Optical Neural Networks: Principles, Challenges, and Future Prospects in Computing and Astrophotonics

The rapid development of optical neural networks (ONNs) has led to the introduction of new research avenues for computing power enhancement. Because of the characteristics of optical signals, which include low power consumption, low latency, high parallelism, and large bandwidths, optical computing based on neural network architectures is showing promise for processing of spatial signals, temporal signals, and on-chip information. At present, there is a lack of a unified ONN computing architecture, and because of the limitations of the physical characteristics of these networks, different application scenarios have led to proposals of different requirements for the structural design, device selection, integration method, and signal processing method of the network. In this paper, we systematically elaborate on the practical value of ONNs, analyze their computational fundamentals in depth, discuss the challenges faced in computational and astrophotonics applications in detail, and simultaneously emphasize the important position and broad prospects of optical computing in the future information society.