2D Covalent Organic Framework-Based Heterostructure for Neuromorphic Visual Processing

The increasing demands of edge computing and Internet of Things (IoT) applications have highlighted critical limitations in conventional von Neumann-based vision systems, particularly regarding energy efficiency and processing latency. Drawing inspiration from biological vision, an artificial visual system with a 2D covalent organic framework (COF)/Molybdenum Disulfide (MoS₂) van der Waals heterostructure-based synaptic device is developed. This architecture integrates ultrathin zinc porphyrin-2,5-dihydroxyterephthalaldehyde (ZnP-2,5-DHa) 2D COF nanosheets, known for their exceptional photoresponsivity and environmental stability, with high-mobility MoS₂ semiconductor channels. The resulting heterostructure exhibits both broadband optical response and light-tunable synaptic plasticity, enabling biomimetic visual information acquisition and preprocessing within a single integrated platform. For static image processing, its spectrally selective response enables RGB-based denoising, improving convolutional neural network (CNN) recognition accuracy from 37.9% to 92.5% in high-noise environments, as evidenced in motion-blurred license plate restoration. For dynamic vision tasks, the integration of relaxation dynamics with 16 distinguishable conductance states facilitates precise spatiotemporal motion encoding, allowing artificial neural networks (ANN) to achieve 95.2% directional recognition accuracy. This study provides a material-level innovation for next-generation bio-inspired vision systems, with transformative potential for autonomous vehicles, intelligent surveillance, and neuromorphic edge computing.