Neural network enabled metasurface design for phase manipulation

AI-Based Metamaterial Design

The use of metamaterials in various devices has revolutionized applications in optics, healthcare, acoustics, and power systems. Advancements in these fields demand novel or superior metamaterials that can demonstrate targeted control of electromagnetic, mechanical, and thermal properties of matter. Traditional design systems and methods often require manual manipulations which is time-consuming and resource intensive. The integration of artificial intelligence (AI) in optimizing metamaterial design can be employed to explore variant disciplines and address bottlenecks in design. AI-based metamaterial design can also enable the development of novel metamaterials by optimizing design parameters that cannot be achieved using traditional methods. The application of AI can be leveraged to accelerate the analysis of vast data sets as well as to better utilize limited data sets via generative models. This review covers the transformative impact of AI and AI-based metamaterial design for optics, acoustics, healthcare, and power systems. The current challenges, emerging fields, future directions, and bottlenecks within each domain are discussed.

Recent advances in metasurface design and quantum optics applications with machine learning, physics-informed neural networks, and topology optimization methods

As a two-dimensional planar material with low depth profile, a metasurface can generate non-classical phase distributions for the transmitted and reflected electromagnetic waves at its interface. Thus, it offers more flexibility to control the wave front. A traditional metasurface design process mainly adopts the forward prediction algorithm, such as Finite Difference Time Domain, combined with manual parameter optimization. However, such methods are time-consuming, and it is difficult to keep the practical meta-atom spectrum being consistent with the ideal one. In addition, since the periodic boundary condition is used in the meta-atom design process, while the aperiodic condition is used in the array simulation, the coupling between neighboring meta-atoms leads to inevitable inaccuracy. In this review, representative intelligent methods for metasurface design are introduced and discussed, including machine learning, physics-information neural network, and topology optimization method. We elaborate on the principle of each approach, analyze their advantages and limitations, and discuss their potential applications. We also summarize recent advances in enabled metasurfaces for quantum optics applications. In short, this paper highlights a promising direction for intelligent metasurface designs and applications for future quantum optics research and serves as an up-to-date reference for researchers in the metasurface and metamaterial fields.

A Fast Design Method of Anisotropic Dielectric Lens for Vortex Electromagnetic Wave Based on Deep Learning

Orbital angular momentum (OAM) has made it possible to regulate classical waves in novel ways, which is more energy- or information-efficient than conventional plane wave technology. This work aims to realize the transition of antenna radiation mode through the rapid design of an anisotropic dielectric lens. The deep learning neural network (DNN) is used to train the electromagnetic properties of dielectric cell structures. Nine variable parameters for changing the dielectric unit structure are present in the input layer of the DNN network. The trained network can predict the transmission phase of the unit cell structure with greater than 98% accuracy within a specific range. Then, to build the corresponding relationship between the phase and the parameters, the gray wolf optimization algorithm is applied. In less than 0.3 s, the trained network can predict the transmission coefficients of the 31 × 31 unit structure in the arrays with great accuracy. Finally, we provide two examples of neural network-based rapid anisotropic dielectric lens design. Dielectric lenses produce the OAM modes +1, -1, and -1, +2 under TE and TM wave irradiation, respectively. This approach resolves the difficult phase matching and time-consuming design issues associated with producing a dielectric lens.

Intelligent design of the chiral metasurfaces for flexible targets: combining a deep neural network with a policy proximal optimization algorithm

Recently, deep reinforcement learning (DRL) for metasurface design has received increased attention for its excellent decision-making ability in complex problems. However, time-consuming numerical simulation has hindered the adoption of DRL-based design method. Here we apply the Deep learning-based virtual Environment Proximal Policy Optimization (DE-PPO) method to design the 3D chiral plasmonic metasurfaces for flexible targets and model the metasurface design process as a Markov decision process to help the training. A well trained DRL agent designs chiral metasurfaces that exhibit the optimal absolute circular dichroism value (typically, ∼ 0.4) at various target wavelengths such as 930 nm, 1000 nm, 1035 nm, and 1100 nm with great time efficiency. Besides, the training process of the PPO agent is exceptionally fast with the help of the deep neural network (DNN) auxiliary virtual environment. Also, this method changes all variable parameters of nanostructures simultaneously, reducing the size of the action vector and thus the output size of the DNN. Our proposed approach could find applications in efficient and intelligent design of nanophotonic devices.

Design of All-Dielectric Metasurface-Based Subtractive Color Filter by Artificial Neural Network

Structural colors produced by light manipulating at subwavelength dimensions have been widely studied. In this work, a metasurface-based subtractive color filter (SCF) is demonstrated. The color display of the SCF is confirmed by finding the complementary color of colors filtered by SCF within the color wheel. In addition, two artificial neural network (ANN) models are utilized to accelerate the metasurface forward prediction, and the long short-term memory (LSTM) shows much better performance than traditional multilayer perceptron (MLP). Meanwhile, we train an inverse ANN model established with LSTM to recover the optimal geometric parameter combinations of the meta-atoms. With the variation of the geometric parameters of meta-atoms, versatile color displays of structural colors are realized. The metasurface we propose exhibits good performance of transmissive-type structural color in visible range. The work provides a method for high-efficiency geometric parameter prediction, and paves the way to nanostructure-based color design for display and anticounterfeiting applications.

