Intelligent Metasurface Design for Quantum Optics

Metasurfaces, the two-dimensional counterparts of metamaterials, have recently gained unprecedented attention for their ability to precisely manipulate electromagnetic waves. Traditional metasurface design relies on forward design approaches based on rigorus coupled wave analysis,  finite-element or finite-difference time-domain simulations with manual parameter optimization, which is computationally expensive and often deviates from ideal meta-atom responses in practice. Moreover, coupling between neighboring meta-atoms introduces undesired functionalities that hinder device performance. Recently, photonic inverse design has emerged as a viable methodology for complex metasurface design to tackle these challenges. Inverse design can produce both structured and freeform metasurfaces, where interactions between the elements are leveraged to create rich design spaces in which each pixel contributes to mode excitations. In this work, we explore and compare scientific machine learning, physics-informed neural networks, and topology optimization as intelligent design frameworks in terms of their physical accuracy, computational time, and degrees of freedom, enabling a foundation for next-generation metasurfaces in quantum optics applications.