Very thin YIG [Y₃Fe₂(FeO₄)₃] films (16 nm) were fabricated on Gd₃Ga₅O₁₂ (GGG) wafers by PLD. X-ray diffraction (XRD) and atomic force microscopy (AFM) showed that our YIG are high quality (See Fig. S1). The in-plane FMR spectrum shows only one uniform mode with linewidth about a few Oersteds, and 4πM_s is about 0.203 T from Kittel formula (See Fig. 1(a) and Fig. S2(a)). These values were very closed to one obtained by Lin et al.^[26] Py, Pt, Cu, Ta or Al with high purity (4N) was then deposited on YIG by magnetron sputtering to create an NM strip with a mask of 0.2 mm × 2.3 mm. All samples were cut into 1 mm × 3 mm for DC-voltage and microwave absorption measurements in a homemade X-band microwave absorption spectrometer. The microwave power was fixed at 32 mW.

Fig. 1.

Figure Option ViewDownloadNew Window

Fig. 1. (color online) (a) The experimental setup for the microwave absorption measurement and electrical detection of FMR in which DC-voltage along the long edge of the sample was measured. (b) The zoom-in of sample and copper-pads. The total thickness of the sample including GGG substrate is about 1 mm. (c) The in-plane rotation geometry of sample. The microwave of frequency f = 9.7 GHz propagated along the z axis, and the external field H was along the x axis. e and h are the electric and magnetic components of the microwave, respectively along the x axis and the y axis. The angle between the y axis and the long edge of sample was denoted as θ that varies as the shorted copper plate rotates in the xy-plane. (d) FMR signals of 3-nm Py thin film strip when the Al wires were connected/disconnected to the Cu-pads (with/without antenna effect). The inset is the equivalent circuit with antenna effect.

The experimental setup is shown in Fig. 1. The FMR microwave absorption and the DC-voltage were measured at frequency f = 9.7 GHz of TE₁₀ mode in the X-band cavity. The sample with size of 1 mm × 3 mm was mounted in the middle of a shorted copper plate at one end of the cavity that can rotate in the xy-plane as illustrated in Fig. 1. The angle between the y axis and the long edge of the sample is denoted as θ . Two thin rectangular copper sheets of 1.5 mm × 8 mm were symmetrically placed on the both sides (1 mm away from the sample) of the sample in the cavity as illustrated in Fig. 1. These two small copper sheets were isolated from the shorted copper plate and acted as electrical connection pads that connected to an SR-530 lock-in amplifier of Stanford Research Systems or Keithley-2182 nanovoltmeter. It should be pointed out that these two thin copper sheets did not affect the X-band microwave distribution from the angular dependence measurements of FMR for Py (see Fig. S3). Two Al wires of diameter about 30 micrometers were attached to two long-edge ends of the sample. As illustrated in Fig. 1, the electrical measurement circuit was switched on when the other ends of Al wires were connected to the two Cu-pads. The in-plane external field and AC microwave field were always orthogonal with each other in order to have a maximal precessing magnetization.

The DC-voltage from ISHE, AMR, and/or AHE of a ferromagnetic metal near the FMR has a symmetric Lorentz line-shape and an asymmetric dispersive lineshape^[27–29]

Because our computation resources do not allow us to perform reliable calculations on YIG/Pt and/or YIG/Cu systems due to the huge unit cells, we performed the calculations on Ni (111)/Cu and Ni (111)/Pt to analogize YIG/Cu and YIG/Pt systems using the same method presented in Ref. [19] for calculating MPE of Pt in Ni (111)/Pt systems.

The black circles in Fig. 2(a) are the derivative microwave absorption spectrum (dI/dH) of a pure YIG film as a function of external magnetic field H. The lineshape of the FMR derivative absorption spectrum of the pure thin YIG sample follows a standard differential Lorentzian line with FMR peak at H = 2.497 kOe (1 Oe = 79.5775 A/m), and peak width of Γ = 10 Oe which shows high quality of our thin YIG samples.^[16] The blue circles in Fig. 2(a) are the FMR derivative absorption spectra of a thin YIG(16 nm)/Pt(10 nm) bilayer strip when the electrical detection circuit is switched off (a conventional FMR configuration used by many groups). As compared with the FMR of the pure YIG film, the FMR derivative microwave absorption spectrum of thin YIG/Pt bilayer sample appears to shift to a lower field with a seemingly broaden peak width that was observed in previous studies^[5,16,31] and was used as an essential evidence of spin pumping from YIG. A careful examination showed that the above observed FMR absorption curves of switch-off circuit are better described by two independent FMR signals. One of them with a relative amplitude of A₁ = 60.5% was from the free YIG because it has the same peak position and peak width of H₁ = 2.497 kOe and Γ₁ = 10 Oe as the free YIG. The other with peak position H₂ = 2.488 kOe, peak width Γ₂ = 12 Oe, and relative amplitude A₂ = 39.5% was naturally attributed to the YIG covered by Pt. Because Pt modifies magnetic properties of YIG,^[31] the peak position and peak width of the second FMR signal differ slightly from those of the free YIG. In fact, our FMR results on pure YIG and YIG(16 nm)/Pt(10 nm) films agree completely with those observed by many other groups.^[5,16,31]

