They have important potential for future optical and photonic devices, as they allow versatile control over light waves: for example, to efficiently absorb light in sensors or solar cells, to steer infrared light beams for autonomous driving applications and light-based communication methods, and to refract waves in ultra-compact optical imaging devices. It is however notoriously difficult to design metasurfaces for such specific functions, as the relation between structure and function is nontrivial and the design space enormous.
This project aims to develop new computational tools for the automated design of metasurfaces, based on deep learning techniques. Specifically, we target to infuse such artificial intelligence with rigorous knowledge of physical laws and models, to enhance the speed and accuracy of the methods and to expand the range of structures and functions that they can be applied to. We test the effectiveness of the developed methods on specific target applications with high technological potential, such as light detection, imaging, and beam steering.