Feistel-PUF: Sequential Obfuscation-Based Machine Learning Attack-Resistant Physical Unclonable Function for IoT Device Security Authentication

doi:10.1109/JIOT.2025.3638868

Feistel-PUF: Sequential Obfuscation-Based Machine Learning Attack-Resistant Physical Unclonable Function for IoT Device Security Authentication

2025-12-01

0

PRE

AI

DOI：10.1109/JIOT.2025.3638868

原文链接

原文求助

摘要

En 中文

Device authentication protocols based on a strong physical unclonable function (PUF) show promise for enhancing Internet of Things (IoT) security. However, a strong PUF isvulnerable tomachine learning (ML) attacks. This article proposes a Feistel structure sequential obfuscation-based PUF (Feistel-PUF) to resist ML attacks. When the PUF is obfuscated by the Feistel structure, the challengeresponse relationship resembles that of sequential logic circuits. Specifically, each response depends on both current and historical challenges, thereby significantly increasing the complexity of the challengeresponse mapping. Experimental results showed that even with one million collected challengeresponse pairs (CRPs), the prediction accuracies of ML attacks on Feistel-PUF remained at approximately 50%. Notably, the obfuscation process excluded response data, thereby preserving the randomness, reliability, and uniqueness of PUF with negligible performance degradation. The Feistel structure was iteratively reused for sequential obfuscation processing, maintaining constant and minimal hardware overhead. We propose a novel approach that combines the Feistel-PUF with an authentication protocol. This system synchronizes obfuscation parameters during the initial registration phase and implements dynamic updates throughout the authentication process. This approach allows the obfuscation structure to be made public, thereby overcoming the reliance on structure or algorithm secrecy in traditional dynamic challenge obfuscation schemes. The security of the protocol was confirmed by subjecting it to ProVerif verification and security analysis.

Keyword:

Feistel structure

machine learning (ML)

physical unclonable function (PUF)

security authentication protocol

sequential obfuscation

Feistel-PUF: Sequential Obfuscation-Based Machine Learning Attack-Resistant Physical Unclonable Function for IoT Device Security Authentication

摘要

期刊

机构