TrojanSAINT: Gate-Level Netlist Sampling-BasedInductive Learning for Hardware Trojan Detection

Training/Validation/Test Preparation

To prepare your training, validation and test sets, first ensure you have the train, val, and test directories. If not, you can simply make them by running

mkdir train
mkdir val
mkdir test

Go into your nets directory

cd nets

Run split.py. Be careful that you have nothing important in your train, val, and test as this script will empty them. split.py accepts a single integer as argument indicating which file will be used for testing. This integer indicates the position of the file in alphabetical order (so an argument of 0 will select rs232t1000.v, for example)

python split.py

Converting Into Graph

We provide a parser that takes the verilog files as input and converts them into the format needed for use with GraphSAINT

To run it, you first need to prepare your training, validation and test sets as described above. Then simply run:

python parser.py

Note that the script needs verilog_lib.v in the same directory. Once it is run, adj_full.npz, adj_train.npz, class_map.json, feats.npy, role.json will be generated. Those can then be used with GraphSAINT

Training and Inference with GraphSAINT

Training and inference is done on GraphSAINT, which is included here as submodule.

We make slight modifications to the graphsaint/pytorch_version/train.py file to save the outputs, and we include the modified version in this repo, so make sure to use this submodule here. To run training, simply follow the instructions on the GraphSAINT repository using the new files and the DATE21.yaml