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

HL: Fachverband Halbleiterphysik

HL 13: Poster Session III

HL 13.23: Poster

Dienstag, 28. September 2021, 13:30–16:30, P

A Memristive Circuit for a Delay-Based Supervised Classifier — Dennis Michaelis, •Sebastian Jenderny, and Karlheinz Ochs — Ruhr University Bochum, Chair of Digital Communication Systems, Bochum, Germany

Supervised learning based on artificial neural networks is a major principle for many pattern recognition tasks. Corresponding circuit implementations are often based on implementing synaptic weight changes. In this work, we propose a different approach based on learning delays instead of synaptic weights. For this purpose, we synthesize an electrical circuit for a dynamic axon model. The resulting circuit is based on memristive Jaumann structures in combination with delay elements. We utilize this circuit to design a neural network for the supervised learning of gait patterns. Here, the learning is based on the circuit selecting delay lengths in a self-organized way, which further introduces an additional degree of freedom compared to the synaptic weight approach. A wave digital emulation verifies our approach by showing that the axonal delays associated with the trained gait patterns are successfully learned, leading to correct classification results.