Novelty Detection for SilGeo Hardware Assurance

Abstract

In today's world, electronic hardware-level threats have become increasingly common. These threats can range from the infiltration of counterfeit or malicious hardware in the supply chain to the use of electronic attack tools. Due to the ever-evolving nature and customizability of inauthentic electronics, it is difficult to validate the authenticity of hardware. While measurement devices have been created to detect these threats, their successful deployment requires a high level of expertise.

This thesis addresses these challenges by proposing a novelty detection method for the \gls{silgeo} hardware validation platform with low deployment barriers. It has been shown that this method can be trained on as few as three valid devices, and the entire training and application process is fully automated, requiring no expertise. The method incorporates Bayesian statistical models that rely on carefully selected assumptions and domain knowledge. Furthermore, maximum false positive rates can be estimated and adjusted without additional data.

The presented method is tested in several case studies on devices ranging from surface-mount integrated circuits to Wi-Fi-enabled disguised attack tools. In each case, the estimated maximum false positive rate exceeded the observed false positive rate, and most counterfeit and malicious devices were identified. This thesis presents a practical solution to detecting and validating hardware in a rapidly changing threat landscape.