The Effects of Thermal, Strain, and Neutron Irradiation on Defect Formation in AlGaN/GaN High Electron Mobility Transistors and GaN Schottky Diodes

Transcription

1 The Effects of Thermal, Strain, and Neutron Irradiation on Defect Formation in AlGaN/GaN High Electron Mobility Transistors and GaN Schottky Diodes DISSERTATION Presented in Partial Fulfillment of the Requirements for the Degree Doctor of Philosophy in the Graduate School of The Ohio State University By Chung-Han Lin, B.S, M.S. Graduate Program in Electrical and Computer Engineering The Ohio State University 2013 Dissertation Committee: Professor Leonard J. Brillson, Advisor Professor Steven. A. Ringel Professor Wu Lu

2 Copyright by Chung-Han Lin 2013

3 Abstract We use depth-resolved cathodoluminescence spectroscopy (DRCLS), Kelvin probe force microscopy (KPFM), and surface photovoltage spectroscopy (SPS) on a nanometer scale to map the temperature, stress, and defects inside GaN high electron mobility transistors (HEMTs). DRCLS maps temperature at localized depths, in particular within the twodimensional electron gas (2DEG) region during device operation. KPFM maps surface electric potential across the device, revealing lower potential patches that decrease rapidly with increasing off-state stress. CL spectra acquired at these patches exhibit defect emissions that increase with both on- and off-state stress and that increase with decreasing surface potential. SPS also reveals features of deep level gap states generated after device operation that reduce near band edge (NBE) emission and increase surface band bending. These techniques also reveals that electronic defects form in AlGaN layer due to field-induced stress accompanied with the decrease of the surface potential and the increase of the gate leakage current. The splitting of the AlGaN emission and the enhancement of yellow and blue band luminescence indicate crystal quality deterioration caused by stress. Our nanoscale measurements are consistent with defect mainly generation by inverse piezoelectric field-induced stress at the gate edge on the drain side at high voltage. Our DRCLS, SPS, time-resolved SPS (t-sps), current-voltage-temperature (I-V-T) shows that fast and thermal neutron preferentially affect device properties. Fast neutron ii

4 will induce defects in GaN by recoil and displacement damage whereas thermal neutron tends to enhance the interaction between metal/semiconductor interfaces due to heat. Time-resolved surface photovoltage spectroscopy (t-sps) results reveal a defect evolution of GaN under fast neutron irradiation that indicates low fast neutron dosage will enhance GaN properties a result which is confirmed by DRCLS results. XPS results show that Ti and Ni are more resistant than other metal but will interact with GaN at higher thermal neutron fluence. Our results show that fast and thermal neutrons are both detrimental electronic devices without proper protection. iii

5 Dedication Dedicated to my family, Pik Kei, Bella, and Jason. iv

6 Acknowledgments This work would not have been possible without the facilities, financial support, and especially academic advice from my advisor, Professor Leonard Brillson. I would like to express my sincere appreciation for his guidance. Former and current group member Daniel Doutt, Tyler Merz, Michael Hetzer, Evan Katz, and Zhichun Zhang, Mitchell Rutkowski, Yu-Feng Dong, and Dr. Snjezana Balaz are also acknowledge for useful and insightful discussion and equipment knowledge. Furthermore, I sincerely thank Daniel, Tyler, and Michael s help on experimental setups and training. At OSU, I would like to thank Dr. Jie Qiu and Professor Lei Cao at Department of Mechanical and Aerospace Engineering, Nuclear Engineering for assistance of neutron irradiation knowledge and experiments and Dr. Aaron R. Arehart for data analysis and discussion. Additionally, I would also thank Professor Jesús A. del Alamo at MIT for supplying samples and discussion. Professor Umesh K. Mishra at UC-Santa Barbara for sample providing. To my wife Pik Kei, thank you so much for supporting me, fed me, and encourages me when I am pursing my PhD. Thanks for taking care of our two lovely kids, Bella and Jason. v

7 Vita April 16, Born Tainan, Taiwan B.S. Electrical Engineering, Chang Gung University M.S. Electrical Engineering, Chang Gung University Army R.O.C Graduate Teaching Associate, Department of Electrical and Computer Engineering, The Ohio State University Publications 1. C.-H. Lin, T. A. Merz, D. R. Doutt, J. Joh, J. A. del Alamo, U. K. Mishra, and L. J. Brillson, Strain and temperature dependence of defect formation at AlGaN/GaN high electron mobility transistors on a nanometer scale, IEEE Trans. Electron Devices, vol. 59, no. 10, pp , Oct C.-H. Lin and L. J. Brillson, Scanning Probe Techniques Identify GaN HEMT Reliability Issues, Compound Semiconductor magazine, Jan 2011 Available: vi

8 d= C.-H. Lin, D. R. Doutt, U. K. Mishra, T. A. Merz, and L. J. Brillson, Field-Induced Strain Degradation of AlGaN/GaN High Electron Mobility Transistors on a Nanometer Scale, Appl. Phys. Lett., vol. 97, no. 3, pp , Nov C.-H. Lin, T. A. Merz, D. R. Doutt, M. J. Hetzer, J. Joh, J. A. del Alamo, U. K. Mishra, and L. J. Brillson, Nanoscale mapping of temperature and defect evolution inside operating AlGaN/GaN high electron mobility transistors, Appl. Phys. Lett., vol. 95, no. 3, pp , Jul T.-E. Nee, J.-C. Wang, H.-T. Shen, C.-H. Lin, and Y.-F. Wu, Observations of electrical and luminescence anomalies in InGaN/GaN blue light-emitting diodes, J. Vac. Sci Technol. A, vol 24, no. 4, pp , Jun R.-M. Lin, J.-C. Wang, C.-H. Lin, and T.-E. Nee, Improvement of carrier confinement in blue light emitting diode with InGaN/GaN multiquantum barriers, Proc. SPIE. Vol. 5722, pp , Jan T.-E. Nee, J.-C. Wang, C.-H. Lin, R.-M. Lin, C.-A. Huang, B.-R. Fang, and R.-Y. Wang, Cross sections for the investigation of the electroluminescence excitation of InGaN/GaN quantum wells in blue light-emitting diodes with multiquantum barriers, J. Vac. Sci. Technol. B, vol. 23, no. 3, pp , May J.-C. Wang, C.-H. Lin, R.-M. Lin, T.-E. Nee, B.-R. Fang, and R.-Y. Wang, Mater. Res. Soc. Symp. Proc., Characterization of the carrier confinement for InGaN/GaN light emitting diode with multiquantum barriers, vol. 831, E , 2005 vii

9 9. R.-M. Lin, C.-H. Lin, J.-C. Wang, T.-E. Nee, B.-R. Fang, and R.-Y. Wang, J. Cryst. Growth, Study of electroluminescence quenching in the InGaN/GaN blue diode with multi-quantum barrier structure, vol. 278, no. 1-4, , Feb T.-En Nee, K.-T. Chien, Y.-L. Chou, L.-C. Chou, C.-H. Lin, R.-M. Lin, B.-R. Fan, and S.-S. Chang, J. Vac. Sci. Technol. B, Effect of current spreading on luminescence improvement in selectively oxidized AlGaInP light-emitting diodes, vol. 21, no. 3, pp , May 2003 Fields of Study Major Field: Electrical and Computer Engineering viii

10 Table of Contents Abstract... ii Dedication... iv Acknowledgments... v Vita... vi List of Tables... xii List of Figures... xiii Chapter 1 Introduction Motivation Objectives Research outline... 6 Chapter 2 Background Material properties of GaN-related materials Crystal structure Polarization and 2DEG formation Reliability issues of AlGaN/GaN HEMTs Temperature ix

11 Hot electron degradation Defects and traps Electro-chemical reactions Inverse piezoelectric effect Diffusion Stability of Schottky and Ohmic contacts Metal-Semiconductor contacts Schottky barrier diode Current transport processes Thermionic Emission Theory Thermionic Field Emission Theory Field Emission Theory Recombination in the space charge region Neutron irradiation effect on GaN-based detectors Chapter 3 Experimental Techniques and Setups Depth resolved cathodoluminescence spectroscopy (DRCLS) Atomic force microscopy and Kelvin probe force microscopy Surface photovoltage spectroscopy and time-resolved surface photovoltage spectroscopy x

12 3.4 AlGaN/GaN HEMT electrical measurement and current-voltage-temperature experimental setups Device details and experimental procedures Chapter 4 Result and Discussion Reliability issues investigation of AlGaN/GaN HEMTs AlGaN/GaN HEMT operating temperature characterization AlGaN/GaN HEMT field induced stress characterization AlGaN/GaN HEMT defect characterization Reliability issues of AlGaN/GaN HEMTs due to on- and off-state stress Discussion Neutron irradiation effects of GaN Schottky diodes Neutron irradiation on GaN Neutron irradiation on GaN Schottky diode Discussion Chapter 5 Conclusion Chapter 6 Future work List of References xi

13 List of Tables Table 3.1 Experimental details of neutron irradiation on GaN Schottky diode. The thermal neutron capture cross section of cadmium is large and thus can be used to filter out thermal neutrons during neutron irradiation Table 4.1 Summary of electrical properties extracted from I-V-T measurement of GaN Schottky diode before and after neutron irradiation xii

14 List of Figures Figure 1.1 New devices using GaN-based semiconductor devices. From Ref. [1] Figure 2.1 Hexagonal wurtzite structure of GaN and AlN crystal. From Ref. [57] Figure 2.2 Schematic representation of c-plane (polar), a- and m-plane (non-polar), and γ- plane (semi-polar). From Ref. [58] Figure 2.3 Directions of spontaneous polarization P SP and piezoelectric polarization P PE of AlGaN/GaN heterostructures with Ga- and N- terminations. AlGaN and GaN layers are either strained or relaxed. The +σ indicates the difference of P SP and P PE in AlGaN and GaN layer induce positive charge at interface and free electrons tends to compensate and form 2DEG. From Ref. [61] Figure 2.4 (a) EL intensity distribution of an operating AlGaN/GaN HEMT. (b) EL spectrum of the same device. An equivalent temperature = 1700 K was extracted by fitting Maxwellian distribution. From Ref. [20] xiii

15 Figure 2.5 Polarization dependent EL spectra from an operating AlGaN/GaN HEMT. The intensity of EL parallel component is greater that the perpendicular one. From Ref. [67] Figure 2.6 Schematic diagram of an AlGaN/GaN HEMT 2DEG depletion caused by negative charge accumulates at surface and the equivalent circuit. (b) Band diagram of AlGaN/GaN heterostructure show before and after the negatively charged virtual gate forms. From Ref. [16] Figure 2.7 Source-drain current v.s. bias of AlGaN/GaN HEMTs: (a) with increasing virtual gate length X Q and (c) with increasing trap charge density N Q. From Ref. [18] Figure 2.8 Transmission electron microscope (TEM) images of defects formed in the drain side after reliability test where drain current I D drop (a) 19% and (b) 58% [24]. (c) TEM image of device showed a large defect and nano energy dispersive X-ray (EDX) results measured at point A and B [24]. A higher O, C, and Si concentration can be observed. From Ref. [24] Figure 2.9 (a) Theoretical representation of the formation of lattice disruption that the gate current may assist the surface contamination or oxygen residues interacts with AlGaN crystal. (b) TEM image of degraded AlGaN/GaN HEMTs showing lattice disruption region. From Ref. [25] xiv

16 Figure 2.10 Changes in (a) normalize I Dmax and (b) gate leakage current I G-OFF with V DS = 0 state (no current flow, drain and source site stress simultaneously), OFF state (high V DS, low I D ), and high-power state (high V DS, high I D ) [70]. (c) Pictures show GaN HEMT under electrical stress with high V DG and the conceptual I G degradation mechanism [21]. The tensile stress induced by stress voltage may break the crystal and induce a pathway for electrons flow from gate to 2DEG Figure 2.11 (a) Elastic energy density in AlGaN/GaN HEMT under OFF-state stress [23]. (b) Strain ε zz and stress σ xx in GaN determined from E 2 phonon shift and the averaged simulated E z -filed component at V SD = 40 V. Inset shows the average strain from the experimental data vs. applied voltage [71]. A concentration of elastic energy (or stress) in the AlGaN crystal close to the gate edge drain side area can be observed. The location of the highest elastic energy indicates the place where the crystal damages first Figure 2.12 Schematic of AlGaN/GaN HEMT degradation caused by diffusion: (a) diffusion along dislocation, (b) diffusion enhanced by inverse piezoelectric effect, (c) diffusion along the pit created by field-induced stress. From Ref. [27] Figure 2.13 (a) A metal/n-type semiconductor before (left) and after (right) contact with each other. (b) A metal/p-type semiconductor before (left) and after (right) contact each xv

17 other. The alignment of the Fermi levels between metal and semiconductor indicate equilibrium condition has been reached. From Ref. [78] Figure 2.14 Energy band diagram of metal on n-and p-type semiconductors under (a) no bias, (b) forward bias, and (c) reverse bias. From Ref. [79] Figure 2.15 Energy band diagram and carrier transport process of a Schottky barrier under forward bias: (1) thermionic emission, (2) thermionic field emission, (3) field emission, and (4) recombination in the space charge region Figure 2.16 Gallium and Nitrogen atoms transmute to Germanium and Oxygen due to thermal neutron. σ is thermal neutron capture cross-section and T 1/2 is the half life time Figure 2.17 Energy released by Gallium transmutes to germanium and nitrogen transmute to oxygen due to thermal neutron interaction. Q is the energy released by nuclear reaction Figure 3.1 An example of electron beam incident upon AlGaN/GaN heterostructure with electron probing energy E B varies from 1 to 10 kev with 90 incident angle. The depth indicated by peak electron-hole pair generation show that with DRCLS technique one can xvi

18 investigate properties from surface to interface then inside the semiconductor bulk volume Figure 3.2 Schematic of DRCLS measurement setups using JEOL JAMP-7800F SEM and optical signal collection systems. (Bottom) A close look of electron beam, half parabolic mirror, and sample positions located inside UHV SEM chamber. The stage position can be changed through a X-Y-Z manipulator. The CL signal from electron beam is gathered by mirror, reflected by the optical path, through monochromator, and analyzed by photo multiplier tube (PMT) Figure 3.3 (a) Schematic of an AFM system including cantilever, tip, laser, and position sensitive photodetector. (b) Metal and semiconductor with work function and with compensation voltage V CPD that aligns the metal and semiconductor vacuum level and. From Ref. [90] Figure 3.4 A Park XE-70 AFM/KPFM system Figure 3.5 (a) Schematic of KPFM and SPS system. The monochromatic incident light is from a white light source through a monochromator and optical fiber. (b) SPS spectrum from an n-type semiconductor with two defect levels E C - E 1 and E V + E 2 with band gap E g. (c) Electronic band diagram for an optical transition from defect level to conduction xvii

19 band with incident light energy equal to E 1. (d) Electronic band diagram for an optical transition from valence band to defect level with incident light energy equal to E Figure 3.6 An example of time resolved SPS (t-sps) measurement result showing CPD varies with light turns on and off. An rising and decaying CPD which can be used to extract of,,,, and for defect density calculation Figure 3.7 Park XE-70 AFM/KPFM system with optical add-ons for SPS and t-sps measurement. (b) Enlarged view shows positions of optical fiber, sample holder and cantilever Figure 3.8 Two types of EBIC stages that use for temperature and stress measurement. 55 Figure 3.9 and (b) EBARA closed cycle helium cooled cryogenic system with Keithley 2400 source meter. (c) Lakeshore 330 temperature controller. (d) Wire-bonded AlGaN/GaN HEMTs device mounted on the cold finger for I-V-T measurement Figure 4.1 Temperature distribution of U-08 operated at V DS = 3.5 V and V GS = -2 V. P1 is nearest the gate while P10 is furthest away. Inset photo shows SEM image of sourcedrain region measured. From Ref. [99] xviii

20 Figure 4.2 (a) SEM image of virgin sample M-01 source-gate-drain areas across which temperatures were measured. Dashed red rectangles mark extrinsic drain and source areas. (b) Temperature distribution across extrinsic drain and source area at V DS = 6 V, V GS = -1V, I D = 1 A/mm with E B = 10 kev. Dashed black lines are guides to the eye. Temperature increases monotonically from drain to gate overhang edge. From Ref. [100] Figure 4.3 DRCLS cross-sectional temperature distribution of sample M-01: (a) Whole device, (b) Device drain side close to gate overhang. V DS = 6 V, V GS = -1 V, I D = 1 A/mm, and 8 < E B < 22 kev. Multiple hot spots are apparent. From Ref. [100] Figure 4.4 (a) Corresponding NBE shift with applied OFF-sate stress at drain side gate edge area. (b) External stress caused by applied voltage under OFF-state stress. From Ref. [104] Figure 4.5 Field-induced-stress distribution of a virgin device caused by applied bias under off-state stress. S, G, D denote source, gate, and drain metal contact. Dashed lines are guides to the eye. A maximum ~ 0.29 GPa compressive stress increase is evident at the gate edge drain side area. Field-induced off-state stress increases the most at drainside gate edge region. From Ref. [100] xix

21 Figure 4.6 For U-08, (a) SEM image shows the Source (S), Gate (G) and Drain (D) region. (b) CL map of NBE emission (middle) and KPFM result (side) taken in the same (a) region. Red dashed circles show some of the same higher defect, higher potential region for CL and KPFM map, respectively, where devices degrade faster. From Ref. [99] Figure 4.7 DRCLS spectrum at one of lowest potential areas (inset) within extrinsic drain region after Off-state stress showing 2.2 ev YB, BB, and 3.45 ev NBE peaks. From Ref. [104] Figure 4.8 DC output characteristics and I G-OFF as a function of (a) Off-state stress V DS = V and V GS = - 6 V, and (b) on-state stress V DS = 6 V, V GS = 0 V, and I D = 0.75 A/mm. Inset show the DC-IV output characteristics before and after (a) off-state and (b) on-state stress. Green arrows indicate the point monitored by AFM/KPFM. An 2.6x increase of the I G-off and decrease of I D after off-state stress can be observed, faster after V cri = 28 V. I G-off and I D show an opposite trend after on-state stress. Changes in output characteristics are correlated directly with surface potential and morphology. From Ref. [104] Figure 4.9 (a) KPFM results which shows the evolution of surface potential under Offstate stress at upper and lower region of AlGaN/GaN HEMTs. AFM images show upper and lower scanning areas at V DG = 36 V. The SEM image in (b) shows the corresponding xx

