A feature vector-based modeling attack method on symmetrical obfuscated interconnection PUF

Abstract

Physical unclonable function (PUF) are widely used in solutions such as device authentication and lightweight encryption due to their tamper-resistant, key-free storage and lightweight properties. However, the security of PUFs is threatened by modeling attacks. In this paper, we propose a novel modeling attack method for the symmetrical obfuscated interconnection physical unclonable function (SOI PUF) based on feature vectors. The proposed method introduces an innovative feature vector transformation technique and vector response pair to capture higher-order relationships with complex PUF architectures. Meanwhile, we propose two important principles for designing deep neural network (DNN) attack models. The experiments are systematically validated for the novel SOI PUF and cSOI PUF architectures, and the results show that, under equivalent dataset conditions, the proposed method achieves a higher attack success rate compared to the traditional challenge-response pair-based modeling approaches, achieving an accuracy of 98.42% in modeling SOI PUF. This study provides valuable theoretical and practical insights for enhancing PUF security and designing attack-resistant PUF architectures.