Paper 2021/296

Revisiting Fault Adversary Models - Hardware Faults in Theory and Practice

Jan Richter-Brockmann and Pascal Sasdrich and Tim Güneysu

Abstract

Physical attacks are serious threats to hardware implementations of any strong cryptographic primitive. Particularly, fault injection attack is considered as a powerful technique to successfully attack embedded cryptographic implementations since various fault injection mechanisms from simple clock glitches to more advanced techniques like laser fault injection can lead to devastating attacks, even with just a single successfully injected fault. Given these critical attack vectors, researchers in academia and industry came up with a long list of dedicated countermeasures to thwart such attacks. However, the validation of proposed countermeasures is mostly performed on custom adversary models that are often not tightly coupled with the actual physical behavior of available fault injection mechanisms and techniques and, hence, fail to model the reality accurately. Furthermore, using custom models complicates comparison between different designs and evaluation results. As a consequence, we aim to close this gap by proposing a simple, generic, and consolidated fault injection adversary model in this work that can be perfectly tailored to existing fault injection mechanisms and their physical behavior in hardware. To demonstrate the advantages of our adversary model, we apply it to a cryptographic primitive (i.e., an ASCON S-box) and evaluate it based on different attack vectors. We further show that our proposed adversary model can be used and integrated into the state-of-the-art fault verification tool VerFI. Finally, we provide a discussion on the benefits and differences of our approach compared to already existing evaluation methods and briefly discuss limitations of current available verification tools.