Effect of Mechanical Stress on Gate Current and Degradation in AlGaN/GaN HEMTs

Transcription

1 Effect of Mechanical Stress on Gate Current and Degradation in AlGaN/GaN HEMTs Andrew Koehler, Min Chu, Amit Gupta, Mehmet Baykan, Scott Thompson, and Toshikazu Nishida Florida MURI Review November 10, 2011

2 Outline Motivation Stress Dependence of Gate Current and Characterization of Electric Field in AlGaN Barrier Simulation of the Dominant Stress-Altered Gate Leakage Current Gate Current Degradation under Mechanical and Electrical Stress Summary 2

3 Mechanical Stress on GaN HEMT Reliability FLOORS Stress Measurements Gate Leakage Modeling Wafer Bending Stress Measurements 3 ΔR/R (%) Compression ΔR(σ) Tension Stress (MPa) I G (A) ΔJ G /J G (%) 9 J G0 = 603 x 10 3 A/m G G ΔJ G (σ) Time (s) 4 3 Stress (MPa) DFT Simulation E C E T LLO Devices V G (V) t=0, As Built E V t>0, Degradation 3

4 Gate Current Degradation Large E AlGaN Inverse PZ Mechanical Stress Defect formation Device Breakdown Increase in J G High Power State On State J. Joh, et al., IEDM 2007 (Courtesy, Erica Douglas) V DS = 0 V State Off State Systematically investigate impact of mechanical stress on J G 4

5 Outline Motivation Stress Dependence of Gate Current and Characterization of Electric Field in AlGaN Barrier Simulation of the Dominant Stress-Altered Gate Leakage Current Gate Current Degradation under Mechanical and Electrical Stress Summary 5

6 Wafer Bending (4-Point Bending) Stress on upper surface at the center of the substrate: σ E t yx a L 2a 2a 2 3 = = [Strength of Materials, edited by S. Timoshenko Krieger, Melbourne, FL,1976] E = Young s modulus t = sample thickness y = vertical displacement a, L = rod spacing t a L Neutral Axis Tension Compression Force Force y Sample 6

7 GaN HEMT Wafer Bending Setup Probe tip Gold wire Conductive epoxy GaN HEMT wafer samples are too small for 4- point bending HEMT sample attached to high carbon stainless steel strip and inserted into bending setup Device is wirebonded and connected to probe tips to take electrical measurements while simultaneously applying stress Stress is calibrated using strain gauge mounted on sample (~20 MPa/graduation) Stress is applied longitudinal to channel direction GaN HEMT wafer under stress Epoxy Steel rods High carbon steel strip 7

8 J G Stress Measurement Procedure 1. Bias device in dark measuring J G V G = -0.1 to 4 V J G stable Incrementally apply stress Stress Tension Compression Incrementally release stress Avg. J G at each stress Time 8

9 Experimental Results V G = V V G = -4 V 9

10 Bias and Stress Sensitivity of J G Stress sensitivity of J G decreases Characterize J G mechanism and impact of stress on J G 10

11 S Device Operation ON Condition OFF Condition AlGaN G D S AlGaN G D GaN 2DEG GaN 2DEG depleted under gate V G = 0 Ni AlGaN GaN 2DEG High reverse bias 2DEG depleted under gate Built-in field at V G = 0 2DEG is not inversion electrons (from surface states) No flatband during operation Device is turned off by depleting 2DEG V T is defined when 2DEG is depleted 11

12 Gate Leakage Transport Mechanisms ON Condition OFF Condition Ni AlGaN GaN Ni AlGaN GaN 2DEG E C 2DEG Depleted Poole-Frenkel emission Fowler-Nordheim tunneling Trap-assisted-tunneling Phonon-assisted tunneling E C To characterize gate leakage, need E AlGaN 12

13 1D E AlGaN Calculation Ni AlGaN GaN = 1.4 ev V G = 0 = 0.4 ev E AlGaN is a function of ψ s 2DEG density (n s ) is used to calculate E AlGaN Gauss s Law E AlGaN depends on σ int and n s σ int is constant n s depends on gate voltage 13

