Synaptic variation reduction via embedding Au nanocrystals in resistive switching layer and bottom electrode interface for CuTe/CuO/TiN-stacked synaptic device

Abstract

Reliable artificial synaptic devices are essential for the stable and fast training of artificial neural networks (ANNs). Specifically, synaptic devices should be robust during the training and testing of ANNs to embed them in the hyper-scale synaptic cores of neuromorphic computing architectures. In this study, a highly reliable artificial synaptic device based on a CuTe/CuO/TiN-stacked conductive-bridge random-access memory cell having forming-free property was developed via embedding Au nanocrystals in the CuO resistive switching layer and TiN bottom electrode interface. Forming-free property was achieved by precisely designing the diameter of Au nanocrystals implementing the interface between the CuO resistive switching layer and TiN bottom electrode. In particular, this synaptic device exhibited multilevel current states when the compliance current level was varied. In addition, the synaptic device embedding Au nanocrystals (i.e., ~ 17.7 nm in diameter) showed a remarkable reduction of the variation in synaptic modulation. Furthermore, the test accuracy of image recognition via a deep neural network simulation was dramatically improved up to 91.95% using practical synaptic modulation data of the synaptic device embedding Au nanocrystals (i.e., ~ 17.7 nm in diameter).