Hybrid Quantum-Classical Inverse Design of Metasurfaces for Tailored Narrow Band Absorption

Abstract

The inverse design of metasurfaces poses a considerable challenge because of the intricate interdependencies that exist between structural characteristics and electromagnetic responses. Traditional optimization methods require significant computational resources and frequently do not produce the most effective solutions. This study presents a hybrid quantum-classical machine learning approach known as Latent Style-based Quantum GAN (LaSt-QGAN). This method integrates a Variational Autoencoder (VAE) with a Quantum Generative Adversarial Network (QGAN) to enhance the optimization of metasurface designs aimed at achieving narrow-band absorption and unidirectionality. The proposed method results in a reduction of training time by 10X and a decrease in data requirements by 40X when compared to traditional GAN-based approaches. The produced metasurface designs demonstrate a high fidelity in relation to the target absorption spectra compared to the classical GAN based methods. Additionally, the integration of a material look-up table facilitates manufacturability by allowing for the substitution of predicted material properties with viable alternatives, all while preserving performance accuracy. Moreover the model is able to generate Q-factor upto the order of $10^4$, while the training dataset has Q-factor upto the order of $10^3$.