HTnet: Transfer Learning for Golden Chip-Free Hardware Trojan Detection

This is the companion repository for our paper titled "HTnet: Transfer Learning for Golden Chip-Free Hardware Trojan Detection" published in [The Design, Automation, and Test in Europe (DATE'2021)] conference.

Data

The data used in this project comes from the following source:

The Our archive, which contains side-channel signals for multiple hardware Trojan Infected circuits.

Code

The code is divided as follows:

The trainOnAllExecptTwoModality_different_classes.py python file to train base HTnet.
The tl_learning_tests_TwoModality_double_net.py python file to use the base base HTnet for feature extraction.
The utils folder contains the necessary functions to read the datasets and summarize the results.
The classifiers folder contains python files deep neural network models. The [htnet.py] is used when finding the best model structure. The htnetNarrowed.py is the final model published in the paper and used for training the base model. The transferLearningDoubleNetAnomalyDetectorGroup.py present the double net transfer learning approach provided in the paper.

To run the code, the data should be placed in the HT_Data folder. Then, the base model should be trained and tested as follows:

python3 trainOnAllExecptTwoModality_different_classes.py
python3 tl_learning_tests_TwoModality_double_net.py

Prerequisites

All python packages needed are listed in pip-requirements.txt file and can be installed simply using the pip command.

Acknowledgment

This work was supported by the Office of Naval Research (ONR) under Award N00014-17-1-2499.

Reference

If you re-use this work, please cite:

@article{faezi2021htnet,
  Title                    = {HTnet: Transfer Learning for Golden Chip-Free Hardware Trojan Detection},
  Author                   = {Faezi, Sina and Yasaei, Rozhin and Al Faruque, Mohammad},
  journal                = {The Design, Automation, and Test in Europe},
  Year                     = {2021}
}