Machine Learning Approach to the Topological Optimization of Metasurfaces

Paper i proceeding, 2022

We present our work on using machine learning for the topological optimization of metasurfaces. First, we show that deep neural networks can be used to predict the scattering properties of metasurfaces. Subsequently, we demonstrate the inverse design of free-form metasurfaces using a modified CGAN machine learning method that balances the accuracy of desired optical properties with experimental feasibility. Our method allows constraints imposed by the nanofabrication to be integrated in the optimization.