Résumé
In 2019, ANSSI released a protected software implementation of AES running on an STM32 platform with ARM Cortex-M architecture, publicly available on Github. The release of the code was shortly followed by a first paper written by Bronchain and Standaert at Ches 2020, analyzing the security of the implementation and proposing some attacks. In order to propose fair comparisons for future attacks on this target device, this paper aims at presenting a new publicly available dataset, called ASCADv2 based on this implementation. Along with the dataset, we also provide a benchmark of deep learning based side-channel attacks, thereby extending the works of Bronchain and Standaert. Our attacks revisit and leverage the multi-task learning approach, introduced by Maghrebi in 2020, in order to efficiently target several intermediate computations at the same time. We hope that this work will draw the community's interest toward the evaluation of highly protected software AES, whereas some of the current public SCA datasets are nowadays reputed to be less and less challenging.