Selenium Vacancy-Enabled 2D Bi2O2Se Optoelectronic Synapses for Neuromorphic Vision Processing.

Abstract

Neuromorphic visual computing aims to fuse sensing, memory, and processing into a single hardware stack, enabling reliable analog weight updates. Defect engineering in two-dimensional (2D) semiconductors is an effective strategy for achieving such synaptic functionality. However, in most 2D materials, the defects that confer plasticity also compromise lattice stability and degrade intrinsic performance. This work demonstrates that Bi2O2Se provides a robust, defect-tolerant platform. The growth process introduces selenium vacancy (VSe) without compromising covalent structure or the approximately 0.8 eV bandgap, yielding stable and programmable trap centers. Combined with a low-trap hexagonal boron nitride (h-BN) gate dielectric that suppresses extra interfacial states, the resulting Bi2O2Se/h-BN transistor exhibits analog conductance modulation, excitatory postsynaptic currents, double-pulse facilitation, and long-term potentiation/inhibition, supporting hybrid electro-optic learning via vacancy-assisted charge trapping and persistent photoconductivity. Mapping measured device characteristics onto designed convolutional neural networks achieves 90.1% accuracy in complex image recognition, establishing an optoelectrical neuromorphic multifunctional platform for edge intelligence systems.