A Side-Channel Attack on Masked and Shuffled Implementations of M-LWE and M-LWR Cryptography : A case study of Kyber and Saber

Abstract: In response to the threat of a future, large-scale, quantum computer, the American National Institute of Standards and Technology (NIST) initiated a competition for designs of quantum-resistant cryptographic primitives. In 2022, the lattice-based Module-Learning With Errors (M-LWE) scheme Kyber emerged as the winner to be standardized. The standardization procedure and development of secure implementations call for thorough evaluation and research. One of the main threats to implementations of cryptographic algorithms today is Side-Channel Analysis (SCA), which is the topic of this thesis. Previous work has presented successful power-based attacks on implementations of lattice cryptography protected by masking and even masking combined with shuffling. Shuffling makes SCA harder as the order of independent instructions is randomized, reducing the correlation between operations and power consumption. This randomization is commonly implemented by shuffling the order of the indexes used to iterate over a loop, using the modern Fisher-Yates algorithm. This work describes a new attack that defeats the shuffling countermeasure by first attacking the generation of the index permutation itself. The attack first recovers the positions of the first and last indexes, 0 and 255, and then rotates the encrypted messages using a ciphertext malleability applicable to many ring-based LWE schemes to shift two bits into the known positions from which they can be recovered. This procedure is repeated to recover full messages in 128 rotations. The attack is tested and evaluated on masked and shuffled implementations of Kyber as well as Saber, another similar finalist of the NIST competition which is based on the Module-Learning With Rounding (M-LWR) problem. Compared to the previous attack on masked and shuffled Saber, which required 61,680 traces, the 4,608 needed for this attack demonstrates a 13-fold improvement.