An Enhanced Photonic Spiking Neural Network Based on the VCSEL-SA for Recognition and Classification

Neuromorphic computation based on physical devices has attracted wide attention due to its high performance in solving specific tasks. The chosen devices usually imitate the way the brain transmits information and calculates. However, the majority of existing encoding methods for neuromorphic computation rely on either rate or temporal information, resulting in inefficiency in accomplishing computing tasks. Here, we present an efficient and biologically plausible encoding method for a photonic spiking neural network (PSNN) based on vertical-cavity surface-emitting lasers with an embedded saturable absorber (VCSELs-SA). In the proposed method, the date information is encoded into the strength of the injection rectangular pulse and then converted into spikes emitted by VCSELs-SA. We can obtain timing and threshold characteristics by adjusting the injection intensity which is beneficial to process the data in the following steps. The encoding method is verified by recognizing four spiking patterns and four expressions. Furthermore, we classify the Iris data set based on the encoding method with only four pre-neurons and two post-neurons. The encoding method and structure further explore the application of PSNN to pave the way for the hardware implementation.