ABSTRACT The human retina capable of extracting key feature information is a crucial sensory element in the visual system. Constructing bionic devices with multitype feedback to emulate the retina behaviors in complex environments has been a persistent pursuit to broaden the visual range. However, the definition and regulation of synaptic weights persist as a bottleneck problem in the terahertz (THz) devices to achieve neuromorphic function. Here, we have proposed bismuth‐based THz photonic neuromorphic devices with picosecond short‐term plasticity and constructed the THz‐optical neural network (THz‐ONN) to imitate retina function. Crucially, the weight embodied by THz photoresponse can be precisely regulated via incremental optical pulses, delivering an incredibly simple yet powerful approach that heralds systems with a continuously variable plasticity. Further development of diverse neuromorphic devices for various scenarios could be realized through the combination of band alignment engineering and substrate effects to control photocarrier transport. The corresponding neuromorphic computing based on THz‐ONN indicates the high recognition accuracy of hardware. The present study provides an exciting paradigm for the realization of THz neuromorphic devices and opens an avenue for mimicking biological sensory system.
Zhang et al. (Thu,) studied this question.