Spin‐Charge Conversion in Fe/Au/Sb2Te3 Heterostructures as Probed By Spin Pumping Ferromagnetic Resonance

Large‐area antimony telluride (Sb2Te3) thin films are grown by a metal organic chemical vapor deposition technique on 4” Si(111) substrates, and their topological character probed by magnetoconductance measurements. When interfaced with Fe thin films, broadband ferromagnetic resonance spectroscopy (BFMR) shows a clear increase of the damping parameter in Fe/Sb2Te3 when compared to a reference Fe layer, which may suggest the occurrence of spin pumping (SP) into Sb2Te3. Simultaneously, X‐ray reflectivity and conversion electron Mossbauer spectroscopy evidence the development of a chemically and magnetically pure Fe/Sb2Te3 interface. However, by conducting SP‐FMR, it is shown that no spin‐to‐charge conversion (S2C) occurs in Fe/Sb2Te3, while a clear SP signal develops by introducing a 5 nm Au interlayer between Fe and Sb2Te3, with a measured inverse Edelstein effect conversion efficiency of λIEE = 0.27 nm. The results shed some light on the correlation among the chemical‐structural‐magnetic properties of the Fe/Sb2Te3 interface, the broadening of the magnetic damping parameter as detected by BFMR, and the occurrence of S2C, as probed by SP‐FMR.


DOI: 10.1002/admi.202101244
view of realizing a step toward technology transfer, developing fabrication methods suitable to guarantee the large-scale production of TI is necessary.
The Metal Organic Chemical Vapor Deposition (MOCVD) process has recently been successful in growing epitaxialquality TI, such as antimony telluride (Sb 2 Te 3 ) and bismuth telluride (Bi 2 Te 3 ) on 4" Si(111) substrates, being characterized by the typical Van der Waals stacking order. [5,6] In particular, Sb 2 Te 3 layers are characterized by the expected rhombohedral crystalline structure belonging to the R-3m space group. [7][8][9] In the present manuscript, we show that epitaxial Sb 2 Te 3 layers on Si(111) manifest improved magnetoconductance (MC) performances when compared to the "granular" Sb 2 Te 3 obtained on SiO 2 , [10,11] and we probe S2C in heterostructures formed by Fe/Sb 2 Te 3 and Fe/Au/Sb 2 Te 3 stacks by making use of broadband ferromagnetic resonance spectroscopy (BFMR) and spin pumping FMR (SP-FMR).
Following the deposition of Fe layers in direct contact with Sb 2 Te 3 , BFMR evidences a clear increase of the damping parameter from 23 · 10 −3 in the reference Au/Fe/Si(111) to 41 · 10 −3 in Au/Fe/Sb 2 Te 3 , where epitaxial Sb 2 Te 3 is obtained following proper thermal processing. [5] Simultaneously, X-ray reflectivity (XRR) and conversion electron Mössbauer spectroscopy (CEMS) evidence the development of a chemically and magnetically pure interface. However, by conducting SP-FMR, we show that no S2C occurs when Fe is in direct contact with epitaxial Sb 2 Te 3 , while a clear SP signal emerges when a 5 nm Au interlayer is introduced to fabricate an Au(5 nm)/Fe(5 nm)/ Au(5 nm)/Sb 2 Te 3 /Si(111) heterostructure. By interpreting the S2C conversion with the inverse Edelstein effect (IEE), a conversion efficiency of λ IEE ≈ 0.27 nm is obtained, in good agreement with that previously observed in Co/Sb 2 Te 3 systems. [4] Our results point to the conclusion that the observed S2C, as detected by SP-FMR in Fe/Au/Sb 2 Te 3 , originates from the topologically protected surface states (TSS) of Sb 2 Te 3 , which are likely detrimentally affected when depositing Fe in direct contact with Sb 2 Te 3 . Table 1 summarizes the structure of the samples being the subject of the present work. In the AS DEP, PRE ANN, and POST Large-area antimony telluride (Sb 2 Te 3 ) thin films are grown by a metal organic chemical vapor deposition technique on 4" Si(111) substrates, and their topological character probed by magnetoconductance measurements. When interfaced with Fe thin films, broadband ferromagnetic resonance spectroscopy (BFMR) shows a clear increase of the damping parameter in Fe/Sb 2 Te 3 when compared to a reference Fe layer, which may suggest the occurrence of spin pumping (SP) into Sb 2 Te 3 . Simultaneously, X-ray reflectivity and conversion electron Mossbauer spectroscopy evidence the development of a chemically and magnetically pure Fe/Sb 2 Te 3 interface. However, by conducting SP-FMR, it is shown that no spin-to-charge conversion (S2C) occurs in Fe/Sb 2 Te 3 , while a clear SP signal develops by introducing a 5 nm Au interlayer between Fe and Sb 2 Te 3 , with a measured inverse Edelstein effect conversion efficiency of λ IEE = 0.27 nm. The results shed some light on the correlation among the chemical-structural-magnetic properties of the Fe/Sb 2 Te 3 interface, the broadening of the magnetic damping parameter as detected by BFMR, and the occurrence of S2C, as probed by SP-FMR.

