The frequency difference between two oppositely propagating spin waves can be used to probe several interesting magnetic properties, such as the Dzyaloshinskii-Moriya interaction (DMI). Propagating spin wave spectroscopy is a technique that is very sensitive to this frequency difference. Here, we show several elements that are important to optimize devices for such a measurement. We demonstrate that for wide magnetic strips, there is a need for de-embedding. Additionally, for these wide strips, there is a large parasitic antenna-antenna coupling that obfuscates any spin wave transmission signal, which is remedied by moving to smaller strips. The conventional antenna design excites spin waves with two different wave vectors. As the magnetic layers become thinner, the resulting resonances move closer together and become very difficult to disentangle. In the last part, we therefore propose and verify an alternative antenna design that excites spin waves with only one wave vector. We suggest to use this antenna design to quantify the DMI in thin magnetic layers.
This is no longer useful for thicknesses larger than the skin depth.
We use 4 symmetric- and antisymmetric Lorentzians to fit the curves. We need a fourth curve to properly fit the background for some measurements. We believe there to be a fourth spin wave resonance—the quantization of the ks peak—at higher fields, but we are not able to reliable fit this peak.
A simple way to check this would entail removing the magnetic strip between the antennas only. However, as detailed in the supplementary material, this parasitic coupling is also mediated by the magnetic strip. Therefore, it still remains to be explicitly verified that the parasitic coupling is independent of spin waves being transmitted.
Moving the antennas closer together should remedy (at least part of) this issue.
We fitted all resonance fields, similar to the analysis in Fig. 2(b), using g = 2.17, Ms = 1.44 MA m−1, and ks,m = 2.16, 6 μm−1, and the fit finds Meff = 1.02± 0.01 MA m−1, weff = 1.2 ± 0.1 μm, and t = 10 ± 0.6 nm. These values agree well with the data from Fig. 2(b). The fit uncertainties for the quantization and secondary peak are quite high—to avoid getting non-nonsensical uncertainties we therefore normalize the fit uncertainties by χ2.
We use only 1 symmetric and antisymmetric Lorentzian to fit the spectrum. The higher order mode for the alternative design is not fitted. It is visible, but the fits consistently placed the peak at a different location.