially 320 x 256 x 32 = 2.6 million operation) search algorithm is why real time replacement is not chosen as the default option, though in practice the CPU load is manageable for reasonable quality arrays w

The 2-Point Image Non-Uniformity Correction window is very similar to the 1-poin- correction dialog with the exception of containing an additional capture button and a histogram display of the pixel gain. It will be necessary to capture two frames, one bright and one dark to get a full 2-point correction. The mean values of the bright and dark frames are used to calculate a gain and an offset term for each pixel. The values calculated are displayed in the upper two histograms. The bottom histogram contains a running variance calculated on the dark frames. Similarly to i-point correction, all three of these histograms can be used to select bad-pixels for addition to the replacement map.

252

Figure 130. Two-point NUC and bad-pixel detection user interface

The camera system is capable of recording corrected video to the computer hard drive for latter review. In order to conserve hard drive space on-the-fly compression is performed on the video stream. The software automatically determines the list of available video compressors installed on the host computer. The user must select from among the available video compressor presented in the dialog box. The software does not check the suitability of each compressor for compressing RGB32 video and poor choice of compressor may require that additional filters be inserter to convert the data to a suitable format for compression, potentially

253 hurting performance significantly. The default video compression settings will be used for that

compressor, and no attempt is made to allow control of the video compression parameters.

Figure 131. Video capture user interface

10.5.2.2. Camera Interface Unit:

The camera interface unit is a custom Microsoft DirectShow filter responsible for managing communications between the camera head and the software. Its output is a 16 bit anthropomorphic video stream reconstructed from the network communications with the camera. To facilitate the control of the camera, the camera interface unit exposes a custom property page with four separate tabs. However, many of these features are indirectly controllable through the video setup windows in a more abstracted manner via the sliders. The window can be spawned by selecting the [Camera System] menu and selecting [Network

Setup]. Alternately the keyboard combination of [Ctrl + N] can be used, where “N” stands for

“Networking.

254 Frames are sent by the portable camera system to the laptop when the system is operating. Due to the 15000 byte limitation of the Ethernet protocol the frame data must be

split up across a number of different packets. The software is then responsible for collecting

these packets, reassembling the frame data in the commuter’s memory, and handling frame

synchronization and timing. In the case of a 320 x 256 FPA 128 packets of 1310 byte each will

be required to generate a single frame. This corresponds to about 40 Mbps at 30 FPS and

requires a minimum of 100 Mbps connectivity, however the camera system can operate over

Gigabit Ethernet for much higher frame rates. Each packet consists of a small header used to

allow for the MAC level transport, and two 6 byte numbers that contain the current scan line

and the frame to which that scan-line goes. By default the Portable Camera System software

will cease building a frame and display it as soon as any portion of the next frame is received.

The Portable Camera System is designed for direct connection to the laptop computer.

Transmission over a public network is possible but not encouraged due to the high probability

of image corruption.

Figure 132. FPA camera system image transport packet layout.

255 The first tab of the camera control interface contains the advanced bias adjustment controls. This window lists the name, value, and limits of each of the available biases. The biases are listed by their hardware addresses with the first 8 being the DC biases available on the bias card and the clock card. The ADC offsets are also included on this list in their raw 16- bit form; all of the other biases listed are in terms of voltage at the output terminal. The final bias listed is the clock bias. This is voltage used to dive the clock driver and must be in the range of 3.5 volts to 5.5 volts to prevent damage to the clock driver chip.

Figure 133. User interface for low-level control of programmable camera biases

To change a bias the user merely selects the edit box immediately adjacent to the bias

name of interest and enters a value. To actually tell the camera to set the bias the user then

clicks the adjacent set box. This causes the software to check the bias against the limits

specified. If the specified value is larger than the high limit then the value of the high limit will

256 be used instead of the specified value. If the value specified is less than the lower limit then the

lower limit will be used: thus the data is effectively clipped at these limits.

The second tab of the Camera Control interface is the Serial Word control interface.

This tab contains elements that can be used to control the command word used in ROICs that support that functionality. Many ROICs support advanced features through the serial word. IN

The case of the Indogo 9809, the serial word is used to send a 32 bit control word to the ROIC.

The breakdown of bit functionality is shown graphically and the user can change a select set of the ROIC features though the list boxes on the right of the screen. The details of the ROIC are

given in AppendixX 3: ROIC Specifications and Camera Interfacing.X

Figure 134. Control serial word interface for Indigo 9705 & 9809 ROICS

The third tab of the Camera Control interface is the Windowing Word control

interface. This tab contains elements that can be used to control the windowing implemented

257 by the ROIC. However, not all ROICs support this functionality, and the Portable Camera

System will still attempt to read out a full frame even if the windowing is used to reduce the region of interest.

Figure 135. Windowing serial word interface for Indigo 9705 & 9809 ROICs

The fourth and final tab of the Camera Control interface is the FPA Cameras Source control interface. This tab primarily contains diagnostic information on the rendering, such as the current frame rate being rendered and the last frame sent. The only control it contains is a read network check box that is identical to the pause feature available though the main windows interface.

258

Figure 136. FPA Camera Source interface, showing diagnostic information

10.5.2.3. Image Processing Unit:

The Image Processing Unit is another custom Microsoft DirectShow filter responsible for

taking the raw 16bit data from the network interface unit, and applying various image

corrections to the data. It is the most computationally intensive part of the camera system, and

thus contains a specially tuned vectorized processing pipeline that takes advantage of the SSE2 extension available on modern processors. Most of the image processing is controlled directly thorough the one and two point correction interfaces, however this component exposes a custom property page to allow for advanced adjustment of the image processing. In addition, a number of debug related image processing steps are available from this interface.

The actual image processing path is broken down into four stages: Pre-processing, Image

Correction, Post, Processing, and Finishing. Up to 7 distinct operations can be performed of the data in the worst case scenario, though in general practice, only three are commonly used.

259

Figure 137. Advanced image processing user interface

The pre-processing steps are really used only for debugging purposes: the default

option of No Effect is used in the majority or cases. Flat will ignore the incoming image data

and replace it with uniform pixel intensity across the entire frame; this can be useful for

inspecting the image correction in place. Grid will discard the incoming pixel data and replace it with an alternating black/white checkerboard pattern; this is useful for seeing the pixel replacement patterns in effect. The last three options create various gradients that can be used for inspection of the LUT’s image transfer characteristics.

The image correction stage is performed after the Pre-processing and contains check boxes that allow for selectively enabling and disabling the various image corrections that can be applied to the data. Background Subtraction applies a 16 bit floating point addition to the

pixel data. Uniformity Scaling then multiplies that value by a 32bit scale factor and divides by

2^16 to yield a properly scaled pixel value. Bad Pixel Replacement then reads through the

array and performs pixel-reordering as specified by the bad-pixel replacement map.

Post processing contains more options that are primarily of use for debugging purposes.

The default option of No Effects should be used in the vast majority of cases. The ADC

260 options are only useful for a quad-output ROIC; they will select data from only one of the four

outputs, replacing the other three. This can be useful for checking the uniformity of the various

ROIC outputs. One pixel stats will replace the entire field with the last pixel interrogated with a right click, and write the numeric value of this pixel to a file for each frame; this is a useful debugging output for analyzing the noise of a single pixel, particularly if the saved data is post- processed.

The finishing steps are useful for improving the quality of mediocre images.

Averaging performs a temporal filter on the output data by averaging together the selected

number of frames before rendering. This can significantly reduce the high frequency noise of

the image at the expense of frame rate, and is a good step to use before taking still images if

quality is a concern and the scene can be sufficiently composed before capture. The spatial

filter performs a Gaussian blur on the image with the selected q-factor. The actual weighting used is displayed above the drop down menu, and is based upon integration of the two nearest

neighbors, and requires a two-pass approach due to the required 2D integration. Spatial

filtering averages together adjacent pixels and can significantly reduce the effect of isolated bad

pixels or non-correctable pattern noise. However, it does this at the expense of edge crispness without sacrificing the frame rate. Generally a small amount of spatial and temporal filtering can be implemented to achieve an improved image with out noticeable affecting the imaging.

These values are all updated in real time and trial and error is encouraged to obtain an acceptable tradeoff.

261 10.5.2.4. Brightness and Contrast Adjustment:

The video is inherently a 16-bit grayscale image after image correction. However modern computer displays are not capable of displaying more than about 8bits worth or grayscale information. The brightness and contrast window is responsible for controlling the transition from 16-bit grayscale pixel intensities to 32-bit RGB pixels. This is accomplished by using a 32 bit look-up table with 2^16 entries. This allows the complicated floating point math necessary to create the LUT, to be performed once, and the value saved. This reduces the computational strain on the pixel pipeline, but can introduce a slight jitter when the values are adjusted, and the LUT need to be recreated. To aid in controlling the pixel mapping specified by the LUT, a custom property page is exposed that contains three sliders, for Brightness, contrast, and Gamma. The brightness and contrast adjustment window can be spawned by selecting the [Image Processing] menu and selecting the first item, [Brightness & Contrast].

Alternately the keyboard combination of [Ctrl + C] can be used, where “C” stands for

“brightness and Contrast”.

Figure 138. Brightness and contrast user interface. Allows adjustment of gamma and colorization

262 Brightness, Contrast, and Gamma are controlled by the three sliders at the top.

Brightness controls the median value of the video. Contrast controls the spacing between different intensities assigned from the raw 16-bit values. Gamma controls the curvature of the transfer function used to partition the available intensities, and isn’t as intuitive as the previous two. More simply gamma controls where in the intensity range the available contrast is more concentrated. In practice gamma is used to correct non-linearity of the response of commercial displays, but in working with infrared video it is extremely useful: careful adjustment of the gamma can allow high contrast imaging of relatively cold sources like humans while still having available contrast not to saturate for hot objects that may be in the environment. The formula used to generate the LUT is as follows:

263

//Quasi Normalized the constants to generate floating point numbers float f_c = Contrast; // Range = 0 to 2, 0..1 => Decrease f_c /= 0x7FFF; // 1..2 => Increase float f_b = Brightness;//Range = 0 to 2, 0..1 => Decrease f_b /= 0x7FFF; // 0..2 => Increase float f_g = Gamma; // Range = 0 to 2, 0..1 => Positive Curviture f_g /= 0x7FFF; // 1..2 => Negative Curviture for(iPixel=0; iPixel <= 0xFFFF; iPixel++) { if (iPixel < m_DataMin) { // Clip the data to the DataMin LUT[iPixel].rgbRed = (BYTE)0; LUT[iPixel].rgbGreen = (BYTE)0; LUT[iPixel].rgbBlue = (BYTE)0; } else if (iPixel > m_DataMax) {//Clip the Data to the DataMax LUT[iPixel].rgbRed = (BYTE)0; LUT[iPixel].rgbGreen = (BYTE)0; LUT[iPixel].rgbBlue = (BYTE)0; } else { // Generate lookup table values f_i = iPixel; // Normalize pixel index f_i /= m_DataMax;

// Adjust the Brightness & Contrast // switch algorythms in order to uniformaly cover range if (f_c < 1){ temp = (f_i - 0.5) * f_c;//Center on zero => expand symetricly temp += (f_b-0.5); } else { temp = (f_i - 0.5) / (2 - f_c); temp += (f_b - 1)/(2 - f_c) + 0.5; }

// Clip the data to the range 0 to 1 // (Gamma Correction goes Asymtotic and can cause overlows otherwise) if (temp > 1){ temp = 1; } else if (temp < 0) { temp = 0; }

// Adjust the Gamma // switch algorythms in order to uniformaly cover the whole range. if (f_g < 1){ temp = powf((float)temp, f_g); } else { temp = powf((float)temp, (1/(2-f_g))); }

// Colorize the floating point data Colorize(temp, iPixel); } }

Colorize controls the way the brightness and contrast adjusted video intensities are converted to RBG colors using one of the available colorization schemes listed at the bottom.