14 Ideal Device: n s and E AlGaN n s0 No trapped charge in ideal device n s no accumulation of holes (n i ~ cm -3 ) E AlGaN V T built-in field E AlGaN saturates below V T 14

15 Ideal vs. Actual Device Ideal device Actual Device σ int σ int No charge traps (σ int = σ PZ ) Trapped charge reduces σ int (σ int = σ PZ Q int ) Lower n s lower V T Can experimentally calculate E AlGaN 15

16 Calculating E AlGaN of Actual Device n s x10 12 n s (V G ): σ int Integration of the C V curve C F/m V T 3x x x10 12 n s (cm -2 ) Calculate n s0 Take the difference of n s0 and charge induced by gate bias V G (V) Can determine experimentally the necessary parameters for 1D E AlGaN 16

17 1D E AlGaN of Actual Device Q it = 0 Q it 0 Q it reduces σ int Lowers V T and E SAT Increases E AlGaN (0) Can experimentally calculate E AlGaN (matches 1D model) 17

18 Physics of E AlGaN Saturation 1D model predicts saturation of E AlGaN Verify using Sentaurus 2D simulations Self-consistently solve: Poisson s equation Electron/hole continuity equations Piezoelectricity: Piezoelectric/spontaneous polarization implemented using fixed interface charge * Used same constants as 1D calculation 18

19 Surface Potential in GaN Potential extracted 0.5 nm below AlGaN/GaN interface V G = -10 V V G = -4 V V G = -2 V V G = 0 V ψ S increases with V G at center of gate explains E AlGaN saturation Large potential drop at gate edges 19

20 2D E AlGaN Simulation at Center of Gate 2D simulated E AlGaN saturates at center of the gate Must analyze the gate edges in more detail 20

21 E AlGaN at Gate Edge E AlGaN at 1 nm below gate contact Source Gate Drain Potential drop at gate edges results in horizontal electric-field increasing E AlGaN Gate V G = -10 V V G = -4 V V G = -2 V V G = 0 V [J. Joh, et al., IEDM 2007] E AlGaN peaks at gate edge could cause device breakdown (inverse piezoelectric effect) 21

22 Section Summary Stress sensitivity of J G decreases with increasing reverse gate bias Explained physics of E AlGaN Saturation under center of gate Peaks at gate edges Validated 1D model with experimental data and 2D simulations 22

23 Outline Motivation Stress Dependence of Gate Current and Characterization of Electric Field in AlGaN Barrier Simulation of the Dominant Stress-Altered Gate Leakage Current Gate Current Degradation under Mechanical and Electrical Stress Summary 23

24 Literature Review on HEMT Gate Leakage [Yan et al. APL, 97, (2010)] [Sathaiya et al. JAP, 99, (2006)] [Mitrofanov et al. JAP, 95, 6414 (2004)] [H Zhang et al. JAP, 99, (2006)] FN PFE 24

25 Direct Tunneling Formula Fowler-Nordheim Tunneling metal AlGaN GaN Thermionic Field Emission J q m m m q = E exp 8πhφb 3qhE ( / ) 8π 2 ( φ ) 3 2 * * e n 2 n b [Zhang et al. JAP, 99, (2006)] metal AlGaN GaN * qa T = k φ 0 ( ) ( ) B J f P d f FD FD φ φ φ * 1 8π 2mq 3/2 =, P= exp φ φ 3hE B φ 1+ exp kt / q [Karmalkar et al. APL, 82, 3976 (2003)]

26 Direct Tunneling Current Density Gate Current Density (A/m 2 ) 1E+5 1E-4 1E-13 Experiment TFE T = 300K m* = 0.23m e ɸ b = 1.4V Δ image lowering 1E Direct tunneling is unlikely to dominate the reverse gate leakage until at least 2.5MV/cm. FN Electric Field (MV/cm) 26 26

27 Bulk Trap-Assisted Gate Leakage 0 ɸ t ɸ B E d R1 R2 Step 1: Direct tunneling Thermal assisted tunneling Step 2: Direct tunneling Poole Frenkel emission [O Mitrofanov et al. JAP, 95, 6414 (2004)] Phonon assisted tunneling [P Pipinys et al. JAP, 99, (2006)] metal AlGaN GaN [Sathaiya et al. JAP, 99, (2006)] Assume uniform trap distribution Assume triangular potential within AlGaN Only one defect level is considered 27 27

