Large spin-to-charge conversion at room temperature in extended epitaxial Sb2Te3 topological insulator chemically grown on Silicon

Spin-charge interconversion phenomena at the interface between magnetic materials and topological insulators (TIs) are attracting enormous interest in the research effort towards the development of fast and ultra-low power devices for the future information and communication technology. We report a large spin-to-charge conversion efficiency in Au/Co/Au/Sb2Te3/Si(111) heterostructures based on Sb2Te3 TIs grown by metal organic chemical vapor deposition on 4 inches Si(111) substrates. By conducting room temperature spin pumping ferromagnetic resonance, we measure an inverse Edelstein Effect length {\lambda}IEE up to 0.75 nm, a record value for 3-dimensional chalcogenide-based TIs heterostructures. Our results open the path toward the use of chemical methods to produce TIs on large area Si substrates and characterized by highly performing spin-charge conversion, thus marking a milestone toward future technology-transfer.


Introduction
Information and Communication Technologies (ICT) are deeply changing our lives and working routines, and this trend got remarkably boosted during the Covid-19 pandemic.
Governments' digital agendas consider expanding the use of ICT products and services 1 at all levels. In the 2005-2019 period, the number of individuals using the Internet grew from 1.1 billion to 4 billion, representing the 51% of the world population. 2,3 The ever-expanding ICT will have a huge impact in terms of power consumption. In 2020, the electricity consumption due to ICT was ~ 3000 TWh i.e. 11% of the total, with a foreseen increase up to 8000 TWh in 2030. 4,5 This constant increase could have a strong impact on climate change, which is one of greatest challenges of the 21 st century. 6 In order to improve the overall efficiency and lower the power consumption of any electronic circuit and device, new materials with enhanced functionalities must be brought to a maturity level.
Topological Insulators (TIs) represent a state of matter in which the material bulk has insulating properties while the surface hosts highly conducting states. 7 In TIs, electrons are characterized by a Dirac-like dispersion energy and very strong spin-orbit coupling determine the electron spin orientation with respect to their momentum thus generating topologically protected surface states (TSS). 7 TIs are therefore considered a very plausible solution to bring spintronics to the next level in the future ICT, 5,8 in which the devices' functionalities can be driven by a collection of spin-orbit coupling phenomena such as spin Hall effects (SHE). 9 Thanks to their TSS, TIs provide an efficient alternative to the typically used heavy metals (HM) for exploiting spin-charge interconversion effects in heterostructures where TIs and magnetic materials are interfaced. 10,11 The second generation of 3-dimensional (3D)-TIs, such as bismuth and antimony chalcogenides-based Bi 2 Se 3 , Bi 2 Te 3 and Sb 2 Te 3, is attracting huge interest. [12][13][14] They are narrow band-gap semiconductors with rhombohedral crystalline structures belonging to the R-3m space group. 12,14 In principle, exploiting TSS in these 3D-TIs requires epitaxial quality thin films, feature most commonly achieved by the widely-reported Molecular Beam Epitaxy (MBE) deposition method, [15][16][17][18][19] with several reports about the use of magnetron sputtering also available. [20][21][22] In order to fill the gap between research and technology, a firm and decisive effort to develop methods to grow TIs on large-area Si substrates, by simultaneously controlling their functional properties, is highly required.
Recently, chemical methods, such as Atomic Layer Deposition, Chemical Vapor Deposition (CVD), and Metal-Organic CVD (MOCVD) were shown to allow cost-effective depositions and complex 3D structures on large areas. 23,24 In a recent review by Zabaveti et al. 25 a comparison between growth methods for the synthesis of chalcogenides thin films in terms of their lateral dimension, has showed the clear advantage in using chemical methods (i.e. costeffectiveness, complex 3D structures).
We recently developed a MOCVD process to grow epitaxial-quality Antimony Telluride (Sb2Te3) on 4" Si(111) substrates 24 (Supplementary Info. -Fig. S1). When compared to granular-Sb2Te3 grown on SiO2, 26 the epitaxial-Sb2Te3 on top of Si(111) shows improved magnetoconductance (MC) performances especially upon proper annealing, providing clearer and more robust TSS (Supplementary Info -Fig. S2). The next fundamental step is therefore to quantify and optimize spin-charge interconversion phenomena at the interface of the developed TIs with magnetic materials.
