Dynamic Metasurface Structural Colour: Lateral Hybrid System

Abstract

This thesis explores the development and optimization of tunable structural colours using advanced metasurface configurations and deep learning techniques. Traditional metasurfaces, typically composed of single-material resonators, face significant limitations in achieving high sensitivity and tunability. These metasurfaces interact with light base on their geometry, material and surface profile, which restricts their ability to dynamically adjust optical properties. This limited tunability presents challenges for applications requiring adaptive optical functionalities. Although single-material metasurfaces, such as those made from noble metals (eg. Al, Au) and high refractive index dielectrics (eg. Si, Ge), can provide some level of performance, they fall short when it comes to achieving the dynamic and highly responsive tuning necessary for more advanced applications. To address these limitations, researchers have increasingly turned to hybrid metasurfaces, which combine multiple materials to leverage the unique optical properties of each. By integrating materials with distinct plasmonic or dielectric properties, hybrid metasurfaces offer the potential for enhanced performance through the coupling of different resonant modes. However, despite these advancements, traditional hybrid metasurfaces still exhibit certain constraints, such as limited sensitivity and adaptability, due to their reliance on layered configurations. In response to the limitations of tunability and sensitivity in traditional layered metasurfaces, this thesis introduces a novel lateral hybrid metasurface design that combines dielectric and metal resonators in a planar arrangement In this study, a Si-Au hybrid metasurface was used as a case study. Unlike conventional layered metasurfaces, which typically exhibit sensitivity (change in resonating wavelength over change in lattice size) values around 2 or lower, the lateral hybrid configuration achieves significantly higher sensitivity, reaching up to 26. Additionally, the sRGB coverage of the Si-Au lateral system reached 14%, compared to just 1% for its layered counterpart. This lateral arrangement offers more precise control over light-matter interactions and results in a broader and more dynamic range of tunable colours. The lateral hybrid design effectively addresses the limitations of both single material and traditional layered metasurfaces, offering a robust and highly sensitive solution for real-time tunable optical performance. Deep learning is central to this research, significantly enhancing both forward and inverse design processes. Forward design models predict the optical performance of metasurfaces based on specific input parameters, while inverse design models optimize configurations to achieve desired optical properties. By utilizing AI, the design process is vastly more efficient, with forward predictions up to 105 times faster than conventional FEM simulations. The inverse model enables custom design of metamaterials without relying on cumbersome trail and error approach, automating and optimizing what was previously dominated by human intuition. The deep learning model developed in this thesis achieved up to 90% accuracy, leading to the design of an efficient and highly tunable lateral hybrid metasurface. This approach eliminates guesswork and significantly reduces the time and resources required for metasurface design. The lateral hybrid system proposed in this dissertation could facilitate the development of metasurfaces with tailored properties and extend their applications beyond the laboratory to real-world environment.