Hybrid photonic integrated circuits for neuromorphic computing [Invited]

Abstract

The burgeoning of artificial intelligence has brought great convenience to people’s lives as large-scale computational models have emerged. Artificial intelligence-related applications, such as autonomous driving, medical diagnosis, and speech recognition, have experienced remarkable progress in recent years; however, such systems require vast amounts of data for accurate inference and reliable performance, presenting challenges in both speed and power consumption. Neuromorphic computing based on photonic integrated circuits (PICs) is currently a subject of interest to achieve high-speed, energy-efficient, and low-latency data processing to alleviate some of these challenges. Herein, we present an overview of the current photonic platforms available, the materials which have the potential to be integrated with PICs to achieve further performance, and recent progress in hybrid devices for neuromorphic computing.