264 Grayscale and inverse grayscale are the most common. IR Colors does a sepia tint such as is

commonly used for AFM images. Blue-Gray-Red displays middle intensities in grayscale and

highlight the extremes of this range in red or blue to aid in picking out hot and cold objects in

the environment. Custom 1 & Custom 2 are reserved for future use. Rainbow applies a full- spectrum colorization to the video. RGB565(debug) is only used for debugging in conjunction

with the FPA camera simulator and interpreters the 16-grayscale intensities as if they were

packed RGB values.

10.5.3. Hardware Design:

All of the hardware implemented to support the portable camera system has been

custom designed specifically for use in this system. The boards use a two layer design with the

bulk of one layer dedicated to a common ground plane for improved noise performance. Plated

vias are used to make connections between the layers. The boards were made using an external

contractor to achieve accurate registration of the layers and a minimum feature size of 5 mills.

Separate analog and digital power rails and grounds are maintained to minimize noise. In

addition all components are bypassed with the parallel combination of a 0.1 μF and 10 μF

capacitor on the supply rails. A 10 ohm series resistor is also added to the power supply input

of all high frequency devices to increase the RC constant and provide better power supply

isolation.

10.5.3.1. Modular Back & Front Plane design:

In order to facilitate the ease of reconfiguration a modular front and back plane design was adapted. All cards have an 80 back plane connector that provides power, ground, and

265 digital connectivity between the cards. At the front of some of the cards is a 20 pin connecter

that connects to the front plane: the front plane is responsible for distributing the various card’s inputs and outputs to the dewar connectors. This means that cards within the camera can easily be swapped out as necessary, and adapting to a new ROIC or dewar only requires the creation of a new front plane card to manage distribution of the new signals.

Figure 139. Cross-sectional view of the camera system electronics showing the front and back planes showing the attachment to the dewar.

10.5.3.2. Embedded CPU Daughter card and power supply.

The top card in the system is the only card that is not connected to the front plane. This card is responsible for generating the necessary voltages for operation of the camera system electronics and houses the embedded CPU (Virtex-4 mini-module) that controls the operation of the camera head.

The input power connector supplies this board with +/- 12 VDC and +5 VDC from the

external power supply. This is then used to generate the +/- 12 VDC analog supplies, the +3.3

VDC analog power supply, the +3.3 VDC digital power supply, the +2.5 VDC digital supply,

266 and the +1.2 VDC digital supply. The analog supplies and the +3.3 VDC digital supplies are

connected to the backplane on this board. In addition, the analog and digital ground planes are joined together on this board and connected linked to the power supply.

Figure 140. Embedded CPU Daughter card and power supply board.

This card is also responsible for connecting the two 14 bit ADC busses, the 8 bit clock

bus, and the two-wire I2C bus, and the relay control output to the mini-module. Data is

transferred from the backplane to and from the mini-module using a number of traces laid out

on this board.

The embedded CPU daughter card and power supply board also contains the debugging

interfaces for the camera system. A RS232 transceiver is provided to display debug output

strings on a debugging terminal. A JTAIG headed is also provided to allow for updating the

embedded FLASH ROM or using the debug facilities of the mini-module.

This board provides LEDs to indicate the operational status of each of the digital power

supplies. A done LED is provided to indicate when the embedded software has completed

booting, and is ready to start receiving data. Three additional LED indicate the operational

267 status of the camera system. In addition to the LEDs on the Main board the Virtex-4 Mini-

Module has several LEDs associated with the Ethernet PHY that indicate the presence of an

Ethernet link, the speed of the link, and any traffic present on the link.

10.5.3.3. 2-chanel 20 MSPS ADC cards with programmable Offset

There are two dual-channel ADC cards in the system for a total of 4 analog inputs. For each channel, two high-speed AD8021AR video op-amps are used to buffer the input and apply approximately 2x gain to the signal. The first op-amp is set up for unity gain with a 10 MZh low pass filter on the input. The output of this op-amp is mixed with an adjustable DC signal to allow for +/- 10 volts of offset. A 0 to 3.3 VDC signal is generated with a 16 bit serial DAC.

The output is adjusted to span a rang of +/- 10 VDC using an OPA227 op-amp. The scaled value is then buffered by another OPA227 op-amp and fed into the analog. The combined analog signal and offset are then fed into the second AD8021 video op-amp set up to obtain

~2x gain. The output of this op-amp is clamped at -0.7 VDC below the negative rail to protect the ADC. The output is then fed through a 50 ohm impedance matching resistor and clamped again at 4 VDC and -0.7 VDC, again to protect the ADC. A RC filter limits the input frequency range to 10 MHz at the ADC input for more stable operation.

268

Figure 141. Custom designed analog card with video gain pipeline, programmable offset generator, and two channel 14 bit 20MSPS ADC.

Each of the analog singles is fed into one of the + inputs of the AD9248 analog-digital converter. The – inputs are tied to the reference voltage to give an input range of 0 to 2xVref.

The 1.0 V reference is internally generated on the ADC. The digital output is of the two ADC

channels are multiplexed onto a 14 wire parallel bus and fed to the embedded CPU via the

backplane. There are two 14 bit busses to allow taking one reading from each of the ADCs

simultaneously.

10.5.3.4. 6 Channel programmable bias card

Various ROICs can require up to 8 different DC biases. To facilitate versatility, the portable camera system has 8 biases, each of which can be individually programmed to any value between 0 and 10 volts. 6 of the biases are located on the bias card, the other 2 are located on the clock card due to space constraints. A FET based relay is provided to disconnect each of the bias outputs and protect the ROIC during startup and shutdown of the camera system. A 10 ohm precision resistor is provided on each output to limit the current and to act

269 as a shunt so that the current consumption can be measured using a pair of clips placed across the resistor.

Figure 142. 6 channel programmable bias card

The biases originated at 0 to 3.3 VDC output from a 16 bit serial DAC mapped onto the

I2C bus. An RC filter stabilizes the DAC output. An OPA227 precision op-amp scales this value to the 0 to 10 VDC range. A high current BUF634 unity gain buffer amplifier is used to amplify the current sourcing capability to 250 mA for each of the outputs.

10.5.3.5. Clock driver and 2 channel programmable bias card

The clock signals originate from the embedded CPU, but these 3.3 VDC high speed clocks are not suitable for directly driving most ROICs. The main purpose of the clock driver card is to provide an adjustable clock skew (to reduce high frequency noise in the ROIC) and to allow for programmable control of the clock voltage. An additional DC bias circuit similar to those found on the bias card is used to power a SN74LVCC3245 bus driver. This chip translates the clocks from 3.3 VDC to a programmable value in the range of 3 to 5.5 VDC. An

270 RC circuit on the output of this chip is used to load the clock line going to the ROIC acting as a low pass filter preventing high frequency clock edge ringing from being generated in the ROIC.

In addition to the adjustable clock driver, this card also contains the remaining 2 bias channels. Both of these bias channels as well as the clock rail voltage are fed from a single 4- channel serial DAC mapped onto the I2C buss.

Figure 143. Clock driver and 2 channel programmable bias card

271 10.6. Appendix 6: Development of a Semiconductor Wafer

Cleaning System

As part of this work a completely custom semiconductor wafer-cleaning system has been designed, built, installed, and qualified. Cleaning is critical to realization of high-quality low leakage photodetectors. The target application of this system is the ultra high-level cleaning of the surface of semiconductor wafers during the various semiconductor processing steps necessary to realize finished devices. The system is designed to accomplish this using multiple high-pressure solvent sprays impingent upon a rotating semiconductor wafer. The system is classified as an automatic high-pressure solvent cleaning system. This is not a novel cleaning concept; however this system is a one of a kind design created specifically to fit the unique space and facilities available for use at the center for quantum devices. The total cost to fabricate, install and qualify the system was just shy of $60,000.

272

Figure 144. Wafer holder (center) and multi-nozzle spray arm (top) showing the

Semiconductor wafer cleaning system installed in a chemical hood.

During operation, the wafer is held on a large rotating vacuum chuck. The chuck is sufficiently large to fully support the back of the wafer and prevent deflection of the wafer while cleaning. The rotational speed of the wafer is ramped up to a programmable speed at a programmable ramp rate. A robotic moving arm containing 6 high-pressure spray nozzles, one low-pressure rinse nozzle, and a N2 blow dry nozzle will then be moved over the wafer. The moving arm holds the nozzles normal to the wafer surface, and can control the nozzle position anywhere over the wafer. In operation, the first spray nozzle (usually stripper) will be placed above the center of the wafer. The valve will open and a high-pressure spray will be emitted from the nozzle. This spray will be moved from the center to the edge of the rotating wafer, sweeping the contaminants off of the wafer. Immediately thereafter, the flow will stop and the

273 next nozzle will be placed above the center of the wafer. The process will continue through the

remaining nozzles progressively cleaning the wafer to higher and higher levels. This will then

be immediately followed by a high-purity N2 blow dry. This blow-dry moves from the center

to the edge pushing the final solvent or DI wafer off the wafer and preventing the formation of

water spots.

10.6.1. Systems Overview

The system consist of separate units, the wafer fixture and spray arm, the solvent handling and pressurization system, the DI water system, the waste collection system, and the electrical control system. The majority of the system is hidden from the user view with only the wafer fixture and spray arm shown above directly visible. The remained of the system is housed in the support chase adjacent to the system as shown in the diagram below.

Figure 145. General layout of the semiconductor wafer cleaning system, showing the

location of support equipment in the chase, away from the main use interface which occupies

minimum space in the hood.

274 General Capabilities

• Clean Wafer size: up to 2” Fully Supported, Up to 5” Possible with additional chucks

• Automation: Manual load and unload, fully automatic cleaning process

• User Interface: Labview interface for intuitive graphical system control.

• Recipes: Can control solvent, delivery time, head rastering, and temperature.

Heated Stripper Capabilities

The system employs two custom fabricated high purity heat exchangers for heating of the stripper used. Heating is by means of a 3 meter long counter-current heat exchanger. The entire line is heated to within 1.5” of the valve, ensuring that the exit temperature is well within specifications after the first few seconds of dispensing. A photo of the heat exchangers and schematic diagram are shown below:

Figure 146. Left) picture of the insulated counter-current heat exchangers where they meet the valves. Right> schematic diagram of the three heat exchanger located behind the hood.

275 • Strippers (2): The system will be capable of dispensing (1)AZ 400T and (2)AZ Kwik-Strip.

o • Temperature Control: Controllable from ambient to 90 P C.P

• Pressure: Individually controlled dispense pressure of 5 to 125 PSIG.

• Flow rate: Approximately 500 mL/min, (exact value depends upon pressure)

• Storage: The system will have a storage capacity of 16L for each of these stripers.

• Purity: all wetted components will be 316 SS, Teflon, or Kel-F (a fluropolymer)

• Limitations: Both strippers must have the same temperature.

Solvent Capabilities

Solvents are stored in 316 SS electro-polished casks. For safety they are kept in a

flammables storage cabinet that is vented outside the building. The pressure in each of these

casks is individually controllable via a manually set regulator. Teflon sealed quick connects

are provided for easy filling of the tanks. A picture of the solvent storage facility is shown below along with a schematic diagram of the plumbing.

Figure 147. Left) Photo of the solvent storage tanks located within the flammable storage

cabinet. Right) Schematic diagram of the solvent pressurization plumbing (green), flexible lines

(blue) and tanks (red).

276

• Solvents (3): The system is capable of dispensing (1)Acetone, (2)Methanol, and (3)Propanol.

• Temperature Control: The system dispenses these solvents at ambient temperature.

• Pressure: Individually controlled dispense pressure of 5 to 125 PSIG.

• Flow rate: Approximately 500 mL/min, (exact value depends upon pressure)

• Storage: The system has a storage capacity of 16L for each of these solvents.

