Research Of Strong Physical Unclonable Function And Attack Techology

Hardware Security is a key issue with the emergence and proliferation of embedded systems( like smart cards, RF-id tags, and mobile phones). In traditional cryptography, adversaries carry out physical attacks such as invasive, semi-invasive attacks access to the devices, can lead to key exposure and bring huge threat to hardware security. Physical Unclonable Functions(PUF) is a new, hardware-based security primitive appeared in recent years. PUF use the random differences caused by fabrication variations in the process of IC. So it can effectively avoid the disadvantage of digital keys.Firstly, the paper deeply studied the principle of PUF and analyses the background of the PUF in internationl, realized two kind of Strong PUF under the SMIC 65 nm CMOS technology, analysed and compared their Uniformity, Uniqueness and Reliability by Monte Carlo simulation. In order to verify the feasibility of PUF in practice, the paper implemented Ring Oscillator PUF in Spartan6 FPGA. At the same time, we extract a lot of Challenge-Response Pairs to attack the PUF.The paper applied Logistic Regression(LR) and Support Vector Machine(SVM) respectively to attack different struture of Arbiter PUF. We compared three parameters optimization methods — Gradient Descent Algorithm, Gradient Descent Algorithm and Resilient Backpropagation(RPROP) Algorithm in Logistic Regression. The result showed that RPROP algorithm is the best optimization method because of high prediction rate and less time. For any length of Arbiter PUF or certain complexity of XOR-Arbiter PUF, the prediction rate is about 99%. In Support Vector Machine thchnology, we mainly used different kernel function and parameters in Lib SVM to attack Arbiter PUF. The results showed that prediction rate is around 99% when attack Arbiter PUF, but prediction rate is very low when attack complex XOR-Arbiter PUF. So Logistic Regression is the optimal choices, its efficiency and prediction rate have reached a very high level.