We will discuss the design and manufacturing of freeform metasurfaces that exhibit high efficiency, multi-functional capabilities. First, we will discuss local and global gradient-based optimization algorithms that can produce non-intuitive, curvilinear designs utilizing multi-scattering light-matter interactions to achieve high performance. Second, we will discuss how physics-augmented deep networks can be trained with a combination of data and physical constraints to serve as accurate surrogate electromagnetic solvers that can produce solutions three to four orders of magnitude faster than with conventional methods. Third, we will introduce a concept termed reparameterization that can enforce hard design constraints, such as fabrication-dictated minimum feature sizes, into the design process. Finally, we discuss future challenges and opportunities to freeform photonics implementation.
|