Stochasticity-aware modeling methodology of mott memristors validated on vanadium dioxide devices

Abstract Volatile memristors based on Mott insulators, such as vanadium dioxide (VO 2 ), are envisaged as key components for neuromorphic computing e.g. to implement the Hodgkin-Huxley model. A deterministic electro-thermal model known as the Picket-Williams (PW) model is adapted to our micro-fabricated VO 2 -based memristors. The tuned model accurately captures the experimental memristive behaviors of our device for different operating temperatures (35 and 45 °C). The transitions voltages and current of the reversible metal-insulator transitions are notably well reproduced. We further establish the mathematical relations between the fluctuations of the intrinsic material parameters and the cycle-to-cycle variations of the switching voltages. The such-obtained stochastic propagation model is then applied to feed a Monte-Carlo simulation on the tuned PW-model. The combination reproduces the experimental device fluctuation statistics (with an error smaller than 0.5 mV on the standard deviation). It is further demonstrated capable to simulate the transient operation of a typical spiking circuit prone to material stochasticity. The proposed data-driven methodology is general and could be applied to other models to simulate the impact of memristors device stochasticity on the circuit-level fluctuations, paving the way for faithful statistical modelling of neuromorphic circuits.