Silicon Photonics for High-Performance Computing and Beyond

Cover
Half Title
Title Page
Copyright Page
Contents
Preface
	References
Editors
Contributors
Section 1: High-Performance Computing Interconnect Requirements and Advances
1. Silicon Photonic Modulation for High-Performance Computing
	1.1 Introduction
	1.2 A Board Problem with an Optical Solution
	1.3 Petabyte Optical Interconnection
	1.4 Multiplexing
		1.4.1 Time and Space Division Multiplexing
		1.4.2 Wavelength Division Multiplexing
	1.5 Generating Multiple Optical Carriers
		1.5.1 Optical Spectral-Slicing
		1.5.2 Noise in a Slice from an SLED
	1.6 Optical Comb Generation
		1.6.1 Properties of an Optical Comb
		1.6.2 Generating Combs from Lasers and RF Sources
			1.6.2.1 Single-Spaced Optical Frequency Comb Generators
			1.6.2.2 Double-Spaced Comb Generation
		1.6.3 General Model of a Comb Generator
	1.7 Modeling a Transmitter in a High-Speed Optical Interconnection
	1.8 Summary
	References
2. Laser Modulation Schemes for Minimizing Static Power Dissipation
	2.1 Introduction
	2.2 Pros and Cons of On-Chip Optical Networks
	2.3 Background
		2.3.1 Optical Devices
		2.3.2 Optical Buses
	2.4 Laser Modulation in CPUs
		2.4.1 Probe
		2.4.2 ColdBus
		2.4.3 PShaRe
			2.4.3.1 Basic Architecture
			2.4.3.2 Prediction
			2.4.3.3 Reconfiguring the Network
			2.4.3.4 Tuning Network
		2.4.4 BigBus
			2.4.4.1 Architecture
			2.4.4.2 Prediction
	2.5 Laser Modulation in Multi-Socket Systems
	2.6 Laser Modulation in GPUs
		2.6.1 Architecture
		2.6.2 Prediction Mechanism
		2.6.3 Reconfiguration
	2.7 Conclusion
	References
3. Scalable Low-Power High-Performance Optical Network for Rack-Scale Computers
	3.1 Introduction
	3.2 Related Work
	3.3 Architecture
		3.3.1 RSON Architecture Overview
		3.3.2 FODON Switch Design
		3.3.3 Preemptive Chain Feedback Control Scheme
	3.4 Evaluations
		3.4.1 Evaluation Setup and Methodology
		3.4.2 Performance Evaluation
		3.4.3 System Energy Consumption
		3.4.4 Optical Energy Efficiency
		3.4.5 System Performance per Energy
	3.5 Conclusions
	References
4. Network-in-Package for Low-Power and High-Performance Computing
	4.1 Data Movement in Advanced Computing Systems
	4.2 Wideband Wireline Communications
	4.3 Inter-symbol Interference
	4.4 Choice of Signaling Method
	4.5 OMWS
	4.6 Transceiver Architecture
	4.7 Analog Encoder and Decoder Circuits
	4.8 OMWS Issues
	4.9 Summary and the Future Directions
	Acknowledgment
	References
Section II: Device- and System-Level Challenges and Improvements
5. System-Level Management of Silicon-Photonic Networks in 2.5D Systems
	5.1 Introduction
	5.2 Thermal and Process Sensitivities of Optical Devices
		5.2.1 Thermal Variations (TV) Induced MRR Resonance Shifts
		5.2.2 Process Variations (PV) Induced MRR Resonance Shifts
	5.3 Device- and Design-Level Techniques to Mitigate MRR Resonance Shifts
		5.3.1 Device-Level Techniques
		5.3.2 Design-Level Techniques to Mitigate MRR Resonance Shifts
	5.4 System-Level Management Techniques
	5.5 Conclusion
	Acknowledgment
	References
6. Thermal Reliability and Communication Performance Co-optimization for WDM-Based Optical Networks-on-Chip
	6.1 Introduction
	6.2 Chapter Overview
	6.3 Thermal Sensitivity of Micro-ring Resonators Under Wavelength Tuning Mechanism
	6.4 PV-OTS: PV-Tolerant Optical Temperature Sensor
	6.5 Thermal Reliable ONoCs
		6.5.1 Methods for Reliable Optical Communication