The use of spin-pumping ferromagnetic resonance (SP-FMR) to investigate spin-tocharge (S2C) conversion at ferromagnets (FM)/HM interfaces has been theoretically described for a long time, 27,28 and widely demonstrated. [29][30][31][32][33][34][35] Alternatively, also spin torque -FMR (ST-FMR) [36][37][38] and second harmonic longitudinal voltage 17,39,40 measurements have been reported. In the case of FM/TIs systems, several reports have recently emerged with studies by ST-FMR, 41-43 spin Seebeck effect, 44 or SP-FMR. 20, 29,31,34,35,45,46 In this work, we report a large S2C conversion occurring at room temperature (RT) in Au/Co/Au/Sb2Te3/Si(111) heterostructures, by making use of broadband FMR (BFMR), also known as all-electrical spin wave spectroscopy, and SP-FMR. In SP-FMR, a pure spin current is generated in the Co layer and perpendicularly pumped into the adjacent 3D-Sb2Te3, through the Au interlayer, which is found essential for suppressing interfacial non-linear effects due to two magnon scattering (TMS). As a figure of merit for the S2C conversion efficiency quantification, we measure the inverse Edelstein effect length 47 , which is found to range from 0.28 nm to 0. 75 Fig.S1). In order to promote an epitaxial order, the Sb2Te3 films are subjected to specific in-situ thermal treatments. 24 The Au(5nm)/Co and Au(5nm)/Co/Au(5nm) capping layers are prepared by e-beam evaporation on pre-cut ~1 x 1 2 Sb2Te3 pieces using an Edwards Auto306 facility, producing Au(5nm)/Co(t)/Sb2Te3 and Au(5nm)/Co(t)/Au(5nm)/Sb2Te3 heterostructures, with the nominal thickness (t) within the 2 -30 nm range ( Fig. 1(a)). The BFMR and SP-FMR experiments are conducted using a home-made setup as depicted in Fig. 1(b), where the sample is 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 "flip-chip" configuration for in-plane (IP) measurements. 48 To induce an oscillating magnetic field in the FM layer, the sample is fixed on a custom grounded coplanar waveguide (GCPW) ( Fig. 1(b,c)) connected to a broadband Anritsu RF-source (Supplementary Info. - Fig. S3 and Fig. S4). The FMR signal for a fixed RF frequency is performed by measuring the derivative of the absorption power downstream of the electrical transmission line as a function of H ext through a lock-in amplifier ( Fig. 1(b)). In the SP-FMR experimental configuration, the sample edges are connected to a nano-voltmeter with Ag wires soldered with Ag paint and a voltage signal (V mix ) is measured as a function of H ext (Fig. 1(c)).
In Fig. 1 where α represents the damping constant of the FM magnetization, γ the gyromagnetic ratio and ∆ 0 the inhomogeneous broadening. The latter parameter provides information about the magneto-structural quality of a FM film, and it is fundamental to confirm the reliability of the physical properties obtained by BFMR. 51 From the best-fit of the of experimental data to Eq.
(1), the damping parameter α for each Co thickness in both the Au(5nm)/Co(t)/Sb 2 Te 3 and Au(5nm)/Co(t)/Au(5nm)/Sb 2 Te 3 systems are extracted, and the values are plotted in Fig.   2(c) as a function of the inverse of the Co thickness (1/ ).
Typically, in the framework of the SP theory, 28,52 the α(1/ ) curve follows a linear trend as described by the first two terms on the right-hand side of Eq. (2), where α bulk represents the damping constant of the bulk material, the Bohr magneton, the saturation magnetization, g the g-factor, the thickness of the FM layer and Re(g e ↑ f ↓ f ) is the real part of the effective spin-mixing conductance. The latter quantity plays a central role in the description of the SP phenomena, being directly proportional to the spin current density generated in the FM layer and pumped into the adjacent non-magnetic material, here Sb 2 Te 3 , at resonance condition.
