Regensburg 2019 – wissenschaftliches Programm

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

MA: Fachverband Magnetismus

MA 15: Magnetism Poster A

MA 15.45: Poster

Dienstag, 2. April 2019, 10:00–13:00, Poster E

An analog magnon adder for all-magnonic neurons — Thomas Brächer and •Philipp Pirro — Fachbereich Physik and Landesforschungszentrum OPTIMAS, Technische Universität Kaiserslautern, D-67663 Kaiserslautern

Neuromorphic computing is one of the most promising more-than-Moore technologies that can greatly outperform conventional CMOS computing architecture for an important set of tasks like pattern recognition, machine learning and cognitive tasks. Spintronics constitutes a highly interesting platform for neuromorphic networks since its non-volatility allows for a straight-forward integration of the data processing and memory functionality on the same level, one of the key features of brain-inspired computation. Spin-waves are a highly promising data carrier to convey information in a neural network, as they are ultra-low in energy and as they can travel over large distances without the need for spin-to-charge or charge-to-spin interconversion. In this work, we demonstrate that a leaky resonator together with a parametric amplifier can perform the action of an analogue addition over incoming spin-wave pulses. For this operation, the losses in the resonator are just compensated by the parametric amplification process. The signal integration in the spin-wave domain is similar to the signal integration in neurons in spiking neural networks and the biological original, where a nonlinearity triggers the neuron to ’fire’ and release its energy. By adjusting the gain of the amplifier, various applications for such an adder can be envisioned. Together with the intrinsic nonlinearity of the spin-wave dynamics, magnonic neurons can, thus, be envisioned.