Metal oxide resistive memory with a deterministic conduction path

Abstract

Resistive random access memories (RRAMs) with minimal power dissipation, high speed, and matrix-vector multiplication capability are potentially ideal for data-centric applications such as neuromorphic computing. However, RRAMs still suffer from instability caused by uncontrolled filament growth and random oxygen vacancy distribution. In this study, a Ge–Sb–Te ternary chalcogenide layer that functions as a conductive lead is added to a HfO₂-based RRAM layer to confine the subsequent filament formation to the initially determined site. Based on the DC and pulse measurement data, this technique is confirmed to improve the memory switching reproducibility without compromising its endurance and retention. Such deterministic behavior will be important in improving the sensing margin and multi-level capability of RRAM technology as the switching characteristics become more unstable with extreme device scaling.