Clearly, the trend observed for the Au(5 nm)/Co(t)/Sb 2 Te 3 stacks (green data in Fig.   2(c)) does not follow a linear dependence in the whole thickness range. Indeed, by applying the conventional SP fitting model (first two terms in Eq. (2)), an α bulk = (5 ± 1) · 10 −3 is obtained, which is in disagreement with the (8 ÷ 11) · 10 −3 range expected for bulk Co. 49,53 Being ↑↓ a fundamental parameter to judge spin pumping functionalities, the observed nonlinearity in the Au(5 nm)/Co(t)/Sb 2 Te 3 system must be carefully addressed in order to avoid the extraction of unphysical ↑↓ values from BFMR experiments. 53 The non-linear α enhancement can origin from magneto-structural disorder in the Co thin films and/or at the Co/Sb2Te3 interface. Indeed, for the thinnest samples, the obtained inhomogeneous term ∆H 0 shows a slight enhancement when compared to the thicker samples, see Fig. 2 On the other hand, the XRR analysis (Supplementary Info. -Fig. S6) evidences a high chemical-structural quality of the Co layers, suggesting that the divergence observed in Fig. 2(c) for the Au/Co/Sb2Te3 set (green stars) likely has other origins. Actually, L. Zhu et al.(2019) 53 has recently reported and analyzed the BFMR response in several FM/Pt heterostructures, pointing out that, in the majority of the studied systems, the SP is a relatively minor contribution to α, when measured in the GHz frequency region. Indeed, they suggested that two further terms should be accounted to properly describe the α(1/t Co ) curve: Spin Memory Loss (SML) and Two-Magnon Scattering (TMS). SML is an interface effect manifesting with an additional linear contribution to that in Eq. (2). Due to SML, the spin current pumped from the precessing magnetization in a FM is partially suppressed at the interface with an adjacent layer, as a result of back-scattering. Recently, the main source of SML was attributed to the presence of an abrupt interruption (i.e. at the interface) between a FM and a material with high SOC, such as HM or TIs. 54 Differently, the TMS is an energy transfer mechanism between the FMR uniform precessional mode and degenerate spin waves. [55][56][57][58][59] As discussed in Refs. 57,60 , the source of the TMS is the presence of defects and imperfections at the surfaces and interfaces of FM thin films, which act as a source of scattering for the precessing magnetization. Indeed, the TMS is often related to the morphological and magnetic roughness at the FM/(HM or TIs) interface. According to Ref. 53 , the total damping can be seen as α = α bulk + α SP + α T MS , thus giving the full expression in Eq. (2), where β TMS is the TMS coefficient, proportional to ( ) 2 (with , as the interfacial magnetic anisotropy density and the saturation magnetization, respectively) and to the density of the magnetic defects at the FM/(HM or TIs) interface. 60  (2), we obtain α Bulk = (8.7± 0.9) · 10 −3 , g eff = (0.8 ± 1) · 10 19 m −2 and β T MS = (4.5 ± 0.9) · 10 −19 m −2 . The α Bulk value perfectly agrees with those expected for bulk Co, thus demonstrating how the inclusion of the TMS contribution is necessary to interpret our FMR data set over the whole range of thicknesses. Therefore, the adopted fitting strategy provides reliable ↑↓ values, which are comparable to those previously reported in FM/TIs systems ( Table 2). In Fig. 2 (Table 2).
If a FM thin film is in contact with a good spin sink (i.e. HM, TIs), the generation of pure spin currents from FM into HM or TIs, is associated with a high ↑↓ value. In principle, the insertion of an interlayer between FM and the non-magnetic layer, could lead to a reduction of SP depending on the spin diffusion length (λ s ) value characterizing the particular interlayer used. 63 On the other hand, in the case of TIs, the direct contact with magnetic materials could also have a detrimental effect on the TSS, 64  interface roughness has been shown to play a key role in the S2C conversion efficiency. 20,31,32,53,65,66 Therefore, choosing an appropriate interlayer and finding the best trade-off in maintaining the TIs' TSS while keeping an efficient spin transport across the FM/interlayer/TIs interface, is mandatory but also impressively challenging. By comparing our ↑↓ with other available results (Table 2), it can be concluded that there is certainly still some room to further enhance the spin mixing at the Co/Au/Sb2Te3 interface. A complete overview of different interlayer options to optimize the SP in Co/Sb2Te3-based systems is out of the scope of the present paper and may be the subject of future studies.

