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.

With the advent of the Internet of Things, security has become indispensable. Physical unclonable functions (PUFs) are emerging as a promising alternative to classical cryptographic algorithms as it provides a lightweight and cost-effective solution for implementing a keyless security mechanism. Before adopting a PUF for real-world applications, a thorough examination of all important properties of PUF is necessary, and security and reliability are two of the important properties. The multiplexer based PUF (MPUF) was recently designed to improve upon the reliability while maintaining a similar resistance to machine learning (ML) attacks as compared with XOR PUFs of certain sizes. Recently, feed-forward neural network (NN) methods were found to be an effective tool for studying PUFs’ security against ML attacks, and a study in 2019 found that some MPUFs are insecure against NN attack methods. In this paper, we try to gain further insight into various factors of NNs that affect the predictive power of NN as PUF attack methods. We investigate a NN that employs different optimizers at different stages of the machine learning process, leading to what is called a hybrid-optimizer-enhanced NN. We implemented the new NN for ML attack of MPUFs, and experimental results have shown it converges faster than a traditional NN with a single optimizer on attacking MPUFs, and the new method also requires less training data as compared with a recent NN-based attack study of MPUFs.