PhD in Computer Science

The crypto algorithm is safe in theory, but its implementations still leak information to side-channels and can be attacked with Differential Power Analysis (DPA), Correlation Power Analysis (CPA) and state-of-the-art machine learning (ML). Even though SCA countermeasure method like masking is applied to make the implementations safer under DPA and CPA, ML can learn the features of the mask and the sensitive data and be able to attack the protected implementations. The training and attacking process of ML does not show what and where those features come from, and so cannot explain how an ML model can successfully combine those leakages in analysis. It leads to the requirement of explainable ML models to identify leakage locations that are combined in the attack, so that additional countermeasure methods can be applied to the correct executions to strengthen the security of crypto under SCA.