Abstract:The computational demands of modern AI have spurred interest in optical neural networks (ONNs) which offer the potential benefits of increased speed and lower power consumption. However, current ONNs face various challenges,most significantly a limited calculation precision (typically around 4 bits) and the requirement for high-resolution signal format converters (digital-to-analogue conversions (DACs) and analogue-to-digital conversions (ADCs)). These challenges are inherent to their analog computing nature and pose significant obstacles in practical implementation. Here, we propose a digital-analog hybrid optical computing architecture for ONNs, which utilizes digital optical inputs in the form of binary words. By introducing the logic levels and decisions based on thresholding, the calculation precision can be significantly enhanced. The DACs for input data can be removed and the resolution of the ADCs can be greatly reduced. This can increase the operating speed at a high calculation precision and facilitate the compatibility with microelectronics. To validate our approach, we have fabricated a proof-of-concept photonic chip and built up a hybrid optical processor (HOP) system for neural network applications. We have demonstrated an unprecedented 16-bit calculation precision for high-definition image processing, with a pixel error rate (PER) as low as $1.8\times10^{-3}$ at an signal-to-noise ratio (SNR) of 18.2 dB. We have also implemented a convolutional neural network for handwritten digit recognition that shows the same accuracy as the one achieved by a desktop computer. The concept of the digital-analog hybrid optical computing architecture offers a methodology that could potentially be applied to various ONN implementations and may intrigue new research into efficient and accurate domain-specific optical computing architectures for neural networks.

Towards Constructing a DOE-based Practical Optical Neural System for Ship Recognition in Remote Sensing Images

Optical Analog Computing Devices Designed by Deep Neural Network

On-chip 4F-System Based on Concave Mirrors for Optical Neural Networks

Integrated Diffractive Optical Neural Network with Space-Time Interleaving

On-chip Photonic Diffractive Optical Neural Network Based on a Spatial Domain Electromagnetic Propagation Model.

An Optical Satellite Controller Based on Diffractive Deep Neural Network.

Scientific Computing with Diffractive Optical Neural Networks

A Compact Butterfly-Style Silicon Photonic–Electronic Neural Chip for Hardware-Efficient Deep Learning

Space-efficient optical computing with an integrated chip diffractive neural network

Optical neural networks: progress and challenges

Multi-Dimensional Reconfigurable, Physically Composable Hybrid Diffractive Optical Neural Network

Neuromorphic Computing Network for Underwater Image Enhancement and Beyond

A multichannel optical neural network for advanced machine vision

Digital-analog hybrid matrix multiplication processor for optical neural networks

Integrated multi-operand optical neurons for scalable and hardware-efficient deep learning

Optical-electronic neural network for multi-modality and high-accurate lensless optics design and image reconstruction

Free-Space Optical Spiking Neural Network

LightRidge: An End-to-end Agile Design Framework for Diffractive Optical Neural Networks

Multimodal deep learning using on-chip diffractive optics with in situ training capability

Strategies for Training Optical Neural Networks

High speed and reconfigurable optronic neural network with digital nonlinear activation