Design Space Reparameterization Enforces Hard Geometric Constraints in Inverse-Designed Nanophotonic Devices

Abstract

Inverse design algorithms are the basis for realizing high-performance, freeform nanophotonic devices. Current methods to enforce geometric constraints, such as practical fabrication constraints, are heuristic and not robust. In this work, we show that hard geometric constraints can be imposed on inverse-designed devices by reparameterizing the design space itself. Instead of evaluating and modifying devices in the physical device space, candidate device layouts are defined in a constraint-free latent space and mathematically transformed to the physical device space, which robustly imposes geometric constraints. Modifications to the physical devices, specified by inverse design algorithms, are made to their latent space representations using backpropagation. As a proof-of-concept demonstration, we apply reparameterization to enforce strict minimum feature size constraints in local and global topology optimizers for metagratings. We anticipate that concepts in reparameterization will provide a general and meaningful platform to incorporate physics and physical constraints in any gradient-based optimizer, including machine learning-enabled global optimizers.

Benchmarking deep learning-based models on nanophotonic inverse design problems

“…Therefore, the observation that all three models perform similarly in terms of robustness is not surprising. Although we only consider two specific inverse design problems, these neural networks models and introduced evaluation metrics are applicable for many other nanophotonic inverse design problems with different structures and materials, including the multilayer thin films 35 , plasmonic nanostructures 56 , and metasurfaces 37 , etc, where their structures can be described either by a vector or an image when processed by appropriate neural networks. Therefore, our conclusions are generalizable to a wide range of nanophotonic inverse design problems.…”

A vertically-loaded diamond microdisk resonator spin-photon interface

“…We note that the 𝐿 = 1 geometry already achieves good overlap with a Gaussian mode, and expect that this could be improved with direct optimization of the grating structure for this particular figure of merit. Other optimization procedures could be used, such as inverse design [28,29], to numerical evaluation based on the analytical proposal described here. The VLDMoRt can also be designed to enhance 𝜂 of diamond group-IV emitters and even other solid-state defects such as rare-earth ions and quantum dots.…”

Intelligent designs in nanophotonics: from optimization towards inverse creation

“…Recently growing attention has focused on the hybrid tactics to customize metagratings [97], e.g., to combine NN with conventional optimizers. Besides the above-noted contributions [87], another seminal work of Liu et al [88] proposed a fusion scheme connecting compositional pattern-producing network and a cooperative coevolution to tailor molecule-resembled meta-surfaces (see the algorithm diagram in Fig.…”