Practical aspects on non-profiled deep-learning side-channel attacks against AES software implementation with two types of masking countermeasures including RSM

Abstract Deep-learning side-channel attacks, applying deep neural networks to are known that can easily attack some existing countermeasures such as masking and random jitter. While there have been many studies on profiled deep-learning a new approach involves learning non-profiled attacks was proposed in 2018. In our study, we investigate the structure of multi-layer perceptrons points interest for using ANSSI database with countermeasure. The results investigations indicate it is better use simple network model, apply regularization prevent over-fitting, select wide range power traces contain information interest. We also implemented AES-128 software implementation protected Rotating Sboxes Masking countermeasure, which has never attacked by Xmega128 microcontroller carried out against it. Non-profiled successfully recovered all partial keys while conventional analysis could not. showed least significant bit adequate selection successful but best labeling method may vary depending countermeasure algorithm. conducted two experimental analyses clarify learn mask values used One gradient visualization previous studies, other removal traces.