Spin pumping in Au/Co/Au/Sb2Te3 heterostructures
In a SP experiment a 3D spin current density 3 is generated at resonance in the Co layers, longitudinally injected into Sb2T3 across the Au interlayer, and detected through IP SP-FMR. 28,34,52,54,67,68 The general expression for 3 (in units of A/ 2 ) is given by Eq. (3).
where ℏ is the reduced Plank constant, the frequency of the RF-signal, the charge of the electron and ℎ the oscillating magnetic field generated by the GCPW.
Following the spin pumping into the Sb2Te3 layer, a charge current is generated in the Sb2Te3 layer and detected as a potential drop across the measured sample. 69,70 The electronic transport in our Sb2Te3 layers mainly occurs in 2D, as demonstrated by the MC measurements conducted before the Au/Co/(Au) deposition, and interpreted in the framework of the Hikami-Larkin-Nagaoka model (Supplementary Info. -Fig. S2). Therefore, the charge current density 2 that is generated by the 3 pumping, can be expressed with Eq. (4).

= (4)
where W is the width of the sample ( Fig. 1(c)), R is the sheet resistance as measured separately at four point in the Van der Pauw configuration in the same setup used for MC studies, and VSP is the voltage that is generated across the sample purely due to the SP from Co into the Sb2Te3 layer. The VSP is obtained from the generated transverse Vmix, being the quantity directly accessible in a SP-FMR experiment ( Fig. 1(c)). The first step is therefore to fit the detected Vmix with Eq.(5). 71,72 where VSym and VAsym are the symmetric and anti-symmetric Lorentzian functions, respectively, Hres is the value of the magnetic field at the resonance and ∆ is the half-width at half-maximum (HWHM). From the SP theory, 27,69,71,73 the symmetric Lorentzian extracted from the fit in Eq. (5) can be originated only from the SP contribution to the curve, and ideally = .
However, this term could also contain the thermal Seebeck effect, 44 and in order to extract the pure SP contribution, VSP is typically obtained through Eq. (6).
The so-called "spin rectification terms" contribute to the VAsym part, being originated from the anisotropic magnetoresistance and anomalous Hall effect in the Au/Co/Au trilayer. 69,[71][72][73] The adopted fitting procedure of the SP-FMR data are reported in the Supporting Information (Fig. S7) for an Au(5nm)/Co(20nm)/Au(5nm)/Sb2Te3 stack.
To assess the intrinsic role played by Sb2Te3 in boosting the S2C conversion efficiency, the new set of samples listed in Table 1  According to the SP theory 27 , by reversing the direction of the applied magnetic field, the DC voltage relative to the SP contribution must change sign. This is observed for all the samples in Table 1, with Fig. 3(b) showing the case of sample S1.  Table 1, and the extracted 2 (from Eq. (4)) are depicted in Fig. 3(d) and listed in Table 1. As expected, in our measured 2 there is a certain contribution from Au, as demonstrated by the different 2 detected in S2 and S3. Neverthless, the presence of Sb2Te3 in sample S1 provides a gigantic extra contribution to the S2C conversion, with a 250% enhancement when compared to the reference S2 sample.
The different 2 values obtained in samples S2 and S3 indicate that the spin current 3 is simultaneously pumped from Co in both the Au layers. Thus, most likely, in sample S1 the spin current pumped into the Au capping layer is reflected at the Au/air interface and then partially absorbed by the Sb2Te3 substrate. Considering that for Co and Au is ∼ 10 nm and ∼ 35 nm respectively, 69,76 a tentative sketch of the 3 scheme in S1, S2 and S3 is depicted in Fig. 4. Here, the 3 backflows at the Au/Sb 2 Te 3 and Au/Si(111) interfaces, are not considered. In the case of sample S3, the larger HS3 (175 ± 3 Oe) when compared to both S1 (86.5 ± 0.8 Oe) and S2 (75.5 ± 2.6 Oe), is attributed to the partial Co oxidation due to air exposure.
This induces additional structural and magnetic disorder that reflects into a higher magnetic damping.

