The Future of Memristors: Materials Engineering and Neural Networks

Abstract

From Deep Blue to AlphaGo, artificial intelligence and machine learning are booming, and neural networks have become the hot research direction. However, due to the size limit of complementary metal–oxide–semiconductor (CMOS) transistors, von Neumann‐based computing systems are facing multiple challenges (such as memory walls). As the number of transistors required by the neural network increases, the development of neural networks based on the von Neumann computer is limited by volume and energy consumption. As the fourth basic circuit element, memristor shines in the field of neuromorphic computing. The new computer architecture based on memristor is widely considered as a substitute for the von Neumann architecture and has great potential to deal with the neural network and big data era challenge. This article reviews existing materials and structures of memristors, neurophysiological simulations based on memristors, and applications of memristor‐based neural networks. The feasibility and advancement of implementing neural networks using memristors are discussed, the difficulties that need to be overcome at this stage are put forward, and their development prospects and challenges faced are also discussed.

Advances in Memristor-Based Neural Networks

“…As a non-linear two-terminal passive electrical component, studies have shown that the conductance of a memristor is tunable by adjusting the amplitude, direction, or duration of its terminal voltages. Memristors have shown various outstanding properties, such as good compatibility with CMOS technology, small device area for high-density on-chip integration, non-volatility, fast speed, low power dissipation, and high scalability (Lee et al, 2008;Waser et al, 2009;Akinaga and Shima, 2010;Wong et al, 2012;Yang et al, 2013;Choi et al, 2014;Sun et al, 2020;Wang et al, 2020;Zhang et al, 2020). Thus, although memristors took many years to transform from a purely theoretical derivation into a feasible implementation, these devices have been widely used in applications such as machine learning and neuromorphic computing, as well as non-volatile random-access memory (Alibart et al, 2013;Liu et al, 2013;Sarwar et al, 2013;Fackenthal et al, 2014;Prezioso et al, 2015;Midya et al, 2017;Yan et al, 2017Yan et al, , 2019bAmbrogio et al, 2018;Krestinskaya et al, 2018;Li C. et al, 2018Wang et al, 2018aWang et al, , 2019aUpadhyay et al, 2020).…”

Recent advances on crystalline materials-based flexible memristors for data storage and neuromorphic applications

“…Recently, memristors have emerged as promising contenders for next-generation high-capacity information storage and computing systems, attributing to their advantages of fast data transfer rate, short access time, low power consumption, and the compatibility with complementary metal-oxide-semiconductor (CMOS) technology [8][9][10][11][12][13][14]. More importantly, they have exhibited great potential in the applications of nonvolatile memory, logic computing and brain-inspired neuromorphic hardware [15][16][17][18][19][20][21][22].…”

“…More importantly, they have exhibited great potential in the applications of nonvolatile memory, logic computing and brain-inspired neuromorphic hardware [15][16][17][18][19][20][21][22]. These three interrelated technologies provide a feasible route for developing a novel in-memory computing architecture that integrates information storage and processing in one system [12,13], which can break through the existing von Neumann bottleneck and memory wall of traditional computing systems. A typical memristor device generally composes of two electrodes and a switching layer between them, which can switch between high and low resistance states (RSs) in response to an external electric voltage.…”

Structural Analysis and Performance in a Dual‐Mechanism Conductive Filament Memristor

“…[ 10,11 ] According to previous literature, the switching behavior in RRAM corresponds to the formation and rupture of conductive filaments. [ 12 ] To explore the applications of RRAM devices, an in‐depth understanding of the filament formation mechanism is significant.…”

“…[10,11] According to previous literature, the switching behavior in RRAM corresponds to the formation and rupture of conductive filaments. [12] To explore the applications of RRAM devices, an in-depth understanding of the filament formation mechanism is significant.Two theories are based on the composition of the conductive filaments, the electrochemical mechanism (ECM) and the valence change mechanism (VCM). [13,14] Devices with ECM exhibit a large memory window; however, the retention is insufficient.…”