Aristotle University Of Thessaloniki (AUTH)
Experimental evaluation of photonic integrated circuits for multilayer photonic neural networks
Abstract
dc:descriptionThis dissertation presents the design and experimental validation of three modular photonic subsystems that support key signal processing functions for integrated photonic neural networks. The work focuses on the implementation of distinct building blocks for optical weighting, nonlinear activation, and on–chip amplification. These subsystems were developed for compatibility with the silicon nitride (Si3N4) platform, emphasizing energy efficiency, high–speed operation, signal fidelity, and scalable integration. The first subsystem implements a reconfigurable optical weighting mechanism based on Mach–Zehnder interferometersincorporating ferroelectric barium titanate phase shifters. These devices were integrated on alow–loss Si3N4 platform, achieving nanowatt–level static power consumption, propagation lossesbelow 1 dB/mm, and response times under 20 ns. Fabricated via a hybrid multi–material stack, the weighting units were experimentally validated in terms of electro–optic phase tuning efficiency and modulation dynamics. The results demonstrate their potential as low–power alternatives to conventional thermo–optic phase shifters commonly employed in scalable photonic networks. The second subsystem realizes a nonlinear activation function unit implemented in bothopto–electronic and opto–electro–optic configurations. It combines a semiconductor optical amplifier, a high–speed uni–traveling–carrier photodiode, and a BiCMOS transimpedance amplifier to produce reconfigurable analog responses with sigmoid–like characteristics. This subsystem was validated employing both structured step–wise and stochastic optical inputs,confirming its ability to maintain signal integrity and stable dynamic behavior. Experimental measurements demonstrated operation at bandwidths exceeding 10 GHz, with energy–per–bitvalues ranging from 24–36 pJ/bit. The third subsystem is a monolithically integrated optical amplifier based on an erbium–doped alumina waveguide, fabricated directly on a Si3N4 platform. It was developed to compensate for signal degradation and restore optical power in complex photonic chips with multiple on–chip components, organized into multi–stage or multi–layer architectures and supporting dense wavelength–division multiplexing (WDM). The amplifier demonstrated internal optical gain of 1.44 dB/cm at 1532 nm and enabled high–speeddata transmission across multiple WDM configurations, reaching aggregate throughput upto 480 Gb/s. Both static and dynamic experiments confirmed its performance in broadband, multi–channel environments. Collectively, the three subsystems provide a modular and experimentally validated foundation for multilayer photonic neural architectures. The design approach emphasizes practical integration and signal interoperability across components, addressing key requirements for inference–mode photonic computing. Although full subsystem integration into a unified neural network was not realized within the scope of this dissertation, the presented work delivers a compatible and energy–efficient set of building blocks that meet the functional demands of neuromorphic photonic systems. This contribution lays the groundwork for future co–integration, system–level validation, and the eventual deployment of programmable, high–throughput photonic neural hardware for advanced computing applications.
Degree
thesis:*- Grantor dc:publisher
- Aristotle University Of Thessaloniki (AUTH)
- Year dc:date
- 2025
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Chrysostomidis, Themistoklis
- Χρυσοστομίδης, Θεμιστοκλής
Subjects
dc:subject × 18- Φωτονικά Νευρωνικά Δίκτυα
- Φωτονικά ολοκληρωμένα κυκλώματα
- Φωτονικά στοιχεία βάρους
- Φωτονικές μη γραμμικές συναρτήσεις ενεργοποίησης
- Ενισχυτές ντοπαρισμένου κυματοδηγού με ιόντα ερβίου
- Πειραματική φωτονική
- Photonic Neural Networks
- Photonic integrated circuits
- Photonic weighting elements
- Photonic nonlinear activation function
- Erbium doped waveguide amplifiers
- Experimental photonics
- Φυσική
- Φυσικές Επιστήμες
- Οπτική
- Physical Sciences
- Natural Sciences
- Optics
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/60032