DLSCA results on Riscure Pinata (ARM Cortex-M4F)

All the code is available in the code folder.

The results are divided by target (SBOX_OUT or HW_SO), number of devices (1d or 2d) and type of dataset (fixed or random key)

• plots contains the training/validation accuracy and loss
• ge_plots_comparison contains the plots of the evolution of Guessing Entropy (GE) for increasing number of traces
• multi_ge, if present, contains numerous runs of the GE computation to see the run to run variability

filenames are structured as _ where

• TRAINDEV is the training device(s) used, can be D1 or D1D2
• TARGET is either SBOX_OUT or HW_SO
• SUFFIX depends on the specific file
  • hp.json contains the hyperparamters for the specific model
  • model contains the model in hd5 format
  • if it contains ptx it means that it is referred to a network trained with trace data + plaintext information. The plaintext can be appended as is or scaled in [0,1]

Acknowledgments

This work was supported in part by project SERICS (PE00000014) under the NRRP MUR program funded by the EU – NGEU. We also thank Luca Castellazzi for his initial realization of our NN models.