최기훈, 한동국, 김주환, 오충연, 박혜진,

Vol. 35, No. 1, pp. 37-46, 2월. 2025
10.13089/JKIISC.2025.35.1.37, Full Text:
Keywords: Deep Learning, Side-Channel Analysis, Template Attack, FPGA
Abstract

The performance of deep learning-based side-channel analysis is highly affected by the target intermediate value. The security analyzer should choose an intermediate value related to the side-channel information as a label. In general, the difference between the value stored in a register and the value to be stored is chosen as an intermediate value for a hardware cryptographic implementation. It requires a data flow of the data register; the efficient analysis method could be different by the hardware architecture. This paper proposes a universal deep learning-based profiling attack method for hardware cryptographic implementation. The proposed method reveals the secret key by combining several intermediate values related to the typical hardware implementations. We demonstrate the proposed method by performing the proposed and existing deep learning-based profiling attacks. The existing method, which utilizes a single intermediate value, failed to reveal the secret key for a specific implementation. Besides, the proposed method discloses each key of two implementations. This result implies that the proposed method is universally utilizable.

Statistics
Cite this article