Enhancement of spin Seebeck effect in Fe₃O₄/Pt thin films with α-Fe nanodroplets

The spin Seebeck effect (SSE), first observed in 2008,¹ is defined by spin accumulation in a magnetic material as a result of a temperature gradient and has been observed in a wide variety of materials, notably yttrium iron garnet (YIG) as an insulator with low damping as well as magnetite (Fe₃O₄), a ferrimagnetic semiconductor at room temperature that becomes insulating below the Verwey transition at 120 K.² The possible applications of the spin Seebeck and other magnetothermal effects (such as the anomalous Nernst effect, ANE) include their use in heat flux sensors,³ thermoelectrics,^4,5 and as a spin current source in spintronics devices. However, for spin Seebeck devices, we require utilization of this spin accumulation by spin pumping into an adjacent paramagnetic layer, such as Pt, where it is converted to a measurable voltage via the inverse spin Hall effect (ISHE).⁶ As such, the efficiency of this spin injection has a significant role to play in the application of any spin Seebeck based devices.

There have been several studies that have demonstrated that the nature of the interface between magnetic and paramagnetic layers has an impact on the efficiency of spin injection into the paramagnetic layer.^7–12 For example, for YIG/Pt bilayers, a strong dependence of the spin pumping efficiency was observed with YIG surface treatment prior to Pt deposition (with spin mixing conductance, g_r, varying from 0.02 × 10¹⁹ to 3.43 × 10¹⁹ m⁻²).⁷ With regard to interfacial roughness, there are conflicting reports of improvement in spin injection: e.g., a linearly increasing relationship between g_r and roughness of Fe₃O₄/Pt bilayers,¹² or an order of magnitude decrease in ISHE measured in YIG/Pt bilayers with increasing roughness.⁸ As a separate route to improve spin injection, the insertion of ultrathin magnetic layers of NiO, Mn, IrMn, or Ru was shown to enhance the measured spin Seebeck voltage of YIG/Pt bilayers by up to 100%; however, this was partly due to the modification of the resistance.¹⁰ Theoretically, first principle calculations modeling the different types of disorder at a Permalloy/platinum interface have indicated that introducing a foreign magnetic metal layer into the Pt could result in an increase in the spin injection efficiency,¹³ or that a surface-roughness-induced increase in the spin Hall effect (thus increased efficiency of spin current conversion to charge current) could occur due to side jump scattering.¹⁴ In this Letter, we will show an enhancement of the spin Seebeck coefficient measured for Fe₃O₄/Pt bilayers with the inclusion of α-Fe nanodroplets.

Fe₃O₄/Pt thin films with nominal thicknesses of 80/5 nm were deposited simultaneously onto four 10 × 10 mm² borosilicate glass substrates by pulsed laser deposition (PLD). Deposition occurred under UHV conditions (<10⁻⁹ mbar) from a polycrystalline Fe₂O₃ target (Pi-kem, purity 99.9%), with λ = 532 nm, a pulse width of 8 ns, and a 10 Hz repetition rate. We have shown previously that 80 nm thick Fe₃O₄ films grown by PLD in this way are typically preferentially textured out of plane ⟨111⟩, with a grain size equivalent to the film thickness (80 nm), suggesting columnar growth.¹⁵ In this work, however, we report on films produced from heavily textured/ablated targets that resulted in an increase in droplet formation.^16,17 Note that the substrate carrier was not rotated during deposition, which results in the small variations in the film deposition observed here.

Structural characterization was carried out using a combination of x-ray diffraction (XRD) on a Bruker D2 phaser and x-ray reflectivity (XRR) on a Siemens D5000. In addition, polarized neutron reflectometry (PNR) measurements were performed on the PolRef (ISIS, UK) beamline, with the OSMOND 1D detector,¹⁸ and fit using the Refl1D software package.¹⁹ 

The microstructure and chemical distribution of the thin film layers were examined using a FEI Tecnai F20 field emission gun scanning transmission electron microscope (FEGSTEM). Images of the cross sections of the layers were recorded in scanning mode with a probe size of ∼1 nm using the high angle annular dark field (DF) (HAADF, Fischione) and bright field (BF, Gatan) detectors. The TEM was also equipped with an Oxford Instruments X-MaxN 80 TLE windowless x-ray detector to allow energy dispersive x-ray spectroscopy (EDS) for chemical mapping. Cross sections of thin films for STEM were prepared using a FEI Nova 600 nanoLab focused ion beam (FIB)/scanning electron microscope (SEM).^15,20 The TEM lamella had a thickness of approximately 100 ± 20 nm for EDS measurements and 30 nm for high resolution imaging.

