International Association
for Cryptologic Research

We present a side-channel attack on CRYSTALS-Dilithium, a post-quantum secure digital signature scheme, with two variants of post-processing. The side-channel attack exploits information leakage in the secret key unpacking procedure of the signing algorithm to recover the coefficients of the polynomials in the secret key vectors ${\bf s}_1$ and ${\bf s}_2$ by profiled deep learning-assisted power analysis. In the first variant, one half of the coefficients of ${\bf s}_1$ and ${\bf s}_2$ is recovered by power analysis and the rest is derived by solving a system of linear equations based on ${\bf t} = {\bf A}{\bf s}_1 + {\bf s}_2$, where ${\bf A}$ and ${\bf t}$ are parts of the public key. This case assumes knowledge of the least significant bits of the vector ${\bf t}$, ${\bf t}_0$. The second variant waives this requirement. However, to succeed, it needs a larger portion of ${\bf s}_1$ to be recovered by power analysis. The remainder of ${\bf s}_1$ is obtained by lattice reduction. Once the full ${\bf s}_1$ is recovered, all the other information necessary for generating valid signatures can be trivially derived from the public key. We evaluate both variants on an ARM Cortex-M4 implementation of Dilithium-2. The profiling stage (trace capture and neural network training) takes less than 10 hours. In the attack assuming that ${\bf t}_0$ is known, the probability of successfully recovering the full vector ${\bf s}_1$ from a single trace captured from a different from profiling device is non-negligible (9%). The success rate approaches 100% if multiple traces are available for the attack. Our results demonstrate the necessity of protecting the secret key of CRYSTALS-Dilithium from single-trace attacks and call for a reassessment of the role of compression of the public key vector ${\bf t}$ in the security of CRYSTALS-Dilithium implementations.