Machine Learning Methods to Detect Voltage Glitch Attacks on IoT/IIoT Infrastructures

Abstract

A majority of modern IoT/IIoT digital systems rely on cryptographic implementations to provide satisfactory levels of security. Hardware attacks such as side-channel analysis attacks or fault injection attacks can significantly degrade and even eliminate the desired level of security of the infrastructure in question. One of the most dangerous attacks of this type is voltage glitch attacks (VGAs), which can change the intended behavior of a system. By effectively manipulating the voltage at a specific time, an error can be injected that can change the intentional conduct and bypass system security features or even extract confidential information such as encryption keys by analyzing incorrect outputs of the firmware. This study proposes an innovative VGAs detection system based on advanced machine learning. Specifically, an innovative semisupervised learning methodology is used that utilizes a hybrid combination of algorithms. Specifically, a heuristic clustering method is used based on a linear fragmentation of group classes. In contrast, the ELM methodology is used as an algorithm for retrieving hidden variables through convex optimization.