Sheet resistance measurements of the bilayers were obtained using the van der Pauw geometry. Finally, spin Seebeck measurements were obtained using the heat flux method at room temperature.^21,22 All samples were the same geometry (8 × 8 mm² active area) and mounted similarly; further details can be found elsewhere.^15,21,27

Figure 1 shows a schematic of the substrate orientation with respect to the target and laser direction during deposition, and XRD for the four films that were deposited (labeled G50#1–#4). Initial analysis of the XRD patterns only indicated subtle differences between the four samples: (1) minor differences in the height and full width maximum of the Pt (111) peak at 2θ = 39.65°; (2) changes in the height of the Bragg peak at 2θ = 44.7°, which we attribute to an α-Fe phase; and (3) emergence of a minor feature at 2θ = 42.4° for samples G50#2 and G50#4. The difference in the Pt (111) peak is likely due to a combination of slightly different Pt thicknesses and strain or disorder. Corresponding analysis of XRR confirmed that the Pt thicknesses, t_Pt, differed slightly between samples, as summarized in Table I, which also manifested in variations of the measured sheet resistance of each sample, R_s. The resistivity, ρ = R_st_Pt, was consistently around 2.2 × 10⁻⁷ Ω m, as expected for these bilayers.¹⁵ 

Figure 2 shows electron microscopy measurements of G50#1 and G50#2. Top–down SEM and EDS indicated a stark difference in surface texture: whereas G50#1 had multiple plate like formations, and G50#2 had a lower density of droplets that were more well-formed; corresponding EDS indicated that these droplets were strongly Fe rich. Figures 2(c) and 2(d) show STEM/BF, and Figs. 2(e) and 2(f) show HAADF images of the typical cross sections of G50#1 and #2. The corresponding EDS maps of the Fe₃O₄/Pt area are also shown for Pt, Fe, O, and Si. The BF images showed a polycrystalline magnetite layer with a thickness of approximately 85.2 ± 2.8 nm and an estimated Pt thickness of 8.7 ± 0.2 nm. The errors here come from the HRTEM lattice, but we expect that the surface waviness across the 100 nm slice resulted in differences between values estimated by XRR and TEM.

In line with the Fe maps, a number of Fe-rich, O-deficient spherical regions were present throughout the magnetite layer, varying from 30 to 80 nm in diameter, indicating that they are likely to be Fe droplets. This is in contrast to previous TEM of Fe₃O₄/Pt bilayers produced using the same system and method, but with a much less textured target.¹⁵ The major difference in the two samples lies in the population density of the Fe droplets, i.e., from this small slice G50#1 appeared to have 4–5 times more Fe droplets throughout the layer than G50#2, as highlighted in the Fe (and O by loss of contrast) EDS maps.

As mentioned already, the films presented here were deposited from a Fe₂O₃ target that had been previously ablated sufficiently enough to cause significant texturing of the target. Electron microscopy confirmed that this results in an increase in droplet formation, which agrees with previous observations as the deposition target is degraded.^17,23,24 While it is a common practice to use a dense or freshly polished target in order to suppress droplet formation,²⁵ we use this feature here to explore the impact of the inclusion of α-Fe nanodroplets on measured magnetothermal effects (SSE + ANE). We argue that due to alignment of the substrates, as the target became more heavily textured, the plume was redirected so that substrates G50#1 and #3 were situated closer to the edge of the plume [see Fig. 1(c)]. It has been shown that droplets are likely to be found on the circumference zone of the laser,²⁶ hence the higher droplet density observed in G50#1 (Fig. 2, top panel).

Figure 3 shows representative magnetothermal measurements for these samples as well as a similar series produced from a previously textured target (G18#1–4), and a summary of the spin Seebeck coefficient obtained. Nominal thicknesses of 80 nm Fe₃O₄ and 5 nm Pt were chosen for reproducibility with prior work, in particular the observed saturation of the spin Seebeck signal for t_Fe3O4 > 80 nm at approximately 47 nV m/W.²⁷ Any small differences in the Fe₃O₄ thickness should, therefore, not have an impact of the observed spin Seebeck signal.

To summarize, there is a clear grouping of S_Jq for samples deposited at the same time: a subset that demonstrates S_Jq $≈38$ nV m/W, similar to previously reported values,²⁷ and a second subset that exhibits a 64% increase to approximately 63 nV m/W. This is a significant enhancement of the voltage generated for films deposited at the same time. Also shown in Fig. 3 is S_Jq measured for another set of samples, G18. In this case, a similar trend was observed, but due to sample degradation and limited information on substrate placement with respect to the plume, we were unable to confirm that an increase in Fe droplets was the source of this enhancement. We include here only to demonstrate the reproducibility of this effect for highly textured targets.

