A Fast Deep Learning Method for Security Vulnerability Study of XOR PUFs

Abstract

Physical unclonable functions (PUF) are emerging as a promising alternative to traditional cryptographic protocols for IoT authentication. XOR Arbiter PUFs (XPUFs), a group of well-studied PUFs, are found to be secure against machine learning (ML) attacks if the XOR gate is large enough, as both the number of CRPs and the computational time required for modeling n-XPUF increases fast with respect to n, the number of component arbiter PUFs. In this paper, we present a neural network-based method that can successfully attack XPUFs with significantly fewer CRPs and shorter learning time when compared with existing ML attack methods. Specifically, the experimental study in this paper shows that our new method can break the 64-bit 9-XPUF within ten minutes of learning time for all of the tested samples and runs, with magnitudes faster than the fastest existing ML attack method, which takes over 1.5 days of parallel computing time on 16 cores.

Security Evaluation of n-Choose-k-Sum PUFs Against Modeling Attacks

“…Secondly, for 4-XOR-nCk-PUF, the neuron number of hidden layer used in [35] is (2 3 , 2 4 , 2 3 ), which is smaller than that of our method (64, 32, 16). From Table . 6 we can see the prediction accuracy of our method is higher and training time is shorter, taking 4-XOR-100(10)-PUF for example, the prediction accuracy of our method is 99.01% compared with Khalid et al of 98.53% with 95 × 10 4 CPRs talking part in training.…”

“…However, many studied have reported XPUFs can be successfully attacked. For example, a logisitic regression (LR) ML attack method was published [34], its prediction accuracy can reach 99% for 9 Arbiter PUFs XORed; a neural network-based attack was reported [35], which can attack XPUFs with fewer CRPs and shorter learning time when compared with ML attack methods. In this section, Multiple Layer Perceptron (MLP) based ML method will be leveraged for l-XOR-nCk-PUF attacking, because MLP is more effective than using the LR in terms of the training time [35]- [38].…”

“…The hidden layer parameter of MLP: As it mentioned in section IV-B, the MLP in previous attacks has three hidden layers with 64, 32 and 16 neurons in each hidden layer, respectively. Khalid et al [35] present a neural networkbased method that can successfully attack XOR-PUFs with significantly fewer CRPs and shorter learning time when compared with other ML attack methods [30], [34], [38], [40], the MLP comprised of three hidden layers, with 2 l−1 neurons for the first and last layers and 2 l for the middle layer, l is the number of XOR operations. Therefore, we will compare the prediction accuracy and timing of those two method for 4 to 9-XOR-nCkPUF, the simulation results as shown in Table.…”

“…For 5-XOR-nCk-PUF, the comparison of the two ) is hidden layer parameter of our attack method, it means the neural network has three hidden layers, and the number of neurons in each hidden layer is 64, 32 and 16, respectively. (2 l−1 , 2 l , 2 l−1 ) is hidden layer parameter of the attack method proposed in [35], the neurons number of its three hidden layers is 2 l−1 , 2 l and 2 l−1 ), respectively, where refers to the times of XOR operation, e.g. when l = 6, it is (2 5 , 2 6 , 2 5 ).…”

Machine Learning and Fuzzy Logic in Electronics: Applying Intelligence in Practice

“…Mursi et al show one work from the area of crypto analysis as they present a method, based on neural networks, for attacking XOR Arbiter Physical unclonable functions (XPUF) (it is a hardware security primitive) [46]. XPUFs are used for the generation of a cryptographic key or for device authentication.…”