Neural network driven optimization of targeted metasurface colors (Conference Presentation)

Tailored sub-wavelength structures of select materials, such as plasmonic metasurfaces and polymeric bragg reflectors, enable strong electromagnetic (EM) interactions that can be tuned to select frequencies across the spectrum. While such systems can be independently tailored towards precise colorimetric outputs in both static and dynamic configurations, new opportunities emerge in hybridized forms that enable more precise spectral control. Optimization of the nanophotonic response was achieved through the use of gaussian process regressions (GPRs) and artificial neural networks (ANNs) to predict colors and spectra of nanoscale metasurface patterns. In addition to the forward model, the inverse functions were also approximated for use in direct design prediction. Spectra were measured, as well as simulated with Lumerical.software Machine-learned models were trained to approximate each respectively. Corresponding experiments were performed to validate and improve the models, isolating metasurface parameters to targeted colorimetric response.