Presynaptic Optical Modulation Enabled by Dielectric-Embedded Type-II PbS/PbSe Quantum Dots for Near-Infrared Neuromorphic Vision.

In nighttime perception and near-infrared (NIR) neuromorphic vision, a central challenge lies in encoding weak optical inputs into stable and controllable temporal states at the sensor front end. In nocturnal mammalian retinas, AII amacrine cells achieve this function by relying on stable chemical intermediate states, presynaptic modulation, and subthreshold temporal integration. Functionally inspired by selected information-processing principles of this mechanism, we propose and demonstrate a bio-inspired device design strategy. Through a chemical-structural codesign approach, type-II PbS/PbSe core-shell quantum dots (QDs) are used for the first time and embedded into the dielectric layer of flexible carbon-nanotube (CNT) synaptic transistors, forming stable yet decaying photo-generated intermediate states. These states modulate the CNT channel conductance via photogating, realizing a presynaptic, nonconductive modulation. The synaptic device exhibits NIR perception covering 395-808 nm and operates with an energy consumption of 0.325 fJ per synaptic event. An 8 × 8 synaptic array enables history-dependent temporal processing, generating 32 distinguishable output states under five-bit optical inputs. Based on these properties, a reservoir computing (RC) framework is validated on a dataset exceeding 105 samples, achieving 94.6% accuracy in nighttime human action recognition tasks. These results establish a bio-inspired material-device-system pathway for NIR near-sensor temporal perception and computation.