Inverse design of topological Fano resonance via deep learning based multi-objective optimization

Due to the complicated physical mechanisms of topological systems and the pronounced sensitivity and intrinsic asymmetry of Fano resonance, achieving structural designs for specific high-quality topological Fano resonances is particularly challenging. In this Letter, we propose a framework that integrates deep learning with multi-objective particle swarm optimization (MOPSO) to achieve the on-demand structural design of high-Q Fano resonance in a topological system. In our framework, the residual and fully connected neural networks serve as surrogate models to predict the Fano spectrum and the quality factor, achieving accuracies of 97.04% and 98.96%, respectively. The effectiveness of the framework is demonstrated by designing topological Fano resonance at frequencies of 98, 100, and 102 THz. Our results may offer a versatile route for multi-performance optimization and inverse design of topological photonic crystals.