• Purity: all wetted components will be 316 SS, Teflon, or Kel-F (a fluropolymer).

Heated DI water Capabilities

• DI water (1): (1)The System uses 18 Mega-ohm DI from the house system.

o • Temperature Control: Thermostatically set at 80 P C,P decreases with flows above 1 GPM.

• Pressure: Controllable, from 30 PSIG to 100 PSIG.

• Flow rate: Approximately 500 mL/min, (exact value depends upon pressure)

• Purity: all wetted components will be 316 SS, Teflon, PolyPro, or other fluropolymers

• Limitations: User must manually open purge valve to get high-purity hot DI.

• Bonus: An additional hot DI water spigot was added for general use and system purge.

Cold DI water Capabilities

• DI water (1): (1) The System use the 18 Mega-Ohm DI directly from the house system.

• Temperature: Fixed at the ambient temperature of the DI water supply.

• Pressure: Fixed at system pressure of 30 PSIG.

• Flow rate: 2 L /min

• Purity: all wetted components will be 316 SS, Teflon, or other fluropolymers

277

Nitrogen Blow Dry

• Nitrogen (1): (1) The System uses high purity house-nitrogen.

• Temperature: Manually selectable between ambient, or the same as the two strippers.

• Pressure: Fixed at system pressure of 50 PSIG.

• Flow rate: manually adjustable via a needle valve, up to the limit of the nozzle orifice.

• Purity: all wetted components will be 316 SS, Teflon, or other fluropolymers

Waste Collection

The waste collection system is critical to the operation of the semiconductor wafer cleaning system. Due to the nature of the solvents and strippers used, this material cannot be dumped directly down the drain, and must instead be collected for disposal. However the DI water used to rinse is not required to be collected, and is far too large in volume for that to be an option. Instead a novel diversion and lift pump strategy was implemented to collect solvent waste by gravity drain, and pump waste water up to a level where it can drain by gravity into a standard laboratory cup-sink. This system is shown in the picture and diagram below, with the specification listed in the bullet point below that.

278

Figure 148. Left) Photo of bottom of wafer fixture showing water lift pump (top shelf), and waste collection container (bottom shelf), Right) schematic diagram of the waste collection plumbing showing the lift pump draining waste water into a nearby acids sink.

• Chemical Waste: Stripper and solvent waste are combined together and collected in a standard 19 L waste container. The container is vented into the hood, and is fitted with a capacitive level sensor to inhibit the execution of a new cleaning cycle if container is full thereby preventing the system from overflowing.

• Water waste: Rinse water will be separated from the solvent storage, and pumped up and allowed to drain into the general libratory acids drain.

• Automation: The system will automatically divert the waste to the proper destination depending upon the nozzles selected.

• Vacuum waste: the vacuum pump line is fitted with an automatic blow down collection trap. This will collect any solvent or water that is inadvertently sucked into the vacuum line. At the completion of every run the vacuum separator tank is pressurized and will drain via check valve into the general solvent waste container.

• Ventilation waste: The exhaust booster is fitted with a filter element and coalesced waste will collect in the 5-gallon drum. This will need to be emptied periodically.

279 • Limitations: The ventilation waste is not automatically combined with the solvent waste, and must be manually processed on a weekly basis.

Safety Systems

• Storage: The three flammable solvents and the two strippers are housed in a ventilated flammable safety storage cabinet approved for the storage of flammable liquids. It contains the storage tanks and prevents their tipping over. It provided sufficient secondary containment to contain the rupture of one of the storage tanks. The waste solvents are collected in a 19L waste container. This container is also housed in a flammable safety storage cabinet that provides secondary containment for a partial leak of the vessel. The door to the storage cabinet is pneumatically interlocked to prevent opening the cabinet while the solvent tanks are pressurized.

• Pressure: All piping and system components are rated for a working pressure of 125 PSI or higher, and fitted with appropriate safety relief devices to prevent rupture. In the event of a serious fire the storage tanks will be capable of venting their contents without exceeding 120% of their rated working pressure.

• E-stop: The system is provided with two E-stop buttons, one by the cleaning head, and one by the remote storage container. If actuated, they will immediately remove power from all pressurized and/or moving components. This will cause the system to be freely movable, and will close all supply valves, and vent the residual pressures in the storage tanks to less than 1.5 PSIG thereby stopping any flow.

• Ventilation: The storage cabinet is connected to cleanroom general exhaust. In the event of an over pressurization, the flammable vapor will be contained by the cabinet, and exhausted via this vent. In the event of an emergency shutdown, all flammable vapors in the tanks will be vented to the cleanroom exhaust via double check valves; the outlets of these vent pipes are individually fitted with porous screens to prevent flashback along the vent lines into the tanks. The delivery head itself is housed within a chemical hood. Extra ventilation is provided at the cleaning bowl to prevent the accumulation of flammable vapors.

• Labeling: All storage tanks are appropriately labeled according to the characteristics of the material inside. The flammable safety cabinet is additionally labeled to indicate the presence of flammable liquids. _PR

10.6.2. Control and Software Design

The system is fully automatic, employing a sophisticated Labview interface to allow for control of the valves, pumps, heaters, spinner, and spray arm. Everything is fully automated

280 except for the setting of the solvent tank pressures. In normal operation all the user needs to do

is load a wafer and press the start button. Processes are defined using plain text recipes that dictate the operation of the machine. A custom recipe interpreter was written to parse these files and operate the machine in a reproducible manner. The user has full control of all setting

via the recipe, though in practice only about 20 commands will ever be necessary. To aid in

writing recipes a separate recipe editor with real-time syntax highlighting is provided.

The system interfaces the electro-mechanical components via a National Instruments

data acquisition card (model PCI-6259). This card provides 32 analog inputs, 4 analog outputs,

and 48 digital inputs/outputs. It has a number of built in timers and counter that are used to

control the operation of the stepper motor and maintain accurate position control. Analog inputs

are used to read the temperature through the system, and the read-back the spinner speed.

Analog outputs are used to control the spinner speed. Digital outputs control the valves

through relays and solenoids that supply compressed air to the valve actuators. Electrical

components are controlled directly via relays. Software interlocking is provided by a master enable relay in addition to mechanical E-Stop button. Digital Inputs are used to count the stepper motor position, read the waste level sensor, and detect an E-stopped condition. All other control is supplied through soft buttons provided by the software. All of the electro- mechanical hardware necessary to operate the semiconductor wafer cleaning system is housed in an electronics control rack attached to the side of the flammable solvent storage unit. This contains all the relays, solenoids, power supplies, stepper motor drivers, and input modules to interface the system with the DAC card in the computer. A picture of the electronics control package is shown below.

281

Figure 149. Semiconductor wafer cleaning system electronics and pneumatics control

package.

The Labview control interface presents the user with a schematic diagram of the system.

Custom visual elements have been specifically constructed to maintain the visual metaphor of a

schematic diagram. All controls are quad-stated so that selected state and interlocked state

information can be given for each. At the bottom of the diagram is a text box that contains the

previous, current, and next step in any running recipe. An image of the main software

application is shown below. From the tabs at the top of the screen, the user can select from

among the other four tabs. The recipe tabs is used for loading and error checking recipe files.

The consumption tab provides an interface for calculating the solvent consumption. Since the tanks are opaque, it is necessary to track the consumption by indirectly recording the elapsed delivery time, and applying consumption factors to determine the amount remaining. The last

282 tab contains debugging information. It contains a full table showing all of the inputs and

outputs, both as set, and as actually allowed by the interlocks. Status information for the recipe

parser is also given showing the current step, and how it has been interpreted by the compiler.

There are also panes to show the status of the thermal bath and the wafer fixture rotation.

Figure 150. Main control panel for interfacing with the semiconductor wafer cleaning system.

283 11. References

1 TP PT F.S. Johnson. “Space Materials Handbook.” Lookheed Missiles and Space Co., Jan (1962).

2 TP PT C Cordoba ,J A Munoz , V Cachorro, I Aguirre de Carcer, F Cusso, and F Jaque. “The detection of solar ultraviolet-C radiation using KCl:Eu 2+ thermoluminescence dosimeters.” Journal of Physics D Applied Physics, 30, pp. 3024–3027, (1997).

3 TP PT A. Lacis, J. Hansen. “A parameterization for the Absorption of Solar radiation in the Earth’s atmosphere.” Journal of Atmospheric Sciences, 31, pp. 118-133 (1974).

4 TP PT P. Kung, A Yasan, R. McClintock, S.R. Darvish, K. Mi, M. Razeghi. “Future of AlxGa1-xN materials and device technology for ultraviolet photodetectors.” Proceedings of the SPIE, 4650, pp.199-206, Jan. (2002).

5 TP PT M. Razeghi. “Short-wavelength solar-blind detectors- status, prospects, and markets.” Proceedings of the IEEE, 90, pp. 1006-1014, June (2002).

6 TP PT M. Ulmer, M. Razeghi, E. Bigan. “Ultra-Violet Detectors for Astrophysics, Present and Future.” in Optoelectronic Integrated Circuit Materials, Physics, and Devices, ed. Manijeh Razeghi, Yoon-Soo Park and Gerald L. Witt, Proceedings of the SPIE, 2397 pp. 210-216, Bellingham, Wash.: SPIE-The International Society for Optical Engineering, (1995).

7 TP PT DARPA, “Vehicle Self-Protection Technology (VSPT)”, BAA 98-14, January (1998).

8 TP PT H. Morkoc. “GaN-Based Modulation-Doped FETs and UV Detectors.” Naval Research Reviews, 51 (1), pp. 26-42 (1999).

9 TP PT DARPA, “Semiconductor Ultraviolet Optical Sources (SUVOS)”, BAA 02-03, January (2002).

10 TP PT G.A. Shaw and M. Nischan. “Short-range NLOS ultraviolet communication test bed and measurements.” Proceedings of the SPIE, 4396, pp. 31-40 (2001).

11 TP PT A. Boin, P. Lagadec, E. Michel-Kerjan, W. Overdijk. “Critical Infrastructures under Threat: Learning from the Anthrax Scare.” Journal of Contingencies and Crisis Management, 11 (3), pp. 99–104, (2001).

12 TP PT L. Gilbert and S. Maurer. “Implementation of a Bioaerosol Warning System.” Proceedings of the SPIE, 6398, pp. 639807.1-639807.8 (2006).

284

13 TP PT J. C. Carrano, A. J. Maltenfort. “Semiconductor ultraviolet optical sources for biological agent detection.” Proceedings of the SPIE, 4743, pp. 232-238 (2002).

14 TP PT V. Sivaprakasam, A. L. Huston, C. Scotto and J. D. Eversole. “Multiple UV wavelength excitation and fluorescence of bioaerosols.” Optics Express, 12 (19), pp. 4457-4466 (2004).

15 TP PT C. Richards, C. L. Benner, K.L. Butler-Purry, and B. Russell. “Electrical Behavior of Contaminated Distribution Insulators Exposed to Natural Wetting.” IEEE Transactions on Power Delivery, 18 (2), pp. 551-558 (2003).

16 TP PT Jungang Liu, Wei He, Liang Zhou, and Fan Yang. “Application of Ultraviolet Detector in Discharge Detection.” International Journal of Emerging Electric Power Systems, 7 (3), pp. 2.1-2.8 (2006).

17 TP PT M. Ulmer, M. Razeghi, E. Bigan. “Ultraviolet Detectors for Atrophysics, Present and Future.” Proceedings of the SPIE, 2397, pp. 210-218 (1995).

18 TP PT P. Kung, A. Saxler, X. Zhang, D. Walker, M. Razeghi, M. Ulmer. “GaN, GaAlN, and AlN Detectors for Astrophysics, an Update.” Proceedings of the SPIE, 2685, pp. 126-131 (1996).

