Current spectral adaptation of vision systems is volatile operation that relies on cascading optical filters and electronic components, resulting in bulky architectures and high energy consumption. Inspired by the spectrally tunable vision of a migratory fish, we propose a spectral-adaptive nonvolatile-operating device based on a two-dimensional MoS2 channel with CuInP2S6 (CIPS) gate, in which the ferroelectric-photosensitive synergy of CIPS routes the carriers, emulating retina's adaptive feedback. The ferroelectric polarization dynamically tunes spectral synaptic plasticity and keeps the high spectral suppression ratio up to 102 without constant gate voltage or optical filters, which enhances target spectral feature extraction and elevates image recognition accuracy in cluttered scenes from 71.4 to 95.2%. Furthermore, the ferroelectric-photosensitive synergy of CIPS gate endows the Weber contrast (>102) on-demand switching in spectral dynamic scene, enabling autonomous driving seamless adaptation from glare to low-light environments. Nonvolatile reconfiguration of spectral adaptation presents a power-efficient non-von Neumann vision sensor.
Jiang et al. (Fri,) studied this question.