Integrated Si-Based Photonics James S. Harris Stanford University Peking University Summer School Beijing, China July 19, 2013 Demands for Optics at Shorter Distances Internet, Wide Area Local Area Network (WAN) Network (LAN) On-card Rack-to-Rack Inter-chip STANFORD Card-to-Card Intra-chip Distance Multi-km 10 - 2000m 30+m 1m # of Lines 1 1-10 ~100 ~100 - 1000 Use of Optics Since 1980s to early Since late Present 1990s 1990s Distance 0.1 - 0.3m 10 - 100mm <10mm Present ++ # of Lines ~1000 ~10,000 ~100,000 Use of Optics 2010 - 2015 Probably after 2015 Sometime in the future Adapted fromtechnology IBM Research Can evolution of telecom address Inter/Intra chip applications? Peking University Summer School, July 19, 2013 JSH 2 Architecture change STANFORD • Multiple cores on a chip are already available – Trend: increase # of cores NOT speed or complexity • Parallel architectures increased bandwidth • Nanophotonic communication is a credible solution D. Fattal & M. Fiorentino HP Labs Peking University Summer School, July 19, 2013 JSH 3 Communications Challenge Intel Microprocessor, 2005 Peking University Summer School, July 19, 2013 STANFORD Broadway, New York City, 1887 JSH 4 On-chip Interconnects STANFORD 100 Gate Delay FO = 4 Relative Delay Local 10 ITRS Roadmap 2005 Global W Repeaters Global WO Repeaters Global interconnects RC limited 1 CMOS device Local connects 100 250 190 130 90 66 45 32 Process Technology Node (nm) What is required to solve this challenge? LOW COST, LOW POWER, INTEGRATED, CMOS COMPATIBLE, OPTICAL TRANSCEIVERS Peking University Summer School, July 19, 2013 JSH 5 Interconnect Performance STANFORD Energy/bit Latency Wmin for Cu CNT from ITRS for optics = 0.6µm Cdet=Cmod=10fF - Electrical interconnects power dissipated by wire and repeaters latency by wire and repeaters - Optical interconnect (1 Channel) power dissipated by end devices latency by end devices • Cu, CNT: small wire width → Energy per bit decreases as wire pitch is scaling (CV2). Latency increases as wire pitch scales down • Optics favorable for longer wires Koo, Kapur and Saraswat, IEEE Trans. Electron Dev., Sept. 2009 Peking University Summer School, July 19, 2013 JSH 6 Photonic Integrated Circuit-1993 STANFORD Soref, Proc. IEEE, 1687 (1993) Waveguide architecture with butt coupled fibers III-V edge emitting lasers, modulators, detectors and high-speed electronics (HBT or HEMT) All off-chip and Mostly III-V devices Peking University Summer School, July 19, 2013 JSH 7 Silicon-Compatible Photonics: A Materials Challenge STANFORD Integrate the required photonic devices on silicon Intel http://www.research.ibm.com/photonics/images/soi_phwire.jpg Y.-H. Kuo, et.al., Nature 437 (2005) http://www.bit-tech.net/news/2007/09/18/intel_has_worlds_fastest_sige_photo_detector/1 Can a new material be engineered to suit our needs? Peking University Summer School, July 19, 2013 JSH 8 Band Structures of GaAs, Si & Ge STANFORD E [111] [100] GaAs E Global Minima at zone center k Local Minima at zone center [100] [111] Si k E [100] k [111] Ge Poor Efficient Efficient Emission & Emission & Absorption Absorption Absorption Silicon Based Germanium Peking University Summer School, July 19, 2013 JSH 9 Unique Multiple Band Ge/SiGe QW STANFORD Deep direct band gap, QW Ec, e- Ec,L Ev,lh Ev,hh Relaxed Si1-yGey buffer Strained Si1-xGex