Near-infrared photovoltaic gating enables polarity-reconfigurable WSe2 phototransistors for in-sensor computing.

Neuromorphic in-sensor computing requires scalable, energy-efficient optoelectronic devices with reconfigurable functionality, yet most retinomorphic platforms rely on complex van der Waals stacking and remain largely restricted to the visible spectrum. Here we demonstrate a simple substrate-engineered near-infrared (NIR) reconfigurable phototransistor by integrating an ambipolar few-layer WSe2 channel with a Si PN junction substrate through a 10 nm HfO2 dielectric. Under 980 nm illumination, the Si PN junction generates a photovoltaic potential that is capacitively coupled to the WSe2 channel as an effective photogating voltage, inducing a pronounced threshold-voltage shift and enabling gate-programmable switching between negative (NPC) and positive photoconductance (PPC). The device achieves responsivities of 62.3 A W-1 (PPC) and 14.5 A W-1 (NPC), microsecond-level response times, and 13 distinguishable programmable states. By mapping PPC, NPC and zero photoconductance (ZPC) states to positive, negative and zero convolution weights, the device enables in-sensor convolution for image preprocessing such as edge extraction, offering a scalable route toward NIR-capable neuromorphic vision hardware.