近日，国科大杭州高等研究院陈效双团队研究了由原始黑色砷磷实现的偏振敏感神经形态视觉感应。相关论文于2026年2月3日发表在《光：科学与应用》杂志上。

偏振敏感神经形态视觉传感技术在区分光偏振态方面表现卓越，其内在优势在于能够降低复杂光照环境中的眩光干扰并提升视觉清晰度，为自动驾驶、光通信及可见光至红外光谱范围的仿生成像等前沿应用提供可能。

研究组展示了一种基于高质量本征各向异性二维黑砷磷纳米片的偏振敏感神经形态光电晶体管。该器件在近红外通信波段内展现出优异的光电性能：峰值响应率达2.88 A W^-1，偏振比达4.7，动态范围达40 dB。通过偏振态与栅压等多维度输入调控，该光电晶体管成功模拟了类人脑神经响应视觉刺激的突触行为，成对脉冲易化值可达201%。尤为关键的是，器件表现出栅压可调的短期可塑性，而光持续性则可触发稳定的长期可塑性状态，为记忆巩固奠定基础。

基于其神经形态特性，研究组构建了光电混合神经网络，在Fashion-MNIST数据集上实现超90%的分类准确率，并利用耶鲁人脸数据库数据在0°线偏振条件下实现71.38%的重建准确率。研究组进一步通过斯托克斯参量提取与线偏振度映射，演示了基于该黑砷磷光电晶体管的偏振分辨成像方法，能够高保真重建隐藏目标，揭示传统成像系统无法观测的精细偏振特征。该工作为集成偏振成像、计算与通信功能的高性能神经形态视觉系统奠定了基础平台，为应对规模化类脑光电子技术中的关键挑战提供了解决方案。

附：英文原文

Title: Polarization-sensitive neuromorphic vision sensing enabled by pristine black arsenic-phosphorus

Author: Zhang, Shi, Zhu, Shuguang, Tian, Shijian, Zhang, Libo, Chen, Cheng, Xiao, Kening, Mo, Wenqi, Hou, Shicong, Zhang, Yunduo, Wen, Yuanfeng, Tan, Yiran, Zhang, Kaixuan, Han, Jiayue, Liu, Changlong, He, Jiale, Tang, Weiwei, Wang, Jun, Li, Guanhai, Zhang, Kai, Wang, Lin, Chen, Xiaoshuang

Issue&Volume: 2026-02-02

Abstract: Polarization-sensitive neuromorphic vision sensing excels in distinguishing light polarization states, offering intrinsic advantages in reducing glare and enhancing visual clarity in complex lighting environments, enabling advanced applications in autonomous driving, optical communication, and bioinspired imaging across the visible-to-infrared spectrum. Here, we present a polarization-sensitive neuromorphic phototransistor based on a high-quality, intrinsically anisotropic two-dimensional black arsenic-phosphorus nanosheet, which exhibits exceptional optoelectronic performance with a peak responsivity of 2.88AW-1, a polarization ratio of 4.7 and a dynamic range of 40dB within the near-infrared communication band. Through multidimensional input control, including polarization and gate voltage, the phototransistor successfully simulates synaptic behaviors analogous to human neural responses to visual stimuli, with paired-pulse facilitation values reaching 201%. Critically, the device demonstrates gate-tunable short-term plasticity, with optical persistence triggering stable long-term plasticity states that underpin memory consolidation. The neuromorphic properties enable the development of a hybrid optical-electronic neural network which achieves a classification accuracy of over 90% on the Fashion-MNIST dataset and a reconstruction accuracy of 71.38% using data from the Yale Face Database under 0o linear polarization. We demonstrate a polarization-resolved imaging approach utilizing the black arsenic-phosphorus phototransistor to reconstruct hidden targets with high fidelity through Stokes parameter extraction and degree of linear polarization mapping, revealing intricate polarization features invisible to conventional imaging systems. Our work establishes a foundational platform for high-performance neuromorphic vision systems with integrated polarization imaging, computation, and communication functionalities, addressing critical challenges in scalable brain-inspired optoelectronic technologies.

DOI: 10.1038/s41377-025-02125-0

Source: https://www.nature.com/articles/s41377-025-02125-0