When Similarities Among Devices are Taken for Granted: Another Look at Portability

The original idea of profiling implies attacking one device with a leakage model generated from an “identical copy”, but this concept cannot be always enforced. The leakage model is commonly generated with traces from an “open device”, assuming that a model which works for one device should work for another copy as well. In practice, applying a leakage model to a different copy of the same device (commonly called portability) is a hard problem to deal with, as intrinsic differences in the devices or the experimental setups used to obtain the traces cause behavioural variations which lead to an unsuccessful attack. In this paper we propose a novel similarity assessment technique that allows evaluators to quantify the differences among various copies of the same device. Moreover, we support this technique with actual experiments to show that this metric is directly related to the portability issue. Finally, we derive a method that improves the performance of template attacks.