The Application Of Iterative Principal Component Analysis In Differential Power Analysis

With the effective attacking on integrated circuits, Side Channel Analysisbecomes one of the important research focuses. On one hand, people keep onproposing new SCA methods, from Simple Power Analysis (SPA) andDifferentiated Power Analysis to Correlation Power Analysis and TemplateAttack. Among all these SCA methods, Differentiated Power Analysis iswidely used because of simple steps and excellent analysis performance. Onthe other hand, people make further study on the current methods and thenoise elimination in DPA is one of the key research focuses.The real power signals have always been submerged by the noise and thusthe elimination of noise became the key factor for a successful attacking. Theprevious research reveals that it can not effectively eliminate noise only withmore sampled signals, while it is not possible to acquire the sampled signalsas more as we need. Therefore, it becomes important to eliminate the noiselying in sampled signals to the best with limited sampled signal number.Several noise-elimination methods were proposed for DPA, such as themulti-bit selection function method, filter method, high-order cumulantmethod and Principal Component Analysis (PCA) method. Among all thesemethods, principal component analysis method provides the best performance.But the decomposition of autocorrelation matrix of the input signals duringthe computation brings the high computational complexity and high memoryaccupation problem, which reduces the effectiveness of the principalcomponent analysis method for real application. To solve this problem, weintroduce the Iterative Principal Component Analysis (IPCA) to DPAapplication. IPCA proves to greatly reduce the computational complexity and memory occupation. We make discussion on the computational complexityand memory occupation of the noise-elimination methods and IPCA is themost effective one.While we make discussion on the advantages of IPCA in DPA, we alsofocus on the defects of IPCA in DPA. IPCA has the stability and convergenceproblems when processing the complicated high dimensional signals, whichmatters in analysis judgement. For this problem, we add weighted processingto the normal NIC criterion to acquire our new Weighted Parameter Criterionand finally derive the new IPCA method which can guarantee the stable andconvergent iterative computation. We demonstrate the effectiveness of ourmethod in DPA with experiments and compute the relationship betweenweighted parameter matrix and the input signals. On account of therelationship, we can compute the weighted parameter matrix during thepre-processing step and use this matrix in our actual attack.To make our research more convincible, we set up the SCA test platform.With plenty experimental results, we make comparations between differentnoise-elimination methods, and the comparation indexes include thecomputational complexity, memory occupation, successful rate, peak ratioand signal-noise ratio. Our modified IPCA method proves to be the one withthe best comprehensive performance.