Research On Side Channel Attack And Domain Adaptive Learning In Complex Scenarios Based On Lightweight Algorithms

The ultimate goal of a side-channel attack is to recover the encryption key on the attacking device using a model trained from the analyzing device.However,most previous work is based on publicly available datasets,does not consider the portability of the environment,simplifies the experimental setup by using only a single device for analysis and attack,and typically overestimates the practical applicability of the analysis attack.In this paper,we consider three practical scenarios,namely,across devices,across key bytes,and across encryption algorithms,to evaluate the impact of changes in the environment on the efficiency of the attack.The experimental results show that the change of environment leads to a significant overestimation of the attack efficiency.To address this situation,this paper proposes a preprocessing method for EM leakage signals,which can effectively extract information points from EM leakage signals that portray similar information in these three scenarios,while other noninformation points hinder the migration learning of models in different scenarios.The results show that the portability of different devices,different key bytes,and different encryption algorithms can be achieved by these information points.In this paper,we firstly build an experimental acquisition platform,and secondly carry out a deep learning-based bypass attack scheme design,followed by portability side channel analysis in different scenarios and evaluate performance metrics such as accuracy and guessing entropy.The main contributions are as follows:First,in this paper,we use a lightweight cryptographic algorithm as the object of study,and use a deep learning method to perform a side-channel attack,using the accuracy of the model as a measure of the attack efficiency,and quantify the attack efficiency of the encryption algorithm using the results of different selection functions as intermediate values.This move aims to find the selection function with the largest leakage,and use this selection function as the attack point to successfully recover the full key.Second,the impact of the actual scenario change conditions of encryption devices,encryption algorithms,and key bytes on the attack difficulty of electromagnetic leakage is studied,and this paper investigates comparing the cracking difficulty of trajectory tracking obtained based on these different scenarios.Third,the impact of environmental changes on evaluation indexes such as accuracy and guess entropy is studied,and the data preprocessing method proposed in this paper solves the portability problem caused by domain differences in different scenarios.Fourth,The proposed method does not result in an increase in the number of training templates,and even solves the problem of repeated training of multiple templates caused by the "divide and conquer" approach.