Design And Implementation Of APUF Anti-modeling Attack Based On Nonlinear Processing

With the frequent occurrence of security vulnerabilities in recent years,the importance of system security,especially the physical security,has gradually received attention.The introduction of physical unclonable functions can provide a good solution for physical security,including identity authentication,key storage and so on.However,the Arbiter-based physical unclonable function(Arbiter PUF,APUF)itself has a good linear model,and it is easy to be modeled by the machine learning algorithm to its response behavior,thus losing its unclonable nature.Therefore,it is necessary to design protection structures for APUF,and enhance its anti-modeling attack capability,thereby ensuring the security of the application system.Based on APUF,this thesis considers the resource occupation and anti-modeling attack capability of the protection structure,and designs two protection structures: tree-shaped protection structure and cascade weak PUF protection structure.The tree-shaped protection structure mainly distributes and recombines the output of the APUF,and can combine a plurality of APUF models,thereby enhancing the anti-modeling attack capability of the original APUF.The cascade weak PUF structure mainly uses the weak PUF characteristic to process the original excitation,so that the relationship between original challenges and the processed challenges is a multi-mapping relationship,which disrupts the linear relationship between the original challenges and the APUF,thereby enhancing the anti-modeling attack capability.In this thesis,four APUF structures have been implemented on the Altera FPGA platform,including the original APUF,the APUF with tree-shaped protection structure,the APUF with cascade weak PUF,and the APUF with the two protection structures.Modeling attacks on four APUF structures are performed by Linear Regression,Logic Regression,SVM,and Back Propagation Neural Network algorithms.Experiments show that under the 100000 CRPs modeling attacks,the modeling accuracy of the original APUF structure is as high as 98%.Under the 150000 CRPs modeling attacks,the highest modeling accuracy of APUF with treeshaped protection structure is About 67%,the highest modeling accuracy of APUF with cascade weak PUF is about 66%,and the highest modeling accuracy of APUF with two protection structures is about 61%.Experiments show that the two protection structures have strong anti-modeling attack capabilities,while the resource overhead of the two protection structures is small,60% and 30% respectively.Finally,this thesis completes the layout implementation with two protection structures under the GF130 nm technology.