DeepGate4: Efficient and Effective Representation Learning for Circuit Design at Scale

Overview

Our paper is avaiable at Arixv and OpenReview.

Environment

To install the library deepgate, please refer to python-deepgate.

Dataset Preparation

We provide sample raw data and corresponding processed data in ./raw_data and ./raw_sample_data respectively.

To prepare your own data:

• 1. cd ./simulator and bash ./build.sh
• 2. prepare your own raw aig data in ./YOUR_RAW_DATA
• 3. python ./src/dg_dataset/data_preparation.py --aig_dir ./YOUR_RAW_DATA --save_path ./YOUR_DATASET_DIR

Training

You can run experiment with ./run/train_large.sh and ./run/train_large_baseline.sh

• ./run/train_large.sh denotes running model with our updating strategy
• ./run/train_large_baseline.sh denotes running model with its original strategy

We further offer various baseline models:

• ./run/train_large.sh offers models with baseline(DeepGate2), plain(DeepGate3), sparse(DeepGate4), GraphGPS, Exphormer and DAGformer
• ./run/train_large_baseline.sh offers models with PolarGate, DeepGate2, GraphGPS, Exphormer, DAGformer, GCN, GraphSAGE, GAT and PNA