Fig. 2.

Figure Option ViewDownloadNew Window

Fig. 2. (color online) (a) The FMR derivative absorption spectra of free YIG and YIG/Pt bilayer strip with θ = 0° (θ is the angle between microwave field and the sample long edges). The lower panel is the corresponding DC-voltage (bottom) spectra of YIG/Pt bilayer schematically illustrated on the right. The left inset is the schematic diagram of the electrical. The right inset shows the sample structure YIG/Pt stripe. (b) The fit of FMR spectrum of YIG/Pt strip obtained with antenna effect by three FMR signals respectively for free YIG, YIG covered with Pt, and magnetized Pt surface in contact with YIG. The DC signal agrees with the assertion of spin pumping from the magnetized Pt surface. (c) The FMR derivative microwave absorption (upper) and DC-voltage (lower) spectra of fully covered YIG/Pt bilayer (illustrated in the lower left) with θ = 170° (field was reversed so that the voltage polarity is changed). The inset shows the YIG sample fully covered by Pt. (d) The FMR spectrum of fully covered YIG/Pt bilayer with the antenna effect is best fitted by two FMR signals from YIG covered with Pt and magnetized Pt surface. The corresponding DC-voltage signal was from the magnetized Pt surface. The unit a.u. is short for arbitrary unit.

The green circles in Fig. 2(a) are the FMR derivative absorption spectra of the thin YIG(16 nm)/Pt(10 nm) bilayer strip when the electrical detection circuit is switched on (see the left inset of Fig. 1(a) and the methods below). Strikingly, as compared with the FMR spectra when the circuit is switched off, the microwave absorption signals were greatly amplified. It is worthy to note that the FMR absorption curves of switch-on circuit were best fitted by three independent FMR signals as shown in Fig. 2(b). Among the three signals in Fig. 2(b), two signals (with A₁ = 54.5% and A₂ = 36.4%) are exactly those of free YIG and Pt-covered YIG, and the third signal of H₃ = 2.477 kOe, Γ₃ = 14 Oe, and A₃ = 9.1% came from the amplification of a very weak signal (hidden in the blue circles) originated from the MPE-induced magnetized Pt surface that was in contact with YIG. It agrees with those reports^[18–24] that the paramagnetic Pt is magnetized at the interface when it directly contacts with a ferromagnetic material. Furthermore, the corresponding DC-voltage detected in Pt was from the magnetized Pt surface since their peak positions and peak widths match exactly with each other as shown in Fig. 2(b).

To substantiate this interpretation, we fabricated also thin YIG/Pt bilayer samples in which YIG was fully covered by Pt layer. As shown in Fig. 2(c), the blue and green circles are the FMR absorption signals when the electrical measurement circuit was switched on (green) and off (blue). As expected, the signal from the free YIG was absent. In Fig. 2(d), the FMR absorption curves of switch-on circuit now consist of two signals respectively from the Pt-covered YIG and the magnetized Pt surface. Again, the DC-voltage relates to the signal of the magnetized Pt surface since their peak positions and widths match exactly with each other, and cannot be from the spin pumping of thin YIG.

The above conclusion could also be reached from the change of the shape of voltage-H curves as angle θ between AC-magnetic field and sample long edge varies. The DC-voltage originated from the spin pumping of YIG should follow a Lorentzian lineshape since the FMR absorption is described by the Lorentzian function.^[31] Thus the DC-voltage lineshape would be symmetric about its peak for any angle θ if the spin pumping was from YIG. However, as plotted in Fig. 3(a), it is clear that most spectra consist of a superposition of a Lorentz- and a dispersive-type resonance lineshape.^[27,28]

Fig. 3.

	Figure Option ViewDownloadNew Window
	Fig. 3. (color online) (a) The DC-voltage spectra of a YIG(16 nm)/Pt(10 nm) strip at various angle θ. The symmetrical (b) and asymmetrical (c) components of DC-voltages defined in Eq. (1) were extracted. The solid lines are the best fits of Eq. (2) with . The inset illustrates experimental setup and angle θ.

