Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

AKPIK: Arbeitskreis Physik, moderne Informationstechnologie und Künstliche Intelligenz

AKPIK 3: Research with AI: Hardware, Software, Tools

AKPIK 3.1: Hauptvortrag

Dienstag, 18. März 2025, 11:00–11:30, H5

3D Integration Towards Autonomous Optical Neural Networks — •Adrià Grabulosa, Anas Skalli, and Daniel Brunner — Institute Femto-ST, Université Marie et Louis Pasteur, CNRS UMR6174, 15B Avenue des Montboucons, Besançon, France.

In the last decades, modern electronic integrated circuits has reached a fundamental limit at 2 nm feature sizes. At the same time, emerging computing concepts such as neural networks (NNs), which are already playing a major role in modern societies, further amplifying this challenge. Adopting the third dimension is a promising strategy for achieving scalability of connections over the microchip's dimensions. Here, based on additive one- (OPP) and two-photon polymerization (TPP) processes and combined with direct-laser writing (DLW) settings, a complete toolbox comprising photonic waveguides, splitters and bends towards three-dimensional (3D) photonic integration is presented. The concept CMOS is validated by printing 3D photonic waveguides onto semiconductor (GaAs) substrates and silicon-on-insulator (SOI) platforms. Finally, we show first steps towards the 3D integration of a fully autonomous NN using spatially multiplexed modes of an injection locked large area vertical cavity surface emitting laser (LA-VCSEL) capable of performing NN tasks with >98% accuracy, fully realized in hardware using off-the-shelf, low energy consumption components. Overall, these building blocks are highly appealing for realizing fully-parallel and efficient communication throughout a densely-connected network, which are pivotal concepts for future NN computing topologies.

Keywords: 3D photonic integration; Artificial Intelligence; Optical neural networks; Additive fabrication