2022
DOI: 10.1021/acsphotonics.2c01160
|View full text |Cite
|
Sign up to set email alerts
|

Reparameterization Approach to Gradient-Based Inverse Design of Three-Dimensional Nanophotonic Devices

Abstract: We propose a three-dimensional freeform nanophotonic platform in which wavelength-scale domains comprise basic geometric structures with explicitly defined dimensions, positions, orientations, and minimum feature size constraints. Given a desired wavefront shaping objective, these parameters can be collectively optimized using gradient-based shape optimization with full accounting of near-field interactions between structures. We apply our concept to a variety of metagratings supporting high diffraction effici… Show more

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
1
1
1

Citation Types

0
6
0

Year Published

2023
2023
2024
2024

Publication Types

Select...
6

Relationship

2
4

Authors

Journals

citations
Cited by 7 publications
(6 citation statements)
references
References 43 publications
0
6
0
Order By: Relevance
“…Therefore, the improvement in the merit function would be smaller and potentially hinder the optimization process. To address this, we leverage a latent matrix that extends the domain to all real numbers (−∞, ∞) instead of restricting it to [0, 1], which is inspired by work on reparameterization to latent variables and the concept of latent noise vectors prevalent in generative adversarial networks (45)(46)(47). This latent matrix is mapped to the device density using the pattern generator function.…”
Section: Optimization Results and Working Principlesmentioning
confidence: 99%
“…Therefore, the improvement in the merit function would be smaller and potentially hinder the optimization process. To address this, we leverage a latent matrix that extends the domain to all real numbers (−∞, ∞) instead of restricting it to [0, 1], which is inspired by work on reparameterization to latent variables and the concept of latent noise vectors prevalent in generative adversarial networks (45)(46)(47). This latent matrix is mapped to the device density using the pattern generator function.…”
Section: Optimization Results and Working Principlesmentioning
confidence: 99%
“…To ensure ease of fabrication in a manner that streamlines with gradient-based freeform optimization, we utilize a reparameterization scheme in which the dimensions and positions of nanoscale elements within a super-pixel are represented as latent variables that can be transformed to physical layouts using analytic expressions 31 . In this manner, minimum feature size and gap constraints are hard coded into the element representations.…”
Section: Design Principlementioning
confidence: 99%
“…Finally, a finite number of building blocks with desired phase shifts are collected from the pre-built meta-library and arranged with periodicity to realize targeted wavefront shaping. The forward design approach is widely used due to its interpretable framework and ease of implementation [ 19 , 20 ]. However, as the design constraints and design degrees of freedom scale up, it becomes less useful and may result in limit device performance.…”
Section: Introductionmentioning
confidence: 99%