SPICE Simulation of the Neuromorphic System Composed of Neuron Circuit and Synaptic Device

Abstract

Artificial intelligence, such as voice recognition, face recognition, and autonomous motion that can artificially replace the human brain has caused a big issue, and the artificial intelligence industry in the future will bring about a big change in the global trend. However, computing system based on the existing von Neumann structure and artificial intelligence implementation through it show problems in terms of power consumption and efficiency. Especially, when performing higher-order operations, von Neumann bottleneck appears when processing large amounts of data due to the large power consumption and inefficiency which caused the emergence of a new artificial intelligence system. To confirm the efficiency of the neuromorphic system, Spiking-Neural-Network system is closely investigated by using synaptic device and neuron circuit. In the SPICE simulation, neuromorphic system feature equivalent output with artificial neural network MATLAB simulation when performing MNIST pattern recognition test. This means any software-based artificial intelligence can be implemented in hardware. Furthermore, in order to enhance accuracy of the MNIST pattern recognition, overflow retaining neuron circuit is proposed. Through a simple circuit structure change, virtual membrane node is properly operated with minimizing the wasted signals differently from conventional integrate and fire neuron circuit. It is confirmed by comparing the raster plot extracted from the MATLAB simulation and SPICE simulation output which implicit the hardware implementation of all other artificial neural network.