Stacked cross-point memory using IGZO thin film for synaptic elements

Abstract

We conducted research and development of a hardware neural network by using oxide semiconductors of α-In-Ga-Zn-O (IGZO). We have made a 2 layers cross-point type device that is a hardware neural network that uses electrical characteristics of IGZO as a synapse. It is possible to enable parallel computing, high operating speed, low power consumption, high integration, and robustness, that were not possible with software. By the experiments, we evaluated the condition closest to the synaptic characteristics and found the optimal learning voltage. We also made a 3 layers cross-point type device that is a hardware neural network with the aim of achieving higher integration. These studies will enable highly integrated hardware neural networks of two or three layers.