		6.5.2 Routing Criterion: I/L/Z-Shaped Routing Path
	6.6 Conflict-Avoidance Routing Approaches
		6.6.1 Problem Formulation
			6.6.1.1 Communication in ONoC
			6.6.1.2 Problem Definition
		6.6.2 MILP-Based Routing Approach
		6.6.3 Conflict-Aware Routing Heuristic Algorithm
	6.7 Performance Evaluation
		6.7.1 Accuracy of the Thermal Sensitivity Model
		6.7.2 Performance of the PV-OTS Design
		6.7.3 Effectiveness of Our Routing Approaches
	6.8 Conclusion
	Acknowledgment
	References
7. Exploring Aging Effects in Photonic Interconnects for High-Performance Manycore Architectures
	7.1 Introduction
	7.2 VBTI Aging in Microrings (MRs)
		7.2.1 Analytical Models for VBTI Aging Mechanism
	7.3 Impacts of VBTI Aging
		7.3.1 Impacts of VBTI Aging on MRs' Resonance Characteristics
		7.3.2 Impacts of VBTI Aging on DWDM-Based OOK Links
			7.3.2.1 VBTI Aging in Modulator MRs of Source Node
			7.3.2.2 VBTI Aging in Receiver MRs of Destination Node
	7.4 Mitigating the Impacts of VBTI Aging
		7.4.1 Reactive Mitigation of VBTI Aging Impacts
		7.4.2 4-PAM Signaling: A Technique for Proactive Mitigation of VBTI Aging Impacts
	7.5 Evaluation
		7.5.1 CLOS PNoC Architecture
		7.5.2 Evaluation Setup
		7.5.3 Modeling of Fabrication Process Variations in MRs
		7.5.4 Modeling of Signal Degradation-Related Power Penalty
		7.5.5 Evaluation Results
	7.6 Conclusions
	References
8. Improving Energy Efficiency in Silicon Photonic Networks-on-Chip with Approximation Techniques
	8.1 Introduction
	8.2 Related Work
	8.3 Background: Floating Point Format
	8.4 Lorax Framework: Overview
		8.4.1 Loss-Aware Laser Power Management
		8.4.2 Integrating Multilevel Signaling for Approximation
	8.5 Experiments
		8.5.1 Experimental Setup
		8.5.2 Application-Specific Approximation Sensitivity Analysis
		8.5.3 Comparative Results for Laser Power and EPB
	8.6 Conclusions
	References
Section III: Novel Design Solutions and Automation
9. Automated, Scalable Silicon Photonics Design and Verification
	9.1 History/Context of Traditional IC Design and Fabless Model
	9.2 Process Design Kits
	9.3 Electronic PDKs
		9.3.1 Layer Map
		9.3.2 Characterized Devices/Device Models and Symbols
		9.3.3 Pre-characterized Cells
		9.3.4 Intellectual Property
		9.3.5 Design Rules and Technology Files
	9.4 Physical and Electrical Verification
		9.4.1 Design Rule Checking
		9.4.2 Layout vs. Schematic Verification
		9.4.3 Antenna Checks
		9.4.4 Parasitic Extraction
		9.4.5 Electrical Rule and Reliability Checking
		9.4.6 Design for Manufacturing
	9.5 A PDK for Integrated Silicon Photonics
		9.5.1 Pre-characterized Devices
		9.5.2 Layout Implementation
	9.6 Automated Photonics Design Methodologies
		9.6.1 Layout-Centric Design
		9.6.2 Schematic-Based Design
		9.6.3 Schematic-Driven Layout
		9.6.4 Automated Layout
			9.6.4.1 Scripted
			9.6.4.2 Schematic-Driven
			9.6.4.3 Test Chip
	9.7 Silicon Photonics Verification
	9.8 PIC Physical Layout Verification
	9.9 Best Practices for Photonics DRC
	9.10 PIC LVS Verification
	9.11 Best Practices for Photonics LVS
	9.12 Litho-Friendly Design Simulation
	9.13 Conclusion
	References
10. Inverse-Design for High-Performance Computing Photonics
	10.1 System-Level Motivation
	10.2 Components of an Optical Interconnect
	10.3 Inverse-Design Method Formulation
	10.4 Practical Photonic Devices with Inverse-Design
	10.5 Outlook of Inverse-Design for High-Performance Computing
