Practical attacks on masked hardware

Abstract

In this paper we analyze recently introduced questions for masked logic styles in general and for one such logic style called MDPL in particular. The DPA resistance of MDPL suffers significantly from a problem called early propagation, which denotes a data-dependent time of evaluation of logic cells depending on input signal-delay differences. Experiments on a prototype chip show that in case of specific MDPL modules like the analyzed AES coprocessor, early propagation does not unconditionally break the DPA resistance of MDPL. Investigations indicate that this might be due to the regular structure of the particular MDPL circuit, which is assumed to cause only relatively ″small″ signal delay differences. Furthermore, in this article it is shown that the recently proposed, so-called PDF-attack could not be turned into a successful practical attack in our environment. Finally, the recently raised question whether MDPL has special requirements in terms of the generation of random mask bits or not is discussed theoretically.