Neutron: Neural Particle Swarm Optimization for Material-Aware Inverse Design of Structural Color

Haozhu Wang, L. Jay Guo

Abstract: Designing optical structures for generating structural colors is challenging due to the complex relationship between the optical structures and the color perceived by human eyes. Machine learning-based inverse approaches have been applied to expedite the structural color design process due to their exceptional ability to learn complex relationships from data. However, existing methods often assume that the materials for the optical structures are fixed, which could lead to sub-optimal performance of color generation due to the inability to search for and include the best materials. To address this issue, a hybrid approach termed Neural Particle Swarm Optimization is proposed in this paper. The proposed method combines a multitask mixture density network that predicts the materials and associated design parameter values and particle swarm optimization to finetune the predicted design parameter values. The proposed method demonstrates exceptional design accuracy and efficiency on two practical tasks of designing environmental-friendly alternatives to chrome coatings and reconstructing pictures with structural colors based on multilayer optical thin films. With the proposed method, several designs that could be used to replace the chrome coating have been discovered; pictures with more than 200,000 pixels can be reconstructed within 2 to 3 hours with an almost unnoticeable difference from the original picture.