Intelligent vision applications increasingly require sensors that not only capture optical scenes with high speed and sensitivity but also perform in-sensor processing and memory. Neuromorphic vision sensors (NVSs) that monolithically integrate photodetection and synaptic functions within a single device have thus emerged as a promising route toward low-redundancy, low-latency visual processing. Although significant progress has been made, existing NVSs still leave considerable room for improvement in reducing residual redundancy and achieving higher response speed and sensitivity. Here, we report a reconfigurable two-terminal vertical neuromorphic vision sensor (NVS) based on a PBDB-T:ITIC-Th organic heterojunction, in which the thickness of the ZnO hole-blocking layer is engineered to balance trap-mediated synaptic dynamics and high-performance photodetection. In photodetector mode, the NVS achieves an external quantum efficiency (EQE) of 78%, a detectivity (D*) of 1.2 × 1013 Jones, and fast rise/fall times of 7.5/9.6 μs. Additionally, the device supports two reconfigurable synaptic operation modes, enabling both event-driven processing and conventional synaptic behavior. System-level simulations further demonstrate that both synaptic modes enable high-accuracy vehicle recognition via a convolutional neural network (CNN), whereas the event-driven mode efficiently extracts dynamic contours under strong background illumination. The results could provide a compact, low-latency, and event-driven hardware platform for next-generation neuromorphic vision and intelligent sensing systems in complex environments.
Shen et al. (Mon,) studied this question.