ABSTRACT Emerging applications in the Internet of Things (IoT), autonomous vehicles, and robotics expose energy and latency bottlenecks in conventional photo‐sensing based on the von Neumann architecture. In‐sensor computing has emerged as a promising alternative by integrating multiple functions into a single device. 2D materials are widely used for prototypes owing to their excellent optoelectronic properties; however, free‐space illumination poses several challenges. First, the limited light–matter interaction length‐yields low absorption. Second, although logic and neuromorphic functions can combine electrical and optical signals, free‐space coupling hinders dense interconnection. Finally, these applications often require infrared operation, which remains challenging for commonly used transition metal disulfides (TMDs). Here, we report a waveguide‐integrated multifunctional photodetector for on‐chip in‐sensor computing that integrates light detection, nonvolatile memory, and reconfigurable logic/synaptic operations. Efficient infrared detection at 1552.5 nm is achieved, with a normalized photocurrent‐to‐dark‐current ratio (NPDR) up to 6.4 × 10 3 m W −1 . After negative control‐gate pulses programming, the device exhibits a projected retention of 21.1% after 10 years. We further realize an “AND” logic and synaptic short‐ and long‐term memory (STM/LTM). These results advance on‐chip in‐sensor computing compatible with high‐density photonic‐electronic interconnects and highlight a route toward systems that can sense, store, and compute at the point of capture.
Zhang et al. (Thu,) studied this question.