Modeling of oxide-based ECRAM programming by drift-diffusion ion transport

This work presents a new numerical model for the electrochemical random-access memory (ECRAM), a three terminal analogue memory device where the defect concentration controls the conductance of a metal oxide channel. The model relies on partial differential equations for ionic drift-diffusion across the ECRAM stack, causing the potentiation and depression of the synaptic channel. The model accounts for experimentally-observed linear programming and temperature acceleration of potentiation/depression. The model is useful for predicting the device characteristics and optimizing the device geometry for analogue-domain neural network accelerators.