Predictive Design of Threshold Switching Voltage in Electron Doped VO2 for Neuromorphic Computing

Vanadium dioxide undergoes a insulator-metal transition (IMT) in the vicinity of room temperature. In two-terminal devices, the phase transition can be initiated by applying bias voltage that results in a threshold switching behaviour. Such devices are of interest for use as artificial neurons and synapses in neuromorphic computing hardware. The transition temperature can be reduced from nearly 340K in pure crystals to 300K by choice of electron dopants such as tungsten. Understanding the atomic mechanisms of tuning the transition temperature with dopants is an active area of study as this has significant implications for design of low power switches for artificial neurons in neuromorphic computing. Here, we will first present our results on the rate of decrease of the transition temperature with increasing tungsten concentration and compare against experimental studies on threshold switching voltage scaling with doping. We find the slopes are quite similar (21 K/at% compared to 19.6 V/at%) suggesting a strong correlation in the phase switching mechanism related to Joule heating. We will discuss a predictive model using combination of semiconductor impurity conduction theory and symbolic regression (SR) to extract a general relationship between doping density and insulator-metal transition temperature that extends beyond the linear regime. This analysis strongly indicates the effectiveness of electron doping in de-stabilizing the insulating state compared to structural distortions from iso-valent dopants (e.g. Ti) . Then, we will present results on high energy ion implantation modelling of various electron donors into VO2 for systematic tuning of the transition properties and compare against experiments. Combined, these studies will be useful for the design and fabrication of low power switching oxide IMT devices in neuromorphic computing.