For a given θ, the DC-voltage curve was fitted to Eq. (1) so that both symmetric and asymmetric components of the DC-voltages U_sym and U_asy were obtained. Their angle-dependences were plotted in Fig. 3(b) that fit well with theoretical prediction of Eq. (2) (see the methods below). Voltages (for symmetric component) and (for asymmetric component) due to the spin rectification are and . The voltage U_sp from the ISHE due to spin pumping is U_sp = 1.02 µV. Thus it shows that substantial amount of the DC-voltage came from the AMR and AHE of a ferromagnetic metal that resulted in an asymmetric line-shape, and the natural conclusion is that the Pt surface in contact with YIG was magnetized and generated a spin rectification voltage.^[29] The DC-voltage signal from the spin pumping is larger than that from spin rectification, and this magnitude cannot be from spin current pumped from the uniform precession of the YIG magnetization as shown in a simple estimate in the supplementary materials (see Appendix A). These results further confirmed that the electrical signal is from the magnetized Pt surface in our thin YIG/Pt system.

Figure 4 is the H-dependence of the derivative microwave absorption with the switch-on circuit and the DC voltage of a typical thin YIG/Pt strip sample for various frequencies at θ = 0°. The DC voltages (Fig. 4(a)) have a symmetric Lorentzian shape while the microwave absorption has multipeaks (Fig. 4(b)) due to different FMR sources. This implies that only one FMR source pumped spin current into Pt and generated the DC-voltage. Clearly, the peak position and width of the DC-voltage match well with those of the tiny FMR signal. In order to show that this peak is not from YIG, the saturation magnetization and the gyromagnetic ratio associated with the observed electrical signal were extracted from our experimental data. Figure 4(c) shows the frequency dependence of the peak field of the DC-voltage that fits well with the Kittel formula,^[32,33] f = (γ/2π)(H_res(H_res + 4πM_s))^1/2, with the gyromagnetic ratio γ = gµ_B/h = 1.738 × 10¹¹ rad·T⁻¹ ·s⁻¹, slightly away from the known YIG value of 1.785 × 10¹¹ rad · T⁻¹ · s⁻¹, and the saturation magnetization M_s = 0.248 T. It gave a Lander factor g = 2.08 for magnetized Pt surface. These values are different from M_s = 0.200 T for typical thin YIG grown by pulsed laser deposition (PLD) method due to the deficiency of Fe ions.^[41] Our experimental value of M_s for MPE of Pt layer in YIG/Pt system, slightly larger probably because the induced moment of Pt in YIG/Pt is slightly higher than that in Ni/Pt since Fe³⁺ in YIG has 5 µ_B, is comparable with the reported M_s = 0.224 T in Ni/Pt by XMCD measurement.^[19] Thus, this result further supports our assertion that the precessing magnetization of thin YIG did not pump spin current to Pt layer, contrary to the popular belief.^{[5,14–16,34]}

Fig. 4.

	Figure Option ViewDownloadNew Window
	Fig. 4. (color online) The H-dependence of the DC-voltage (a) and FMR spectra (b) of YIG/Pt stripe line at θ = 0° and for various frequencies with antenna effect. (c) The frequency dependence of peak position of DC-voltages. The solid line is the fit to the Kittel formula.

To further substantiate our claim that the DC-voltage in thin YIG/Pt bilayer is due to the spin pumping from the MPE-induced magnetized Pt layer, a 5-nm thick Cu was inserted between YIG and Pt (Ta) so that Pt (Ta) surfaces were not in contact with YIG and no MPE is possible for Pt (Ta). The upper panel of Fig. 5 is the typical FMR derivative microwave absorption spectrum of one of our thin YIG/Cu/Pt samples. The blue circles are the results when the electrical detection circuit was switched off while the green circles are those when the circuit was switched on. The signal from the magnetized Pt surface was obviously absent, and was replaced by a new FMR signal at an even lower field of H = 2.46 kOe, far below YIG resonance peak, and with a peak width of Γ = 12 Oe. Although this new signal was seen in a 50-times enlarged figure as shown by the black circles in the top panel, it was extremely weak when the electrical detection circuit was switched off. Similar to thin YIG/Pt bilayer samples, an extra signal can be clearly observed when the electrical detection circuit was switched on. As expected this time, the DC voltage of thin YIG/Cu/Pt and/or YIG/Cu/Ta (middle and lower panels of Fig. 5) was observed at H = 2.46 kOe, exactly corresponding to the new FMR signal. The peak widths of the FMR and DC-voltage signals matched again with each other as shown in the middle (for YIG/Cu/Pt) and lower (for YIG/Cu/Ta) panels. In contrast, there were no DC-voltage signals at the YIG resonant fields, confirming that the DC-voltage was not due to the spin pumping of YIG, but due to that of the magnetized Cu surface (in contact with YIG).^[35–38] Moreover, the signs of the DC-voltages of thin YIG/Cu/Pt and YIG/Cu/Ta samples are opposite due to the opposite sign of spin Hall angle for Pt (positive) and Ta (negative).^[16]

