Semiconductor lasers for photonic neuromorphic computing and photonic spiking neural networks: A perspective

Abstract

Photonic neuromorphic computing has emerged as a promising avenue toward building a high-speed, low-latency, and energy-efficient non-von-Neumann computing system. Photonic spiking neural network (PSNN) exploits brain-like spatiotemporal processing to realize high-performance neuromorphic computing. Linear weighting and nonlinear spiking activation are two fundamental functions of a SNN. However, the nonlinear computation of PSNN remains a significant challenge. Therefore, this perspective focuses on the nonlinear computation of photonic spiking neurons, including numerical simulation, device fabrication, and experimental demonstration. Different photonic spiking neurons are considered, such as vertical-cavity surface-emitting lasers, distributed feedback (DFB) lasers, Fabry–Pérot (FP) lasers, or semiconductor lasers embedded with saturable absorbers (SAs) (e.g., FP-SA and DFB-SA). PSNN architectures, including fully connected and convolutional structures, are developed, and supervised and unsupervised learning algorithms that take into account optical constraints are introduced to accomplish specific applications. This work covers devices, architectures, learning algorithms, and applications for photonic and optoelectronic neuromorphic computing and provides our perspective on the challenges and prospects of photonic neuromorphic computing based on semiconductor lasers.