A lightweight general PUF framework for resisting machine learning attacks

Physical Unclonable Function (PUF) is an attractive and low-cost security primitive that requires no storage and is resistant to reverse engineering. However, classical PUFs are highly vulnerable to machine learning attacks, and most attempts to resist these attacks consume excessive resources. To address this challenge, a lightweight general PUF framework is proposed in this paper. Firstly, the framework adopts segmentation processing to introduce structural nonlinearities for the purpose of self-protection. Secondly, the pre-segment response, pre-segment challenges and post-segment challenges undergo XOR processing to introduce challenges obfuscation, which greatly enhances the machine learning resistance of the PUF. In addition, for configurable RO PUF, a novel MUX-based RO (called MRO) is proposed in this paper, which can save resources by 50 %. Implementing a two-segment MRO-MRO instance based on the proposed PUF framework results in reliability, uniformity, and uniqueness that are close to the ideal values. Comprehensive experiments demonstrate that the proposed PUF has the advantages of scalable framework, low resource overhead, and strong resistance to machine learning attacks.