High-Yield and Uniform NbOx-Based Threshold Switching Devices for Neuron Applications

Abstract

Threshold switching (TS) devices based on NbOx materials show intriguing potential for constructing artificial neurons in a neuromorphic machine. However, the high electroforming voltage, the low TS yield, and the poor device uniformity hinder the practical application of NbOx-based TS devices. In this work, we systematically investigate the effect of film composition on device performance by adjusting the oxygen contents in the NbOx films. The electroforming voltage decreases with lowering the oxygen content, and the forming yield for activating TS behavior increases without an additional reset process. Moreover, we propose a stacked method by inserting a NbOy layer with high oxygen content between the low oxygen NbOx layer and the bottom electrode. The intercalated NbOy layer serves as a virtual bottom electrode after breakdown, enhancing the local electrical field and improving cycle-to-cycle stability and device-to-device uniformity. These results demonstrate that the device performances are greatly improved by optimizing the oxygen content and structure, guiding for practical applications of NbOx-based TS devices in neuromorphic computing.