Gate-Tunable Highly Linear Bipolar Photoresponse in Se@SWCNT Adaptive Neurons for Dynamically Programmable Neuromorphic Computing.

The development of tunable and highly controllable photoconductive devices for brain-inspired optical neuromorphic systems remains challenging. Previous neuromorphic devices are limited by asymmetric and nonlinear conductive properties, which impose specific restrictions on training tasks and weight learning rules in dynamic and complex visual environments. A programmable synaptic transistor based on a Se@SWCNT 1D van der Waals heterojunction, enabling gate-controlled positive and negative responses is presented. This approach eliminates the need for multilayer heterojunctions or complex circuits, simplifying array integration and wafer-scale fabrication. This phototransistor shows improved symmetry and linearity (R2 > 0.99) in weight variation following optical stimulation, and simultaneously achieves linear persistent photoconductivity and negative photoconductivity with over 128 memory states, which is not reported previously. By adjusting light intensity and wavelength range, consistent weight rule processing across three tasks of increasing complexity is demonstrated. Notably, different visual tasks require distinct neural structures and decay rates. The proposed transistor facilitates transitions between bio-inspired brain regions via optical hybrid programming, adapting to dynamic visual environments. This innovation contributes significantly to brain-like computing and bio-inspired vision, due to its exceptional accuracy and dynamic switch models.