The data-science revolution is poised to transform the way photonic systems are simulated and designed. Photonic systems are, in many ways, an ideal substrate for machine learning: the objective of much of computational electromagnetics is the capture of nonlinear relationships in high-dimensional spaces, which is the core strength of neural networks. Additionally, the mainstream availability of Maxwell solvers makes the training and evaluation of neural networks broadly accessible and tailorable to specific problems. In this Review, we show how deep neural networks, configured as discriminative networks, can learn from training sets and operate as high-speed surrogate electromagnetic solvers. We also examine how deep generative networks can learn geometric features in device distributions and even be configured to serve as robust global optimizers. Fundamental data-science concepts framed within the context of photonics are also discussed, including the network-training process, delineation of different network classes and architectures, and dimensionality reduction. Neural networks can capture nonlinear relationships in high-dimensional spaces and are powerful tools for photonic-system modelling. This Review discusses how deep neural networks can serve as surrogate electromagnetic solvers, inverse modelling tools and global device optimizers.

Deep neural networks for the evaluation and design of photonic devices

Optimization techniques have been indispensable for designing high-performance meta-devices targeted to a wide range of applications. In fact, today optimization is no longer an afterthought and is a fundamental tool for many optical and RF designers. Still, many devices presented in recent literature do not take advantage of optimization techniques. This paper seeks to address this by presenting both an introduction to and a review of several of the most popular techniques currently used for meta-device design. Additionally, emerging techniques like topology optimization and multi-objective optimization and their context to device design are thoroughly discussed. Moreover, attention is given to future directions in meta-device optimization such as surrogate-modeling and deep learning which have the potential to disrupt the fields of optical and radio frequency (RF) inverse-design. Finally, many design examples from the literature are presented and a flow-chart that provides guidance on how best to apply these optimization algorithms to a given problem is provided for the reader.

Review of numerical optimization techniques for meta-device design [Invited]

Metasurfaces have promised great possibilities in full control of the electromagnetic wavefront by spatially manipulating the phase characteristics across the interface. Here, we report a scheme to realize broadband backward scattering reduction through diffusion-like microwave reflection by utilizing a flexible indium-tin-oxide (ITO)-based ultrathin coding metasurface (less than 0.1 wavelength thick) with high optical transparence. The diffusion-like scattering is caused by the destructive interference of the scattered far-field electromagnetic wave, which is further attributed to the randomly distributed reflection phases on the metasurface composed of pre-designed meta-atoms arranged with a computer-generated pseudorandom coding sequence. Both simulation and measurement on fabricated prototype sample have been carried out to validate its performance, demonstrating a polarization-independent broadband (nearly from 8 GHz to 15 GHz) 10 dB scattering reduction with good oblique performance. The excellent performances can also be preserved to conformal cases when the flexible metasurface is uniformly wrapped around a metallic cylinder. The proposed metasurface may create new opportunities to tailor the exotic microwave scattering features with simultaneously high transmittance in visible frequencies, which could provide crucial benefits in many practical uses, such as window and solar panel applications.

Coding metasurface for broadband microwave scattering reduction with optical transparency.

Coding metasurfaces, based on the combination of two basic unit cells with out-of-phase responses, have been the subject of many recent studies aimed at achieving diffuse scattering, with potential applications to diverse fields ranging from radar-signature control to computational imaging. Here, via a theoretical study of the relevant scaling-laws, the physical mechanism underlying the scattering-signature reduction is elucidated, and some absolute and realistic bounds are analytically derived. Moreover, a simple, deterministic suboptimal design strategy is introduced that yields results comparable with those typically obtained by approaches based on brute-force numerical optimization, at a negligible fraction of their computational burden, thereby paving the way to the design of structures with arbitrarily large electrical size. Results are corroborated by rigorous full-wave numerical simulations and microwave experiments, and may be of interest in a variety of application fields, such as the design of low-scattering targets and illumination apertures for computational imaging, not necessarily restricted to electromagnetic scenarios.

