A Generative Machine Learning-Based Approach for Inverse Design of Multilayer Metasurfaces

Abstract

Electromagnetic metasurface design based on farfield constraints without the complete knowledge of the fields on both sides of the metasurface is typically a time consuming and iterative process, which relies heavily on heuristics and ad hoc methods. This paper proposes an end-to-end systematic and efficient approach where the designer inputs high-level farfield constraints such as nulls, sidelobe levels, and main beam level(s); and a 3-layer nonuniform passive, lossless, omega-type bianisotropic electromagnetic metasurface design to satisfy them is returned. The surface parameters to realize the far-field criteria are found using the alternating direction method of multipliers on a homogenized model derived from the method of moments. This model incorporates edge effects of the finite surface and mutual coupling in the inhomogenous impedance sheet. Optimization through the physical unit cell space integrated with machine learning-based surrogate models is used to realize the desired surface parameters from physical meta-atom (or unit cell) designs. Two passive lossless examples with different feeding systems and far-field constraints are shown to demonstrate the effectiveness of this method.

Multiobjective Statistical Learning Optimization of RGB Metalens

“…56,57 Novel approaches, known as generative models, [57][58][59] have been adopted to accelerate the training process and have been utilized to optimize multifunctional nanophotonic devices. [60][61][62] Despite this intense research effort, there is no yet a general relation defining the required number of simulations to train the network according to the number of objectives/parameters, indicating that multiple objective optimizations with training networks still require a considerable amount of resources. Besides, deep network is not inherently an optimization tool and suffers from convergence issues, notably in the case of competing objectives.…”

Towards 3D-Printed Inverse-Designed Metaoptics

“…While advanced (inverse-)design techniques (e.g. density-based topology optimization) have generated designs achieving optimized performance for single-layer structures [8][9][10], it is commonly acknowledged that 3D structures, for instance embodied in a compact (wavelength-scale) arrangement of multiple layers of a dielectric material, would enable a quantum leap in terms of multifunctionality, while conserving the main value proposition of metasurfaces: compactness [7,[11][12][13][14][15][16][17][18]. Despite this observation, there has previously been very little experimental work dedicated to compact multi-layer structures (which we will refer to as volumetric optics).…”

Conditional Generative Adversarial Networks for Inverse Design of Multi-functional Microwave Metasurfaces

“…In these studies, the machine learning algorithms helped reduce the computational time required for the numerical simulations of classical metamaterials and metasurfaces. Inspired by the recent rapid advancements of machine learning tools in optics, several research studies elaborately explored machine learning-based inverse design of microwave metasurfaces to overcome the sophistic challenges of metasurfaces design [37,38,39,40,41,42,43].…”

“…Generative models were used as a solution to this shortcoming in the metasurfaces inverse design [39,40,41]. Hodge et al [39] trained a deep convolutional generative adversarial network on a dataset of known unit cells to generate new unit cells tailored to the design objectives.…”

“…Hodge et al [39] trained a deep convolutional generative adversarial network on a dataset of known unit cells to generate new unit cells tailored to the design objectives. Similarly, a generative machine learning model based on a variational autoencoder for designing multi-layer metasurfaces in transverseelectric (TE) and transverse-magnetic (TM) polarizations was developed [40]. However, these designs were trained using variations in geometric parameters of a limited number of known unit cells such as Jerusalem cross, rectangular and elliptical patches, and ring resonators that restrict the access to various EM functionalities by the designed metasurfaces.…”