Digital-analog hybrid matrix multiplication processor for optical neural networks

Abstract The computational demands of modern AI have spurred interest in optical neural networks (ONNs) which offers the potential benefits of increased speed and lower power consumption. However, current ONNs running analog optical signals face a fundamental limitation in calculation precision, typically around 4 bits, due to accumulated noise from electro-optical components. This obstacle is inherent to the analog computing nature of ONNs. The analog signals representing the inputs and the computational results require high-resolution signal format converters (digital-to-analogue conversions (DACs) and analogue-to-digital conversions (ADCs)), which pose a major obstacle 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 is significantly enhanced. The DACs for inputs are removed. And the resolution requirement of the ADCs becomes independent of the bitwidth of the inputs. This can increase the operating speed at a high calculation precision and facilitate 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 1.8×10 −3 at a 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.