Analysis of Variability in Resistive Memory Originated from Initial Defect Configuration by Fully Coupled Electrothermal and Phase-Field Models

Recently, resistive random-access memory (RRAM) has been profoundly studied to apply to neuromorphic computing. So far, it is hard to use the RRAM as a synapse device because its characteristics are sensitively affected by various factors. In other words, the variability is too severe. Therefore, reducing the variability is one of the major challenges in the RRAM device research. In this work, we investigate the factors affecting the device variability in conductive filament behaviors using a multi-physics simulation that integrates the electrothermal and phase-field models. The initial defect configuration is the source of the variability. With the given defects, the cell architecture, both cell geometry and constituent materials, to reduce the variability in the device characteristics are investigated.