Statistical learning approaches for the design of highly effecient gradient metasurfaces at visible regime

Abstract : In this work, we exploit two advanced optimization techniques based on statistical learning and evolutionary strategies together with a fullwave high order Discontinuous Galerkin Time-Domain (DGTD) solver in order to optimize 3D gradient metasurfaces with different shapes. Unlike what is usually provided in the literature, first we introduce rigorously to the community the methodologies of our advanced optimization techniques that outperform most of the available techniques in the literature, especially for complex problems that contain several global minima/maxima. Second, we demonstrate both numerically and experimentally optimal designs for 3D real life gradient meta-surfaces based on GaN semiconductor at the visible regime. Our numerical results reveal that for spherical shaped antenna, one can achieve more than 85% of diffraction efficiency for both TM and TE polarization using only 150 solver runs. In addition, for rectangular shaped antenna, an efficiency above 88% can be achieved for TM polarization using less than 150 fullwave simulations. To the best of our knowledge, this is the highest diffraction efficiency reported so far at the visible regime for real life 3D structures.
Complete list of metadatas

Cited literature [44 references]  Display  Hide  Download

https://hal.archives-ouvertes.fr/hal-02156881
Contributor : Mahmoud Elsawy <>
Submitted on : Friday, June 14, 2019 - 6:15:57 PM
Last modification on : Thursday, July 11, 2019 - 1:32:42 AM

File

arXiv_submission_optimization_...
Files produced by the author(s)

Identifiers

  • HAL Id : hal-02156881, version 1

Collections

Citation

Mahmoud Elsawy, Stéphane Lanteri, Régis Duvigneau, Peinan Ni, Gauthier Brière, et al.. Statistical learning approaches for the design of highly effecient gradient metasurfaces at visible regime. 2019. ⟨hal-02156881⟩

Share

Metrics

Record views

208

Files downloads

108