Fig. 5.

Figure Option ViewDownloadNew Window

Fig. 5. (color online) The FMR spectra of a YIG/Cu bilayer sample (upper panel) and DC-voltage spectra of YIG/Cu/Pt (middle panel) and YIG/Cu/Ta (lower panel) with θ = 0°. In the upper panel, the green (blue) circles are the FMR derivative microwave absorption spectra when the electrical detection circuit is switched on (off). A weak signal, as shown by the black circles that is the zoom-in (50 times enlarged) of the blue circles inside the red rectangle, was amplified when the electrical circuit was switched on. The DC-voltage in Pt (middle panel) and Ta (lower panel) can be fitted well by the Lorentzian function (solid lines).

Again, above experiments do not support the general belief that precessing YIG magnetization at FMR pumps spin current into Pt or Ta in thin YIG/Pt, YIG/Cu/Pt, and YIG/Cu/Ta systems. Our results are consistent with the assertion that it was the magnetized NM surface pumping spin current into the Pt or Ta layer. The clear evidences include 1) DC-voltage peaks were far from the FMR peaks of the YIG, but were exactly overlapped with the FMR peak of MPE-induced magnetized NM surface; 2) the angle-dependence of the DC-voltage lineshape that shows big contribution from AMR and AHE of a magnetized metal. Furthermore, the MPE of both Pt and Cu were confirmed by our first-principle calculations (see the Methods). It was found that a few Cu atomic layers adjacent to Ni was magnetized with average moment about −0.02 µ_B/atom, which was only about 1/5 of the average moment of Pt (0.11 µ_B/atom) in Ni (111)/Pt system.^[19] If one assumes that Cu/Ni and Cu/YIG (Pt/Ni and Pt/YIG) have the similar MPE, then the precession of this small moment can pump spin current into Pt or Ta layer. We can naturally interpret our observed DC-voltage as the ISHE. This smaller magnetized Cu moment explains the much smaller voltage than that in thin YIG/Pt system as shown in Fig. 2(a). The small negative Cu magnetic moment is also consistent with lower resonance field for the magnetized Cu due to the negative exchange field at FMR.^[32]

To further verify the above assertion, we did a controlled experiment with thin YIG(16 nm)/Al(5 nm)/Pt(10 nm) samples. It is clear that the thin YIG(16 nm)/Al(5 nm)/Pt(10 nm) samples only show one FMR peak as shown in Fig. 6, which is almost the same as that of pure YIG in Fig. 2(a). Moreover, the derivative microwave absorption spectrum of a typical thin YIG/Al/Pt samples in Fig. 6 does not change whether the electrical detection circuit was switched on or off, in contrast to that for YIG/Pt or YIG/Cu/Pt(Ta) systems. Especially, no detectable DC-voltage was observed as shown in the bottom panel of Fig. 6.

Fig. 6.

	Figure Option ViewDownloadNew Window
	Fig. 6. (color online) Top: The FMR derivative microwave absorption spectra of a YIG/Al(5 nm)/Pt(10 nm) sample when the electrical detection circuit was switched on (red) and off (blue). No MPE-induced magnetized Al was observed. Bottom: No DC-voltage signal in Pt was observed.

According to our first-principle calculations on Ni/Al system, there is no significant spin polarization in Al in YIG/Al/Pt system, and correspondingly there is no additional FMR peak from the magnetized Al layer. No observable DC-voltage in Fig. 6 indicates that the FMR of the YIG layer could be hard to pump a detectable spin current into Al/Pt layers in YIG/Al/Pt system efficiently. Therefore, the controlled experiments with YIG(16 nm)/Al(5 nm)/Pt(10 nm) further reveal that spin current in NM layer(s) was not from YIG in thin YIG/NM₁ bilayer or YIG/NM₁/NM₂ multilayer samples, but from magnetized NM1 surface (in contact with YIG).