Reinforcement learning for logic synthesis.

This is the source codes for our paper "Exploring Logic Optimizations with Reinforcement Learning and Graph Convolutional Network", published at 2nd ACM/IEEE Workshop on Machine Learning for CAD (MLCAD), Nov. 2020.

The authors include Keren Zhu, Mingjie Liu, Hao Chen, Zheng Zhao and David Z. Pan.

The project is modified to run under:

python 3.12.3
Pytorch 2.2.0

The project has other dependencies such as numpy, six, etc. Please installing the dependencies correspondingly.

The project requires the Python API, abc_py, for Berkeley-abc.

Please refer to the Github page of abc_py for installing instruction.

Benmarks can be found in url.

The current version can execute on combinational .aig and .blif benchmarks. To run the REINFORCE algorithm, please first edit the python/rl/testReinforce.py for the benchmark circuit. And execute python3 testReinforce.py

Keren Zhu, UT Austin (keren.zhu AT utexas.edu)