Digital Implementation of the Spiking Neural Network and Its Digit Recognition

Abstract

Motivated by biological principles of neural systems, spiking neural network (SNN) shows a tremendous potential in solving pattern recognition and cognitive tasks in recent years. In this study, a biologically inspired SNN composed of three layers is implemented on a reconfigurable FPGA with high computational efficiency and low hardware cost. The proposed SNN is consists of spiking neurons simulated by leaky-integrate-and-fire neuron model. In addition, spiking-time-dependent-plasticity based on event-driven is utilized to train the constructed network. The real-time hardware realization of the proposed SNN demonstrates powerful and efficient learning scheme. Results on different datasets shows that the proposed SNN implementation has the merit of capability of coping with pattern recognition tasks. Furthermore, the proposed implementation with remarkable performance could be applied and embed in bio-inspired neuromorphic platform such as robots for recognition tasks and on-line applications.