Introduction
Topological insulators (TI) are gaining attention from a technological point of view due to their foreseen highly efficient capability to control adjacent magnetic media and to act as ideal spin sinks through spin-to-charge conversion (S2C) phenomena. [1] Clarifying the role played in S2C by the specific chemical bonds and/or the presence of magnetically dead-layers and/or intentionally-grown interlayers at the interface between TI and ferromagnetic (FM) layers is highly demanding. [2][3][4] Additionally, in ANN heterostructures, the only difference is in the Sb 2 Te 3 preparation (see Experimental Section). [5] The evaporated Fe and Au layers are polycrystalline, see Figure S1 in the Supporting Information. The post-growth annealing of Sb 2 Te 3 triggers a reorganization of the film grains, favoring the formation of highly-oriented layers, being characterized by a surface mean roughness (as detected by atomic force microscopy) around 1 nm. [5] A scanning electron microscopy image of the post-annealed Sb 2 Te 3 is shown in Figure S2 in the Supporting Information. Moreover, the total reflection X-ray fluorescence spectrometry analysis showed mainly the presence of Sb and Te, with no additional elemental signals detected, thus demonstrating the absence of undesired contaminants in the Sb 2 Te 3 layers, see Supporting Information ( Figure S3, Supporting Information)." Figure 1a shows the MC (expressed by Δσ in units of e 2 /h) recorded at 10 K for the Sb 2 Te 3 layers in AS DEP, PRE ANN, and POST ANN samples prior to the Fe/Au depositions. A clear non-parabolic perturbation is observed for the POST ANN sample. As a matter of fact, the MC data in all samples display a deviation from pure Lorentzian parabolic contribution, which is attributed to weak antilocalization (WAL) emerging at low temperature in the presence of highly conductive 2D states, due to intense spin-orbit coupling (SOC). Typically, the presence of WAL in TI is associated with the existence of TSS that contribute to the global conduction mechanisms. [12][13][14][15] In the framework of the Hikami-Larkin-Nagaoka (HLN) model, [11] the parameters α HLN (being connected to the number of 2D conducting channels), and the spin coherence length (l ϕ ), can be extracted following the fit of the MC curves, and the results are shown in Figure 1b.

Magnetotransport in Extended Sb 2 Te 3 on Si(111)
Regardless of the thermal processing to which they are subjected, [5] all the Sb 2 Te 3 layers exhibit WAL, and the extracted α HLN ranges from 0.1 to 0.3. These values are lower than the 0.5 expected in the case of a single conductance channel (likely the surface), indicating that there is still a contribution from the bulk states. Indeed, all the layers show a consistent decrease of their sheet resistance (R S ) upon cooling (not shown) that, together with the high holes density (≈10 20 cm −3 ), suggests a relevant participation of bulk states in the global conduction mechanism. The R S of Sb 2 Te 3 layers lowers with thermal processing (insets in Figure 1b), due to their improved crystalline quality. [5] This is reflected in the marked increase of l ϕ from 31 nm for asdeposited Sb 2 Te 3 up to 60 nm in the post-annealed Sb 2 Te 3 layer.
It is interesting to note how, even in the as-deposited Sb 2 Te 3 , there is a relatively high α HLN , being much higher (at same T) than that observed in as-deposited Sb 2 Te 3 grown on SiO 2 , while the l ϕ is lower. [11] We attribute the higher α HLN on top of Si(111) as due to a better intra-granular structural quality when compared to Sb 2 Te 3 films on top of SiO 2 , while the lower l ϕ may be connected to a higher density of grain boundaries, limiting the transverse spin-coherence length.

