An efficient neural optimizer for resonant nanostructures: demonstration of highly-saturated red silicon structural color

Abstract

Freeform nanostructures have the potential to support complex resonances and their interactions, which are crucial for achieving desired spectral responses. However, the design optimization of such structures is nontrivial and computationally intensive. Furthermore, the current "black box" design approaches for freeform nanostructures often neglect the underlying physics. Here, we present a hybrid data-efficient neural optimizer for resonant nanostructures by combining a reinforcement learning algorithm and Powell's local optimization technique. As a case study, we design and experimentally demonstrate silicon nanostructures with a highly-saturated red color. Specifically, we achieved CIE color coordinates of (0.677, 0.304)-close to the ideal Schrodinger's red, with polarization independence, high reflectance (>85%), and a large viewing angle (i.e., up to ~ 25deg). The remarkable performance is attributed to underlying generalized multipolar interferences within each nanostructure rather than the collective array effects. Based on that, we were able to demonstrate pixel size down to ~400 nm, corresponding to a printing resolution of 65,000 pixels per inch. Moreover, the proposed design model requires only ~300 iterations to effectively search a 13-dimensional design space - an order of magnitude more efficient than the previously reported approaches. Our work significantly extends the free-form optical design toolbox for high-performance flat-optical components and metadevices.