1. Ultra High Speed InP Heterojunction Bipolar Transistors Mattias Dahlström Trouble is my business, (Raymond Chandler)

2. Ultra High Speed InP Heterojunction Bipolar Transistors Introduction to HBT’s How to make a fast HBT… –Delay terms –The graded base –The base-collector grade Recent results –Record f max mesa DHBT* –Record f  DHBT *details regarding this to follow

3. The transistor Schematic of an HBTTypical common-emitter characteristics Small change in base current  large change in collector current

4. InP lattice structure Nearest neighbor: 2.5 A Lattice constant: 5.86 A

5. InP and InGaAs have  -L separations of ~0.65 eV, vs ~0.4 eV for GaAs→ larger collector velocity InGaAs has a low electron effective mass → lower base transit time InGaAs InP

6. Objectives and approach Objectives: fast HBTs → mm-wave power, 160 Gb fiber optics desired: 440 GHz f t & f max, 10 mA/  m 2, C cb /I c <0.5 ps/V better manufacturability than transferred-substrate HBTs improved performance over transferred-substrate HBTs Approach: narrow base mesa → moderately low C cb very low base contact resistance required, and good alignment → carbon base doping, good base contact process high f t through high current density, thin layers bandgap engineering: small device transit time with wide bandgap emitter and collector

7. Potential uses of InP HBT Communication systems: wireless communication, fiber optics transceivers, digital processing in radar (ADCs, DACs) Types of circuits: broadband amplifiers, power amplifiers, laser/modulator drivers comparators, latches, fast logic Circuit characteristics 1-10 000 HBTs per IC Very high demands for speed (40-200 GHz) Fast logic with moderate power consumption (~20 mW/gate) Moderate Output Power mmwave power amps, optical modulator drivers ~6 V at J c =4 mA/μm 2, ~2 V at J c =8 mA/μm 2

8. DHBT band diagram: under bias base emitter collector

9. High speed HBT: some standard figures of merit Small signal current gain cut-off frequency (from H 21 ) Maximum power gain ( from U) Collector capacitance charging time when switching :

10. Scaling laws for fast HBTs for x 2 improvement of all parasitics: f t, f max, logic speed… base  2: 1 thinner collector 2:1 thinner emitter, collector junctions 4:1 narrower current density 4:1 higher emitter Ohmic 4:1 less resistive Challenges with Scaling: Collector mesa HBT: collector under base Ohmics. Base Ohmics must be one transfer length sets minimum size for collector Emitter Ohmic: hard to improve…how ? Current Density: dissipation, reliability Loss of breakdown avalanche V br never less than collector E g (1.12 V for Si, 1.4 V for InP) ….sufficient for logic, insufficient for power narrow collector mesa transferred-substrate

11. Contact resistance: tunneling through barrier High doping: 1-9 10 19 cm -3 Small bandgap: InAs<InGaAs<InP<GaN Surface preparation: no interstitial oxide Metal reactions Theory: idealized contact

12. Pd-based contacts Pd/Pt reacts with III-V semiconductor: InGaAs+Pd  As + (In,Ga)Pd+(In,Ga)(Pd,As) Pd reaction depth ~4 x thickness 25 Å Pd for 300 Å base Contact resistance: 100-500  -  m 2  1-20  -  m 2 from TLM and RF-extraction Ohmic contact to p-type material 10-100 times worse than n-type. Work function line-up, electron/hole effective mass Yu, J.S.; Kim, S.H.; Kim, T.I. ‘’PtTiPtAu and PdTiPtAu ohmic contacts to p-InGaAs’’, Proceedings of the IEEE Twenty-Fourth International Symposium on Compound Semiconductors, San Diego, CA, USA, 8-11 Sept. 1997

13. Emitter resistance

14. Emitter resistance: grades removed At degenerate doping levels grades are not necessary Contact resistance: 50  m 2  25  m 2  15  m 2 High doping  3 10 19 cm -3 No InGaAs-InP grade necessary at very high doping Thin undepleted n- emitter Small emitter area increases R ex InGaAs cap layer InP emitter light doping heavy doping

15. Base resistance R bb is a critical parameter for f max, and in npn HBT the base contact resistance dominates. R bb is minimized through high base doping and improved base contact metallization, small undercut W gap, and long emitter L e TLM measurement

16. Problems with very thin bases Etching and depletion effects reduce the effective base thickness T b, and increases the base resistance. At 500 nm scaling generation, best base thickness is 30-40 nm better f max, lower R bb -related delay terms in gate delay, minimal improvement in f  between 25 & 30 nm High resistance