barrier Strained Ge QW Peking University Summer School, July 19, 2013 <1ps tunneling >100 GHz modulation Lower, shallow indirect band L minima Direct band gap transition h+ ∆EC, direct = 0.4 eV ∆EV = 0.1 eV for heavy hole Strain causes valence band splitting JSH 10 Quantum-Confined Stark Effect STANFORD l l Electro-absorption and electro-optic modulation by tuning electron-hole coupling in quantum wells More pronounced for excitons (bound electron-hole pairs) Ec Ev No E-field E-field 1. Red shift of absorption edge 2. Smaller wave function overlap – lower α 3. Change of n through Kramers-Kronig relationship Peking University Summer School, July 19, 2013 JSH 11 SiGe and GE QW Growth on Si l Graded SiGe buffer is widely used l l l Direct growth of SMOOTH, THIN buffer l l Si Graded buffer Low surface roughness Post anneal reduces dislocation density Buffer thickness is critical for single mode waveguide devices on SOI High Dislocation Density 400nm 10µm l Graded SiGe Low defect density Thick buffer layer Large surface roughness-Critical for QWs Ge or SiGe Si Single-Tgrowth direct growth Peking University Summer School, July 19, 2013 400nm Ge or SiGe l STANFORD High-T Ge or SiGe Anneal Low-T SiGe Si Two-Tgrowth direct growth JSH 12 SiGe Surface Morphology STANFORD QWs require surface roughness ≤ 0.2nm 2-Temp-step Ge-on-Si by MBE l As-grown roughness RMS ~ 0.2nm Peking University Summer School, July 19, 2013 Single-Temp-step SiGe-on-Si by CVD l l As-grown roughness RMS ~ 2.5nm Annealed roughness RMS ~ 0.228nm JSH 13 Strain-balanced Structure STANFORD growth direction n+ SiGe cap layer Undoped SiGe buffer layer Ge/SiGe MQWs Compressive Tensile Undoped SiGe buffer layer Strain force ε p+ Relaxed SiGe buffer layer Silicon Substrate Average Si concentration in Ge/SiGe QWs equals that of SiGe buffer Y.-H. Kuo, et al, Nature 437, 1334 (2005) Peking University Summer School, July 19, 2013 JSH 14 Ge/SiGe Modulator on Si STANFORD Ge 10nm/ Si0.15Ge0.85 16nm Y.-H. Kuo, et al, Nature 437, 1334 (2005) Materials, Processes and Temperature are all CMOS-compatible Peking University Summer School, July 19, 2013 JSH 15 Strong QCSE in Ge/SiGe QWs STANFORD Y.-H. Kuo, et al, Nature 437, 1334 (2005) Peking University Summer School, July 19, 2013 JSH 16 Fabrication Process STANFORD Peking University Summer School, July 19, 2013 JSH 17 Small Signal Modulation STANFORD Bias: 2.5V Device top view size: 6µm *6 µm Response limited by contact resistance Peking University Summer School, July 19, 2013 JSH 18 Integrated Waveguide Modulator, Detector and Laser STANFORD Ge Quantum Well(s) N-SiGe P-SiGe P-Si Waveguide modulator SiO2 Source, Modulator and Detector have identical QWs Function determined by bias polarity Light source Modulator Photodetector SiGeSn cap layer GeSn QWs SiGeSn buffer layer Si waveguide SiO2 Si substrate Peking University Summer School, July 19, 2013 JSH 19 Outline STANFORD l l Introduction Ge/SiGe QCSE Electroabsorption Modulator l l l l Strained Ge and GeSn Emitters l l l SiGe Growth and Characterization Device Fabrication and Measurement Optical Characterization Growth & Characterization of Tensile Strained Ge Growth & Characterization of GeSn Summary Peking University Summer School, July 19, 2013 JSH 20 Si Based Laser Ge direct band gap engineering STANFORD GeSn material - Sn is semi-metal - Reported