Skip to Main content Skip to Navigation
Conference papers

Unprofiled expectation-maximization attack

Julien Béguinot 1, 2, 3 Wei Cheng 1, 2, 3 Sylvain Guilley 3, 2, 4, 5 Olivier Rioul 3, 2, 1 
1 COMNUM - Communications Numériques
LTCI - Laboratoire Traitement et Communication de l'Information
4 SSH - Secure and Safe Hardware
LTCI - Laboratoire Traitement et Communication de l'Information
Abstract : Block ciphers are often protected against side-channel attacks by masking. When traces are available for each key hypothesis, the attacker usually resorts to templates attacks with a profiling phase. Lemke-Rust & Paar suggested at CHES2007 a way to profile templates for Gaussian mixture models, with the use of the well-known Expectation-Maximization (EM) algorithm. In this work, we present a new attack, “unprofiled-EM” (U-EM) that does not use the knowledge of the masks nor requires a profiling phase. This is done by “on-the-fly” regression of the coefficients of a stochastic model using the EM algorithm. Compared to previous methods, it is easy to implement, computa- tionally tractable and efficient in terms of success rate or guessing entropy. We discuss several variations of U-EM and compare their performances on simula- tions and on real DPA contest traces. The best attack scenario depends on the trade-off between measurement noise and epistemic noise.
Document type :
Conference papers
Complete list of metadata
Contributor : Olivier Rioul Connect in order to contact the contributor
Submitted on : Saturday, July 9, 2022 - 7:51:53 AM
Last modification on : Saturday, August 13, 2022 - 3:10:07 AM


  • HAL Id : hal-03718705, version 1


Julien Béguinot, Wei Cheng, Sylvain Guilley, Olivier Rioul. Unprofiled expectation-maximization attack. 18th International Workshop on Cryptographic Architectures Embedded in Logic Devices (CryptArchi 2022), May 2022, Porquerolles, France. ⟨hal-03718705⟩



Record views