Modeling Attack Resistant PUFs Based on Adversarial Attack Against Machine Learning

The Physical Unclonable Function (PUF) has been proposed for the identification and authentication of devices and cryptographic key generation. A strong PUF provides an extremely large number of device-specific challenge-response pairs (CRP) which can be used for authentication. Unfortunately, the CRP mechanism is vulnerable to modeling attack, which uses machine learning (ML) algorithms to predict PUF responses. Many methods have been developed to strengthen strong PUFs; however, recent studies show that they are still vulnerable under refined ML algorithms with enhanced computing power. In this article, we propose to defend PUFs against modeling attacks from a different perspective. By modifying the CRP mechanism, a PUF can provide contradictory data such that an accurate prediction model of the PUF under attack cannot be built. Three different levels of threats are analyzed, and experimental results show that the proposed method provides an effective countermeasure against ML based modeling attacks. The proposed protection mechanism is validated using FPGA, and the results show that the performance of PUFs is also improved with the help of the proposed protection mechanism. In addition, the proposed method is compatible with hardware strengthening schemes to provide even better protection for PUFs.