Photontorch is a photonic simulator for highly parallel simulation and optimization of photonic circuits in time and frequency domain. Photontorch features CUDA enabled simulation and optimization of photonic circuits. It leverages the deep learning framework PyTorch to view the photonic circuit as essentially a recurrent neural network. This enables the use of native PyTorch optimizers to optimize the (physical) parameters of the circuit.
- Floris Laporte [floris.laporte@ugent.be]
- Website: photontorch.com
Photontorch can be installed with pip:
pip install photontorch
During development or to use the most recent Photontorch version, clone the repository and link with pip:
git clone https://git.photontorch.com/photontorch.git
./install-git-hooks.sh # Unix [Linux/Mac/BSD/...]
install-git-hooks.bat # Windows
pip install -e photontorch
During development, use pytest to run the tests from within the root of the git-repository:
pytest tests
Read the full documentation here: https://docs.photontorch.com
- Python 2.7 (Linux only) or 3.6+. It's recommended to use the Anaconda distribution.
pytorch>=1.5.0:conda install pytorch(see pytorch.org for more installation options for your CUDA version)numpy:conda install numpyscipy:conda install scipy
tqdm:conda install tqdm[progress bars]networkx:conda install networkx[network visualization]matplotlib:conda install matplotlib[visualization]pytest:conda install pytest[to run tests]pandoc:conda install pandoc[to generate docs]sphinx:pip install sphinx nbsphinx[to generate docs]torch-lfilter:pip install torch-lfilter[faster lfilter for detectors]
If you're using Photontorch in your work or feel in any way inspired by it, we ask you to cite us in your work:
Floris Laporte, Joni Dambre, and Peter Bienstman. "Highly parallel simulation and optimization of photonic circuits in time and frequency domain based on the deep-learning framework PyTorch." Scientific reports 9.1 (2019): 5918.
- Complex tensor support. Complex tensors are not supported in PyTorch/Photontorch. Wherever complex tensors would be applicable, Photontorch expects a real-valued tensor with the real and imag part stacked in the first dimension. The Photontorch issue can be followed here and the PyTorch issue here.
- Sparse tensor support. A lot of memory usage can probably be avoided when transitioning to a sparse tensor representation for the connection matrices and scatter matrices. The Photontorch issue can be followed here
Photontorch used to be available under a custom Academic License, but Since October 2020, Photontorch is now fully open source and available under the AGPLv3.
Copyright © 2020, Floris Laporte - UGent - AGPLv3