Thesis defence of Amir ALI POUR (AMfoRS team): PUF based Secure Computing for Resource Constrained Cyber Physical Objects

Amir ALI POUR - AMfoRS team

Composition of the jury
Giorgio DI NATALE - Thesis director - Research director - CNRS - TIMA Grenoble (France)
Vincent BEROULLE - Co-thesis director - Full professor - Grenoble Institute of Technology (Grenoble INP) - LCIS Valence (France)
David HELY - Co-thesis director - Full professor - UGA -Grenoble INP
Alberto BOSIO, Ecole Centrale Lyon (Reviewer)
Jean-Luc DANGER, Telecom Paris ((Reviewer)
Fatemeh AFGHAH, Clemson University
Laurent FESQUET, UGA-Grenoble INP
Lilian BOSSUET, Jean Monnet Saint Etienne University

PUF based Secure Computing for Resource Constrained Cyber Physical Objects
There is a tendency for cyber-physical system designers to cultivate the physical characteristics of the system as primitives for cyber-security protocols. This is a similar to how biometric data are used for humans for identification and confidential data encryption. For cyber-physical systems, a concept known as Physically Unclonable Function (PUF) is founded for such matter. In silicon chips, PUF is an implementation of a function over the unique physical features of manufactured devices. Based on its structure, PUF can generate from a few, to very large number of digital fingerprints. The type known as strong PUF typically has the structure that can provide very large number of digital fingerprints to an extent that storing all of them is practically infeasible. In this thesis, we study how we can cultivate the potential of strong PUFs by utilizing machine learning as a medium to enroll PUF, and later recover digital fingerprints via random access. Primarily, we talk about a method for PUF enrollment using machine learning. We discuss how ML can fit as an industrial method for enrolling PUF by highlighting some of the important cost parameters. Then we talk about several optimization techniques for ML specifically designed for PUF modeling with the aim to reduce the cost of enrollment. After that, we talk about security protocol design. We specifically aim at designing a protocol that is not leaking PUF data through publicly auditable channels. We first provide a security countermeasure for an existing key generation protocol based on PUF. Then we move forward with an idea of a new protocol for authentication and key generation based on strong PUF that is specifically using ML model of PUF for PUF data recovery. Through this protocol, we discuss a novel technique for PUF data recovery which in turn requires no exchange of any offset of PUF data, making it ultimately a secure protocol against man in the middle attacks.