Announcing a new publication from Opto-Electronic Advances; DOI 10.29026/oea.2026.250350.

The rapid advancement of artificial intelligence, big data analytics, and scientific computing has imposed unprecedented demands on computational power, pushing traditional electronic architectures to their physical limits. The slowdown of Moore’s Law and the escalating “memory wall” bottleneck have made it increasingly difficult for von Neumann-based processors to deliver the energy efficiency and throughput required by modern deep neural networks. In contrast, the human brain achieves exceptional computational efficiency and cognitive capability with remarkably low power consumption, inspiring the development of novel computing paradigms. Neuromorphic computing, which emulates the structure and information-processing principles of the brain, integrates interdisciplinary advances from neuroscience, computational neuroscience, and materials science, etc. aiming to achieve high-efficiency, low-latency computing systems.

Optical platforms, with their inherent advantages of broad bandwidth, low latency, high parallelism, and low power consumption, have emerged as a promising alternative in the post-Moore era, infusing new momentum into neuromorphic computing. By encoding information across multiple physical dimensions of light, e.g., wavelength, amplitude, phase, and polarization, optical systems can perform large-scale computations within a single time step. Photonic neuromorphic computing thus combines the architectural principles of neuromorphic computing with the intrinsic strengths of photonics, in order to realize high-speed, low-power, and highly integrable brain-inspired systems.

Despite notable progress, current photonic neuromorphic platforms still face limitations in achieving universal acceleration. On one hand, single optimization strategies lack adaptability across diverse hardware architectures. On the other hand, reducing samples in pursuit of computational speed often comes with a loss of accuracy, that is, an inherent trade-off between computational throughput and accuracy. Therefore, developing a solution that enables universal acceleration without sacrificing accuracy holds significant theoretical and practical importance, which will offer a pathway to overcome the rate-performance bottleneck and advance photonic neuromorphic computing toward practical and scalable applications.

The research group of Prof. Nianqiang Li and Yu Huang, from the School of Optoelectronic Science and Engineering at Soochow University, proposed a performance-compensated acceleration strategy based on a compression-decompression architecture, aiming to overcome the inherent limitations of conventional photonic neuromorphic computing in information processing speed and system scalability.

The core innovation of this framework lies in the construction of an adaptive, eigenvector-driven compression-decompression synergistic network. In the preprocessing stage, the compression process projects input data into a low-dimensional subspace highly relevant to the target task through eigenvector analysis, thereby significantly reducing the data volume processed by the hidden layer while preserving critical features. The decompression stage then effectively compensates for information loss incurred during compression to improve final performance.

To validate the generality and effectiveness of the proposed framework, the research group conducted experiments on two representative photonic neuromorphic platforms: a continuous-wave system based on commercial off-the-shelf lasers and a spiking system built on a specifically fabricated photonic neuro-synaptic chip. Experimental results demonstrate that the framework enables substantial speed improvements in photonic neuromorphic computing within the optical domain. The system achieves a recognition speed of 100 million images per second for image classification tasks and a processing throughput of 600 million data points per second for time-series tasks. Notably, while latency and hardware resource requirements are reduced by an order of magnitude, the proposed system’s recognition accuracy and temporal processing capability can still be effectively restored to levels comparable to those of uncompressed architectures. These results confirm that the compression-decompression framework can achieve order-of-magnitude acceleration while effectively compensating for information loss introduced by compression, thereby breaking the inherent trade-off between speed and performance in conventional photonic neuromorphic computing. This work establishes a universal acceleration paradigm for high-throughput, low-power photonic neuromorphic chips and offers a novel pathway for software-hardware co-development.

Keywords: reservoir computing, compression-decompression framework, neuromorphic computing accelerator, semiconductor laser
# # # # # #

Nianqiang Li Professor and Doctoral Supervisor, Associate Dean of the School of Optoelectronic Science and Engineering at Soochow University. He is a recipient of the "Shuangchuang Talent" program of Jiangsu Province, a core member of a Jiangsu Province Shuangchuang Team, the winner of the Young Science and Technology Award from the Optical Society of Jiangsu Province, and first Prize of the 11th Natural Science Award of the Chinese Society for Optical Engineering. He has undertaken over 10 research projects, including 3 projects from the National Natural Science Foundation of China (NSFC) as principal investigator and 1 sub-project from the National Key Research and Development Program of the Ministry of Science and Technology. As first author or corresponding author, he has published over 100 SCI papers in prestigious academic journals such as Opto-Electronic Advances, Ultrafast Science, ACS Photonics, Photonics Research, APL Photonics, Neural Networks, and Chaos, Solitons & Fractals. Several of these papers are ESI Highly Cited Papers, Editor’s Picks, or featured as covers of IEEE JQE. He has delivered nearly 20 invited and oral presentations and has applied for or been granted over 20 national invention patents. Current research interests include laser chaos generation and applications, optical spiking neural networks, reservoir computing, and microwave signal generation and applications. Doctoral students under his supervision have received the 18th Wang Daheng Optics Award (Student Award), the 9th and 11th Young Optical Science and Technology Awards from the Optical Society of Jiangsu Province, and the 2024 CAST Program for Young Talents Support (Doctoral Student Special Project). (Homepage: https://web.suda.edu.cn/nli/ )

Yu Huang Postdoctoral Fellow and Associate Research Fellow. Recipient of the 18th Wang Daheng Optics Award (Student Award) and the 9th Young Optical Science and Technology Award (University Student Category) from the Optical Society of Jiangsu Province. He leads several research initiatives, including a Young Scientists Fund project from the National Natural Science Foundation of China, a Young Provincial Fund project from the Jiangsu Provincial Natural Science Foundation, and a China Postdoctoral Science Foundation General Program. As first author or corresponding author, he has published over 30 SCI papers in major academic journals such as ACS Photonics, Photonics Research, and Chaos, Solitons & Fractals. Current research interests include laser chaos generation and applications, convolutional neural networks, and reservoir computing. (Homepage: https://web.suda.edu.cn/hy2/ )
# # # # # #
Opto-Electronic Advances (OEA) is a high-impact, open access, peer reviewed SCI journal with an impact factor of 22.4 (Journal Citation Reports 2024). OEA has been indexed in SCI, EI, DOAJ, Scopus, CA and ICI databases, and expanded its Editorial Board to 41 members from 17 countries.
# # # # # #

More information: http://www.oejournal.org/oea
Editorial Board: http://www.oejournal.org/oea/editorialboard/list
All issues available in the online archive (http://www.oejournal.org/oea/archive).
Submissions to OEA may be made using ScholarOne (https://mc03.manuscriptcentral.com/oea).
ISSN: 2096-4579
CN: 51-1781/TN
Contact Us: oea@ioe.ac.cn
Twitter: @OptoElectronAdv (https://twitter.com/OptoElectronAdv?lang=en)
WeChat: OE_Journal
# # # # # #

Zhou CD, Cai DY, Huang Y et al. A compression-decompression framework for universal acceleration of photonic neuromorphic computing. Opto-Electron Adv, 250350 (2026). DOI: 10.29026/oea.2026.250350