Artificial Synapse Based on Black Phosphorus/SnS₂ Heterostructure Transistor for Neuromorphic Computing with High...

Artificial synaptic devices have garnered significant attention for overcoming the limitations of conventional von Neumann architecture in neuromorphic computing. However, the reported nonlinear weight update of artificial synapses prevents a hardware neural network from achieving the same high training and learning accuracy levels as those attained by a soft neural work. Despite the large-area fabrication constraints, two-dimensional (2D) materials that have tunable electronic band alignment and optoelectronic properties have emerged as promising candidates for artificial synapses. Here, we demonstrate an artificial synaptic device based on a black phosphorus (BP)/SnS₂ van der Waals heterostructure. This device achieves a large memory window and successfully emulates various synaptic functionalities. Notably, the nonlinearity in the synaptic weight update of this device can be electrically modulated, and the underlying mechanism is analyzed by density functional theory (DFT) calculations. Furthermore, we successfully integrate BP/SnS₂-based artificial synapses into an artificial neural network, achieving a high recognition accuracy of 96.3% on the MNIST data set for handwritten digit recognition. These findings introduce an effective method for high-accuracy neuromorphic applications through the band structure engineering of 2D materials.