22 AFM/KPFM scanning area. The red dashed circles show regions where potentials change faster. (c) The SEM image indicates where upper scan area (rotated 45 clockwise) failure occurs with increasing Off-state stress. From Ref.[104] Figure 4.10 (a) KPFM potential maps before and after 11 minute on-state stress. Surface potential varies less with on-state vs. off-state stress. AFM images show upper and lower scanning areas at time = 11 min Figure 4.11 (a) Representative pre-stress DRCLS spectra with YB, BB, and NBE peaks. (b) Gate current characteristics before (black) and after (red) stress of sample M-01. The increase of the gate leakage current indicates the degradation of gate Schottky contact after stress. (c) Position-dependent averaged YB/NBE ratio shows largest increase at region under gate-drain side after 12 hours, V DS = 10 V, V GS = - 2 V, I D = 0.47 A/mm stress. Averaged BB/NBE ratio shows a much weaker response to local stress. The horizontal dashed lines represent the reference points. From Ref. [100] Figure 4.12 (a) DRCLS results of sample U-08 taken at region (1) with E B = 5 kev at 12 K, (b) SPS maps taken from region (1) - (3), (c) CL map of NBE emission (middle) and KPFM map of potential (side). Dashed lines delineate extrinsic drain and source areas. Both red and black circles show similar higher defect, higher potential regions for CL and KPFM maps. SPS spectra reveal a defect 1.2 ev above the valence band that increases with DRCLS defect emission intensities linked to device degradation. From Ref. [100] 83 xxi

23 Figure 4.13 (a) KPFM maps of sample U-01 from two representative areas of the same device along the width of the transistors show surface potential distribution and averaged defect emission after off- state stress. Device layout is indicated in the figure: source (S), gate (G), and drain (D). (b) and (c) Averaged YB & BB /NBE ratio correspond to areas denoted in (a). In general, regions with lower potential correlate with higher YB or BB defect emission. The horizontal dashed lines represent the reference points. From Ref. [100] Figure 4.14 (a) KPFM maps of sample U-01 from two representative areas of the same device along the width of the transistors show surface potential distribution after on-state stress. Device layout is indicated in the figure: source (S), gate (G), and drain (D). (b) and (c) Averaged YB and BB defect emission correspond to areas denoted in (a). The horizontal dashed lines represent the reference points. From Ref. [100] Figure 4.15 (a) OFF-state KPFM maps showing numbered upper and lower low potential regions and (b) corresponding surface potential and average YB/NBE intensity ratio increases with OFF-state stress. From color potential scale (red higher, blue lower), YB/NBE increases most at lowest potential regions 6, 8, 9, and 10. Higher potential patches display slower changes. 1 and 2 correspond to extrinsic drain and drain-side gate foot of an unstressed reference device. (c) Surface potential variation vs. V DG and corresponding self-consistent electrostatic defect density (smooth lines). From Ref. [104] xxii

24 Figure 4.16 Combined off- and on-state surface potential vs. (a) YB/NBE and (b) BB/NBE ratio from individual point spectra. Dashed lines are guides to the eye. The ~ V surface potential corresponds to pre-stress potential, e.g. R1 and R2. Dashed circle denotes the FEA area where defect emissions are strongly perturbed by catastrophic lattice disruption. In general, YB/NBE and BB/NBE ratios increase linearly as surface potential decreases. From Ref. [100] Figure 4.17 Surface potential evolution with applied stress voltage V DG = 0 33 V. The gate edge drain-side area exhibits a threshold behavior with increasing V DG. Note that the surface potential starts to decrease before V crit ~ 25 V. Dashed lines are guides to the eye. Inset shows I G-Off increases slowly with lower stress voltage, sharply rising if it exceeds Vcrit, and a 6.5x increase can be observed. The changing of surface potential and the degrading of electrical properties indicate the formation of defects during stress Figure 4.18 (a) D-DRCLS probing depth extracted by Monte Carlo simulation. (b) Low temperature D-DRCLS spectrum before stress with different probing depth. A 3.48 ev GaN NBE, a ~ 4.1 ev AlGaN, and two obscured defect related yellow and blue luminescence. The lack of the YB and BB indicate good material properties before stress xxiii

25 Figure 4.19 Low temperature D-DRCLS spectrum after stress. A splitting of the AlGaN emission, i.e. the Franz-Keldysh redshift and crystal relaxation blue shift, indicted the deterioration of the crystal quality due to stress. A 3.75 ev defect emission appears in the spectrum and can be precisely located in AlGaN layer because of nanometer scaled depth resolution of DRCLS. A rigid red shift of BB and AlGaN emission indicates the defect responsible for BB enhancement close to surface is in AlGaN. The overall results reveals field-induced stress cracks the AlGaN crystal and induces crystallographic defects Figure 4.20 The distribution of (a) YB and (b) BB in AlGaN/GaN HEMT before and after stress extracted by D-DRCLS technique. Dashed lines are guides to the eye. A 10x increase of YB at surface and 1.7x increase of BB in AlGaN layer can be observed. These results shows field-induced stress affects the crystal quality not only at surface, but also extends into devices Figure 4.21 SPS spectra at drain side gate edge region (a) before, and (b) after V DG = 33 V stress. E C 1.35 ev and E V ev defects form at drain side gate edge compare with SPS spectra before stress. The SPS spectra are similar before and after stress. These defects correlate with our D-DRCLS results showing YB BB, and 3.75 ev defects increasing after off-state stress. These defects are electrically-active because they not only change surface potential but also degrade electrical properties of AlGaN/GaN HEMTs xxiv

26 Figure 4.22 (a) A summary of defect distribution measured by DRCLS. Dashed lines are guides to the eye. The results shows YB is significant at surface, quickly fades away when probing into the device whereas BB starts to increase at g-algan/algan interface, peaks at AlGaN layer, and decreases when approaching AlGaN/GaN interface ev defect emission is mainly in AlGaN layer, neither in g-algan nor in GaN layer. (b) The correlation between our scanning probe techniques and other techniques. Our results are comparable to previous published results but with nanometer scale depth resolution Figure 4.23 Current-voltage (I-V) characteristics of Schottky diodes irradiated with (a) fast, and (b) fast + thermal neutron irradiation. Inset shows same curve but in linear scale Figure 4.24 DRCLS spectrum at region between Schottky and ohmic contact before neutron irradiation showing 2.2 ev yellow band (YB), blue band (BB), and 3.45 ev near band edge emission (NBE) peaks. Inset shows the schematic of GaN Schottky diode that we use in this dissertation. A slightly increase of YB and a dramatic increase of BB accompanied with the disappearance of NBE phonon replica indicate the GaN crystal quality degrades due to n/cm 2 neutron irradiation. Dashed lines are guides to the eye Figure 4.25 DRCLS results of (a) YB/NBE ratio and (b) BB/NBE ratio variation with different neutron fluence. A decrease of YB/NBE and BB/NBE ratio may indicate the xxv

27 improvement of the crystal quality. However, the increase of the YB/NBE and BB/NBE ratio with further increasing fluence implies that irradiation-induced defect may deteriorate device properties at high neutron dosage Figure 4.26 SPS spectra at drain side gate edge region (a) before, and after (b) fast (c) fast + thermal neutron irradiation. E C 0.6 ev and E V ev defects form due to neutron irradiation compare with SPS spectra before stress. E C 0.6 ev can be correlated with BB luminescence in DRCLS results. Fast and thermal neutron seems interact with defect differently. Energy levels in blue, red, and black color are defect exist before irradiation, induced by neutron irradiation, and complementary transitions Figure 4.27 The evolution of defect densities with (a) fast, and (b) fast + thermal neutron irradiation. DRCLS and t-sps confirm that low dosage of neutron irradiation may improve the GaN crystal quality. Thermal neutron seems to suppress defects induced by fast neutron irradiation Figure 4.28 Ideality factor (n) as a function of temperature for Schottky diode irradiated with different neutron fluence. n is between 1 and 2 for temperature above 100 K and increase dramatically with temperature below 80 K. Inset shows measured ideality factor with the prediction of TFE model. The characteristic energy E 00 varies from 8 28 mev with 2 mev/step xxvi

28 Figure 4.29 Plot of k B T/q vs. nk B T/q for devices with various neutron irradiated fluence. Various dashed lines refer to the prediction of thermionic emission (TE, n =1), generation-recombination (G-R, n = 2), thermionic field emission (TFE, E 00 ), and field emission (FE, E 00 ) theories Figure 4.30 Modified Richardson plot (ln(i STE /A e T 2 ) vs. 1/nk B T ) for devices (a) before neutron irradiation and irradiated with (b) n/cm 2 fast, (c) n/cm 2 fast, (d) n/cm 2 fast + thermal, and (e) n/cm 2 fast + thermal neutrons. The Schottky barrier height ( ) and Richardson constant (A * ) are extracted from curves at temperature range dominated by TE transport Figure 4.31 Sheet resistance (R sheet ), contact resistance (R contact ), and series resistance (R series ) as a function of (a) fast and (b) fast + thermal neutron irradiation fluence Figure 4.32 The correlation between YB and BB luminescence and defects which are found by SPS technique. New defect transitions appear at 1.4 ev and 0.6 ev due to fast and fast + thermal irradiation Figure 4.33 XPS spectra of GaN (a) before and (b) after thermal neutron irradiation. The core level spectra show Ge 2p 1/2 peak at bounding energy around ev suggests some Ga has been transmute into Ge due to thermal neutron. SEM images of 40 nm (c) Ni/GaN and (d) Ti/GaN after thermal neutron irradiations. The melting area shown in (c) xxvii

29 and pinwheel and bubbles shown in (d) imply that localized temperature may be high enough for atoms inter-diffusions to occur, which is detrimental for device performance Figure 6.1 AlGaN/GaN electric field measured by KPFM technique. The device is biased when doing KPFM scan. Black dashed line shows where gate and drain supposed to be. It is obvious that high electric filed regions appear at region close to the gate Figure 6.2 (a) SEM, (b) AFM, and (c) KPFM of U-series device after FIB cutting. (d) SEM image of M-series device after FIB milling. By using low kev gallium beam, the damage at the cut facet may be reduced xxviii

30 Chapter 1 Introduction 1.1 Motivation Wide band gap III-nitride material (III-N) systems such as aluminum nitride (AlN), gallium nitride (GaN), and indium nitride (InN) have been subject of great interest as a practical cure-alls for applications that demand optical, high power, high speed, temperature-insensitive, and radiation tolerant devices. When alloyed with their ternary and quaternary alloys, the materials system with direct band gap ranges from ev making group III-N particularly advantageous for the fabrication of optoelectronic devices such as light emitting diodes (LED), lasers and solar blind ultraviolet (UV) photo detectors. In additions to optoelectronic applications, aluminum gallium nitride/gallium nitride (AlGaN/GaN) based transistors enables technology for future high power and high speed electrical applications such as broad band wireless networks, electric hybrid vehicles, and defense radar systems due to the capability of handling high breakdown field, high saturation velocity and high current density. The other emerging field for GaN-based devices are aerospace-based and homeland security applications such as outer space satellite systems and neutron detectors because bulk AlGaN and GaN is apparently 1

31 Figure 1.1 New devices using GaN-based semiconductor devices. From Ref. [1] 2

32 more radiation hard than Si and GaAs. Figure 1 shows some potential applications of GaN-based devices. Although GaN-based devices, especially AlGaN/GaN high electron mobility transistors (HEMTs), play an important role in high power applications, high current density in the channel region leads to significant Joule heating that degrades device performance by decreasing carrier mobility, [2-6] thermal conductivity, and thermodynamic stability. AlGaN/GaN device temperatures have been measured previously by visible or ultra violet (UV) micro-raman methods [7-12] and scanning thermal microscopy (SThM). [13] Both experimental [7, 8, 10-12] and simulation [3, 8, 11, 12] results show that hot spots and dislocations form near the gate on the drain side. Semiconductor dislocations [14] and defect states [15] may also degrade mobility, heat distribution and device characteristics. Besides temperature, high voltage stress due to high power operation which may also become another important factor contribute to device degradation. High electric fields may induce point defects or structural damage at device surface/subsurface regions that degrade reliability [16-30]. Notwithstanding extensive studies by many groups, the physical mechanisms underlying GaN HEMT reliability remain elusive, in large part due to the interactions between high electric fields, mechanical stress, and temperature inside the device structure. These effects combine with piezoelectric strain and occur in highly localized regions of the transistor so that device failure is hard to predict. Techniques that are able to measure temperature, fieldinduced stress, and defects inside operating AlGaN/GaN HEMTs in a nanometer scale are urgently needed. 3

33 Given that many satellite systems are required to operate in the high radiation environment of space, a clear understanding of radiation effects is essential for successfully integrating GaN-based system into platforms in Earth s orbit or outer space. The Earth s magnetosphere is bombarded by energetic charged particles such as protons, electrons. These charged particles, named cosmic radiation, interact with the Earth s atmosphere continuously and generate secondary cosmic ray particles such as neutrons that can be detected at surface. Critical integrated circuits exposed to these high energy particles may be damaged and the whole system will malfunction. Another application of GaN-based devices is neutron detection for nuclear reactors or homeland security due to neutrons play an important role in many nuclear reactions such as fission of uranium-235 or plutonium-239 which people use in both nuclear power plant and nuclear weapons. Although extensive research has been done on irradiation effects of GaN devices by high energy photon [31-34], proton [35-43], and electron [44-48], there is only scarce information on GaN devices irradiated by neutron. Researcher found that deep levels forms in GaN after neutron irradiation due to recoiling damage. The cascades of collisions will produce disordered regions and deep level traps which can be detected by deep level transient spectroscopy with electrical injection (DLTS) or optical injection (ODLTS) [49-51], thermally stimulated current (TSC) spectroscopy [52-53], photoluminescence (PL) [33, 54], and cathodoluminescence (CL) techniques [49]. Sometimes GaN becomes more insulating due to carrier removal after fast neutron irradiation [50]. Additionally, thermal neutron irradiation may actually doped GaN based on the conversion of Gallium atoms to Germanium [52, 55-56]. However, the lack of 4

34 proper understating of failure mechanisms after neutron irradiation may hinder both commercial and military applications of GaN-based devices. A systematic investigation of GaN device behavior: (1) irradiated with different irradiation dosage, (2) irradiated with different types of neutron, and (3) the interaction between radiation defects and dopants and impurities and their effects on interfaces, is highly needed. 1.2 Objectives The first goal is to measure the temperature, field-induced stress, and defects distribution inside operating AlGaN/GaN HEMT and their variation with different stress conditions in order to understand the physical mechanisms that degrade these leading communications devices. That information can help identify designs that prevent the damage caused by these mechanisms in order to improve the device reliability. My dissertation involves the use of depth-resolved cathodoluminescence spectroscopy (DRCLS), atomic force microscopy (AFM), Kelvin probe force microscopy (KPFM), regular and time-resolved surface photovoltage spectroscopy (SPS and t-sps) to investigate the distributions of temperature, stress, and defect generation inside state-ofthe-art AlGaN/GaN-based HEMTs on the nanometer scale. The second goal is to investigate the effects of GaN Schottky diodes under different dosage and types of neutron irradiation. My dissertation involves the use of temperature dependent current-voltage (I-V-T) measurement, DRCLS, SPS, t-sps, and x-ray 5

35 photoemission spectroscopy (XPS) to investigate the damage created by neutron irradiation and their effect on the current transport mechanisms and material properties. 1.3 Research outline An introduction of GaN material properties, carrier transport mechanisms of Schottky barrier, and several AlGaN/GaN HEMT degradation mechanism are addressed in chapter 2. In chapter 2 we also explain the nuclear interaction between GaN and neutrons. Chapter 3 describes basic theories and experimental setups of techniques which we use to investigate AlGaN/GaN HEMTs performance change with respect to on- and off-state stress and the electrical and material properties of GaN Schottky diodes changes with fast and/or thermal neutron irradiation. Chapters 4 suggest major degradation mechanisms which affect AlGaN/GaN reliability the most and the effect of thermal and fast neutron irradiation on GaN and metal contacts. Chapter 5 summarizes all the results and chapter 6 lists future work that need to be done are listed in chapter 6. 6

36 Chapter 2 Background 2.1 Material properties of GaN-related materials III-nitride semiconductors system is promising for many applications including LEDs, lasers, full color display, future power distribution system, cell phone base station, and electric vehicles. The lack of the crystal symmetry, the properties of different plane, and the strong bonding between Al/Ga and N allows several unique properties appears in the III-nitride families such as two dimensional electron gas (2DEG) formation at AlGaN/GaN interface without intentionally doping. In order to utilize the properties of III-nitride, the knowledge of material properties such as crystal structure and polarization inside the crystal need to be addressed Crystal structure Figure 2.1 shows a typical GaN or AlN crystal in wurtzite structure. Each of the two neighboring atoms forms a hexagonal close pack (HCP) type package. Due to the non- 7

37 Figure 2.1 Hexagonal wurtzite structure of GaN and AlN crystal. From Ref. [57]. Figure 2.2 Schematic representation of c-plane (polar), a- and m-plane (non-polar), and γ- plane (semi-polar). From Ref. [58]. 8

38 centrosymmetric configuration, materials with wurtzite structure can have properties such as piezoelectricity. However, for optical devices, the existence of piezoelectric field, such as device grown on c-plane, may cause quantum confine stark effect (QCSE) which not only decrease the efficiency but also cause emission photon energy shift [59]. To overcome this problem, optical devices were grown on some non-polar or semi-polar facet. Figure 2.1 and 2.2 show a-, m-, and γ- planes in III-nitride crystal which exhibit none or medium polarization field. In my thesis, we will focus on the device grown on c-plane substrate due to advantage the 2DEG formation induced by strong spontaneous and piezoelectric field, which are crucial feature for nitride heterostructure devices Polarization and 2DEG formation The polarization filed in AlGaN and GaN material can be separated in two parts: spontaneous polarization P SP in equilibrium lattice and the strain-induced polarization P PE due to the lattice mismatch between AlGaN and GaN. Assume the linear interpolation of physical properties between GaN and AlN, P SP of Al x Ga 1-x N different aluminum concentration can be expressed as [60-61]: ( ) (2.1) with unit C/m 2. Piezoelectric polarization P PE can be calculated as [60-61] ( ) (2.2), (2.3) 9

39 where a 0 and c 0 are the unstrained lattice parameters, is the strain tensor of c-axis, and are the in-plane strain tensors. e 33 and e 31 are piezoelectric coefficients. The relationship between the lattice constant of a and c can be shown as ( ) (2.4) where C 13 and C 33 are elastic constants which are ( ) and ( ) with different aluminum concentration x and unit GPa. Inputting Eq. (2.3) and Eq. (2.4) into Eq. (2.2) and the P PE in the c-direction can be determined by ( ) (2.5) The orientation of the P SP and P PE is defined assuming that the positive direction is from metal atom to the nearest nitrogen atom along c-axis. The polarization charge sheet density σ cause by the sum of P SP and P PE difference between top and bottom material can be expressed as ( ) ( ) [ ( ) ( )] [ ( ) ( )] [ ( ) ( )] [ ( ) ( )] (2.6) Assuming a thin, strained AlGaN layer grown on a thick, relaxed GaN layer, Eq. (2.6) can also be rewritten as [ ( ) ( )] [ ( )] [ ( ) ( )] ( ) (2.7) 10