28 Bulk Trap Leakage Model Details ( 1 ) R = C N f f P R 1 1 t FD 1 = C N f P 2 2 t 2 R1 R2 metal AlGaN GaN R: tunneling/emission rate P: tunneling/emission probability f: probability of a defect state being occupied by an electron At steady state: t FD 1 2 t 2 Cf P Cf P+ CP 1 FD 1 ( 1 ) R = R C N f f P = C N f P f = 1 FD R CC N f PP 1 1 N t FD 1 2 = = t + Cf 1 FDP1 CP 2 2 Cf 1 FDP1 CP q J = R d Ed Overall leakage current density: ( ) φ φ E φ t 28 28

29 Gate Leakage Simulation Result Gate Current Density (A/m 2 ) 1E+3 1E+2 1E+1 1E+0 1E-1 1E-2 1E-3 Experiment (E T =0.5eV) Simulation (2 nd step: FN, PFE, PAT) 1E Gate Electric Field (MV/cm) Ns = cm 3 E T = 0.82eV T = 300K The bulk trap-assisted leakage process is limited by step-1 in which electrons tunnel from metal gate to defect levels

30 Poole-Frenkel Emission from Surface States J PF I PF J PF ET ΔφT = CEexp, rkt J gate where Δ φ = β E = T πε [Yeargan et al. JAP, 39, 5600 (1968)] 3 qe AlGaN J gate J back metal AlGaN GaN J gate =J PF -J back [Harrell, et al. Thin Solid Films, 2002] α m* I back Jback = C'exp E [Yan et al. APL, 97, (2010)] I PF increases dramatically and I back is negligible once V g <

31 Poole-Frenkel Emission from Surface States 1E-5 1E-6 Symbols (experiment results) Dash lines (simulation results) 375K Log(J/E) 1E-7 1E-8 1E-9 300K r = 1.25 ± 0.06, E T = 0.49 ± 0.04 ( ) C = E (MV/cm) 1.6 1E E 1/2 0 J (A/m 2 )

32 Stress-Altered Effective Mass and Field Out of Plane m* Change (%) Effective mass change Uniaxial stress ~ 0.3% change for 400MPa -200 Biaxial stress ~ 0.5% change for 400MPa Stress (GPa) Polarization Change (%) Electric field change Uniaxial stress ~0.7% change for 400MPa -200 Biaxial stress ~1.4% change for 400MPa Stress (GPa) sp 3 d 5 tight binding model was used to simulate m out ΔE field = ΔP piezoelectric 32 32

33 Stress-Altered J G Based on Δm* and ΔE Experiment Simulation V G =-0.25V 33 33

34 Stress-Altered E T Qualitative Argument Lattice Mismatch Wafer Bending Ga N No stress Biaxial tension (2.8GPa) Uniaxial tension (0.4GPa) Uniaxial compression (0.4GPa) Θ 2 Θ 2 Θ 2 Θ 1 Θ 1 Θ 1 ΔΘ 1 <0, ΔΘ 2 >0 ΔL 1 <0, ΔL 2 >0 ΔΘ 1 <0, ΔΘ 2 >0 ΔL 1 <0, ΔL 2 >0 ΔΘ 1 >0, ΔΘ 2 <0 ΔL 1 >0, ΔL 2 <0 AlGaN AlGaN AlGaN AlGaN [Choi et al. APL vol. 92, pp (2004)] 34 34

35 Stress-Altered r-parameter No Stress Tensile Stress Compressive Stress r original Less compensation Δr > 0 Heavy compensation Δr < 0 ΔE T and Δr are treated as fitting parameters Stress affects gate leakage current mainly through ΔE T and Δr

36 Stress-Altered Gate Leakage Calculation ET ΔφT α m* Jσ = 0 = CEexp C'exp rkt E ( * +Δ *) ( ) ( ET +ΔET) Δφ α m m T Jσ 0 = C( E+ΔE) exp C'exp ( r r) kt +Δ E+ΔE ΔJG J J = J J G σ 0 σ= 0 σ =

