Abstract:

Masking is a countermeasure against differential power analysis (DPA) attacks on cryptographic devices by using random masks to randomize the leaked power of sensitive information.Template attacks (TA) against cryptographic devices with masking countermeasure by far require attackers have knowledge of masks at the profiling phase.This requirement not only increase the prerequisite of template attacking,but also lead to some sort of difference between the experimental encryption codes of the profiling device and the codes of commercial cryptographic devices,which might degrade performance in real world attacking.Blind mask template attack directly learns templates for the combination of no mask intermediate values without the need of knowing the masks of training power traces,and then uses these templates to attack masked cryptographic devices.Both traditional Gaussian distribution and neural network were adopted as the templates in experiments.Experimental results verified the feasibility of this new approach.The success rate of neural network based blind mask template attacking against masked cryptographic devices is very close to that of traditional template attacks against cryptographic devices without masking countermeasure.

Key words: side channel attack, template attack, blind mask template attack, masking countermeasure, neural network