Deep-learning-enabled inverse engineering of multi-wavelength invisibility-to-superscattering switching with phase-change materials

Multifrequency spherical cloak in microwave frequencies enabled by deep learning

Designing invisibility devices for required frequency bands is important in anti-detection methods in various fields such as communications, construction, and others. However, traditional design methods are time-consuming, with manual adjustment of parameters and continuous trial and error. Fortunately, the data-driven approach based on deep learning has revolutionized the field. In this article, we demonstrate that utilizing a trained deep neural network can handle the problem efficiently. It can accurately predict the scattering cross section (SCS) of a multilayer sphere under given structural parameters and reversely design the structural parameters corresponding to the target spectrum. Using the predicted parameters, three-dimensional full-wave simulations are conducted, achieving perfect invisibility performance under transverse electric (TE), transverse magnetic (TM) waves, and point source illumination at multiple frequencies. Our results provide a compelling case for utilizing deep learning in cloaking design.

Large-scale photonic inverse design: computational challenges and breakthroughs

Recent advancements in inverse design approaches, exemplified by their large-scale optimization of all geometrical degrees of freedom, have provided a significant paradigm shift in photonic design. However, these innovative strategies still require full-wave Maxwell solutions to compute the gradients concerning the desired figure of merit, imposing, prohibitive computational demands on conventional computing platforms. This review analyzes the computational challenges associated with the design of large-scale photonic structures. It delves into the adequacy of various electromagnetic solvers for large-scale designs, from conventional to neural network-based solvers, and discusses their suitability and limitations. Furthermore, this review evaluates the research on optimization techniques, analyzes their advantages and disadvantages in large-scale applications, and sheds light on cutting-edge studies that combine neural networks with inverse design for large-scale applications. Through this comprehensive examination, this review aims to provide insights into navigating the landscape of large-scale design and advocate for strategic advancements in optimization methods, solver selection, and the integration of neural networks to overcome computational barriers, thereby guiding future advancements in large-scale photonic design.

Exploring localized ENZ resonances and their role in superscattering, wideband invisibility, and tunable scattering

While the role and manifestations of the localized surface plasmon resonances (LSPRs) in anomalous scattering, like superscattering and invisibility, are quite well explored, the existence, appearance, and possible contribution of localized epsilon-near-zero (ENZ) resonances still invoke careful exploration. In this paper, that is done along with a comparison of the resonances of two types in the case of thin-wall cylinders made of lossy and loss-compensated dispersive materials. It is shown that the localized ENZ resonances exist and appear very close to the zero-permittivity regime, i.e., at near-zero but yet negative permittivity that is similar to the ENZ modes in thin planar films. Near- and far-field characteristics of the superscattering modes are investigated. The results indicate that the scattering regimes arising due to LSPRs and localized ENZ resonances are distinguishable in terms of the basic field features inside and around the scatterer and differ in their contribution to the resulting scattering mechanism, e.g., in terms of the occupied frequency and permittivity ranges as well as the sensitivity to the wall thickness variations. When the losses are either weak or tend to zero due to the doping with gain enabling impurities, the sharp peaks of the scattering cross-section that are yielded by the resonances can be said to be embedded into the otherwise wide invisibility range. In the case of lossy material, a wide and continuous invisibility range is shown to appear not only due to a small total volume of the scatterer in the nonresonant regime, but also because high-Q superscattering modes are suppressed by the losses. For numerical demonstration, indium antimonide, a natural lossy material, and a hypothetical, properly doped material with the same real part of the permittivity but lower or zero losses are considered. In the latter case, variations of permittivity with a control parameter can be adjusted in such a way that transitions from one superscattering mode to another can be achieved. In turn, transition from the strong-scattering to the invisibility regime is possible even for the original lossy material. The basic properties of the studied superscattering modes may be replicable in artificial structures comprising natural low-loss materials.