As electron microscopy only provides a small snapshot of the films, we also used PNR to probe the depth-dependent structure and magnetization of the layers, averaged in the $xy$-plane. Data were obtained at a positive field saturation (700 mT) and in a smaller field of 14 mT after negative field saturation of −700 mT. As seen in Fig. 4, a clear splitting in the PNR up (u) and down (d) scattering cross sections was observed, which is a direct response of the magnetic scattering from the sample. There were also clear fringes that primarily arose from the Fe₃O₄/Pt bilayer. [We show XRR data for the for samples in Fig. 4(a) as well, to demonstrate the contrast observed due to the Pt layer.] These data were first fit using a standard “slab” bilayer model, where the interfaces were modeled using analytical error function profiles. As the neutron scattering length for Fe₃O₄ and Pt is closely matched, we used data from XRR and SEM to constrain the fit of the Pt layer to be within 10% of the estimates summarized in Table I. The scattering length density (SLD) profile obtained from the best fits using this model confirmed a slightly thinner Fe₃O₄ layer for G50#3, which is consistent with the assumption that it lays on the circumference of the plume (hence lower deposition rate and higher droplet density). However, the posterior distributions of fit parameters for thickness of the Pt layer (t_Pt) showed a strong preference to decrease down to the lower bound of the limit imposed by our model. Furthermore, the two roughness parameters (σ_i and σ_t) show a strong correlation (Pearson correlation $p≈$ 0.6), indicating that for this model parameterization, there was not enough information in the data to uniquely determine these values.

The SLDs obtained using the morphological model are shown in Fig. 5(e). While we fixed the value of $ρ Fe$ to that of bulk Fe, the value of $ρ Pt$ was allowed to vary to consider that the Pt layer may not be continuous across the entire film. Figure 5(f) shows the posterior distributions for $r 0$⁠, $t Pt,$ and the derived area particle density $n A= f N 2 3 r 0 2$⁠.

The mean values of $r 0$ were 45.6 ± 2.3 nm and 47.1 ± 1.7 nm for G50#3 and G50#4, respectively, which correlates well with typical particle sizes observed in SEM and TEM. Typical particle area densities, $n A$⁠, ranged from about 25 to 55 particles/μm², which again agreed with particle counts in SEM images (Fig. 2). From the magnetic SLD, $ρ M$⁠, we calculated the saturation magnetization as 497 ± 19 (G50#3) and 528 ± 16 kA/m (G50#4) for the Fe₃O₄ film, similar to previously measured moment of 465 kA/m.¹⁵ For the Fe droplets, we found 1110 ± 11 (G50#3) and 1190 ± 65 kA/m (G50#4), which is lower than the bulk saturation magnetization of 1700 kA/m, but not unexpected for α-Fe nanoparticles. The values of $t Pt$ were less than suggested from XRR and HRTEM, which is likely due to thickness variations in the flat and curved areas of the film that were not accounted for with our model. Regardless of the simplified assumptions, the morphological model allowed us to extract general parameters describing the surface morphology.

To summarize, a series of Fe₃O₄/Pt thin films were prepared with varying density of Fe nanodroplets. It was shown that for the samples with higher density of these nanodroplets, there was a 64% increase in the observed magnetothermal voltage, which far exceeds any expected change due to additional ANE in the Fe nanodroplets. We showed using PNR that there was no significant change in the magnetization of the Fe₃O₄ layer, and that the Fe nanodroplets could also be contributing some magnetic moment to the device. As the contribution due to ANE in the Fe₃O₄, Pt layers, and Fe nanodroplets can be considered negligible in this case, and PNR does not show a significant difference in magnetization at the interface, we suggest that the observed enhancement of the SSE is due to a combination of modification of the interface roughness and impact of the Pt coated Fe nanodroplets on spin injection efficiency. Previous observations of an increase in spin mixing conductance in Fe₃O₄/Pt layers with interface roughness would suggest an upper limit increase of 20% (based on the RMS roughness doubling).¹² While possible ANE contributions from the Fe nanodroplets are expected to be negligible due to the low volume content and comparatively lower signal,²⁹ they could result in local magnetization of the Pt layer that would enhance any proximity induced ANE. In addition, we argue that the inclusion of Fe in the Pt may have resulted in an increase in spin injection efficiency as theorized by Zhang et al.¹³ This relationship between the surface morphology and spin caloritronic effects suggests a route to engineer the efficiency of spin Seebeck-based devices, e.g., for waste heat recovery, and determining the exact nature of this increase will be the focus of future work.

The authors acknowledge Dr. Keith Yendall, Mr. Sam Davis, and the Loughborough Materials Characterisation Centre for assistance and use of the facilities. This work was supported by the Engineering and Physical Sciences Research Council (No. EP/P006221/1). N.L. received funding from the European Research Council (ERC) under European Union's Horizon 2020 Program, Contract No. 101001290 (3DNANOMAG).