direct bandgap for SnxGe1-x is between 10% and 20% Sn - Lattice relaxed or compressive strained layer Strain - Theoretically,1.8% tensile strained Ge is direct bandgap - Thin layer of Ge - Potential buffer layer (larger lattice constant) Relaxed GeSn, GaAsSb, InGaAs M. Bauer et al., APL, 81, 2992 (2002) He and Atwater, PRL, 97(10), 1937 (1997) M.V. Fischetti et al., JAP. 80(4) 2234 (1996) Peking University Summer School, July 19, 2013 JSH 21 Ge Laser STANFORD Si-Ge Laser Structure Si-Ge Laser Spectrum Camacho-Aguilera-MIT OptExp 20 11317 (2012) Good News—Ge can be made to lase Bad News—Insanely high threshold current Peking University Summer School, July 19, 2013 JSH 22 Role of Heterostructures and Dimensionality on Lasers STANFORD Impact of epitaxy, improved materials Zh. Alferov, IEEE JSTQE, 6 832 (2000) Nobel Lecture FOUR orders of magnitude decrease in threshold current density as a result of heterojunctions and energy band engineering Peking University Summer School, July 19, 2013 JSH 23 Highly Strained Ge Nano-bridge STANFORD Nano-bridge Structure Calculated Gain & Loss Süess-PSI Nature Photon 10 1038 (2013) Free carrier absorption increases with carrier densities and creates high laser threshold current More sophisticated band engineering & QWs are essential Peking University Summer School, July 19, 2013 JSH 24 The Potential of Ge/GeSn: Direct Bandgap STANFORD Advantages of Ge: • Si-compatible material • Low effective mass in Γ (0.038m0) • Nearly direct-bandgap and band engineer-able Large effective mass (0.22m0) Γ Eg = 0.8eV Inefficient optical transitions L Eg = 0.664eV Simplified Ge Bandstructure Peking University Summer School, July 19, 2013 JSH 25 The Potential of Ge/GeSn: Direct Bandgap STANFORD Advantages of Ge: The Biaxaial Tensile-Strain Effect • Si-compatible material ~1.5%Large Strain Required effective mass (0.22m0) • Low effective mass in Γ (0.038m0) Γ Inefficient optical transitions • Nearly direct-bandgap and band engineer-able L Eg = 0.8eV Eg = 0.664eV Y. Huo, et al., APL (2011) Simplified Ge Bandstructure Peking University Summer School, July 19, 2013 JSH 26 The Potential of Ge/GeSn: Direct Bandgap STANFORD The Sn-Alloying Effect ~6-8% Sn Required Γ L R. Chen, et al., Applied Physics Letters 99 (2011) Simplified Ge Bandstructure Peking University Summer School, July 19, 2013 JSH 27 Tensile strained Ge (TEM) STANFORD In0.3Ga0.7As 10nm tensile strained Ge 10nm In0.3Ga0.7As 300nm In0.15Ga0.85As 200nm GaAs substrate 100 nm InGaAs buffer layers: Defects are terminated at interface Ge layer: 2.46% in-plane tensile strain Peking University Summer School, July 19, 2013 Ge 10 nm JSH 28 Measured Strain & PL in Ge/InGaAs STANFORD Strain (Raman) Normalized intensity (a.u.) 1 Photoluminescence 1 Straine d Ge 0.8 0.8 InGaAs 0.4 Intensity (a.u.) 0.6 Bulk Ge 0.2 0 260 270 280 0.34% 0.92% 1.81% 2.33% 290 300 310 0.6 0.4 320 0.2 Raman shift (cm-1) Normalized intensity (a.u.) 1 0.8 0.6 In0.1Ga0.9A s In0.2Ga0.8A s In0.3Ga0.7A s In0.4Ga0.6A s 0 1300 0.4 0.2 0 260 280 300 raman shift (cm-1) 320 Peking University Summer School, July 19, 2013 1350 1400 1450 1500 (nm) 1550 1600 Indium concentration Ge Raman shift (cm-1) Strain (Raman) 10% 1.05 0.26% 20% 3.63 0.91% 30% 7.13 1.78% 40% 9.60 2.35% JSH 29 Photoluminescence Ge/In0.4Ga0.6As STANFORD 1 20K 30K 40K 50K 75K 100K 150K 200K 300K Intensity (a.u.) 0.8 0.6 0.4 0.2 40% InGaAs Strained Ge 40% InGaAs 27% InGaAs 13% InGaAs GaAs 0 1200 1400 1600 1800 2000 2200 (nm) Strained Ge/In0.4Ga0.6As is a Type II Heterojunction Peking University Summer School, July 19, 2013 JSH 30 The Issue for GeSn: Solid-Solubility STANFORD 3.5% Sn Attempted SGTE Alloy Database, http://www.crct.polymtl.ca/fact/phase_diagram .php?file=Ge-Sn.jpg. Increasing Sn 400oC Sn HighTemperature Segregation >4.5% High-Strain Precipitation Y. Shimura, et. al. Jpn. J. Appl. Phys. 48 (2009) Peking University Summer School, July 19, 2013 JSH 31 MBE Ideal for Investigative Tool STANFORD GeSn and SiGeSn Challenges: 1. Low solid-solubility (1%) of Sn in Ge – MBE can decouple source and substrate growth temperatures 2. Challenges in finding precursors that decompose at low temperature – Very high-purity (99.999% or better) solid sources available for evaporation 3. Lattice constant changes greatly with Sn or Si alloying, adversely affecting the bandstructure and film quality III-V Chamber (InGaAs/GaAs ) Group IV Chamber (GeSiSn, GeSn) Peking University Summer School, July 19, 2013 Group IV Stack Strain Control with III-V JSH 32 Want High-Quality, DirectBandgap GeSn STANFORD Goal: Explore basic material properties and unravel competing strain and composition bandgap effects to provide basis for quantum well device design Our Method: MBE Growth on GaAs/lattice relaxed InGaAs • Ability to control strain with Indium composition • GaAs/InGaAs & GeSn optically distinguishable • Higher Ge strain and higher Sn incorporation using low-temperature MBE growth (200oC) Peking University Summer School, July 19, 2013 GeSn InGaAs Anneal InGaAs GaAs JSH 33 TEM of 7% GeSn Layers STANFORD 7 X greater than equilibrium solubility Ge0.07Sn0.93 10% InGaAs Buffer strained Ge or GeSn 5 nm InxGa1-xAs 200nm GaAs substrate GaAs Substrate High quality Ge93%Sn7% epi layer: • No defects • No precipitation (phase segregation) Peking University Summer School, July 19, 2013 H. Lin, et al., Thin Solid Films 520 (2012) JSH 34 Surface Quality Maintained w/High Sn Fraction Increasing Sn percentage 4.5% Sn, 100oC 7.0% Sn, 200oC STANFORD 8.8% Sn, 100oC RMS = 0.529nm RMS = 0.403nm RMS = 0.626nm 4.5% and 7.0% Samples grown on In0.10Ga0.90As, ~50nm GeSn 8.8% Sn Sample grown on In0.25Ga0.75As Surface RMS roughness changes only slightly with H. Lin, et al., Thin Solid increasing Sn %. Peking University Summer School, July 19, 2013 JSH 35 Films 520 (2012) Great Material Quality Possible with MBE STANFORD GeSn with 10.5% Sn, low-T growth GeSn Film InGaAs Buffer RMS=0.519nm Peking University Summer School, July 19, 2013 H. Lin, et al., Thin Solid Films 520 (2012) JSH 36 Where Does GeSn Become Direct Bandgap? STANFORD R. Chen, et al., APL 99 (2011) Bowing = 2.1 eV, ~7% Sn H. Lin, et al., APL 100 (2012) Bowing = 2.4 eV, ~6.5% Sn Peking University Summer School, July 19, 2013 J. Mathews, et al., APL 97 (2010) Consensus: It’s a lot less than people thought! Experimental data suggests it’s around 5.5-7% Sn – very achievable!! JSH 37 SiGeSn/GeSn/SiGeSn Quantum Well STANFORD SiGeSn 50nm InGaAs buffer GeSn/Si GeSn InGaAs buffer STEM-EDX Intensity for Si (a.u.) SiGeSn 30nm GeSn 30nm Intensity for Ga and Ge (a.u.) Glue Ge Si Ga Position (arb. unit) GaAs substrate Peking University Summer School, July 19, 2013 JSH 38 Strain and