Machine Learning based Side Channel Selection for Time-Driven Cache Attacks on AES

Abstract

Side-channel attacks use indirect information about cryptographic operations from the targeted system. This makes the attacks highly effective on the system. In this paper, we explore the time-based cache attack that uses the time feature and cache information as secondary channel information. We selected AES algorithm to accomplish the time-based side channel attack. This time-based side channel attack targets the secret key in the last cycle of AES algorithm. We use machine learning models to extract information from secondary channels to determine vulnerabilities of the system. We use tree models on the time profiles created during the attack then we evaluated the most significant characteristics of the attack. Since Decision Tree, Random Forest, Gradient Boosting Model, and Extreme Gradient Boosting algorithms are very sensitive to processing tasks, we selected them as tree algorithms. Analysis results show that “cycle on average” information helps to predict the time-driven cache attacks. Moreover, Extreme Gradient Boosting algorithm provides better results.