Improved convergence in planar nanophotonic topology optimization via the multigradient

Adjoint Topology Optimization is a physics-informed inverse-design strategy which has been widely applied in electromagnetics and nanophotonics to create highly performant meta-devices. By interfering the electric fields of both forward and reverse (i.e., adjoint) simulations within a device, it is possible to compute a permittivity gradient which will improve its performance. In the specific case of planar metasurface design, the permittivity gradient will vary depending on the z-height, which presents a problem for designing devices suitable for single-layer lithographic fabrication, since such products have uniform material properties in the z-direction. State-of-the-art applications of adjoint topology optimization address this issue by averaging together the gradients at multiple z-heights. This article proposes an alternative strategy for combining these same gradients based on the multigradient. This modification produces a notable speedup of as much as 2–3 times for a representative supercell design problem, indicating this new strategy may be preferable for planar nanophotonic design problems in general.