Review of machine learning based hardware Trojan detection methods

Abstract

As integrated circuit (IC) design gets more complicated, outsourcing parts of the IC design and fabrication is commonly applied to simplify the production and reduce the cost. This leads to the threat of malicious manipulation to the design by the third parties involved. Such threat is considered as hardware Trojan attack, which could pose adverse impacts to a system or network. Recently, hardware Trojan is gaining more interest as a research subject, especially pre-silicon detection. Various kinds of hardware Trojan detection approaches have been proposed to detect different Trojan types in different circuits. This study summarises the existing hardware Trojan detection methods and discusses the attributes of the methods to better distinguish them between each other. Existing presilicon detection methods are reviewed, which includes verification-based, threshold-based and machine learning-based feature analysis techniques. The objective of this study is to ease the future hardware-Trojan-related research by providing an organised summary of detection techniques, especially pre-silicon detection.