Cypress: VLSI-Inspired Scalable GPU-Accelerated PCB Placement

Cypress is a scalable, GPU-accelerated PCB placement method inspired by VLSI. It incorporates tailored cost functions, constraint handling, and optimized techniques adapted for PCB layouts.

How to Build

You can build in two ways:

• Build without Docker by following the instructions of the DREAMPlace build at README_DREAMPlace.md.
• Use the provided Dockerfile to build an image with the required library dependencies.

Run PCB Placement

cd /Cypress
python dreamplace/Placer.py path/to/your/config.json

Run Hyperparameter search

./tuner/run_tuner.sh 1 1 	test/tune/pcb-configspace.json 	/path/to/aux_file 	test/tuner/tuner-ppa.json "" 	100 4 0 0 10 	./tuner 	./results/result_path

Acknowledgement

Cypress builds upon and extends the following open-source projects:

• DREAMPlace: A deep learning toolkit-enabled VLSI placement framework that provides the foundation for our GPU-accelerated placement engine.

• AutoDMP: An automated design methodology platform that inspired aspects of our hyperparameter tuning approach.

We thank the authors and contributors of these projects for their valuable work that made Cypress possible.

Citation

If you find this work useful, please cite the following paper:

@inproceedings{zhang2025cypress,
  title={Cypress: VLSI-Inspired PCB Placement with GPU Acceleration},
  author={Zhang, Niansong and Agnesina, Anthony and Shbat, Noor and Leader, Yuval and Zhang, Zhiru and Ren, Haoxing},
  booktitle={Proceedings of the 2025 International Symposium on Physical Design},
  pages={xx--xx},
  year={2025}
}