Paper 2022/1659

A Deep Learning aided Key Recovery Framework for Large-State Block Ciphers

Abstract

In the seminal work published by Gohr in CRYPTO 2019, neural networks were successfully exploited to perform differential attacks on Speck32/64, the smallest member in the block cipher family Speck. The deep learning aided key-recovery attack by Gohr achieves considerable improvement in terms of time complexity upon the state-of-the-art result from the conventional cryptanalysis method. A further question is whether the advantage of deep learning aided attacks can be kept on large-state members of Speck and other primitives. Since there are several key points in Gohr’s key-recovery frameworks that seem not fit for large-state ciphers, this question stays open for years. This work provides an answer to this question by proposing a deep learning aided multi-stage key-recovery framework. To apply this key-recovery framework on large-state members of Speck, multiple neural distinguishers (NDs) are trained and carefully combined into groups. Employing the groups of NDs under the multi-stage key-recovery framework, practical attacks are designed and trialed. Experimental results show the effectiveness of the framework. The practical attacks are then extended into theoretical attacks that cover more rounds. To do that, multi-round classical differentials (CDs) are used together with the NDs. To find the CDs’ neutral bits to boost signals from the distinguishers, an efficient algorithm is proposed. As a result, considerable improvement in terms of both time and data complexity of differential key-recovery attacks on round-reduced Speck with the largest, i.e., the 128-bit state, is obtained. Besides, efficient differential attacks are achieved on round-reduced Speck with 96-bit and 64-bit states. Since most real-world block ciphers have a state size of no less than 64 bits, this work paves the way for performing cryptanalysis using deep learning on more block ciphers. The code is available at https://github.com/AI-Lab-Y/NAAF.