ANN Approach to Predict the Responses of 64 bits XOR APUF with High Accuracy

Physically unclonable functions (PUFs) are applied in hardware security applications. During the integrated circuits fabrication process, a unique response mechanism gets created. These characteristics are utilized during the PUF implementation. Delay-based arbiter PUFs are one kind of PUF that uses the time difference of two delay-line signals. The machine learning-based approaches can predict arbiter PUF (APUF) responses. Exclusively-OR (XOR) APUFs can tolerate the existing attack methods unless a large dataset size and quite a long training time are applied. This paper uses an artificial neural network (ANN)-based strategy to attack large-sized XOR APUFs. The implemented ANN-based model accomplished 99.5% accuracy in predicting the PUF responses within 10 minutes. These vulnerabilities of PUF need to be disclosed so that the PUF designers can re-evaluate the existing PUF designs and avoid potential risks for IoT devices.