Geiger-mode LiDAR with InP-based SPADs: From Airborne Platforms to Driverless Cars

Size: px

Start display at page:

Download "Geiger-mode LiDAR with InP-based SPADs: From Airborne Platforms to Driverless Cars"

Meagan Kelly
5 years ago
Views:

1 Geiger-mode LiDAR with InP-based SPADs: From Airborne Platforms to Driverless Cars Mark Itzler Argo AI Sep 2015

cars built with Ford Four office locations: Pittsburgh, Pennsylvania Dearborn, Michigan

2 Overview of Argo AI Developing Virtual Driver System for Autonomous Vehicles Founded by Bryan Salesky and Peter Rander in late 2016 Ford investment announced in Feb generations of cars built with Ford Four office locations: Pittsburgh, Pennsylvania Dearborn, Michigan Mountain View, California Cranbury, New Jersey Acquired in Oct 2017 In-house Geiger-mode LiDAR technology 2

3 Outline Basics of InGaAs/InP SPADs Device performance attributes anode contact SiN x passivation p + -InP diffused region i-inp cap multiplication region n-inp charge n-ingaasp grading i-ingaas(p) absorption n + -InP buffer n + -InP substrate anti-reflection coating cathode contact GmAPD cameras for airborne 3D LiDAR imaging FPA integration and camera performance GmAPDs for 3D LiDAR in autonomous vehicles Design considerations and demonstrator performance 3

4 Why detect single photons in the SWIR? Minimal loss in optical fiber (e.g., at 1.5 µm) Covertness (to human vision and I 2 night vision goggles) Reflective imaging (+ spectral information) Greater eye safety for active imaging (beyond 1.4 µm) 0.4 μm Environmental factors Less solar background than visible/nir Better atmospheric transmission Technological factors Maturity of pulsed laser sources (e.g., 1.06 µm, 1.5 µm) Maturity of SWIR optics (e.g., from telecom) InGaAsP Si 0.9 μm 1.7 μm 1.0 μm 4

(e.g., quantum cryptography and computing) Eve Bob Encryption using quantum properties of single photons

5 Single photon communications: initial driver for InGaAs/InP SPADs Exploit quantum mechanical nature of photons quantum information processing (e.g., quantum cryptography and computing) Eve Bob Encryption using quantum properties of single photons Alice single-photon detector module single-photon GmAPD array Communications in photon-starved environments free-space optical communications NASA/Jet Propulsion Lab deep-space optical comm 5

6 InGaAs(P) APD design for SWIR detection Two key device regions: Multiplication region: Create additional carriers by avalanche gain Absorption region: Absorb photon to create electrical carrier multiply electrical carriers absorb photons, create electrical carriers anode contact p + -InP diffused region i-inp cap multiplication region n-inp charge n-ingaasp grading i-ingaas(p) absorption n + -InP buffer n + -InP substrate anti-reflection coating SiN x passivation cathode contact E InGaAs for 1.5 µm InGaAsP for 1.06 µm optical input 6

7 1.5 μm SPAD DCR vs. PDE Performance Fundamental trade-off: DCR and PDE both increase with bias State-of-the-art DCR: ~1 khz at 20% PDE, ~2 khz at 30% PDE Higher PDE accessible with larger bias Dark Count Rate (Hz) 1E+5 1E+4 1E+3 1E+2 PLI production data for 25 μm dia. SPADs 1550 nm T = 218 K 0% 5% 10% 15% 20% 25% 30% 35% Photon Detection Efficiency 7 1E 4 1E 5 1E 6 1E 7 T = -55 C Dark Count Probability (ns 1 )

8 DCR performance distribution DCR distribution for ~1300 production 1.5 µm GmAPDs at 20% PDE ~90% of devices: 1 khz 10 khz ~10% of devices > 10 khz (significant outliers) Median DCR = 2.0 khz Dark Count Rate (khz) µm diameter SPADs 600 PDE = 20% PDE = 20% 500 Number of Devices X Device Number DCR (khz) More 8

9 Timing Jitter Several factors affect GmAPD detection timing ~100 ps jitter for typical operation Jitter often circuit-limited avalanche build-up (vertical and laterial) residual discontinuity short transit p + -diffusion i - multiplier i-charge i-grading i - absorber long transit K FWHM ~ 50 ps Timing Jitter (ps) 100 ~100 ps Zappa, Tosi, Cova, SPIE 65830E (2007) 10 InGaAs/InP SPAD Overbias Voltage (V) Itzler, et al., J. Modern Opt. 54, 283 (2007) 9

