Keynote Talk: Machine Learning for Hardware Security - from Application to Transformative Impact

With a broad spectrum of modern hardware security attacks, integrated circuit (IC) trust has become a primary concern across levels of design abstraction and fabrication. Cutting edge benefits of ML enable opportunities for ML-guided, secure electronic design and manufacturing. However, the complexity and variety of ML models, algorithms, and architectures pose a significant challenge - application of ML without advance ML expertise is at best sub-optimal. In addition, in-depth insight into modern complex circuits, design and fabrication processes, and hardware security threats is required to develop resilient ICs. How to create a transformative impact in hardware security with ML is still an open question. This talk will provide an overview of various security vulnerabilities as well as related design challenges, needs, and learning-guided trust methods. We will discuss recent ML-based approaches for on-chip detection of side-channel attacks and emerging directions for trusted fabrication in untrusted foundries, considering threats such as reverse engineering, overbuilding, counterfeiting, and IP piracy. We will then use a framework for hardware Trojan detection as a demonstration vehicle for closing the expertise gap among IC design, cybersecurity, and ML researchers as part of a transformative ML hardware security effort. The talk will conclude with a perspective on future opportunities at the intersection of these fields.