Engineering colors in all-dielectric metasurfaces: metamodeling approach

Abstract

In this Letter, we engineer the colors of all-dielectric metasurfaces by means of a metamodel-based optimization approach. This algorithm combines heuristic optimization and neural networks to retrieve the metasurface’s optimal geometrical parameters that serve to reproduce a prescribed color. The metasurfaces were fabricated and experimentally validated through dark field optical microscope images. We present typical results for periodic arrays of nanoparticles with arbitrary cross section. The approach is well-suited for color reproduction and is computationally inexpensive.

Deep learning in nano-photonics: inverse design and beyond

“…Therefore, replacing conventional physics simulations by surrogate ANNs is a natural solution to speed-up the inverse design of photonic nano-structures via global optimization heuristics [57,58]. This concept has recently been applied by several groups to the design of individual photonic nanostructures or metasurfaces [59][60][61][62][63][64].…”

Deep Learning the Electromagnetic Properties of Metamaterials—A Comprehensive Review

“…In metal‐based metamaterials, for example, DNNs have been used to predict the resonant optical behavior in a number of SRR, [ 111 ] cross‐based, [ 108 ] chiral, [ 111–114 ] and coded metasurfaces. [ 115 ] In ADMs, DL has been applied to problems in color generation, [ 116–118 ] efficient metagrating design, [ 119,120 ] and modeling the complex resonant structure of cylindrical meta‐atoms, [ 109,121,122 ] supercells, [ 14 ] and multilayer nanostructures. [ 123–125 ] In photonic crystals, DNNs have been used to optimize the Q‐factor in nanocavities, [ 99 ] waveguide properties in fibers, [ 126 ] compute the band structure in 1D [ 127 ] and 2D [ 128–130 ] PCs, and predict edge states in topological insulators.…”

“…Instead of using a pure DL model, surrogate DNNs can also be combined with the above mentioned optimization methods to solve AEM inverse design problems, which we categorically label as a hybrid approach. There are many such reports of hybrid optimization models in the DL AEM literature, [ 14,115,116,119,120,132,135,136,140,141,143,175,176,189,197,208–212 ] which we summarize very generally here while noting that the individual models may vary substantially due to the number of optimization techniques available.…”

“…For example, population‐based algorithms in metaheuristics, such as evolutionary strategies (ES), genetic algorithms (GA), and particle swarm optimization (PSO), were shown to be effective in the inverse design of a range of AEM systems including metasurfaces, [ 189,211 ] digitally coded metamaterials, [ 115,208 ] thin films, [ 176,209 ] graphene‐based nanostructures, [ 132 ] and gratings. [ 116,120,142,175 ] Another class of hybrid models involves combining generative networks with a secondary optimization procedure to iteratively refine the solution space, based on either gradient‐free ES [ 189 ] or adjoint‐based topology optimization. [ 190,194,197 ] Other conventional strategies that were also shown to facilitate inverse AEM design with DNNs include Bayesian, [ 140,212,213 ] interior point method, [ 119 ] truncated Newtonian optimization, [ 143 ] as well as search‐based lookup tables.…”

Deep neural networks for the prediction of the optical properties and the free-form inverse design of metamaterials

“…The laws of physics are often described by partial differential equations and a formidable number of computational techniques have been developed to solve partial differential equations given definite values of their coefficients. The inverse problem [4][5][6][7][8][9], determining the coefficients that produce an a priori given solution, is a much more complex problem, but a very relevant one since it may offer technological solutions that go beyond the human imagination. Free-form inverse design in physics, also known as topology optimization, is well established in continuum mechanics [10], but only at its infancy in paradigms of physics that rely on the wave equation [11].…”