Memristive devices for improving AI energy efficiency

Abstract

Resistive switching memories (ReRAM) are emerging devices that can exhibit both volatile and non-volatile behaviour, making them promising candidates for next-generation memory and neuromorphic computing applications. In this work, inkjet printed Ag/h-BN/Au memristive devices based on an hexagonal boron nitride dielectric, which is a two-dimensional material, are experimentally investigated. Electrical characterization is performed through current–voltage measurements, where reproducible switching between high- and low-resistance states is observed, associated with the formation and rupture of conductive filaments driven by ionic migration. Both volatile and non-volatile regimes are obtained depending on the applied electrical conditions. The role of current compliance in filament stabilization during the switching process is considered. In addition, the devices are analysed in terms of cycle-to-cycle and device-to-device variability to evaluate their reproducibility and reliability. The results highlight the stochastic nature