Title:Hybrid PUF: A Novel Way to Enhance the Security of Classical PUFs

Abstract:Physical unclonable functions provide a unique 'fingerprint' to a physical entity by exploiting the inherent physical randomness. With the help of quantum information theory, this paper proposes solutions to protect PUFs against machine learning-based attacks. Here, based on the querying capability, we first divide the adversaries into two classes, namely adaptive and weak adversaries. We also modify an existing security notion, universal unforgeability, to capture the power of those two classes of adversaries. We then introduce the notion of a hybrid PUF, using a classical PUF and quantum conjugate coding. This construction encodes the output of a classical PUF in non-orthogonal quantum states. We show that the indistinguishability of those states can significantly enhance the security of the classical PUFs against weak adversaries. Moreover, we show that learning the underlying classical PUF from the outputs of our HPUF construction is at least as hard as learning the classical PUF from its random noisy outputs. To prevent the adversaries from querying the PUFs adaptively, we borrow ideas from a classical lockdown technique and apply them to our hybrid PUF. We show that the hybrid PUFs, together with the lockdown technique, termed as hybrid locked PUF, can provide a secure client authentication protocol against adaptive adversaries and are implementable with the current day quantum communication technology. Moreover, we show that HLPUF allows the server to reuse the challenges for further client authentication, providing an efficient solution for running a PUF-based client authentication protocol for a longer period while maintaining a small-sized challenge-response pairs database on the server-side. Finally, we explore the lockdown technique with quantum PUF and show that the direct adaptation of the classical lockdown technique will not work with the fully quantum PUFs.