17. Increase of sheet resistance with thin base layers InGaAs base doped 6 10 19 cm -3, surface pinned at 0.18 eV. Surface depletion decreases base thickness 40 Å. R b,extrinsic =800-1000 Ω/sq R b,intrinsic =600-750 Ω/sq Base protected by E/B grade (contacts diffused through 160 Å grade) Surface depletion Wet etching Base surface exposed :

18. Collector resistance R c : access resistance between collector contact and the mesa. Minimized by large collector contacts, and low resistance subcollector

19. Subcollector design Some still use all InGaAs subcollector… Subcollector resistivity 500 A InGaAs + 2000 A InP ~ 11  /sq 125 A InGaAs + 3000 A InP ~ 9  /sq Etching selectivity of InGaAs vs. InP main limit  50 A InGaAs Contact resistance better to 125 A than 50 A after annealing Goals: minimize electrical resistance minimize thermal resistance limit thickness to improve manufacturability Thermal conductivity of InGaAs ~5 W/mK Thermal conductivity of InP ~68 W/mK T subc Etch stop layer provides collector undercut – less C bc - 53 % of thermal resistance

20. Base-emitter capacitance C je is the junction capacitance between the emitter and base C je corresponds to ~100 Å depletion thickness Minimized by shrinking the emitter area at fixed or at increasing current I c

21. Base-collector capacitance C bc is the junction capacitance between the base and subcollector.

22. Base-collector capacitance T c = 3000 A  2150 A  1500 A A bc must be kept small: narrow emitter narrow base contacts undercut of base contacts implant or regrowth Breakdown limits thickness Thickness (A)Breakdown (V) 21507.5 15004-5 Collector thickness reduced due to speed requirements: C cb increases !

23. Theory of the base If gain is limited by Auger recombination in the base: is 100-250 fs is 10-50 Decreasing increases. High N a and T b for low  s decreases Grade gives 30-50 % improvement The base sheet resistance: The base transit time: p s is 400-900  /sq

24. Base Transit Time Fitting of relevant parameters of the form With doping as Intrinsic carrier concentration Diffusivity Kroemer’s double integral: Drift-Diffusion equation for base current: Exit term Solution used for evaluation of the base transit time: Ballistic injection:

25. Base grading Graded bandgapGraded doping Doping 8  5 10 19 cm -3 Change in In:Ga ratio InAs: E g =0.36 eV GaAs: E g =1.43 eV

26. Base grading: induced electric field Induced electric field accelerates electrons towards collector – decreases base transit time and increases gain Limits: strain Limits: Bandgap narrowing, needs degenerate doping

27. The effect of degenerate doping Evidence: Observed V be increase V on ~ φ bi, increases with E v N b =4 10 19 cm 3  0.75 V N b =8 10 19 cm 3  0.83 V for graded base-emitter Strong variation in Fermi-level with doping at high doping levels

28. Base bandgap narrowing Model after V. Pavlanovski Bandgap grade Doping grade BGN provides an electric field opposing the doping-induced field. ~1:5 in magnitude

29. Base Transit time Ballistic effects may arise when T b <180-200 @5 10 19 cm -3 (Tessier, Ito) Results:Bandgap gradedDoping graded DC gain2518 ft250 GHz282 GHz Bandgap grade and doping grade give same  b

30. Collector design Transit time: Close inspection show velocity near base most important GradeNo Grade -Use grade -Use setback

31. Base-collector grade Early grade designs: Too coarse No setback layer Recent grade designs: 15 A period 200 A setback layer Gain:7 f  :128 GHz (T c =3000 A) J kirk :1.3 mA/μm 2 Gain:27 f  :282 GHz (T c =2150 A) J kirk :4 mA/μm 2

32. InAlAs/InGaAs super lattice Why super lattice? –MBE is more suited for super lattice than quaternaries. –InP/InGaAs gives poor quality material due to phosphorous-arsenic intermixing MOCVD growth → InGaAsP grade GaAsSb base needs no grade

33. Quantum well trapping Electron/hole in the InGaAs well 500 meV InAlAs potential barrier A rough approximation: the infinite potential well. If E n > 500 meV (InGaAs/InAlAs potential)  no electron confinement ~31 A is the maximum allowed InGaAs width by this model Quantum mechanical trapping in grade

34. The delta-doping H. Kroemer : a conduction band difference can be offset with a grade and a delta-doping With this choice the conduction band will be smooth No delta-dopingDelta-doping V bc =0.3 V

35. The setback layer An InGaAs layer beneath the base –Margin for Base dopant diffusion –Increases Electron speed at SL Setback V bc =0.3 V No setback V bc =0.3 V

36. Collector design: doping

37. Collector design: velocity and scattering No  -L scattering  -L scattering possible Collector band profile designed for greatest possible distance without  -L scattering

38. Collector under current (simulation) N c reduced by J c /q/v sat Current blocking

39. Metal resistance Resistance of e-beam deposited metals higher than “book” values. Metal resistance increases when T<1000 A Problem for base contact (PdTiPdAu with 600 A gold)  sm =0.5 Ω/sq3-8 Ω added to R bb TiPdAu 200/400/9000 A PdTiPdAu 30/200/400/600 A TiPdAu 200/400/4000 A Reduces f max Thermal stability?

