Machine Learning Methods for Hardware Security

Abstract

In the context of VLSI design systems, experts have often emphasized aspects, such as cost, power consumption, performance, and reliability in the integrated circuit (IC) design process. Until the last decade, security has always been an afterthought. However, with increasing count and power of hardware-based attacks, such as hardware Trojans and side-channel analysis, the need for securing the entire IC supply chain and design process has become imperative. Simultaneously, applications of machine learning (ML) have become an inseparable part of thwarting such threats and attacks. Over the past decade, an emerging wave has inspired researchers in the hardware security domain to apply ML-based algorithms for addressing security challenges. With the presence of an enormous number of offshore vendors for cost-effective design and fabrication procedure of ICs, emergence of malicious agents in the IC design process has become a persistent problem. In this chapter, we present existing challenges in hardware security and outline opportunities offered by ML algorithms to overcome such challenges. We investigate the impact of ML methods in terms of both hardware-for-security as well as security-of-hardware. In addition, we demonstrate supervised learning and unsupervised learning methods that are widely applied against potential threats.