Down the Rabbit Hole: Revisiting the Shrinking Method

The paper is about methodology to detect and demonstrate impossible differentials in a block cipher. We were inspired by the shrinking technique proposed by Biham et al. [2,3] in 1999 which recovered properties of scalable block cipher structures from numerical search on scaled down variants. Attempt to bind all concepts and techniques of impossible differentials together reveals a view of the search for impossible differentials that can benefit from the computational power of a computer. We demonstrate on generalized Feistel networks with internal permutations an additional clustering layer on top of shrinking which let us merge numerical data into relevant human-readable information to be used in an actual proof. After that, we show how initial analysis of scaled down TEA-like schemes leaks the relevant part of the design and the length and ends of the impossible differentials. We use that initial profiling to numerically discover 4 15-round impossible differentials (beating the current 13-round) and thousands of shorter ones.