EUVIMEDEUVIMED entdecken
Crossref

Super-compact scalable optical matrix operations by modularized inverse design

Meizhen Jiang, Lu He, Xiangdong Zhang

Photonics Research · 01.11.2026 · 14 (11) · 4873

DOI: 10.1364/prj.605078 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Photonics Research, online erschienen am 30.10.2026. Verfasst von Meizhen Jiang und 2 weiteren Autor:innen. Die Arbeit nennt 72 Literaturangaben. Gefördert von: National Key R&D Program of China, National Natural Science Foundation of China, China Postdoctoral Science Foundation.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Optical matrix computing, as a pivotal technology for overcoming the limitations of conventional electronic computation, offers distinct advantages such as high parallelism, low power consumption, and large bandwidth. However, existing on-chip photonic devices face the challenges of large footprints and poor scalability, and inverse-design methodologies suffer from excessive design time and computational resource consumption when applied to large-scale matrix synthesis. To address these challenges, we propose an innovative “modularized inverse-design” approach, which optimizes 2×2 unitary-matrix cells and constructs a database containing thousands of unitary matrices with distinct beam-splitting ratios, enabling the flexible assembly of optical real orthogonal matrices of arbitrary dimensions. Leveraging this method, we realize super-compact N × N optical unitary matrices whose footprints are several orders of magnitude smaller than those of conventional Mach–Zehnder interferometer architectures for fixed matrix transformations. We successfully implement the optical vector-matrix multiplications of 10×10, 20×20, and 30×30 dimensions and perform classification tasks on the MNIST and Fashion-MNIST datasets. This scheme markedly enhances both the integration density and design efficiency of optical matrix computing, presenting an efficient and scalable solution for future large-scale, highly integrated artificial intelligence chips.

Förderung

  • National Key R&D Program of China
  • National Natural Science Foundation of China
  • China Postdoctoral Science Foundation
  • Postdoctoral Fellowship Program
  • China National Postdoctoral Program for Innovative Talents

Zitieren

APA

Jiang, M., He, L., & Zhang, X. (2026). Super-compact scalable optical matrix operations by modularized inverse design. Photonics Research, 14(11), 4873. https://doi.org/10.1364/prj.605078

Vancouver

Jiang M, He L, Zhang X. Super-compact scalable optical matrix operations by modularized inverse design. Photonics Research. 2026;14(11):4873. doi:10.1364/prj.605078

BibTeX
@article{Jiang2026,
  author = {Jiang, Meizhen and He, Lu and Zhang, Xiangdong},
  title = {Super-compact scalable optical matrix operations by modularized inverse design},
  journal = {Photonics Research},
  year = {2026},
  volume = {14},
  number = {11},
  pages = {4873},
  doi = {10.1364/prj.605078}
}
Kostenlos starten →