Electrically Driven Insulator-Metal Transitions in Vanadium Dioxide for Neuromorphic Computing

The field of artificial intelligence (AI) envisions to create intelligent machines inspired by the human brain. Over the past two decades, the rapid development of artificial intelligence has enabled machines to learn and perform complicated variety of tasks and even outperforming humans in some aspects. Design of semiconductor device hardware for AI is an active area of research. Ideally, these materials and devices should display behaviors similar to the operation of neurons and synapses, such as learning and forgetting, summation of inputs, threshold firing, etc. Correlated oxide materials such as VO₂ are a good candidate system due to the wide variety of physical phenomena they can possess. The metal to insulator transition of VO₂ can be triggered by electrothermal effect, which is important for building artificial neurons. Additionally, it can retain several metastable states by well controlled doping, which is desirable for artificial synapses. Three studies based on VO₂ are presented here. In the first study, the switching dynamics of in-series connected VO₂ devices under triangular electric pulses is presented. Experimental data shows the temperature of one of the connected devices can affect the input voltage to trigger E-IMT of both connected VO₂ device. Furthermore, this property is employed in building neuron models in SNN simulations to show case its application in neuromorphic networks. In the second work, selective area hydrogen doping of VO₂ device by electrode patterning is shown. With this technique, both artificial neurons and synapses can be realized in a single chip. Finally, the effect of hydrogen doping on volatile electrically triggered insulator to metal transition of VO₂ is discussed. Modulation of threshold voltages and stochasticity is demonstrated. These studies highlight the opportunities presented by phase transitions in oxides for emerging computing technologies.