Side-Channel Analysis: Combining (Quantitative) Statistical Analysis with Engineer’s (Qualitative) Intuition

In the last decade side-channel cryptanalysis (timing attacks, power attacks, electromagnetic radiation attacks, cache-based attacks etc.) has become an important branch in cryptology. The goal of any side-channel attack is to extract key-dependent information from the leakage signal. For simplicity, we focus on power attacks in the following although our considerations could be adapted to radiation attacks. The ’classical’ approach are DPA attacks, which require only little set-up work. On the negative side their efficiency is limited since they are usually based on a simple but restrictive model (e.g. Hamming weight model), which the attacked device often does not fulfil. A further problem is how to weight the side channel information gained at several instants. Template attacks interpret power measurements as values taken on by random variables. ’Classical’ template attacks do not employ any model assumptions besides that the power signals measured at time instants t1 < t2 < . . . < tm are jointly normally distributed. In the following x denotes the relevant part of the plaintext (e.g. a byte that affects a particular S-box), z the masking value and k the relevant subkey. To estimate the density fx,z,k(·) of the m-dimensional random vector (It1(x, z, k), . . . , Itm(x, z, k)). (1)