Physics-constrained active learning diffusion framework for inverse design of MIM metasurfaces
Optics Express · 10.08.2026 · 34 (16) · 30165
DOI: 10.1364/oe.605206 ↗Kurz gefasst
Fachartikel in der Zeitschrift Optics Express, online erschienen am 04.08.2026. Verfasst von Tan Qu und 4 weiteren Autor:innen. Die Arbeit nennt 28 Literaturangaben. Gefördert von: National Natural Science Foundation of China.
Abstract (Originalsprache)
The inverse design of metal-insulator-metal (MIM) remains challenging due to the inefficient parameter sweeping and stringent physical requirements. While deep learning improved efficiency, it often lacks interpretability and struggles with limited data. The paper proposes a physical-informed generative framework that synergizes diffusion models with active learning to achieve high-performance MIM absorbers. Starting from a limited dataset, the diffusion model can learn the distribution characteristics of high-performance designs in the parameter space and generate candidate asymmetric structures with high absorption potential. By embedding geometric constraints as well as physical metrics including peak absorption, quality factor and figure of merit (FoM) into the optimizing loop, our method improved the training data quality and overall structural performance. The generated structures exhibit good geometric continuity, structural stability, narrowband single-peak absorption, and multi-resonance responses, with overall performance superior to the initial dataset. Electric field distributions and equivalent optical circuit analysis further reveal the resonance modes and absorption mechanisms, providing physical interpretability for the data-driven design process. The designed metasurfaces demonstrate excellent narrowband and multi-peak absorption, establishing a powerful paradigm for efficient photonic device design.
Förderung
- National Natural Science Foundation of China
Zitieren
APA
Qu, T., Zhang, J., Zhang, Y., Wu, J., & Wu, Z. (2026). Physics-constrained active learning diffusion framework for inverse design of MIM metasurfaces. Optics Express, 34(16), 30165. https://doi.org/10.1364/oe.605206
Vancouver
Qu T, Zhang J, Zhang Y, Wu J, Wu Z. Physics-constrained active learning diffusion framework for inverse design of MIM metasurfaces. Optics Express. 2026;34(16):30165. doi:10.1364/oe.605206
BibTeX
@article{Qu2026,
author = {Qu, Tan and Zhang, Jiayi and Zhang, Yan and Wu, Jiaji and Wu, Zhensen},
title = {Physics-constrained active learning diffusion framework for inverse design of MIM metasurfaces},
journal = {Optics Express},
year = {2026},
volume = {34},
number = {16},
pages = {30165},
doi = {10.1364/oe.605206}
}
EUVIMED