40 Figure 2.3 Directions of spontaneous polarization P SP and piezoelectric polarization P PE of AlGaN/GaN heterostructures with Ga- and N- terminations. AlGaN and GaN layers are either strained or relaxed. The +σ indicates the difference of P SP and P PE in AlGaN and GaN layer induce positive charge at interface and free electrons tends to compensate and form 2DEG. From Ref. [61] 11

41 If the polarization induced sheet charge density is positive, free electrons will tend to compensate the induced charge to keep the charge neutrality of the semiconductor, thus forming the 2DEG at AlGaN and GaN interface. The 2DEG has higher mobility than the bulk electron mobility and the 2DEG sheet carrier density is in general between and cm -2 [62] Reliability issues of AlGaN/GaN HEMTs Throughout the development of III-nitride, substantial effort has spent to identify physical mechanisms responsible for device degradation. In an accelerated life test of GaN-based HEMTs, the mean time to failure (MTTF) can be expressed as the general reliability equation [25]: ( ) ( ) ( ) (2.8) (2.9) where C is a constant, and AF(F) is an acceleration factor related to current(i), voltage(v),and temperature (T). E a is the activation energy of the degradation mechanism, k B is Boltzman s constant and T is the channel temperature. Recent report shows MTTF value of 10 7 hours at a junction temperature of 150 C for device operates at 40 V [63]. Besides lifetime, the more important issue is to understand mechanisms of electrical degradation or even predict device failure. This is very crucial for future GaNbased HEMTs applications. The analysis of failure mechanisms of GaN HEMTs could start from investigating the most common degradation mechanisms. 12

42 Temperature High current density in the channel region leads to significant Joule heating that degrades device performance by decreasing carrier mobility [2-6] thermal conductivity, and thermodynamic stability. AlGaN/GaN device temperatures have been measured previously by visible or ultra violet (UV) micro-raman methods [8-12, 64], nematic liquid crystal technique [66], and scanning thermal microscopy (SThM) [13]. Since the longitudinal optical (LO) phonon peak shifts with temperature, Raman spectroscopy is useful to extract temperature distribution of AlGaN/GaN HEMTs. Another method is photoluminescence (PL) since band gap is a function of temperature [65]. Due to the polarity changes from anisotropic to isotropic caused by heat, nematic liquid crystal can be used to detect hot region of AlGaN/GaN HEMTs. The other method is SThM. Atomic force microscope (AFM) with Pd tip can measure the temperature distribution across the surface. Most of the results show hot spots that appear close to the drain side gate edge Hot electron degradation Device aging due to hot electron is a well-known failure mechanism of GaAs-based device. Due to high drain bias and sub-micron gate geometries, extremely high electric 13

43 (a) (b) Figure 2.4 (a) EL intensity distribution of an operating AlGaN/GaN HEMT. (b) EL spectrum of the same device. An equivalent temperature = 1700 K was extracted by fitting Maxwellian distribution. From Ref. [20] Figure 2.5 Polarization dependent EL spectra from an operating AlGaN/GaN HEMT. The intensity of EL parallel component is greater that the perpendicular one. From Ref. [67] 14

44 field can be researched. One way to investigate hot electron effect is through electroluminescence (EL) since the ratio of the EL intensity to drain current is a function of 1/(V DS -V DSAT ) where V DS is the drain source voltage [20]. Generally speaking, this EL spectrum does not have very obvious features. Also, there is no e/h recombination corresponding to the GaN or AlGaN band gap. Many authors attribute EL of AlGaN/GaN HEMTs to intraband transitions of highly energetic electrons that acquire kinetic energy to accelerate/decelerate in the high field region of the channel. In figure 2.5, Nakao et al. found that polarized EL intensity parallel to drain current is higher than that perpendicular to the current [67]. This is consistent with this rapid acceleration/deceleration. Buoya et al. found that this EL intensity is stronger at higher temperature. Their EL spectra show clear peak features indicating that EL signals in operating AlGaN/GaN HEMTs are due to intraband transitions (due to charge center scattering) or to sub-band gap defects [68] Defects and traps Figure 2.6 and 2.7 show how surface defects affect the device performance [16, 18]. The existence of negative charges at surface caused by surface trap may deplete the 2DEG channel and lead to an extension of the gate depletion region. In other words, an additional negative charge forms at the negatively biased metal gate, and the magnitude of the induced negative bias is controlled by the total amount of the trap charge at the extrinsic drain region. This additional virtual gate will decrease drain current. 15

45 (a) (b) Figure 2.6 Schematic diagram of an AlGaN/GaN HEMT 2DEG depletion caused by negative charge accumulates at surface and the equivalent circuit. (b) Band diagram of AlGaN/GaN heterostructure show before and after the negatively charged virtual gate forms. From Ref. [16] 16

46 (a) (b) Figure 2.7 Source-drain current v.s. bias of AlGaN/GaN HEMTs: (a) with increasing virtual gate length X Q and (c) with increasing trap charge density N Q. From Ref. [18] 17

47 According to the simulation results shown in Figure 2.7, the increase of virtual gate length (X Q ) and trap charge density (N Q ) will reduce drain current. Results shown here indicate that surface traps can affect the device output performance dramatically. Device simulations show that defects in different regions produce different device output characteristics [68]. Defects in the surface layer or AlGaN layer only decrease drain current slightly. Defects in the buffer layer produce larger drain current decreases. With all traps are included, the decreases are even larger. So defects generated inside the AlGaN/GaN HEMT can cause major device problems [69] Electro-chemical reactions Very high electric fields with high power operation will also cause morphological defects such as cracks and pits to form at the drain-side gate edge [24, 25]. Figure 2.8(a) and (b) show transmission electron microscope (TEM) results of the formation of a pit under drain bias V DS = 40 V and drain current I D = 250 ma/mm. Each correlates with a different I D drop (19% and 58%). Nano-scaled energy dispersive X-ray (EDX) results show higher Si, O and C concentration in the defect area. The passivation layer flowed into the cavity and oxidation occurred indicates that high temperature may involve in this failure [24]. Smith et al. [25] suggest an electro-chemical reaction may happen at gate edge drain side area. An electrical path formed due to the shift of the polarization charge creates an electrical path for gate current. Figure 2.9(a) shows this theoretical interpretation where this current interacts with oxygen contaminant to react with the 18

48 (a) (b) (c) Figure 2.8 Transmission electron microscope (TEM) images of defects formed in the drain side after reliability test where drain current I D drop (a) 19% and (b) 58%. (c) TEM image of device showed a large defect and nano energy dispersive X-ray (EDX) results measured at point A and B. A higher O, C, and Si concentration can be observed. From Ref. [24] 19

49 (a) (b) Figure 2.9 (a) Theoretical representation of the formation of lattice disruption that the gate current may assist the surface contamination or oxygen residues interacts with AlGaN crystal. (b) TEM image of degraded AlGaN/GaN HEMTs showing lattice disruption region. From Ref. [25] 20

50 crystal along the strain gradients and results in a lattice disruption region shown in Figure 2.9(b) Inverse piezoelectric effect Not only does I D drop, but also gate leakage current (I G-off ) increases after stress. Figure 2.10 (a) and (b) show an increase of I G-off and decrease of I Dmax with High power state, Off-state, and V DS = 0 state stress after passing a critical voltage [70]. Since there is only very small current at Off-state and V DS = 0 state stress, another degradation mechanism such as the inverse piezoelectric effect must be present. Even without any applied voltage, there are already stresses present inside AlGaN & GaN crystals. Generally speaking, AlGaN on GaN is under tensile strain. During device operation, high applied voltages induce large mechanical stresses inside the crystal, especially in the gate edge area. Figure 2.11 (a) and (b) show theoretical and experimental results and they both show that a very high stress occurs close to the drain-side gate edge [22, 71]. The intrinsic strain already present may exceed the critical elastic energy and crystallographic defects may form. Stressing beyond the critical voltage creates traps that become a path way for electrons to flow from the gate to the 2DEG channel. 21

51 (a) (c) (b) Figure 2.10 Changes in (a) normalize I Dmax and (b) gate leakage current I G-OFF with V DS = 0 state (no current flow, drain and source site stress simultaneously), Off state (high V DS, low I D ), and high-power state (high V DS, high I D ) [70]. (c) Pictures show GaN HEMT under electrical stress with high V DG and the conceptual I G degradation mechanism [21]. The tensile stress induced by stress voltage may break the crystal and induce a pathway for electrons flow from gate to 2DEG. 22

52 (a) (b) Figure 2.11 (a) Elastic energy density in AlGaN/GaN HEMT under Off-state stress [23]. (b) Strain ε zz and stress σ xx in GaN determined from E 2 phonon shift and the averaged simulated E z -filed component at V SD = 40 V. Inset shows the average strain from the experimental data vs. applied voltage [71]. A concentration of elastic energy (or stress) in the AlGaN crystal close to the gate edge drain side area can be observed. The location of the highest elastic energy indicates the place where the crystal damages first. 23

53 Diffusion Due to high power operation, large amount of current and field induced stress may be generated inside AlGaN/GaN HEMTs. Thus, the temperature inside the device my reach more than 150 C and the diffusion process may occur. As illustrated in figure 2.12, Kuball et al. show that the diffusion constant will increase two orders of magnitude with temperature rise from 22 C to 150 C [27]. Furthermore, the high field induced stress may also assist this diffusion process by either create pits by breaking AlGaN crystal because energy in general is the lowest at dislocations, or accelerate the impurities or contaminations diffuse into metal/semiconductor interface such as oxygen or carbon. The formation of this high trap density area will decrease I Dmax and increase I G-Off which are similar to what we discuss before Stability of Schottky and Ohmic contacts Contacts are important for AlGaN/GaN HEMTs since they are directly related to device performance. Chou et al. [72] found that device under 48 hours temperature step stress cycle at V DS = 10 V and I D = 500 ma/mm, with a junction temperature around 390 C, show no obvious I-V characteristics can be found on Schottky contact (Pt/Au) and no interdiffusion of ohmic metal (Ti/Al/Pt/Au) can be detected. Other researchers [73] report that an increase of the Schottky barrier height after lifetime tests caused by an interfacial layer formed between the Ni/Au Schottky metallization. This will cause a 24

54 (a) (b) (c) Figure 2.12 Schematic of AlGaN/GaN HEMT degradation caused by diffusion: (a) diffusion along dislocation, (b) diffusion enhanced by inverse piezoelectric effect, (c) diffusion along the pit crated by field-induced stress. From Ref. [27] 25

55 shift in pinch-off voltage. The other researchers [74] show good stability of ohmic contact (Ti/Al/Ni/Au) but an increase of Schottky barrier (Mo/Au) height after 2000 hours, 340 C thermal test. However, gate sinking can also be observed in T-gate devices operating in the dry nitrogen ambient [75]. 2.2 Metal-Semiconductor contacts The detection of very high and low dosage of neutron flux is crucial for power plant and national border security. High displacement energy (~ 20 ev) and low intrinsic carrier concentration at high temperature render III-nitride material more suitable for these applications than commonly used semiconductor for detection purpose such as Si and GaAs [46, 76-77]. Furthermore, wide band gap should result in detectors with lower leakage current than an equivalent Si detector Schottky barrier diode One of the simple ways to produce a detector is through a Schottky barrier diode which can be achieved by depositing metal with appropriate work function on a given semiconductor surface. When metal and n-type semiconductor contact with each other, the charge will flow from metal to semiconductor. As shown in Figure 2.13(a), the Fermi 26

56 levels of the metal and semiconductor will eventually line up indicating that the equilibrium is established. The Schottky barrier created at metal/semiconductor interface can be roughly explained by the work function difference between the two assume no interface charges are involved. The work function is defined as the difference between vacuum level and the Fermi level. The quantity is shown as for metal. is the electron affinity of the semiconductor which is the difference between Fermi level and vacuum level in semiconductor. is defined as ( ). According to Figure 2.13, the Schottky barrier height: For n-type semiconductor: ( ) (2.10) For p-type semiconductor: ( ) (2.11) In metal/n-type semiconductor contact, if we apply a positive bias to the metal respect to the semiconductor, this situation is defined as forward bias. Under forward bias condition with amount of, the metal to semiconductor barrier is reduced to q ( ) and the electron can more easily flow from semiconductor to metal. If we apply a negative bias to the metal respect to the semiconductor, this situation is defined as reverse bias. Under reverse bias condition with amount of, the metal to semiconductor barrier is enlarged to ( ). For p-type semiconductor, the forward (reverse) bias situation is defined as a negative (positive) bias applied to the metal respect to semiconductor. The band diagram changes for both biased are similar to n-type counterparts. It should be noted here that we assume an idealized case that remains constant with different 27

57 (a) (b) Figure 2.13 (a) A metal/n-type semiconductor before (left) and after (right) contact with each other. (b) A metal/p-type semiconductor before (left) and after (right) contact each other. The alignment of the Fermi levels between metal and semiconductor indicate equilibrium condition has been reached. From Ref. [78] 28

58 amount of forward and reverse bias. The detailed electronic band diagram changes with respect to bias voltage of metal on n-type of p-type semiconductor are illustrated in Figure Current transport processes The current transport in metal-semiconductor is mainly due to majority carriers. When bias is applied, the carriers in the metal or semiconductor may gain enough kinetic energy to overcome the barrier, tunneling through the barrier, or governed by the combination of the two. Figure 2.15 shows four major current conduction processes (electrons) under forward bias. These four processes are: (1) electron transport from semiconductor to metal by surmounting the potential barrier at metal/semiconductor interface. This process is likely to happen at moderate temperature (~ 300 K) with device doping density N D around cm -3 and depletion region is relatively wide. This process is called thermionic emission (TE), (2) for moderate doping (~ cm -3 < N D < ~10 18 cm -3 ) semiconductor, the barrier is not thin enough for direct quantum-mechanical tunneling of electrons. This process requires electrons first gain some kinetic energy from bias then tunnel through the barrier with electron energy equal to E m where the barrier is thin enough to allow quantummechanical tunneling. This process is called thermionic field emission (TFE), (3) electrons cross the barrier through direct quantum-mechanical tunneling. This process in general occurs at heavily doped (N D > cm -3 ) semiconductor. Due to narrow depletion 29

59 (a) (b) (c) Figure 2.14 Energy band diagram of metal on n-and p-type semiconductors under (a) no bias, (b) forward bias, and (c) reverse bias. From Ref. [78] 30

60 (1) qv bi E m (2) qv n qφ Bn (3) (4) E C EFM E FB qv F E V Figure 2.15 Energy band diagram and carrier transport process of a Schottky barrier under forward bias: (1) thermionic emission, (2) thermionic field emission, (3) field emission, and (4) recombination in the space charge region. 31

61 region and very thin barrier thickness, cold and cool electrons may be able to tunnel from semiconductor to metal. This process is called field emission (FE). It should be noted here that this process may be responsible for most ohmic contacts in the absence of a good match between metal and semiconductor work function, (4) generationrecombination in the space charge region which is similar to the p-n junction case Thermionic Emission Theory The current density from semiconductor to the metal can be expressed as [79]: (2.12) assume where is Boltzmann s constant and T is the temperature, is the carrier velocity, and the current flow does not affect the equilibrium. The electron density is given by [79]: ( ) ( ) ( ) ( ) (2.13) where N(E) is the density of state, F(E) is the distribution function, h is Planck s constant, and m * is the effect mass. Assume all the energy of electrons in the conduction band is kinetic energy, then eq. (2.13) can be also rewritten as: ( ) ( ) [ ( ) ] ( ) [ ] [ ]( ) (2.14) combining eq. (2.13) and eq. (2.14), one can get [79]: 32

62 ( ) [ ( ) [ ] [ ] [ ] (2.15) where ( ) is the effective Richardson constant for thermionic emission. m 0 is the free electron mass. For GaN with electron effective mass equal to 0.222m 0, the theoretical Richardson contact is found to be A K -2 cm -2 [80]. The current flow from metal to semiconductor can be obtained from eq. (2.15) with V = 0: ( ) (2.16) The total current will be [79]: [ ] { [ ] } { [ ] } (2.17) [ ] (2.18) where is the area of the Schottky diode. Eq. (2.18) can also be expressed as: [ ] (2.19) ( ) (2.20) can be extracted from temperature dependent current-voltage (I-V-T) measurement. By plotting vs. ( ), one can obtained barrier height from the slope and Richardson constant from y-intercept. 33

63 Thermionic Field Emission Theory In TFE transport process, the electrons tunnel from semiconductor to metal with an energy E m above the conduction band gaining from the applied bias. The current density from semiconductor to metal through TFE at intermediate temperature range can be expressed as [78, 81]: [ ] [ ( )] (2.21) The energy is chosen to satisfy (2.22) and b m, c m and f m are Taylor expansion coefficients which are defined as: ( ) ( ) (2.23) [ ] (2.24) ( ( )) ( ) (2.25) The energy E 00 is given by: (2.26) where is Plank s constant, is the doping density, is the semiconductor dielectric constant, is effective mas, and is Boltzmann s constant. From eq. (2.22) and (2.24) E m can be evaluated as: ( ) ( ( )) (2.27) 34

64 It can be observed that the current should be dominated by the exponential term in eq. (2.21), thus by combining eq. (2.22) eq. (2.24): ( ) (2.28) ( ) (2.29) then input eq. (2.25) and eq. (2.28) into eq. (2.21), and neglecting the error function term, eq. (2.21) can be expressed as: ( ) ( ) [ ( ) ] [ ] [ ] (2.30) if we consider the electrons from metal to semiconductor which can be obtained from eq. (2.21) with V = 0, then combine with eq. (2.30) we can obtain the total current of a Schottky diode under forward bias through TFE: ( [ ] ) (2.31) ( ) (2.32) For the current transport of a Schottky diode under reverse bias, the current-voltage relationship can be expressed as [78]: ( ( )) ( ) [ ] [ ] [ ] (2.33) ( ) 35 (2.34)

65 where A is the Richardson constant of metal. If we include the current from semiconductor to metal, then eq. (2.33) will be [78]: ( [ ] ) (2.35) ( ) (2.36) A unique property of forward and reverse bias condition is [78]: (2.37) Field Emission Theory The current tunneling though the barrier can be expressed as [78, 81]: ( )( ) (2.38) ( )( ) (2.39) where ( ) is the transmission coefficient, and for low temperature and high doping levels, ( ) is roughly equal to [ ]. The total current density for the forward current through FE transport is given by [81]: { [ ( )]} { [ ( )]} [ ] [ ] [ ] (2.40) 36

