Photonic neuromorphic computing: from device classification to system reality

Abstract

Integrated photonic neural networks promise to relieve the bandwidth, latency and energy limits of electronic artificial-intelligence hardware by computing directly with light. In a recent review, Han, Shen, Gu and Zhang organize the field around photonic synapses, neurons and memristors, and map these devices onto coherent, wavelength-parallel, diffractive and reservoir-computing architectures. This News & Views argues that the review is most valuable when read as both a device classification and a reminder that practical neuromorphic photonics must be judged at full-system scale.

FullText(HTML)