Erlangen 2022 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
Q: Fachverband Quantenoptik und Photonik
Q 31: Photonics I
Q 31.8: Talk
Wednesday, March 16, 2022, 12:15–12:30, Q-H15
Development of a nanophotonic nonlinear unit for optical artificial neural networks — •Jan Riegelmeyer1, Alexander Eich1, Marlon Becker2, Benjamin Risse2, and Carsten Schuck1 — 1Institute of Physics, University of Münster, Germany — 2Institute of Computer Science, University of Münster, Germany
Coherent nanophotonic circuit implementations of artificial neural networks (ANNs) try to mimic signal processing in biological brains and hold great potential for fast and energy efficient computing. However, the realization of nonlinear nanophotonic components, which are utilized as activation function, remains a major challenge.
In our work, we plan on employing solution-based photoresponsive molecular systems as nonlinear building blocks of optical ANNs, for which we design and fabricate nanophotonic interfaces. We perform finite difference time domain simulations of 3D waveguide-to-free-space coupling structures that create a free-space optic link on-chip, which can be filled with photoresponsive solutions realizing tunable attenuation. The corresponding structures are produced via Direct Laser Writing in photopolymer. To confine solution in-between the couplers we fabricate micrometer-sized reservoirs made from epoxy-based photoresist. Our device realizes a new platform for optically interfacing with solution-based photoresponsive systems via multiple nanophotonic channels benefitting not only ANN-implementations but integrating novel soft matter systems into nanophotonic circuits.