Fortified-Edge 5.0: Federated Learning for Secure and Reliable PUF in Authentication Systems

Physical Unclonable Functions (PUFs) are widely studied for the security of devices in the largely heterogenous Internet-of-Things ecosystem. The need for low-power and low-cost yet robust and reliable security systems is of prime importance in resource-constrained environments like smart villages. Using PUFs as a security primitive has the limitation of environmental effects that lead to bit flipping in the PUF response, the challenge in using PUFs is to overcome the bit errors without adding to the area overhead or computational overhead. This research proposes a novel bit error detection and correction algorithm implemented using Federated Learning (FL). The error detection and correction model uses the N-gram concept of Natural Language Processing (NLP). The FL model is implemented on Flower AI, the global model gets the locally trained model's parameters, updates itself, and shares the updated models with all the local models. At the edge, the use of FL for model training and updating enhances the efficiency of the authentication system that uses PUF Challenge-Response Pairs (CRPs), reduces the area overhead and power consumption, and improves the security of the PUF-based authentication system.