unravel11 / iccad2019benchmarks

ICCAD-2019 Benchmarks

Machine Learning-Based Hotspot Detection: Fallacies, Pitfalls and Marching Orders

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.

Technical details of the proposed ICCAD-2019 benchmarks

• 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.

Simple ML-based hotspot detection flow (discussed in section 3 of the paper)

• 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 State-Of-The-Art (SOTA) methods

• Source code of DAC'17 [12] and TCAD'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 of VTS'18 [26], please contact us directly.

Citation

@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}
}