An Obfuscated Challenge Design for APUF to Resist Machine Learning Attacks

• Main Authors: Chen, Bo, Wang, Pengjun, Li, Gang
• Format: Conference Proceeding
• Language: English
• ISSN: 2162-755X
• DOI: 10.1109/ASICON47005.2019.8983648

Physical unclonable function (PUF) has emerged as a lightweight hardware security primitive for resource constrained devices. The arbiter PUF (APUF) is a typical kind of strong PUF. However, conventional APUF is vulnerable to machine learning (ML) attacks. In this paper, we propose an obfuscated challenge design for APUF (OC-APUF), which exchanges the bit positions in the challenge according to our design rules. The subsequent recovery of the obfuscated challenge by a recovery circuit is guaranteed. Then the corresponding response is produced by the APUF with the real challenge. The goal is to obscure the direct relationship between challenges and responses to prevent ML attacks. Most importantly, the unclonability of the APUF is preserved, and there is almost no increase in hardware complexity while still maintaining a high level of security. Experiment results show that 64-bit APUF with obfuscated challenge can resist ML attacks with a maximum prediction rate of 60% using the logistic regression (LR) strategy.