Abstract

Metasurfaces enable precise control of optical phase, amplitude, and polarization within an ultrathin form factor. Yet many large-area devices are still designed by assembling unit cells characterized under the local periodic approximation, which sometimes neglects important electromagnetic interactions between dissimilar neighboring elements. Full-metasurface inverse design has recently emerged as a powerful alternative because it treats the entire aperture as a coupled electromagnetic system and directly exploits inter-element interactions.

In this review, we organize large-area metasurface design methods according to the spatial extent of the jointly optimized region, progressing from reusable unit cells to supercells and full-metasurface inverse design. We also examine neural-network-based approaches, including surrogate electromagnetic solvers, inverse design networks, and data-efficient and fabrication-aware methods. In addition, we assess whether reusable supercell designs can provide computational savings comparable to those of meta-atom libraries.

Finally, our hardware-aware scaling analysis estimates the memory and runtime required for full-wave full-metasurface optimization across a range of aperture sizes and computing systems. These results clarify the trade-off between electromagnetic accuracy and computational cost and show how advances in computing hardware may expand the practical limits of full-aperture metasurface design.