Detecting HLS Hardware Trojans using Random Forest Classifier

Abstract

High-level Synthesis (HLS) generated IP designs are widely and effectively used in several image/video processing applications, consumer electronics applications as well as multimedia domains. However, several prior works have resolutely established major security vulnerabilities that HLS design process exposes. HLS framework/design flow can be maliciously exploited by an adversary/attacker. These vulnerabilities sanction malicious Trojan to be furtively injected during HLS design phases such as scheduling, mux-interconnect design stage, RTL datapath etc, resulting into compromised IP designs. This paper presents novel technique for detecting HLS hardware Trojans (HLS-HT) using random forest classifier based machine learning. The proposed detection framework is capable of detecting various types of state-of-the-art HLS-HT including performance degradation hardware Trojan (PD-HT), denial of service hardware Trojan (DoS-HT), battery exhaustion hardware Trojan (BE-HT), downgrade attack hardware Trojan (DA-HT), functional hardware Trojan (F-HT) and Time-bomb hardware Trojan (TB-HT). The proposed approach achieves high detection accuracy with zero false negatives for specific HLS-HTs considered here, ensuring sturdy IP design security. © 2013 IEEE.