19 TP PT J.-Y. Duboz. “GaN as Seen by the Industry.” Physica Status Solidi (a), 176(1), pp. 5-14 (1999).

20 TP PT R. Beal. “New tools for safeguarding utility boiler furnaces.” ISA Proc. 4th ,Vat. Power Instrumentation Conference (Instrument Society of America), pp 131-40 (1961).

21 TP PT A. Jones. “Flame failure detection and modern boilers.” Journal of Physics E: Science Instrumentation, 21, pp. 921-928 (1988).

22 TP PT K. Rygh. “Fire safety research in microgravity: How to detect smoke and flames you cannot see.” Fire Technology, 31(2), pp. 175-185 (1995).

23 TP PT M. Razeghi and A. Rogalski. “Semiconductor ultraviolet detectors.” Journal of Applied Physics, 79 (10), pp. 7433-7473 (1996).

24 TP PT S. Nakamura, G. Fasol, S. Pearton. “The Blue Laser Diode.” Springer-Verlag, Germany (1997).

25 TP PT R. McClintock, J. L. Pau, K. Minder, C. Bayram, P. Kung, M. Razeghi. “Hole-initiated multiplication in back-illuminated GaN avalanche photodiodes.” Applied Physics Letters, 90(14), pp. 141112-1 - 141112-3 (2007).

285

26 TP PT J. B. Limb, D. Yoo, J. H. Ryou, W. Lee, S. C. Shen, and R. D. Dupuis, M. L. Reed, C. J. Collins, and M. Wraback, D. Hanser, E. Preble, N. M. Williams, K. Evans. “GaN ultraviolet avalanche photodiodes with optical gain greater than 1000 grown on GaN substrates by metal-organic chemical vapor deposition.” Applide Physics Letters, 89 (011112), pp. 1-3 (2006).

27 TP PT D. Walker, X. Zhang, P. Kung, A. Saxler, S. Javadpour, J. Xu, and M. Razeghi. “AlGaN ultraviolet photoconductors grown on sapphire.” Applide Physics Letters, 68 (15), pp. 2100- 2101 (1996).

28 TP PT D. Walker, X. Zhang, A. Saxler, P. Kung, J. Xu, and M. Razeghi. “AlxB GaB 1-xB NB (0≤x≤1) ultraviolet photodetectors grown on sapphire by metal-organic chemical-vapor deposition.” Applied Physics Letters, 70 (8), pp. 949-951 (1997).

29 TP PT E. Monroy, F. Calle, J. A. Garrido, P. Youinou, E. Munoz, F. Omnes, B. Beaumont, and P. Gibart. “Si-doped Al xGa1 −xN photoconductive detectors.” Semiconductor Science and Technology, 14, pp. 685-689 (1999).

30 TP PT C. H. Qiu and J. I. Pankove. “Deep levels and persistent photoconductivity in GaN thin films.”, Applied Physics Letters, 70 (15), pp. 1983-1985 (1997).

31 TP PT J. Z. Li, J. Y. Lin, H. X. Jiang, A. Salvador, A. Botchkarev, and H. Morkoc. “Nature of Mg impurities in GaN.” Applied Physics Letters, 69 (10), pp. 1474-1476 (1996).

32 TP PT A. Y. Polyakov, N. B. Smirnov, A. V. Govorkov, M. G. Milvidskii, J. M. Redwing, M. Shin, M. Skowronski, D. W. Greve, and R. G. Wilson. :Properties of Si donors and Persistent Photoconductivity in AlGaN.” Solid-State Electronics, 42 (4), pp. 627-635 (1998).

33 TP PT A. Y. Polyakov, N. B. Smirnov, A. S. Usikov, A. V. Govorkov, and B. V. Pushniy. “Studies of the Origin of he Yellow Luminescence Band, the Nature of Nonradiative Recombination and the Origin of Persistent Photoconductivity in N-GaN Films.“ Solid-State Electronics, 42 (11), pp. 1959-1967 (1998).

34 TP PT J. A. Garrido, E. Monroy, I Izpura, and E Munoz. “Photoconductive Gain Modeling of GaN photodetectors.” Seciconductor Science and Technology, 13, pp. 563-568 (1998).

35 TP PT D. Walker, E. Monroy, P. Kung, J. Wu, M. Hamilton, F. J. Sanchez, J. Diaz, and M. Razeghi. “High-speed, low-noise metal-semiconductor-metal ultraviolet photodetectors based on GaN.” Applied Physics Letters, 74 (5), pp. 762-764 (1999).

36 TP PT P. Kung, D. Walker, P. Sandvik, M. Hamilton, J. Diaz, I.-H. Lee, and M. Razeghi. “Schottky MSM photodetectors on GaN films grown on sapphire by lateral Epitaxial overgrowth.” Proceedings of the SPIE, 3629, pp. 223-229 (1999).

286

37 TP PT J.C. Carrano, T. Li, P.A. Grudowski, C.J. Eiting, R.D. Dupuis, and J.C. Campbell. “High quantum efficiency metal-semiconductor-metal ultraviolet photodetectors fabricated on single-crystal GaN epitaxial layers.” Electronics Letters, 33 (23), 1980-1981 (1997).

38 TP PT T. Li, D.J.H. Lambert, A.L. Beck, C.J. Collins, B. Yang, M.M. Wong, U. Chowdhury, R.D. Dupuis, and J.C. Campbell. “Solar-blind AlxGa1-xN based metal-semiconductor-metal ultraviolet photodetectors.” Electronics Letters, 36 (18), pp. 1581-1583 (2000).

39 TP PT B. Yang, D.J.H. Lambert, T. Li, C.J. Collins, M.M. Wong, U. Chowdhury, R.D. Dupuis, and J.C. Campbell. “High-performance back-illuminated solar-blind AlGaN metal- semiconductor-metal photodetectors.” Electronics Letters, 36 (22), pp. 1866-1867 (2000).

40 TP PT A. Osinsky, S. Gangopadhyay, B. W. Lim, M. Z. Anwar, M. A. Khan, D. V. Kuksenkov and H. Temkin. “Schottky barrier photodetectors based on AlGaN.” Applied Physics Letters, 72 (6), pp. 742-744 (1998).

41 TP PT M. Razeghi. DARPA Review, Feb. (2000).

42 TP PT C. I. Wu and A. Khan. “Electronic states and effective negative electron affinity at cesiated p-GaN surface.” Journal of Applied Physics, 86, pp. 3209–3212 (1999).

43 TP PT F. Machuca, Y. Sun, Z. Liu, K. Ioakeimidi, P. Pianetta, R. Pease. “Prospect for high brightness III–nitride electron emitter.” Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures, 18 (6), pp. 3042-3046 (2000).

44 TP PT C. Wu, A. Kahn, E. Hellman, D. Buchanan. “Electron affinity at aluminum nitride surfaces.” Applied Physics Letters, 73 (10), pp. 1346-1349 (1998).

45 TP PT C. Wu, A. Kahn. “Negative electron affinity and electron emission at cesiated GaN and AlN surfaces.” Applied Surface Science, 162–163, pp. 250–255 (2000).

46 TP PT C. Wu, A. Kahn. “Negative electron affinity at the Cs/AlN(0001) surface.” Applied Physics Letters, 74 (10), pp. 1433-1435 (1999).

47 TP PT M. Ulmer, B. Wessels, O. Siegmund. “Progress in the Fabrication of GaN Photocathodes.” ASP Conference Series, 000, pp. 1 – 4 (2002).

48 TP PT F. Shahedipour, M. Ulmer, B. Wessels, C. Joseph, T. Nihashi. “Efficient GaN Photocathodes for Low-Level Ultraviolet Signal Detection.” IEEE Jouranl of Quantum Electronics, 38 (4), pp. 333-336 (2002).

287

49 TP PT D. J. Leopold, J. H. Buckley, P. Rebillot. “High quantum efficiency ultraviolet/blue AlGaN/InGaN photocathodes grown by molecular-beam epitaxy.” Journal of Applied Physics, 98 (043525), pp. 1-5 (2005).

50 TP PT M. Ulmer, B. Wessels, B. Han, J. Gregie, A. Tremsin, O. Seigmund. “Advances in Wide- Band-Gap Semiconductor Based Photocathode Devices for Low Light Level Applications.” Proceedings of the SPIE, 5164-18, pp. 1-11 (2003).

51 TP PT S. Uchiyama, Y. Takagi, M. Niigaki, H. Kan, H. Kondoh. “GaN-based photocathodes with extremely high quantum efficiency.” Applied Physics Letters, 86, pp. 103511-1 – 103511-3 (2005).

52 TP PT R. McClintock, P. Sandvik, K. Mi, F. Shahedipour, A. Yasan, C. Jelen, P. Kung, and M.

Razeghi. “AlxB GaB 1-xB NB materials and devices technology for solar blind ultraviolet photodetector applications.” Proceedings of the SPIE, 4288, pp. 219-229 (2001).

53 TP PT P. Kung, X. Zhang, D. Walker, A. Saxler, and M. Razeghi. “GaN p-i-n photodiodes with high visible-to-ultraviolet rejection ratio.” Proceedings of the SPIE, 3287, pp. 214-220 (1998).

54 TP PT E. Monroy, M. Hamilton, D. Walker, P. Kung, F. J. Sanchez, and M. Razeghi. “High-quality visible-blind AlGaN p-i-n photodiodes.” Appied Physics Letters, 74 (8), pp. 1171-1173 (1999).

55 TP PT D. Walker, P. Kung, P. Sandvik, J. Wu, M. Hamilton, I.-H. Lee, J. Diaz, and M. Razeghi.

“AlxB GaB 1-xB NB p-i-n photodiodes on sapphire substrates.” Proceedings of the SPIE, 3629, pp. 193-198 (1999).

56 TP PT J.C. Carrano, T. Li, P.A. Grudowski, C.J. Eiting, D. Lambert, J.D. Schaub, R.D. Dupuis, and J.C. Campbell. “Low dark current pin ultraviolet photodetectors fabricated on GaN grown by metal organic chemical vapour deposition.” Electronics Letters, 34 (7), 692-694 (1998).

57 TP PT G. Parish, S. Keller, P. Kozodoy, J. P. Ibbetson, H. Marchand, P. T. Fini, S. B. Fleischer, S. P. DenBaars, U. K. Mishra, and E. J. Tarsa. “High-performance (Al,Ga)N-based solar-blind ultraviolet p–i–n detectors on laterally epitaxially overgrown GaN.” Appied Physics Letters, 75 (2), pp. 247-249 (1999).

58 TP PT C. Pernot, A. Hirano, M. Iwaya, T. Detchprohm, H. Amano, and I. Akasaki. “Solar-Blind UV Photodetectors Based on GaN/AlGaN p-i-n Photodiodes.” Japanese Journal of Applied Physics, 39 (5A), pp. L387-L389 (2000).

59 TP PT T. Li, S. Wang, A. L. Beck, C. J. Collins, B. Yang, R. D. Dupuis, J. C. Carrano, M. J.

Schurman, I. T. Ferguson, and J. C. Campbell. “High quantum efficiency AlxB GaB 1-xB N/GaN-B

288

based ultraviolet p-i-n photodetectors with a recessed window structure.” SPIE: Optoelectronics 2000 Technological Digest, 3948, pp. 304-310 (2000).

60 TP PT W. Yang, T. Nohova, S. Krishnankutty, R. Torreano, S. McPherson, and H. Marsh. “Back- illuminated GaN/AlGaN heterojunction photodiodes with high quantum efficiency and low noise.” Applied Physics Letters, 73 (8), pp. 1086-1088 (1998).

61 TP PT E. J. Tarsa, P. Kozodoy, J. Ibbetson, B. P. Keller, G. Parish and U. Mishra. “Solar-blind AlGaN-based inverted heterostructure photodiodes.” Applied Physics Letters, 77 (3), pp. 316-318 (2000).