Multi-Parameter Inversion of AIEM by Using Bi-Directional Deep Neural Network

A novel multi-parameter inversion method is proposed for the Advanced Integral Equation Model (AIEM) by using bi-directional deep neural network. There is a very complex nonlinear relationship between the surface parameters (dielectric constant and roughness) and radar backscattering coefficient. The traditional inverse neural network, which is constructed by using the backscattering coefficients as the input and the surface parameters as the output, leads to bad convergence and wrong results. This is because many sets of surface parameters can get the same backscattering coefficient. Therefore, the proposed bi-directional deep neural network starts with building an AIEM-based forward deep neural network (AIEM-FDNN), whose inputs are the surface parameters and outputs are the backscattering coefficients. In this way, the weights and biases of the forward deep neural network can be optimized and predicted, which can be used for the backward deep neural network (AIEM-BDNN). Then, the multi-parameters are updated by minimizing the loss between the output backscattering coefficients with the measured ones. By inserting a sigmoid function between the input and the first hidden layer, the input multi-parameters can be efficiently approximated and continuously updated. As a result, both the forward and backward deep neural networks can be built with these weights and biases. By sharing the weights and biases of the forward network, the training of the inverse network is avoided. The bi-directional deep neural network can not only predict the backscattering coefficient but can also inverse the surface parameters. Numerical results are given to demonstrate that the RMSE of the backscattering coefficients calculated by the proposed bi-directional neural network can be reduced to 0.1%. The accuracy of the inversion parameters, including the real and imaginary parts of the dielectric constant, the root mean square height and the correlation length, can be improved to 97.56%, 91.14%, 99.04% and 98.45%, respectively. At the same time, the bi-directional neural network also has good accuracy for the inversion of the POLARSCAT measured data.

Inverse design of coupled subwavelength dielectric resonators with targeted eigenfrequency and Q factor utilizing deep learning

Subwavelength all-dielectric resonators supporting Mie resonances are promising building blocks in nanophotonics. The coupling of dielectric resonators facilitates advanced shaping of Mie resonances. However, coupled dielectric resonators with anisotropic geometry can only be designed by time-consuming simulation utilizing parameter scanning, hampering their applications in nanophotonics. Herein, we propose and demonstrate that a combination of two fully connected networks can effectively design coupled dielectric resonators with targeted eigenfrequency and Q factor. Typical examples are given for validating the proposed network, where the normalized deviation rates of eigenfrequency and Q factor are 0.39% and 1.29%, respectively. The proposed neutral network might become a useful tool in designing coupled dielectric resonators and beyond.

Dielectric metalens for miniaturized imaging systems: progress and challenges

Lightweight, miniaturized optical imaging systems are vastly anticipated in these fields of aerospace exploration, industrial vision, consumer electronics, and medical imaging. However, conventional optical techniques are intricate to downscale as refractive lenses mostly rely on phase accumulation. Metalens, composed of subwavelength nanostructures that locally control light waves, offers a disruptive path for small-scale imaging systems. Recent advances in the design and nanofabrication of dielectric metalenses have led to some high-performance practical optical systems. This review outlines the exciting developments in the aforementioned area whilst highlighting the challenges of using dielectric metalenses to replace conventional optics in miniature optical systems. After a brief introduction to the fundamental physics of dielectric metalenses, the progress and challenges in terms of the typical performances are introduced. The supplementary discussion on the common challenges hindering further development is also presented, including the limitations of the conventional design methods, difficulties in scaling up, and device integration. Furthermore, the potential approaches to address the existing challenges are also deliberated.

Comparison of Different Neural Network Architectures for Plasmonic Inverse Design

The merge between nanophotonics and a deep neural network has shown unprecedented capability of efficient forward modeling and accurate inverse design if an appropriate network architecture and training method are selected. Commonly, an iterative neural network and a tandem neural network can both be used in the inverse design process, where the latter is well known for tackling the nonuniqueness problem at the expense of more complex architecture. However, we are curious to compare these two networks' performance when they are both applicable. Here, we successfully trained both networks to inverse design the far-field spectrum of plasmonic nanoantenna, and the results provide some guidelines for choosing an appropriate, sufficiently accurate, and efficient neural network architecture.

Broadband achromatic metalens design based on deep neural networks

For the design of achromatic metalenses, one key challenge is to accurately realize the wavelength dependent phase profile. Because of the demand of tremendous simulations, traditional methods are laborious and time consuming. Here, a novel deep neural network (DNN) is proposed and applied to the achromatic metalens design, which turns complex design processes into regression tasks through fitting the target phase curves. During training, x-y projection pairs are put forward to solve the phase jump problem, and some additional phase curves are manually generated to optimize the DNN performance. To demonstrate the validity of our DNN, two achromatic metalenses in the near-infrared region are designed and simulated. Their average focal length shifts are 2.6% and 1.7%, while their average relative focusing efficiencies reach 59.18% and 77.88%.

Time-Effective Simulation Methodology for Broadband Achromatic Metalens Using Deep Neural Networks

Metasurface has demonstrated potential and novel optical properties in previous research. The prevailing method of designing a macroscale metasurface is based on the local periodic approximation. Such a method relies on the pre-calculated data library, including phase delay and transmittance of the nanostructure, which is rigorously calculated by the electromagnetic simulation. However, it is usually time-consuming to design a complex metasurface such as broadband achromatic metalens due the required huge data library. This paper combined different numbers of nanofins and used deep neural networks to train our data library, and the well-trained model predicted approximately ten times more data points, which show a higher transmission for designing a broadband achromatic metalens. The results showed that the focusing efficiency of designed metalens using the augmented library is up to 45%, which is higher than that using the original library over the visible spectrum. We demonstrated that the proposed method is time-effective and accurate enough to design complex electromagnetic problems.