arXiv Science⌕ Search

arXiv subjects

Sinan Shu

Publications and source records attributed to Sinan Shu.

2 recordsLinked to original sources

Sign reversal of ST-FMR signals in CoFeB/Pt explained by contributions beyond spin-orbit torque

Spin-orbit torques (SOT) provide an efficient route for electrical control of magnetization and is of central interest for low-power magnetic memory applications. Spin-torque ferromagnetic resonance (ST-FMR) is a widely used technique to quantify SOT efficiencies. However, the rectified voltage measured in ST-FMR can contain contributions that are not directly related to SOTs, which affects the accuracy of the extracted torque efficiencies. This may contribute to the large variation of reported SOT efficiencies in literature. In this study, we investigate the damping-like torque efficiencies in CoFeB/Pt heterostructures as a function of Pt thickness. Conventional ST-FMR analysis yields an apparent sign reversal of the extracted efficiency at a Pt thickness of approximately 10 nm. By combining complementary measurement geometries and analysis approaches, we separate the torque-induced signal from additional voltage contributions that share the same angular dependence. Our results show that non-torque voltage contributions can produce an apparent sign reversal in conventional ST-FMR analysis, highlighting the need to distinguish SOT-induced signals from parasitic contributions and we show how to reliably determine the SOT efficiencies.

cond-mat.mtrl-sci↗

Noise-based Local Learning using Stochastic Magnetic Tunnel Junctions

Brain-inspired learning in physical hardware has enormous potential to learn fast at minimal energy expenditure. One of the characteristics of biological learning systems is their ability to learn in the presence of various noise sources. Inspired by this observation, we introduce a novel noise-based learning approach for physical systems implementing multi-layer neural networks. Simulation results show that our approach allows for effective learning whose performance approaches that of the conventional effective yet energy-costly backpropagation algorithm. Using a spintronics hardware implementation, we demonstrate experimentally that learning can be achieved in a small network composed of physical stochastic magnetic tunnel junctions. These results provide a path towards efficient learning in general physical systems which embraces rather than mitigates the noise inherent in physical devices.

cs.ET↗