Abstract

Inspired by the low energy consumption of the human brain, neuromorphic engineering aims to develop novel hardware to aid the development of machine learning algorithms [1]. Integrated photonics is regarded as a formidable candidate to develop a low-power, ultra-fast optical neural network, due to the high bandwidth, low crosstalk [2] and temporal characteristics governed by fast carrier dynamics. In an all-optical photonic spiking neural network (SNN), a laser neuron is triggered by an optical pulse of sufficient energy generated by previous neurons. From a dynamics perspective this thresholding functionality is an excitable response to an input perturbation and can be realized through integrated two-section lasers [3].

© 2023 IEEE