Strategy for the Integrated Design of Ferroelectric and Resistive Memristors for Neuromorphic Computing Applications

Abstract

Implementing bimodal memristor operations using different operating principles and multifunctional thin films is a promising neuromorphic system strategy in terms of efficiency, versatility, and flexibility. In this study, we perform preliminary investigations to determine whether the ferroelectric and resistive memristor can be intentionally selected in one cell. The conversion process from ferroelectric to resistive memristor and the distinction between the two devices are explained based on systematic analyses. Based on a variety of measurements and analyses, the conversion process from ferroelectric to resistive memristor is investigated. Additionally, we experimentally demonstrate that both devices can emulate a variety of synaptic plasticity. We utilize different pulse schemes to improve the weight update linearity of both devices and then compare the recognition rates of both devices using the Fashion Modified National Institute of Standards and Technology (MNIST) data set and software-based simulations. Finally, using the short-term memory characteristics of the ferroelectric memristor, we experimentally demonstrate the memory/forgetting process of the human brain and simulate a reservoir computing system utilizing a ferroelectric/resistive memristor, fabricated with the same materials and processes, as the reservoir layer/readout layer, respectively.