Measured Traces Reduction Using SNR of Leakage for Tolerance Evaluation to Deep Learning-based Side-channel Attack

Side-channel Attacks (SCAs) are threats to decrypt cryptography by analyzing the switching noise from cryptographic ICs. Additionally, Deep Learning-based SCAs (DLSCAs) can break masking-protected cryptography, a popular countermeasure to conventional SCAs. Hence, it is required to evaluate tolerance to DL-SCAs for possible countermeasures. In this paper, we propose a method to reduce the number of leakage trace measurements for evaluation. In the proposed method, we first measure the leakage traces when a SNR of leakage is high and low. Next, we add Gaussian noise on a high SNR traces to make its SNR equal to a low SNR. This method applies correlation power analysis (CPA) to identify a low SNR, and it requires a smaller number of traces than when running an actual attack. Therefore, this method enables the reduction of measured traces. We measured leakage traces of a microcontroller implementing the masking-protected AES-128. As the results, this method predicts that 9 out of 12 secret key bytes are recovered, and this is almost consistent with the measured one.