Yiyang Li1
University of Michigan1
In-memory computing using analog resistive memory promise substantial improvements in the energy efficiency of deep neural networks. Valence-change memory (VCM) and electrochemical random-access memory (ECRAM) are two promising analog resistive memories that stores information through the migration of oxygen vacancies. In this work, we show how materials thermodynamic engineering and especially phase separation can be used to control oxygen transport and create nonvolatile memory devices. We first investigate two-terminal filamentary VCM, and show that the widely used transport model based on Fickian diffusion does not accurately describe oxygen migration during retention. Instead, oxygen transport is governed by phase separation and spinodal decomposition, which enables certain resistance states to be indefinitely nonvolatile. Next, we apply these thermodynamic principles to design three-terminal ECRAM cells and achieve the first nonvolatile ECRAM cell that operate under short circuit configurations without a switch.