40. Results 2150 A collector  high f max, high V br,CEO IPRM 2002, Electron Device Letters, Jul. 2003; M. Dahlström et al, ''Ultra-Wideband DHBTs using a Graded Carbon-Doped InGaAs Base'' 1500 A collector  high f , high f max, high J c Submitted to DRC 2003; M. Dahlstrom, Z. Griffith et al.,“InGaAs/InP DHBT’s with ft and fmax over 370 GHz using Graded Carbon-Doped Base”

41. InGaAs 3E19 Si 400 Å InP 3E19 Si 800 Å InP 8E17 Si 100 Å InP 3E17 Si 300 Å InGaAs graded doping 300 Å Setback 2E16 Si 200 Å InP 3E18 Si 30 Å InP 2E16 Si 1700 Å SI-InP substrate Grade 2E16 Si 240 Å InP 1.5E19 Si 500 Å InGaAs 2E19 Si 500 Å InP 3E19 Si 2000 Å 300 A doping graded base Carbon doped 8*10 19  5* 10 19 cm -2 200 Å n-InGaAs setback 240 Å InAlAs-InGaAs SL grade Thin InGaAs in subcollector High f max DHBT Layer Structure and Band Diagram V be = 0.75 V V ce = 1.3 V Emitter Collector Base

42. InGaAs 3E19 Si 400 Å InP 3E19 Si 800 Å InP 8E17 Si 100 Å InP 5E17 Si 400 Å InGaAs graded doping 300 Å Setback 3E16 Si 200 Å InP 3E18 Si 30 Å InP 3E16 Si 1030 Å SI-InP substrate Grade 3E16 Si 240 Å InP 1.5E19 Si 500 Å InGaAs 2E19 Si 125 Å InP 3E19 Si 3000 Å Thinner InP collector Collector doping increased to 3 10 16 cm -3 Thinner InGaAs in subcollector Thicker InP subcollector High f  DHBT Layer Structure and Band Diagram V be = 0.75 V V ce = 1.3 V Emitter Collector Base

43. Results: DC High f max DHBTHigh f  DHBT Gain: 23-28 n b /n c :1.05/1.44 V br,CEO : 7 V Gain: 8-10 n b /n c :1.04/1.55 V br,CEO :4 V No evidence of current blocking or trapping

44. Results: RF High f max DHBTHigh f  DHBT Highest f max for mesa HBT Highest f  for mesa DHBT Highest (f , f max ) for any HBT High current density

45. Results: Base width dependence Emitter junction 0.6 x 7  m, V c e =1.3 V T b =300 A. T c =1500 A

46. Results: RF - trends Variation of f  vs. I c and V ce, of an HBT with a 0.54  m x 7.7  m emitter, and a 2.7  m width base- collector junction. Variation of f  and f max vs. V ce, of an HBT with a 0.54  m x 7.7  m emitter, and a 2.7  m width base- collector junction. I c =20 mA. Need higher V ce for high current f  drops at high V ce high V ce for full collector depletion

47. Results: evolution ff f max Final grade Old grade New grade Strong improvement in f  and J opt J opt f  and f max > 200 GHz at J c >10 mA/  m 2 T c =1500 A

48. Capacitance vs. current Emitter junction 0.5x7.6 um T c = 1500 A, N c =3 10 16 cm -3 DHBT 20 Graded emitter base junction DHBT 17 Abrupt emitter base junction Emitter junction 0.54x7.6 um and 0.34x7.6 um. T c = 2150 A, N c =2 10 16 cm -3 J max ~3 mA/  m 2 J max ~6.5 mA/  m 2 48 % J max ~3.2 mA/  m 2 for T c =2150 A

49. Area dependence on capacitance reduction WeWe BB C EW bc Ccb from Y-parameters at 5 GHz V ce =1.3 V V ce =1.5 V Extrapolating with linear fit gives 55 % for r=1 Ccb is reduced where the current flows  reduce extrinsic base

50. Max current density vs. emitter size The current at which Ccb increases (J max ) as a function of emitter width for two different HBT Narrow emitters have higher critical current density Not necessarily higher f t (due to R ex ) - Current spreading

51. Calculation of current spreading Poisson’s equation with depth dependant current J(x) Solving double integral provides Kirk threshold correction term J now has emitter width dependence at J kirk Lateral diffusion One-dimension Kirk condition

