prefab 0.2.5

Machine learning based prediction of photonic device fabrication

PreFab

PreFab models fabrication process induced structural variations in integrated photonic devices using deep learning. New insights into the capabilities of nanofabrication processes are uncovered and device design fidelity is enhanced in this virtual nanofabrication environment.

Prediction

PreFab predicts process-induced structural variations such as corner rounding (both over and under etching), washing away of small lines and islands, and filling of narrow holes and channels in planar photonic devices. The designer then resimulates their (predicted) design to rapidly prototype the expected performance and make any necessary corrections prior to nanofabrication.

Predicted fabrication variation of a star structure on a silicon-on-insulator e-beam lithography process.

Correction

PreFab also makes automatic corrections to device designs so that the fabricated outcome is closer to the nominal design. Less structural variation generally means less performance degradation from simulation to experiment.

Corrected fabrication of a star structure on a silicon-on-insulator e-beam lithography process.

Models

Each photonic foundry requires its own predictor and corrector models. These models are updated regularly based on data from recent fabrication runs. The following models are currently available:

Foundry	Process	Latest Version (Date)	Type	Status	Name	Usage
ANT	NanoSOI	v4 (Apr 12 2023)	Predictor	Beta	p_ANT_NanoSOI_v4	Open
ANT	NanoSOI	v4 (Apr 12 2023)	Corrector	Beta	c_ANT_NanoSOI_v4	Open

This list will update over time. Usage is subject to change. Please contact us or create an issue if you would like to see a new foundry and process modelled.

Installation

Local

Install the latest version of PreFab locally using:

pip install prefab

Alternatively, you can clone this repository and then install in development mode using:

git clone https://github.com/PreFab-Photonics/PreFab.git
cd PreFab
pip install -e .

This will allow any changes made to the source code to be reflected in your Python module.

Online

You can also run the latest version of PreFab online by following:

Getting Started

Please see the notebooks in /examples to get started with making predictions.

Performance and Usage

Currently, PreFab models are accessed through a serverless cloud platform that has the following limitations to keep in mind:

• 🐢 Inferencing is done on a CPU and will thus be relatively slow. GPU inferencing to come in future updates.
• 🥶 The first prediction may be slow, as the cloud server may have to perform a cold start and load the necessary model(s). After the first prediction, the server will be "hot" for some time and the subsequent predictions will be much quicker.
• 😊 Please be mindful of your usage. Start with small examples before scaling up to larger device designs, and please keep usage down to a reasonable amount in these early stages. Thank you!

License

This project is licensed under the terms of the LGPL-2.1 license. © 2023 PreFab Photonics.