In the information era, conventional silicon-based chips face significant challenges to meet the demands of real-time data processing, device miniaturization, and low-power operation due to the short-channel effects, leakage current control, and quantum effects. The von Neumann architecture, with its separation of computing and storage units, leads to a bottleneck that the massive data transfer between units consumes a substantial energy and reduces a system efficiency. To address these issues, neuromorphic computing systems based on memristors have attracted much attention. These systems integrate storage and computing together, thus enhancing a computational efficiency and reducing an energy consumption. Two-dimensional (2D) materials, with their atomic-scale thickness, free of dangling bonds on the surface, and tunable properties, emerge as promising candidates for key materials in neuromorphic computing.

This review provides a comprehensive overview on the applications of 2D materials in neuromorphic computing. The synthesis methods of 2D materials and their heterojunctions are introduced, and then various architectures and operating principles of 2D neuromorphic devices are discussed. The review also represents the unique advantages of 2D materials, such as their high carrier mobility and excellent mechanical flexibility, which make them suitable for a wide range of applications, including wearable devices, biosensors, and implantable medical devices. In terms of device structures, the review mainly covers two-terminal memristors and three-terminal synaptic transistors. Two-terminal memristors, with their simple structure and high scalability, are ideal for high-density memristor crossbar arrays. Three-terminal synaptic transistors offer a more controllable performance and a better stability. The review explores emerging multi-terminal heterosynaptic devices, which can simulate more complex synaptic behaviors.

This review further examines the integration of 2D materials into neuromorphic systems, including in-memory computing architectures and sensory-processing-computing integrated systems. In-memory computing architectures, such as memristor crossbar arrays, leverage the non-volatile storage and multi-resistance states of memristors to directly simulate neural network node weights and perform parallel matrix operations. Sensory-processing-computing integrated systems utilize the unique response properties of 2D materials to collect environmental signals, such as light, gas, and sound, as well as integrate sensing modules with in-memory computing to achieve efficient data processing.

This review addresses the challenges of applying 2D materials to neuromorphic computing, including material preparation, device design, and system integration. While laboratory-scale preparation methods like mechanical exfoliation can yield high-quality 2D materials, they are not suitable for large-scale integration. Wafer-scale growth methods, such as chemical vapor deposition (CVD) and molecular beam epitaxy (MBE), offer a better scalability but face some challenges in controlling the number of layers, achieving a uniform thickness or a high production efficiency. In terms of device design, 2D neuromorphic devices exhibit diverse architectures and operating principles, such as conductive filaments, phase changes, and defect engineering. However, further efforts are needed to reduce power consumption and improve cyclic stability.

The review outlines future research directions, such as the development of large-area material growth and transfer techniques, the design of novel device structures, and the integration of multi-functional systems. 2D materials are expected to play a crucial role in the advancement of neuromorphic computing through interdisciplinary collaboration and continuous innovation, paving a way for the development of artificial intelligence, the Internet of Things, and big data technologies.