66 When, TE is the dominant transport process. When, FE is the dominant transport process. When, TFE is the main mechanism which is a combination of TE and FE Recombination in the space charge region The recombination rate of excess electrons and holes in the space charge region is given by Shockley-Read-Hall recombination theory which is [82-83]: ( ) ( ) ( ) (2.41) where n and p are electron and hole concentration, n and p are constants related to trap energy, C n and C p are constants related to electron and hole capture rate and N t is the trap density. Eq. (2.41) can be rewritten as: ( ) ( ) (2.42) where and, assume, the recombination current can be evaluated as [82-83]: [ ] [ ] (2.43) Neutron irradiation effect on GaN-based detectors 37

67 Radiation damage is a detrimental alteration of detector operational properties especially at high levels. The severity of radiation damage depends on the energy and types of incident particles. Neutrons can be roughly classed as fast (kinetic energy ~ 1 MeV) and thermal (kinetic energy ~ ev) neutrons. These may produce several types of effects within semiconductor devices: (1) Displacement damage: Neutron with high enough kinetic energy may generate disorder region due to cascades of high-energy recoil atoms. This will cause the formation of lattice disorder regions and produce less ordered structure and thus affect the properties of GaN-based devices such as HEMTs or detectors. The lattice disorder region will pin the Fermi level at around E C ev [84]. Besides lattice disorder region, the displacement of atoms sometimes creates interstitials, vacancies, and structural defects in the lattice and can migrate through the semiconductor crystal lattice and form complex point defects or defect clusters [33, 49-54, 85-86]. (2) Transmutation: Neutron with low kinetic energy may interact with atoms and form new elements which is a useful method for the controlled homogenously impurity doping by nuclear reaction due to strong penetration of neutrons [51, 55-56, 87]. When GaN is bombard with thermal neutron, the Ga and N will form isotopes such 69 Ga, 71 Ga, 14 N, and 15 N, and follow with the decay to stable isotope: 69 Ga(n,γ), 70 Ga(T 1/2 = min) 70 Ge ; 71 Ga(n,γ)72Ga(T 1/2 = 1.41 hour) 72 Ge ; 15 N (n, γ) 16 N(7.13s) 16 O. Details of GaN transmutation due to thermal neutron are illustrated in figure 2.16 and figure

68 Figure 2.16 Gallium and Nitrogen atoms transmute to Germanium and Oxygen due to thermal neutron. σ is thermal neutron capture cross-section and T 1/2 is the half life time. 39

69 Reaction Q-value (MeV) Gamma ray (MeV) Recoil atom energy (MeV) Atom displaced by recoil atom Ga( n, ) Ga (65%) MeV Total displacement: 25,624/ion; Total vacancies: 24,778/ion; Replacement collisions:847/ion Ga( n, ) Ga (96%) MeV Total displacement: 25,041/ion; Total vacancies: 24,212/ion; Replacement collisions:829/ion N( n, ) N (50%) MeV Total displacement: 172/ion; Total vacancies: 166/ion; Replacement collisions:5/ion N( n, ) N Damage could be ignored due to the small cross-section Figure 2.17 Energy released by Gallium transmutes to germanium and nitrogen transmute to oxygen due to thermal neutron interaction. Q is the energy released by nuclear reaction. 40

70 Chapter 3 Experimental Techniques and Setups 3.1 Depth resolved cathodoluminescence spectroscopy (DRCLS) Cathodoluminescence (CL) is analogous to photoluminescence (PL) in that electrons vs. photons incident upon the sample excite electron hole pairs at designed depths, and then the electron-excited electron hole pairs recombine at either band gap or defect levels inside characterized material and emit photons with different energy. These radiative recombination transitions may include above band gap, band to band, dopant to valence band, conduction band to acceptor, donor to acceptor, and transitions include defect levels. Due to the different origins of the electron-hole pairs (electron-excited of CL vs. photon-excited of PL), CL have several advantages upon PL technique: (1) the ability to more efficiently excite wider or indirect band gap semiconductor, insulator, and sometimes even above band gap transition than PL, (2) good spatial resolution on the scale of nanometers. This resolution can be achieved by an electron beam focused by several magnetic lens, and this technique is already been applied to some electron microscope such as scanning electron microscope (SEM). The spatial resolution of PL is determined by the laser spot size and is usually in a range of μm due to the optical 41

71 Figure 3.1 An example of electron beam incidents upon AlGaN/GaN heterostructure with electron probing energy E B varies from 1 to 10 kev with 90 incident angle. The depth indicated by peak electron-hole pair generation show that with DRCLS technique one can investigate properties from surface to interface then inside the semiconductor bulk volume. 42

72 diffraction limit, (3) the penetration depth of the incident electron beam is tunable. By tuning the electron beam energy from hundreds of ev to several kev, one can characterize the surface, interface, and bulk properties based on the variation of the electron-hole pair generation depth. This depth distribution can be simulated by a Monte Carlo simulation program such as CASINO [88]. Figure 3.1 shows the Monte Carlo simulation results of an AlGaN/GaN heterostructure. By changing the electron beam probing energy E B, one can easily probe the surface, interface, and bulk properties of a given semiconductor with multiple layers. The DRCLS experiment in this dissertation was achieved by a JEOL JAMP-7800F ultrahigh vacuum (UHV) scanning electron microscope (SEM) with nm-scale resolution excited electron-hole pair recombination and cathodoluminescence with electron beam energies E B ranging from 1 to 22 kev. We used layer-dependent Monte Carlo simulations [88] to select E B that probe the GaN and AlGaN/GaN heterstructure. An Oxford MonoCL grating monochromator and Hamamatsu photomultiplier tube (PMT) collected optical emission signals from samples at low (12 K and 80 K) and room temperature with photon energy ranging from ev. The spatial resolution of our DRCLS experiment setup is between nm. Several experimental SEM components can be also found in figure

73 Monochromator Electron beam Signal PMT X-Y-Z stage 2 nd electron detector Mirror Sample holder Figure 3.2 Schematic of DRCLS measurement setups using JEOL JAMP-7800F SEM and optical signal collection systems. (Bottom) A close look of electron beam, half parabolic mirror, and sample positions located inside UHV SEM chamber. The stage position can be changed through a X-Y-Z manipulator. The CL signal from electron beam is gathered by mirror, reflected by the optical path, through monochromator, and analyzed by photo multiplier tube (PMT). 44

74 3.2 Atomic force microscopy and Kelvin probe force microscopy Atomic force microscopy (AFM) is a very high resolution scanning probe microscopy operating by measuring the forces between a probe and the sample. The instrument consists of a cantilever with a sharp tip mounted on its end. When the tip approaches to the sample surface, the van der Waals force between the tip and sample causes a deflection of the cantilever. In general, this deflection is measured using a laser spot reflecting from the cantilever top surface and detected by a position sensitive photodetector. Generally speaking, AFM can be operated in three different modes: contact mode, noncontact mode, and tapping mode. We operate only in non-contact mode. In contact mode operation, the topography mapping is achieved by the tip contact the surface though scanning process. In noncontact mode, the cantilever is oscillated at a frequency. The van der Waals force and other long range force acts to decrease this frequency on the cantilever. A feedback loop from the controller will try to maintain the frequency and amplitude that original applied to the cantilever by adjust the averaged distance between tip and the sample surface. Measuring the tip-to-sample distance at each x and y generates the topography of the sample. Generally speaking, operating under noncontact mode will reduce the resolution but generate less damage than operating under contact mode. In tapping mode operation, the cantilever oscillates at an amplitude around nm. When the tip approaches the surface, a relatively abrupt change occurs due to a strong repulsive force when tip touches the surface and is sensed by the piezoelectric actuator. The actuator will control the height of the cantilever and 45

75 maintain the cantilever oscillation amplitude. The topography is produced by imaging the force of this intermittent contact. Detailed explanation of the AFM theory can be found elsewhere [89]. The schematic of the AFM cantilever, tip, laser, and detector can be found in figure 3.3(a) [90]. Figure 3.3(b) illustrates the basics of the Kelvin probe force microscopy (KPFM). This technique is accomplished by measuring the work function difference between metal and semiconductor by modulating a capacitance probe. Due to the work function difference between metal and semiconductor and the oscillation of the cantilever, a frequency-dependent displacement current will be generated. The difference between metal (tip of AFM) and semiconductor (sample) is termed contact potential difference (CPD). When an external DC bias, equal and opposite to the CPD, is applied to the capacitor, the Fermi levels on metal and semiconductor will back to the isolated case which metal and semiconductor are far away from each other. This will cause the displacement current to become zero so that the surface potential can be measured, i.e. The CPD can be found by determining the external bias that cancels this CPD so that no displacement current can be detected [91-92]. The AFM/KPFM experiment in this dissertation was achieved by Park XE-70 system with the Enhanced EFM module. During KPFM surface potential mapping, a noncontact mode allows us successfully maintains the tip-sample distance of only a few nanometers allowing for measurements that do not disrupt the surface or its features. Details of experimental setups can also be found in figure

76 Figure 3.3 (a) Schematic of an AFM system including cantilever, tip, laser, and position sensitive photodetector. (b) Metal and semiconductor with work function and with compensation voltage V CPD that aligns the metal and semiconductor vacuum level and. From Ref. [90]. 47

77 Lock-in amplifier AFM/KPFM control console XE-70 AFM/KPFM PC Vibration isolating table Figure 3.4 A Park XE-70 AFM/KPFM system. 48

78 3.3 Surface photovoltage spectroscopy and time-resolved surface photovoltage spectroscopy Surface photovoltage spectroscopy (SPS) and time-resolved surface photovoltage spectroscopy (t-sps) are very useful techniques that are widely used of measuring surface or near surface material properties such as electronic structure, surface band bending, gap states, and defect density [91-96]. The basic working principle of SPS technique is measuring the surface potential variation with respect to the incident monochromatic light. Figure 3.5 (a) illustrate the setup of an AFM/KPFM/SPS system. In such setup, one can easily measure topography, surface potential, and defect levels at from AFM, KPFM, and SPS at the same time. Figure 3.5 (b) shows a typical SPS spectrum from an n-type semiconductor with two distinct defect levels E 1 and E 2 within the band gap E g. Figure 3.5 (c) shows the newly promoted electrons to conduction band, due to incident light energy hν = E 1 populates from defect level with energy E 1 below conduction band, reduce the n-type band banding at surface and increase the CPD which can be measured by KPFM. This phenomenon can be observed as an upward turn in Figure 3.5 (b). In another way, figure 3 (d) shows the newly promoted electrons form valence band, due to incident light energy hν = E 2 populates electrons to defects with energy E 2 above valence band, increase the n-type band banding at surface and decrease the CPD. This phenomenon can be observed as a downward turn in figure 3(b). The time resolved SPS (t-sps) technique can provide more than energy level and band bending 49

79 Figure 3.5 (a) Schematic of KPFM and SPS system. The monochromatic incident light is from a white light source through a monochromator and optical fiber. (b) SPS spectrum from an n-type semiconductor with two defect levels E C - E 1 and E V + E 2 with band gap E g. (c) Electronic band diagram for an optical transition from defect level to conduction band with incident light energy equal to E 1. (d) Electronic band diagram for an optical transition from valence band to defect level with incident light energy equal to E 2. 50

80 information. t-sps measures individual defects at selected incident photon energies and yields information about surface defect densities and photoionization cross sections. The model which is used to calculate the surface defect state densities is list below [93, 96]: ( (3.1) ) where is the defect density, is the Boltzmann constant, T is temperature, is the dielectric constant, is the permittivity, is the bulk doping density. The definitions of,,,, and are shown in figure 3.6. The SPS and t-sps experiment in this dissertation was achieved by Park XE-70 system with optical add-ons. The selected monochromatic light was generated by a tungsten white light source, through an Oriel 260i monochromator, f number matcher, and a fiber, then shine under the tip. All the data acquirement was controlled by a computer. A Detailed experimental setup can is shown in figure AlGaN/GaN HEMT electrical measurement and current-voltagetemperature experimental setups In order to measure temperature distribution inside operating AlGaN/GaN HEMTs, the sample was first mounted on a TO-05 can and three wire connections between pins and source (S), gate (G), and drain (D) contact pads was made by wire bonding technique. The sample was then mounted on an electron beam induced current (EBIC) stage and a 51

81 Figure 3.6 An example of time resolved SPS (t-sps) measurement result showing CPD varies with light turns on and off. An rising and decaying CPD which can be used to extract,,,, and for surface defect density calculation. 52

82 White light source Monochromator (a) F # matcher Optical fiber (b) Optical fiber Optical microscope Cantilever Sample holder X-Y-Z stage Figure 3.7 Park XE-70 AFM/KPFM system with optical add-ons for SPS and t-sps measurement. (b) Enlarged view shows positions of optical fiber, sample holder and cantilever. 53

83 copper heat sink in order to enhance thermal conducting. The schematic is shown in figure 3.8 (a). For stress measurements, the sample was mounted on another type of EBIC stage which is shown in figure 3.8 (b). The temperature and stress measurement was done at room temperature (300 K) and low temperature (12 K), respectively. The sample biasing for both temperature and stress measurement was achieved by HP4145B semiconductor parameter analyzer. The current-temperature-voltage (I-V-T) measurement was achieved by EBARA closed cycle helium cooled cryogenic system (Figure 3.9 (a) and (b)) with temperature range from K controlled by Lakeshore 330 temperature controller (Figure 3.9(c)). The sample was first mounted on the copper board with three metal bonding pads. Then the Schottky and ohmic contacts were connected to the pads through wire bonding technique and the copper board was then mounted on the cold finger (Figure 3.9 (d)). The current-voltage (I-V) measurement was done by Keithley 2400 source meter with wire connected to output ports. 3.5 Device details and experimental procedures Two sets of GaN HEMT devices termed U-series, grown on sapphire with gate length ~ 1.25 µm and gate width ~ 70 µm, and M-series, grown on a SiC substrate with gate length ~ 0.25 μm and gate width ~ 40 μm. U-series heterostructures consist of an unintentionally doped (UID) GaN layer, a 40 nm Al 0.22 Ga 0.78 N and 10 nm cm -3 Sidoped graded Al x Ga 1-x N layer (0 < x < 0.22) sandwiched by a 0.7 nm AlN interfacial 54

84 (a) Copper heat sink (b) EBIC stage TO-05 can Sample Figure 3.8 Two types of EBIC stages that use for temperature and stress measurement. 55

85 (a) EBARA cryogenic cold head (b) (c) Keithley 2400 Temperature controller Compressor Electrical Cold finger feed through Sample (d) Copper board Figure 3.9 and (b) EBARA closed cycle helium cooled cryogenic system with Keithley 2400 source meter. (c) Lakeshore 330 temperature controller. (d) Wire-bonded AlGaN/GaN HEMTs device mounted on the cold finger for I-V-T measurement. 56

86 layer and a 250 nm GaN cap. The GaN cap layer was etched to form drain (D), gate (G), source (S) contacts. M-series devices consists of a thick GaN buffer layer, followed by 1 nm AlN, 16 nm Al 0.28 Ga 0.72 N, 3 nm GaN cap and Si 3 N 4 passivation layer. Thus the etched U-series samples are nearly equivalent in structure to the M-01 heterostructure. In order to measure defect generation and cross-sectional temperature distribution on a nanometer scale, we use a JEOL JAMP-7800F ultra high vacuum (UHV) scanning electron microscopy (SEM) with beam energy 1 kev < E B < 22 kev and temperature 12 K < T< 300 K. An E B = 5 kev beam excites the 2DEG in the U-series, whereas the thicker overlayer of the M-series requires 10 kev. Monte Carlo [88] simulations to select an E B that probes the 2DEG region and to estimate penetration depths for mapping 2- dimensional, cross-sectional temperature distributions. A Park XE-70 AFM/KPFM with ~20 nm lateral resolution, tungsten lamp and monochromator provided simultaneous topography, potential maps, and SPV spectra. An Agilent 4145B analyzer enabled device operation and DC on- (high I D, low V DS ) and off-state (low I D, high V DS ) stress during measurements. On-state device conditions were: V DS = 10 V, V GS = - 2 V, I D = 0.47 A/mm (M-01) for 12 hours or V DS = 6 V, V GS = 0 V, I D = 0.75 A/mm (U-01) for 11 minutes. The longer stress time for the device on SiC is to ensure similar degree of degradation as in the device on sapphire. Off-state stress conditions were: (1) V DS = V, increasing in 2 V steps in order to insure a transition through the critical voltage point [70], V GS = - 6 V, I D = A/mm. Each step period was ~ 1 minute. A key failure indicator is gate leakage current I G-Off, taken at V DS = 0.5 V, V GS = -6 V. (2) V DS = 0 25 V, V GS = -8 V, I D = 1*10-4 A/mm, 30 minutes/step. In this stress condition, we 57

87 monitor the gate leakage current I G-off at V DS = 0.1 V, V GS = - 8 V. Experiments were performed on unpackaged, on-wafer devices and at room temperature. Off-state stress measurements permit us to factor out temperature effects since heating is negligible with the minimal currents involved. The Schottky diodes for neutron irradiation experiments were grown by metal-organic chemical vapor deposition (MOCVD). The thin film structures were grown on sapphire substrate followed by a 2 μm GaN layer doped with cm -3 silicon. Schottky contacts consist of 30 nm Ni and 400 nm Au. Ohmic contacts were made by 20 nm Ti, 150 nm Al, 40 nm Ni, and 50 nm Au. The ohimc contacts were given at 30 second, 850 C rapid thermal anneal. Samples were irradiation by OSU research reactor (OSURR) which is a 500 kw, natural convention, pool-type reactor. The neutron fluence are: (1) fast neutron: n/cm 2, n/cm 2, and n/cm 2, (2) fast + thermal neutron: n/cm 2, n/cm 2, n/cm 2 with fast : thermal neutron ratio = 1:4 and 1:2 for n/cm 2 case. A cadmium (Cd) cap was used in order to filter out thermal neutrons. Details of the irradiation experimental parameters are also shown in table 3.1. To obtain defect related optical spectra due to neutron irradiation, a JEOL JAMP- 7800F ultrahigh vacuum (UHV) scanning electron microscope (SEM) with nanometerscaled resolution excited electron-hole pair recombination and cathodoluminescence with 2, 5, and 10 kev electron beam energies E B at 80 K. Surface photovoltage spectrum (SPS) and time-resolved SPS (t-sps) results were achieved by a Park XE-70 AFM/KPFM system with selected monochromatic light illuminated under the cantilever tip. This light was obtained from a tungsten white light source through an Oriel 260i 58

88 monochromator and an optical fiber. A Keithley 2400 source meter and EBARA closed cycle helium cryogenic system were used to measure the current-voltage (I-V) characteristics of Schottky barrier diodes under neutron irradiation with temperature at 10K, and from 20 K to 300 K with 20 K/step which are controlled by Lakeshore 330 auto tuning temperature controller. 59

