A novel intelligent photonic design method enabled by metamaterials and k-nearest neighbor

Abstract

The utilization of metamaterials plays a pivotal role in integrated photonics. The precise design of metamaterials enables them to finely manipulate light, resulting in an ultra-compact footprint and exceptional performance that cannot be achieved by traditional structures. The conventional methods for metamaterial design, however, encounter challenges from intricate targets. Although attempts have been made to apply inverse design to metamaterials, there is still a need for a highly intelligent, low-computation method, and easy-to-fabricate metamaterial structure. Here, we present an efficient methodology that combines metamaterials, heuristic algorithms, and machine learning to facilitate the rapid development of intricate devices. The method is used to design 1 × N power splitters with arbitrary power ratios, as an application example. Specifically, 1 × 2, 1 × 3, 1 × 4 power splitters with arbitrary ratios are fabricated and experimentally demonstrated. The application of this method in arbitrary power splitter highlights its appropriateness for the design and optimization within integrated photonics devices.