Link to the paper: [https://ieeexplore.ieee.org/document/8942128]
The Datasets and the source code mentioned in the paper are shared in this repository.
- Layer Numbers
- 0: Extent
- 10: Metal polygons
- 21: Hotspot core marker
- 23: Non-Hotspot core marker
- Hotspots and Non-Hotspots can be identified in one of the following ways:
- Through their corresponding marker layers.
- Cellnames: Every pattern in the file has a unique name in the form of a cellname. Hotspot patterns contain the keyword
_hotspot
in their cell names, whereas, Non-Hotspot patterns contain_nonhotspot
. - Labels: Every pattern contains a text label at its center (in layer 0). The label is same as the cell name.
- Identifying Truly-Never-Seen-Before (TNSB) hotspots within the Testing Dataset - 1:
- A CSV file containing the cell names of TNSB hotspots is included in the same folder as the Testing Dataset - 1.
- The source code (Jupyter notebook (Python 2.7)), training and testing datasets, and the pre-trained models are made available.
- Users can either use the pre-trained models or re-train them locally. Instructions to switch between the two modes, to change dataset paths etc., are provided in the main code.
- Source code of
DAC'17 [12]
andTCAD'18 [11]
can be found in link - Source code of
SMACD'18 [13]
can be found in link - We have not publicly released the source code of
VTS'18 [26]
yet. Therefore, to obtain the source code ofVTS'18 [26]
, please contact us directly.
@inproceedings{reddy2019machine,
title={Machine Learning-Based Hotspot Detection: Fallacies, Pitfalls and Marching Orders},
author={Reddy, Gaurav Rajavendra and Madkour, Kareem and Makris, Yiorgos},
booktitle={2019 IEEE/ACM International Conference on Computer-Aided Design (ICCAD)},
pages={1--8},
year={2019},
organization={IEEE}
}