Neural Network for Inverse Design in Nanophotonics

This project demonstates training a neural network for inverse design of nanophotonic gratings. The project is inspired by the work of Liu et al. in ACS Photonics journal.

Generating Data

• To be able to run DNN training scripts, you should have dataset dataset.npz file in the project root folder.
• If you want to use pre-generated dataset, you can download the dataset file from here. Note: The file size is ~1 GB.
• In case you wish to generate the dataset from scratch, you can run either produce_data.py python script or produce_data.ipynb Jupyter notebook.

Forward Model

• See forward_model.ipynb Jupyter notebook for loading, training, and saving forward model
• Saved forward model states can be found in forward_model folder.

Inverse Model

• See inverse_model.ipynb Jupyter notebook for loading, training, and saving inverse model
• Saved inverse model states can be found in inverse_model folder.