Bioplausible Photonic-Electronic Neural Networks

Photonic-electronic neural networks employ photonics for high-bandwidth communication and electronics for computing. In this paper we present a programmable photonic-electronic architecture for bioplausible neuromorphic computing to achieve 775±15 spikes/inference corresponding to 915 pJ/inference and 28.6 pJ/inference in the current GF45SPCLO and future 7nm designs, respectively.