Abstract
The von Neumann architecture is encountering challenges, including the "memory wall" and "power wall" due to the separation of memory and central processing units, which imposes a major hurdle on today's massive data processing. Neuromorphic computing, which combines data storage and spatiotemporal computation at the hardware level, represents a computing paradigm that surpasses the traditional von Neumann architecture. Artificial synapses are the basic building blocks of the artificial neural networks capable of neuromorphic computing, and require a high on/off ratio, high durability, low nonlinearity, and multiple conductance states. Recently, two-dimensional (2D) materials and their heterojunctions have emerged as a nanoscale hardware development platform for synaptic devices due to their intrinsic high surface-to-volume ratios and sensitivity to charge transfer at interfaces. Here, the latest progress of 2D material-based artificial synapses is reviewed regarding biomimetic principles, physical mechanisms, optimization methods, and application scenarios. In particular, there is a focus on how to improve resistive switching characteristics and synaptic plasticity of artificial synapses to meet actual needs. Finally, key technical challenges and future development paths for 2D material-based artificial neural networks are also explored.