A Visual Bionic Neural Encoder Based on NPC Memristor for Spiking Neural Network

As a new generation of neural networks, neuron-based spiking neural network (SNN) neuromorphic computing has become the focus of research, which utilizes biological plausible neurons to provide faster inference and lower energy consumption. In the hardware implementation of the spiking neural network, a hardware encoder part similar to the sensory neuron is needed to convert external stimuli into pulsed signals (spike train) according to the neural algorithm and the corre-sponding learning rules. Unfortunately, most traditional Spiking neural network circuits use digital circuit technology as the input, but most of the information in the real world is in the form of analog information, so neural encoders need to be developed to meet the growing demand for neuromorphic computation. This paper presents a hardware encoding design for neurons based on a negative photoconductive memristor that converts optical signals into count-based pulsed neural encoding. Finally, the proposed bionic encoding device is used to encode the MNIST dataset, and then the trained SNN is used to make an accuracy judgment of over 97.38%. Based on this, a bionic hierarchical visual perception system is designed.