SNN Implementation Based on Systolic Arrays

As a neural network simulating the firing behavior of biological neurons, a spiking neural network (SNN) has the advantages of low power consumption, high efficiency and high fault tolerance, which is a vital way to realize brain-like computing. However, the implementation of SNN confronts many challenges, such as how to determine input coding, training methods and how to build network structure, hardware platform and so on. This paper presents an input coding scheme, network model and hardware implementation. The input coding scheme of SNN is based on Poisson coding, which can realize the transmission of information, effectively use time information, reduce redundant data, and improve the efficiency and robustness of coding. The network model is based on a three-layer network, consisting of an input layer, a hidden layer, and an output layer. The hardware implementation is based on Systolic arrays, which can realize large-scale parallel computing and improve the throughput and the speed of the hardware computing.