62 TP PT T. Li, D.J.H. Lambert, M.M. Wong, C.J. Collins, B. Yang, A.L. Beck, U. Chowdhury, R.D.

Dupuis, and J.C. Campbell. “Low-noise back-illuminated AlxB GaB 1-xB N-basedB p-i-n solar-blind ultraviolet photodetectors.” Journal of Quantum Electronics, 37 (4), pp. 538-545 (2001).

63 TP PT J.D. Brown, J.Z. Li, P. Srinivasan, J. Matthews, and J.F. Schetzina. “Solar-blind AlGaN heterostructure photodiodes.” MRS Internet Journal of Nitride Semiconductor Research, 5 (9), (2000).

64 TP PT T J.T L. Pau, E. Monroy, M. A. Sánchez-García, E. Calleja, and E. Munoz. “AlGaN ultraviolet photodetectors grown by molecular beam epitaxy on Si(111) substrates.”

Materials Science and Engineering B, 93 (1), pp. 159-162 (2002). T

65 TP PT V. A. Dmitriev, K. G. Irvine, C. H. Carter, N. I. Kuznetsov, E. V. Kalinina. “Electric breakdown in GaN p-n junctions.” Applied Physics Letters, 68(2), pp. 229-231 (1996).

66 TP PT A. Osinsky, M.S. Shur, R. Gaska, Q. Chen. “Avalanche breakdown and breakdown luminescence in p-π-n GaN diodes.” IEEE Electronics Letters, 34 (7), pp. 691-693 (1998).

67 TP PT K. McIntosh, R. Molnar, L. Mahoney, M. Geis, K. Molvar, I. Melngailis, R. Aggarwal, W. Goodhue, S. Choi, and D. Spears. “GaN avalanche photodiodes grown by hydride vapor- phase epitaxy.” Applied Physics Letters, 75, pp. 3485-3487 (1999).

68 TP PT J. Carrano, D. Lambert, C. Eiting, C. Collins, T. Li, S. Wang, A. Beck, R. Dupuis, and J. Campbell. “GaN avalanche photodiodes.” Applied Physics Letters, 76, pp. 924-297 (2000).

69 TP PT B. Yang, T. Li, K. Heng, C. Collins, S. Wang, J. Carrano, R. Dupuis, J. Campbell, M. Schurman, and I. Ferguson. “Low dark current GaN avalanche photodiodes.” IEEE Journal of Quantum Electroics, 36, pp. 1389-1391 (2000).

70 TP PT S. Verghese, K. McIntosh, R. Molnar, L. Mahoney, R. Aggarwal, M. Geis, K. Molvar, E. Duerr, and I. Melngailis. “GaN avalanche photodiodes operating in linear-gain mode and Geiger mode.” IEEE Transactions on Electronic Devices, 48 (3), pp. 502-511 (2001).

289

71 TP PT J.B. Limb, D. Yoo, J.H. Ryou, W. Lee, S.C.Shen, R.D. Dupuis, M.L. Reed, C.J.Collins, M. Wraback, D. Hanser, E. Preble, N.M. Williams, and K. Evans. “GaN ultraviolet avalanche photodiodes with optical gain greater than 1000 grown on GaN substrates by metal-organic chemical vapor deposition.” Applied Physics Letters, 89, pp. 11112-1 – 11112-3 (2006).

72 TP PT R. McClintock, J. L. Pau, K. Minder, C. Bayram, P. Kung, M. Razeghi. “Hole-initiated multiplication in back-illuminated GaN avalanche photodiodes.” Applied Physics Letters, 90, pp. 141112-1 – 141112-3, (2007).

73 TP PT P. Kung, R. McClintock, J. L. Pau, K. Minder, C. Bayram, M. Razeghi. “III-nitride avalanche photodiodes.” Proceedings of the SPIE, 6479, pp. 64791J-1 – 64791J-12, (2007).

74 TP PT R. McClintock, A. Yasan, K. Minder, P. Kung, and M. Razeghi. “Avalanche multiplication in AlGaN based solar-blind photodetectors.” Applied Physics Letters , 87, pp. 241123-1 – 241123-3, (2005).

75 TP PT T. Tut, S. Butun, B. Butun, M. Gokkavas, H. Yu, and E. Ozbay. “Solar-blind AlxGa1–xN- based avalanche photodiodes.” Applied Physics Letters, 87, pp 223502-1 – 223502-3, (2005).

76 TP PT M. S. Shur, and M. Asif Khan. “GaN and Ultraviolet Detectors.” chapter 10 of Gallium Nitride (GaN) II, edited by J. I. Pankov, and T. D. Moustakas, Academic Press, San Diego, pp. 407-439, (1999).

77 nd TP PT S. M. Sze. Physics of Semiconductor Devices, 2P P edition, John Wiley and Sons, Inc., pp. 743-789, (1981).

78 TP PT S. L. Chuang. Physics of Optoelectronic Devices. John Wiley and Sons, Inc., pp. 583-615, (1995).

79 TP PT V. V Kuryatkov, H. Temkin, J. C. Campbell, and R. D. Dupuis. “Low-Noise photodetectors based on heterojunctions of AlGaN-GaN.” Applied Physics Letters, 78 (21), pp. 3340-3342, (2001).

80 TP PT C. J Collins, T. Li, D. J. H. Lambert, M. M. Wong, R. D. Dupuis, and J. C. Campbell. “Selective Regrowth of AlGaN[0.30]Ga[0.70]N p-i-n Photodiodes.” Applied Physics Letters, 77 (18), pp. 2810-2812, (2000).

81 TP PT G. Stillman and C. Wolfe. Chap. 5 “Avalanche photodiodes” in Semiconductors and Semimetals, ed. R Willardson (Academic, New York, 1977), Vol. 12, p.333.

290

82 TP PT Voxtel Inc. “Photon Counting APDs, a Primer,” http://www.voxtel-HTU

inc.com/white_papers/TechNotes_Photon-countingAPDs.pdfUTH (2005).

83 TP PT C.H. Chen, H. Liu, D. Steigerwald, W. Imler, C.P. Kuo, M.G. Craford, M. Ludowise, S.

Lester, and J. Amano. “A study of parasitic reactions between NH3B B and TMGa or TMAl.”, . Electronic Materials, 25, pp. 1004 - 1012, (1996).

84 TP PT L. Schowalter, S. Schujman, W. Liu, M. Goorsky, M. Wood, J. Grandusky,F. Shahedipour- Sandvik. “Development of native, single crystal AlN substrates for device applications.” Physica Status Solidi (a), 203 (7), pp.1667-1671, (2006).

85 TP PT E. Oh, S. K. Lee, S. S. Park, K. Y. Lee, I. J. Song, J. Y. Han. “Optical properties of GaN grown by hydride vapor-phase epitaxy.” Applied Physics Letters, 78 (3), pp. 273-275, (2000).

86 TP PT A. Yasan, R. McClintock, K. Mayes, S. R. Darvish, H. Zhang, P. Kung, M. Razeghi, S. K. Lee, J. Y. Han. “Comparison of ultraviolet light-emitting diodes with peak emission at 340 nm grown on GaN substrate and sapphire.” Applied Physics Letters, 81 (12), pp. 2151-2154, (2002).

87 TP PT F.A. Ponce, J.S. Major, W.E. Plano, D.F. Welch. “Crystalline structure of AlGaN epitaxy on sapphire using AlN buffer layers.” Applied Physics Letters, 65 (16), pp. 2302-2304, (1994).

88 TP PT C. J. Sun, P. Kung, A. Saxler, H. Ohsato, K. Haritos, and M. Razeghi. “A crystallographic model of (00·1) aluminum nitride epitaxial thin film growth on (00·1) sapphire substrate.” Journal of Applied Physics, 75, pp. 3964 – 3972, (1994).

89 TP PT K. Hiramatsu, S. Itoh, H. Amano, I. Akasaki, N. Kuwano, T. Shiraishi, and K. Oki. “Growth mechanism of GaN grown on sapphire with AlN buffer layer by MOVPE.” Journal of Crystal Growth, 115, pp. 628 – 636, (1991).

90 TP PT H. Amano, N. Sawaki, I. Akasaki, Y. Yoyoda. “Metalorganic vapor phase epitaxial growth of a high quality GaN film using an AlN buffer layer.” Applied Physics Letters, 48 (5), pp. 352-354, (1986).

91 TP PT This work was performed in collaboration with V.P. Dravid and Z.G. Zheng of the Northwestern University Electron Probe Instrumentation Center (EPIC).

92 TP PT M. A. Khan, J. N. Kuzina, R. A. Skogman, D.T. Olson, M. MacMillan, W.J. Choyke. “Low pressure metalorganis chemical vapor deposition of AlN over sapphire substrates.” Applied Physics Letters, 61, pp. 2359 - 2361, (1992).

291

93 TP PT H. Wang, J. Zhang, C. Chen, Q. Fareed, J. Yang, A. Khan. “AlN/AlGaN superlattices as dislocation filter for low-threading-dislocation thick AlGaN layers on sapphire.” Applied Physics Letters, 81, pp. 604 – 606, (2002).

94 TP PT J. Neugebauer, C. G. Van de Walle. Proc. ICPS-22 World Scientific, Singapore, p. 2327, (1995).

95 TP PT C. Stampfl and C. Van de Walle. “Doping of AlxB GaB 1B – xN.”B Applied Physics Letters, 72, pp. 459 – 461, (1998).

96 TP PT M. D. McCluskey, N. M. Johnson, C. G. Van de Walle, D. P. Bour, M. Kneissl, and W. Walukiewicz. "Metastability of oxygen donors in AlGaN." Physics Review Letter, 80, pp. 4008, (1998).

97 TP PT S. T. Bradley, S. H. Goss, L. J. Brillson, J. Hwang, and W. J. Schaff. “Deep level defects and doping in high Al mole fraction AlGaN.” Journal of Vacuum Science and Technology. B 21, pp. 2558, (2003).

98 TP PT S. Fisher, C. Wetzel, E. E. Haller, and B. K. Meyer. “On p-type doping in GaN—acceptor binding energies.” Applied Physics Letters. 67, pp. 1298 (1995).

99 TP PT C. G. Van de Walle, C. Stampfl, J. Neugebauer, M. D. McCluskey, and N. M. Johnson. “Doping of AlGaN alloys.” MRS Internet J. Nitride Semiconductor Resserarch, 4S1, pp. G10.4, (1999).

100 TP PT B. Pödör. “On the dislocation scattering in silicon-on-insulator films.” Physica Status Solidi A, 16, pp. K197 – K200, (1970).

101 TP PT R. McClintock, A. Yasan, K. Mayes, D. Shiell, S.R. Darvish, P. Kung, and M. Razeghi. “High Quantum Efficiency AlGaN Solar-Blind p-i-n Photodiodes.” Applied Physics Letters, 84 (8), p.1248-1250, (2004).

102 th TP PT M. Razeghi, 7P P Wide Bandgap III-Nitride Workshop, Richmond, VA, March 10-12, (2002).

103 TP PT P. Cantu, S. Keller, U. Mishra, S. DenBaars. “Metalorganic chemical vapor deposition of highly conductive Al0.65Ga0.35N films.” Applied Physics Letters, 82, pp. 3683-3685, (2003).

104 TP PT V. Adivarahan, G. Simin, G. Tamulaitis, R. Srinivasan, J. Yang, A. Khan, M. Shur, R. Gaska. “Indium–silicon co-doping of high-aluminum-content AlGaN for solar blind photodetectors.” Appied Physics Letters, 79, pp. 1903-1905, (2001).

292

105 TP PT S. Yamaguchi, M. Kariya, S. Nitta, H. Amano, I. Akasaki. “Strain relief by In-doping and its effect on the surface and on the interface structures in (Al)GaN on sapphire grown by metalorganic vapor-phase epitaxy.” Applied Surface Science, 159-160, pp. 414-420, (2000).

