Attacks in Reality: the Limits of Concurrent Error Detection Codes Against Laser Fault Injection

Abstract

As a prominent attack approach against the security modules of integrated circuits, fault injection attacks (FIA) are able to breach thecryptographic primitives by analyzing the intentionally induced computation errors by adversaries. Parity-based Concurrent Error Detection (CED) techniques are often deployed as a countermeasure, owing to their low-overhead. Advanced linear and non-linear randomized encodings can be employed for constructing varying CED schemes. In this paper, we first evaluate the detection capability of linear parity-protected ciphers implemented in commercial FPGA, using laser fault injection (LFI) technique. A single-bit linear parity scheme is shown to be ineffective for error detection, since the LFI can typically flip multiple bits that are close to each other. On the other hand, a linear randomized parity scheme, with multiple bits parity, shows higher detection rates. Further, we study existing (randomized) non-linear encoding-based CED. With practical fault distributions on PRESENT cipher, non-linear randomized codes are extensively tested against fault injection. Although, known to have better theoretical detection bounds, non-linear encodings do not provide much improvements over simple randomized linear codes.