Machine learning enabled rational design for dynamic thermal emitters with phase change materials

Dynamic thermal emitters have attracted considerable attention due to their potential in widespread applications such as radiative cooling, thermal switching, and adaptive camouflage. However, the state-of-art performances of dynamic emitters are still far below expectations. Here, customized to the special and stringent requirement of dynamic emitters, a neural network model is developed to effectively bridge the structural and spectral spaces and further realizes the inverse design with coupling to genetic algorithms, which considers the broadband spectral responses in different phase-states and utilizes comprehensive measures to ensure the modeling accuracy and computational speed. Besides achieving an outstanding emittance tunability of 0.8, the physics and empirical rules have also been mined qualitatively through decision trees and gradient analyses. The study demonstrates the feasibility of using machine learning to obtain the near-perfect performance of dynamic emitters, as well as guiding the design of other thermal and photonic nanostructures with multifunctions.

A deep neural network for general scattering matrix

The scattering matrix is the mathematical representation of the scattering characteristics of any scatterer. Nevertheless, except for scatterers with high symmetry like spheres or cylinders, the scattering matrix does not have any analytical forms and thus can only be calculated numerically, which requires heavy computation. Here, we have developed a well-trained deep neural network (DNN) that can calculate the scattering matrix of scatterers without symmetry at a speed thousands of times faster than that of finite element solvers. Interestingly, the scattering matrix obtained from the DNN inherently satisfies the fundamental physical principles, including energy conservation, time reversal and reciprocity. Moreover, inverse design based on the DNN is made possible by applying the gradient descent algorithm. Finally, we demonstrate an application of the DNN, which is to design scatterers with desired scattering properties under special conditions. Our work proposes a convenient solution of deep learning for scattering problems.

Inverse Design of Nanophotonic Devices Using Generative Adversarial Networks with the Sim-NN Model and Self-Attention Mechanism

The inverse design method based on a generative adversarial network (GAN) combined with a simulation neural network (sim-NN) and the self-attention mechanism is proposed in order to improve the efficiency of GAN for designing nanophotonic devices. The sim-NN can guide the model to produce more accurate device designs via the spectrum comparison, whereas the self-attention mechanism can help to extract detailed features of the spectrum by exploring their global interconnections. The nanopatterned power splitter with a 2 μm × 2 μm interference region is designed as an example to obtain the average high transmission (>94%) and low back-reflection (<0.5%) over the broad wavelength range of 1200~1650 nm. As compared to other models, this method can produce larger proportions of high figure-of-merit devices with various desired power-splitting ratios.

Free-form optimization of nanophotonic devices: from classical methods to deep learning

Nanophotonic devices have enabled microscopic control of light with an unprecedented spatial resolution by employing subwavelength optical elements that can strongly interact with incident waves. However, to date, most nanophotonic devices have been designed based on fixed-shape optical elements, and a large portion of their design potential has remained unexplored. It is only recently that free-form design schemes have been spotlighted in nanophotonics, offering routes to make a break from conventional design constraints and utilize the full design potential. In this review, we systematically overview the nascent yet rapidly growing field of free-form nanophotonic device design. We attempt to define the term "free-form" in the context of photonic device design, and survey different strategies for free-form optimization of nanophotonic devices spanning from classical methods, adjoint-based methods, to contemporary machine-learning-based approaches.

Deep learning assisted design of high reflectivity metamirrors

The advent of optical metasurfaces, i.e. carefully designed two-dimensional nanostructures, allows unique control of electromagnetic waves. To unlock the full potential of optical metasurfaces to match even complex optical functionalities, machine learning provides elegant solutions. However, these methods struggle to meet the tight requirements when it comes to metasurface devices for the optical performance, as it is the case, for instance, in applications for high-precision optical metrology. Here, we utilize a tandem neural network framework to render a focusing metamirror with high mean and maximum reflectivity of R_mean = 99.993 % and R_max = 99.9998 %, respectively, and a minimal phase mismatch of Δϕ = 0.016 % that is comparable to state-of-art dielectric mirrors.