89 Table 3.1 Experimental details of neutron irradiation experiment of GaN Schottky diodes. The thermal neutron capture cross section of cadmium is large and thus can be used to filter out thermal neutrons during neutron irradiation. 60

90 Chapter 4 Result and Discussion 4.1 Reliability issues investigation of AlGaN/GaN HEMTs The study of device degradation at the nanoscale involves both on-state and off-state stress conditions which give rise to temperature and field-induced stress effects, respectively. First we consider the effects of temperature under on-state conditions. We then examine the effects of stress under off-state conditions. Furthermore, we also demonstrate the capabilities of measuring defects in devices. The combination of this knowledge allows us to investigate the reliability issues of the state-of-the-art AlGaN/GaN HEMTs AlGaN/GaN HEMT operating temperature characterization Figure 4.1 shows the U-08 temperature distribution between gate and drain, termed the extrinsic drain region, with beam current I B ~ 0.5 na. Peak electron-hole pair excitation depth U 0 derived from Monte Carlo simulation is ~125 nm for E B = 5 kev, just 75 nm below the 2DEG channel region. DRCLS provides local temperatures since the GaN 61

91 Figure 4.1 Temperature distribution of U-08 operated at V DS = 3.5 V and V GS = -2 V. P1 is nearest the gate while P10 is furthest away. Inset photo shows SEM image of sourcedrain region measured. From Ref. [99] 62

92 near-band-edge (NBE) emission energy varies as ( ) ( ) ( ) for temperature T and band gap E g [97]. Under low power (source-drain voltage V DS = 3.5 V, gate-source voltage V GS = -2 V, and drain-source current I DS = 367 ma/mm), figure 4.1 shows temperature increasing from drain to gate along 10 equallyspaced points along a perpendicular to and between gate and drain, consistent with micro- Raman results [7-10] and other techniques but with dramatically higher spatial and depth resolution. Localized electron beam heating produces no measureable NBE temperature rise at room temperature. At I B = 0.5 na and E B = 5 kev, induced beam currents are ~ 0.2 μa, negligible compared to our device currents. Piezoelectric strain can shift NBE energies by 24 mev/gpa [98] = 40 ºC/GPa, but V SD = 40 V produces strains of ~0.4 GPa at the drain [71] that extrapolate linearly to < 1 ºC for V SD = 6 V here. Figure 4.2 shows a temperature distribution with 100 nm spatial resolution for virgin device M-01 operating at V DS = 6 V, V GS = -1 V, I D = 1 A/mm with beam current I B ~ 4.5 na. Regions between gate-source and gate-drain are termed extrinsic source and extrinsic drain, respectively. Peak electron-hole excitation depth U 0 derived from a Monte Carlo simulation for E B = 10 kev is 655 nm below the Si 3 N 4 passivation surface, i.e., close to 2DEG channel region. The red dashed lines show the edge of the gate. The temperature increases from the drain side to the gate overhang edge at the extrinsic drain region. The temperature distribution at the extrinsic source region exhibits a similar trend. Furthermore, the temperature is higher within the extrinsic drain region compared with the extrinsic source region. These results agree with results extracted by a micro-raman method [64]. Note that the electron beam-induced current under our measurement 63

93 Figure 4.2 (a) SEM image of virgin sample M-01 source-gate-drain areas across which temperatures were measured. Dashed red rectangles mark extrinsic drain and source areas. (b) Temperature distribution across extrinsic drain and source area at V DS = 6 V, V GS = -1V, I D = 1 A/mm with E B = 10 kev. Dashed black lines are guides to the eye. Temperature increases monotonically from drain to gate overhang edge. From Ref. [100] 64

94 conditions (I B = 4.5 na, E B = 10 kev) is ~ 0.04 A/mm, negligible compared with device currents so that localized electron beam heating is not significant. Likewise, piezoelectric strain can also shift the NBE emission peak [71], but its effect is very small at V DS = 6 V and does not affect our temperature measurement. Neither this current nor this strain has a measurable effect on the NBE emission energy. DRCLS measurements also provided in-situ measurements of temperature versus depth across the extrinsic source and drain, in effect yielding cross-sectional temperature maps during device operation, e.g., in Figure 4.3. U 0 from the top of the Si 3 N 4 passivation layer with E B between 10 and 22 kev in 2 kev step are 655 nm, 770 nm, 920 nm, 1070 nm, 1250 nm, 1460 nm, and 1730 nm, respectively, while we used E B = 8 kev to probe the GaN cap surface. The corresponding positions are also marked in Figure 4.3 (a) and (b). The lateral and depth resolution for 8 kev is ~ 100 nm, proportionally larger for higher beam energies. According to the contour map, the temperature distribution varies from 300 K to 370 K (27-97 ºC). Monte Carlo simulations show that the rate of CL excitation peaks at characteristic depths, increasing with increasing incident beam voltage E B [101]. Since electron-hole recombination at shallower depths contributes to the CL spectra, we subtract these contributions with spectra at slightly lower E B to achieve even finer depth resolution of the temperature distribution inside the device. Fig. 4.3 (b) shows a magnified view of the region close to the edge of the gate overhang). Here, multiple hot spots are evident inside the operating GaN HEMT. One is near the gate overhang edge, where electric field and current density are high. The other two are 260 nm and 800 nm below the extrinsic drain side 2DEG channel. Pomeroy et al. [71] proposed that hot spots 65

95 (a) Gate Figure 4.3 DRCLS cross-sectional temperature distribution of sample M-01: (a) Whole device, (b) Device drain side close to gate overhang. V DS = 6 V, V GS = -1 V, I D = 1 A/mm, and 8 < E B < 22 kev. Multiple hot spots are apparent. From Ref. [100] 66 (Continued)

96 Figure 4.3 Continued Figure 4.3 (b) 67

97 will form around defects in GaN HEMTs. We suggest that point defects or dislocations may impede thermal transfer, leading to hot spots AlGaN/GaN HEMT field induced stress characterization DRCLS measures electric-field-induced stress in Off state (no heating) from the NBE peak, which shifts to higher energy by 26 mev/gpa [102]. Figure 4.4 (a) and (b) shows NBE energy increases up to 7 mev corresponding to a 0.27 GPa compressive stress at the edge of the gate on the drain side within the 2DEG channel under Off-state stress. For a 28 V critical V DG, the corresponding stress is ~ 0.25 GPa compared with this transistor before stress. Figure 4.5 shows the field-induced-stress distribution caused by the external bias from source to drain with various bias conditions of a virgin device at U-01. With E B = 5 kev, we are able to probe the region close to the 2DEG region. In order to prevent any selfheating effects, gate voltage (V GS ) was kept at - 6 V to limit the channel current to < 5*10-6 A/mm. DRCLS measures the field-induced stress at the AlGaN/GaN interface under off-state stress condition from NBE peaks, which shift by 26 mev/gpa [102] with no self-heating. DRCLS shows a 7.6 mev blueshift corresponding to a 0.29 GPa compressive stress in the GaN layer at the gate edge drain side with V DS = 26 V. The line scan results also show that the field-induced-stress increases faster at gate edge drain side regions than at regions close to the drain metal contact. del Alamo et al. proposed an inverse piezoelectric degradation mechanism to explain the device performance 68

98 Figure 4.4 (a) Corresponding NBE shift with applied Off-sate stress at drain side gate edge area. (b) External stress caused by applied voltage under OFF-state stress. From Ref. [104]. 69

99 Figure 4.5 Field-induced-stress distribution of a virgin device caused by applied bias under off-state stress. S, G, D denote source, gate, and drain metal contact. Dashed lines are guides to the eye. A maximum ~ 0.29 GPa compressive stress increase is evident at the gate edge drain side area. Field-induced off-state stress increases the most at drainside gate edge region. From Ref. [100]. 70

100 deterioration due to high applied bias [21-22]. The high field can induce mechanical stress that exceeds the crystal s elastic energy, creating lattice defects especially at the gate edge drain side area. While field-induced stress appears on both sides of the gate, figure 4.5 shows that the external stress is the highest at the AlGaN/GaN interface close to the gate edge drain side area, so it is the likeliest place for defects to appear first. Regarding the effect of stress on our temperature measurements, figure 4.5 shows that the field-induced stress is negligible for the bias voltage used for our temperature distribution measurements (V DS = 6 V, V GS = - 1 V). The piezoelectric strain can shift NBE energies by 26 mev/gpa = 43.5 ºC/GPa. The field-induced stress distribution in this case should be close to the V DS = 0 V, V GS = - 6 V condition in figure 4.5 so that the temperatures extracted from DRCLS require little or no correction (0.03 GPa = ~ 1.3 ºC). However, the thermal strain acts to impact the measure temperature by DRCLS by ~ 6.5 ºC at 370 K, proportionally lower at lower temperature [103] AlGaN/GaN HEMT defect characterization Figure 4.6 (a) shows the SEM image of another U-08 GaN HEMT operated to failure (V DS = 3.5 V, V GS = 0 V, I DS = 633 ma/mm, 4 hours) in UHV. AFM revealed a nm deep, ~60 nm wide crater in the extrinsic drain region. A CL map of NBE (~ 3.42 ev at 12 K) emission in Figure 4.6 (b) shows: (i) NBE disappearance at the failure region and (ii) variations in NBE intensity within the extrinsic drain and source. Bright CL map areas exhibit 1000-fold higher NBE intensity than dark areas in the extrinsic drain versus 71

101 Figure 4.6 For U-08, (a) SEM image shows the Source (S), Gate (G) and Drain (D) region. (b) CL map of NBE emission (middle) and KPFM result (side) taken in the same (a) region. Red dashed circles show some of the same higher defect, higher potential region for CL and KPFM map, respectively, where devices degrade faster. From Ref. [99]. 72

102 uniform intensity of the virgin sample. NBE intensity decreases apparent in figure 4.6 (b) signify free carrier recombination at sub-band gap energies. KPFM maps in figure 4.6 (b) show non-uniform potential distribution. Dashed rectangles define the same extrinsic drain and source regions as in figure 4.6 (a), while an ellipse outlines the crater. Surface potentials vary by more than 100 mev within the unbiased extrinsic drain and source regions, with higher potentials in general corresponding to regions of reduced NBE emission (higher defect density). Figure 4.7 shows the DRCLS result at one of the lowest surface potential areas in the extrinsic drain region after OFF-state stress. Besides the 3.45 ev NBE, a 2.2 ev yellow band (YB) and ev blue band (BB) are evident. YB emission is often associated with Ga vacancies [105]. BB emission can be associated with surface or bulk defects [105]. Figure 4.6 and 4.7 demonstrate that DRCLS can not only measure defects in the AlGaN/GaN HEMT, but also maps the defect induced by on- or off-state stress which we will discuss in detail later. Combining with other techniques that we introduced before, we will be able to investigate AlGaN/GaN HEMTs reliability issues due to different operating conditions Reliability issues of AlGaN/GaN HEMTs due to on- and off-state stress Reliability is an important concern in GaN-based HEMTs due to the extreme current densities and strain under operation at high applied bias. Figure 4.8 (a) shows DC-IV and I G-Off before and after Off-state stress. I G-Off increases by 2.6 with V DG, rising sharply 73

103 Figure 4.7 DRCLS spectrum at one of lowest potential areas (inset) within extrinsic drain region after Off-state stress showing 2.2 ev YB, BB, and 3.45 ev NBE peaks. From Ref. [104]. 74

104 above a 28 V critical voltage after Off-state stress. However, the device under on-state stress shows I G-off decrease by 20%. The drain current I D also shows an opposite trend where I D decrease ~ 10% after off-state stress but varies ~ 2% after on-state stress. The observations of the electrical properties changes with respect to on- and off-state stress indicated device may degrade in different ways under different types of stress. Figure 4.9 shows AFM/KPFM results used to monitor unpassivated device characteristics with different stress conditions. With increasing Off-state stress, KPFM reveals micron-scale patches of lower potential and Fermi level (E F ). Figures 4.9 (a), (b) and (c) display the KPFM, atomic force microscope (AFM) and SEM images, respectively, of gate (G), drain (D), and source (S) in a full gate s upper and lower scanned parts of sample U-01. The KPFM images show a striking evolution of surface potential with increasing stress voltage V DS as low surface potential patches begin to grow. Initially, the lowest potential area ( upper half) extends along and on both sides of the gate. Between V DG = 24 and 32 V, the lowest potential area shifts to the gate foot on the drain side. As V DG increases further, this lowest potential area expands and extends from the gate edge across the extrinsic drain region. The surface potential at this extended area decreases from ~ 0.4 V to -1.6 V. AFM shows no surface morphology changes during these potential changes from the rectangular gate edge to the drain, unlike pits measured previously from HEMTs with a T-shaped gate etched away [106]. Offstate V DS > 32 V results in a broken gate, a crater forming at the extrinsic drain region and device failure (Figure 4.9 (c)). Although the surface potential changes dramatically as a result of high-voltage stress, no topography changes are observed before failure 75

105 Off-state On-state Figure 4.8 DC output characteristics and I G-OFF as a function of (a) Off-state stress V DS = V and V GS = - 6 V, and (b) on-state stress V DS = 6 V, V GS = 0 V, and I D = 0.75 A/mm. Inset show the DC-IV output characteristics before and after (a) off-state and (b) on-state stress. Green arrows indicate the point monitored by AFM/KPFM. An 2.6 increase of the I G-off and decrease of I D after off-state stress can be observed, faster after V cri = 28 V. I G-off and I D show an opposite trend after on-state stress. Changes in output characteristics are correlated directly with surface potential. From Ref. [104] 76

106 occurs and a crater appears. Figure 4.10 (a) and (b) show surface potential and topography changes before and after on-state stress. For on-state stress, similar surface potentials occur but without clear trend or expansion. There are no changes in topography can be observed during on-state stress. Koley et al. found that low surface potential areas form close to the drain side gate edge after stress, which they attribute to accumulation of negative charge from gate tunneling electrons under high bias stress [107]. Instead, we propose that the appearance of low surface potential patches is due to on- and off-state stress-induced defect formation. Comparing the potential and morphology results with electrical stress, one can link device failure with lower surface potential area. As stress voltage increases, device properties gradually degrade toward failure - faster for V DG above a ~28 V critical voltage. Although no AFM or SEM morphology changes are evident, surface potential changes dramatically with stress. Furthermore, failure occurs close to the region of lowest potential. Failure is defined here as a crater in the extrinsic drain region. Likewise, the lower the surface potential, the more probable failure occurs there. Therefore, surface potential appears to be a strong indicator for device failure: (i) low potential (e.g., decreasing by > 1.5 ev) patches indicate likely failure points, (ii) a patch with the lowest potential is likely the initial point of failure, and (iii) the lower the potential, the closer to device failure. DRCLS measurements within the extrinsic source and drain provide further evidence for stress-induced defect formation that correlates with device degradation. 77

107 Figure 4.9 (a) KPFM results which shows the evolution of surface potential under Offstate stress at upper and lower region of AlGaN/GaN HEMTs. AFM images show upper and lower scanning areas at V DG = 36 V. The SEM image in (b) shows the corresponding AFM/KPFM scanning area. The red dashed circles show regions where potentials change faster. (c) The SEM image indicates where upper scan area (rotated 45 clockwise) failure occurs with increasing Off-state stress. From Ref.[104] 78

108 Figure 4.10 (a) KPFM potential maps before and after 11 minute on-state stress. Surface potential varies less with on-state vs. off-state stress. AFM images show upper and lower scanning areas at time = 11 min. 79

109 The DRCLS spectrum in figure 4.11 (a) shows 2.2 ev yellow band (YB), ev blue band (BB), and 3.45 ev NBE emission from our M-01 GaN HEMT devices before semi on-state (V DS = 10 V, V GS = - 2 V, I D = 0.47 A/mm for 12 hours) stress. Post-stress DRCLS shows higher defect emissions [99]. YB emission is often associated with Ga vacancies where BB emission may be due to bulk or surface defects [105]. Figure 4.10 (b) shows a x increase in the gate current at reverse bias after on-state operation for 12 hours in room ambient. We used DRCLS to correlate defect emissions with degradation of the gate-edge extrinsic drain during device operation with two neighboring devices on the same die termed device under test (DUT) and reference (Ref). DUT device on sample M-01 was on-state stressed for 12 hours vs. the unstressed Ref device Labeled regions in figure 10 (c) are: #1 (extrinsic drain, DUT), #2 (under gate-drain side, DUT), #3 (under gate-source side, DUT), #4 (extrinsic drain, ref), and #5 (under gate, Ref). An SEM image of the defined regions appears in [100]. The YB/NBE ratios in regions #2 and 3 increase 4.3X and 2.6X, respectively. In contrast, figure 4.11 (c) inset shows much weaker correlations for BB/NBE ratios. YB defect emission exhibits much larger and systematic increases than defects characterized by BB emission, indicating a link between this defect and device degradation. These results demonstrate that stress during device operation generates defects. Furthermore, stress affects different defects preferentially. We used DRCLS and KPFM to correlate the defect emissions with potential changes further. Here U-08 was operated to failure (V DS = 3.5 V, V GS = 0 V, I D = 0.63 A/mm, 4 hours), i.e., crater formation, under UHV condition (2*10-9 torr). UHV avoids uncertainties due to heating in air. The SEM image [99] shows that failure occurs at a 80

110 Figure 4.11 (a) Representative pre-stress DRCLS spectra with YB, BB, and NBE peaks. (b) Gate current characteristics before (black) and after (red) stress of sample M-01. The increase of the gate leakage current indicates the degradation of gate Schottky contact after stress. (c) Position-dependent averaged YB/NBE ratio shows largest increase at region under gate-drain side after 12 hours, V DS = 10 V, V GS = - 2 V, I D = 0.47 A/mm stress. Averaged BB/NBE ratio shows a much weaker response to local stress. The horizontal dashed lines represent the reference points. From Ref. [100] 81

111 crater at the extrinsic drain region. Figure 4.12 (a) shows the DRCLS spectrum acquired at 12 K with E B = 5 kev in region (1) before stress. The two major peaks are due to GaN with phonon replica (~ 3.45 ev) and AlGaN (~ 4.08 ev). No defect-related peaks are evident. Before device operation, the NBE emission is distributed uniformly. After device operation to failure, the DRCLS map shown in the center part of figure 4.12 (c) with E B = 5 kev of GaN NBE (3.45 ev) intensity reveals major lateral variations, e.g., bright and darker regions (1) and (2), respectively, and the darkest region (3). NBE emission intensity decreases as defect density and alternate recombination pathways increase. The NBE and AlGaN peak intensities decrease X between regions (1) vs. (3), indicating that a high density of non-radiative defects accumulate around/inside the crater region. Straddling the DRCLS maps in figure 4.12 (c) are KPFM potential distributions at extrinsic drain and source regions, marked with dashed yellow rectangles, without any external bias. Here, surface potential varies by ±80 mv with higher potential regions aligning with reduced NBE emissions. These are outlined by red and black ellipses. We then used SPS to determine the sub-band gap defect position of sample U-08 operated to failure. Figure 4.12 (b) shows for all regions a clear surface photovoltage (SPV) change at ~3.4 ev (blue dash-dot line), assigned to the GaN band gap at room temperature. SPS spectra acquired from regions (1), (2) and (3) display an increasing slope change at ~1.2 ev (denoted by red dashed line) with proximity to the failure site. The sign of slope change at this energy corresponds to a defect level located 1.1 ev above the valence band [108]. This result agrees with our DRCLS mapping result that defect densities are lower in brighter areas, e.g., region 1. 82

