High-resolution EM attacks against leakage-resilient PRFs Explained: And an improved construction

Achieving side-channel resistance through Leakage Resilience (LR) is highly relevant for embedded devices where requirements of other countermeasures such as e.g. high quality random numbers are hard to guarantee. The main challenge of LR lays in the initialization of a secret pseudorandom state from a long-term key and public input. Leakage-Resilient Pseudo-Random Functions (LR-PRFs) aim at solving this by bounding side-channel leakage to non-exploitable levels through frequent re-keying. Medwed et al. recently presented an improved construction at ASIACRYPT 2016 which uses “unknown-inputs” in addition to limited data complexity and correlated algorithmic noise from parallel S-boxes. However, a subsequent investigation uncovered a vulnerability to high-precision EM analysis on FPGA. In this paper, we follow up on the reasons why such attacks succeed on FPGAs. We find that in addition to the high spatial resolution, it is mainly the high temporal resolution which leads to the reduction of algorithmic noise from parallel S-boxes. While spatial resolution is less threatening for smaller technologies than the used FPGA, temporal resolution will likely remain an issue since balancing the timing behavior of signals in the nanosecond range seems infeasible today. Nonetheless, we present an improvement of the ASIACRYPT 2016 construction to effectively protect against EM attacks with such high spatial and high temporal resolution. We carefully introduce additional key entropy into the LR-PRF construction to achieve a high remaining security level even when implemented on FPGAs. With this improvement, we finally achieve side-channel secure LR-PRFs in a practical and simple way under verifiable empirical assumptions.