Right or wrong collision rate analysis without profiling: full-automatic collision fault attack

In CHES 2010, Fault Sensitivity Analysis (FSA) on Advanced Encryption Standard (AES) hardware circuit based on S-box setup-time acquired by injecting clock glitches is proposed. Soon after, some improvements of FSA were presented such as colliding timing characteristics from Moradi et al. However, the acquisition of timing characteristics requires complex procedure due to the very gradual decrease of clock glitch cycle and the heavy requirements of setup-time samples. In HOST 2015, Wang et al. presented template-based right or wrong collision rate attack to improve the efficiency of FSA, but its profiling and plaintexts-choice procedures required too many encryptions. In this paper, we fix only one specific clock glitch cycle, and take the right or wrong collision rate as a collision distinguisher. So, the whole process is a non-profiling collision attack which can be executed automatically without massive pre-computations and interactions between PC and signal generator. According to the experiments, 256 encryptions are enough for exactly deciding whether two plaintext bytes can induce an S-box collision. Compared with the existing power analysis and FSA-based attacks on AES hardware, it costs negligible time (about 6.65 s) and storage space (only one byte), and no offline computations for finding the collision between two masked S-boxes. Furthermore, our study shows that the signal-to-noise ratio in FSA-based attacks is much higher than power-based attacks.