This repository is the official repository for our paper "CNN-Cap: Effective convolutional neural network based capacitance models for full-chip parasitic extraction" and "CNN-Cap: Effective convolutional neural network based capacitance models for interconnect capacitance extraction". This repository provides several datasets and models for 2-D and 3-D interconnect capacitance extraction.
- git clone git@github.com:ydc123/CNNCap.git
- pip install -r requirements.txt
For the 2-D pattern, we generated 10 datasets on 5 layer combinations, consisting of Pattern-B and Pattern-C. Please refer to dataset.py to learn how to load these datasets. For each datasets, we provide 2 models to predict the total capacitance and coupling capacitance respectively.
We have released these datasets and models on Baidu Netdist (link, 链接 with password 6sp8). After downloading and extracting the compressed file, you will get the following files:
data
├── 15nm_B_2_4_6.json
├── 15nm_B_2_4_9.json
├── 15nm_B_3_6_8.json
├── 15nm_C_2_4_6.json
├── 15nm_C_2_4_9.json
├── 15nm_C_3_6_8.json
├── 55nm_B_2_3_6.json
├── 55nm_B_2_4_6.json
├── 55nm_C_2_3_6.json
└── 55nm_C_2_4_6.json
saved_models
├── best.model_CNNCap_mse_15nm_B_2_4_6_total.pth.tar
├── best.model_CNNCap_mse_15nm_B_2_4_9_total.pth.tar
├── best.model_CNNCap_mse_15nm_B_3_6_8_total.pth.tar
├── best.model_CNNCap_mse_15nm_C_2_4_6_total.pth.tar
├── best.model_CNNCap_mse_15nm_C_2_4_9_total.pth.tar
├── best.model_CNNCap_mse_15nm_C_3_6_8_total.pth.tar
├── best.model_CNNCap_mse_55nm_B_2_3_6_total.pth.tar
├── best.model_CNNCap_mse_55nm_B_2_4_6_total.pth.tar
├── best.model_CNNCap_mse_55nm_C_2_3_6_total.pth.tar
├── best.model_CNNCap_mse_55nm_C_2_4_6_total.pth.tar
├── best.model_CNNCap_msre_15nm_B_2_4_6_env.pth.tar
├── best.model_CNNCap_msre_15nm_B_2_4_9_env.pth.tar
├── best.model_CNNCap_msre_15nm_B_3_6_8_env.pth.tar
├── best.model_CNNCap_msre_15nm_C_2_4_6_env.pth.tar
├── best.model_CNNCap_msre_15nm_C_2_4_9_env.pth.tar
├── best.model_CNNCap_msre_15nm_C_3_6_8_env.pth.tar
├── best.model_CNNCap_msre_55nm_B_2_3_6_env.pth.tar
├── best.model_CNNCap_msre_55nm_B_2_4_6_env.pth.tar
├── best.model_CNNCap_msre_55nm_C_2_3_6_env.pth.tar
└── best.model_CNNCap_msre_55nm_C_2_4_6_env.pth.tar
For the 3-D pattern, we generate datasets from a real layout design. See here for details of the datasets.
You can run the following command to predict the total capacitance of 15nm_B_2_4_6 dataset with the model we provide.
python test.py --model saved_models/best.model_CNNCap_mse_15nm_B_2_4_6_total.pth.tar \
--logfile log/log_eval_model_CNNCap_mse_15nm_B_2_4_6_total.txt \
--data_path data/15nm_B_2_4_6.json \
--goal total
You can also train a model for predicting dominant capacitance on the 15nm_B_2_4_6 dataset by running the following command.
python train.py --lr 1e-5 \
--batch_size 64 \
--savename model_tmp.pth \
--logfile log/log_tmp.txt \
--epoch 100 \
--goal env \
--data_path data/55nm_C_2_3_6.json \
--loss msre
For more details on training and testing code, please refer to train.py and test.py.
You can run the following command to predict the total or coupling capacitance with the model we provide.
The checkpoint files can be downloaded from here.
After downloading and extracting the compressed file, you will get resnet34_total.pth and resnet34_env.pth,
which are the pretrained parameters of resnet34 models for total capacitance and coupling capacitance.
cd 3d/model
python infer.py --model_type resnet34 \
--goal env \
--filter_threshold 0.05 \
--bs 32 \
--pretrained path/to/checkpoint
If you benefit from our work in your research, please consider to cite the following paper:
@inproceedings{yang2021cnn,
title={CNN-Cap: Effective convolutional neural network based capacitance models for full-chip parasitic extraction},
author={Yang, Dingcheng and Yu, Wenjian and Guo, Yuanbo and Liang, Wenjie},
booktitle={2021 IEEE/ACM International Conference On Computer Aided Design (ICCAD)},
pages={1--9},
year={2021},
organization={IEEE}
}
@article{yang2022cnn,
title={CNN-Cap: Effective Convolutional Neural Network Based Capacitance Models for Interconnect Capacitance Extraction},
author={Yang, Dingcheng and Li, Haoyuan and Yu, Wenjian and Guo, Yuanbo and Liang, Wenjie},
journal={ACM Transactions on Design Automation of Electronic Systems (TODAES)},
year={2022},
publisher={ACM New York, NY}
}
Please feel free to contact us if you have any questions.