Spike Sequence Learning in a Photonic Spiking Neural Network Consisting of VCSELs-SA With Supervised Training
Ziwei Song, Shuiying Xiang, Zhenxing Ren, Genquan Han, Yue Hao
Abstract
We propose a fully-connected photonic spiking neural network (SNN) consisting of excitable vertical-cavity surface-emitting lasers with an embedded saturable absorber (VCSELs-SA) to implement spike sequence learning by a supervised training. The photonic spike-timing-dependent plasticity (STDP) is incorporated into a classical remote supervised method (ReSuMe) algorithm to implement supervised training of a photonic SNN for the first time. The computation model of the photonic SNN is derived based on the Yamada model. To optimize the learning process, we further propose a novel measure, the so-called spike sequence distance, to quantitatively evaluate the effects of controllable parameters. The numerical results show that, the photonic SNN successfully reproduces a desirable output spike sequence in response to a spatiotemporal input spike pattern by means of the iteration algorithm to update synaptic weights continuously. These results contribute one step forward toward the device-algorithm co-design and optimization of the all-VCSELs-based energy-efficient photonic SNN.