A challenge-screening strategy for enhancing the stability of strong PUF based on machine learning

Physical unclonable function (PUF) is a hardware security primitive with wide application prospects. Strong PUFs have unique advantages in the field of the internet of things (IoT) device authentication due to their lightweight and exponential input and output characteristics. However, under the same conditions, the responses generated by some challenges of strong PUFs can be altered, thus affecting the success rate of device authentication. Affected by this, the threshold of matching degree in the protocol needs to be lowered to avoid the high false rejection rate (FRR). In this way, the difficulty for attackers to crack the protocol will be greatly reduced, and the security risks will be increased. To enhance the stability of strong PUFs, this study proposes a method of using machine learning (ML) to screen the stable challenges of strong PUFs. First, multiple sets of challenges are randomly generated and tested for stability. Then, the challenge and the corresponding stability are used as the input and output of the ML model, and the parameters in the model are optimized. Finally, the model can be used to calculate whether the challenge is stable and screen out the unstable to get a stable challenge dataset. The experimental results show that the stability of different strong PUFs has been improved to varying degrees after screening by the above method, indicating that the method has good applicability and can be widely used to improve the stability of strong PUFs.