Chemical, Structural, and Magnetic Characterization of the Fe/Sb 2 Te 3 Interface
In Figure 2, the chemical-structural analysis carried out by XRR on the whole set of samples listed in Table 1 is shown. Here, the collected data (black solid line) are fitted with a multilayer model and the best fit is represented by a red solid line. For the AS DEP sample, it is not possible to conduct a useful fit. This is caused by its high surface roughness of 3.9 nm. [5] This value is comparable to the nominal 5 nm-thickness of the 57 Fe layer evaporated on top (Table 1), thus originating a non-uniform distribution of the Fe thickness, likely promoting a partial coverage of the Sb 2 Te 3 surface. Indeed, the Croce-Nevot XRR model can be successfully employed for samples with a stacking order not too far from a multilayered system, where the levels of the interdiffusion or structural inhomogeneities are restrained. [16] Differently, clear XRR oscillations emerge in the PRE ANN and POST ANN samples (Figure 2a), showing how the improved structure and morphology of the developed epitaxial Sb 2 Te 3 directly reflects in the development of a well-controlled Au/Fe/Sb 2 Te 3 heterostructure. From XRR, a high chemicalstructural quality is observed also for the REF-IL and POST ANN-IL samples, demonstrating the successful integration of the Au interlayer in separating the Fe layer from the Sb 2 Te 3 at the bottom of the stack (Figure 2b).
The extracted XRR parameters for all the samples are summarized in Table 2 Table 1 (black solid line). The best fit of the curves is represented by the red solid lines and the extracted parameters are summarized in Table 2. As substantiated in the text, for the AS DEP sample a reliable XRR model could not be found due to the relatively high roughness of un-treated Sb 2 Te 3 . Table 2. XRR parameters extracted from the fit shown in Figure 2. The error bars are ± 0.1 nm for the thickness (and roughness), and ± 0.05 e − Å −3 for the electronic densities. samples, demonstrates the good reproducibility of the Au/Fe/Au deposition process on top of Sb 2 Te 3 and Si(111). Conversion-electron Mössbauer spectroscopy (CEMS) is conducted on all the samples indicated in Table 1, and the results depicted in Figure 3. In the Fe/Sb 2 Te 3 structures that we previously discussed in refs. [7,10], pulsed laser deposition was employed to grow a ≈1 nm 57 Fe layer in direct contact with Sb 2 Te 3 , then capped by a few nm of non-CEMS sensitive 54 Fe. The main difference with the set of samples here under study (Table 1) is that 100% of isotopically enriched 57 Fe layers are used. Consequently, CEMS probes both the top Au/Fe and bottom Fe/Sb 2 Te 3 interfaces in AS DEP, PRE ANN and POST ANN, the two Au/Fe and Fe/Au interfaces in REF-IL and POST ANN-IL, and of course the whole Fe layer in all the samples. It is demanding to separate the contribution in the spectra from the top Au/Fe, in order to monitor the changes occurring at the bottom Fe/Sb 2 Te 3 interface depending on the Sb 2 Te 3 preparation (Table 1).
We adopt the following fitting strategy. First, the REF sample of Figure 3a is analyzed, with the aim of identifying the hyperfine parameters for the components originating at the top Au/Fe interface. These are used, and fixed (except where indicated), in the AS-DEP, PRE ANN, and POST ANN samples in Figure 3b-d, since the Au/Fe deposition is done at RT in all the samples following identical procedures and without any postgrowth processing.
The CEMS-spectrum for the REF sample is dominated by a magnetically-split sextet (green SEXTET in Figure 3) that is attributed to α-Fe, with a B hf slightly lower than the bulk case (32.9 T) as often observed in thin films. [17] Additionally, a small magnetic component DIST is detected (yellow in Figure 3), together with two paramagnetic doublets DOUB-1 and DOUB-2, respectively red and orange in Figure 3. All these components are still present in the PRE ANN and POST ANN samples, plus an additional paramagnetic doublet DOUB-3 (shown in Figure 3). In the REF-IL and POST ANN-IL samples, we detect α-Fe, DIST and only the DOUB-1 paramagnetic component. All the parameters extracted from the fit of all the CEMS-spectra depicted in Figure 3, are summarized in Table 3.
