Paper 2020/1056

Automated enumeration of block cipher differentials: An optimized branch-and-bound GPU framework

Wei-Zhu Yeoh, Je Sen Teh, and Jiageng Chen

Abstract

Block ciphers are prevalent in various security protocols used daily such as TLS, OpenPGP, and SSH. Their primary purpose is the protection of user data, both in transit and at rest. One of the de facto methods to evaluate block cipher security is differential cryptanalysis. Differential cryptanalysis observes the propagation of input patterns (input differences) through the cipher to produce output patterns (output differences). This probabilistic propagation is known as a differential; the identification of which is a measure of a block cipher’s security margins. This paper introduces an optimized GPU-based branch-and-bound framework for differential search. We optimize search efficiency by parallelizing all branch-and-bound operations, completing the entire search on the GPU without communicating with the CPU. The meet-in-the-middle (MITM) approach is also adopted for further performance gains. We analyze the financial and computational costs of the proposed framework using Google Cloud VM to showcase its practicality. When optimized for performance, we can attain up to 90x speedup while saving up to 47% of the running cost as compared to a single CPU core. When optimized for cost, the proposed framework can save up to 83% of financial costs while retaining a speedup of up to 40x. As a proof of concept, the proposed framework was then applied on 128-bit TRIFLE-BC, 64-bit PRESENT, and 64-bit GIFT. Notably, we identified the best differentials for PRESENT (16 rounds) and 64-bit GIFT (13 rounds) to date, with estimated probabilities of $2^{-61.7964}$ and $2^{-60.66}$ respectively. Although the differential results for TRIFLE-BC were incremental, the proposed framework was able to construct differentials for 43 rounds that consisted of approximately 5.8x more individual trails than previous work, making it one of the most efficient approaches for larger block ciphers.