Generalized Power Attacks Against Crypto Hardware Using Long-Range Deep Learning

To make cryptographic processors more resilient against side-channel attacks,engineers have developed various countermeasures. However, the effectiveness ofthese countermeasures is often uncertain, as it depends on the complexinterplay between software and hardware. Assessing a countermeasure'seffectiveness using profiling techniques or machine learning so far requiressignificant expertise and effort to be adapted to new targets which makes thoseassessments expensive. We argue that including cost-effective automated attackswill help chip design teams to quickly evaluate their countermeasures duringthe development phase, paving the way to more secure chips. In this paper, we lay the foundations toward such automated system byproposing GPAM, the first deep-learning system for power side-channel analysisthat generalizes across multiple cryptographic algorithms, implementations, andside-channel countermeasures without the need for manual tuning or tracepreprocessing. We demonstrate GPAM's capability by successfully attacking fourhardened hardware-accelerated elliptic-curve digital-signature implementations.We showcase GPAM's ability to generalize across multiple algorithms byattacking a protected AES implementation and achieving comparable performanceto state-of-the-art attacks, but without manual trace curation and within alimited budget. We release our data and models as an open-source contributionto allow the community to independently replicate our results and build onthem.