MERL - Mitsubishi Electric Research Laboratories

Share:

TR2020-130

Inverse Design of Nanophotonic Devices using Deep Neural Networks

    •  Kojima, K., Tang, Y., Koike-Akino, T., Wang, Y., Jha, D., Parsons, K., TaherSima, M., Sang, F., Klamkin, J., Qi, M., "Inverse Design of Nanophotonic Devices using Deep Neural Networks", Asia Communications and Photonics Conference (ACP), September 2020.
      BibTeX TR2020-130 PDF
      • @inproceedings{Kojima2020sep,
      • author = {Kojima, Keisuke and Tang, Yingheng and Koike-Akino, Toshiaki and Wang, Ye and Jha, Devesh and Parsons, Kieran and TaherSima, Mohammad and Sang, Fengqiao and Klamkin, Jonathan and Qi, Minghao},
      • title = {Inverse Design of Nanophotonic Devices using Deep Neural Networks},
      • booktitle = {Asia Communications and Photonics Conference (ACP)},
      • year = 2020,
      • month = sep,
      • url = {https://www.merl.com/publications/TR2020-130}
      • }
  • MERL Contacts:
  • Research Areas:

    Artificial Intelligence, Electronic and Photonic Devices, Machine Learning, Optimization, Signal Processing

We present two different approaches to apply deep learning to inverse design for nanophotonic devices. First, we use a regression model, with device parameters as inputs and device responses as outputs, or vice versa. Second, we use a novel generative model to create a series of improved designs. We demonstrate them to design nanophotonic power splitters with multiple splitting ratios.