10 Afterpulsing and Geiger-mode reset time Some carriers are trapped during avalanche, then de-trap at later times Mitigate with longer hold-off (or reset) time but limits APD counting rate # of trapped carriers primary avalanche short hold-off time afterpulse p + -diffusion i - multiplier i-charge i-grading i - absorber trap sites located in multiplication region # of trapped carriers E c E v Long hold-off time 10

11 State-of-the-art DCR & PDE (PLI SPADs) Univ. Geneva (H. Zbinden) 5 Hz DCR at 20% PDE T = -110 C, 1550 nm Low temperature P app ~ 2% (20 μs deadtime) Free-running (negative feedback APD) 5 Hz Korzh, et al., APL 104, (2014) Toshiba / Cambridge, UK 55% PDE (at ~800 khz) T = 20 C, 1550 nm High temperature 55% P d ~ 3 x 10 4 per 360 ps gate P app ~ 15% ~90 ps jitter Self-differencing with 1 GHz gating Maximum count rate of 500 MHz Comandar, et al., JAP 117, (2015) 11

Silicon vs InGaAs/InP SPAD performance Compare underlying material properties of Si and InGaAs/InP SPADs Remove role of bandgap compare at T for which E g /2kT is equivalent Progress on

12 Silicon vs InGaAs/InP SPAD performance Compare underlying material properties of Si and InGaAs/InP SPADs Remove role of bandgap compare at T for which E g /2kT is equivalent Progress on absolute performance in InGaAs/InP using smaller active regions DCR: ~5X superiority in material properties for Silicon Afterpulsing: ~10X superiority in material properties for Silicon 12

13 Outline Basics of InGaAs/InP SPADs Device performance attributes anode contact SiN x passivation p + -InP diffused region i-inp cap multiplication region n-inp charge n-ingaasp grading i-ingaas(p) absorption n + -InP buffer n + -InP substrate anti-reflection coating cathode contact GmAPD cameras for airborne 3D LiDAR imaging FPA integration and camera performance GmAPDs for 3D LiDAR in autonomous vehicles Design considerations and demonstrator performance 13

contact p + -InP diffused region i-inp cap multiplication region n-inp charge n-ingaasp grading

14 Geiger-mode APD detector array design APDs scalable to large-format arrays semiconductor scaling GmAPD in every pixel 100 µm pixel pitch GmAPD schematic GmAPD pixel 32 x 32 GmAPD array Indium bump anode contact p + -InP diffused region i-inp cap multiplication region n-inp charge n-ingaasp grading i-ingaas(p) absorber n + -InP buffer n + -InP substrate anti-reflection coating SiN x passivation cathode contact 14

Focal plane array integration: 32 x 32 Focal Plane Array (FPA)

(PDA), GaP Microlens array (MLA) and CMOS Readout Integrated

interposer Microlens array (MLA) on top ROIC Ceramic interposer

15 Focal plane array integration: 32 x 32 Focal Plane Array (FPA) integration of three semiconductor chips: InP Photodiode array (PDA), GaP Microlens array (MLA) and CMOS Readout Integrated Circuit (ROIC) Lid 32 x 32 PDA 100 µm pitch FPA chip stack on interposer Microlens array (MLA) on top ROIC Ceramic interposer TEC Housing MLA GmAPD PDA FPA solid body model Full FPA assembly without lid 15

interposer TEC Housing FPA solid body model 50 µm pitch 128x32 FPA

16 Focal plane array integration: 128 x 32 Same FPA assembly platform for 128 x 32 (50 µm pixel pitch) Lid MLA GmAPD PDA ROIC Ceramic interposer TEC Housing FPA solid body model 50 µm pitch 128x32 FPA assembly 128x32 FPA chip stack on interposer Microlens array (MLA) on top 16

17 Turn-key camera-level integration Modular three-board design (FPA, FPGA, I/O) GUI interface supports all camera functions 10 cm x 10 cm x 9 cm assembled boards convection cooling 17