The DIST in REF is due to Fe atoms partially mixing with Au at the Au/Fe interface. [18] In the AS DEP, PRE ANN and AS DEP samples, DIST now also contains the contribution from the bottom Fe/Sb 2 Te 3 interface. [7,10] However, due to the relatively low spectral intensity of the DIST component, we keep a unique distribution of sextets in the fit of the AS-DEP, PRE ANN and POST ANN CEMS signal, where the contribution from the bottom Fe/Sb 2 Te 3 interface becomes dominant, as indicated by the lower isomer shift when compared to the DIST in REF (Table 1), which is now very similar to that observed at the Fe/Sb 2 Te 3 interfaces. [7,10] The nature of the DIST component seems not to be very much affected by the Sb 2 Te 3 preparation ( Figure 3 and Table 3). DOUB-1 in the REF sample could potentially originate from both the top Au/Fe or/and the bottom Fe/Si interface(s). It is well known that Fe and Si have a certain reactivity even at RT, and paramagnetic doublets arise in the CEMS spectra [19,20] Indeed, FeSi and FeAu components may have a similar quadrupole splitting around 0.7 mm s −1 as we detect. [18,20] Being the observed isomer shift of 0.4 mm s −1 (Table 3) very close to that expected at the Au/Fe interface, [18] we tend to attribute DOUB-1 mostly to the top Au/Fe interface, even if a partial contribution to the spectral intensity from FeSi cannot be ruled out. DOUB-1 is indicated with "FeAu" in The AS DEP CEMS-spectrum ( Figure 3b) has an overall shape much well-resembling those previously observed at the interface in the 54 Fe/ 57 Fe(1 nm)/Sb 2 Te 3 /SiO 2 /Si heterostructures. [7,10] AS DEP displays a dominant central paramagnetic component, for which the hyperfine parameters do not match those of DOUB-1 in REF. As anticipated, through the fully 57 Feenriched Fe layer, CEMS now explores the whole 5 nm-thick Fe layer and the additional top Au/Fe interface (Table 1). Therefore, in the fit of AS DEP we first include the DOUB-1 attributed to the top Au/Fe interface, as identified in the REF sample (Table 1) and keep fixed its hyperfine parameters in the fit of all Table 3. Hyperfine parameters for all the spectral components used to interpret the CEMS data shown in Figure 3, for the full set of samples of Table 1. (1) 1.80 (9) 1.020(4) (2) 1.80 (7) 1.02 (3) <Γ> the Sb 2 Te 3 -containing samples (  (Table 3). In Fe/Sb 2 Te 3 (granular)/SiO 2 /Si, the central paramagnetic doublet has been assigned to the formation of a FeTe-type of the compound at the Fe/Sb 2 Te 3 interface, [7,10] and we attribute DOUB-3 to the formation of a "FeTe" coordination that affects 43% of the Fe atoms in the AS DEP sample. The isomer shift of the DOUB-3 component is in good accordance with those previously reported, with a larger quadrupole splitting, [7,10] thus indicating a highly disordered environment around Fe. This is most likely due to the very large surface roughness up to 3.9 nm of the as deposited Sb 2 Te 3 on Si(111), [5] being basically of the order of the nominal Fe thickness. Indeed, the surface roughness of the as-deposited Sb 2 Te 3 on SiO 2 was of the order of ≈2 nm, [11] on top of which, most likely, Fe layers grow with a slightly higher conformality, thus generating Fe components with lower quadrupole splitting than in the case of AS DEP. Also, in refs. [7,10] the Fe thickness was ≈10 nm, thus certainly providing a more uniform layer.
A marginal Fe oxidation is detected through DOUB-2 in AS DEP, PRE ANN, and POST ANN samples, see Figure 3e Figure 3e shows the evolution of the relative spectral intensity for all the components identified from the fit of the AS DEP, PRE ANN, and POST ANN samples. The main effect of the Sb 2 Te 3 thermal processing is to induce a collapse of the "FeTe" contribution, promoting a good quality Fe/Sb 2 Te 3 interface, with the CEMS data of the POST ANN (Figure 3d) basically resembling that of the REF sample (Figure 3a). We underline a remarkable difference between the evolution of the quality of the Fe/Sb 2 Te 3 interface (both structurally and magnetically) following the Sb 2 Te 3 growth on SiO 2 , [10] with that now observed on Si(111). In the former case, the annealing of the Sb 2 Te 3 prior to Fe deposition has been shown to have no influence on the relative abundance of the interface FeTe-type of bonding, with a constant fraction of 15-20% detected. [10] This is strikingly different from the case of Sb 2 Te 3 on Si(111) here presented, where already the pre-annealing of the Si(111) before the Sb 