Compositional Analysis STANFORD • Composition and Strain measured by SIMS and XRD-RSM – SiGeSn: Si = 5.58%; Sn = 9.16% Eg = 0.785 eV – GeSn: Sn = 7.91%; strain = 0.3% Compressive • Previous studies1,2 decoupled Sn composition and strain effects – Eg = 0.548 eV calculated for Ge0.92Sn0.08 1) H. Lin, et al., Appl. Phys. Lett. 100 141908 (2012) 2) H. Lin, et al., Appl. Phys. Lett. 100 102109 (2012) Peking University Summer School, July 19, 2013 Direct Indirect In-plane tensile strain JSH 39 GeSn Low-Temperature Photoluminescence STANFORD T=20K T=294K Peking University Summer School, July 19, 2013 JSH 40 Lattice-Matched Options for GeSn QWs STANFORD Direct Bandgap Energy (eV) Unstrained Quantum Wells possible with the addition of Si Peking University Summer School, July 19, 2013 JSH 41 The Stage Is Set – What About Lasers? STANFORD Mirror Mirror Gain Region Photon Emission > Photon Absorption Onset of lasing when optical gain ≥ loss Require low-threshold lasers LOW LOSS Optical Losses: • Minimize mirror losses -> Ge difficult to cleave, high-Q resonators • Free carrier absorption µ n, p -> Minimize doping to reach threshold • Optical scattering and mode confinement: Good design and fabrication Carrier Recombination and Threshold Current: • Reduce SRH recombination µ n, p -> Maintain high material quality, reduce doping • Auger recombination µ n 2 p -> Minimize doping to reach threshold With competing L-Valley occupation, n-type doping of 2-5 x 1018 cm-3 is optimum Peking University Summer School, July 19, 2013 JSH 42 Challenges for a GeSn Laser STANFORD High Carrier Concentration Produces Free Carrier Absorption Effect of FCA on Laser threshold: • Large internal losses increase threshold since required carrier concentration at threshold is an exponential function of αi • Even worse for threshold current, (Ideal case), (Auger Recombination dominant) Peking University Summer School, July 19, 2013 JSH 43 Gain Spectra for GeSn QWs STANFORD Gain Spectrum for p=n=2.4e19 cm-3 Gain Spectrum for 8% Sn (GeSn) Increasing Sn Increasing Carrier Concentration Pure Ge Addition of Sn greatly increases the net material gain for fixed carrier concentration!! MUCH LOWER threshold current lasers! Only need carrier density of ~5e18 cm-3 for 1000 cm-1 of gain for 8% Sn Peking University Summer School, July 19, 2013 JSH 44 High-Quality Material is Paramount STANFORD Relative Laser Threshold vs. Carrier Lifetimes in Just Direct-Bandgap GeSn (ΔEc = 0) Relative Threshold, log10 Due to Density of States, ~98% of carriers still in the L-valley • Non-radiative lifetimes critical for both valleys • Need high-quality material to reduce defect states • Moderate n-type doping Peking University Summer School, July 19, 2013 JSH 45 The Benefits of Direct-Gap Materials ΔEc GeSn Photoluminescence STANFORD Γ L R. Chen, et al., Applied Physics Letters 99 (2011) Increase in PL with Sn because more carriers occupy the direct Γ-valley! Sn alloying results in increased optical efficiency Peking University Summer School, July 19, 2013 JSH 46 carrier concentration (cm -3) Carrier Confinement (for Ge/SiGe) Energy (eV) 1 0.5 0 -0.5 -1 450 500 Y (um) 550 STANFORD 20 10 10 10 450 500 550 Bias = 0.76V Simulation: •10nm Ge QW in Si0.2Ge0.8 pn junction Carrier concentration: •1.5E19 in QW and <1E18 in barrier •Carriers are confined in QW (15 – 50 