Inverse design of large-scale freeform meta-optics by breaking the memory wall of full-wave simulation
Abstract
Gradient-based design of large optical surfaces hits a memory wall: a stored-history time-domain adjoint needs 24Nt bytes per cell for Nt steps, and every field update is bound by memory bandwidth. Here we present a validated framework that removes this wall and makes globally optimized freeform meta-optics practical. Its finite-difference time-domain adjoint reconstructs interior fields backwards from two recorded planes, so volume memory no longer grows with Nt, and fused kernels raise memory-bandwidth use on an H200 GPU from 6.4% to a remarkable 56% of peak. Overlapping tiles, coupled by angular-spectrum propagation and distributed over GPUs, keep tile memory independent of the aperture. Within hours on one GPU node, we optimize, against one global objective, every density variable of single-layer metasurfaces 100 to 200 μm across, and design a nine-wavelength lens, a color hologram and a polarization-switched hologram. Confirmed by an independent solver within 1.6%, they substantially outperform meta-atom designs, reaching 1.64, 1.65 and 1.24 times their objectives.
Related
- Label-Efficient Generative Inverse Design and Design-Rule Discovery in Freeform Topological Photonics
- Polymer Membrane Tensegrity: Inverse Design of Polymer Films Morphing into Freeform 3D Surfaces with Digital Photopatterning Technique
- Efficient Modeling of 3D-Printed Freeform Waveguides with a Transformation Optics-Based Beam Propagation Method
- Quantum simulation of wave optics in weakly inhomogeneous media using block encoding
- Wave-optics numerical simulation of atmospheric turbulence in 10,000-km-scale satellite–ground optical links with spatially adaptive phase screens
BibTeX
@misc{park2026inverse,
title = {{Inverse design of large-scale freeform meta-optics by breaking the memory wall of full-wave simulation}},
author = {Hyoseok Park and Yeonsang Park},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2609.30039},
eprint = {2609.30039},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.30039},
}
RIS
TY - UNPB
TI - Inverse design of large-scale freeform meta-optics by breaking the memory wall of full-wave simulation
AU - Hyoseok Park
AU - Yeonsang Park
PY - 2026
DA - 2026/10/06
PB - arXiv
DO - 10.48550/arxiv.2609.30039
UR - https://arxiv.org/abs/2609.30039
AB - Gradient-based design of large optical surfaces hits a memory wall: a stored-history time-domain adjoint needs 24N_t bytes per cell for N_t steps, and every field update is bound by memory bandwidth. Here we present a validated framework that removes this wall and makes globally optimized freeform meta-optics practical. Its finite-difference time-domain adjoint reconstructs interior fields backwards from two recorded planes, so volume memory no longer grows with N_t, and fused kernels raise memory-bandwidth use on an H200 GPU from 6.4% to a remarkable 56% of peak. Overlapping tiles, coupled by angular-spectrum propagation and distributed over GPUs, keep tile memory independent of the aperture. Within hours on one GPU node, we optimize, against one global objective, every density variable of single-layer metasurfaces 100 to 200 μm across, and design a nine-wavelength lens, a color hologram and a polarization-switched hologram. Confirmed by an independent solver within 1.6%, they substantially outperform meta-atom designs, reaching 1.64, 1.65 and 1.24 times their objectives.
ER -