Spin-to-charge conversion efficiency in Au/Co/Au/Sb2Te3 stacks
Our main interest is now to translate the observed additional giant 250% increase in the SP contribution due to Sb2Te3 (Fig. 3(c)), into S2C conversion efficiency. In the case of the 2D-type of conduction occurring in our epitaxial Sb2Te3 (Supplementary Info. -Fig. S2), the S2C conversion is dominated by the IEE 47 , and = 2 / 3 is the S2C conversion efficiency figure-of-merit. 29,35,77 In order to extract the pure contribution due to the Sb2Te3, the ↑↓ required to   ( In our opinion, the first approach (Eq. 7(•)) is the most accurate since  can be obtained from a linear fit of the FMR broadening change as a function of the resonance frequency, while the second approach (Eq. 7(••)) only considers the difference of the FMR broadening at a fixed frequency. On the other hand, the latter strategy is still at the basis of several reports about SP efficiency in FM/(HM,TIs) systems. 31,70,71,77,79 In fact, the FMR measurements have been typically conducted by adapting cavity electron paramagnetic resonance facilities, with a single RF excitation frequency. 51 It is also not uncommon to see reports of S2C efficiencies extracted from samples having a single FM thickness, and measurements based on a single frequency. 78,79 In the following, we extract by following both approaches. Figure 5(a) shows the evolution of the ( ) curves measured in S1 and S2 and fitted with the Kittel equation for the IP configuration, from which we obtain: 1 = 603 ± 46 3 , 1 = 2.64 ± 0.08 , 2 = 653 ± 29 3 , 1 = 2.20 ± 0.04. From the linear best-fit of the FMR signal linewidth as a function of the resonant frequency reported in Fig. 5(b), 1 = (25.5 ± 0.6) · 10 −3 and 2 = (20.3 ± 0.2) · 10 −3 are extracted. According to Eq. 7(•), these values give ,Sb2Te3 ↑↓ = 8.34 · 10 18 m -2 , which from Eq. (3) provides 3 − 2 3 = 6.4 · 10 5 A m -2 as the pure accumulation due to the presence of Sb2Te3 in S1. By considering the 2 measured for S1 (Table 1) Table 2 reports a collection of relevant ↑↓ and data as obtained by FMR-based methods for heterostructures including TIs, and a selection of HM. The different methods used to interpret the FMR data (Eq. 7(•) vs (••)) are also indicated, with the aim to highlight the need of a standardized procedure of data reporting. only to that reported for stanine 29 (Table 2). The lower limit ~0.28 is at least of the same order of magnitude (and often higher) of those observed in 3D-TIs produced by MBE or sputtering (Table 2), thus proving the suitability of MOCVD to produce highly performing 3D-TIs on large-area Si substrate. According to the obtained values, the system here presented may be of interest in the development of magnetoelectric spin-orbit logic devices. 86 The key to understand the origin for this very large S2C efficiency may lie in the heterostructures, where a "FeTe" type of bonding at the interface is highly favored. Being FeTe a paramagnetic compound, it could hinder any S2C conversion effect at the interface, or at least largely limit the efficiency of such conversion. As a matter of fact, this is one of the main motivations for our choice of a 5 nm Au buffer layer at the Co/Sb2Te3 interface. The Au interlayer efficiently suppresses several detrimental effects at the Co/Sb2Te3 interface, the main one being certainly the TMS (Fig. 2(c)).
The transport properties of the TSS for several free-standing TIs can be studied by different techniques such as angle-resolved photoemission spectroscopy and scanning tunnel microscopy (STM). 29,90 According to the calculation carried out by Fert and Zhang in Ref. 91

Large-area MOCVD-grown epitaxial Sb2Te3 thin films
Antimony Telluride (Sb2Te3) thin films growth is exploited by MOCVD with an AIXTRON 200/4 system, operating with ultra-high pure Nitrogen carrier gas and equipped with a cold wall horizontal deposition chamber, accommodating a 4'' IR-heated graphite susceptor (Fig. S1). Amongst the available antimony and telluride sources, antimony trichloride (SbCl3) and bis(trimethylsilyl)telluride (Te(SiMe3)2) are selected as MOCVD precursors, because their intrinsic chemical reactivity, unlike precursors such as the most commonly In order to obtain the best crystalline quality, the Sb 2 Te 3 films are subjected to two thermal processes. The

Magnetotransport measurements on Sb2Te3 thin films
Magnetoconductance (MC) measurements constitute a powerful tool for the investigation of the topological properties of a TI. The typical MC curve has a parabolic shape, but in specific materials, due to quantum effects dominating at low magnetic field, a deviation from the canonical parabola can be observed.
