Enhancing Arbiter PUF Against Modeling Attacks Using Constant Weight Encoding

Physically Unclonable Functions (PUFs) have been promoted as a lightweight security solution for IoT devices, offering authentication without the need to store a secret key or perform complex cryptography. However, PUFs are susceptible to modeling attacks, where adversaries use machine learning to replicate the PUF. Research efforts have focused on improving PUF designs to enhance their resistance to these attacks. In this work, we propose a method to bolster the security of PUFs against modeling attacks. Our approach first obfuscates the challenges using an XOR operation, followed by transforming the challenges into constant weight vectors through encoding. This method introduces randomness by dynamically generating random bits for the XOR step and also uses these bits to generate the weight for encoding. By obscuring the relationship between Challenge-Response Pairs (CRPs), our method makes modeling significantly harder. Testing with varying numbers of CRPs, our results show that modeling accuracy changes only subtly, with a maximum accuracy of 56