Pascal STARK : Integrated photonic computing

Abstract : Massive amounts of data are generated each year, and the growth in volume accelerates quickly. Most of the data is unstructured, rendering processing and information extraction challenging. This led to the revival of artificial intelligence (AI), for example in the form of artificial neural networks. As models have been rapidly growing in size, processing with traditional digital electronic processors became increasingly slow and energy inefficient. Optical computing offers great potential to overcome these bottlenecks, as it inherently provides large bandwidth, massive parallelism and low latency. Furthermore, it might drastically improve energy efficiency, thereby reducing the environmental impact of AI workloads. To break out of niche applications cost-effective scaling and integration with existing systems is critical. Integrated Si photonics has majored as a technology over the last decades and provides the means for economical large-scale fabrication. It can be integrated with electronic systems with reasonable overhead, finally rendering the wide adaption of optical computing in data centers tangible. In this thesis we first present the heterogeneous co-integration of barium titanate and III-V technology for efficient modulation and light emission on a Si photonics platform. We show low loss light coupling between the different functional layers using inverse taper structures. In the second part of the thesis, we propose a concept for an integrated all-optical reservoir computing system and verify its performance in an extensive simulation study. Finally, we demonstrate a Si photonic convolution accelerator based on a finite impulse response filter in two-port lattice form and compare its performance to state-of-the-art electronic systems.

Jury Composition:

Daniel BRUNNER, Chargé de recherche, Université Bourgogne Franche-Comté, PhD Director

Wim BOGAERTS, Professeur, Ghent University – IMEC, Reporter

Guy VAN DER SANDE, Professeur, Vrije Universiteit Brussel, Reporter

Bert Jan OFFREIN, Manager Neuromorphic Devices and Systems Group, IBM Research Europe, Zurich, PhD Co-Director

Hervé MAILLOTTE, Directeur de recherche, Université Bourgogne Franche-Comté, Reviewer

Laurent VIVIEN, Directeur de recherche, Université Paris-Saclay, Centre de Nanosciences et de Nanotechnologies (C2N), Reviewer

Localization : Institut FEMTO 15 B, avenue des Montboucons 25000 BESANCON - salle Amphithéâtre Jean-Jacques Gagnepain