Achieving Error-Free Lightweight Authentication With DRAM-Based Physical Unclonable Functions
Nico Mexis, Nikolaos Athanasios Anagnostopoulos, Stefan Katzenbeisser, Elif Bilge Kavun, Fatemeh Tehranipoor, Tolga Arul · IEEE Transactions on Circuits and Systems I Regular Papers · 2024
In this article, we introduce a novel approach to achieving lightweight device authentication through the use of a low-complexity Convolutional Neural Network (CNN). In our work, we improve the False Authentication Rate (FAR) by transforming the standard CNN into a Bayesian CNN (BCNN or BNN). This transformation enables the use of probabilistic modelling techniques, increasing the model’s robustness and its confidence in authentication decisions. Regardless of the model used, clients authenticate with a retention-based Dynamic Random Access Memory Physical Unclonable Function (DRAM PUF) response. Our approach integrates the low computational complexity of the CNN with the intrinsic security characteristics of the DRAM PUF, offering a robust solution for lightweight and secure device authentication.