Design of Reliable and Modeling-Attack Resistant Strong PUFs for Lightweight Applications

Physical Unclonable Functions (PUF) have been proposed as security primitives for security applications. However, previous studies indicate that strong PUFs are vulnerable under machine learning (ML) based modeling attacks as such algorithms can achieve extremely high prediction accuracy. The reliability of PUFs is also a concern. In this paper, we propose to enhance the reliability and resistance to modeling attacks from the system perspective. With the help from the server, we can improve the PUF reliability by using error correction code. Furthermore, we propose to use S-boxes to confuse the attackers. A novel attack strategy based on stronger neural networks is proposed to assess the security. Experimental results show that the proposed method can improve PUF reliability and unpredictability with low hardware overhead.