COMSOL-in-the-Loop Automated Metasurface Inverse Design with Bayesian Optimization
Metasurface design often relies on labor-intensive geometric tuning and repeated full-wave simulations, making the exploration of sample spaces labor-intensive. In this article, we present an end-to-end automated framework for the inverse design of metasurface absorbers and circular-dichroism (CD) devices using COMSOL Multiphysics. The framework accepts user-defined performance objectives, such as minimum band-averaged absorptance or a specified circular-dichroism threshold. Candidate patterns are constructed from parameterized geometric primitives through translation, rotation, scaling, and Boolean operations. For each candidate, the framework automatically generates and validates the geometry, performs a full-wave finite-element simulation, extracts the optical response, and evaluates the design objective. Bayesian optimization uses these results to select subsequent candidates until the performance requirement or computational budget is reached, forming a closed design loop that requires no manual intervention. Applied to a gold-chromium-silicon dioxide-silicon metasurface, the framework independently identifies distinct geometries that realize broadband absorption and strong circular dichroism, demonstrating that the same material platform can support different optical functionalities. This training-free approach provides a compact, interpretable, and physics-grounded route to automated metasurface design.
