Volatile and Nonvolatile Memristive Devices for Neuromorphic Computing
Guangdong Zhou,Zhongrui Wang,Bai Sun,Feichi Zhou,Linfeng Sun,Hongbin Zhao,Xiaofang Hu,Xiaoyan Peng,Jia Yan,Huamin Wang,Wenhua Wang,Jie Li,Bingtao Yan,Dalong Kuang,Yuchen Wang,Lidan Wang,Shukai Duan
DOI: https://doi.org/10.1002/aelm.202101127
IF: 6.2
2022-02-10
Advanced Electronic Materials
Abstract:Ion migration as well as electron transfer and coupling in resistive switching materials endow memristors with a physically tunable conductance to resemble synapses, neurons, and their networks. Four different types of volatile memristors and another four types of nonvolatile memristors are systemically surveyed in terms of the switching mechanisms and electrical properties that are the basis of different computing applications. The volatile memristor features spontaneous conductance decay after the cease of electrical/optical stimulations, which are closely related to the surface atom diffusion, metal–insulator–transition (including charge–density–wave), thermal spontaneous emission, and charge polarization. Such unique dynamic state evolution at the edge of chaos has enabled them to emulate certain synaptic and neural dynamics, leading to various applications ranging from spiking neural networks to combinatorial optimizations. Nonvolatile resistive switching behavior originated from the electron spins, ferroelectric polarization, crystalline‐amorphous transitions or interplay between ions and electrons enables the memristor array to implement the vector–matrix multiplication, which is the key convolutional operation in artificial neural networks. The progress, challenges, and opportunities for both volatile and nonvolatile memristor in the level of materials, integration technology, algorithm, and system are highlighted in this review. Four volatile memristors and four nonvolatile memristors in terms of the switching dynamics, application in neuromorphic computing, and architectures of memristor arrays are systemically summarized. The motivation of this review is to refine the conception of volatile and nonvolatile memristor, summarizes the important progress in hardware technology, and proposes the possible solutions in different level for memristor‐based neuromorphic computing.
materials science, multidisciplinary,physics, applied,nanoscience & nanotechnology