Abstract:In traditional metasurface structure design, it heavily relies on electromagnetic simulations to obtain transmission and phase spectral, followed by empirical adjustments. This iterative trial-and-error process, especially when dealing with multi-objective optimization tasks, demands intensive and time-consuming computations, which to a certain extent restricts the development of the metasurface research field. In this paper, a proposed method achieves rapid prediction of spectral responses corresponding to structural units by seeking analytical solutions within the constructed neural network model. The proposed deep learning-based method for predicting transmission and phase spectral of metasurface units consists of metasurface unit dataset construction and a ResNet-based network framework. In the dataset construction approach, an overhead view of the unit structure is extracted and transformed into a binary image, where scaling factors are coupled into the two-dimensional image to increase dimensionality. This enables the representation of different structures such as square pillars, elliptical cylinders, and varying sizes of metasurface units using the same data format, significantly enhancing network generalization. Within the network framework, ResNet is employed to predict the real and imaginary parts of the S-21 parameter, which are then inverted to obtain transmission and phase information. The progressive training method employed in combination with this framework yields high prediction accuracy. The deep learning-based method for predicting transmission and phase spectral of dielectric metasurface units, as revealed in this paper, achieves a 7200-fold increase in prediction speed compared to traditional electromagnetic simulation methods.

A Freeform Dielectric Metasurface Modeling Approach Based on Deep Neural Networks

A Deep Learning Approach for Objective-Driven All-Dielectric Metasurface Design

A Novel Modeling Approach for All-Dielectric Metasurfaces Using Deep Neural Networks

On the Application of Deep Learning in Modeling Metasurface

Neural Network-Based Surrogate Model for Inverse Design of Metasurfaces

Deep-Learning-Empowered Inverse Design for Freeform Reconfigurable Metasurfaces

Symmetry Enhanced Network Architecture Search for Complex Metasurface Design

Dual-band optical collimator based on deep-learning designed, fabrication-friendly metasurfaces

A Deep-Learning Approach for Rapid Prediction of Spectral Responses of Meta-Atoms

Deep Convolutional Neural Networks to Predict Mutual Coupling Effects in Metasurfaces

Freeform Diffractive Metagrating Design Based on Generative Adversarial Networks

A Physics Based Approach for Neural Networks Enabled Design of All-Dielectric Metasurfaces

Pushing the Limits of Multifunctional Metasurface by Deep Learning

Deep-learning-assisted intelligent design of terahertz hybrid-functional metasurfaces with freeform patterns

Global reconstruction of form-free metasurfaces for customized demand via a learning paradigm

Deep neural network enabled active metasurface embedded design

Deep Learning: A Rapid and Efficient Route to Automatic Metasurface Design

On-demand Quick Metasurface Design with Neighborhood Attention Transformer

Fast design of plasmonic metasurfaces enabled by deep learning

Physics-Driven Machine-Learning Approach Incorporating Temporal Coupled Mode Theory for Intelligent Design of Metasurfaces

Multifunctional Metasurface Design with a Generative Adversarial Network