Improved Linear Distinguishers for SNOW 2.0

In this paper we present new and more accurate estimates of the biases of the linear approximation of the FSM of the stream cipher SNOW 2.0. Based on improved bias estimates we also find a new linear distinguisher with bias 2−86.9 that is significantly stronger than the previously found ones by Watanabe et al. (2003) and makes it possible to distinguish the output keystream of SNOW 2.0 of length 2 words from a truly random sequence with workload 2. This attack is also stronger than the recent distinguishing attack by Maximov and Johansson (2005). We also investigate the diffusion properties of the MixColumn transformation used in the FSM of SNOW 2.0 and present some evidence why much more efficient distinguishers may not exist.