2 Te 3 deposition has an immediate effect in reducing the central paramagnetic component at the Fe/Sb 2 Te 3 interface, see Figure 3c-e. We think this is mainly originated by the very efficient action of the substrate thermal treatment in reducing the Sb 2 Te 3 surface roughness when grown on Si(111), as previously reported. [5] This condition leads to an almost epitaxial growth of Sb 2 Te 3 with clear van der Waals layers already in the PRE ANN case. [5] As a consequence, the subsequent Fe growth occurs more regularly, with marginal chemical mixing, finally giving an almost pure magnetically-ordered Fe layer, comparable to a reference growth directly on Si(111), see Figure 3a,c,d. Therefore, the better structural quality of Sb 2 Te 3 in the PRE ANN and POST ANN samples, when compared to AS-DEP (and to granular Sb 2 Te 3 on SiO 2 ), has a major role in limiting the Fe chemical interaction with Te close to the Sb 2 Te 3 surface. Additionally, a role could be also played by the higher energy of the impinging Fe atoms in the case of a pulsed laser deposition, [10] when compared to e-beam evaporation (present results). To develop a theoretical model for the evolution of the Fe/Sb 2 Te 3 interface upon Sb 2 Te 3 annealing is of certain interest, with a special focus on the thermally induced tuning of the chemical, structural or magnetic properties. We hope our results will further boost the use of first-principles multiscale modeling in this direction. [21]

Broadband FMR and Spin Pumping FMR in Au/Fe/(Au/)Sb 2 Te 3
BFMR is employed to study the magnetization dynamics of samples REF, AS DEP, PRE ANN, and POST ANN reported in Table 1, and the acquired data are depicted in Figure 4. For the AS DEP sample, the BFMR signal is too weak to be easily revealed, likely due to the disordered morphology of the Au/Fe bilayer, as emerged from the discussion of the XRR and CEMS data (Section 2.2). Figure 4a reports the evolution of the resonant frequency (f res ) as a function of the resonant magnetic field (H res ), which is interpreted with Equation (1), for all samples.
where γ is the gyromagnetic ratio and M eff the effective magnetization. The M eff value can be expressed as 4πM eff = 4πM s -H k, where M s and H k are the saturation magnetization and the magnetic anisotropy field, respectively. The extracted M eff (g-factor) values are 586±38 emu cc −1 (2.47±0,02) and 594±33 emu cc −1 (2.47±0,02) for the PRE ANN and POST ANN, respectively, and 682±42 emu cc −1 (2.56±0,02) for the REF sample, being lower than in bulk Fe (M eff ≈1700 emu cc −1 ). On the other hand, the XRR and CEMS results point to a relatively good quality of the Fe layers, therefore excluding a significant contribution from magneto-structural disorder and/or magnetic dead layers in reducing M eff . We attribute the H k increase mainly to surface-induced anisotropy. [22][23][24] Even in very sharp interfaces, an additional source of magnetic anisotropy can also originate from the hybridization occurring at the Au/Fe and Fe/Sb 2 Te 3 interfaces, giving rise to a local modification of the materials band structure. [22] Figure 4b summarizes the full width at half maximum (FWHM) ΔH of the BFMR signal as a function of f res , and the linear fit is performed using Equation (2).
where α indicates the damping constant (or Gilbert parameter) and ΔH 0 the so-called inhomogeneous broadening term, a quantity accounting for the magneto-structural disorder present in a FM thin film.  (Figure 4b), show that these two samples are magnetically very similar, which is in accordance with their almost identical chemical, structural and magnetic properties, as probed by XRR and CEMS (Section 2.2). The measured ΔH 0 < 30 Oe for all samples (y-intercept in Figure 4b) is significantly lower or comparable with many reports present in the literature. [25][26][27][28] Again, this corroborates the good magneto-structural quality of the Fe layer anticipated by XRR and CEMS results.
Often, the increase of the FMR damping parameter α in FM/ TI systems, when compared to FM-reference structures, is attributed to the generation of a pure spin current in the FM layer that is pumped orthogonally in the TI. [29][30][31][32][33][34] When this happens, following SP, a longitudinal charge current should be generated at the FM/TI interface. In the case of an occurring pure SP mechanism, the slope of the linear fit in Figure 4b is related to the socalled spin mixing conductance g eff , as described by Equation (3).
