A Surrogate-Assisted Extended Generative Adversarial Network for Parameter Optimization in Free-Form Metasurface Design

TL;DR

This work addresses the challenge of rapidly designing high-DOF free-form metasurfaces for 5G applications by introducing XGAN, an extended GAN guided by a surrogate model. The surrogate, F-ResNet, maps 2D metasurface patterns to their spectral responses, enabling physics-aware feedback during inverse design, while the generator outputs ternary free-form patterns conditioned on target spectra. The approach achieves an average spectral-prediction accuracy of 0.9734 on 20,000 designs and a 500× speedup over conventional optimization, with F-ResNet delivering ACC_ave ≈ 0.99 and XGAN delivering MAE well below 0.1 in inverse design. This surrogate-assisted framework accelerates metasurface library construction and can extend to other inverse-design domains, such as optical metamaterials and nanophotonic devices.

Abstract

Metasurfaces have widespread applications in fifth-generation (5G) microwave communication. Among the metasurface family, free-form metasurfaces excel in achieving intricate spectral responses compared to regular-shape counterparts. However, conventional numerical methods for free-form metasurfaces are time-consuming and demand specialized expertise. Alternatively, recent studies demonstrate that deep learning has great potential to accelerate and refine metasurface designs. Here, we present XGAN, an extended generative adversarial network (GAN) with a surrogate for high-quality free-form metasurface designs. The proposed surrogate provides a physical constraint to XGAN so that XGAN can accurately generate metasurfaces monolithically from input spectral responses. In comparative experiments involving 20000 free-form metasurface designs, XGAN achieves 0.9734 average accuracy and is 500 times faster than the conventional methodology. This method facilitates the metasurface library building for specific spectral responses and can be extended to various inverse design problems, including optical metamaterials, nanophotonic devices, and drug discovery.

Figures (10)

• Figure 1: The network of the forward/surrogate model F-ResNet.
• Figure 2: The modeling of XGAN. (a) The training processes of the generator and discriminator of XGAN. (b) The loss functions of XGAN consist of the discriminator loss $L_{D}$ and the generator loss $L_{G}$.
• Figure 3: Proposed XGAN architecture consists of (a) a generator and (b) a discriminator.
• Figure 4: Data collection process. (a) Discretization and coding of the metasurface pattern. (b) The process of collecting data based on Python-PEEC co-simulation. The collected data are utilized for training neural networks that replace the manual work in forward and inverse processes.
• Figure 5: A sketch of metasurface design via XGAN.