Paper 2018/898

Differential Cryptanalysis in ARX Ciphers with specific applications to LEA

Ashutosh Dhar Dwivedi and Gautam Srivastava

Abstract

In this paper we focus on differential cryptanalysis dedicated to a particular class of cryptographic algorithms, namely ARX ciphers. We propose a new algorithm inspired by the Nested Monte-Carlo Search algorithm to find a differential path in ARX ciphers. We apply our algorithm to a round reduced variant of the block cipher LEA. We use the concept of a partial difference distribution table (pDDT) in our algorithm to reduce the search space. This methodology reduced the search space of the algorithm by using only those differentials whose probabilities are greater than or equal to pre-defined threshold. Using this concept we removed many differentials which are not valid or whose probabilities are very low. By doing this we decreased the time of finding a differential path by our nested algorithm due to a smaller search space. This partial difference distribution table also made our nested algorithm suitable for bigger block size ARX ciphers. Finding long differential characteristics is one of the hardest problems where we have seen other algorithms take many hours or days to find differential characteristics in ARX ciphers. Our algorithm finds the differential characteristics in just a few minutes with a very simple framework. We report the differential path for up to 9 rounds in LEA. To construct differential characteristics for a large number of rounds, we divide long characteristics into short ones, by constructing a large characteristic from two short characteristics. Instead of starting from the first round, we start from the middle and run experiments in forward as well as in the reverse direction. Using this method we improved our results and report the differential path for up to 12 rounds. Overall, the best property of our algorithm is that it has potential to provide state-of-the-art results but within a simpler framework as well as less time. Our algorithm is also very interesting for future aspect of research, as it could be applied to other ARX ciphers with a very easy going framework.