Gate-tunable in-sensor computing vdW heterostructures for infrared photodetection

Abstract

Convolutional preprocessing is feasible for feature extraction and accurate recognition. In-sensor computing, which requires a photodetector with a computation function, is a potential candidate for hardware-implemented preprocessing. However, limited by the high carrier concentration in infrared sensing materials, reconfigurable manipulation of photocarriers is hardly complemented. Thus, previous works mostly focused on preprocessing in the visible range. Here, we propose a gate-tunable BP/MoS₂ heterostructure. With an elaborate design on the material’s thickness, the depletion region can be precisely controlled, resulting in multiple and reconfigurable responsivity states. With a sharp and clean interface, our device shows strong linear dependence over the broadband spectrum, which is the prerequisite for constructing convolutional kernels. Furthermore, observing the maximum photocurrent in the V_g sweeping process demonstrates strong regulation of carrier concentration in the infrared sensing material, BP layer. Since it has superior performance in high linearity and multiple states construction, our device is suitable for realizing computation in photodetector for convolutional preprocessing, underscoring its superiority in intelligent infrared perception and preprocessing.