A Single-Trace Fault Injection Attack on Hedged ML-DSA : Assessing the security of a post-quantum signature algorithm to physical attacks
Year: 2024
Type: dissertation
Abstract: Digital signature algorithms allow a signing party to prove the authenticity of digital information to a verifying party. Digital signatures are used for a variety of purposes, including to ensure the authenticity of firmware updates and as a building block of public key infrastructure (PKI). The security of current digital signature algorithms relies on the hardness of certain fundamental problems, such as the discrete logarithm or integer factorisation problems, which may be solved in polynomial time using a cryptographically relevant quantum computer. In recognition of the emerging threat posed by advances in quantum computing, recent efforts in the field of post-quantum cryptography have focused on designing replacements for existing schemes that remain secure even if cryptographically relevant quantum computers are built. The National Institute of Standards and Technology (NIST) held a competition that ultimately selected CRYSTALS-Dilithium for standardisation as a postquantum module-lattice-based digital signature algorithm under the name ML-DSA. With devices running cryptographic algorithms being increasingly physically accessible to attackers, implementations must consider the threat of physical attacks, such as side-channel analysis and fault injection. Previous work has shown that certain implementations of CRYSTALS-Dilithium are susceptible to physical attacks. As part of the standardisation, NIST introduced a hedged mode for ML-DSA, that changes the sampling of a private random seed to increase the resistance to physical attacks. The hedged mode uses the SHAKE256 extendable output function to derive the seed. This work presents a fault injection attack on hedged ML-DSA in ARM Cortex-M4 targeting SHAKE256. First, voltage glitching is performed to skip absorption of input data into the sponge during the sampling of the seed in the signing procedure, fixing it to a known constant value. We identified settings that consistently cause the necessary function to be skipped without crashing the device. After the successful fault injection, the secret key vector is derived directly from the resulting signature.
Author: Jenral, Sönke
Institution: KTH Royal Institute of Technology
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7006003693
