Feature-Assisted Neuro-CMT Approach to Fast Design Optimization of Metasurfaces

The neuro-coupled mode theory (i.e., neuro-CMT) approach has been recently reported for the intelligent design of metasurfaces. This letter presents an advance, that is, the feature-assisted neuro-CMT approach, to address the issue of bad starting points and to increase the optimization efficiency further. We define the resonant frequencies in the original neuro-CMT surrogate as feature parameters and identify them as additional outputs. Then, we formulate a feature-based objective function to guide the optimization to automatically identify and move the resonant frequencies into the desired frequency band at the initial stage, while ensuring that the electromagnetic (EM) response meets the design specification in the subsequent optimization process. The proposed approach is applied to the design of two metasurface microwave absorbers, showing increased convergence speed and solution optimality compared with the existing neuro-CMT approach. Numerical simulations and experimental measurements further verify the accuracy of the proposed approach.