Visible-to-mid-infrared in-sensor computing with a reconfigurable black phosphorus photodiode

In-sensor computing offers a solution for communication latency and power consumption in traditional machine vision systems with separate image acquisition and processing. Most sensors with internal computing capability operate in the visible or near infrared regime. An extension to the mid-infrared regime is desired. We present a reconfigurable black phosphorus (BP) photodiode for visible-to-mid-infrared in-sensor convolutional processing. Based on the split-gate induced spatially differential doping, there appears a reconfigurable isotype homojunction in the BP channel, leading to a well controllable bipolar responsivity that is linearly dependent on the gating voltage. The device operates at room temperature. The convolutional preprocessing based on our device is consistent with the simulation. The handwritten digit recognition with our device exhibits a high accuracy. This study advances broadband in-sensor computing and intelligent detection technology.