Threshold voltage controllable high-order anodized NbO$_x$ Mott memristor and charge trap HfO$_2$ memcapacitor for neuromorphic computing

Abstract

With the advancement of machine learning technologies, the demand to efficiently process large amounts of data in parallel is increasing. However, because the current von Neumann computing architecture has separate memory and processing units, a bottleneck phenomenon occurs and data parallel processing is limited. Therefore, there is a growing need for the development of new computing hardware that enables efficient computation of large amounts of data. Among them, neuromorphic computing, implementing artificial neural networks through analogue computation, is gaining significant attention. To achieve this, the development of high-performance synapses and neuron components is essential. This work presents the implementation of high-order neuron devices by adjusting the height of threshold switching (TS) region in anodized NbO$_x$ Mott memristor. Through this device, the threshold variability and spike adaptation of neurons were implemented. Furthermore, this work demonstrates the utilization of trap sites at the hafnium dioxide-silicon interface as capacitive synaptic devices. Additionally, low-power MAC operation with no static power consumption was implemented through these devices.