Impossible Differentials Automation: Model Generation and New Techniques

In this paper, we propose new techniques for impossible differential cryptanalysis. The first one is a hybrid model for finding distinguishers on block ciphers that have both bit-oriented and word-oriented components; we apply this model to LBlock, and build an improbable differential for 18 rounds, improving over the previous 17-round results. Our second model builds impossible differential attacks for ARX ciphers automatically, including, for the first time, hash table based optimizations into the complexity evaluation of the attack. We apply this model to the HIGHT block cipher, and improve complexity of the state-of-the-art 27-round attack. Finally, we include these techniques in the cryptanalysis tool CLAASP, building the needed decryption functions automatically from the block cipher’s graph representation; this inversion technique is of independent interest to other similar libraries, such as TAGADA.