52. Summary of delay terms

53. Emitter heat sinking Emitter interconnect metal  2 μm to 7 μm

54. Process improvements: local alignment Machine alignment provides <0.2 μm alignment in good weeks

55. Process improvements: lift-off Improved hardening of top resist surface 0.4 x 8 μm emitters, ~1 μm thick

56. What to do in the future: short term Have new material with InAs rich emitter cap  less R ex  increased f  Doping grade and combined grade  less  b  increased f  ? Small scale circuits by Z. Griffith and others Write paper on Kirk effect / collector current spreading Hålls me slåttern

57. What to do in the future: long term Need a more SiGe like processing technology –Lift-off –Isolation –Emitter regrowth Work on HBT design –Emitter design –Base grade See circuits come out …

58. Summary of work Linear base doping grade New base-collector grade Pd based base ohmics Narrow base mesa HBT –Record f max –Record f  InAs HEMT’s

59. Conclusion Mesa HBT can achieve superior performance to T.S. InAlAs/InGaAs S.L. grade permits use of InGaAs for base and InP for collector Excellent transport characteristics in collector InGaAs setback layer improves b-c grade PdTiPdAu base ohmics can achieve p- type contact resistance as good as n-type

60. in case of questions

61. Results: base-collector capacitance Full depletion Variation of C cb vs. I c and V ce. Note that V be =0.85-0.90 volts over the same bias range.

62. Hole mobility extraction With measured base sheet resistance and doping level the base hole mobility can be estimated

63. Collector velocity from Kirk threshold Slope corresponds to collector saturation velocity

64. Collector velocity from  bc

65. InP-InGaAs and InP-GaAsSb Base-collector grades necessary Grades not necessary

66. H21 at 5 GHz vs. current Emitter junction 0.5x7.6 um E0.7 B05 Gain does not depend on V ce, but on bias. Max gain around 26.5

67. Current RF gain vs. voltage Heating likely cause

68. Results: Gummel

69. DHBT 20: Capacitance cancellation data Not max f t,f max (current too low for that, but wanted to avoid blowing)cc Theory: G-L scattering reduces collector transit time and heating

70. Capacitance cancellation Previous slide 4 fF reduction from ft vs. Vce relation, very close to measured

71. Results: RF validity W-band measurements one week apart Re-measurements show similar f t and f max. Roll-off is very close to -20 dB/decade in the 75-110 GHz band.

72. Resistance vs. doping InGaAs and InP n-type doping : 1-3 10 19 cm -3 InGaAs p-type doping 1.2 10 20 cm -3 : no p-InP with C doping

73. Mesa HBT mask set: first iteration Emitters 0.4, 0.5, 0.6, 0.7, 1.0, 2.0 μm wide, 8 μm long for RF measurements Base extends 0.25, 0.5 and 1.0 μm on each side of base Base plug in revision 1 Emitter ground metal 2 μm wide

74. Mesa HBT mask set: second iteration Emitters 0.4, 0.5, 0.6, 0.7, 1.0, 2.0 μm wide, 8 μm long for RF measurements Base extends 0.35, 0.5 and 1.0 μm on each side of base Base plug now on smaller tennis-racquet handle Emitter ground metal extended to 7 μm width

75. RF measurements: CPW structure 230  m

76. RF measurements: air bridges 120  m New m : /4=137 um 117  m120  m

77. RF measurements: calibration LRL calibration using on wafer Open, Zero-length through line, and delay line OLTS used to check U in DC-50 GHz band Probe pads separated by 460  m to reduce p-p coupling RF environment not ideal, need: thinning, air bridges, vias for parasitic mode suppression

78. RF parameter extraction Emitter resistance (Error page 101 eq. 5.4) Base collector capacitance Base collector resistance Base collector delay time, ideality factor and capacitance

79. Switching speed limited by output capacitance How do we get speed improvement Design Specifications set ΔV and R L  sets I Reduce C by decreasing A C  Increase in J since I fixed  J limited by Kirk Effect  Increase in J increase dissipated power density Formula simplistic  insight

80. Can we measure R th (Method of Lui et al ) Ramp I B for different V CE Measure V BE and I C Depends on current density Large uncertainty in values. Fitting regression curves helps to reduce error

81. Validation of Model Caused by Low K of InGaAs Max T in Collector Ave Tj (Base-Emitter) =26.20°C Measured Tj=26°C Good agreement. Advice Limit InGaAs Increase size of emitter arm

82. Ultra High Speed InP Heterojunction Bipolar Transistors Why this title? Some recent conference results show transistor f  of 130 GHz… InP is a brittle semiconductor with a metallic luster. We mix it with GaAs and AlAs. Use Si and C as dopants Heterojunction: contains junctions of different materials

83. DHBT carrier profile quick comment that this is unbiased....under bias both DR will fill with E