Advancing Machine Learning Optimization of Chiral Photonic Metasurface: Comparative Study of Neural Network and Genetic Algorithm Approaches

TL;DR

The paper tackles the challenge of designing chiral photonic metasurfaces with strong circular dichroism (CD) while maintaining high reflectivity. It compares a two-output neural-network (NN) optimization pipeline with a genetic-algorithm (GA) approach, augmented by enantiomer-based data augmentation, adaptive training, and a unified fitness function $F=|\Delta R_{\rm CD}|\cdot R_{\rm pref}$, across GaP/air and PMMA/air platforms using RCWA simulations. The study finds that the 2-output NN reduces the trade-off between CD and reflectivity and, together with data augmentation, achieves efficient convergence, while the GA offers robustness for more complex geometries at higher computational cost; a hybrid NN–GA strategy is proposed to leverage both strengths. The results demonstrate substantial improvements in CD magnitude and spectral reflectivity, with practical implications for fabricable chiral mirrors and polarization-selective photonic devices, and establish a scalable design workflow applicable to other inverse-design challenges in photonics.

Abstract

Chiral photonic metasurfaces provide unique capabilities for tailoring light-matter interactions, which are essential for next-generation photonic devices. Here, we report an advanced optimization framework that combines deep learning and evolutionary algorithms to significantly improve both the design and performance of chiral photonic nanostructures. Building on previous work utilizing a three-layer perceptron reinforced learning and stochastic evolutionary algorithm with decaying changes and mass extinction for chiral photonic optimization, our study introduces a refined pipeline featuring a two-output neural network architecture to reduce the trade-off between high chiral dichroism (CD) and reflectivity. Additionally, we use an improved fitness function, and efficient data augmentation techniques. A comparative analysis between a neural network (NN)-based approach and a genetic algorithm (GA) is presented for structures of different interface pattern depth, material combinations, and geometric complexity. We demonstrate a twice higher CD and the impact of both the corner number and the refractive index contrast at the example of a GaP/air and PMMA/air metasurface as a result of superior optimization performance. Additionally, a substantial increase in the number of structures explored within limited computational resources is highlighted, with tailored spectral reflectivity suggested by our electromagnetic simulations, paving the way for chiral mirrors applicable to polarization-selective light-matter interaction studies.

Figures (7)

• Figure 1: Sketched chiral metasurface optimization task: (a) the polarization selective reflectance, (b) The geometry (top view unit cell, lattice and cross section), and (c) two distinct algorithms for the inference of a high CD and suitable spectral reflectivity performance of the patterned air--substrate interface (top: neural-network algorithm; bottom: evolutionary algorithm). Drawings after Ref. mey2022machine.
• Figure 2: Workflow of the Genetic Algorithm
• Figure 3: Improvement of the few-layer perceptron reinforcement-learning pipeline: from a single output neuron (a) to two output neurons and an enhanced fitness evaluation function (b). Insets: The structures shown in green represent the best candidate identified after the first iteration (left side), during which 64 randomly generated configurations are evaluated. The structures shown in blue (right side) correspond to the optimized designs obtained at the end of the full optimization process.
• Figure 4: Results of the thickness optimization for the two tested compositions, GaP/Air in blue and PMMA/Air in green. Each dot represents the best individual structure found at a specific thickness. The GaP/Air composition achieved the highest fitness value at thickness $\lambda_0/3$ while the PMMA/Air composition achieved the highest fitness at thickness $\lambda_0$.
• Figure 5: Structures complexity increase: comparison of polarization-sensitive structure performance obtained from neural network and genetic algorithm optimizations.