This repository extends the official C++ implementation of CFIRSTNET: Comprehensive Features for Static IR Drop Estimation with Neural Network. The original "Comprehensive Features" proposed in the paper have been modified to support not only Nangate45 but also sky130hd, asap7, and custom design data. We express our gratitude to the original authors for sharing their excellent research.

Overview

This project enhances the CFIRSTNET implementation to support IR Drop Estimation across multiple process technologies and custom designs. Key modifications include:

• Added support for sky130hd and asap7 process presets
• Custom metal layer and orientation configuration
• Automatic orientation detection
• Optimized channel configuration (M=3, generating 11 channels)

Note: The code was modified with assistance from ChatGPT 5 Pro and compared with the original implementation using vis.ipynb. Please be aware that unexpected errors may occur, and caution is advised.

License

This project adheres to the Apache License 2.0, as specified in the original CFIRSTNET repository. See the LICENSE file for details.
Note: The modified code also follows the Apache License 2.0, allowing commercial use, provided the license conditions are met.

Dependencies

• Compiler: g++-11 or later (with C++17 support)
• Library: Boost (specifically boost_iostreams)
• Operating System: Linux (Ubuntu 20.04 or later recommended)
• Optional: libstdc++fs (required in some environments)

Installing Dependencies

To install Boost on Ubuntu, run:

sudo apt update
sudo apt install libboost-all-dev

Build Instructions

Build the project on Linux using the following command:

g++ -O3 -std=c++17 -Wall -Wextra -pedantic main.cpp -o irdump -lboost_iostreams

Troubleshooting Build Errors

If you encounter filesystem-related errors, try:

g++ -O3 -std=c++17 -Wall -Wextra -pedantic main.cpp -o irdump -lboost_iostreams -lstdc++fs

Known Issues:

• Ensure libstdc++fs is available if filesystem errors occur (check g++ version).
• Verify that Boost is properly installed (see installation instructions above).

Usage

Run the program with the following example command:

./irdump /your_sp_path/sample.sp 384 384 --out-dir ./dump

Presets and Options

• sky130 Preset: Uses metal layers met1, met4, met5 (V, H, V) and saves channel data to CSV

  ./irdump /path/to/file.sp 384 384 --preset sky130 --out-dir ./dump

• asap7 Preset: Uses metal layers m5, m8, m9, m10 (H, V, H, V) with automatic orientation detection

  ./irdump /path/to/asap7.sp.gz 384 384 --preset asap7 --auto-ori --print-ori

• nangate45 Preset: Uses metal layers m1, m4, m7, m8, m9 (V, H, V, H, V)

  ./irdump /path/to/n45.sp.gz 256 256 --preset nangate45

• Custom Configuration: Specify custom metal layers and orientations (H/V or 2/1)

  ./irdump my.sp 512 512 --preset custom --metals m1,m3,m5 --ori V,H,V --out-dir ./dump --print-ori
  

Channel Configuration

• sky130: M=3 (Total 11 channels)
  • 0–4: Hypothetical IR Drop (HIRD, 5 channels)
  • 5–7: Resistive Distance (RD, 3 channels)
  • 8–10: Wire Resistance (WR, 3 channels)
• nangate45, asap7: M=5 (Total 23 channels)
  • 0–8: Hypothetical IR Drop (HIRD, 9 channels)
  • 9–15: Resistive Distance (RD, 7 channels)
  • 16–22: Wire Resistance (WR, 7 channels)

Usage Tips

• Automatic Orientation Detection: Use the --auto-ori flag to automatically detect metal layer orientations.
• Orientation Output: Add --print-ori to display the orientation settings.
• Output Directory: Results are saved to the directory specified by --out-dir.
• Comparison Analysis: Refer to vis.ipynb to compare the original and modified implementations.

Notes

• While the modified code has been tested, unexpected errors may occur in certain environments. Please report any issues.
• Ensure input files (e.g., .sp, .sp.gz) are correctly formatted.
• For details on the .sp file format and content, please refer to the BeGAN paper or the ICCAD-2023 contest C invitation paper.