Research On Strong PUF Defense Technology Against Machine Learning Attacks

In recent years,the frequent occurrence of Io T security incidents has aroused people's attention to the research of lightweight hardware security mechanisms in resource-constrained situations.Among them,physical unclonable function(PUF)is attracting attention as a new hardware security primitive,which uses the random physical characteristics generated during IC manufacturing to generate a unique challenge response pair(CRP)for each device and has unclonable characteristics.In many PUF designs,strong PUF is more secure due to its exponential number of CRPs,and is often used for device authentication and identity recognition,but recent studies have shown that attackers are able to clone strong PUFs through machine learning attacks while collecting a small number of CRPs,and other strong PUF structures that are effective against machine learning attacks also face the pain point of high hardware overhead.In order to solve the problems faced by the current strong PUF structure,this thesis proposes two low-overhead strong PUF defense structures that resist machine learning attacks.In addition,for the current situation that there is no unified test standard and platform for PUF security testing,this thesis designs a general machine learning-based PUF security verification system.The main work of this thesis is summarized as follows:(1)A dynamic reconstruction of chaotic PUF structure(CLC-PUF)is proposed,which is realized by a reconfigurable linear shift register(RLFSR)and a reconfigurable unit composed of LFSR,strong PUF and chaos modules.After receiving a challenge,the reconfigurable unit dynamically updates the feedback function and initial value of the RLFSR,and outputs its initial state as the response.The mapping relationship between the response and the challenge is dynamic and random,and can be used to resists modeling attacks.Through machine learning attack experiments,performance evaluation and analysis,and comparison with other strong PUF defense structures,it shows that CLC-PUF can effectively resist machine learning attacks,has excellent performance,and has the advantages of low hardware overhead.(2)A PUF structure based on a novel lightweight streaming encryption algorithm is proposed.Its core idea is to use a new logical lightweight streaming encryption algorithm to obfuscate the mapping relationship between challenge and response,the algorithm applied in the structure is complex and the hardware overhead is small enough.We prove through attack experiments that the structure can effectively improve the robustness of strong PUF against machine learning attacks,and evaluate the uniformity,uniqueness and reliability of the structure,the data show that they are all close to the ideal state.(3)Developed a machine learning-based PUF security verification system.In this prototype system,we integrated commonly used machine learning attack methods to evaluate the security of PUFs;at the same time,users can also view the evaluated PUF data in the system,which is beneficial for users to compare PUFs...