where μ 0 is the Bohr magneton, t FM the thickness of the ferromagnetic layer, α FM/TI and α FM are the damping constant of the samples with and without the TI layer, respectively. g eff is a fundamental figure of merit, which quantifies the spin current that accumulates at the FM/TI interface. Other than SP effects, an increased α value at FM/TI, when compared to FM, can also originate from a spin memory loss at the FM/TI interface and/or structural inhomogeneities of the FM layer. [35] The subtraction of a proper reference is considered a reliable strategy to exclude contributions not directly related to the SP process. [28,36] Indeed, this approach allows to eliminate the SP arising at the Au/Fe interface, thus isolating the contribution solely from the bottom Fe/Sb 2 Te 3 interface. The two-magnon scattering (TMS) cannot be fully accounted for with the subtraction method, and it can generate an over-or under-estimation of g eff . [4,35] On the other hand, our extracted g eff value for both PRE ANN and POST ANN samples is g eff = 2.2 · 10 19 m 2 , being very well in accordance with those reported for similar heterostructures, [25][26][27][28][29][30][31][32][33][34][35][36][37][38][39] thus suggesting a negligible contribution of TMS at the Fe/Sb 2 Te 3 interfaces. In order to investigate the relationship between the measured g eff in the POST ANN sample and the occurrence of S2C, and to estimate the SP efficiency, we refer to the procedure described in ref. [4]. In particular, SP-FMR is conducted on the POST ANN sample, and the results are depicted in Figure 4c. Here, the so-called mixing voltage (V mix ) is recorded, and the V mix (H res ) curve is fitted with the Lorentzian function, Equation (4 where V sym and V asym are the symmetric and anti-symmetric components, respectively due to SP and rectification effects. [4] The symmetric contribution can also arise from the thermal Seebeck effect. Being such a thermal contribution independent from the H res direction, [40] it can be eliminated by considering the effective SP signal (V SP ), as extracted by the following Equation (5).   Table 1. a) represents the evolution of the Kittel's curves in the in-plane (IP) configuration, from which γ and M eff are extracted. In panel b), the BFMR signal linewidth as a function of the resonant frequency is plotted. The red solid lines indicate the linear fit from which α and ΔH 0 are extracted. c) V mix acquired for the POST ANN sample at RF frequency of 12.5 GHz and RF power of 132 mW. The presence of an almost totally anti-symmetric signal for negative magnetic fields (red dots) indicates a negligible SP effect, as discussed in the main text.
absence of any S2C in POST ANN. This happens despite the good structural and magnetic quality of such sample (Section 2.2), and the presence of a clear enhancement of the Gilbert damping constant α, as detected by BFMR when compared to the REF sample (Figure 4b). Our results constitute a clear warning in concluding about the occurrence of SP (i.e., a successful S2C) only on the basis of (B)FMR. [29][30][31][32][33][34] The enhanced α value is due to the large transverse 3D spin current density ( J S D 3 ) that is generated in the Fe layer when in contact with Sb 2 Te 3 . On the other hand, this is not converted into charge current through S2C conversion, suggesting that the high SOC in Sb 2 Te 3 may not be sufficient to generate S2C.
It is well known how the nature of TSS is ideally preserved when TI are interfaced with non-magnetic materials (i.e., nonmagnetic impurities), while the direct contact with FM layers could prevent the conservation of the TSS. [36,41,42] In the case of S2C mainly occurring through the TSS, it would then be clear how such a process can be detrimentally affected when interfacing Fe with Sb 2 Te 3 . To get more insight into the above scenario, we conduct BFMR and SP-FMR on POST ANN-IL and REF-IL, where an Au interlayer between Fe and Sb 2 Te 3 is used to protect the TSS (Table 1), and the results are summarized in Figure 5.
From the analysis of the Kittel dispersion (Figure 5a), the extracted M eff (g-factor) values are 1183±15 emu cc −1 (2.12±0,01) and 1213±11 emu cc −1 (2.01±0,01) for the POST ANN-IL and REF-IL, respectively. When compared to the set of samples without the Au interlayer (Table 1), the M eff values are almost doubled. Due to the similar chemical, structural, and magnetic quality of the Fe layers in POST ANN and POST ANN-IL (Section 2.2), we attribute such a higher M eff to a strong reduction of the surface magnetic anisotropy due to the Au interlayer. Also the lower g-factors, when compared to the samples without Au interlayer, are in accordance with the higher crystallographic symmetry of the Fe/Au interface with respect to Fe/Sb 2 Te 3 , being characterized by a body center cubic (BCC)/BCC and BCC/rhombohedral symmetry, respectively. [22] Figure 5b shows the corresponding ΔH(f res ) curves for the POST ANN-IL and REF-IL samples. From the linear fit with Equation (2), it emerges that both samples have an extremely low ΔH 0 < 4 Oe, an indication of a very good magneto-structural quality. POST ANN-IL shows α = 8.75 · 10 −3 , higher than the α = 7.35 · 10 −3 in REF-IL. Such a measured Δα = 1.4 · 10 −3 generates a spin mixing conductance g eff = 5.3 · 10 18 m −2 , a value which is 76% lower than in POST ANN. Such a reduction is due to the introduction of the Au interlayers, which give rise to additional dissipative phenomena for the produced spin current (i.e., interfacial spin memory loss, scattering with impurities, thermal dissipation). [43] The almost perfect superposition of the Kittel curves showed in Figure 5a and the ΔH 0 < 4 Oe value extracted from Figure 5b further confirm that the Au/Fe/Au trilayers deposited on Sb 2 Te 3 and Si(111) substrates, have almost the same magneto-structural properties, a fundamental prerequisite to make a reliable comparison between the two cases, to estimate the S2C due to the presence of Sb 2 Te 3 .
