Artificial Neuromorphic Synapse Based on MoS 2 /h‐BN/CrOCl/Graphene for Non‐Volatile Memory in Visual Recognition Applications

ABSTRACT Non‐volatile memory devices and neuromorphic computing systems based on two dimensional (2D) materials have emerged as promising platforms for real time artificial intelligence processing. However, conventional floating gate architectures often suffer from limited multi state programmability due to tunneling layer instability and interface defects, hindering the development of multifunctional bio inspired neuromorphic hardware. In this work, a MoS 2 /h‐BN/CrOCl/graphene floating‐gate field effect transistor that integrates high performance non‐volatile memory with optoelectronic synaptic functionalities is demonstrated. The incorporation of a 2D wide bandgap CrOCl layer as a tunneling pump, vertically stacked with h‐BN, raises the electron tunneling barrier and significantly suppresses charge leakage, thereby improving retention and endurance. As an electronic synapse, the device exhibits an ultra‐wide memory window exceeding 170 V and a program/erase ratio greater than 10 4 . As a photonic synapse, it emulates short term and long‐term plasticity, enabling Reservoir Computing based recognition of MNIST handwritten digits with 97.8 % accuracy. Furthermore, by decoding time encoded optical pulses, the system achieves dynamic visual recognition of vehicle motion directions with 93.7% accuracy. These results highlight the potential of the proposed heterostructure for constructing reliable, multi modal neuromorphic synapses suitable for future edge vision and intelligent sensing systems.