Hardware Implementation of Multi-Layer Photonic Spiking Neural Network With Three Cascaded Photonic Spiking Neurons

Abstract

Photonic technologies offer great prospects for novel ultrafast, energy-efficient and hardware friendly neuromorphic computing platforms. Moreover, Photonic spiking neural network (PSNN) emerged as a promising approach toward building a low-latency and energy-efficient non-von-Neuman computing system is of particular interest. In this paper, a fabricated Fabry–Pérot laser with saturable absorber (FP-SA) is employed as the photonic spiking neuron of the PSNN, and a multi-layer PSNN with such three cascaded photonic spiking neurons is proposed. The cascadability of the proposed multi-layer PSNN is demonstrated experimentally. It is found that the performance of the pattern recognition task using the hardware-algorithm collaborative computing is improved in the multi-layer PSNN. Besides, the nonlinear neuron-like dynamics including temporal integration and threshold in such multi-layer PSNN are also experimentally investigated, and better nonlinear neuron-like dynamics can be achieved. Such proposed multi-layer PSNN potentially opens up the prospect of employing simple hardware structure and cooperating with algorithm design to realize a photonic depth neural network to solve more complex problems.