SLMGAN: Single-layer metasurface design with symmetrical free-form patterns using generative adversarial networks

Page view(s)
8
Checked on Apr 09, 2025
SLMGAN: Single-layer metasurface design with symmetrical free-form patterns using generative adversarial networks
Title:
SLMGAN: Single-layer metasurface design with symmetrical free-form patterns using generative adversarial networks
Journal Title:
Applied Soft Computing
Keywords:
Publication Date:
24 September 2022
Citation:
Dai, M., Jiang, Y., Yang, F., Xu, X., Zhao, W., Dao, M. H., & Liu, Y. (2022). SLMGAN: Single-layer metasurface design with symmetrical free-form patterns using generative adversarial networks. Applied Soft Computing, 130, 109646. https://doi.org/10.1016/j.asoc.2022.109646
Abstract:
The metasurfaces offering the required spectral responses have ushered in a revolution of manipulating the light in a prescribed manner. A single-layer metasurface design is more appealing than a multi-layer one due to the fabrication complexities. To date, various research groups have explored on architected metasurfaces with general shapes of cubes, crosses, or octothorpes, while a few works utilized evolutionary algorithms to search for metasurfaces with free-form patterns, which relied on the quality of the initial guess. In this paper, a solution is presented to replace the intuition-based approach with generative adversarial networks. The constructed generative networks mathematically formulate the virtual mappings between the pairs of optical spectra and symmetrical patterns with user-defined geometric structures. When fed a time sequence of spectra, the designed networks assimilate the physical property and generate on-demand patterns to match the desired responses. The output patterns are proved to yield matching optical responses with an average accuracy of 0.9. Generative Adversarial Networks are firstly applied to single-layer metasurface designs with symmetrical free-form patterns for desired optical spectra in an inverse-design system.
License type:
Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0)
Funding Info:
This research / project is supported by the A*STAR - AME Programmatic Fund
Grant Reference no. : A20H5b0142
Description:
ISSN:
1568-4946
Files uploaded:

File Size Format Action
elsarticle-template-num-namesclean.pdf 14.52 MB PDF Open