PO-PUF: A Circuit Based on Dynamic Challenge Obfuscation and Tail-Path Perturbation for Resistance to Machine Learning Attacks

Abstract

Physical Unclonable Functions (PUFs) have been widely used in hardware security. Although strong PUFs can provide a large number of challenge-response pairs (CRPs), their underlying mappings are more vulnerable to machine-learning-based modeling attacks. To address this issue, this paper proposes a strong PUF design, termed Path-Obfuscated PUF(PO-PUF), which integrates dynamic challenge obfuscation with tail-path perturbation. In the proposed design, the input challenge adaptively generates the obfuscation mechanism to enable dynamic challenge reconstruction, thereby weakening the mapping relationship between the original challenge and path selection. Meanwhile, multi-mode crossover perturbations are introduced in the tail paths to alter the critical delay-race relationship and further increase the complexity of the CRP mapping. Experimental results show that the proposed PO-PUF achieves prediction rates close to the ideal random-guessing level of 50% under Logistic Regression (LR), Support Vector Machine (SVM), and Artificial Neural Network (ANN) attacks. In addition, its randomness, uniqueness, and reliability reach 49.98%, 49.83%, and 97.73%, respectively.