Demonstration of advanced neuromorphic computing using low current memristive arrays

Abstract

As artificial intelligence technology using artificial neural networks develops, the development of new computing methods using memristors, which are resistive switching elements, is being promoted. However, not much research has been done to demonstrate the operation of memristors on an array basis. In this dissertation, a demonstration of the operation of a network based on a memristor crossbar array was demonstrated, and at the same time, an energy efficient network was constructed by developing an algorithm that mimics the brain. In addition, based on material properties, dynamic routings in artificial neural network topology were made, enabling advanced computing techniques to be directly operated in hardware.