This is the official open source repository for "A Timing Engine Inspired Graph Neural Network Model for Pre-Routing Slack Prediction" (DAC 2022).

Please refer to our Dataset repo for download instructions and documentation.

First, download the dataset and extract to <repo root>/data/.

Then, download our pre-trained weight here or here or here and put it under <repo root>/checkpoints/08_atcd_specul/15799.pth. (Note that Google drive is showing it as an archive but please download it as a whole.)

Finally, run the following command:

python3 train_gnn.py --test_iter 15799 --checkpoint 08_atcd_specul

This prints two tables indicating the net delay and arrival time prediction results, same as that in our paper.

data_graph.py: the data reader and pre-processor.

model.py: our customized GNN model implementation (please refer to the class NetConv, SignalProp, and TimingGCN). also inside it is our GCNII baseline implementation.

train_gnn.py: the model training and testing logic.

train_stat_rf.py, train_deepgnn.py: model training and testing for random forest and deep GCNII baselines.

Please cite our work if you find our code useful.

@inproceedings{mltimerdac22,
 author = {Guo, Zizheng and Liu, Mingjie and Gu, Jiaqi and Zhang, Shuhan and Pan, David Z. and Lin, Yibo},
 booktitle = {Proceedings of the 59th Annual Design Automation Conference 2022},
 organization = {ACM},
 title = {A Timing Engine Inspired Graph Neural Network Model for Pre-Routing Slack Prediction},
 year = {2022}
}