Deep-Learning-Based Security Evaluation on Authentication Systems Using Arbiter PUF and Its Variants
- R. Yashiro, T. Machida, M. Iwamoto, and K. Sakiyama
- IWSEC 2016
- LNCS 9836
Fake integrated circuit (IC) chips are in circulation on the market, which is considered a serious threat in the era of the Internet of Things (IoTs). A physically unclonable function (PUF) is expected to be a fundamental technique to separate the fake IC chips from genuine ones. Recently, the arbiter PUF (APUF) and its variants are intensively researched aiming at using for a secure authentication system. However, vulnerability of APUFs against machine-learning attacks was reported. Upon the situation, the double arbiter PUF (DAPUF), which has a tolerance against support vector machine (SVM)-based machine-learning attacks, was proposed as another variant of APUF in 2014. In this paper, we perform a security evaluation for authentication systems using APUF and its variants against Deep-learning (DL)-based attacks. DL has attracted attention as a machine-learning method that produces better results than SVM in various research fields. Based on the experimental results, we show that these DAPUFs could be used as a core primitive in a secure authentication system if setting an appropriate threshold to distinguish a legitimate IC tags from fake ones.