An Enhanced Encoding Method for Photonic Spiking Neural Network Based on the VCSEL-SA

The concept of neuromorphic computation, implemented by physical devices, has obtained significant attention owing to its exceptional performance in handling specific tasks. These devices have the characteristics of neuron-like output and can mimic the brain's methodology of information transmission and computation. However, a significant limitation lies in the prevalent encoding methods for neuromorphic computation, which primarily rely on either rate or temporal encoding, thereby causing inefficiencies in the execution of computational tasks. In this article, we introduce an effective and biologically inspired encoding approach for a photonic spiking neural network (PSNN), utilizing vertical-cavity surface-emitting lasers with an embedded saturable absorber (VCSELs-SA). Our method encodes data information by modulating the intensity of an injection rectangular pulse, which subsequently triggers the emission of spikes from the VCSELs-SA. The proposed PSNN can recognize four spiking patterns and classify the Iris dataset. What's more, we get an accuracy of 94.67% in the classification task with only six neurons.