Photonic Associative Learning Neural Network Based on VCSELs and STDP

Abstract

In this article, we propose a photonic neural network (PNN) to emulate associative learning and forgetting for the first time. The PNN contains three photonic neurons based on 1550nm long-wavelength vertical-cavity surface-emitting lasers (VCSELs) subject to double polarized optical injection. A computational model of the PNN is derived based on the well-known spin-flip model. According to the learning rule named spike-time-dependent plasticity (STDP), the connection weight between neurons in the PNN is modified by self-learning. The simulation results show that, during the associative learning process, an association of feeding food and bell ringing is established within the Long-term synaptic potentiation window of the STDP; while, the association is forgotten within the Long-term depression window of the STDP in the forgetting process. Moreover, the effects of time interval between two pre-synaptic spikes on the speed of associative learning and forgetting are considered in detail. Finally, we further realize pattern recall based on the photonic associative learning, which shows promises for emerging applications of a large-scale energy-efficient PNN.