Training Deep Neural Networks for the Inverse Design of Nanophotonic Structures

By Dianjing Liu, Yixuan Tan, Erfan Khoram, Zongfu Yu. Conference on Lasers and Electro-Optics, 2017

1,025 citations recorded by the source corpus.

We demonstrate a tandem neural network architecture that tolerates inconsistent training instances in inverse design of nanophotonic devices. It provides a way to train large neural networks for the inverse design of complex photonic structures. © 2019 The Author(s)

Subfields: Artificial Intelligence, Atomic and Molecular Physics, and Optics, Electronic, Optical and Magnetic Materials

View Paper: https://www.semanticscholar.org/paper/8daa6189ed0c3e5fa4ffecc7aeed93e35297a737

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. John Peurifoy
2. Jiaqi Jiang
3. Rohit Unni
4. Marin Soljačić
5. Zhicheng Wang
6. Max Tegmark
7. Feng Cheng
8. Yongmin Liu

In Atomic and Molecular Physics, and Optics

1. Marin Soljačić
2. John D. Joannopoulos
3. Steven G. Johnson
4. Anuradha Murthy Agarwal
5. Jiaqi Jiang
6. Rohit Unni
7. Max Tegmark
8. Feng Cheng

In Electronic, Optical and Magnetic Materials

1. Yongmin Liu
2. Wei Ma
3. Yihao Xu
4. John D. Joannopoulos
5. Marin Soljačić
6. Rohit Unni
7. Zhicheng Wang
8. Max Tegmark

Citations in the corpus

• Global optimization of dielectric metasurfaces using a physics-driven neural network (2019, background): “26,29 One challenge is that the computational cost of creating the training dataset itself can be immense.”
• Probabilistic Representation and Inverse Design of Metamaterials Based on a Deep Generative Model with Semi‐Supervised Learning Strategy (2019, methodology): “To remedy this issue, tandem training strategy has been used to avoid instable training loss and force the inverse network to converge to one possible solution [53], but this operation sacrifices the design varieties and leads to limited generalization ability.”
• Nanophotonic particle simulation and inverse design using artificial neural networks (2017, background): “(16) was able to overcome degeneracy problems wherein the same spectrum can be generated by particles of different geometrical arrangements.”
• Intelligent nanophotonics: merging photonics and artificial intelligence at the nanoscale (2018, methodology): “A similar procedure has also been used to study transmission of multilayer thin films, where hundreds of thousands of samples were generated for training [152].”
• Interfacing photonics with artificial intelligence: an innovative design strategy for photonic structures and devices based on artificial neural networks (2021, background): “In early 2018, D. Liu et al. introduced, for the first time, a tandem network architecture for the inverse design problem [46].”
• A Deep Learning Approach for Objective-Driven All-Dielectric Metasurface Design (2019, positive): “To achieve this goal, we connected the fully-trained PNN to a meta-filter generator to form a cascaded network and avoid the non-convergence problem resulting from non-unique solutions (21, 22).”
• Physics-informed neural networks with hard constraints for inverse design (2021, methodology): “…in the supervised-learning paradigm to learn the nonlinear mapping from arbitrary designs to their associated functional properties [25, 42, 29, 59, 47, 54, 39, 40, 15] or vice versa [29, 49, 54], i.e., neural networks are used as surrogate models of PDE solvers to accelerate the…”