二维材料与人工视觉系统的多维融合: 前沿突破与范式革新

Artificial visual system (AVS) has received increasing attention for their transformative potential in fields such as medical diagnostics, intelligent robotics, and machine vision. Traditional silicon-based imaging technologies, however, face significant limitations, including high energy consumption, limited dynamic range, and integration challenges. Two-dimensional (2D) semiconductor materials, such as MoS₂, WSe₂, and black phosphorus have emerged as promising alternatives due to their atomically thin structure, tunable bandgaps, high carrier mobility, and superior optoelectronic properties. In this work, recent breakthroughs in the integration of 2D materials with AVS are investigated. Highlighted is the development of a reconfigurable four-terminal phototransistor array based on WSe₂ and IGZO heterostructures, which enables monocular 3D disparity reconstruction without the need for multiple cameras or active light sources. The system demonstrates a dynamic imaging rate exceeding 33 frames per second and supports real-time sensing, memory storage, and ambipolar mode switching with ultralow power consumption (as low as 142 pW). Key innovations include multifunctional device architectures that simulate the retinal photoreceptors, bipolar cells, and even neural synapses, achieving functions such as image sensing, real-time adaptation, color recognition, motion tracking, and multimodal perception. Furthermore, by simulating the human neurovisual pathways, these 2D material-based devices can potentially realize in-sensor computing and neuromorphic processing, which substantially reduce data transfer bottlenecks and energy overhead. Nonetheless, the field is still in its formative stage. Here, several critical bottlenecks are emphasized: the lack of scalable, defect-controlled synthesis of 2D heterostructures; the limited spectral bandwidth and color fidelity of current photonic components; the immature state of neuromorphic elements, which often lacks stability, long-term memory, and bio-realistic plasticity. Moreover, the practical integration with real-world applications requires compatibility with high-density manufacturing and dynamic, multi-modal environments. In the future, artificial vision platforms, empowered by engineered 2D materials and heterostructures, will develop into highly compact, intelligent, and context-aware agents capable of autonomous perception and interaction in complex real-world settings.