To directly evaluate the S2C efficiency in the POST ANN-IL sample, SP-FMR experiments are conducted. According to Equation (6) Figure 4c for the POST ANN, it is clear that a large V SP appears thanks to the Au insertion, which clearly changes sign upon the reversal of the applied magnetic field, thus demonstrating the occurrence of S2C.
Our results demonstrate that the observed S2C cannot originate from a bulk spin Hall effect in Sb 2 Te 3 , otherwise it should be observed in POST ANN as well. The immediate explanation is that inserting a 5 nm Au layer between Fe and Sb 2 Te 3 protects the TSS in Sb 2 Te 3 (Figure 1), finally allowing to electrically detect S2C. This is in accordance with what we have previously observed in Co/Sb 2 Te 3 systems, [4] thus showing that the S2C is not dependent on the FM nature (Fe or Co). Indeed, the successful use of Au in protecting the TSS of other TI (Bi 2 Se 3 ) has been previously suggested, [44] with basically unaltered spinmomentum locking shown in Au/Bi 2 Se 3 . When compared to other metals such as Ag or Cr, Au has turned out to be more efficient in producing a chemically-inert Au/Bi 2 Se 3 interface. [44][45][46] Our results show that this may well be the case also in the Au/Sb 2 Te 3 system.
Within the above scenario, the S2C can be described by the so-called Inverse Edelstein Effect (IEE), [47,48] where the IEE length represents the figure of merit to quantify the S2C efficiency. For sample POST ANN-IL we obtain λ IEE = 0.27 nm, which is in perfect agreement with the value obtained in Au/Co/Au/Sb 2 Te 3 , [4] thus demonstrating the intrinsic role played by Sb 2 Te 3 and its preserved TSS to origin such a large S2C. When compared to other 2 nd class chalcogenide-based TI, [26,[37][38][39][48][49][50][51][52][53][54] the extracted λ IEE is comparable to the highest reported so far.
Our results show that realizing a chemically and magnetically pure Fe/Sb 2 Te 3 interface may not be enough to achieve a successful S2C, which is, on the other hand, achieved through the insertion of an Au interlayer to protect the TSS of Sb 2 Te 3 .

Conclusion
In summary, MC experiments show that the annealed epitaxial Sb 2 Te 3 develops TSS-connected conduction. When interfaced with Fe, the enhanced chemical-structural quality of Sb 2 Te 3 is connected to a remarkable suppression of intermixing occurring at the Fe/Sb 2 Te 3 interface. Indeed, BFMR evidences that such an enhanced chemical, structural and magnetic quality of the Fe/Sb 2 Te 3 interface gives rise to a clear broadening of the magnetic damping, which is often attributed to a successful S2C. However, our results show that this may not be the case, since no S2C is electrically detected in Fe/Sb 2 Te 3 samples. On the other hand, the use of an Au interlayer at the Fe/Sb 2 Te 3 interface turns out to be successful in achieving quite a large S2C, thus strongly suggesting that the observed effect mainly relates to the TSS of Sb 2 Te 3 , which are finally driving S2C, due to the IEE. We extract quite a large λ IEE = 0.27 nm conversion factor, in very good agreement with what we have observed by using an Au/Co/Au injector on top of Sb 2 Te 3 . [4] This demonstrates that a successful S2C conversion can be achieved in Sb 2 Te 3 -based systems by employing an Au interlayer, and such a conversion does not depend on the nature of the FM electrode (Fe, Co).

Experimental Section
Sb 2 Te 3 thin films were grown by MOCVD with an AIXTRON 200/4 setup operating with ultra-pure nitrogen carrier gas and equipped with a cold wall horizontal deposition chamber, accommodating a 4'' IR-heated graphite susceptor. The Sb 2 Te 3 crystalline quality improved by appropriate thermal treatment of the substrate (at 500 °C) prior to the deposition (PRE ANN sample in Table 1), followed by in situ postdeposition annealing at 300 °C (POST ANN sample in Table 1). All the details about the Sb 2 Te 3 samples preparation can be found in ref. [5].
