AnalogGenie

A generative engine for automatic discovery of analog circuit topologies by representing circuits as Eulerian circuits and using a decoder-only transformer to predict the next device pin.

About This Work

For more details, please refer to our ICLR'25 paper: AnalogGenie: A Generative Engine for Automatic Discovery of Analog Circuit Topologies

Processed Dataset and Model Checkpoint

Processed dataset: https://huggingface.co/datasets/JianGao666/AnalogGenie

Model checkpoint: https://huggingface.co/JianGao666/AnalogGenie

How to Use

Environment Setup

This setup requires Anaconda. Run the following command below:

conda env create -f environment.yml

To activate the environment:

conda activate AnalogGenie

Dataset

Check data_categorization.md file for dataset categorization

Data Preprocessing

Convert SPICE netlist to adjacency matrix (SPICE2GRAPH_full.py will result in a more expressive but dense graph)

python SPICE2GRAPH_compress.py

Convert adjacency matrix to Eulerian circuit and perform augmentation

python Augmentation.py

Stack Eulerian circuits to NumPy array for training and validation

python Stack.py

Pretraining and Inference

Perform pretraining

python Pretrain.py

Perform inference and generate circuits

python Inference.py

Citation

If you use this framework for your research, please cite our ICLR'25 paper:

@inproceedings{
gao2025analoggenie,
title={AnalogGenie: A Generative Engine for Automatic Discovery of Analog Circuit Topologies},
author={Jian Gao and Weidong Cao and Junyi Yang and Xuan Zhang},
booktitle={The Thirteenth International Conference on Learning Representations},
year={2025},
url={https://openreview.net/forum?id=jCPak79Kev}
}

Contact Information

If you have any questions regarding using this framework, please feel free to contact us at gao.jian3@northeastern.edu.

Version History

• 0.1
  • Initial Release

License

This framework is licensed under the MIT License - see the LICENSE.md file for details