Infrared (IR) vision systems are essential for intelligent sensing in complex environments where visible-light perception becomes ineffective, such as low illumination, obscurants, and low-contrast scenes. However, most existing infrared imaging systems rely on conventional architectures that physically separate sensing, memory, and computation, resulting in substantial data-transfer overhead, latency, and energy consumption. In this work, an infrared in-sensor computing platform is developed based on a large-area ferroelectric photoelectric memristor array that integrates infrared sensing, nonvolatile memory, and weighted computation within a single hardware system. By coupling an infrared-sensitive ReSe2 channel with in-plane ferroelectric polarization, the devices exhibit highly linear (99%), nonvolatile, multilevel modulation of self-powered photoresponsivity, enabling stable analog photoelectric weight programming with near-infrared sensitivity. Leveraging these properties, a 5 × 9 photoelectric memristor crossbar array is employed for direct infrared image sensing and processing at the sensor level. A proof-of-concept infrared image recognition task is demonstrated with in situ sensing and classification, without intermediate data conversion, external computing units, or bias-assisted operation, achieving high recognition accuracy and robust classification margins even in the presence of noise. This work establishes a practical pathway toward compact, low-latency, and energy-efficient infrared in-sensor computing systems, highlighting the potential for next-generation intelligent vision hardware.
Sheng et al. (Fri,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: