Detection of HLS-based Hardware Trojans using Logistic Regression Binary-Classifier

Abstract

Literature has established that an attacker can exploit malicious/Trojan inserted high level synthesis (HLS) framework/tool for compromising hardware intellectual property (IP) designs. This paper presents a novel logistic regression (LR) based machine learning (ML) detection methodology capable of detecting the following seven state-of-the-art (SOTA) HLS Trojan variances: timer-sensitive trojan trigger based automated logic-transformation (TST-HT), time-bomb hardware Trojan (TB-HT), performance degradation hardware Trojan (PD-HT), battery exhaustion hardware Trojan (BE-HT), downgrade attack hardware Trojan (DA-HT), denial of service hardware Trojan (DoS-HT), and functional hardware Trojan (F-HT). The proposed detection technique is capable of very high accuracy with 0% FN and > 0% FP rates. © 2009-2012 IEEE.