遗传算法能量分析中初始化与变异机制研究

The combination of artificial intelligence and side-channel analysis brought new research direction to cryptanalysis. In recent ten years, genetic algorithm has been introduced into side channel analysis, and a series of related research results have emerged in the world. However, the existing power analysis based on genetic algorithm had the problem of local optimization and low efficiency. This paper aimed to make a connection between local optimization and success rate, choose better initialization and mutation mechanism, and increase the efficiency of artificial-intelligence-based side-channel analysis. In this paper, we first analyzed the success reason of genetic-algorithm-based power analysis, and then discussed why the existing power analysis method of genetic algorithm fell into the local optimum. Accordingly, we introduced correlation-power-analysis-based initialization, heuristic mutation mechanism, random byte mutation, and random initialization, and then combined and compared them. Through some experiments, such as parameter selection, success rate comparison and calculation cost comparison, it is concluded that the method of correlation-power-analysis-based initialization combined with random byte mutation has the highest success rate and the lowest calculation cost. At the same time, this paper summarizes the limitations of genetic algorithm-based correlation power analysis method: not suitable for software implementation, difficult to analyze large bit-width operation, high complexity in attack protection countermeasures, high complexity in low signal-to-noise ratio. It is suggested that the value calculated in bytes or bits should not be stored in the register directly during the hardware calculation of cryptographic algorithm, so as to protect against the power analysis attack based on genetic algorithm. At last, the future work is prospected, and we think that the new method has high practicability in analyzing the block cipher algorithm implemented by non-protected hardware, and it is recommended to be applied to the actual side channel analysis and evaluation.