DE-HNN: An effective neural model for Circuit Netlist representation

This is the github repository of The AIStats 2024 paper:
"DE-HNN: An effective neural model for Circuit Netlist representation. Z. Luo, T. Hy, P. Tabaghi, D. Koh, M. Defferrard, E. Rezaei, R. Carey, R. Davis, R. Jain and Y. Wang. 27th Intl. Conf. Artificial Intelligence and Statistics (AISTATS), 2024." arxiv

Important: Recent Updated:

A new version for UT Austin Data is updated at "de_hnn_tx/"

This new version is easier to use, more compatible, and require much less RAM.

Please see detailed instruction in New README

Following the environment setup is still strongly recommended.

Environment Setup

We built based on python 3.8.8, CUDA 11.6, Linux Centos 8.

Please consider using virtual environment like anaconda for easier use.

The dependencies are recorded in requirements.txt.

Notice that we put some CUDA-related packages in

cuda_related_package.txt

Those packages might need to be manually installed to fit your device's CUDA version.

If you believe your CUDA version fits the requirements, please run:

pip install dgl-cu113 -f https://data.dgl.ai/wheels/repo.html
pip install torch==1.13.1+cu116 torchvision==0.14.1+cu116 torchaudio==0.13.1 --extra-index-url https://download.pytorch.org/whl/cu116

Other packages can be installed by run:

pip install -r requirements.txt

How to Download Netlist Data

A seperate README for Netlist Data is available here: Data README

Raw data is available at link.

After downloading the raw data, please extract and put the raw data directory to "/de_hnn/data/" directory.

After that, please run following command to generate the full processed dataset. The whole process can last for hours.

source run_all_python_scripts.sh

Skip Processing Data

If one does not want to process the raw data for any reason but just need the processed data, the full processed data is also available at link.

After downloading the processed data, please extract and put the data directory to "/de_hnn/data/" directory.

How to train

After the data is processed, the experiments can be run following the instrcutions below.

Simple Test

First, enter directory "/de_hnn/experiments/cross_design/".

A simple test on cross-design full-DE-HNN can be run using the following command:

source run_simple_test.sh

A training process will be initiated immediately. This is a simple test just to test if codes are ready to use.

A more complete test run with full number of epoches can be run using the following command:

source run_full_test.sh

The whole process might take 30 minutes to hours depending on devices, which will produce the full results for full-DE-HNN for cross-design.

For full single-design experiments

All files are ready in directory "/de_gnn/experiments/single_design/".

Please run the commands in this format "source run_all_{model}.sh" depending on which model you want to run.

For full cross-design experiments

All files are ready in directory "/de_gnn/experiments/cross_design/".