112 Figure 4.12 (a) DRCLS results of sample U-08 taken at region (1) with E B = 5 kev at 12 K, (b) SPS maps taken from region (1) - (3), (c) CL map of NBE emission (middle) and KPFM map of potential (side). Dashed lines delineate extrinsic drain and source areas. Both red and black circles show similar higher defect, higher potential regions for CL and KPFM maps. SPS spectra reveal a defect 1.2 ev above the valence band that increases with DRCLS defect emission intensities linked to device degradation. From Ref. [100] 83

113 With device operation, defects (native point defects and/or dislocations) may increase locally, decreasing or eliminating NBE luminescence. Significantly, the 2.2 ev YL defect emission from DRCLS and the ~1.2 ev SPS feature of the same regions are near complements of the 3.45 ev GaN band gap, indicating a DRCLS optical transition from the conduction band 2.2 ev above. For sample U-08, the additional slope change at ~1.6 ev in region (3) may be related to subsurface features created by the crater formation. Figures 4.13 and 4.14 show the correlation between surface potential and defect emission of sample U-01 from two representative areas of same device along the width of the transistors before and after on- state (V DS = 6 V, V GS = 0 V, I D = 0.75 A/mm, 11 minutes) and off- state (V DS = V in 2V/step, V GS = - 6 V, I D = 5*10-6 A/mm) stress. Potential maps and defect intensity increases for off- and on-state stress in figures 4.13 and 4.14 show that decreasing surface potential correlates with increased YB and BB intensities. Each data point of defect emission in figures 4.13 and 4.14 represents an average of multiple spectra. For the off-state stressed device, areas 1, 2 and 3 are in the middle of extrinsic drain and region close to the drain side where surface potential is higher (~ V). Areas 4, 5 and 6 are at the gate edge drain side with V and ~ V potentials. Area 7 has the second lowest potential (~ V). Area 8 is at the crater edge where device failure occurs (failure edge area, FEA). For on-state stress, area (i) and (ii) are the middle of extrinsic drain (~ 0.06 V), area (iii) and (iv) are the gate edge drain side area with higher surface potential (~ V), and area (v), (vi) and (vii) are the gate edge drain side area with lower surface potential (~ V). Regions 4, 5 7, and 8 exhibit monotonically increasing YB and BB defect intensities with decreasing potential. 84

114 Areas 1, 2 and 3 exhibit negligible defect increases along with higher surface potential (~ 0.45 V), consistent with the defect versus potential correlation. Unstressed devices (R1 & R2) exhibit comparable defect emission to area 1 and 2. Figures 4.13 and 4.14 also show that on-state potential changes are less pronounced, and indicate correspondingly smaller defect increases. We observe no BB increases with on-state stress. Sample M-01 and U- 01 results show that device performance degrades during device operation (i.e., under onand off-state stress) due to defect formation and figure 4.11 shows that stress enhances defect emission preferentially. Within the same off and on-state stress duration (11 minutes), the device subjected to off-state stress exhibits larger variation in surface potential than with on-state stress. This is due to either the time of on-state stress is not long enough to generate defects, or the off-state stress has much more prominent effect on degrading device properties. We believe the voltage stress [21] is the primary force to cause device failure than current or time. The effect of stress on I G-off and I D-max is also described in [104]. KPFM maps for sample U-01 under Off-state stress show that decreasing surface potential correlates with increased YB and BB intensities. Figures 4.15 (a) and (b) show numbered KPFM patches of an Off-state stressed transistor and corresponding changes in DRCLS defect intensity for each patch, respectively. Each data point in figure 4.15 (b) represents an average of multiple spectra. With increasing Off-state stress and decreasing potential, regions 6, 8, 9, and 10 exhibit monotonically increasing defect intensity. Thus area 9 at the extrinsic drain region has the second lowest potential and the second highest 85

115 Figure 4.13 (a) KPFM maps of sample U-01 from two representative areas of the same device along the width of the transistors show surface potential distribution and averaged defect emission after off- state stress. Device layout is indicated in the figure: source (S), gate (G), and drain (D). (b) and (c) Averaged YB & BB /NBE ratio correspond to areas denoted in (a). In general, regions with lower potential correlate with higher YB or BB defect emission. The horizontal dashed lines represent the reference points. From Ref. [100]. 86

116 Figure 4.14 (a) KPFM maps of sample U-01 from two representative areas of the same device along the width of the transistors show surface potential distribution after on-state stress. Device layout is indicated in the figure: source (S), gate (G), and drain (D). (b) and (c) Averaged YB and BB defect emission correspond to areas denoted in (a). The horizontal dashed lines represent the reference points. From Ref. [100]. 87

117 Figure 4.15 (a) Off-state KPFM maps showing numbered upper and lower low potential regions and (b) corresponding surface potential and average YB/NBE intensity ratio increases with OFF-state stress. From color potential scale (red higher, blue lower), YB/NBE increases most at lowest potential regions 6, 8, 9, and 10. Higher potential patches display slower changes. 1 and 2 correspond to extrinsic drain and drain-side gate foot of an unstressed reference device. (c) Surface potential variation vs. V DG and corresponding self-consistent electrostatic defect density (smooth lines). From Ref. [104] 88

118 YB increase. Area 10 at the crater edge has the highest CL YB emission but a higher KPFM potential measured prior to crater formation. Areas with higher potentials and unstressed devices exhibit negligible defect increases, e.g., areas 1-5. The BB emission exhibits similar increases in Off-state in regions 9 and 10 as shown in Figure Figure 4.15 (c) shows the variation in surface potential with V DG. For low potential regions such as areas 9 and 10, there is a striking change in the rate of surface potential decrease above OFF-state stress V DG ~ V. The corresponding increase in gap state defect density with lower E F can be calculated using a self-consistent electrostatic model that depends on acceptor level E A and E F positions in the band gap and the E F -dependent density of charged acceptors σ - A [109]. Negatively-charged acceptors induce a dipole qδv = q 2 σ - A d/ε that shifts E F lower in the band gap, depending on the occupancy of charged acceptor sites σ - A = σ A0 /[1+exp[(E F -E A )/k B T]], charge separation d, dielectric permittivity ε, and total acceptor density σ A0. As defect density increases, E F moves closer to the gap state. Using deep level spectroscopy measured trap densities of high cm -2 for similar stressed HEMTs [110], we fit the potential variation to defect densities increasing by nearly two orders of magnitude with a charge separation of approximately 8 nm, corresponding to defects located at or below the AlGaN surface. As with gate leakage current, there are pronounced decreases in potential and increases in defect density above V DG ~25 V [70]. The correlation of stress and defect generation thresholds supports a model of defects forming between gate and drain by Off-state fieldinduced stress normal to the surface that, added to lattice-mismatch strain already present, exceeds the crystal s critical elastic energy density [21-22]. 89

119 Figure 4.16 shows a systematic correlation between surface potential and YB/NBE as well as BB/NBE ratios of individual spectra. We observed a clear trend of stronger defect emission with decreasing surface potential This agrees with previous simulation results showing how an increasing electrically-active acceptor defect density near midgap lowers the surface Fermi level [104]. These potential-defect density correlations mean that: (1) KPFM can map where defects are generated at or close to the device surface and (2) such maps provide a predictive tool for device failure by illustrating the region of lowest potential and highest defect density that corresponds to the initial point of failure. Furthermore, these defect features exhibit a strong increase above a critical voltage [104] similar to device results of refs. [21] and [22]. In general, an increase of I G- off and a decrease of I D-max can be observed after on- and off-state stress. Thus after offstate stress, figures 4.13 and 4.14 display 2.2 YB/NBE and 3.7 BB/NBE increases at the gate edge drain side. The increase is even more obvious at the FEA area, i.e., 4.7 YB/NBE and 10.4X BB/NBE ratio increases. According to the results that we already shown, the main degradation mechanism of AlGaN/GaN HEMTs is field induced stress. This stress induced by high voltage due to high power operation will generate defects and thus degrade electrical properties of the devices. However, there is scarce information about the spatial distribution of degradation related defects, especially with measurement results in a nanometer scale laterally and in depth. In order to investigate the evolution of failure with filed-induced stress in cross section, we use an Agilent 4145B analyzer not only applied DC off-state (V DS = 0 25 V, V GS = -8 V, I D = 1*10-4 A/mm, 30 minutes/step) stress, but also monitor 90

120 Figure 4.16 Combined off- and on-state surface potential vs. (a) YB/NBE and (b) BB/NBE ratio from individual point spectra. Dashed lines are guides to the eye. The ~ V surface potential corresponds to pre-stress potential, e.g. R1 and R2. Dashed circle denotes the FEA area where defect emissions are strongly perturbed by catastrophic lattice disruption. In general, YB/NBE and BB/NBE ratios increase linearly as surface potential decreases. From Ref. [100]. 91

121 the gate leakage current I G-off (I G at V DS = 0.1 V, V GS = - 8 V). Figure 4.17 shows the variation of surface potential with respect to stepped off-state stress V DG. The surface potential at drain side gate edge decreases gradually with increasing V DG, faster if exceeds critical voltage V crit ~ volt. Figure 4.17 inset shows I G-off as a function of V DG. Similar threshold behavior can also be observed in gate leakage current where I G-off exhibits a sharp rise at around the same V crit extracted by surface potential measurement. The 6.5 increase of I G-off accompanied with the decrease of the surface potential from 0.4 V to ~ -2.5 V indicates device performance degrades after off-state stress, possibly due to either the formation of defects or the increase of defect density [100, 104]. From layer dependent Monte Carlo simulations [88], we varied probe energy E B from 1 to 9 kev to measure semiconductor materials properties from different layers and interfaces. Additionally, we applied a technique terms differential DRCLS (D-DRCLS) which we subtracted the contribution of lower E B from higher E B for improving the depth resolution further. Figure 4.18 (a) shows the peak electron-hole excitation depth derived from D-DRCLS. By varying E B, we were able to investigate the material and defect properties of AlGaN, GaN layers, and their interfaces before and after off-state stress with nanometer scale depth resolution. Low temperature (T = 80 K) D-DRCLS results of an AlGaN/GaN HEMT from drain and source side of gate edge are shown in figure 4.18 (b). Besides the 3.48 ev GaN near band edge emission (NBE) and ~ 4.1 ev AlGaN peak, two weak ev yellow band (YB) and ev blue band (BB) luminescence can be observed. In general, YB luminescence is attributed to gallium vacancy (V Ga ) where the BB luminescence might be from oxygen- or structure- related 92

122 Figure 4.17 Surface potential evolution with applied stress voltage V DG = 0 33 V. The gate edge drain-side area exhibits a threshold behavior with increasing V DG. Note that the surface potential starts to decrease before V crit ~ 25 V. Dashed lines are guides to the eye. Inset shows I G-Off increases slowly with lower stress voltage, sharply rising if it exceeds Vcrit, and a 6.5x increase can be observed. The changing of surface potential and the degrading of electrical properties indicate the formation of defects during stress. 93

123 Figure 4.18 (a) D-DRCLS probing depth extracted by Monte Carlo simulation. (b) Low temperature D-DRCLS spectrum before stress with different probing depth. A 3.48 ev GaN NBE, a ~ 4.1 ev AlGaN, and two obscured defect related yellow and blue luminescence. The lack of the YB and BB indicate good material properties before stress. 94

124 defects in AlGaN or GaN layers [105, 111]. Figure 4.19 shows the D-DRCLS results at drain side gate edge after off-state (V DS = 25 V, V GS = -8 V) stress. Comparing the results before and after stress, we found: (1) A 3.95 ev peak appears in AlGaN layer. This peak may correspond to the red shift of AlGaN peak due to Franz-Keldysh effect caused by the inhomogeneity of the strain-induced piezoelectric field [112]. (2) A 4.15 ev peak can be resolved from the shoulder of AlGaN peak. The ~ 0.03 V blue shift of the peak compared to 1 kev spectrum may be attributed to localized crystal relaxation due to the reduction of tensile strain in AlGaN film by field-induced stress. (3) A new ~ 3.75 ev defect forms in AlGaN grading layer (g-algan)/algan interface and AlGaN layer. (4) A rigid 0.2 ev shift of the AlGaN peak with BB luminescence indicates defects responsible for BB close to the surface are from AlGaN layer. (5) A dramatically increase of the YB at very top surface. The Franz-Keldysh red shift accompanied with the relaxation of crystal, the formation of new defects, and the enhancement of BB and YB luminescence reveals that field-induced stress that exceeds the critical elastic energy of AlGaN crystal degrades the AlGaN crystal quality, and the region of defect formation is mainly close to the surface. Figure 4.20 (a) and (b) shows the YB and BB luminescence with respect to probing depth. The YB intensity after off-state stress at drain side gate edge is significant at the surface and is about an order magnitude higher compared to the result before stress. It then quickly fades away when we probe through AlGaN layer, and becomes comparable to the YB intensity from the same device before stress. It is interesting to note that the YB at source side gate edge also increases after off-state stress. 95

125 Figure 4.19 Low temperature D-DRCLS spectrum after stress. A splitting of the AlGaN emission, i.e. the Franz-Keldysh redshift and crystal relaxation blue shift, indicted the deterioration of the crystal quality due to stress. A 3.75 ev defect emission appears in the spectrum and can be precisely located in AlGaN layer because of nanometer scaled depth resolution of DRCLS. A rigid red shift of BB and AlGaN emission indicates the defect responsible for BB enhancement close to surface is in AlGaN. The overall results reveals field-induced stress cracks the AlGaN crystal and induces crystallographic defects. 96

126 Figure 4.20 The distribution of (a) YB and (b) BB in AlGaN/GaN HEMT before and after stress extracted by D-DRCLS technique. Dashed lines are guides to the eye. A 10 increase of YB at surface and 1.7 increase of BB in AlGaN layer can be observed. These results shows field-induced stress affects the crystal quality not only at surface, but also extends into devices. 97

127 The AlGaN BB intensity extracted by D-DRCLS shows a different trend. It starts to increase at g-algan/algan interface, peaks in AlGaN layer which is ~ 1.7 compared to same device before stress, decreases quickly when probed close to AlGaN/GaN interface, then become comparable to the BB intensity before stress. No BB increase at source side can be observed. This confirms that stress will create defects preferentially [100]. Figure 4.21 (a) and (b) illustrate the SPS results before and after off-state stress. For n- type material, the population (de-population) of electrons from defects at surface will increase (decrease) band bending, thus moving the Fermi-level (E F ) down (up), and eventually affect the contact potential. In other words, if the optical transition corresponds to a positive slope change in SPS spectrum, it is characteristic of an electron transition from an occupied defect state E C E T to the conduction band. On the other hand, if the optical transition corresponds to a negative slope change, it indicates a transition from the valence band to an unoccupied defect state located at E V + E T. Comparing figure 4.21 (a) and (b), we found two distinct defect levels, E C 1.35 ev and E V ev, that appear after off-state stress. These new forming defects are either close to surface or locate at AlGaN layer close to g-algan/gan interface. Figure 4.22 summarized the results extracted by our scanning probe techniques: (1) A threshold behavior was found in the evolution of surface potential at drain side gate edge region with applied Off-state stress (figure 4.17). I G-Off increases gradually with Off-state stress V DG, sharply rising above V DG = 21 V and increases by 6.5 with V DG = 33 V (Figure 4.17 inset). 98

128 Figure 4.21 SPS spectra at drain side gate edge region (a) before, and (b) after V DG = 33 V stress. E C 1.35 ev and E V ev defects form at drain side gate edge compare with SPS spectra before stress. The SPS spectra are similar before and after stress. These defects correlate with our D-DRCLS results showing YB BB, and 3.75 ev defects increasing after off-state stress. These defects are electrically-active because they not only change surface potential but also degrade electrical properties of AlGaN/GaN HEMTs. 99

129 Both KPFM and current-voltage (I-V) results show a slow degradation at low off-state stress and a dramatically deterioration if stress voltage exceed V crit ~ V. The field induced stress (~ GPa) [100] adds to build-in stress plus other degradation mechanisms [20, 25, 27] may deteriorate the AlGaN crystal thus increases I G-Off and reduces the surface potential. (2) New defects forms at drain side gate edge region of AlGaN/GaN HEMT after off-state stress. YB increases dramatically at surface whereas the BB and 3.75 ev defect emission peak close to g-algan/algan interface (Figure 4.19 and 4.20). A splitting of the AlGaN emission, which caused by inhomogeneity of the piezoelectric field and crystal relaxation, and the appearance of the BB and 3.75 ev deffect luminescence in AlGaN layer, indicates crystallographic defects form mainly in AlGaN layer and is caused by inverse piezoelectric effect [21] or field-induced diffusion [27]. These defects are electrically active because they move the E F position and have impact on device electrical properties. The amount of these defects can be estimated by KPFM results [100, 104] and two orders of magnitude increase of some defects after offstate stress is possible. (3) SPS spectra show new defects form within E g after V DG = 33 V Off-state stress. SPS shows that defect levels E C 1.35 ev, E V ev, and E C 1.7 ev, form at drain side gate edge compared with SPS spectra before stress. Our results are consistent with previous literature that a E C 0.6 ev [23, 26, 114, 117] can be observed after stress. Some of these defects can also be correlated with YB (E C 2.0 ev) and BB (E C 1.35 ev) in our D-DRCLS results (Figure 4.19 and 4.20) that YB and BB increase after off-state stress. Due to our better depth resolution, we believe that the field-induced of E C 1.35 ev and E V ev defect density may exceed cm -2 and cm -2 at 100

130 localized spacial region. (4) D-DRCLS reveals that field-induced defects distribute mainly within the top 30 nm of the device structure. This value may vary due to different device structure, targeted defect levels or aluminum composition. However, our value is close to previous published transmission electron microscope (TEM) [24] or AFM [106, 113] results. (5) Device performance starts to degrade with stress even before the V crit. This indicates that additional degradation mechanisms must be considered such as leakage current induced defects [25] Discussion Our results relate directly to the various models proposed for AlGaN/GaN HEMT degradation. High mechanical stress during device operation generates electrically-active defects [21 22, 28]. These defects reduce I D and increase I G-off, and become more evident if exceed critical voltage. Hot electron effects also contribute to device degradation [20, 30, 118]. Here, semi-on conditions generate the highest hot-electron rate and the maximum degradation. Defect diffusion can also cause devices to fail [23, 25, 27]. Charged surface traps can diffuse into the device along dislocations and degrade device performance by altering channel potentials. This process may also be enhanced by an inverse piezo-electric effect, temperature, and leakage current. Our observations of device after off-state stress in this paper provide strong support for the inverse piezoelectric effect: (1) The increase in stress with off-state operation and its localization at the drain-side gate as evidenced by figure 4.4 and 4.5. This shows that 101

