Design and Evaluation of Hybrid Physically Unclonable Functions for Enhanced Security

In this paper, the design and the performance evaluation of a novel partially hybridized PUF against advanced ML attacks. Arbiter and Feed-Forward PUFs fall beneath conventional PUF architectures that have been identified to be susceptible to ML-based total modelling attacks that may predict PUF responses. To address these risks, a combined PUF that integrates the positive features presented by Arbiter PUFs with other nonlinear structures like Feed-Forward PUFs and XOR PUFs is suggested. The hybrid layout makes it highly rigorous and more complex to predict by the attacker. The hybrid PUFs mapped using one million challenge and response pairs and its security is analysed. Here outcomes show that the hybrid PUF gets significantly less than 51% incorrect classification, significantly enhancing the accuracy of conventional PUFs. This work suggests a feasible solution to enable secure authentication and key generation in low power devices that are vulnerable to ML attacks.