The memristor is an excellent candidate for nonvolatile memory and neuromorphic computing. Recently, two-dimensional (2D) materials have been developed for use in memristors with high-performance resistive switching characteristics, such as high on/off ratios, low SET/RESET voltages, good retention and endurance, fast switching speed, and low power and energy consumption. Low-power memristors are highly desired for recent fast-speed and energy-efficient artificial neuromorphic networks. This Perspective focuses on the recent progress of low-power memristors based on 2D materials, providing a condensed overview of relevant developments in memristive performance, physical mechanism, material modification, and device assembly as well as potential applications. The detailed research status of memristors has been reviewed based on different 2D materials from insulating hexagonal boron nitride, semiconducting transition metal dichalcogenides, to some newly developed 2D materials. Furthermore, a brief summary introducing the perspectives and challenges is included, with the aim of providing an insightful guide for this research field.

An artificial synapse is essential for neuromorphic computing which has been expected to overcome the bottleneck of the traditional von-Neumann system. Memristors can work as an artificial synapse owing to their tunable non-volatile resistance states which offer the capabilities of information storage, processing, and computing. In this work, memristors based on two-dimensional (2D) MXene Ti3C2 nanosheets sandwiched by Pt electrodes are investigated in terms of resistive switching (RS) characteristics, synaptic functions, and neuromorphic computing. Digital and analog RS behaviors are found to coexist depending on the magnitude of operation voltage. Digital RS behaviors with two resistance states possessing a large switching ratio exceeding 103 can be achieved under a high operation voltage. Analog RS behaviors with a series of resistance states exhibiting a gradual change can be observed at a relatively low operation voltage. Furthermore, artificial synapses can be implemented based on the memristors with the basic synaptic functions, such as long-term plasticity of long-term potentiation and depression and short-term plasticity of the paired-pulse facilitation and depression. Moreover, the “learning–forgetting” experience is successfully emulated based on the artificial synapses. Also, more importantly, the artificial synapses can construct an artificial neural network to implement image recognition. The coexistence of digital and analog RS behaviors in the 2D Ti3C2 nanosheets suggests the potential applications in non-volatile memory and neuromorphic computing, which is expected to facilitate simplifying the manufacturing complexity for complex neutral systems where analog and digital switching is essential for information storage and processing.

A two-terminal memristor can be used for information memory and logic operation as well as serving as an artificial synapse for neuromorphic computing. Selective memory with some enjoyable information to be remembered and other to be screened out can be emulated by an artificial synapse. In this work, a memristor based on a single WO3 nanowire can be constructed, which demonstrates the co-existence of bipolar nonvolatile and volatile resistive switching (RS) behaviors that can be tuned by the amplitude of the operation voltage. For small operation voltages (2 V), the device demonstrates nonvolatile analog RS, which can be utilized as an artificial synapse with long- and short-term plasticity. The learning–forgetting experience of human can be emulated based on the artificial synapse. Moreover, the artificial synapse can be used for image recognition with the recognition accuracy up to 94% for small hand-written image. On the other hand, volatile RS can be observed with large operation voltages (6 V). Furthermore, based on the diverse nonvolatile and volatile RS behaviors, selective memory can be emulated. Our fabricated memristor can be used as an artificial synapse to achieve image recognition and to emulate selective memory, which paves a way to construct smart neuromorphic systems facing complex information.

The WSe2-based memristor demonstrates the controllable digital and analog resistive switching behavior. Moreover, it can be used to emulate the “learning-forgetting-relearning” experience and performs image recognition with high recognition accuracy.

No abstract