Retina-Inspired Dual-Mode Photodetector with Spectral-Tunable Memory Switching for Neuromorphic Visual Systems

Abstract

The development of multifunctional photodetectors that integrate sensing, storage, and computing to mimic the human visual system for efficient image processing is a key area of research. In particular, retina-inspired optoelectronic devices with multispectral information preprocessing capabilities are critical for constructing neuromorphic visual systems; however, achieving this in traditional photodetectors is challenging due to the lack of suitable photoresponse modes. Herein, a graphene/organic photodetector (GOP) with a spectral-tunable photoresponse memory mode switching feature is demonstrated. Benefiting from the unique photogenerated charge transfer and trapping behavior in the heterojunction, the device exhibits memory-free (with recovery times of a few milliseconds) and long-memory (with recovery times of several hundred seconds) photoresponse modes under long-wavelength (650–1064 nm) and short-wavelength (370–520 nm) light stimulation, respectively. Furthermore, the device supports spectral-tunable dual-mode switching between photosynaptic and photodetection under multiple light pulse stimulations, enabling real-time preprocessing of images with mixed green and red dual-wavelength information using a GOP-based 3 × 3-pixel image sensor. We also demonstrate a GOP-constructed neuromorphic visual system for efficient image processing, where the front-end GOP-based image sensor filters out background noise in the input images, significantly improving the image recognition accuracy of the back-end GOP-connected artificial neural network (from 40 to 93%).