Coupled Ferroelectric‐Photonic Memory in a Retinomorphic Hardware for In‐Sensor Computing

TLDR

All‑in‑one artificial visual systems are attractive for their energy efficiency and real‑time processing, and the growth of IoT has heightened the importance of in‑sensor computing at the edge of data‑flow architectures. The study proposes a prototype retina‑inspired visual sensor that integrates ferroelectricity and photosensitivity in 2D α‑In₂Se₃. The device mimics retinal photoreceptors and amacrine cells by using ferroelectric polarization switching in α‑In₂Se₃ for optical reception and memory computation, and implements in‑sensor convolution through a network of phototransistors with five pre‑programmed kernels that perform arithmetic to generate edge‑enhanced images. The prototype achieved ≈94 % accuracy on 12 000 MNIST images using gate‑tunable excitatory/inhibitory short‑term plasticity, demonstrating that ferroelectric α‑In₂Se₃ enables highly compact, efficient retinomorphic hardware even with ambipolar transport.

Abstract

Abstract The development of all‐in‐one devices for artificial visual systems offers an attractive solution in terms of energy efficiency and real‐time processing speed. In recent years, the proliferation of smart sensors in the growth of Internet‐of‐Things (IoT) has led to the increasing importance of in‐sensor computing technology, which places computational power at the edge of the data‐flow architecture. In this study, a prototype visual sensor inspired by the human retina is proposed, which integrates ferroelectricity and photosensitivity in two‐dimensional (2D) α‐In 2 Se 3 material. This device mimics the functions of photoreceptors and amacrine cells in the retina, performing optical reception and memory computation functions through the use of electrical switching polarization in the channel. The gate‐tunable linearity of excitatory and inhibitory functions in photon‐induced short‐term plasticity enables to encode and classify 12 000 images in the Mixed National Institute of Standards and Technology (MNIST) dataset with remarkable accuracy, achieving ≈94%. Additionally, in‐sensor convolution image processing through a network of phototransistors, with five convolutional kernels electrically pre‐programmed into the transistors is demonstrated. The convoluted photocurrent matrices undergo straightforward arithmetic calculations to produce edge and feature‐enhanced scenarios. The findings demonstrate the potential of ferroelectric α‐In 2 Se 3 for highly compact and efficient retinomorphic hardware implementation, regardless of ambipolar transport in the channel.