37 Effect of Stress on J G 2.5 Simulation ΔJ / J (% / 100MPa) Fitting parameters: ΔE T = 1.6±0.3 (mev/gpa) Δr = 0.028±0.004 (/GPa) 300K 300K 375K Electric Field (MV/cm) I gate increases/decreases under uniaxial tension/compression

38 Section Summary Investigated the possible gate current mechanism of the reverse-biased GaN HEMT Determined the PFE from surface state to be the dominant leakage mechanism Simulated and understood the stress dependence of J gate. Collaborate with Erin Patrick for FLOODS gate leakage modeling 38

39 Outline Motivation Stress Dependence of Gate Current and Characterization of Electric Field in AlGaN Barrier Simulation of the Dominant Stress-Altered Gate Leakage Current Gate Current Degradation under Mechanical and Electrical Stress Summary 39

40 Theory I : AlGaN/GaN HEMT Degradation Theory II : Fresh Device Stressed Device Crystallographic defect-formation due to inverse piezoelectric effect [Dr. del Alamo s group] Metal diffusion from gate into AlGaN layer [Dr. Fan Ren s group] 40

41 Si MOS Gate Oxide Breakdown Gate: Polysilicon Conductive path I G t trace for a 1.67 nm oxide during CVS Traps Substrate [Robert O Connor et. al. Semicond. Sci. Technol. 2005] Oxide breakdown is marked by sudden increase of leakage current. Breakdown occurs due to clustering of trapped charges at localized sites, leading to enhanced conduction. Breakdown can be Soft Breakdown (SBD) or Hard Breakdown (HBD). 41

42 Time Dependent Electrical Step Stressing Vary length t of each E-stress step. Perform step stress test on GaN HEMTs, with t=10sec, 100sec, 500sec, 1000sec, and measure I G simultaneously. 42

43 Time Dependent Electrical Step Stressing Results V DS = 0V 100 sec/step Device A V G Step I G V DS = 0V 1000 sec/step Device B V G Step I G V cric = -33 V V cric = -27 V Trapping transients are observed at each stress step prior to breakdown. One large sudden increase in I G is observed, followed by multiple small jumps in I G 43

44 I G -V G after Electrical Step Stressing Device A Device B After stress test V DS =0V After stress test V DS =0V Before stress test Before stress test Increase in I G is observed after device degradation 44

45 Combined Electrical and Mechanical Stressing Measure I G during the application of stress Need to determine step size for E-stress and mechanical stress 45

46 Piezoelectric Constitutive Relation T S = = c s t E. S e. E t E. T + d. E T= Stress matrix (6X1) S = Strain matrix (6X1) E = Field (3X1) c E = stiffness matrix (6X6) s E = compliance matrix (6X6) d t,e t = piezoelectric matrix (3X6) V G (V) V GS =-2V V GS =-1V -1V of V G generates ~10 MPa of biaxial tensile stress 46

47 Experimental Set Up E-Stress Mechanical Stress Courtesy, David Cheney V G step size is -5V, each applied for 8min Uniaxial mechanical stress is applied in steps of 15 MPa, upto 150 MPa max. 47

48 I G Degradation under Electrical-Mechanical stress Breakdown point: V G = -25V σ = 60 MPa E-Stress Mechanical Stress I G 48

49 Section Summary Measured I G under E stress stepping at varying lengths of stress step Device Flow: AFRL E-stress Mechanical Stress for t>0 Integrated mechanical stress and electrical stress to perform degradation experiments 49

50 Summary and Future Work Investigated the dominant leakage mechanism to be PFE in reverse-biased AlGaN/GaN HEMT Stress sensitivity of gate current decreases with increasing reverse bias Determined that inverse piezoelectric effect plays significant role in AlGaN/GaN HEMT failure Device Flow: AFRL E Stress Mechanical Stress (next to LEAP) Theory Flow: Stress dependent gate leakage modeling Electromechanical (FLOORS) 50

51 Acknowledgments MURI FUNDING 51

52 Effect of Mechanical Stress on Gate Current and Degradation in AlGaN/GaN HEMTs QUESTIONS? 52