Prior to the deposition of Fe, resistivity and MC measurements were carried out on the Sb 2 Te 3 thin films by using a four points probe with the van der Pauw method, by making use of a closed-cycle cryostat. [11] Magnetic fields up to 0.8 T were applied and directed perpendicularly with respect to the sample plane (and the current). The MC measurements were conducted at a constant applied current (50 µA into the as-deposited and 100 µA in the pre-and post-annealed Sb 2 Te 3 ), by measuring the voltage changes during the magnetic field scan. In particular, the measurements of the sheet resistance (R S ) as a function of the magnetic field (B) and temperature (T) were recorded on 1 × 1 cm 2 substrates cleaved from the grown wafers without any patterning, thus avoiding potential additional surface degradations.
Following MC, the Sb 2 Te 3 /Si(111) pieces were transferred into an Edwards Auto306 e-beam evaporation tool, where the Au(5 nm)/ 57 Fe (5 nm) bilayers and Au(5 nm)/ 57 Fe(5 nm)/Au(5 nm) trilayers were deposited all in situ. In all the processes the starting value of the vacuum in the deposition chamber was in the range of 5 · 10 −7 ÷10 −6 Pa. For each evaporated element, the electronic gun deposition currents and the corresponding chamber pressures during the growths were for Au: 120 mA and 7.8 · 10 −6 Pa; for Fe: 80 mA and 4.6 · 10 −6 Pa. The Si(111) substrates used as references were cleaned with isopropyl alcohol and treated with HF prior to the evaporation processes. Figure 6. Scheme of the experimental setup used for BFMR and SP-FMR experiments. For more details on the technical aspect of these measurements please refer to ref. [4]. X-Ray Reflectivity (XRR) analysis was conducted by using a Cu radiation with the K α emission at λ = 1.54 Å and the patterns were collected with a commercial scintillator and modelled through a matrix formalism corrected by a Croce-Nevot factor. XRR measurements allowed the determination of the thickness, the roughness, and the electronic density (ρ e ) of each layer composing the sample. In particular, ρ e is related to the measured critical vector Qc by the relation Qc (Å −1 ) 0.0375 ( / e e ρ = − Å 3 ) (for the Cu K α emission). [55] The isotopically enriched 57 Fe layer allowed to perform CEMS analysis, [10] which was carried out at RT on all samples in Table 1. CEMS was conducted in a constant-acceleration drive, with the sample mounted as an electrode in a parallel-plate avalanche detector filled with acetone gas. An α-Fe foil at RT was used for the CEMS velocity scale calibration and all the reported isomer shifts are relative to α-Fe. The relative area fraction in Figure 3e was obtained by assuming the same Debye-Waller factor f = 1 for all the components. Analysis of the CEMS data was carried out with the Vinda software package. [56] BFMR was conducted by a broadband Anritsu-MG3694C power source (1-40 GHz), connected to a home-made grounded coplanar waveguide (GCPW), where the samples were mounted in a flip-chip configuration (with the FM film close to the GCPW surface) with a 20 µm thick mylar foil stacked in between to avoid the shortening of the conduction line. The sample-GCPW system was positioned between the polar extensions of a Bruker ER-200 electromagnet maintaining its surface parallel to the external magnetic field H ext , in the so called in-plane (IP) configuration. During the measurements, an RF current at a fixed frequency was carried toward the GCPW and the transmitted signal was directed to a rectifying diode (Wiltron, Model 70KB50 NEG) which converts the RF-signal in a continuous DC-current, subsequently detected by a lock-in amplifier downward the electronic line. A scheme of the BFMR facility is shown in Figure 6. The same instrumental setup adopted for BFMR was employed to perform SP-FMR measurements. In this case, the edges of the sample were contacted with Ag paint and connected to a nanovoltmeter. For a fixed RF frequency and power, a DC-voltage was detected when the FMR occurred (see Figure 6).
For the SP-FMR measurements carried out in this manuscript, the RF frequency was fixed at 12.5 GHz and the RF power varied from 90 to 132 mW. According to the different RF power, the generated 3D spin current J S D 3 (in units of A m −2 ) was calculated by using Equation (6).  (6) where ℏ is the reduced Plank constant, M S the saturation magnetization, ω the frequency of the RF-signal, e the charge of the electron and h RF the transverse oscillating magnetic field generated by the GCPW.

Supporting Information
Supporting Information is available from the Wiley Online Library or from the author.