X) •Calculated net gain of 200cm-1 Experiment: •Ge 14nm*3QW •200nm Ge grown on SiGe buffer PL signal: •Stronger PL from QWs •Carrier confinement in QWs Xiaochi Chen et al. “Room Temperature Photoluminescence from Ge/SiGe Quantum Well Structure in Microdisk Resonator” [2012] Peking University Summer School, July 19, 2013 JSH 47 Can Lase, but not easy with Q=100 STANFORD 100nm Ge 20nm GeSn (8%) 90nm Ge 15% of TE mode experiences GeSn Gain and FCA 75% of TE mode experiences Ge FCA (no band-to-band absorption) ~30x higher carrier density in GeSn QW than in Ge barriers due to heterostructure Choose low resonator loss to hit threshold, Q of 500 -> ~100cm-1 Peking University Summer School, July 19, 2013 JSH 48 GeSn Photoluminescence 8.00E-07 STANFORD 5% GeSn Mostly Relaxed 7.00E-07 6.00E-07 5.00E-07 3% GeSn Compressive 1% GeSn Compressive 4.00E-07 3.00E-07 0% GeSn Relaxed 2.00E-07 1.00E-07 0.00E+00 1400 1600 1800 Peking University Summer School, July 19, 2013 2000 2200 2400 JSH 49 Microdisk Ge QW Photoluminescence STANFORD PL intensity (a.u.) 8000 10 mW 20 mW 30 mW 40 mW 60 mW 6000 4000 2000 0 1350 1400 1450 1500 Wavelength (nm) •Amplified spontaneous emission pumped by 900nm pulsed laser 1550 0.8 0.6 0.4 0.2 0 -0.2 -0.4 -0.6 -0.8 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 Peking University Summer School, July 19, 2013 • Two small peaks on each fringe TE / TM or higher order mode Ridge waveguide profile, thick disk •PL intensity is super linear, estimated gain of 1500 cm-1 JSH 50 Summary of Initial MBE GeSn Studies STANFORD • Strain-free layers of high-Sn% GeSn alloys – Low-growth temperatures using MBE – Lattice-matched growth using InGaAs buffers • Increased photoluminescence for higherSn samples – – – – Large increase in integrated PL Shrinking ΔEc energy with increased Sn Bandgap mapped out for strain/Sn combinations Only ~7% Sn necessary for direct-bandgap! • GeSn is favorable for lasers! – Low density of states in Γ-valley reduces current to reach transparency optical gain at low carrier concentrations – Low carrier concentrations means reduced free-carrier absorption and Auger recombination, results in lowthreshold lasers Peking University Summer School, July 19, 2013 JSH 51 Summary STANFORD ● Strong quantum confined Stark effect and absorption shift observed in Ge/SiGe quantum well device ● Modulation demonstrated at 30 GHz & 100 GHz possible ● Waveguide modulator can be integrated into SiGe waveguides, eliminating alignment and coupling losses ● Both tensile strain and GeSn alloy will be required to achieve direct bandgap Ge and stimulated emission ● Photonic crystal or optical disks will be required to achieve high-Q cavities ● Strained Ge and GeSn/SiGeSn are all CMOS compatible Peking University Summer School, July 19, 2013 JSH 52 Acknowledgements STANFORD STUDENTS, POSTDOCS and COLLABORATORS Yu-Hsuan Kuo Shen Ren Theodore I. Kamins Yiwen Rong Jonathan E. Roth Marco Fiorentino Yijie Huo Elizabeth Edwards Michael R.T. Tan Hai Lin Rebecca Schaevitz Jae-Hoon Kim Yangsi Ge Onur Fidaner Lars Thylen Yong Kyu Lee Selcuk Yerci Guillaume Huyet Tomasz Ochalski Yiyang Gon Seongjae Cho Ed Fei Suyog Gupta Edris Mohammed Robert Chen Jelena Vuckovic Krishna Saraswat Colleen Shang Ian Young Mark Brongersma SUPPORT DARPA Intel SRC-IFC Thank You Peking University Summer School, July 19, 2013 JSH 53