Inverse Design of Metasurfaces Based on Coupled-Mode Theory and Adjoint Optimization

Abstract

Metasurfaces typically have sizes much larger than the wavelength yet contain a large number of subwavelength features. Thus, it is difficult to design entire metasurfaces using full-wave simulations. However, without full-wave simulations, most existing design approaches cannot accurately model the interactions between the individual elements comprising the metasurface. Here, we demonstrate an approach for the design of resonant metasurfaces based on coupled-mode theory. Our approach fully describes wave dynamics and coupling in metasurfaces and is much more computationally efficient than full-wave simulations. As an example, we show that the combination of coupled-mode theory and adjoint optimization can be used for the inverse design of high-numerical-aperture (0.9) metalenses with sizes as large as 10000 wavelengths. The computation efficiency of our approach is orders of magnitude faster than full-wave simulations. Complex functionalities such as angle-multiplexed metasurface holograms can also be realized. With its accuracy and efficiency, the proposed framework can be a powerful design tool for large-scale resonant flat-optics devices.

Dielectric metalens for miniaturized imaging systems: progress and challenges

“…c Metasurface design procedure using couple-mode-theory (CMT) and adjoint optimization. Reproduced with permission 191 . Copyright 2020, The Authors, published by Springer Nature.…”

“…An example is shown in Fig. 18c 191 . A physical model, the couple-mode-theory (CMT) model, is first built by previous simulations.…”

Fast multi-source nanophotonic simulations using augmented partial factorization

“…However, LPA is inaccurate whenever the cell-to-cell variation is large [71][72][73][74][75] and cannot describe nonlocal responses [76,77] and metasurfaces that are not based on unit cells [78][79][80]. Coupled-mode theory can model the coupling between meta-atoms [81] but requires isolated resonances with high quality factors. Full-wave simulation remains the gold standard but currently requires enormous computing resources.…”

High Speed Simulation and Freeform Optimization of Nanophotonic Devices with Physics-Augmented Deep Learning

“…Maxwell simulators are essential tools for the characterization and design of electromagnetic systems. These systems operate at frequencies spanning the radiowave to X-ray and include a diversity of antennas, − diffractive surfaces, − metamaterials, − and guided wave-based photonic circuits. − Among the most popular frequency domain Maxwell solvers are the finite element method (FEM) , and finite difference frequency domain (FDFD) algorithms. − In both algorithms, the system domain is subdivided into discrete voxels, and the simulator solves for electromagnetic fields by constructing and inverting a sparse matrix with dimensions proportional to the total number of voxels. While these methods can be used to accurately solve general electromagnetics problems, the time and computation cost of matrix inversion serve as practical bottlenecks for the simulation of large domains and for design tasks, where large numbers of electromagnetic simulations are required to perform iterative optimization.…”