18 Timing operation for LIDAR imaging Camera synchronized to pulsed laser 13-bit timing counters in every pixel 0.25 ns time bin resolution (~3 cm) 186,000 frames/sec Laser launch (e.g., trigger signal from flash detectors) Counters start Counters stop µs 2 µs 3.5 µs 8000 bins Camera timing: Programmable delay Range gate Readout 18

19 32 x 32 camera DCR & PDE Performance maps for all pixels of 1.06 μm detection camera (100% yield) Dark Count Rate (khz) DCR Mean = 2.2 khz DC prob. ~ 0.4% / pixel / frame Photon Detection Efficiency (%) PDE Mean = 30.5% Number of Pixels Number of Pixels khz 30% PDE Dark Count Rate (khz) Avg PDE = 30.5% σ(pde) = 4.0% Avg DCR =2.2 khz Avg DCR = 2.2 khz σ(dcr) = 0.4 khz Avg PDE = 30.5% 2X 10% 14% 18% 22% 26% 30% 34% 38% 42% 46% 50% 54% Photon Detection Efficiency -25 C 19

20 FPA DCR contrasted with discrete SPADs FPA distribution over ~1000 pixels much more uniform Array singulation Flip-chip bonding Indium bumps Number of Pixels Avg DCR = 2.2 khz σ(dcr) = 0.4 khz Avg PDE = 30.5% 2X -25 C 4 mm Dark Count Rate (khz) Discrete device singulation Die pick & place Die bonding to carrier Wire bonding Implies DCR degradation due to additional back-end assembly & packaging processes 0.4 mm Number of Devices X DCR (khz) PDE = 20% More 20

21 GmAPDs with reduced size detectors Performance enhancements using reduced active areas DCR, afterpulsing, crosstalk, radiation tolerance Primary trade-off: optical coupling and assembly p-contact metallization SiN x passivation p + -InP diffused region i-inp cap multiplication region n-inp charge n-ingaasp grading i-ingaas(p) absorption n + -InP buffer n + -InP substrate anti-reflection coating n-contact metallization optical input p-contact SiN x passivation p + -InP diffused region i-inp cap multiplication region n-inp charge n-ingaasp grading i-ingaas(p) absorption n + -InP buffer n + -InP substrate anti-reflection coating n-contact metallization optical input 21

22 Scaling of FPA format for 128x32 camera 128 x 32 camera with 50 μm pitch, improved PDE vs. DCR Average DCR (khz) khz 44% 1.06 μm -20 C Number of Pixels X Avg DCR = 1.3 khz σ(dcr) = 0.22 khz Avg PDE = 32.9% 0.0 0% 10% 20% 30% 40% 50% Average PDE (%) 128 x 32 Dark Count Rate (khz)

23 Arrays of GmAPDs have optical crosstalk Single photons emitted during avalanche can trigger neighboring pixels Spatial correlation roughly 1/R 2 Spatial patterning from backside reflections Temporal correlation on ns-scale A Various strategies to mitigate crosstalk 21 x 21 spatial map of crosstalk probability Structure in spatial map B Cumulative Crosstalk Probability 14% 12% 10% 8% 6% 4% 2% 0% % PDE 25% PDE 20% PDE Dimension N for NxN pixel area 23

24 Geiger-mode 3D LiDAR mapping GmAPD-based commercial mapping systems by Harris Corp. based on PLI 128 x 32 GmAPD cameras enables 10X faster data collection than other LIDAR technologies imagery courtesy of aerial photo: Seattle ferris wheel 24

Oct 2015 Retrofit up to 2500 Blackhawks LiDAR for object identification in targeting pods Customized

25 Airborne navigation and target ID LiDAR for autonomous helicopter navigation for Sikorsky Black Hawks United Technologies Aerospace Systems LiDAR based on PLI GmAPD cameras Key self-flying demonstration in Oct 2015 Retrofit up to 2500 Blackhawks LiDAR for object identification in targeting pods Customized camera formats for defense prime contractors Successful flight tests demonstrating technology capability 25

26 Outline Basics of InGaAs/InP SPADs Device performance attributes anode contact SiN x passivation p + -InP diffused region i-inp cap multiplication region n-inp charge n-ingaasp grading i-ingaas(p) absorption n + -InP buffer n + -InP substrate anti-reflection coating cathode contact GmAPD cameras for airborne 3D LiDAR imaging FPA integration and camera performance GmAPDs for 3D LiDAR in autonomous vehicles Design considerations and demonstrator performance 26

