Hybrid Volatile/Nonvolatile Resistive Switching Memory in Ternary Metal Oxide Enabling Hopfield Neural Classification

Abstract

Processing data demands volatile memory, whereas storage necessitates nonvolatile devices. Typically, during the operations, data moves back and forth between them, causing an increase in energy consumption, undesired heating, and von Neumann bottleneck. Therefore, overcoming these obstacles could provide an essential breakthrough for on-demand in-memory processing; however, this would undoubtedly necessitate the unification of both volatile and nonvolatile memory. Ferroelectric materials have the potential for such unification of memory; yet, because of their different physical origins, getting both memories in one device has been difficult. Here, we developed high-performance, two-terminal, hybrid volatile/nonvolatile memory devices using ternary metal oxide as an active material. Particularly, formation of the ferroelectric NiTiO3 phase by solid-state synthesis is revealed by transmission electron microscopy and related tools, which is further corroborated by polarization-voltage measurements. Additionally, devices exhibiting trapping/detrapping-based volatile and ferroelectric polarization governed nonvolatile, remarkably stable (>105 s), and controlled multilevel (>6) analog memory with a configurable switching ratio of 103. Further, intrinsic Hopfield natural learning and object classification of input noisy patterns without any back propagation is demonstrated. The study proposes a route to develop on-demand hybrid volatile/nonvolatile memory devices, which will have a potential impact on multiple applications, including memory storage, processing, and neural classification.