Is Computing with Light All You Need? A Perspective on Codesign for Optical Artificial Intelligence and Scientific Computing

Abstract

Anything and everything can be solved with light—but under what contexts should one use optical computing and how should optical systems be designed for these tasks? In the shadow of reaching physical limits to traditional computer scaling and the large demands from artificial intelligence, research efforts have focused upon a range of optical computing pursuits as of late. This has included advances across a range of approaches, including free-space and on-chip implementations. Amid this excitement, key considerations beyond how to generally harness optical principles to enable computation, are what computations to advantageously pursue and how to design optical systems for these tasks. In particular, this perspective considers free-space optical computing informed by recent research findings to consider select topics, including scene classification, integral differential equations, and many-body simulations. In these contexts, this perspective considers what computations optical computing can and should enable and argues that a codesign approach whereby materials, devices, architectures, and algorithms are simultaneously optimized is needed considered for best performance.