Global optimization of metasurface designs using statistical learning methods

Area of Science:

• Photonics and Nanotechnology
• Computational Electromagnetics
• Materials Science

Background:

• Metasurface performance is critical for optical systems, but current designs often neglect near-field interactions.
• Existing design methods rely on extensive parameter searches, limiting optimization.
• Accurate simulation of optical scattering response is essential for device development.

Purpose of the Study:

• To optimize phase gradient metasurfaces using advanced computational techniques.
• To explore the impact of near-field interactions on metasurface performance.
• To achieve high diffraction efficiency for visible light applications.

Main Methods:

• Utilizing statistical learning and evolutionary strategies for optimization.
• Employing a full-wave high-order Discontinuous Galerkin Time-Domain (DGTD) solver.
• Comparing performance against existing metasurface designs in the literature.

Main Results:

• Demonstrated superior performance of statistical learning and evolutionary strategies.
• Achieved over 88% diffraction efficiency for TM polarization with rectangular nanopillars.
• Achieved over 85% diffraction efficiency for TM and TE polarization with cylindrical nanopillars.
• Optimized designs using a limited number of full-wave simulations (150).

Conclusions:

• Advanced optimization techniques effectively enhance metasurface performance.
• Near-field interactions play a crucial role in metasurface design.
• The developed methods offer a pathway to record diffraction efficiencies for visible light metasurfaces.