Description
The field of side-channel analysis is constantly advancing, revealing vulnerabilities in cryp-
tographic systems. This thesis delves into studying the effectiveness and reliability of
side-channel attacks, particularly through a comparative study of the traditional Linear
Discriminant Analysis (LDA) template attack and potential emerging machine-learning tech-
niques. Renowned for its meticulous analysis of power traces, the LDA template attack
leverages training-derived coefficients for trace prediction. In this research, the use of Mul-
tiple Linear Regression to train these coefficients within the LDA context is proposed to
improve the model’s robustness. Moreover, the objective is to discover a machine-learning
algorithm that could either match or exceed the LDA template attack’s accuracy and noise re-
silience. This comparative study aims to hopefully underscore improvements in side-channel
attack techniques and to address the practical difficulties involved in integrating advanced
algorithms into the traditional template-attack paradigms.
|
Thesis topic in co-operation with the Fraunhofer Institute for Applied and Integrated Security AISEC, Garching