Supervisors info:
Χάρης Μεσαριτάκης, Αναπληρωτής Καθηγητής, Τµήµα Μηχανικών Επικοινωνιακών και Πληροφοριακών Συστηµάτων, Πανεπιστήμιο Αιγαίου
Ιωάννης Τίγκελης, Καθηγητής, Τμήμα Φυσικής, Εθνικό και Καποδιστριακό Πανεπιστήμιο Αθηνών
Αντώνης Μπόγρης, Αναπληρωτής Καθηγητής, Τμήμα Μηχανικών Πληροφορικής και Υπολογιστών, Πανεπιστήμιο Δυτικής Αττικής
Summary:
In this thesis, the filtering characteristics as well as the non linear behavior of a passive microring resonator are investigated, in order to trace the optimal parameters to be used as a building block in a simple, fully optical photonic reservoir computing configuration.
Neuromorphic engineering aims to build low power consumption processing systems, employing basic nervous systems operations, in order to alleviate the device level and system/architectural level challenges faced by conventional computing platforms.On the other hand, choosing the encoding of information in neural networks to be done through spiking signals, promises a great improvement in terms of computational power efficiency and noise tolerance.Photonic systems, due to their inherent advantages, are very well suited for the implementation of a high speed, sufficiently complex neuromorphic spiking information system, either through a standalone photonic approach or in collaboration with electronic approaches, for a wide range of applications.
Reservoir computing serves as a very usefull neuromorphic computational paradigm.Its platform provides a natural framework for implementing learning systems on photonic hardware platforms.While, microring resonators are an excellent passive photonic component which can be used to implement the non-linear activation function in the reservoir computing system’s nodes.
The computational simulation model that we used for the microring resonator performed the necessary arithmetic calculations through waveguiding analysis in order to determine the electric fields at the resonator’s output gates, taking into consideration nonlinear effects such as two-photon-absorption and nonlinear refractive index variation. The thesis is divided in two main parts.
In the first part, the influence of the various structural parameters like the radius, the ring-bus coupling efficiency, the linear absorption coefficient of the microring as well as the operating frequency, is investigated in order to trace the optimal non-linear operating region of the resonator.Where, the optimal non-linear region is defined by two parameters: the maximum slope change of the power transmission function and the minimum required input power.The power transmission functions of each output gate of the resonator are examined, for two different microring resonator configurations (a single bus MRR and an add/drop MRR).Finally, the measurements concerning the recorded second derivative maximums of the power transmission function of each port, as well as the input power corresponding to each of them are presented.
In the second part, the influence of the various structural parameters (ring radius, ring-bus coupling coefficient, the linear absortption coefficient of the microring) on the spectral characteristics of the microring resonator is investigated.Power transmission functions of each output gate of the resonator are examined at the frequency domain, in order to observe the filtering characteristics of the MRR in each case, for the same two configurations.Finally, the measurements concening the extinction ratio and the full width half minimum of each case are presented.
Keywords:
Integrated photonic devices, neuromorphic systems, neural netwroks, machine learning, reservoir computing
