Efficient fuzzy extractor implementations for PUF based authentication

Yuejiang Wen, Yingjie Lao

Abstract: A strong security management of digital system should take all phases of lifecycle into consideration as the adversary can simply take advantage of the weakest link to compromise the systems, regardless of software, network, or hardware. As hardware devices are becoming more and more distributed and unsupervised, hardware security has emerged as an important challenge for cybersecurity. Various techniques have been developed recently to protect systems against malicious attacks from the hardware perspective. One promising hardware security primitive that can be used for authentication or identification is the physical unclonable function (PUF), which extracts chip-unique signatures from manufacturing process variations. However, as environmental variations could affect the characteristics of digital circuits, a PUF can hardly reproduce a consistent response. In the literature, error-correcting codes and fuzzy extractors are often used to correct the PUF errors before using the generated signature for authentication. This paper proposes novel architectures that incorporate different degrees of error-tolerance into PUF response correction to improve the efficiency of PUF-based authentication. One proposed fuzzy extractor implementation improves the reliability by eliminating high-entropy PUF response bits. An alternative method is to input low-entropy bits for authentication directly, while bypassing the error-correcting block. The security can be enhanced by employing this architecture, as the probability of a random guess that could pass the authentication is reduced, while the percentage of correctable keys only decreases slightly.