Volatile nanocomposite memristor with a phase stratification dielectric layer: a threshold switching with rich neuromorphic dynamics

Abstract

Volatile threshold memristive switches are attractive for the fabrication of artificial compact neurons, third–order neuromorphic elements, as well as reservoir computing (RC) systems. The latter have emerged as a powerful neuromorphic framework for efficient temporal signal processing, offering a compelling combination of computational efficiency and low training cost, since only the output layer of the RC system has to be trained. A key challenge, however, lies in finding hardware platforms that possess the necessary short–term memory and nonlinear dynamics to serve as physical reservoirs. In this work, we address this by introducing a volatile bi-layer nanocomposite-based (Co-Fe-B)-LiNbO_х/(NbO_y)-LiNbO_z memristor, which exhibits pronounced and batch-to-batch reproducible short–term memory properties. We experimentally demonstrate that even a compact hardware system built around a single memristor can perform complex tasks. Specifically, we achieve a 97.1% accuracy in handwritten digit recognition from the MNIST dataset and a low normalized root mean square error of 0.028 in predicting the chaotic Hénon map time–series. This study expands the range of promising volatile nanocomposite memristive structures for neuromorphic computing applications, while additionally paving the way toward scalable, high–performance memristor–based RC systems for efficient complex temporal and spatial information processing.