Talanti Salihuojia, Fu Kerui, Zheng Xiaolong, Shi Youzhi, Tan Yimei, Liu Chenxi, Liu Yanfei, Mu Ge, Hao Qun, Weng Kangkang, Tang Xin
School of Optics and Photonics, Beijing Institute of Technology, Beijing, 100081, China.
XinIR Technology (Beijing) Co., LTD, Beijing, 101102, China.
Nat Commun. 2025 May 9;16(1):4311. doi: 10.1038/s41467-025-59446-2.
Simultaneously capturing static images and processing dynamic visual information within a single sensor enables a more comprehensive and efficient acquisition of scene information, thereby enhancing the understanding and processing of complex scenes. However, current artificial visual systems present significant challenges in device integration and multimodal operation. Here, we developed a 640×512-pixel CMOS-integrated organic neuromorphic imager featuring dual modes: standard (frame-based imaging) and synaptic (neuromorphic imaging). In synaptic mode, the system extracts high-resolution spatiotemporal maps (light distribution and motion trajectories) from final frames, decoding temporal sequences of light events through contrast analysis. The neuromorphic device demonstrates adjustable memory behavior through modulation of charge recombination-trapping dynamics, enabling multi-level memory functionality. We further developed a CMOS-compatible photolithography method, which supports high-resolution and non-destructive patterning of organic neuromorphic devices. The fabricated imager allows in-sensor memorization (>18 min) and real-world spatiotemporal imaging with reduced computation resource, demonstrating its potential for industrial monitoring and motion detection.
在单个传感器内同时捕获静态图像并处理动态视觉信息，能够更全面、高效地获取场景信息，从而增强对复杂场景的理解与处理能力。然而，当前的人工视觉系统在设备集成和多模态操作方面面临重大挑战。在此，我们开发了一款640×512像素的CMOS集成有机神经形态成像器，具有两种模式：标准（基于帧的成像）和突触（神经形态成像）。在突触模式下，该系统从最终帧中提取高分辨率时空图（光分布和运动轨迹），通过对比度分析解码光事件的时间序列。这种神经形态器件通过调制电荷复合俘获动力学展现出可调节的记忆行为，实现了多级记忆功能。我们还进一步开发了一种与CMOS兼容的光刻方法，该方法支持对有机神经形态器件进行高分辨率和无损图案化。所制造的成像器允许在传感器内进行记忆（超过18分钟）以及以减少的计算资源进行真实世界的时空成像，展示了其在工业监测和运动检测方面的潜力。
