Regensburg 2022 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
MA: Fachverband Magnetismus
MA 18: Spintronics
MA 18.8: Vortrag
Dienstag, 6. September 2022, 16:45–17:00, H47
Superparamagnetic tunnel junctions for neuromorphic computing — •Leo Schnitzspan1,2, Gerhard Jakob1,2, and Mathias Kläui1,2 — 1Institut für Physik, Johannes Gutenberg Universität Mainz — 2Max Planck Graduate Center, Mainz
Superparamagnetic tunnel junctions (SMTJ) are promising candidates for the implementation of neuromorphic computing. In a SMTJ, the magnetic free layer can switch its orientation induced by thermal activation, leading to a random two-level resistance fluctuation with relaxation times in the order of a few nanoseconds [1]. Their intrinsic stochastic behaviour and additional tunability by external magnetic fields, Spin Transfer Torques (STT) or Spin Orbit Torques (SOT) are key ingredients for low-energy artificial neurons in neural networks. Non-conventional computing, like inverse logic for integer factorization already has been demonstrated based on SMTJs[2]. Measurements of the characteristic stochastic switching behaviour are highlighted and the quality of randomness (according to NIST Statistical Test Suite) for a SMTJ as a potential true random number generator is evaluated. New possible implementation ideas of a stochastic neural network based on SMTJs are proposed and their efficiency is studied in detail.
[1] Hayakawa, K. et al., Phys. Rev. Lett. 126, 117202 (2021). [2] Borders, W. A. et al., Nature 573, 390-393 (2019).