Ultra-wide spectrum photosynapse array with 64k-scale for neuromorphic fusion imaging

Sep 30, 2026

Upload by Jacob

Bio-inspired photosynapse hardware enables wide-spectrum, in-situ optical information processing, yet integrating ultra-wide spectrum responsivity into a single device and scaling to large arrays remains difficult. This study demonstrates a 64k-scale ultraviolet-to-mid-infrared photosynapse array using carbon nanotube/molybdenum oxide heterojunctions. Photosynapse behaviors are verified at 365 nm, 532 nm, 1064 nm, and 10.6 μm, with 99.96% yield via polydopamine-mediated surface treatment. The array improves dynamic-trajectory prediction frame similarity from 0.897 to 0.965 and achieves 99.58% trajectory recognition accuracy under challenging conditions, exceeding visible-only performance and supporting autonomous perception.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages13

Reading Time33 min

AccessFree to download

Introduction

Motivation for wide-spectrum perception beyond visible light

Architectural and device challenges for wide-spectrum sensing

Neuromorphic photosynapse as a solution

Article [https://doi.org/10.1038/s41467-025-66810-9](https://doi.org/10.1038/s41467-025-66810-9) Ultra-wide spectrum photosynapse array with 64k-scale for neuromorphic fusion imaging Received: 22 June 2025 | Accepted: 17 November 2025 | | --- | | | | Check for updates | Guan-Hua Dun 1,2,4 , Jia-He Zhang1,2,4, Xin-Xing Xie3, Ken Qin1,2, Daixuan Wu1,2, Jia-Li Peng1,2, He-Xuan Wang1,2, Zi Wang1,2, Hua-Jing Fang 3, Dan Xie 1,2 , He Tian 1,2 , Yi Yang1,2 & Tian-Ling Ren 1,2 Bio-inspired photosynapse hardware with wide spectrum information fusion imaging and neuromorphic processing is promising for complex environments perception. However, integrating ultra-wide spectrum responsivity within a single photosynapse device and scaling to a large array has remained challenging. Here, we demonstrate a 64k-scale (65,536 pixels) ultraviolet-tomid-infrared photosynapse array based on carbon nanotube/molybdenum oxide heterojunctions. Typical photosynapse behaviors are conﬁrmed in the wide spectrum region of 365 nm (ultraviolet), 532 nm (visible), 1064 nm (near infrared), and 10.6 μm (mid-infrared). A polydopamine-mediated surface treatment enables homogeneous deposition of carbon nanotube/molybdenum oxide ﬁlms across heterogeneous transistor substrates with 99.96% yield. The photosynapse array enhances image frame similarity from 0.897 to 0.965 in dynamic trajectory prediction. Furthermore, the wide spectrum neuromorphic fusion imaging using photosynapse array achieves 99.58% accuracy in trajectory recognition under challenging conditions, surpassing the 63.93% with visible light alone. This work contributes to paving the way for nextgeneration autonomous perception. Next-generation autonomous systems require robust perception capabilities in diverse and challenging environments such as haze, fog, and nighttime1,2. These systems must reliably recognize objects, interpret scenes, and predict trajectories to ensure safe navigation and real-time decision-making3,4. Current automotive sensing relies predominantly on visible (Vis) spectrum complementary metal oxide semiconductor (CMOS) sensors. For example, dynamic vision sensors (DVS) have demonstrates impressive performance in challenging conditions5,6. However, these sensors remain limited to the visible spectrum, whereas many critical autonomous navigation scenarios require perception beyond visible wavelengths7–11. In this regard, wide spectrum fusion imaging beyond Vis information offers a promising solution12–14. For example, the fusion of infrared (IR) with Vis imaging provides complementary sensing capabilities when Vis spectrum performance degrades, as IR wavelengths experience reduced Mie scattering in particulate matter such as fog and smoke compared to Vis light, enabling more effective transmission in diverse and challenging environments14. However, implementing wide spectrum sensing faces signiﬁcant architectural and device challenges15. The separation between sensing and processing units requires complex preprocessing pipelines for each spectrum band16. A typical pipeline includes sequential stages of sensing, signal conditioning, analog-to-digital conversion (ADC), and data storage/retrieval17. For wide spectrum systems, existing methods 1School of Integrated Circuits, Tsinghua University, Beijing, China. 2Beijing National Research Center for Information Science and Technology (BNRist), Tsinghua University, Beijing, China. 3State Key Laboratory for Manufacturing Systems Engineering, Xi’an Jiaotong University, Xi’an, China. 4These authors contributed equally: Guan-Hua Dun, Jia-He Zhang. e-mail: [dunguanhua@tsinghua.edu.cn](dunguanhua@tsinghua.edu.cn); [xiedan@tsinghua.edu.cn](xiedan@tsinghua.edu.cn); [tianhe88@tsinghua.edu.cn](tianhe88@tsinghua.edu.cn); [yiyang@tsinghua.edu.cn](yiyang@tsinghua.edu.cn); [RenTL@tsinghua.edu.cn](RenTL@tsinghua.edu.cn) primarily require multiple discrete sensors for different spectrum12, necessitating numerous sensor-to-proces

FAQs

What does the 64k-scale photosynapse array achieve in this work?

It provides ultraviolet-to-mid-infrared photosynapse operation across a 64k (65,536-pixel) array and enables wide-spectrum neuromorphic fusion imaging for trajectory recognition and prediction tasks.

Which wavelength ranges are demonstrated to produce typical photosynapse behaviors?

Typical photosynapse behaviors are confirmed at 365 nm (ultraviolet), 532 nm (visible), 1064 nm (near infrared), and 10.6 μm (mid-infrared).

How does the proposed approach improve performance under challenging conditions?

Using polydopamine-mediated surface treatment and wide-spectrum neuromorphic fusion imaging, the array boosts frame similarity for dynamic trajectory prediction from 0.897 to 0.965 and reaches 99.58% trajectory recognition accuracy, outperforming visible light alone.

Sep 30, 2026

Upload by Jacob