131 Figure 4.22 (a) A summary of defect distribution measured by DRCLS. Dashed lines are guides to the eye. The results shows YB is significant at surface, quickly fades away when probing into the device whereas BB starts to increase at g-algan/algan interface, peaks at AlGaN layer, and decreases when approaching AlGaN/GaN interface ev defect emission is mainly in AlGaN layer, neither in g-algan nor in GaN layer. (b) The correlation between our scanning probe techniques and other techniques. Our results are comparable to previous published results but with nanometer scale depth resolution. 102

132 stress can preferentially exceed crystal elastic energy at certain areas. (2) The strong and localized changes in electric potential produced by off-state stress as evidenced by figures 4.9, 4.13, and 4.15 and the failure due to crater formation that occurs at the lowest potential induced by stress. (3) The formation and intensity increase of electrically-active defects in the areas exhibiting large potential decreases as shown by figures 4.13, 4.15, and (4) The increase in these defect emissions with proximity to the point of lattice disruption as illustrated by figures 4.13, 4.15, and (5) The systematic correlation between potential changes and defect intensities shown in fig that are consistent with surface band bending and increased free carrier recombination that are induced by charged acceptor states. In order to gain more insight, we need to estimate the build-in stress in our film and also the critical stress that need to bread the crystal. The build-in stress due to lattice match between Al 0.25 Ga 0.75 N and GaN layers can be estimated as below: Assuming AlGaN psudomorphically grows on GaN, σ xy = σ xz = σ yz =0, σ xx = σ yy, the film is under biaxial strain, the strain tensor can be expressed as [123]: [ ( ) ] (4.1) ( ) (4.2) where ε is the strain tensor, σ is the stress, C is the compliance tensor, a 0 is the unstrained lattice constant of GaN, a AlGaN and a GaN are lattice constant of AlGaN and GaN, respectively. Assume GaN layer is totally relaxed, the aluminum concentration of AlGaN layer is 0.25, and the parameters of AlGaN is determined from linear interpolations between GaN and AlN binary which can be found ref. [123], by using eq. 103

133 (1), one can found the stress in Al 0.25 Ga 0.75 N σ xx = 2.75 GPa with a AlGaN = Å, ε xx = , C 11 = GPa, C 12 = GPa, C 13 = GPa, C 33 = 397 GPa. The above calculation show that, the built-in stress due to Al 0.25 Ga 0.75 N/GaN lattice mismatch is 2.75 GPa. This is similar to other recent calculations [22]. We then next estimate the possible amount of stress that stored in the AlGaN with thickness close to critical thickness. The critical thickness can be extracted from parameters list in Ref. [22]: (4.3) (4.4) where W crit is the critical elastic energy, Eγ is Young s module, and h crit is critical thickness. Assuming a strained Al 0.3 Ga 0.7 N/GaN heterostructure, we can calculate the critical Al 0.3 Ga 0.7 N film thickness which is nm and nm with lower and higher bound of critical elastic energy W crit. For simplicity, assume a 30 nm Al 0.3 Ga 0.7 N film pesudomophically grow on relaxed GaN buffer layer, the build-in stress can be calculate from eq. (4.1) and (4.2) to be σ xx = 3.33 GPa with parameters C 11, C 12, C 13, C 33 and ε, from ref. [61] and [123]. The bias-induced stress of 0.3 GPa is well below the critical resolved sheer stress value of 3.33 GPa or higher [ ] needed to cause lattice disruption and defect formation. However, we calculate the built-in stress due to Al 0.25 Ga 0.75 N/GaN lattice mismatch to be 2.75 GPa, similar to other recent calculations [21]. Therefore, the bias-induced stress added to built-in stress may account for our observation of native point defects above the bias threshold reported here [22]. Other mechanisms that may also play a role include thermally-induced stress and impurity diffusion [27]. 104

134 Besides the measurement achieved at the AlGaN/GaN interface region, our D-DRCLS, KPFM, I-V, and SPS also shows: (1) Device performance starts to degrade with stress, even before the V crit. Additional degradation mechanism needs to be considered. (Figure 4.17 and 4.20) The degradation before V crit may be caused by gate leakage current induced defect. (2) The threshold behavior can be attributed to the field-induced-stress that exceeds the critical energy and breaks the crystal structure. E c 1.35 ev and E V ev defects might attribute to field induced defect. (Figure 4.20 and 4.21) (3) D-DRCLS reveals that a new 3.75 ev defect luminescence emerges and BB increases in the AlGaN layer after stress. A raise of the YB close to surface are also observed. These results are consistent with our previous results and other people s observation. (4) The surface potential evolution might be cause by either the increasing of the exiting native defect density or the newly appearing defect levels. (5) These defects are electrically-active because they not only move the surface E F lower in the AlGaN band gap but also have big impact on electrical properties. These results show a strong correlation between stress, surface potential, defects, and device degradation which happens mainly in AlGaN layer close to the device surface. Our observations of devices after on-state stress show that semi-on and high temperature conditions can produce hot electron and diffusion degradation from: (i) the localized potential changes in figure 4.9; (ii) the device failure without high stress apparent in figure 4.12; (iii) the defect increases over time shown in figure 4.11 and On-state stressing results also reveal that devices on SiC is more reliable compared to device grown on sapphire substrate with comparable operating power (M-01 vs. U-01). 105

135 Figure 4.9, 4.13, and 4.14 show both on-state and off-state operating conditions give rise to degradation that manifested itself by electronic changes the appearance of deep level defects and related changes in potential. However, for the realistic off-state and high current on-state operating conditions reported here, we find that off-state conditions induce stronger defect features associated with device failure. These observations provide a self-consistent picture for the device degradation reported in [21, 22]. 4.2 Neutron irradiation effects of GaN Schottky diodes Another factor that can degrade GaN device is nuclear and space radiation-particularly important for GaN since it is out looked for aerospace application. One are of high interest is the effect of neutrons and GaN devices. In order to understand the change of device and material properties due to neutron irradiation, GaN Schottky diodes have been irradiated with various neutron fluence : (1) fast neutron: n/cm 2, n/cm 2, and n/cm 2, (2) fast + thermal neutron: n/cm 2 (Fast = n/cm 2, Thermal = n/cm 2 ), n/cm 2 (Fast = n/cm 2, Thermal = n/cm 2 ), and n/cm 2 (Fast = n/cm 2, Thermal = n/cm 2 ). Figure 4.23 (a) and (b) show current-voltage (I-V) characteristics of GaN Schottky diode at room temperature (T = 300 K) with different types and fluencies of neutron irradiation. According to figure 4.23 (a), both forward and reverse I-V behavior shows negligible influence with low fluence ( n/cm 2 ) neutron irradiation. Higher fast neutron irradiation fluence causes I-V characteristic deterioration at forward bias, but not so much 106

136 at the reverse bias. A ~ 7 decrease of forward current at V = 3 V can be observed. However, with thermal neutron involved, the electrical properties start to degrade at lower fluence shown in figure 4.23 (b) compared with fast neutron case which is shown in figure 4.23 (a). Higher fluence causes a ~ 20 decrease of forward current at V = 3 V and a dramatic increase of the leakage current at reverse bias. Schottky diodes show ohmic behavior after fast ( n/cm 2 ) or fast + thermal ( n/cm 2 ) neutron irrdation. This indicates that GaN or metal/gan interfaces properties change with higher fluence of neutron irradiation. Thus, we investigate the effect of neutron irradiation on GaN or metal separately. The decrease of forward current and increase of the leakage current indicate device properties degrade due to neutron irradiation Neutron irradiation on GaN Figure 4.24 shows the low temperature (T = 80 K) DRCLS result at area between ohmic and Schottky contact before and after neutron irradiation. Besides the 3.45 ev near band edge emission (NBE), a 2.2 ev yellow band (YB) and ev blue band (BB) can be observed. YB emission is often associated with Ga vacancies [105]. BB emission can be associated with surface or bulk defects [105]. A slightly increase of YB, a dramatic increase of BB, and the disappearance of phonon replicas which are in general an indicator of crystal quality, imply that the GaN crystal quality degrades due to neutron irradiation. 107

137 Figure 4.23 Current-voltage (I-V) characteristics of Schottky diodes irradiated with (a) fast, and (b) fast + thermal neutron irradiation. Inset shows same curve but in linear scale. 108

138 Figure 4.25 (a) and (b) shows the YB/NBE and BB/NBE ratio with different neutron fluence. The YB/NBE and BB/NBE ratio for samples under both fast neutron and fast + thermal neutron irradiation first decrease with lower neutron fluence, then increase with increasing neutron fluence. In general, it is expected that the lower the defect/nbe ratio, the higher the crystal lattice properties. Such results impling that some irradiationinduced defect may not only enhance deep level defect emission (YB and BB), but also decrease the NBE. However, lower dosage of fast and/or fast + thermal neutrons may improve the semiconductor material properties by suppressing defects state in semiconductor. Another interesting observation is the YB/NBE and BB/NBE ratios are lower for sample irradiated with fast + thermal neutron than the fast irradiated one. As we mentioned before, for example, the fast neutron fluence of n/cm 2 (fast) and n/cm 2 (fast + thermal) are similar. The only difference is in the fast + thermal neutron irradiation experiment, additional n/cm 2 thermal neutrons are involved in the irradiation. These results indicate that fast neutron and thermal neutron interact with defects in GaN differently. Figure 4.26 (a) - (c) illustrate the SPS results before and after fast (fast neutron fluence = n/cm 2 ) and fast + thermal neutron (fast neutron fluence = n/cm 2, thermal neutron fluence = n/cm 2 ) For n-type material, the population (de-population) of electrons from defects at surface will increase (decrease) band bending, thus moving the Fermi-level (E F ) down (up), and eventually affect the contact potential. If the energy of the incident light is equal the energy of a trap E T below the conduction band E C, the work function of the material at surface will decrease and the contact potential difference (CPD) will increase due to the newly 109

139 Figure 4.24 DRCLS spectrum at region between Schottky and ohmic contact before neutron irradiation showing 2.2 ev yellow band (YB), blue band (BB), and 3.45 ev near band edge emission (NBE) peaks. Inset shows the schematic of GaN Schottky diode that we use in this dissertation. A slightly increase of YB and a dramatically increase of BB accompanied with the disappearance of NBE phonon replica indicate the GaN crystal quality degrades due to n/cm 2 neutron irradiation. Dashed lines are guides to the eye. 110

140 Figure 4.25 DRCLS results of (a) YB/NBE ratio and (b) BB/NBE ratio varies with different neutron fluence. A decrease of YB/NBE and BB/NBE ratio may indicate the improvement of the crystal quality. However, the increase of the YB/NBE and BB/NBE ratio with further increasing fluence implies that irradiation-induced defect may deteriorate device properties at high neutron dosage. 111

141 promoted electrons. Thus, an upward turn of the SPS results indicates a defect with energy E C - E T. On the other hand, if the energy of the incident light is equal the energy of a trap E T above the valence band E V, the work function at surface will increase and the CPD will decrease due to the newly promoted electrons into the defect energy level. Thus, a downward turn of the SPS results indicates a defect with energy E V + E T. According to SPS results, we not only found several defects and their complementary transitions (into and out ot the same defect level) before neutron irradiation, but also comparing from figure 4.21 (a) to (c), we found two newly formed defect levels, E C 0.6 ev and E V ev, that appear after neutron irradiation. These results are consistent with our DRCLS results since: (1) E C 0.6 ev can be correlated with BB luminescence in DRCLS spectra. (2) New defect formation implies a decrease of the crystal quality which is confirmed by the disappearance of the phonon replica of the GaN NBE. Another interesting fact should be noted here: to interact instead of creating new defects in GaN like fast neutrons, thermal neutron seems preferentially with existing defects. In order to get more insight of defect properties with respect to different neutron fluence, we use time-resolved SPS (t-sps) method to estimated defect density of individual defects with different neutron fluence detected by SPS. The density of each defect levels can be estimated as [93, 96]: ( (4.5) ) 112

142 where is the defect density, is the Boltzmann constant, T is temperature, is the dielectric constant, is the free space permittivity, is the bulk doping density, is the changing slope when light turns on, is the CPD before light turns on, is the saturated CPD with light on,, is the changing slope of CPD when light turns off. The detailed definitions of,,,, and can be found elsewhere [93, 96]. Figure 4.27 (a) and (b) show the evolution of defect density with increasing fast and fast + thermal neutron fluence. One can clear observe that under fast neutron irradiation, defects with energy levels E C 1.1 ev and E C 1.2 ev decrease and then increase with increasing neutron fluence. These two defect states can be correlated with YB luminescence in DRCLS results. The density of defects at E C 0.6 ev also increases with increasing fast neutron fluence which can be correlated with BB luminescence as we discussed above. Other defect levels such as E V ev and E C 1.65 ev show similar trends in SPS but didn t appear in our DRCLS results. However, with thermal neutrons involved, densities of defect levels (E C 0.6 ev, E V ev, E V ev, E C 1.65 ev) either decreased or remained at the same levels with increasing fast + thermal neutron irradiation fluence. Only E C 1.1 ev shows a similar trend with fast neutron irradiation case. Another interesting observation is that E V ev defect density dramatically increases with increasing thermal neutron fluence compared with fast neutron case, whose density stayed constant. Our t-sps results show that thermal neutron tends to suppress or heal some of the deep level defects created by fast neutron irradiation. The suppression of these defect levels is also confirmed by DRCLS due to the correlation between defect luminescence (BB and YB) and SPS measured defect levels 113

143 Figure 4.26 SPS spectra at drain side gate edge region (a) before, and after (b) fast (c) fast + thermal neutron irradiation. E C 0.6 ev and E V ev defects form due to neutron irradiation compare with SPS spectra before stress. E C 0.6 ev can be correlated with BB luminescence in DRCLS results. Fast and thermal neutron seems interact with defect differently. Energy levels in blue, red, and black color are defect exist before irradiation, induced by neutron irradiation, and complementary transitions. 114

144 Figure 4.27 The evolution of defect densities with (a) fast, and (b) fast + thermal neutron irradiation. DRCLS and t-sps confirm that low dosage of neutron irradiation may improve the GaN crystal quality. Thermal neutron seems to suppress defects induced by fast neutron irradiation. 115

145 and densities. Combined with results from DRCLS, SPS, and t-sps, we found that fast neutron irradiation with low fluence tends to improve the GaN crystal quality. However, higher fast neutron fluence will compensate this improvement and degrade the properties of GaN. Thermal neutron seems to suppress some of the defects induced by fast neutron irradiation and interact with defects in the GaN differently compared with fast neutron Neutron irradiation on GaN Schottky diode Figure 4.28 shows the ideality factor as a function of temperature from 10 K to 300 K with devices under neutron irradiation. The ideality factor (n) of Schottky diodes at 300 K before neutron irradiated, n/cm 2 fast, n/cm 2 fast, n/cm 2 fast + thermal, and n/cm 2 fast + thermal neutron irradiation are 0.982, 1.001, 1.039, 1.011, and respectively. The ideality factor can t be extracted from I-V characteristics of Schottky diode irradiated with fast and n/cm 2 fast + thermal neutrons due to ohimic behavior and contact melting. For thermionic emission theory, the ideality factor of an ideal Schottky barrier diode should equal unity [124]. Several mechanisms such as image force lowering [ ], generation-recombination interface states [127], thermionic field emission (TFE) transport [81], interfacial layers [128], tunneling [126, 129], and Schottky barrier height inhomogeneity [ , 168] will cause the ideality factor to be greater than 1. Ideality factor greater than unity value at room temperature such as Schottky diode irradiated with n/cm 2 fast + thermal neutron indicates additional carrier transport mechanisms may be involve when carriers 116

146 tunnel through the barrier. In Figure 4.28, we can clearly observed that n for most of the Schottky diodes irradiate with neutrons are between 1-2 with decreasing temperature down to K except one irradiated by n/cm 2 fast + thermal neutron, then increase dramatically with temperature below 80 K. Similar behavior has been observed in GaN Schottky diode I-V-T measurement [ ] and in general can be attribute to different carrier transport mechanisms at different temperature ranges, which commonly can be explained by TFE theory [81, 141]. According to TFE theory, the characteristic energy E 00 is given by [81, 141]: (4.6) where is Plank s constant, is the doping density of bulk GaN, is the GaN dielectric constant, is effective mas, and is Boltzmann s constant. The TFEdominated ideality factor n F can be expressed as: ( ) (4.7) By plotting the ideality factor n F vs. temperature, E 00 can be estimated. Figure 4.28 inset shows a series of lines from (4.7) with E 00 varies from 8 28 mev with temperature from K. The E 00 of Schottky diode before neutron irradiated, n/cm 2 fast, n/cm 2 fast, n/cm 2 fast + thermal neutron irradiation are roughly estimated as 16 mev, 10 mev, 16 mev, and 11 mev, respectively. Figure 4.29 shows the plot of nk B T vs. k B T showing four categories of temperature dependence of the ideality factors and the experimental I-V-T measurement results from five Schottky diodes under neutron irradiation. 117

147 Figure 4.28 Ideality factor (n) as a function of temperature for Schottky diode irradiated with different neutron fluence. n is between 1 and 2 for temperature above 100 K and increase dramatically with temperature below 80 K. Inset shows measured ideality factor with the prediction of TFE model. The characteristic energy E 00 varies from 8 28 mev with 2 mev/step. 118

148 If the diode obeys ideal Schottky theory and TE transport dominates, the experimental data will lie on a straight line labeled n = 1. If the current transport is mainly dominate by generation-recombination (G-R), the experimental data will lie on a straight line labeled n = 2. If the TFE transport dominates the current transport, experimental data will be able to fit by eq. (4.7) with different E 00. If FE dominates, then the experimental data can be fitted by a horizontal straight line shows the current transport is temperature independent. According to the fitting results, most of the carrier transportation in the Schottky diode can be explained by TFE theory. In TFE theory, when, TE is the dominant transport process. When, FE is the dominant transport process. When, TFE is the main mechanism which is a combination of TE and FE [79]. Before radiation, the Schottky diode E 00 energy is roughly 16 mev, indicating that between TE dominates between K and FE starts to play a role with temperature below 180 K. However, due to the low fluence neutron irradiation ( n/cm 2 fast and n/cm 2 fast + thermal), Schottky diode sample E 00 energy reduces to mev indicating that TE transport mechanism is dominant above ~ 100 K. Further increasing neutron fluence will increase E 00 (~ 16 mev with n/cm 2 fast) and a degradation of the Schottky diode can be observed. The Schottky diode irradiated with n/cm 2 fast + thermal neutron shows a trend that can t be well fitted by TE or G-R theory. We tentatively identify FE as dominate transport mechanism below 240 K with E 00 = 28 mev. Having established the knowledge of TE as the dominant transport mechanism at certain temperature range, we can now determine the barrier height and Richardson constant A * from Arrhenius analysis. Due to the small increase of ideality factor n with 119

