A Data-Mining-Assisted Design of Structural Colors on Diamond Metasurfaces

Artificial colors generated by light-matter interactions are widely achieved using metal or dielectric nanostructures. However, those systems are susceptible to reported issues like high optical loss and insufficient robustness, limiting their practical deployment. To address these limitations, the concept of a low-loss and high-saturation color pixel design based on diamond metasurfaces is proposed and demonstrated. The numerical simulation and analytic evaluation are performed to demonstrate the feasibility of realizing high-quality structural colors with diamond metasurfaces. Moreover, a tensor completion-based algorithm is introduced to assist the design of vivid colors, shortening the required runtime of simulation by four orders of magnitude. Such a tensor-based algorithm can precisely predict a complete set of generated colors through partially simulated conditions. Particularly, when the missing rate is lower than 0.8, the obtained prediction accuracy is no less than 85.1%, 90.6%, and 90.9% for cube-, cylinder-, and cone-shaped diamond metasurfaces, respectively. The current study provides a solid theoretical foundation for achieving all-diamond optical metasurfaces, and the present data-mining approach could be expanded to other photonic structure designs and parametric optimization.