Silicon Photonic Architecture for Training Deep Neural Networks with Direct Feedback Alignment

Abstract

There has been growing interest in using photonic processors for performing neural network inference operations; however, these networks are currently trained using standard digital electronics. Here, we propose on-chip training of neural networks enabled by a CMOS-compatible silicon photonic architecture to harness the potential for massively parallel, efficient, and fast data operations. Our scheme employs the direct feedback alignment training algorithm, which trains neural networks using error feedback rather than error backpropagation, and can operate at speeds of trillions of multiply-accumulate (MAC) operations per second while consuming less than one picojoule per MAC operation. The photonic architecture exploits parallelized matrix-vector multiplications using arrays of microring resonators for processing multi-channel analog signals along single waveguide buses to calculate the gradient vector for each neural network layer in situ. We also experimentally demonstrate training deep neural networks with the MNIST dataset using on-chip MAC operation results. Our novel approach for efficient, ultra-fast neural network training showcases photonics as a promising platform for executing AI applications.

arXiv comment15 pages, 6 figures

References in corpus (27)

• Deep Learning with Coherent Nanophotonic Circuits
• Parallel convolution processing using an integrated photonic tensor core
• Photonics for artificial intelligence and neuromorphic computing
• 11 TeraFLOPs per second photonic convolutional accelerator for deep learning optical neural networks
• An Analysis of Deep Neural Network Models for Practical Applications
• Neuromorphic Silicon Photonic Networks
• Convolutional Networks for Fast, Energy-Efficient Neuromorphic Computing
• Single-chip photonic deep neural network for instantaneous image classification
• Brain-inspired computing: We need a master plan
• Training of photonic neural networks through in situ backpropagation
• All Optical Neural Network with Nonlinear Activation Functions
• A one femtojoule athermal silicon modulator
• A Survey of Neuromorphic Computing and Neural Networks in Hardware
• A silicon photonic modulator neuron
• Adding Gradient Noise Improves Learning for Very Deep Networks
• Digital Electronics and Analog Photonics for Convolutional Neural Networks (DEAP-CNNs)
• Direct Feedback Alignment Provides Learning in Deep Neural Networks
• Silicon microring synapses enable photonic deep learning beyond 9-bit precision
• Wavelength tuning and stabilization of microring-based filters using silicon in-resonator photoconductive heaters
• Post-Fabrication Trimming of Silicon Photonic Ring Resonators at Wafer-Scale
• Scaling Up Silicon Photonic-based Accelerators: Challenges and Opportunities
• Fully reconfigurable coherent optical vector-matrix multiplication
• All-silicon light-emitting diodes waveguide-integrated with superconducting single-photon detectors
• Deep neural networks are robust to weight binarization and other non-linear distortions
• Photoconductive Heaters Enable Control of Large-Scale Silicon Photonic Ring Resonator Circuits
• Quantifying power use in silicon photonic neural networks
• Hardware Beyond Backpropagation: a Photonic Co-Processor for Direct Feedback Alignment

Cited by in corpus (11)

• Experimentally realized in situ backpropagation for deep learning in nanophotonic neural networks
• Photonics for Neuromorphic Computing: Fundamentals, Devices, and Opportunities
• Scaling Up Silicon Photonic-based Accelerators: Challenges and Opportunities
• Training neural networks with end-to-end optical backpropagation
• Ultra-high endurance silicon photonic memory using vanadium dioxide
• Training Coupled Phase Oscillators as a Neuromorphic Platform using Equilibrium Propagation
• Mirage: An RNS-Based Photonic Accelerator for DNN Training
• Asymmetrical estimator for training encapsulated deep photonic neural networks
• Self-Contrastive Forward-Forward Algorithm
• Online training and pruning of multi-wavelength photonic neural networks
• Streamlined optical training of large-scale modern deep learning architectures with direct feedback alignment