A Hybrid AI-Driven Framework for Designing Multifunctional Metaphotonic Devices

We introduce a novel hybrid computational framework for the inverse design of large-scale metaphotonic devices by integrating the unique features of topological optimization and advanced machine-learning algorithms. In contrast to conventional optimization methods that often constrain certain design parameters (e.g., thickness, periodicity, or material properties), this hybrid approach enables the complete design of the optimal nanostructures by simultaneously optimizing both structural properties (i.e., geometrical features) and physical parameters. This comprehensive inverse-design strategy enables a more thorough exploration of the entire design space, leading to superior device performance.