Dissertation committee:
Δημήτρης Συβρίδης, Καθηγητής, Τμήμα Πληροφορικής και Τηλεπικοινωνιών, ΕΚΠΑ
Αγγελική Αραπογιάννη, Καθηγήτρια , Τμήμα Πληροφορικής και Τηλεπικοινωνιών, ΕΚΠΑ
Παναγιώτης Ριζομυλιώτης, Αναπληρωτής Καθηγητής, Τμήμα Πληροφορικής και Τηλεματικής, Χαροκόπειο
Ιωάννης Σταυρακάκης, Καθηγητής, Τμήμα Πληροφορικής και Τηλεπικοινωνιών, ΕΚΠΑ
Κωνσταντίνος Χατζηκοκολάκης, Αναπληρωτής Καθηγητής, Τμήμα Πληροφορικής και Τηλεπικοινωνιών, ΕΚΠΑ
Ιωάννης Παναγάκης, Αναπληρωτής Καθηγητής, Τμήμα Πληροφορικής και Τηλεπικινωνιών, ΕΚΠΑ
Έκτορας Νισταζάκης, Καθηγητής, Τμήμα Φυσικής, ΕΚΠΑ
Summary:
In the context of this doctoral dissertation titled "Photonic Technologies for Secure Data Management", the use of photonic unclonable physical functions (Physical Unclonable Functions - PUF) was explored, as central elements for the development of a new generation of systems which will allow the production of cryptographic keys "on demand" and in real time.
This investigation was conducted along 3 main axes: The first axis involved the study of materials and techniques to identify the optimal combination of those, which achieves the maximum security performance of such a system. The second focused on the examination of the reliability of these materials and techniques with regard to their stability to environmental factors and their resistance to computational attacks. Finally, the third axis focused on the comparative evaluation of their performance, which was carried out using algorithms for processing and analysing the experimental data, and through simulation of the relevant phenomena.
More specifically, this thesis initially presents the numerical model that was developed for the simulation of the physical processes that govern the majority of photonic PUFs. This numerical model, which is mainly based on the scalar theory for diffraction and the classical theory of speckle patterns, was successfully used for the design of all experimental setups that were implemented in the context of this dissertation, and for the estimation of their expected performance. Following that, the computational process that was developed for the processing of experimental data is described, through which the extraction of the required cryptographic keys from the available responses of each arrangement was achieved. This computational process includes various signal processing techniques and image hashing functions (Hash Functions), which, in combination with an error correction code (Error Correction Code - ECC) used within the frame of a fuzzy extractor, lead to random and free of errors – caused by the inevitable observation noise – binary sequences. Finally, the experimental results obtained from three different photonic PUF arrangements are presented, which were implemented in order to test three alternative excitation production methods and two different candidate materials. From these experimental results, the main security features of each arrangement were evaluated, which relate to robustness, physical unclonability, and physical unpredictability of their generated keys.
In conclusion, the product of this doctoral dissertation is a fully reliable photonic PUF, the core of which is a conventional optical diffuser. This PUF, which uses a Digital Micromirror Device (DMD) for the production of the required excitations, extracts binary sequences with experimentally proven repeatability that exceeds one month of continuous operation, and cryptographic security certified through the widely accepted NIST randomness tests. The findings of the activities that led to this implementation are estimated to effectively determine the main specifications for the design of a complete laboratory prototype of a photonic PUF. The industrialization of this prototype is expected to overlap, in terms of security performance, the already available conventional random number generators (RNG), as it can lead to portable, low-cost devices of immediate commercial value.