Coding Metasurfaces for Diffuse Scattering: Scaling Laws, Bounds, and Suboptimal Design

Diffusive scatterings of electromagnetic (EM) waves by a thin screen are important in many applications, but available approaches cannot ensure uniform angular distributions of low-intensity scatterings without time-consuming optimizations. Here, we propose a robust and deterministic approach to design metasurfaces to achieve polarization-independent diffusive scatterings of EM waves within an ultrabroad frequency band and for wide-range of incident angles. Our key idea is to use high-efficiency Pancharatnam–Berry meta-atoms to form subarrays exhibiting focusing reflection-phase profiles, that can guarantee nearly uniform diffusive scatterings for arbitrarily polarized EM waves. As an illustration, we design and fabricate two metasurfaces and experimentally characterize their wave-diffusion properties in C, X, and Ku bands. Theoretical, numerical and experimental results demonstrate that our approach can diffuse the incident energy much more uniformly than available approaches based on the uniform-phase s...

Deterministic Approach to Achieve Broadband Polarization-Independent Diffusive Scatterings Based on Metasurfaces

We propose a new strategy to design broadband and wide angle diffusion metasurfaces. An anisotropic structure which has opposite phases under x- and y-polarized incidence is employed as the "0" and "1" elements base on the concept of coding metamaterial. To obtain a uniform backward scattering under normal incidence, Simulated Annealing algorithm is utilized in this paper to calculate the optimal layout. The proposed method provides an efficient way to design diffusion metasurface with a simple structure, which has been proved by both simulations and measurements.

Broadband diffusion metasurface based on a single anisotropic element and optimized by the Simulated Annealing algorithm.

A low-radar cross section (RCS) coding metasurface (MS) with properties of absorption and diffusion for both normal and oblique incidences is proposed in this paper. The coding MS is composed of a miniaturized perfect metamaterial absorber (PMA) and a wideband artificial magnetic conductor (AMC) in a shared aperture. In addition, to avoid strong scattering energy appearing at specific directions, genetic algorithm (GA) is adopted to search the optimal layout of the two MS elements. Simulated and experimental results confirm the properties of coding MS and indicate that the 6-dB RCS reduction bands under TEand TM-polarized normal incident waves are 6.28 GHz–9.16 GHz and 6.33 GHz–9.41 GHz, respectively.

https://www.jpier.org/pierc/pierc75/18.17041201.pdf

A Coding Metasurface with Properties of Absorption and Diffusion for RCS Reduction

The invention discloses a quasi-surface plasmon ultra-wide band low-scattering slot antenna. Radiation apertures have circular structures in same area; the radius of a circle is r1=68mm; the thickness of a dielectric slab of an SSPP slot antenna is 1mm; a metal coats a rear feeding part covering an area of 20mm*40mm on the SSPP slot antenna; a 26mm*2mm gap is formed on a central position of the SSPP slot antenna; the rest positions are not coated with the metal; a dielectric constant of the dielectric slab is 2.65; the loss tangent is less than 0.001. The quasi-surface plasmon ultra-wide band low-scattering slot antenna is realized by the invention; the antenna has an obvious RCS (Radar Cross Section) reduction effect within 1GHz-10GHz; the maximum reduction volume of MRCS is 16.93dB or above; the relative bandwidth of the SSPP slot antenna is 9.4%; the frequency coverage is 5.25GHz-5.77GHz; compared with the traditional antenna, the SSPP antenna has the advantage that the beam width is at least widened for 65 degrees.

Quasi-surface plasmon ultra-wide band low-scattering slot antenna

An optimal-arranged metasurface for broadband and wide-angle RCS reduction is proposed in this paper. To avoid directed scattering beams caused by the chessboard-like metasurface, the optimal arrangement for diffusion effect is searched out using genetic optimization algorithm. Furthermore, the proposed metasurface can obtain a wider 10-dB RCS reduction band which ranges from 5.16 GHz −7.94GHz (42.3% relative bandwidth) compared with the chessboard which ranges from 5.92GHz-6.1GHz (3.0% relative bandwidth). The scattering performances for the oblique incidence are also investigated in the full wave simulation, in which a more than 6-dB bistatic RCS reduction can be achieved from 5.14GHz-7.82GHz for both horizontal and vertical polarizations.

Tong Han

Papers

Broadband diffusion metasurface based on a single anisotropic element and optimized by the Simulated Annealing algorithm.

A Coding Metasurface with Properties of Absorption and Diffusion for RCS Reduction

Quasi-surface plasmon ultra-wide band low-scattering slot antenna

Broadband and wide-angle RCS reduction based on an optimal-arranged metasurface