Datasets and Benchmarks for Nanophotonic Structure and Parametric Design Simulations

Jungtaek Kim, Mingxuan Li, Oliver Hinder, Paul W. Leu
University of Pittsburgh

Conference on Neural Information Processing Systems 2023
Datasets and Benchmarks Track
arXiv Paper GitHub Hugging Face

Abstract

Nanophotonic structures have versatile applications including solar cells, anti-reflective coatings, electromagnetic interference shielding, optical filters, and light emitting diodes. To design and understand these nanophotonic structures, electrodynamic simulations are essential. These simulations enable us to model electromagnetic fields over time and calculate optical properties. In this work, we introduce frameworks and benchmarks to evaluate nanophotonic structures in the context of parametric structure design problems. The benchmarks are instrumental in assessing the performance of optimization algorithms and identifying an optimal structure based on target optical properties. Moreover, we explore the impact of varying grid sizes in electrodynamic simulations, shedding light on how evaluation fidelity can be strategically leveraged in enhancing structure designs.

Contributions

  • Development of a generic simulation scheme and pipeline for nanophotonic structures in Python, based on the open-source software, Meep, and licensed under the MIT license
  • Creation of datasets of a myriad of nanophotonic structures for electromagnetic interference shielding, anti-reflection, and solar cells
  • Investigation into the effects of altering grid sizes in electrodynamic simulations, providing insights into tradeoffs between computational time and simulation accuracy
  • Introduction of benchmarks specifically designed for the optimization of parametric structures, facilitating the evaluation and comparison of different optimization algorithms

Real-World Applications

  • Electromagnetic Inference Shielding: As the usage of electronic devices has grown, there is a growing demand for strategies to shield these devices from external electromagnetic waves and interference.
  • Anti-Reflective Coatings: Light traveling from air to glass partially reflects due to the disparity in index of refraction. Better anti-reflective structures can achieve broad-spectrum and wide-angle anti-reflection.
  • Solar Cells: Nanomaterials are revolutionizing solar cell technology, promising significantly enhanced efficiency and reduced costs.

Nanophotonic Structures

Visualization

Optimization Modes

Our benchmarks support two modes for fast prototyping of optimization methods and one mode for direct running of simulations as follows.

  • Discretized Search Space Mode: A search space is discretized with the increment specified. At each configuration a simulation is run and the outcome of the simulation is recorded.
  • Surrogate Model Mode: Based on the dataset collected from the discretized search space, we fit a surrogate model. This allows us to evaluate any structural configuration in the search space using the trained surrogate model.
  • Simulation Mode: Users can directly run these simulations using our frameworks and perform an optimization algorithm by conducting a simulation at every iteration.

Experiments with the Surrogate Model Mode

Citation


      @inproceedings{KimJ2023neurips,
          title={Datasets and Benchmarks for Nanophotonic Structure and Parametric Design Simulations},
          author={Kim, Jungtaek and Li, Mingxuan and Hinder, Oliver and Leu, Paul W.},
          booktitle={Advances in Neural Information Processing Systems (NeurIPS)},
          volume={36},
          pages={4685--4715},
          year={2023},
          note={Datasets and Benchmarks Track}
      }