CuInP2S6 Heterojunction Based Visible Range Optoelectronic Synapse With Femtojoule Energy Consumption

Abstract

The 2D van der Waals material CuInP₂S₆, characterized by its memory behavior arising from room-temperature ferroelectricity and Cu⁺ ions migration, has emerged as a promising candidate material for artificial synaptic devices. Nevertheless, with a bandgap of 2.7 eV, CIPS-based devices are generally limited to operating in pure electrical mode or under ultraviolet light, making them unsuitable for applications across the entire visible light spectrum. Here, a two-terminal artificial synapse based on CIPS/MoS₂/graphene heterojunction is constructed. Compared to ion migration or ferroelectricity under high bias voltage, photogating due to charge trapping is identified as the working mechanism under low bias voltage (< 1.5 V), which can respond to the shortest pulse (∼5 ms) and least energy consumption of 1.7 / 6.3 fJ per pulse up to date for CIPS-based synapses. Benefiting from the fading memory effect and nonlinear characteristics in visible light range, handwritten digit recognition based on reservoir computing has achieved an accuracy of 90.43% with four times higher efficiency than directly using an artificial neuron network. This work thus paves the way for constructing CIPS heterostructure for artificial vision and neuromorphic computing systems.