Efficient Multi-Byte Power Analysis Architecture Focusing on Bitwise Linear Leakage

• Published in: ACM transactions on embedded computing systems 2024-11, Vol.23 (6), p.1-25, Article 102
• Main Authors: Jiang, Zijing, Ding, Qun, Wang, An
• Format: Article
• Language: English
• Subjects: Security and privacy; Side-channel analysis and countermeasures
• ISSN: 1539-9087; 1558-3465
• DOI: 10.1145/3687484

As the most commonly used side-channel analysis method, Correlation Power Analysis (CPA) usually uses the divide-and-conquer strategy to guess the single-byte key in the scenario of block cipher parallel implementation. However, this method cannot effectively use the power consumption information, resulting in a large number of power consumption traces. Therefore, genetic algorithm-based CPA is proposed, which can efficiently extract keys by multi-byte power analysis. However, genetic algorithm-based CPA tends to sacrifice computational cost to achieve a high key guessing success rate. To solve the above problems, this article focuses on bitwise linear leakage and proposes a multi-byte power analysis architecture based on the raindrop ripple algorithm. First, we propose to complete the key initialization by multiple linear regression. Second, we propose a novel swarm intelligence algorithm, the raindrop ripple algorithm, tailored for multi-byte power analysis based on the principles of “family planning” and “eugenics,” which greatly improves the probability of producing individuals with high fitness values. Third, we further enhance the possibility of the correct key being recovered by traversing the candidate key space in specific conditions. To verify the key guessing efficiency of the multi-byte power analysis architecture based on the raindrop ripple algorithm, comparative experiments are conducted on SAKURA-G with three power analysis methods based on genetic algorithms. Experimental results show that our proposal not only has the efficient power information utilization of multi-byte power analysis but also has a convergence speed comparable to or even faster than that of single-byte CPA. Its efficiency of key guessing is improved by 85.64% compared to EfficiencyGa-CPA, and its convergence speed is even faster than that of single-byte CPA at 725 power traces, and 83.87% faster than single-byte CPA at 1000 power traces, which is astonishing as a multi-byte power analysis.