Detection of Hardware Attacks in Network on Chip Based on Machine Learning

Network on Chip (NoC) is commonly used as the interconnection topology in modern multi-core System on Chip (SoC) architecture. The extensive use of NoC and its pervasive integration across the chip render it a primary target for Hardware Trojan (HT) attacks. Modeling and detecting potential attacks are essential for ensuring the security of NoC. This paper proposes a HT detection scheme based on traffic-attacking HT. The Gem5 platform is utilized to establish attack models for extracting features. Then, using a Bayesian model to identify change points and a random forest algorithm to identify abnormal change points. The experimental results show that the average detection accuracy of Trojans can reach 99.43%.