Tunable Synaptic Plasticity in 2D Ferroelectric Semiconductor Transistor for High-Precision Neuromorphic Computing

Abstract

With the rise of big data and artificial intelligence, the von Neumann architecture’s limitations in computing power and energy efficiency are becoming increasingly evident. Neuromorphic computing, an innovative approach inspired by simulating the workings of the human brain, aims to achieve high computational capabilities with low energy consumption. Two-dimensional (2D) van der Waals ferroelectric semiconductor α-In₂Se₃ exhibits a unique combination of ferroelectricity, semiconductor properties, and the advantages of 2D materials, demonstrating significant potential as an ideal platform for information processing. This work reports a 2D ferroelectric semiconductor synaptic transistor based on α-In₂Se₃, which exhibits nonvolatile characteristics and synaptic plasticity due to the ferroelectric remanent polarization of α-In₂Se₃. The tight coupling between ferroelectric polarization and semiconducting nature allowed the α-In₂Se₃ ferroelectric semiconductor field-effect transistor to achieve a high current on/off ratio of 10⁵, a wide memory window of 81 V, and retention time greater than 600 s. Furthermore, the device demonstrated tunable synaptic plasticity, exhibiting paired-pulse facilitation, long-term potentiation/depression, the transition from short-term to long-term plasticity, as well as learning-experience behavior. Electrically modulated synaptic plasticity enabled an artificial neural network to achieve a peak accuracy of 94.8% on the MNIST handwritten digit data set, maintaining over 80% accuracy under background noise (standard deviation up to 50%), highlighting the robust fault tolerance of the conductance states. These results demonstrate that the 2D ferroelectric semiconductor α-In₂Se₃ holds significant potential for applications in high-performance information storage, processing, and neuromorphic computing.