Experimental Demonstration of Delay-Weight Learning and Pattern Classification With a FP-SA-Based Photonic Spiking Neuron Chip

Optical acceleration of neuromorphic computing has emerged as a platform for achieving low-latency and energy-efficient computation. Due to the rich dynamics, the integrated photonics architecture provides a promising way for implementations of nonlinear computing. In this article, we demonstrate pattern classification of 2 benchmark datasets, the Iris dataset and the Wisconsin Breast Cancer (WBC) dataset, based on the nonlinear neuron-like dynamics of an integrated Fabry-Perot laser chip with a saturable absorption region (FP-SA). The effectiveness and robustness of the algorithm are firstly verified by numerical simulations. With the help of an optimized delay learning method, efficient learning can be achieved based on a single neuron, achieving 96% and 92% in classification accuracy of Iris and WBC dataset respectively. Then, the hardware-algorithm collaborative computing is demonstrated based on a single FP-SA laser chip. The classification accuracy of Iris and WBC dataset could reach 94.67% and 88%, respectively, having a relative low loss compared to the simulation results. This work provides an efficient solution for classification tasks based on optical SNN.