DRL-modelling-of-PUFs

Code to run Deep Reinforcement Learning (DRL) modelling attacks against Delay-based PUFs.

The available PUFs are:

• Arbiter PUF
• k-XOR PUF
• Interpose PUF

The DRL algorithm used is Deep Q-Network (DQN), and the implemented modelling attacks are:

• Supervised Learning:
  • Basic approach
  • Federated learning approach
  • Majority vote approach
• Modified Probably Appoximately Correct (PAC) approach
• IPUF attacks:
  • Naive attack
  • Collaborative attack

How to cite

M. Ferens, E. Dushku, and S. Kosta, "On the Feasibility of Deep Reinforcement Learning for Modeling Delay-based Physical Unclonable Functions," 20th International Conference on Wireless and Mobile Computing, Networking and Communications (WiMob), pp. 1-8, 2024, in press.

Dependencies

The DQN attacks were tested using the following:

• Python 3.8
• pypuf 2.2.0
• numpy 1.23.1
• chainerrl 0.8.0
• tensorflow 2.10.0
• pandas 1.5.1

The LR or MLP attacks were tested using the following:

• Python 3.8
• pypuf 3.2.1
• numpy 1.23.1
• tensorflow 2.4.4
• pandas 1.5.1

How to use

To run single experiments using DRL DQN algorithm use one of the following (use --help for a list of arguments):

python3 DQN_attack_XORPUF.py
python3 DQN_replicant_attack_XORPUF.py
python3 DQN_vote_attack_XORPUF.py
python3 DQN_delay_attack_XORPUF.py
python3 DQN_attack_IPUF.py
python3 DQN_collaborative_attack_IPUF.py

All previous scripts require in the same directory one the classes defined in PUF_env.py and PUF_delay_env.py.

To run single experiments using Supervised Learning Logistic Regression (LR) or Multi-Layer Perceptron (MLP), and the splitting attack on IPUF, use on of the following (use --help for a list of arguments):

python3 LR_attack_XORPUF.py
python3 MLP_attack_XORPUF.py
python3 splitting_LR_attack.py
python3 splitting_MLP_attack.py

References

1. N. Wisiol, C. Gräbnitz, C. Mühl, B. Zengin, T. Soroceanu, N. Pirnay, K. T. Mursi, and A. Baliuka, "pypuf: Cryptanalysis of Physically Unclonable Functions," 2021, version v2. [Online]. Available: https://doi.org/10.5281/zenodo.3901410
2. Y. Fujita, P. Nagarajan, T. Kataoka, and T. Ishikawa, "ChainerRL: A Deep Reinforcement Learning Library," in Journal of Machine Learning Research, vol. 22, no. 77, pp. 1-14, 2021.
3. V. Mnih, K. Kavukcuoglu, D. Silver, A. A. Rusu, J. Veness, M. G. Bellemare, A. Graves, M. Riedmiller, A. K. Fidjeland, G. Ostrovski, S. Petersen, C. Beattie, A. Sadik, I. Antonoglou, H. King, D. Kumaran, D. Wierstra, S. Legg, and D. Hassabis, "Human-level Control through Deep Reinforcement Learning," in Nature, vol. 518, pp. 529-533, 2015.