In particular, the latter phenomenon has been described by  from TSS is -1 if both the interfaces participate to transport and -0.5 if just one of the two surfaces is involved.
Chalcogenide based Tis, such as Sb2Te3, are also heavy materials, where the SOC is relevant, and the bulk states are relatively conductive. For this reason, separating the bulk and surface contribution to the WAL is challenging. To clarify the origin of WAL the magnetic field could be applied also in the plane of the sample, because in this configuration any MC contribution is attributed to bulk states. 3 Our results (not shown here) for α indicate that there exists a mixed contribution of WAL and WL and the measurement performed with the field applied in the film plane suggests that bulk states are not contributing to WAL, but just to WL. In this scenario, α = -0.25 is attributed to a combination of WAL, arising from the TSS, which would give a value of -0.5 and WL, arising from the bulk state, which tends to increase α.
If compared with the granular Sb2Te3 thin films studied by Cecchini et al. 4 , where a value of α = -0.01 at T= 5 K has been reported, the topological properties of the epitaxial Sb2Te3 thin films investigated in this manuscript are largely enhanced, demonstrating the effectiveness of the performed thermal treatments. 5 As discussed in the main text, despite the encouraging results already obtained in terms of spin to charge conversion, it could be possible to further suppress the bulk conductive states contribution, tuning the position of the Fermi level by doping as reported in Ref. 6 .

Home-made grounded coplanar waveguide
The BFMR measurements are performed using a home-made facility obtained from the customization of a Bruker ER-200 instrument, originally adopted for Electron Paramagnetic Resonance (EPR) measurements. The setup is composed by a broadband Anritsu-MG3694C power source (1-40 GHz), which is connected to a home-made grounded coplanar waveguide (GCPW), where the ferromagnetic sample is mounted in a flip-chip configuration (with the FM film close to the CPWG surface), with a 20µm thick mylar foil placed in between, to avoid the shortening of the conduction line. The GCPW is connected to a rectifying diode (Wiltron, Model 70KB50 (NEG), 1 -26.5 GHz, 20 dbm MAX) which converts the RF-signal into a continuous DC-current, in turn sent to a lock-in amplifier for the signal detection. The GCPW is the RF-component used to generate the oscillating h RF magnetic field. A GCPW consists essentially of a central conductor of width w s which carries the RF-current (signal line, S) and two ground planes (G) separated from the signal line by an air gap of thickness w sg (Fig. S3.).
In order to extract the h RF value produced by the GCPW for a fixed RF power, we model the GCPW geometry and calculate the h RF (z) function, where z is the height from the GCPW surface. In Fig. S4

BFMR measurements on the Au(5 nm)/Co(t)/Sb 2 Te 3 and Au(5 nm)/Co(t)/Au(5 nm)/Sb 2 Te 3 samples
In Fig. S5 (a,b) the evolution of the f res (H res ) plots as a function of the Co thickness for the Au(5 nm)/Co(t)/Sb 2 Te 3 stacks is shown. Here, for each Co thickness the acquired dataset (colored squares) is Figure S4: Calculation of the oscillating magnetic field hRF produced with the GCPW for a 73 Fig.3(a)), in this case, the f res (H res ) signal is acquired by Brillouin Light Scattering measurements, as discussed in the section below. In Fig.S5  For the Au/Co/Sb2Te3 stack, the value of the extracted Co g-factors varies with the Co thickness, but not with a clear trend. As pointed out in Ref. 7 , such variation can be attributed to both the difficulty to extract this value from an IP BFMR configuration, due to the non-linear dispersion of the f res (H res ) curve and to the possible modification of the properties of the Co interfaces. Nevertheless, the g-factor values are in the range of g Co = 2.37 − 2.64, which are typical for Co thin films. 8 The M ef f value for each sample and further relevant parameters are summarized in Table S5 below, along with the values of the nominal and real thicknesses of the Co thin films (measured by X-ray Reflectivity), the g-factor, the inhomogeneous broadening ∆H 0 and the damping constant α, as extracted from Fig.2 in the main text.