106 TP PT J.I. Pankove, E.A. Miller, J.E. Berkeyheiser. “GaN electroluminescent diodes.” RCA Review, 32, pp. 383, (1971).

107 TP PT H. Amano, M. Kito, K. Hiramatsu, and I. Akasaki. “P-type Conduction in Mg-Doped GaN Treated with Low-Energy Electron Beam Irradiation (LEEBI).“ Japanese Journal of Applied Physics, 28, pp. L2112, (1989).

108 TP PT S. Nakamura, T. Mukai, M. Senho, N, Iwasa. "Thermal Annealing Effects on P-Type Mg- Doped GaN Films." Japanese Journal of Applied Physics, 31, pp. L139, (1992).

109 TP PT S. Nakamura, N. Iwasa, M. Senoh, and T. Mukai. “Hole compensation mechanism of p-type GaN films.” Japanese Journal of Applied Physics, 31, pp. 1258, (1992).

110 TP PT J. A. Van Vechten, J. D. Zook, R. D. Hornig, and B. Goldenberg. “Defeating Compensation in Wide Gap Semiconductors by Growing in H that is Removed by Low Temperature De- Ionizing Radiation.” Japanese Journal of Applied Physics, 31, pp. 3662, (1992).

111 TP PT P. Kung, A. Yasan, R. McClintock, S. Darvish, K. Mi, and M. Razeghi. “Future of AlxGa1- xN Materials and Device Technology for Ultraviolet Photodetectors.” Proceedings of SPIE 4650, pp. 199 – 211, (2002).

112 TP PT K. B. Nam, M. L. Nakarmi, J. Li, J. Y. Lin, and H. X. Jiang. “Mg acceptor level in AlN probed by deep ultraviolet photoluminescence.” Applied Physics Letters, 83, pp. 878 – 880, (2003).

113 TP PT A. Yasan, R. McClintock, S. R. Darvish, Z. Lin, K. Mi, P. Kung, M. Razeghi. “Characteristics of high-quality p-type AlxGa1–xN/GaN superlattices.” Applied Physics Letters, 80, pp. 2108 – 2100, (2002).

114 TP PT H. Wang, J. Zhang, C. Chen, Q. Fareed, J. Yang, A. Khan. “AlN/AlGaN superlattices as dislocation filter for low-threading-dislocation thick AlGaN layers on sapphire.” Applied Physics Letters, 81, pp. 604-606 (2002).

115 c TP PT A. Yasan, R. MPU Clintock,UP K. Mayes, D. Shiell, L. Gautero, S. R. Darvish, P. Kung, and M. Razeghi. “4.5 mW operation of AlGaN-based 267 nm deep-ultraviolet light-emitting diodes.” Applied Physics Letters, 83, pp. 4701-4703, (2003).

293

116 TP PT J. Zhang, A. Chitnis, V. Adivarahan, S. Wu, V. Mandavilli, R. Pachipulusu, M. Shatalov, G. Simin, J. Yang, A. Khan. “Milliwatt power deep ultraviolet light-emitting diodes over sapphire with emission at 278 nm.” Applied Physics Letters, 81, pp. 4910-4912, (2002).

117 TP PT R. McClintock, A. Yasan, K. Mayes, D. Shiell, S.R. Darvish, P. Kung, and M. Razeghi. “High Quantum Efficiency AlGaN Solar-Blind p-i-n Photodiodes.” Applied Physics Letters, 84 (8), p.1248-1250, (2004).

118 TP PT P. Cantu, S. Keller, U. Mishra, S. DenBaars. “Metal-organic chemical vapor deposition of highly conductive Al0.65Ga0.35N films.” Applied Physics Letters, 82, pp. 3683-3685, (2003).

119 TP PT V. Adivarahan, G. Simin, G. Tamulaitis, R. Srinivasan, J. Yang, A. Khan, M. Shur, R. Gaska. “Indium–silicon co-doping of high-aluminum-content AlGaN for solar blind photodetectors.” Applied Physics Letters, 79, pp. 1903-1905, (2001).

120 TP PT S. Yamaguchi, M. Kariya, S. Nitta, H. Amano, I. Akasaki. “Strain relief by In-doping and its effect on the surface and on the interface structures in (Al)GaN on sapphire grown by metalorganic vapor-phase epitaxy.” Applied Surface Science, 159-160, pp. 414-420, (2000).

121 TP PT C. Collins, U. Chowdhury, M. Wong, B. Yang, A. Beck, R. Dupuis, J. Campbell. “Improved

solar-blind detectivity using an AlxB GaB 1–xB NB heterojunction p–i–n photodiode.” Applied Physics Letters, 80, pp. 3754-3756, (2002).

122 TP PT R. McClintock, A. Yasan, K. Mayes, D. Shiell, S.R. Darvish, P. Kung and M. Razeghi. “High Quantum Efficiency Solar-Blind Photodetectors.” Proceeding of the SPIE, 5359, pp. 434, (2004).

123 TP PT N. Biyikli, I. Kimukin, T. Kartalogl, O. Aytur, and E. Ozbay. “High-speed solar-blind AlGaN-based metal–semiconductor–metal photodetectors.” Physica Status Solidi (C), 0 (7), pp. 2314-2317, (2003).

124 TP PT V. Adivarahan, G. Simin, J. W. Yang, A. Lunev, M. Asif Khan, N. Pala, M. Shur, and R.

Gaska. “SiO2B B -passivated lateral-geometry GaN transparent Schottky-barrier detectors.” Applied Physics Letters, 77 (6), pp. 863-865, (2000).

125 TP PT E.Monroy, F.Calle, J.L.Pau, E.Munoz ,M.Verdu, F.J.Sanchez, M. T. Montojo, F.Omnes, Z.Bougrioua, I.Moerman, and E. San Andres. “Effect of Dielectric Layers on the Performance of AlGaN-Based UV Schottk Photodiodes.” Physica Status Solidi (A), 188 (1), pp. 307–310, (2001).

126 TP PT C. Collins, T. Li, A. Beck, R. Dupuis, J. Campbell, J. Carrano, M. Schurman, I. Ferguson. “Improved device performance using a semi-transparent p-contact AlGaN/GaN

294

heterojunction positive-intrinsic-negative photodiode.” Applied Physics Letters, 75, pp. 2138-2140, (1999).

127 TP PT W. J. Tropf and M. E. Thomas, “Aluminum oxide Al2O3 revisited,” in Handbook of Optical Constants of Solids III, ed. Palik, Academic Press, San Diego, Calif., pp. 653–682, (1998).

128 TP PT O. Ambacher, M. Arzberger, D. Brunner, H. Angerer, F. Freudenberg, N. Esser, T. Wethkamp, K. Wilmers, W. Richter, M. Stutzmann. “AlGaN-Based Bragg Reflectors.” MRS Internet Journal of Nitride Semiconductor Research, 2 (22), pp.1-12, (1997).

129 TP PT J.F. Muth, J.D. Brown, M.A.L. Johnson, Zhonghai Yu, R.M. Kolbas, J.W. Cook, Jr., J.F. Schetzina. “Absorption Coefficient and Refractive Index of GaN, AlN, and AlGaN Alloys.” MRS Internet Journal Nitride Semiconductor Research, 4S1 (G5.2), pp. 1-6, (1999).

130 TP PT C. H. Chen, H. Liu, D. Steigerwald, W. Imler, C. P. Kuo, M. G. Craford, M. Ludowise, S. Lester, and J. Amano. “A study of parasitic reactions between NH3 and TMGa or TMAl.” Journal of Electronic Materials 25, pp. 1004, (1996).

131 TP PT J.D. Brown, Zhonghai Yu, J. Matthews, S. Harney, J. Boney, J. F. Schetzina, J. D. Benson , K. W. Dang, C. Terrill , Thomas Nohava, Wei Yang, and Subash Krishnankutty. “Visible- Blind UV Digital Camera Based On a 32 x 32 Array of GaN/AlGaN p-i-n Photodiodes.” MRS Internet Journal of Nitride Semiconductor Research, 4(9), pp. 1-6, (1999).

132 TP PT J.D. Brown, J. Matthews, S. Harney, J. Boney.” High-Sensitivity Visible-Blind AlGaN Photodiodes and Photodiode Arrays.” MRS Internet Journal of Nitride Semiconductor Research, 5S1(W1.9), (1999).

133 TP PT B. Yang, K. Heng, T. Li, C. J. Collins, S. Wang, R. D. Dupuis, J. C. Campbell, M. J.

Schurman, and I. T. Ferguson. “32×32 Ultraviolet Al0.1B GaB 0.9B N/GaNB p-i-n photodetector array.” Quantum Electronics Letters, 36(11), pp. 1229, (2000).

134 TP PT J.D. Brown, J. Boney, J. Matthews, P. Srinivasan, and J.F. Schetzina. “UV-Specific (320- 365 nm) Digital Camera Based On a 128x128 Focal Plane Array of GaN/AlGaN p-i-n Photodiodes.” MRS Internet Journal of Nitride Semiconductor Research, 5(6), (2000).

135 TP PT P. Lamarre ,A. Hairston, S. P. Tobin, K. K. Wong, A. K. Sood, M. B. Reine, M. Pophristic, R. Birkham, I. T. Ferguson, R. Singh, C. R. Eddy, Jr., U. Chowdhury, M. M. Wong, R. D. Dupuis, P. Kozodoy, and E. J. Tarsa. “AlGaN UV Focal Plane Arrays.” Physica Status Solluti (A), 188 (1), pp. 289, (2001).

136 TP PT J. P. Long, S. Varadaraajan, J. Matthews, and J. F. Schetzina. “UV detectors and focal plane array imagers based on AlGaN p-i-n photodiodes.” Opto-Electronics Review, 10(4), pp. 251, (2002).

295

137 TP PT “ISC9809 320 x 256 Low Background ROIC “, Indego Systems, Inc.,

http://www.indigosystems.com/product/roic_9809.htmlHTU ,UTH (2003).

138 TP PT “SE-IR Detailed camera specification”, SE-IR Corp., HTU http://www.seir.com/detcam.htm UT ,H (2004).

139 TP PT J. L. Vampola, “Readout Electronics for Infrared Sensors” Chapter 5 in The Infrared and Electro-Optical Systems Handbook, Vol. 3, edited by W. D. Rogatto, Executive editors J. S. Accetta, and J. L. Shumaker, Co-publishers: Infrared Information Analysis Center, Michigan, USA, and SPIE Optical Engineering Press, Washington, USA, (1993).

140 TP PT “ISC9809 320 x 256 Low Background ROIC “, Indego Systems, Inc., http://www.indigosystems.com/product/roic_9809.html, (2003).

141 TP PT Z. C. Huang, J. C. Chen, D. B. Mott, and P. K. Shu. “High Performance GaN Linear Array.” Electronics Letters, 32 (14), pp. 1324-1325, (1996).

142 TP PT B.W. Lim, S. Gangopadhyay, J.W. Wang, A. Osinsky, Q. Chen, M.Z. Anwar, and M.A. Khan. “8 x 8 GaN Schottky barrier photodiode array for visible-blind imaging.” Electronics Letters, 33 (7), pp. 633-634, (1997).

143 TP PT Ted Z. C. Huang, David B. Mott, and Anh La. “Development of 256 x 256 GaN Ultraviolet Imaging Arrays.” Proceedings of the SPIE, 3764, pp. 254-260 (1999).

144 TP PT J.D. Brown, J. Boney, J. Matthews, P. Srinivasan, and J.F. Schetzina. “UV-Specific (320- 365 nm) Digital Camera Based On a 128x128 Focal Plane Array of GaN/AlGaN p-i-n Photodiodes.” MRS Internet Journal of Nitride Semiconductor Research, 5(6), (2000).

