基于多种群遗传算法的相关能量分析中个体结合策略探究

Correlation Power Analysis (CPA) is one of the classic and effective methods in side-channel attacks. It recovers the key based on the correlation coefficient between assumed power consumption and actual power consumption. In the scenario of parallel implementation of cryptographic algorithms, CPA’s “divide and conquer” idea of restoring keys will result in a low signal-to-noise ratio, thus, effective information cannot be fully utilized, which will greatly reduce the attack efficiency. The CPA based on simple genetic algorithm can take advantage of the heuristic search feature of genetic algorithm to make full use of effective information and improve attack efficiency. However, genetic algorithms have an inherent shortcoming, i.e., they are easy to converge prematurely, especially in scenarios with a large number of bigger S-boxes. CPA based on multi-population genetic algorithm retains the optimal individual when a single population fails to recover the key, and continues the evolution of a new single population, and the obtained optimal individual and the previously retained optimal individual are “combined” to obtain a better individual, which alleviates the problem of premature convergence to a certain extent. The “primitive method” in this paper is a synonym for this method. This paper explores the combination of outstanding individuals obtained by the evolution of multiple populations, and introduces three new strategies for combining outstanding individuals with multiple populations, namely: group competition, voting method and secondary evolution. The group competition puts two outstanding individuals into a group and then “combines”. The voting method uses fitness as the weight to vote, so that individuals with high fitness have greater decision-making power. The secondary evolution retains the optimal individuals obtained at the end of the evolution of multiple single-populations to form the initial population, and evolves again in a steady-state genetic method. Taking the AES-128 algorithm as an example, the three methods are compared with the original method in terms of success rate and computational cost through simulation experiments with different noise standard deviation and real experiments, and it is found that the secondary evolution gets the best results. In the simulation experiments with the noise standard deviation of 3, the key recovery success rate of the secondary evolution reaches 91% when there are 190 traces, with the computational cost of 0.63 × 10⁶, while the success rate of the original method is only 60%, with the computational cost of 1.60