	References
	Appendices
	A10.1 Estimation of Optical Interconnect System-Level Benefit
		A10.1.1 Memory System Analysis
		A10.1.2 LSTM Application System Analysis
	A10.2 Adjoint Method Derivation
11. Efficiency-Oriented Design Automation Methods for Wavelength-Routed Optical Network-on-Chip
	11.1 Introduction
	11.2 WRONoC Design Criteria
		11.2.1 Common Setting and General Design Rules
		11.2.2 Wavelength Usage
		11.2.3 MRR Usage
	11.3 Design Automation for WRONoCs
		11.3.1 Necessity of Efficient Design Automation Methods for WRONoCs
		11.3.2 Subtraction from Fully Connected Router
		11.3.3 Template-Based Synthesis
	11.4 Conclusion
	References
Section IV: Novel Materials, Devices, and Photonic Integrated Circuits
12. Innovative DWDM Silicon Photonics for High- Performance Computing
	12.1 Introduction
	12.2 Integrated Dense Wavelength Division Multiplexing Architecture
	12.3 Advanced Silicon Photonic Building Blocks
		12.3.1 Multi-Wavelength Lasers
		12.3.2 Efficient Phase Tuner and Modulators
		12.3.3 Robust Si-Ge Avalanche Photodetectors
		12.3.4 High-Gain Heterogeneous Quantum-Dot Avalanche Photodetectors
	12.4 Innovative Integration Platform Development
	12.5 Advanced Wafer-Level Testing and Analysis
	12.6 Novel Fiber Attachment Solution
	12.7 Summary
	References
13. Silicon Photonic Bragg Grating Devices
	13.1 Introduction
	13.2 Integrated Waveguides Bragg Gratings
		13.2.1 Bragg Gratings Theory
		13.2.2 Bragg Gratings Design and Fabrication
			13.2.2.1 Manufacturing Variability
			13.2.2.2 Lithography Effects
		13.2.3 Applications of Bragg Gratings
			13.2.3.1 Hybrid-Integrated Lasers
			13.2.3.2 Biosensors
			13.2.3.3 Dispersion Compensation
	13.3 Contra-Directional Couplers
		13.3.1 Contra-Directional Couplers' Theory
		13.3.2 Applications of Contra-Directional Couplers
			13.3.2.1 Wavelength-Division-(de)-Multiplexer
			13.3.2.2 Broadband Optical Add-Drop (de)-Multiplexers
	13.4 Conclusion
	Acknowledgments
	References
14. Silicon Photonic Integrated Circuits for OAM Generation and Multiplexing
	14.1 Introduction
	14.2 Grating-Assisted Ring-Resonators
		14.2.1 Principle
		14.2.2 Demonstrations
	14.3 Phased Antenna Arrays
		14.3.1 Principle
		14.3.2 Generating Circularly Polarized OAM Modes
		14.3.3 Free-Space Propagation
		14.3.4 Demonstrations
	14.4 In-Plane Mode Convertor
	14.5 Waveguide Surface Holographic Gratings
	14.6 Challenges and Outlook
	References
15. Novel Materials for Active Silicon Photonics
	15.1 Motivation and Overview
		15.1.1 Active Silicon Photonics
		15.1.2 Overview of the Chapter
	15.2 Barium Titanate Electro-Optic Modulators
		15.2.1 Electro-Optic Properties of BaTiO3
			15.2.1.1 Electro-Optic Coefficients
			15.2.1.2 Multi-Domain Structure
		15.2.2 The Waveguide Design: Partially Etched Horizontal Slot Waveguides
			15.2.2.1 Design of the Horizontal Slot Waveguides
			15.2.2.2 On-Chip Grating Couplers
			15.2.2.3 Mach-Zehnder Interferometers
			15.2.2.4 Optical Resonators
		15.2.3 Theoretical Considerations of BTO Modulators
			15.2.3.1 Transmission Loss
			15.2.3.2 Absorption Loss From the Gold Electrodes
			15.2.3.3 Bend Loss
		15.2.4 Electro-Optical Properties
			15.2.4.1 Dependence of the Δn on the Waveguide Width
			15.2.4.2 Dependence of the Δn on the Thickness of the BTO Layer
			15.2.4.3 Dependence of the Δn on the Thickness of the Poly-Silicon Layer
		15.2.5 Material Growth, Characterization and Device Fabrication
