Leveraging integrated photonics to implement AI and computing tasks has been attracting wide attention
due to the ever-growing demand on low-latency and low-power computing resources for AI and machine
learning workloads in data centers. This offers unprecedented opportunities, as well as challenges, for
currently accessible photonic technologies. Different from high-speed transceivers for optical interconnects
where controversy remains for selecting different technology solutions, it reaches extensive consensus
that silicon photonic process technology will be the most promising solution for optical computing
from several aspects such as large scale, integration with electronics, reconfigurability,
interoperability, and miniaturization. In this review, we first introduce our work on utilizing silicon photonic
circuits to perform novel classification computing and its recent extension to electro-photonic dynamic
system, and then discuss our perspectives on the importance, robustness, and challenges of silicon
photonic process technology to implement optical computing.
View full abstract