145 TP PT J.D. Brown, Zhonghai Yu, J. Matthews, S. Harney, J. Boney, J. F. Schetzina, J. D. Benson , K. W. Dang, C. Terrill , Thomas Nohava, Wei Yang, and Subash Krishnankutty. “Visible- Blind UV Digital Camera Based On a 32 x 32 Array of GaN/AlGaN p-i-n Photodiodes.” MRS Internet Journal of Nitride Semiconductor Research, 4(9), (1999).

146 TP PT J.D. Brown, J. Matthews, S. Harney, J. Boney, J. F. Schetzina, J. D. Benson, K. V. Dang, T. Nohava, W. Yang, and S. Krishnankutty. “High-Sensitivity Visible-Blind AlGaN Photodiodes and Photodiode Arrays.” MRS Internet Journal of Nitride Semiconductor Research, 5S1 (W1.9), (1999).

147 TP PT B. Yang, K. Heng, T. Li, C. J. Collins, S. Wang, R. D. Dupuis, J. C. Campbell, M. J. Schurman, and I. T. Ferguson. “32 32 Ultraviolet Al Ga N/GaN p-i-n Photodetector Array.” Quantum Electronics Letters, 36 (11), pp. 1229-1231, (2000).

296

148 TP PT J. P. Long, S. Varadaraajan, J. Matthews, and J. F. Schetzina. “UV Detectors and Focal Plane Array Imagers Based on AlGaN p-i-n Photodiodes.” Opto-Electronics Review, 10 (4), pp. 251-260, (2002).

149 TP PT P. Lamarre, A. Hairston, S. Tobin, K. K. Wong, M. F. Taylor, A. K. Sood, M. B. Reine, M. J. Schurman, I. T. Ferguson, R. Singh, and C. R. Eddy, Jr. “AlGaN p-i-n Photodiode Arrays for Solar-Blind Applications.” Material Research Society Symposium Proceedings, 639, p. G10.9.1, (2001).

150 TP PT P. Lamarre ,A. Hairston, S. P. Tobin, K. K. Wong, A. K. Sood, M. B. Reine, M. Pophristic, R. Birkham, I. T. Ferguson, R. Singh, C. R. Eddy, Jr., U. Chowdhury, M. M. Wong, R. D. Dupuis, P. Kozodoy, and E. J. Tarsa. “AlGaN UV Focal Plane Arrays.” Physica Status Soluti. (A), 188 (1), pp. 289–292, (2001).

151 TP PT E. J. Tarsa, P. Kozodoy, J. Ibbetson, B. P. Keller, G. Parish, and U. Mishra. “Solar-Blind AlGaN-Based Inverted Heterostructure Photodiodes.” Applied Physics Letters, 77 (3), pp. 316-318, (2000).

152 TP PT R. McClintock, A. Yasan, K. Mayes, D. Shiell, S.R. Darvish, P. Kung and M. Razeghi. “High Quantum Efficiency AlGaN Solar-Blind Photodetectors.” Applied Physics Letters, 84, pp. 1248-1250, (2004).

153 TP PT R. McClintock, A. Yasan, K. Mayes, D. Shiell, S.R. Darvish, P. Kung and M. Razeghi. “High Quantum Efficiency Solar-Blind Photodetectors.” Proceeding of the SPIE, 5359, pp. 434, (2004).

154 TP PT K. Mayes, A. Yasan, R. McClintock, D. Shiell, S.R. Darvish, P. Kung, and M. Razeghi. “High Power 280 nm AlGaN Light Emitting Diodes Based on an Asymmetric Single Quantum Well.” Applied Physics Letters, 84, pp. 1046-1048, (2004).

155 TP PT J. Jiang, K. Mi, R. McClintock, M. Razeghi, G.J. Brown, C. Jelen. “Demonstration of 256x256 Focal Plane Arrays Based on Al-free GaInAs/InP QWIP.” Photonics Technology Letters, IEEE, 15(9), pp. 1273, (2003).

156 TP PT Hamamatsu Photonics, K.K., http://usa.hamamatsu.com/, PMTs based upon Cs-Te photocathodes such as R1080.

157 TP PT DARPA BAA06-14, Deep Ultraviolet Avalanche Photodetectors (DUVAP), (2005)

158 TP PT N. Biyikli, O. Aytur, I. Kimukin, T. Tut, E. Ozbay. “Solar-blind AlGaN-based Schottky photodiodes with low noise and high detectivity.” Applied Physics Letters, 81 (17), pp. 3272 – 3275, (2002).

297

159 TP PT J.P. Long, S. Varadaraajan, J. Matthews, and J.F. Schetzina. “UV detectors and focal plane array imagers based on AlGaN p-i-n photodiodes.” Optoelectronics Review, 10 (4), pp. 251- 261, (2002).

160 TP PT J. L. Pau, E. Monroy, M. A. Sánchez-García, E. Calleja, and E. Munoz. “AlGaN ultraviolet photodetectors grown by molecular beam epitaxy on Si(111) substrates” Materials Science and Engineering B, 93, pp. 159, (2002).

161 TP PT K. McIntosh, R. Molnar, L. Mahoney, M. Geis, K. Molvar, I. Melngailis, R. Aggarwal, W. Goodhue, S. Choi, and D. Spears. “GaN avalanche photodiodes grown by hydride vapor- phase epitaxy.” Applied Physics Letters, 75, pp. 3485, (1999).

162 TP PT J. Carrano, D. Lambert, C. Eiting, C. Collins, T. Li, S. Wang, A. Beck, R. Dupuis, and J. Campbell. “GaN avalanche photodiodes.” Applied Physics Letters, 76, pp. 924, (2000).

163 TP PT B. Yang, T. Li, K. Heng, C. Collins, S. Wang, J. Carrano, R. Dupuis, J. Campbell, M. Schurman, and I. Ferguson. “Low dark current GaN avalanche photodiodes.” Journal. Quantum Electronics, IEEE, 36, pp. 1389, (2000).

164 TP PT S. Verghese, K. McIntosh, R. Molnar, L. Mahoney, R. Aggarwal, M. Geis, K. Molvar, E. Duerr, and I. Melngailis. “GaN avalanche photodiodes operating in linear-gain mode and Geiger mode.” Transactions on Electron Devices, IEEE, 48, pp. 502, (2001).

165 TP PT Y. Wang, K. Brennan, and P. Ruden. “Theoretical study of a potential ultraviolet avalanching detector based on impact ionization out of confined quantum states.” Journal of Quantum Electronics, IEEE, 27, pp. 232, (1991).

166 TP PT P. Ruden and S. Krishnankutty. “A solar blind, hybrid III-nitride/silicon, ultraviolet avalanche photodiode.” Transactions on Electron Devices, IEEE, 46, pp. 2348, (1999).

167 TP PT C. Sevik and C. Bulutay. “Gain and temporal response of AlGaN solar-blind avalanche photodiodes: An ensemble Monte Carlo analysis.” Applied Physics Letters, 83, pp. 1382, (2003).

168 TP PT R. McClintock, A. Yasan, K. Minder, P. Kung, and M. Razeghi. “Avalanche multiplication in AlGaN based solar-blind photodetectors.” Applied Physics Letters, 87, pp. 241123-1 – 241123-3, (2005).

169 TP PT R. McClintock, K. Minder, A. Yasan, C. Bayram, F. Fuchs, P. Kung and M. Razeghi. “Solar-blind avalanche photodiodes.” Proceedings of the SPIE, 6127, pp. 61271D-1 – 61271D-10, (2005).

298

170 TP PT J. Zhang, H. Wang, W. Sun, V. Adivarahan, S. Wu, A. Chitnis, C. Chen, M. Shatalov, E. Kuokstis, J. Yang, and M. Asif Khan. “High-quality AlGaN layers over pulsed atomic-layer epitaxially grown AlN templates for deep ultraviolet light-emitting diodes.” Journal of Electronic Materials, 32, pp. 364, (2003).

171 TP PT M. Razeghi, US Patent 5831277, (1997).

172 TP PT I. Oguzman, E. Bellotti, K. Brennan, J. Kolnik, R. Wang, and P. Ruden. “Theory of hole initiated impact ionization in bulk zincblende and wurtzite GaN.” Journal of Applied Physics, 81, pp. 7827, (1997).

173 TP PT R. McClintock, J. L. Pau, K. Minder, C. Bayram, P. Kung, M. Razeghi. “Hole-initiated multiplication in back-illuminated GaN avalanche photodiodes.” Applied Physics Letters, 90, pp. 141112-1 – 141112-3, (2007).

174 TP PT D. Winston and R. Hayes. “SimWindows - A New Simulator for Studying Quantum–well Optoelectronic Devices.” Compound Semiconductors 1994 Institute of Physics Conference Series, 141, pp. 747, (1995).

175 TP PT G. Stillman and C. Wolfe. Chapter 5 “Avalanche Photodiodes” in “Semiconductors and semimetals”, edited by R Willardson (Academic Press, New York, 1977), Vol. 12, p.304.

176 TP PT S. Sze. “Physics of Semiconductor Devices, 2nd edition.” (John Wiley & Sons, New York, 1981), Chap. 13, p.744.

177 TP PT J. C. Carrano, D. J. H. Lambert, C. J. Eiting, C. J. Collins, T. Li, S. Wang, B. Yang, A. L. Beck, R. D. Dupuis, J. C. Campbell. “GaN Avalanche Photodiodes.” Applied Physics Letters, 76, pp. 924 – 927, (2000).

178 TP PT A. Osinsky, M. S. Shur, R. Gaska, Q. Chen, “Avalanche breakdown and breakdown luminescence in p-π-n GaN diodes.” Electronics Letters, IEEE. 34, pp. 691-692, (1998).

179 TP PT T. Tut, S. Butun, B. Butun, M. Gokkavas, H. Yu, E. Ozbay. “Solar-blind AlxB GaB 1–xB N-basedB avalanche photodiodes.” Applied Physics Letters, 87, pp. 223502-1 – 223502-3, (2005).

180 TP PT I. J. Oguzman, E. Belotti, K. F. Brennan, J. Kolnik, R. Wang, P. P. Ruden. “Theory of hole initiated impact ionization in bulk zincblende and wurtzite GaN.” Journal of Applied Physics, 81, pp. 7827-7834, (1997).

181 TP PT DARPA BAA06-14, Deep Ultraviolet Avalanche Photodetectors (DUVAP), (2005).

299

182 TP PT R. McClintock, A. Yasan, K. Minder, P. Kung, M. Razeghi, “Avalanche multiplication in AlGaN based solar-blind photodetectors.” Applied Physics Letters, 87, pp. 241123-1 – 241123-3, (2005).

183 TP PT T. Tut, M. Gokkavas, B. Butun, S. Butun, E. Ulker, and E. Ozbay, “Experimental evaluation

of impact ionization coefficients in AlxB GaB 1B −xNB based avalanche photodiodes.” Applied Physics Letters, 89, pp. 183524-1 – 183524-3, (2006).

184 TP PT A. Nishikawa, K. Kamakura, T. Akasaka, T. Makimoto, “High critical electric field of

AlxB GaB 1–xB NB p-i-n vertical conducting diodes on n-SiC substrates.” Applied Physics Letters, 88, pp. 173508-1 – 173508-3, (2006).

185 TP PT D. Winston and R. Hayes. “SimWindows - A New Simulator for Studying Quantum–well Optoelectronic Devices” Compound Semiconductors 1994 Institute of Physics Conference Series, 141, pp. 747, (1995).

186 TP PT X. A. Cao, H. Lu, S. F. LeBoeuf, C. Cowen, S. D. Arthur, W. Wang. “Growth and characterization of GaN PiN rectifiers on free-standing GaN.” Applied Physics Letters, 87, pp. 053503-1 – 053503-3, (2005).

187 TP PT G. Stillman and C. Wolfe. Chapter 5 “Avalanche Photodiodes” in “Semiconductors and semimetals”, edited by R Willardson (Academic Press, New York, 1977), Vol. 12, p.304.