In Fig. S5(c,d) a complete BFMR study on the Au(5nm)/Co(t)/Au(5nm)/Sb2Te3 stacks is reported. The evolution of the data shows the high quality of the whole set, being in accordance with the FMR theory.
Moreover, as discussed in the main text, the parameters extracted from these measurements demonstrate the high magnetic quality and the thickness control of the investigated samples. From the analysis of the Kittel curve the M s = 921 ± 55 emu/cm 3 and K s = 0.58±0.18 erg/cm 2 values are extracted. These values are lower than those extracted for the Co samples directly in contact with the Sb 2 Te 3 layer. A possible reason can be attributed to the fcc crystalline structure of the Au substrate, which could promote the formation of a higher fraction of cubic crystalline grains in the polycrystalline film, as compared to the same Co deposition on top of the hexagonal Sb 2 Te 3, which typically develops a hexagonal-phase . 9,10 Indeed, as also reported in Ref. 8 , for bulk fcc-Co, M s ∼ 1100 emu/cm 3 , which is lower than in the hex-Co (M s ∼ 1400 emu/cm 3 ). On the other hand, the K s values are in accordance with previous studies on Au/Co/Au sandwiches, 11 suggesting that the Co magnetic moment remains close to the bulk value also for very thin Co thicknesses (down to 2.5 nm in this study). A confirmation of the homogeneity of the Co electronic structure over the whole range of thicknesses values is given by the poorly dispersed values for the g-factors, which are all close to g ∼ 2.5 (Table S6), compatible with typical values for Co thin films. 8,12 In Table S6 the parameters extracted from Fig. S5 (c,d) and Fig.2 in the main text are reported for each sample, besides the nominal thickness of the Co layer.  Fig. S5 and Fig.2 in the main text, and a comparison with real thicknesses extracted by XRR experiments.

Brillouin Light Scattering (BLS) measurements general details
The BLS is an optic technique that makes possible the detection of spin waves traveling within a ferromagnetic film. It is based on the inelastic scattering of monochromatic light from thermally excited spin waves where both energy and momentum are conserved. 14 In order to be used as a complementary analysis to the BFMR, the BLS experiments were conducted by focusing the laser beam at normal incidence upon the sample while the external magnetic field (Hres),applied parallel to the sample surface, was swept from 3500 to zero Oe (as shown in Figure S5). In such a configuration, BLS is totally equivalent to the BFMR adopted in this work. 13,14 Thanks to the high sensitivity of BLS technique, we were able to detect spin waves in ultra-thin ferromagnetic thin films (less than 2 nm) 15 , often too weak to be easily detected by a BFMR measurement.

X-Ray Reflectivity (XRR) measurements and summary of the main BFMR parameters for Au(5nm)/Co(t)/Sb2Te3 stacks
In Fig. S6 (left) the XRR measurements for the Au(5nm)/Co(t)/Sb2Te3 stacks are reported. As it can be observed, the XRR model fits almost perfectly the collected data for all the Co thicknesses, witnessing the reliability of the ferromagnet deposition process. For each layer composing the structure, the thickness, electronic density and roughness are summarized in the table reported in the left side of Fig. S6.  Fig.S5(c,d) and Fig. 2 in the main text.

SP-FMR fitting procedure and power dependence of the spin pumping signal
In order to test the reliability of the experimental setup and the fitting strategy, SP experiments are recorded on various samples with different Co thicknesses. As an example of the adopted fitting procedure, in Fig. S8(a) the SP signal for an Au(5nm)/Co(20nm)/Au(5nm)/Sb2Te3 stack. According to the SP theory 16 , the symmetric and anti-symmetric components of the SP signal virtue of that, the IEE signal VSym, should be linear in the RF power, which is consistent with the trend observed in Fig.S8.