Memristor-based Neuromorphic Implementations for Artificial Neural Networks

Remarkable computing complexity and power is obtained by combining multiple neurons and synapses computational units into high integrated brain-like network system. In order to simulate the biological learning rules in artificial synapses, an artificial neural network (ANN) capable of performing complicated functions is constructed. A significant challenge is to design a circuit that maintains a simple neuron structure, occupies a small silicon area, and implement only one electronic device as an artificial synapse. However, in traditional electronics, the area of silicon occupied by synaptic circuits can vary significantly. As observed and demonstrated, memory resistors exhibit such characteristics, making them the most promising candidates in scalable neural networks. Findings of various memristor network structures that support artificial neural network classification or information storage functions are presented.