Paper 2016/970

Statistical Analysis for Access-Driven Cache Attacks Against AES

Liwei Zhang, A. Adam Ding, Yunsi Fei, and Zhen Hang Jiang

Abstract

In recent years, side-channel timing attacks utilizing architectural behavior have been applied to cloud settings, presenting a realistic and serious cyber threat. Access-driven cache attacks allow the adversary to observe side-channel leakage (cache access pattern) of a critical cryptographic implementation to infer the secret key. However, what the attackers observe may deviate from the real cache footprint of the victim process, affecting the effectiveness of cache-based timing attacks using the observed leakage. Various countermeasures, including secure cache and architectures design, should also be evaluated accurately for their side-channel resilience. To address this need, this paper proposes a mathematical model for access-driven cache attacks, and derives explicit success rate formulas for those attacks. It is the first theoretical model that explicitly considers the misclassification errors for cache access and cache non-access by the victim cryptographic process. We implement several access-driven cache attacks and use our models to evaluate them. We demonstrate that the proposed statistical model predicts the success rate of cache-based timing attacks accurately. We also apply the model onto various cache defense architectures for evaluation.