In-Sensor Convolutional Computation Based on Retina-Inspired Infrared Photodetector Array

In the field of infrared detection, as the resolution of detection wavelengths and spatial imaging improves, traditional architectures that separate sensing and processing struggle to meet the exponentially growing demands of machine vision. Inspired by the human visual system, which performs preprocessing on the retina to reduce the burden of information transmission, we propose an MSM (metal/semiconductor-InGaAs/metal) structure retina-inspired infrared photodetector. This detector allows the modulation of response polarity and amplitude by adjusting the bias voltage, enabling a range of preprocessing operations inside the detector. The Au/InGaAs/Au retinal-like detector exhibits tunable positive and negative photoresponses in the short-wave infrared region, with a response speed of up to 490 ns, facilitating ultra-fast image information preprocessing. We propose a <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$16\times 16$ </tex-math></inline-formula> in-sensor computing hardware system capable of deploying various convolution kernels directly inside the detector to perform convolution operations, offering a unique opportunity for lightweight convolutional neural network hardware.