Berlin 2024 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
MA: Fachverband Magnetismus
MA 20: Poster I
MA 20.45: Poster
Tuesday, March 19, 2024, 16:30–19:00, Poster A
Superparamagnetic tunnel junctions for neuromorphic computing — •Jonas Köhler1, Leo Schnitzspan1,2, Fabian Kammerbauer1, Gerhard Jakob1,2, and Mathias Kläui1,2 — 1Institute for Physics, Johannes Gutenberg University, 55122 Mainz — 2Max-Planck Graduate Center Mainz, 55122 Mainz
Superparamagnetic tunnel junctions (SMTJs) are considered promising candidates for building blocks in neuromorphic computing. Due to thermal excitations, the ferromagnetic free layer can switch its magnetization orientation at the nanosecond timescale. The stochastic behavior of the SMTJs allows them to produce true random numbers with encryption-quality randomness [1]. The state probability and the dwell times may be tuned by an external magnetic field or by an applied current via spin transfer torque. When multiple SMTJs are electrically connected, the spin transfer torque leads to an electrical coupling between the MTJs, with the consequence that their stochastic switching becomes correlated [2].
These properties can be used in neural networks, where MTJs can introduce the noise for noise-based learning methods. In comparison to classical approaches, computing implementations using SMTJs can be more energy-efficient.
[1] L. Schnitzspan, et.al., Phys. Rev. Appl. 20, 024002 (2023)
[2] L. Schnitzspan et. al., arXiv:2307.15165 (2023) (in press Appl. Phys. Lett. 123 (2023))
Keywords: Magnetic Tunnel Junction; Superparamagnetism; Spin-transfer torque; true randomness; Electrical coupling