Leveraging Coprocessors as Noise Engines in Off-the-Shelf Microcontrollers

Securing low-cost microcontrollers against side-channel attacks is an important challenge. One core issue for this purpose is that such devices may exhibit leakages with very limited noise. As a result, standard countermeasures like shuffling or masking, which emulate or amplify noise, have limited effectiveness. In this paper, we investigate the possibility to run hardware coprocessors in parallel to a masked software implementation, in order to generate algorithmic noise. We detail the conditions for such a noise generation to be effective and show experimental evidence that it leads to security improvements compared to masked software implementations running without activated coprocessors. While masking remains expensive, the gains we show in number of traces to recover the key are systematic: an approximate factor two in our experiments, that is raised to the number of masking shares.