An Improved Empirical Mode Decomposition for Power Analysis Attack

Correlation power analysis(CPA) has become a successful attack method about crypto-graphic hardware to recover the secret keys. However, the noise influence caused by the random process interrupts(RPIs) becomes an important factor of the power analysis attack efficiency, which will cost more traces or attack time. To address the issue, an improved method about empirical mode decomposition(EMD) was proposed. Instead of restructuring the decomposed signals of intrinsic mode functions(IMFs), we extract a certain intrinsic mode function(IMF) as new feature signal for CPA attack. Meantime, a new attack assessment is proposed to compare the attack effectiveness of different methods. The experiment shows that our method has more excellent performance on CPA than others. The first and the second IMF can be chosen as two optimal feature signals in CPA. In the new method, the signals of the first IMF increase peak visibility by 64% than those of the tradition EMD method in the situation of non-noise. On the condition of different noise interference, the orders of attack efficiencies are also same. With external noise interference, the attack effect of the first IMF based on noise with 15dB is the best.