Dual-band optical collimator based on deep-learning designed, fabrication-friendly metasurfaces

Abstract

Metasurfaces, which consist of arrays of ultrathin planar nanostructures (also known as “meta-atoms”), offer immense potential for use in high-performance optical devices through the precise manipulation of electromagnetic waves with subwavelength spatial resolution. However, designing meta-atom structures that simultaneously meet multiple functional requirements (e.g., for multiband or multiangle operation) is an arduous task that poses a significant design burden. Therefore, it is essential to establish a robust method for producing intricate meta-atom structures as functional devices. To address this issue, we developed a rapid construction method for a multifunctional and fabrication-friendly meta-atom library using deep neural networks coupled with a meta-atom selector that accounts for realistic fabrication constraints. To validate the proposed method, we successfully applied the approach to experimentally demonstrate a dual-band metasurface collimator based on complex free-form meta-atoms. Our results qualify the proposed method as an efficient and reliable solution for designing complex meta-atom structures in high-performance optical device implementations.

Keywords

• deep learning
• fabrication tolerance
• metasurface
• multiband
• predictive neural network