			15.2.5.1 Epitaxial Growth
			15.2.5.2 Oxygen Annealing
			15.2.5.3 Device Fabrication
		15.2.6 Experimental Results
			15.2.6.1 Grating Couplers
			15.2.6.2 Micro-Ring Resonators
			15.2.6.3 Mach-Zehnder Interferometers
			15.2.6.4 Frequency Response of the BTO Electro-Optic Modulators
		15.2.7 Summary
	15.3 Integrated Photonic Circuits in Gallium Nitride and Aluminum Nitride
		15.3.1 Group III-Nitride Semiconductors as Optical Materials
		15.3.2 Harmonic Generation in GaN Photonic Circuits
			15.3.2.1 GaN Photonic Circuits
		15.3.2.2 Second and Third Harmonic Generation from GaN Photonic Circuits
		15.3.3 AlN Photonic Circuits and Electro-Optic Modulators
			15.3.3.1 AlN Photonic Circuits
			15.3.3.2 AlN Microring Electro-Optic Modulators
			15.3.3.3 Slow Light in AlN 2D Photonic Crystals
			15.3.3.4 AlN One-Dimensional Nanobeam Photonic Crystal Cavity
	15.4 Chapter Summary
	References
Section V: Emerging Computing Technologies and Applications
16. Neuromorphic Silicon Photonics
	16.1 Introduction
	16.2 Silicon Photonic Neurons
		16.2.1 Multiply-Accumulate Operation
		16.2.2 Nonlinear Transformation
	16.3 Silicon Photonic Neural Networks and Applications
		16.3.1 Application I: Neural ODE Solver
		16.3.2 Application II: Nonlinear Programming and Model-Predictive Control
		16.3.3 Application III: Intelligent Signal Processing
	16.4 Conclusion and Future Directions
	References
17. Logic Computing and Neural Network on Photonic Integrated Circuit
	17.1 Introduction of Optical Computing
	17.2 Optical Logic Synthesis
		17.2.1 Optimization Techniques
		17.2.2 Simulation Results
	17.3 Exploiting Waveguide Division Multiplexing
		17.3.1 Proposed Synthesis Flow
		17.3.2 Simulation Results
	17.4 Area-Efficient Optical Neural Networks
		17.4.1 Background of Optical Neural Networks
		17.4.2 Slimmed Optical Neural Network
		17.4.3 Simulation Results
	17.5 Conclusion and Future Directions
	References
18. High-Performance Programmable MZI-Based Optical Processors
	18.1 Introduction and Motivation
	18.2 Theory and Analysis
		18.2.1 2 × 2 Reconfigurable MZI
	18.3 4 × 4 Reck MZI-Based Optical Processor
		18.3.1 Programming the 4 × 4 Reck-Based Optical Processor
		18.3.2 An Example for the Decomposition of an Arbitrary Unitary Transformation Matrix U(4)
	18.4 The Diamond Mesh, a Phase-Error- and Loss-Tolerant MZI-Based Optical Processor for ONNs
		18.4.1 Diamond Topology and Its Programming Process
		18.4.2 Optical Neural Networks for Classification
		18.4.3 4 × 4 Diamond-Topology- and Reck-Topology-Based Single Layer ONNs
		18.4.4 Scalability Investigation of the Reck and Diamond Topologies in Single Layer ONNs
	18.5 Conclusion
	References
19. High-Performance Deep Learning Acceleration with Silicon Photonics
	19.1 Introduction
	19.2 Related Work
	19.3 Noncoherent Photonic Computation Overview
	19.4 CrossLight Architecture
		19.4.1 MR Device Engineering and Fabrication
		19.4.2 Tuning Circuit Design
		19.4.3 Architecture Design
			19.4.3.1 Decomposition Vector in CONV/FC Layers
			19.4.3.2 Vector Dot Product (VDP) Unit Design
			19.4.3.3 Optical Wavelength Reuse in VDP Units
	19.5 Evaluation and Simulation Setup
		19.5.1 Simulation Setup
		19.5.2 Results: CrossLight Resolution Analysis
		19.5.3 Results: CrossLight Sensitivity Analysis
		19.5.4 Results: Comparison with State-of-the-Art Accelerators
	19.6 Conclusion
	References
Index