27 Geiger mode LiDAR for shorter range? Autonomous vehicles: most exciting short range LiDAR application Market size, societal impact Safety imperative for sensors with complementary modalities Wide consensus that driverless car sensor suite will have: LiDAR Cameras RADAR LiDAR Good Low light Performance Good Weather Performance True 3D, High Resolution Little Ability To Read Signs Cameras Poor Low light Performance Worst Weather Performance Inferred 3D, Not True 3D Reads Signs / Sees Colors RADAR Good Low light Performance Best Weather Performance True 3D, Low Resolution Cannot Read Signs 27

28 Airborne vs. Automotive LiDAR Airborne Automotive Range ~10 km ~200 m Resolution 2 10 pts/m cm resolution at 200 m Cost 28

29 Automotive LiDAR design considerations Wavelength: 9xx nm (NIR) vs. >1400 nm (SWIR) Illumination: Flash vs. scanning Laser technology: Fiber/solid state vs. diode Detector technology: Linear mode vs. Geiger mode 29

source 1 ns pulse 700 900 nm 100,000X 900 1 ns pulse 1100 1300 Wavelength (nm) >1400 nm 1500 InGaAsP Silicon based Geiger

30 Wavelength selection for Auto LiDAR Greater eye safety for >1400 nm LiDAR: longer range detection Eye safety constrains 900 nm LiDAR to <100 m for low reflectance objects Eye safe Exposure (J/cm 2 ) 1E+0 1E 1 1E 2 1E 3 1E 4 1E 5 1E 6 1E 7 Point source 1 ns pulse nm 100,000X ns pulse Wavelength (nm) >1400 nm 1500 InGaAsP Silicon based Geiger mode LiDAR ~100 meters Silicon 0.9 μm 1.7 μm 0.4 μm 1.0 μm InP based Geiger mode LiDAR 200+ meters Detector spectral bandwidth 30

elegant but impractical for Auto ~100 FOV

Even given Mpixel array, illuminating all

with ~1000 pixel 1D array, scan to cover

31 Flash vs. scanning for Auto LiDAR Flash illumination: elegant but impractical for Auto ~100 FOV with 0.1 resolution needs 1000 pixels in one direction ~Mpixel 2D array 2D Detector Array Even given Mpixel array, illuminating all pixels takes prohibitive laser energy Laser Scanning provides best balance of laser/detector resources Image vertical FOV with ~1000 pixel 1D array, scan to cover horizontal FOV 1 D Vertical GmAPD Array Return Scanned Photons Scanning Mirror Covering HFOV Diode Lasers Scanned Emitted Photons Fixed VFOV 31

32 Importance of resolution 0.46 (8 mrad) Typical of existing Auto format LiDAR systems today 0.23 (4 mrad) (1 mrad) Capability of Geiger mode Auto LiDAR in development 32

33 Laser and detector technologies Fiber laser/solid state lasers vs. diode lasers Cost! Size/weight Linear mode detectors vs. Geiger mode detectors Transmitter power! Linear mode needs excessive power for long range, diodes not practical System size/weight 33

34 Auto LiDAR system emulation Auto LiDAR demonstrator Color coded for height GmAPD 128 x 32 camera 512 x 64 demo 3D point cloud format Scaling to 2048 x 512 equivalent Color coded for distance 34

35 3D driving video imagery with demonstrator Imagery taken with demonstrator mounted to car roof Google maps view of 300 m driveway through office park 35

capabilities for: High-altitude 3D LiDAR mapping Xtalk, array format/yield

36 SWIR Geiger-mode technology summary Initial performance developed for: Single-photon fiber communications PDE, DCR, afterpulsing, jitter Proven unique capabilities for: High-altitude 3D LiDAR mapping Xtalk, array format/yield Potential for disruptive impact in: High-performance automotive LiDAR range, resolution, SWaP, cost 36

LiDAR Teach-In OSRAM Licht AG June 20, 2018 Munich Light is OSRAM

LiDAR Teach-In OSRAM Licht AG June 20, 2018 Munich Light is OSRAM www.osram.com LiDAR Teach-In June 20, 2018 Munich Light is OSRAM Agenda Introduction Autonomous driving LIDAR technology deep-dive LiDAR@OS: Emitter technologies Outlook LiDAR Tech Teach-In June 20, 2018