188 TP PT A. Osinsky, M. S. Shur, R. Gaska, Q. Chen. “Avalanche Breakdown and Breakdown Luminescence in p-π-n GaN Diodes”, Electronics Letters, 34 (7), pp. 691-692, (1998).

189 TP PT S. Sze. “Physics of Semiconductor Devices”, 2nd edition (John Wiley & Sons, New York, 1981), Chap. 13, p.744.

190 TP PT S. L. Rumyantsev, N. Pala, M. S. Shur, R. Gaska, M. E. Levinshtein, M. Asif Khan, G. Simin, X. Hu, J. Yang, “Thin n-GaN films with low level of 1/f noise.” Electronics Letters, IEEE, 37, pp. 720 - 721, (2001).

191 TP PT S. Verghese, K. McIntosh, R. Molnar, L. Mahoney, R. Aggarwal, M. Geis, K. Molvar, E. Duerr, and I. Melngailis. “GaN avalanche photodiodes operating in linear-gain mode and Geiger mode.” Transactions on Electron Devices, IEEE, 48 (3), pp. 502-511, (2001).

192 TP PT I. J. Oguzman, E. Belotti, K. F. Brennan, J. Kolnik, R. Wang, P. P. Ruden. “Theory of hole initiated impact ionization in bulk zincblende and wurtzite GaN.” Journal of Applied Physics, 81(7827), pp. 7827-7834, (1997).

300

193 TP PT R. McClintock, J. L. Pau, K. Minder, C. Bayram, P. Kung, and M. Razeghi. “Hole-initiated multiplication in back-illuminated GaN avalanche photodiodes.” Applied Physics Letters, 90 (14), pp. 141112-1 – 141112-3, (2007).

194 TP PT J. Goldstien, D. Newbury, D. Joy, C. Lynman, P. Echlin, E. Lifshin, L. Sawyer, J. Michael. “Scanning Electron Microscopy and X-Ray Microanalysis, 3rd Edition.” Kluwer Academic, New York. pp. 23, (2003).

195 TP PT H.C. Casey, Jr., J. Muth, S. Krishnankutty, and J.M. Zavada. “Dominance of tunneling current and band filling in InGaN/AlGaN double heterostructure blue light-emitting diodes.” Applied Physics Letters, 68, pp. 2867 - 2869, (1996).

196 TP PT A. Chitnis, R. Pachipulusu, V. Mandavilli, M. Shatalov, E. Koukstis, J.P. Zhang, V. Adivarahan, S. Wu, G. Simin, and M. Asif Khan. “Low-temperature operation of AlGaN single-quantum-well light-emitting diodes with deep ultraviolet emission at 285 nm.” Applied Physics Letters, 81, pp. 2938 – 2930, (2002).

301 12. Curriculum Vita

12.1. Education

Northwestern University 2001-2007 Ph.D. in Electrical and Computer Engineering, June 2007

Northwestern University 1998-2001 Graduates with a B.S. in electrical engineering June 2001

Mary Institute and St. Louis Country Day School 1995-1998 Graduate with H.S. diploma, June 1998

12.2. Honors

Richter Trust fellowship 2006 – 2007 Presidential Fellowship Finalist 2006 Northwestern university fellowship 2004 – 2005 National defense science and engineering graduate (NDSEG) fellowship 2001 – 2004 SPIE Scholarship grant 2001 Motorola Undergraduate Research Scholarship 2000 – 2001

12.3. Graduate Coursework

2001 Fall (2001-09-24 to 2001-12-14) ECE 402 Quantum Devices A ECE 499 Independent Projects A MAT_SCI 451 Advanced Physics of Materials B

2002 Winter (2002-01-07 to 2002-03-22) ECE 403 Quantum Semiconductors A ECE 499 Independent Projects A MAT_SCI 460 Transmission Electron Microscopy A

2002 Spring (2002-04-02 to 2002-06-14) ECE 328 Numerical Methods for Engineers A ECE 405 Advanced Photonics A MAT_SCI 395 Special Topics: Nanomaterials A

302 2002 Fall (2002-09-25 to 2002-12-13) MAT_SCI 380 Surface Science and Spectrocopy A MAT_SCI 461-2 Diffraction Methods in Mat. Sci. A MECH_ENG 460 Introduction to MEMS A

2003 Winter (2003-01-06 to 2003-03-21) MAT_SCI 361 Crystallography A

2003 Spring (2003-03-31 to 2003-06-13) MAT_SCI 360 Introduction to Electron Microscopy A MECH_ENG 495 Special Topics: Nanotechnology A Cumulative GPA: 3.933

12.4. List of Publications

1. R. McClintock, J. L. Pau, K. Minder, C. Bayram, P. Kung, M. Razeghi. Hole-initiated multiplication in back-illuminated GaN avalanche photodiodes. Applied Physics Letters Vol. 90, 2007.

2. P. Kung, R. McClintock, J. L. Pau, K. Minder, C. Bayram, M. Razeghi. III-nitride avalanche photodiodes. SPIE Conference, January, 2007, San Jose, CA Proceedings – Quantum Sensing and Nanophotonic Devices IV, Vol. 6479, 2007.

3. K. Minder, F. Teherani, D. Rogers, C. Bayram, R. McClintock, P. Kung, M. Razeghi. Etching of ZnO towards the development of ZnO homostructure LEDs. SPIE Conference, January, 2007, San Jose, CA Proceedings – Zinc Oxide Materials and Devices II, Vol. 6474, 2007.

4. M. Razeghi, P.Y. Delaunay, B.M. Nguyen, A. Hood, D. Hoffman, R. McClintock, Y. Wei, E. Michel, V. Nathan and M. Tidrow. First Demonstration of ~ 10 microns FPAs in InAs/GaSb SLS. IEEE LEOS Newsletter 20 (5)-- October 1, 2006.

5. R. McClintock, K. Minder, A. Yasan, C. Bayram, F. Fuchs, P. Kung and M. Razeghi, Solar-blind avalanche photodiodes. SPIE Conference, January 23, 2006, San Jose, CA Proceedings – Quantum Sensing III, Vol. 6127, 2005.

6. R. McClintock, A. Yasan, K. Minder, P. Kung, and M. Razeghi, Avalanche multiplication in AlGaN based solar-blind photodetectors. Applied Physics Letters 87 (241123) December 2005.

7. McClintock R, Yasan A, Mayes K, Kung P, Razeghi M. Back-illuminated solar-blind photodetectors for imaging applications. SPIE Conference, January 25, 2005, San Jose, CA Proceedings – Quantum Sensing II, Vol. 5732, 2005.

303 8. Yasan A, McClintock R, Mayes K, Kung P, Razeghi M. AlGaN-based deep UV light emitting diodes with peak emission below 255 nm. SPIE Conference, January 25, 2005, San Jose, CA Proceedings – Quantum Sensing II, Vol. 5732, 2005.

9. D.J. Rogers, F. Hosseini Teherani, A. Yasan, R. McClintock, K. Mayes, S.R. Darvish, P. Kung, M. Razeghi and G. Garry, ZnO Thin Film Templates for GaN-based Devices. SPIE Conference, January 25, 2005, San Jose, CA Proceedings – Quantum Sensing II, Vol. 5732, 2005.

10. McClintock R, Mayes K, Yasan A, Shiell D, Kung P, Razeghi M. 320x256 Solar-Blind

Focal Plane Arrays based on AlxB GaB 1-xB N.B Applied Physics Letters 86 (1) January 2005, p. 011117.

11. McClintock R and Razeghi M. III-Nitride UV Photodetectors a book chapter in III- Nitride Optoelectronic Devices, edited M. Henini and M. Razeghi, Elsevier Science Publishers, January 9, 2005.

12. McClintock R, Yasan A, Mayes K, Shiell D, Darvish SR, Kung P, Razeghi M. High quantum efficiency solar-blind photodetectors. SPIE Conference, January, 2004, San Jose, CA Proceedings – Quantum Sensing, Vol. 5359, 2004

13. Yasan A, McClintock R, Mayes K, Shiell D, Darvish SR, Kung P, Razeghi M. Growth of deep UV light emitting diodes by metalorganic chemical vapor deposition. SPIE Conference, January, 2004, San Jose, CA Proceedings – Quantum Sensing, Vol. 5359, 2004.

14. McClintock R, Yasan A, Mayes K, Shiell D, Darvish SR, Kung P, Razeghi M. High quantum efficiency AlGaN solar-blind p-i-n photodiodes Applied Physics Letters 84 (8) Febuary 2004, p. 1248-1250.

15. Mayes K, Yasan A, McClintock R, Shiell D, Darvish SR, Kung P, Razeghi M. High- power 280 nm AlGaN light-emitting diodes based on an asymmetric single-quantum well Applied Physics Letters 84 (7) Febuary 2004, p. 1046-1048

16. Yasan A, McClintock R, Mayes K, Shiell D, Gautero L, Darvish SR, Kung P, Razeghi M. 4.5 mW operation of AlGaN-based 267 nm deep-ultraviolet light-emitting diodes Applied Physics Letters 83 (23) December 2003, p. 1701-1703.

17. Yasan A, McClintock R, Mayes K, Kim DH, Kung P, Razeghi M. Photoluminescence study of AlGaN-based 280nm UV LEDs. Applied Physics Letters 83 (20) November 2003, p. 4083-4085.

18. Jiang JT, Mi K, McClintock R, Razeghi M, Brown GJ, Jelen C. Demonstration of 256 x 256 focal plane array based on Al-free GaInAs-InP QWIP. IEEE Photonics Technology Letters 15(9) September 2003, p. 1273-1275.

304

c 19. Yasan A, MP ClintockP R, Mayes K, Darvish SR, Kung P, Razeghi M, Molnar RJ 280 nm UV LEDs grown on HVPE GaN substrates Opto-Electronics Review 10 (4), December 2002, p. 287-289.

c 20. Yasan A, MP ClintockP R, Mayes K, Darvish SR, Zhang H, Kung P, Razeghi M, Lee SK, Han JY. Comparison of ultraviolet light-emitting diodes with peak emission at 340 nm grown on GaN substrate and sapphire. Applied Physics Letters, 81(12), September 16 2002, p. 2152-2153.

c 21. Yasan A, MP ClintockP R, Mayes K, Darvish SR, Kung P, Razeghi M. Top-emission ultraviolet light-emitting diodes with peak emission at 280 nm. Applied Physics Letters, 81(5), July 29 2002, p. 801-802.

c 22. Yasan A, MP ClintockP R, Mayes K, Darvish SR, Kung P, Razeghi M. Top-emission ultraviolet light-emitting diodes with peak emission at 280 nm. Virtual Journal of Nanoscale Science & Technology, 5, August 5 2002.

23. P. Kung, A. Yasan, R. McClintock, S. R. Darvish, K. Mi, M. Razeghi. Future of AlxGa1- xN materials and device technology for ultraviolet photodetectors. Proceedings of SPIE, 4650, May 2002, p. 199-206.

c 24. A. Yasan, R. MP ClintockP , S. R. Darvish, Z. Lin, K. Mi, P. Kung, and M. Razeghi.

Characteristics of high-quality p-type AlxB GaB 1-xB N/GaNB superlattices. Applied Physics Letters, 80 (12), 25 March 2001, p. 2108-2110.

c 25. P. Sandvik, K. Mi, F. Shahedipour, R. MP ClintockP , A. Yasan., P. Kung, M. Razeghi.

AlxB GaB 1-xB NB for solar-blind UV detectors. Journal of Crystal Growth, 231 (3), Oct. 2001, p. 366-370.

c 26. R. MP ClintockP , P. Sandvik, K. Mi, F. Shahedipour, A. Yasan, C. Jelen, P. Kung, M.

Razeghi. AlxB GaB 1-xB NB materials and device technology for solar blind ultraviolet photodetector applications. Proceedings of SPIE, 4288, June 2001 , p. 219-229.