A High-Reliability PUF Solution for Securing RFID Systems Against Machine Learning

For Radio Frequency Identification (RFID) security, reliable keys are essential. Physical Unclonable Functions (PUFs) prevent physical cloning, but they are sensitive to environmental variations and vulnerable to Machine Learning (ML) attacks. In this paper, a security system is proposed that aims to generate keys with high reliability and resistance to ML attacks. The entire system can be integrated into RFID tags. For reliable key generation, the proposed approach utilizes a two-step structure comprising a Coarse PUF and a Fine PUF, along with modified Ring Oscillator (RO) PUFs featuring varying ring counts. This design enhances resistance to machine learning (ML) attacks through challenge obfuscation. To further improve security against ML attacks, real-time power consumption is monitored using a novel analog circuit, and a hardware algorithm is developed based on the monitored power data. The proposed PUF (128-bit key generator) is implemented on an FPGA from the Xilinx family, specifically the Zynq-7 model. The robustness of the proposed PUF is evaluated through voltage and temperature variation tests. Experimental results demonstrate a Bit Error Rate (BER) of 3.42×10.5, with uniqueness and uniformity values of 49.77% and 50.27%, respectively. While a conventional PUF exhibits a vulnerability of 91.23%, the implementation of the proposed system and hardware algorithm reduces this vulnerability to 50.17%. The obtained results confirm that the proposed system offers a significantly more secure and robust solution compared to other competitors.