Broadband In‐Sensor Image Processing Based on Reconfigurable Asymmetric WSe 2 Photodetectors

ABSTRACT Hyperspectral remote sensing, which captures data from the ultraviolet (UV) to near‐infrared (NIR), is critical for applications ranging from environmental monitoring to military reconnaissance. However, the immense data volume creates a bottleneck, causing high latency and power consumption in traditional systems where sensing and processing are physically separated. In‐sensor computing, which performs data pre‐processing directly at the sensor level, offers a promising solution. Here, we demonstrate a novel device architecture for broadband in‐sensor image processing based on an asymmetric bottom‐Au/WSe 2 /top‐Au (b‐Au/WSe 2 /t‐Au) stack. This device leverages a gate‐controlled lateral homojunction within the bipolar WSe 2 channel to achieve a linearly tunable positive and negative photoresponse across a broadband range from 300 to 1800 nm. This makes it ideal for broadband analog multiply‐accumulate operations. Furthermore, the device exhibits linear photoresponse to optical power, with a power law exponent α ranging from 0.97 to 1.05. As a proof‐of‐concept, we fabricated a 3 × 3 array hardwired as a Prewitt operator, successfully demonstrating its capability for real‐time broadband edge detection. This work provides a concise and scalable hardware platform for performing broadband in‐sensor image processing.