Research Of Key Technology Based On Glitch Physical Unclonable Function

The rapid development of integrated circuits(IC)brings convenience to people,at the same time,hardware security of IC has gradually become the focus of social attention.With reference to the unique feature recognition of human,Physical Unclonable Function(PUF)takes the microscopic random variables in the integrated circuit manufacturing process as the unique feature of chips,and becomes a new type of hardware encryption primitive.Different from the traditional key-generation,PUF does not need to store key information in advance,and has the advantages of tampering prevention,physical unclonability and so on.It has great application potential in the field of information security,such as device authentication,key generation and so on.This paper focuses on the Glitch PUF,summarizes the related concepts and properties of Glitch PUF,starting from the attack technology,protection technology and implementation technology and other key technologies.The model of Glitch PUF and the corresponding circuit structure are studied deeply.The main research results are as follows:(1)Firstly,in the aspect of attack technology,the physical architecture of Glitch PUF based on FPGA is analyzed,and the input-output mapping relationship is analyzed in order to solve the problem of static output of unit circuit.Secondly,Python is used to simulate the excitation response behavior of Glitch PUF,and the single heat code is used to preprocess the collected Challenge Response Pairs(CRPs).Finally,the machine learning attack on Glitch PUF using MLP algorithm is proposed.The experimental results show that the MLP algorithm proposed in this paper has stronger attack ability than the logical regression algorithm and the random forest algorithm.Compared with the pre-processed data attack results,the processed data prediction error rate is lower,and has a better attack effect.(2)Protection technology can be divided into two kinds of protection measures,one is to improve the circuit structure of PUF to increase the complexity of the circuit;The other is to use poisoning data to actively guide the wrong classification of machine learning algorithms,so that machine learning can not be correctly modeled and reduce the accuracy of prediction.In the aspect of improving the circuit structure,by establishing the delay model of Glitch PUF and analyzing the model,it is found that the output defect of Glitch PUF circuit seriously affects the safety of PUF.According to the model parameters and analysis,the structure of Glitch PUF circuit is improved.The experimental results show that the reliability,randomness and anti-interference ability of the improved Glitch PUF circuit are enhanced.Aiming at the error classification of machine learning algorithm,this paper proposes a confusing circuit architecture based on poisoning algorithm.It is found that the prediction accuracy decreases by increasing the number of poisoning data.The more the number of unit circuits,the better the poisoning effect,and the prediction error rate of machine learning attack is maintained at about 45%,which effectively improves the anti-modeling attack ability of Glitch PUF.(3)In order to solve the problem that there is a large deviation effect in the implementation of PUF based on FPGA,this paper proposes that the PUF circuit can be implemented based on ASIC.Under the 65 nm CMOS process,the single-bit and multi-bit glitch generation module,delay adjustment module and T flip-flop sampling circuit are designed.The delay adjustment module controls the path delay of the circuit and adjusts the width of glitch.The T flip-flop sampling circuit realizes the bit output of the data.The logic function of the designed circuit is verified by EDA tools.Monte Carlo simulation results show that the uniqueness and stability of the designed circuit are 48.7% and 96.3% respectively.In order to verify the actual performance of the circuit,the designed circuit is implemented on FPGA.At different temperature sampling points,the reliability of the circuit is 93.95%.The circuit consumes less hardware resources and has better lightweight properties.For the security of the circuit,random forest and MLP algorithm are used to attack,and the prediction error rate is more than 30%.