SAMOP 2023 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
Q: Fachverband Quantenoptik und Photonik
Q 44: Integrated Photonics II (joint session Q/QI)
Q 44.7: Vortrag
Mittwoch, 8. März 2023, 18:30–18:45, A320
Towards Reconfigurable Lithium-Niobate-on-Insulator integrated non-von Neumann processors — •Julian Rasmus Bankwitz1,2, Seongmin Jo2, Francesco Lenzini1, and Wolfram Pernice2 — 1Institute of Physics, University of Münster, Germany — 2Kirchhoff Institute for Physics, University of Heidelberg, Germany
In recent years Artificial neural networks (ANNs) showed great advantages in a variety of fields like autonomous driving or language recognition. Fast and efficient efficient matrix-vector-multiplications (MVMs) are the building blocks of ANNs, as they represent the mathematical description of the interconnects of the ANN's neurons. With the exponentially increasing amount of data the world is generating every year, classical von-Neumann structured computers are facing their limits in computation speed and energy consumption. Overcoming those boundaries is a crucial task for modern computing, giving rise to alternative platforms like photonic integrated circuits (PICs). Lithium-Niobate-on-Insulator (LNOI) is an emerging material platform due to its broad optical bandwidth, low propagation loss and high second-order nonlinearity, enabling small footprint electro-optically reconfigurable circuits like adjustable ring resonators for non-classical light sources and Mach-Zehnder-Interferometers (MZIs) for electrically tunable optical switches. Here we demonstrate novel approaches of optical ANN matrices utilizing MZIs from LNOI for ultra-fast MVMs. From high precision fabrication engineering and modular PIC design we show high MZI extinction rations above 24 dB combined with GHz range modulation speed.