149 Figure 4.29 Plot of k B T/q vs. nk B T/q for devices with various neutron irradiated fluence. Various dashed lines refer to the prediction of thermionic emission (TE, n =1), generation-recombination (G-R, n = 2), thermionic field emission (TFE, E 00 ), and field emission (FE, E 00 ) theories. 120

150 Figure 4.30 Modified Richardson plot (ln(i STE /A e T 2 ) vs. 1/nk B T ) for devices (a) before neutron irradiation and irradiated with (b) n/cm 2 fast, (c) n/cm 2 fast, (d) n/cm 2 fast + thermal, and (e) n/cm 2 fast + thermal neutrons. The Shottky barrier height ( ) and Richardson constant (A * ) are extracted from curve at temperature range dominated by TE transport. 121

151 reduced temperature, in order to analyze a wider temperature range and take into account the deviation from the ideality and experimentally observed dependence of n on temperature, we use a modified Richardson plot to extract diodes before and after neutron irradiation. From figure 4.30,.and A * of Schottky is equal to 0.63 ev, 0.65 ev, 0.6 ev, 0.63 ev and 0.33 ev for Schottky diode before and irradiated with n/cm 2 fast, n/cm 2 fast, n/cm 2 fast + thermal, and n/cm 2 fast + thermal neutrons. According to the results, fast neutron irradiation with fluence less than n/cm 2 only slightly increase and then decrease the Schottky barrier height. However, thermal neutron irradiation with similar fluence reduce the Schottky barrier height dramatically. Figure 4.30 shows that A * = AK -2 cm -2, AK -2 cm -2, 9.25 AK -2 cm -2, AK -2 cm -2, and 2.5*10-5 AK -2 cm -2, respectively, for Schottky diodes before and irradiated with n/cm 2 fast, n/cm 2 fast, n/cm 2 fast + thermal, and n/cm 2 fast + thermal neutrons. The theoretical Richardson constant of GaN is given by [79]: ( ) (4.8) where h is Planck s constant, m* is the effective mass, and m 0 is the free electron mass. Thus the theoretical Richardson constant of GaN is equal to AK -2 cm -2 [80]. Previous literatures report Richardson constants extracted from simple I-V model ranging from to 102 [80, , ] and can be attribute to enhanced tunneling current through charged surface state [135, 142], metal induced gap states [135], bulk material defects [146], interfacial oxide layer [142, 144], Schottky barrier height inhomogeneity [80, 136, 143, 145, 147, 168], and chemical reaction at interface [143, 122

152 ]. Schottky diode after n/cm 2 fast neutron irradiation show A * = AK -2 cm -2 which is closer to the theoretical prediction A * value of GaN than before neutron irradiation (A * = AK -2 cm -2 ) and after n/cm 2 fast neutron irradiation (A * = 9.25 AK -2 cm -2 ). The variation of the A * with respect to fast neutron irradiation fluence can be explained by the decrease of the surface defect density according to the t- SPS measurement results shown in figure However, the deviation of A * from the theoretical value for devices under fast + thermal neutron irradiation can t be explained by the surface defect variation. According to t-sps results, defect densities for most of the defects in GaN decrease with increasing fast + thermal neutron irradiated fluence. This should bring A* value of the Schottky diode close to its predicted value. However, an opposite trend was observed. Our preliminary XPS data shows an interaction of oxygen, Ni, and GaN occurs due to thermal + fast neuron irradiation. The A * anomaly caused by metal/gan interfacial chemical reaction has been reported by other researchers [143, 145, 147] and sometimes it is accompanied with barrier height reduction [147]. Therefore, we attribute the A* anomaly observed on devices due to thermal neutron irradiation to metal, most likely Ni, GaN, and oxygen interaction at interface. Another interesting observation of neutron irradiated devices is the increase of the reverse leakage current after neutron irradiation. Reverse leakage current can be attributed to edge related conduction [149], generation recombination in the space charge region [150], or surface and bulk defects [132, ]. A slightly increase of the reverse leakage current of device after fast neutron irradiation may be caused by the GaN surface defect increase. However, a greater increase of the reverse leakage current after 123

153 Figure 4.31 Sheet resistance (R sheet ), contact resistance (R contact ), and series resistance (R series ) as a function of (a) fast and (b) fast + thermal neutron irradiation fluence. 124

154 thermal neutron irradiation may again be attributed to Schottky barrier height inhomogeneity instead of the contribution from defects. Figure 4.31 shows normalized sheet resistance (R sheet ), contact resistance (R contact ), and series resistance (R series ) after (a) fast and (b) fast +thermal neutron irradiation. R sheet and R contact values for devices are extracted from transmission line model (TLM) method and R series is from Schottky diode I-V measurement. R sheet decreases at low fast neutron fluence may be due to GaN recrystallization [86]. Further increase the fast neutron fluence will cause R sheet, R contact, and R series increase and is possibly due to defects generated in GaN or at metal/semiconductor interface, consistent with our t-sps data shown in the earlier section. However, the dramatically increase of R sheet, R contact, and R series for device under thermal neutron irradiation might not originate from defects in GaN generated by thermal neutron, but from thermal neutron interact with metal and cause interfacial layers form at metal/semiconductor interface Discussion Figure 4.32 and table 4.1 summarized deep level defects observed in the SPS results and several electrical parameters extracted by I-V-T measurement before and after neutron irradiation. Several deep level defects have been reported before. The E C 0.6 ev defects has been previously observed in GaN after neutron irradiation which is attributed to N Ga [52, 53, 153] and V N -complex [86, 154]. Another neutron irradiation induced trap E C 0.8 ev is due to Ga i [50-51, 84, 87, 155], lattice disorder region [49, 51, 125

155 156], and Ga-H-V N [86, 157]. E V ev is more likely from V Ga -complex [105]. E C 1.1 ev defect is in general attributed to N i [50-51, 84]. E V ev defect is usually stems from V Ga [105] of V Ga -Ge defect [87, 158]. Researchers also observed that the intensity YB and BB defect emission in photoluminescence (PL) [33, 52, 54, 86-87, 159] or cathodoluminescence (CL) [49] results of GaN may vary with different neutron irradiation fluence. Our observation of Schottky diodes after fast neutron irradiation: (1) the trend of YB and BB to NBE ratio of DRCLS is consistent with that of increasing defect densities with increasing neutron fluence. The decrease of the defect intensity with lower neutron fluence (E V ev, E C 1.65 ev) is probably due to GaN recrystallization caused by fast neutron irradiation [86]. However, further increasing the fluence will compensate this enhancement due to defects or lattice disorder regions created by high energy recoiled neutrons, or the existing atoms displacement induced by the cascade process. The t-sps results show an increase of the E C 1.1 ev which can be correlated with YB in DRCLS, increase with increasing fluence and can possibly attribute to N i [50-51, 84]. (2) Fast neutron irradiation may also affect defects already existing in GaN such as V Ga according to t-sps results. The E V ev is likely due to V Ga in GaN [105]. (3) The increase of DRCLS BB luminescence can be attributed to a newly fast neutron induced E C 0.6 deep level trap. Accompanied with an increase of a E V ev deep level trap density after neutron irradiation, we attribute these two defect levels to edge dislocations [160] or lattice disorder region induced by fast neutron irradiation. This assumption is confirmed by an increase of a E C 0.8 ev defect level density which is previously observed in GaN 126

156 after fast neutron irradiation and is attributed to Ga i and lattice disorder region [49-51, 84, 87]. Furthermore, Li et al. also report that neutron irradiation will cause lattice relaxation in GaN [85]. (4) A * variation of Schottky diodes after fast neutron irradiation can be explained by the variation of defect densities in GaN since defects may cause A* to deviate from its theoretical value [135, 142, 146]. The contribution of TFE to TE transport also indicates that defect densities in GaN Schottky diode may play an important role in determination of carrier transport mechanisms. The dominance of TE transport at metal/gan interface implies that defect density decreases at interface which is observed in our DRCLS and t-sps results. The DRCLS, SPS, t-sps, and I-V-T measurement results show that low fluence of fast neutron irradiation may improve the crystal quality through GaN recrystallization which is confirmed by DRCLS (YB and BB luminescence decrease), t-sps (some of the defect densities decrease), and I-V-T measurement (as TE become more dominant in current transport process, R sheet decreases). However, further increasing the fast neutron irradiation fluence will compensate the crystal quality enhancement of GaN and the displacement damage start to degrade both material (defects and their densities increase in SPS and t-sps) and electrical (TFE is more prominent, A * deviates from theoretical value, R sheet, R contact, and R series all increase) properties of GaN and the Schottky diode. Our observations of Schottky diode after fast + thermal neutron irradiation: (1) the thermal neutron seems to suppress existing defects or defects created by fast neutron irradiation in GaN. Most of the defects and their densities detected by SPS and t-sps measurement decrease with increase neutron irradiation (thermal neutron) fluence except 127

157 E C 0.9 ev defect traps. This observation is consistent with our DRCLS results which show that YB and BB luminescence intensities are lower than GaN after fast neutron irradiation. (2) Figure 4.33 (a) and (b) show XPS results of device before and after highest fluence fast + thermal neutron irradiation. An additional peak with binding energy ~ ev can be barely observed. Researchers reported thermal neutron irradiation can be used as a tool for uniformly doping GaN by transmuting gallium into germanium throughout the material due to large penetration depth of neutrons [52-53, 55-56, 87]. Because the doping concentration can be roughly estimated as 0.13 Φ (Φ, thermal neutron fluence) [55, 87], the Ge doping concentration in our sample would then be around cm -3. The defect density of E C 0.9 ev from t-sps, assuming a 15 nm KPFM detection limit [161], is around cm -3. The density of this defect is around two orders of the magnitude higher than the t-sps detect defect density. However, the γ-ray due to transmutation process only produces relatively shallow defects [31, 163]. Moreover, the density of E C 0.9 ev is relatively stable with increasing fast neutron irradiation fluence shown in figure 4.27 (a). Thus, the E V ev might be from defects related to Ge dopants, possibly bound with V Ga. (3) Figure 4.33 (c) and (d) show SEM images of 40 nm Ni/GaN and 40 nm Ti/GaN samples after fast + thermal neutron irradiation. Ni metal melting patterns are observed that are likely due to thermal neutron irradiation since no melting areas can be observed for samples irradiated with fast neutrons or before irradiation. According to the XPS results, Ga, O, and Ni elements are detected at the melted area while only Ni is detected 128

158 at the undisturbed Ni region. We also observe pinwheels and bubbles that appear on a Ti/GaN sample after thermal neutron irradiation. Ti layers are smooth and without any features for device before and after fast neutron irradiation. Our XPS results show more nitrogen concentration in the Ti layer for the sample irradiated with thermal neutrons. Jeyachandran et al. reported that needle crystallites and bubbles are due to different N 2 concentration in Ti layers [161]. Other researchers find that nitrogen starts to diffuse into Ti layer at around 500 C [164]. Although the thermal foil that we put into the reactor only reads temperature between C, we believe that localized temperature may be much higher. Researchers also found that Ni and GaN inter-diffusion occurs at temperature around 300 C C [143, 147] and affect the electrical properties of the diode. Thus, the deviation of Schottky barrier height and Richardson constant from their theoretical value for Schottky diodes irradiated with thermal neutron, especially for neutron fluence exceeds n/cm 2, can be attributed to Schottky barrier inhomogeneity due to inter-diffusion of metal and semiconductor. The dominance of FE transport than TE transport for a wide temperature range and the increase of R sheet, R contact, and R series also support our proposed model. 129

159 Figure 4.32 The correlation between YB and BB luminescence and defects which are found by SPS technique. New defect transitions appear at 1.4 ev and 0.6 ev due to fast and fast + thermal irradiation. 130

160 Table 4.1 Summary of electrical properties extracted from I-V-T measurement of GaN Schottky diode before and after neutron irradiation. 131

161 Figure 4.33 XPS spectra of GaN (a) before and (b) after thermal neutron irradiation. The core level spectra show Ge 2p 1/2 peak at bounding energy around ev suggests some Ga has been transmuted into Ge due to thermal neutron. Black lines are guides to the eye. SEM images of 40 nm (c) Ni/GaN and (d) Ti/GaN after thermal neutron irradiations. The melting area shown in (c) and pinwheel and bubbles shown in (d) imply that localized temperature may be high enough for atoms inter-diffusions to occur, which is detrimental for device performance. 132

162 Chapter 5 Conclusion We have demonstrated the DRCLS capability to measure the temperature, stress, and defect distributions inside state-of-the-art GaN HEMT devices under realistic operating conditions. Both one-dimensional and cross-sectional characterizations are consistent with previous literature. Cross-sectional temperature mapping shows not only the hot spot location predicted by simulation but also additional hot regions inside the operating device. DRCLS spectra and maps correlated with KPFM mapping show that defects can accumulate locally to change the surface Fermi level position, band bending and hence surface potential after device operation. Our nanoscale external stress measurements are consistent with defects generated by inverse piezoelectric field-induced stress at the gate edge drain side area. Off-state stress degrades devices faster and generates both YB and BB defects, while on-state stress generates only YB defects. SPS identifies defects 1.2 ev above the GaN valence band that accumulate at failure (failure edge area) and other areas within the device after long time on-state stress. D-DRCLS reveals that new 3.75 ev, E c 1.35 ev and E V ev defect appears in AlGaN layer due to field-induced stress. Some of the defects can be correlated with BB and a raise of the YB close to surface are also observed. Device performance starts to degrade with stress, even before the V crit. Additional degradation mechanism needs to be considered. The degradation before V crit 133

163 may be cause by gate leakage current induced defect. These results show a strong correlation between stress, surface potential, defect, and device failure. Overall, the nanoscale depth-resolved optical and scanning probe techniques can be used to describe GaN-based HEMT failure mechanisms and predict the first-to-failure area under realistic situations. In neutron irradiation of GaN and related device, our DRCLS, SPS, t-sps, I-V-T and highlights: (1) Low dosage of fast neutron will enhance GaN crystal quality through recrystallization which is confirmed by lower t-sps measured defect density and the dominance of TE transport through I-V-T measurement. However, increasing the fast neutron fluence further will compensate this improvement by introducing new defects or lattice disorder region which can also degrade the electric properties. The I-V-T measurement also shows TFE instead of TE transport mechanism dominant due to trap formation. The defect density decrease and then increase due to fast neutron irradiation can also be correlated with similar trend of YB and BB defect luminescence extracted by DRCLS. (2) Instead of creating new defects, thermal neutron irradiation seems to improve the GaN crystal quality by suppressing most of the existing or irradiation induced defects. However, electrical properties degrade dramatically with increasing fluence from I-V-T measurement infers that contacts may be affected more by thermal neutron than fast neutron. XPS results confirm that Ti or Ni layers interact with GaN during thermal neutron irradiation. (3) The fast neutron assist defect formation vs. the thermal neutron assist metal-semiconductor inter-diffusion indicates fast and thermal neutron interact with GaN Schottky diode preferentially. (4) Recoil and cascade damages 134

164 induced by fast neutrons and metal/semiconductor interaction induced by transmutationgenerated heat are both detrimental for applications such as outer space and homeland security utilizing GaN-based devices. My results suggest new avenues of research to overcome these obstacles. 135

165 Chapter 6 Future work Although we already achieved AlGaN/GaN HEMTs cross-sectional temperature, stress, and defect measurement by DRCLS, several characterization strategies such as KPFM, SPS or t-sps can t be applied to some of the devices with t-configuration gat and thick passivation layers. This limitation will hinder the AlGaN/GaN HEMT reliability investigations. For example, t-sps measures defect densities close to the surface and KPFM can be used to extract electric field [ ]. The measured surface potential can be used to estimate electric field distribution through [165]: (6.1) ( ) (6.2) where E is an electric field, q is the electric charge, N d is the donor density, N a is the acceptor density, is the permittivity, is the dielectric constant of semiconductor. Figure 6.1 shows some of the preliminary electric field measurement with AlGaN/GaN HEMT biased under KPFM measurement. This experiment is a plane scan and was done on a AlGaN/GaN HEMT without passivation. Our results are at the same range of what experiment [165] and theoretical simulation [167] predict. However, several problems need to be solved before going further. The first one is the drift of the KPFM maps and is possibly due to the way we bias the device. The device is first mounted on a TO

166 can and a gold wire as bonded from the pad on the device to the pins on the can by wire bonding technique. Regular single core wires then are used to bias the TO-05 cylinder which is put on a bread board. The tension stored in the single core wires may cause the sample to rotate during scan. The second problem is the interaction between the cantilever and the bias voltage. This will cause the tip on the cantilever fail to reach the drain side gate edge area with is predicted to be where highest temperature, stress, and electric occur. Moreover, this interaction may affect the measurement results and the effect is unknown now. In order to measure the operating AlGaN/GaN HEMT by our surface characterization techniques, we use focused ion beam (FIB) technique to cut the sample in a ~ 30 angel. Figure 6.2 show (a) SEM, (b) AFM, and (c) KPFM results of a U-series AlGaN/GaN HEMT at an area which is marked with a dashed rectangle in figure 6.2 (a). We use a low energy (3keV) Ga beam to cut the device in order to prevent damage. A flat surface for measurement was achieved. However, unpassivated GaN HEMT device fail after FIB cutting due to the Ga beam bombardment induced damage in AlGaN layer. Instead, we use a M-series sample which has thick SiN passivation layer and repeat the experiment. The DC electrical measurement shows the electrical characteristics are similar to the same device before FIB cutting. This indicates this kind of measurement (FIB cutting) followed by DRCLS, KPFM, SPS, t-sps characterization) can be applied to most of the AlGaN/GaN HEMT nowadays. Another experiment worth doing is to stress the device (on-state, off-state, and semi-on state, V DS = 0 state), under different surroundings such as vacuum, air, oxygen, or nitrogen and measured the device characteristics by DRCLS, 137

167 Figure 6.1 AlGaN/GaN electric field measured by KPFM technique. The device is biased when doing KPFM scan. Black dashed line shows where gate and drain supposed to be. It is obvious that high electric filed regions appear at region close to the gate. 138

168 Figure 6.2 (a) SEM, (b) AFM, and (c) KPFM of U-series device after FIB cutting. (d) SEM image of M-series device after FIB milling. By using low kev gallium beam, the damage at cutting facet may be reduced. 139