Biomimetic Dual‐Stream Neurons with Organic–Inorganic Heterojunction for Efficient Neuromorphic Vision Computing

ABSTRACT The human visual system achieves precise recognition and rapid response across different temporal scales through the coordinated operation of ventral and dorsal streams. However, reported multifunctional artificial neuron devices for visual applications primarily rely on the dynamic characteristics of multiple transmission channels, which increases the complexity of device design and peripheral circuitry. Inspired by this dual‐stream operating regime, we propose a dual‐stream neuromorphic device (D‐SND) capable of realizing two distinct neural response modes: slow accumulation and rapid burst discharge within a single structure. D‐SND achieves stable mode switching based on an organic–inorganic composite heterostructure, maintaining a high on‐off ratio over 10 3 . The internal conductive mechanisms of D‐SND allow in situ reconfiguration without depending on external circuits or predefined process parameters. Furthermore, we constructed various visual tasks for validation. In slow mode, it achieved 97.9% static image recognition accuracy, while in fast mode, it achieved 91.4% spatial perception recognition accuracy. In hybrid mode, the two response states collaborate to achieve multimodal visual fusion similar to the dual pathways in the human brain. Compared to the individual modes, image recognition accuracy improved by 37.2% and 53.3%, respectively. This strategy provides a simple structure, low‐power hardware pathway for constructing task‐reconfigurable neuromorphic visual hardware.