EDA-IRdrop-Prediction
EDA-IRdrop-Prediction copied to clipboard
ycchen218
This is a deep-learning based model for Electronic Design Automation(EDA), predicting the IR drop location on the chip.
EDA IR Drop Prediction
Update
- 2023/08/04
Add the model overview and brief description. - 2023/6/24
Add the IR_drop_train.py file.
Delete model weight
Get model weight at GoogleDrive
Introduce
IR drop analysis is an essential step for IC design, evaluating the power integrity of a chip power delivery network. This is a deep-learning-based EDA tool for predicting the IR Drop heatmap, I achieve high precision and accuracy by inputting the power features. Additionally, for one prediction process only costs about 2 seconds.
Requirement
- python3.8
- scipy
- matplotlib
- numpy
- opencv
- pandas
- pytorch 1.12.0
Model Overview
| feature | discribe |
|---|---|
| $p_{i}$ | Internal power |
| $p_{s}$ | Switching power |
| $p_{sac}$ | Toggle rate scaled power |
| $p_{all}$ | $p_{i}+p_{s}+p_{sac}$ |
| $p_{t}$ | The power of an instance at each time step |
For more feature information V. A. Chhabria et al. [1] have discribe the detail of the feature extraction process.
The CE-FPN model is proposed by Y. Luo et al. [2].
- The purple CNN block is the CoordConv.
- SSF is the Sub-pixel Skip Fusion.
- SCE block is the Sub-pixel Context Enhancement.
- CAG block is the Channel Attention Guided Module.
Train
python IR_drop_train.py
--root_path: The path of the data file
--batch_size: The batch size figure
--num_epochs: The training epochs
--learning_rate: learning rate [0,1]
--weight_path: The path to save the model weight
--fig_path: The path of the figure file
Predict
python IR_drop_predict.py
--data_path: The path of the data file
--fig_save_path: The path you want to save figure
--weight_path: The path of the model weight
--output_path: The path of the predict output with .npy file
--irdrop_threshold: irdrop_threshold [0,1]
--device: If you want to use gpu type "cuda"
Predict result
- Tune your own irdrop_threshold, the defalt is 0.1 as shown in following figure.
- The output coordinate csv file and image array npy file are in the ./output file.
- The model predict cost time is 1.49 ~ 2.2 sec.
Compare with ground truth
Cross validation while evalulate the model
ROC Curve:
SSIM score: 0.863
AUC: 0.991
Percision: 0.9975
by pytorch_msssim.SSIM
by the same metrics code as CircuitNet
Reference
[1] V. A. Chhabria, Y. Zhang, H. Ren, B. Keller, and S. S. Sapatnekar, "Mavirec: mL-aided vectored ir-drop estimation and classification," *Design, Automation & Test in Europe Conference & Exhibition*, pp. 1825-1828, 2021.
[2] Y. Luo, X. Cao, J. Zhang, X. Cao, F. Guo, H. Shen, T. Wang, and Q. Feng, "Ce-fpn: enhancing channel information for object detection," ArXiv Computer Vision and Pattern Recognition, pp. 1-9, 2021.
ycchen218