Absence of magnetic-proximity effect at the interface of Bi$_2$Se$_3$ and (Bi,Sb)$_2$Te$_3$ with EuS

We performed x-ray magnetic circular dichroism (XMCD) measurements on heterostructures comprising topological insulators (TIs) of the (Bi,Sb)$_2$(Se,Te)$_3$ family and the magnetic insulator EuS. XMCD measurements allow us to investigate element-selective magnetic proximity effects at the very TI/EuS interface. A systematic analysis reveals that there is neither significant induced magnetism within the TI nor an enhancement of the Eu magnetic moment at such interface. The induced magnetic moments in Bi, Sb, Te, and Se sites are lower than the estimated detection limit of the XMCD measurements of $\sim\!10^{-3}$ $\mu_\mathrm{B}$/at.

The observation of the quantum anomalous Hall effect (QAHE) in magnetically-doped topological insulators (TI) has raised significant interest for applications in metrology [1,2]. Test measurements demonstrating QAHE quantization accuracy at a record level of better than 1 part-per-million have recently been reported [3]. However, the magnetic impurities are incorporated randomly in the TI lattice and the resulting disorder presumably reduces the temperature and currents at which the QAHE fully develops. It also leads to the observation of superparamagnetic phases and multidomain magnetization switching, which coincide with the QAHE and are not well understood [4]. In this context, imprinting magnetism by proximity effects in thin film heterostructures appears as a promising alternative [5][6][7][8] that could ensure the preservation of the TI crystalline quality and its bulk insulating properties. Magnetic proximity effects in heterostructures comprising TIs and magnetic insulators (MI) have been intensively investigated using optical methods, electronic transport measurements, polarized neutron reflectivity, and muon spin spectroscopy [6][7][8][9][10][11][12][13][14][15][16].
However, while the number of articles reporting signatures of ferromagnetic behavior in TI/EuS interfaces keeps on growing, topological phenomena have yet to be observed. QAHE has been systematically demonstrated in Cr-and V-doped (Bi,Sb) 2 Te 3 [1,2] despite that the ferromagnetic state appears to be much weaker than in Bi 2 Se 3 /EuS. In addition, first-principles calculations of the electronic band structure and magnetic ordering of the Bi 2 Se 3 /EuS interface did not find an induced magnetic moment in the TI or even a significant enhancement of the local magnetic moment of Eu [17,18]. The discrepancy between experiments and theoretical analysis in combination with the absence of direct experimental proof of locally induced magnetism in the TI underscores the need of further experimental investigations.
Electronic and magnetic information regarding microscopic interactions can be obtained using x-ray absorption spectroscopy (XAS) and x-ray magnetic circular dichroism (XMCD). These techniques have been successfully implemented to study magnetically-doped TIs [19][20][21][22][23][24][25][26][27] and-most importantly-are able to detect elementspecific magnetic moments that are induced in the host TI lattice [25,27]. In this Letter, we report XAS and XMCD measurements on TI/EuS interfaces, with TIs of the (Bi,Sb) 2 (Se,Te) 3 family. In contrast to previous investigations, our systematic study provides fundamental local and element-selective information of the magnetic moments associated to Eu, Bi, Sb, Se, and Te atoms. In agreement with theoretical reports, we find no indication of proximity-induced magnetism in the TI. This suggests that the magnetic signatures in TI/MI heterostructures which have been reported do not originate from induced ferromagnetic order in the TI atoms at the TI/MI interface.
The TI/EuS heterostructures were grown on singlecrystal BaF 2 (111) substrates by molecular beam epitaxy under ultrahigh-vacuum (UHV) with a base pressure ∼ 1 × 10 −10 Torr. Bi 2 Se 3 , Bi 2 Te 3 , Sb 2 Te 3 , and (Bi,Sb) 2 Te 3 thin films with a thickness of 10 nm were grown by coevaporation of elemental Bi, Sb, Se, and Te (6N purity), following the procedure in Ref. 28. After growing the TI layer, the samples were transferred in UHV to a second chamber where a layer of 5-nm thick EuS was deposited at room temperature using an electron-beam evaporator, as in Ref. 7. A protective 2-nm thick Al capping layer was subsequently deposited in the same chamber and allowed to oxidize in air. Additional Sb 2 Te 3 /EuS, Sb 2 Te 3 /Bi 2 Se 3 /EuS, and Bi 2 Se 3 /EuS(x ) with x = 1 and 5 nm samples were capped with a 20-nm thick Se layer grown in-situ, which was desorbed in the UHV preparation chamber for XAS measurements by heating the sample to ∼ 180 • C. The structural and crystal quality of the films was assessed using in-situ reflection high energy electron diffraction and ex-situ x-ray diffraction and xray photoelectron spectroscopy (XPS) revealing epitaxial growth and sharp interfaces [29]. The macroscopic magnetic properties were studied by SQUID magnetometry [29].
XAS was performed using a 6-T magnet on beamline BL-29 (BOREAS) at the ALBA synchrotron (Spain), which provides a UHV sample environment with a base temperature of ∼3 K. Measurements were focused on the M 4,5 edges of Eu (1110-1170 eV), Te (570-590 eV), Bi (2560-2700 eV), and Sb (525-545 eV), and L 2,3 edges of Se (1420-1460 eV). The XMCD signal was obtained by subtracting XAS spectra with the photon helicity vector antiparallel and parallel to the applied magnetic field. Measurements used total-electron-yield (TEY) detection, where the drain current was taken from the sample to the ground. The magnetic field was applied along the x-ray beam for two different geometries: normal incidence (perpendicular to the sample plane) and grazing incidence (30 • off the sample plane). Figure 1 depicts XAS and XMCD spectra at the Eu M 4,5 edges for grazing incidence at 3 K under 2 T magnetic field. The M 5 edge peak maximum intensity is normalized to unity in order to compare the spectra from different samples. XMCD is expressed in percentage of the average XAS. The lineshape and amplitude of the normalized XAS and XMCD signals for all TI/EuS samples [ Fig. 1(a) and (d)] are nearly identical, which indicates the same electronic and magnetic state of the Eu atoms. The Eu M 4,5 edges correspond to 3d → 4f electronic transitions. In order to identify the Eu electronic state, the results are compared with the calculated spectra for Eu 2+ 4f 7 ( Fig. 1(a) and Supplementary Material [29]) and the measured spectra in two reference samples:  features (peaks B and D) for Eu 2+ , which is consistent with the expected electronic state for ferromagnetic EuS.
Weak features corresponding to Eu 3+ (peaks A, C, and E) are also visible, revealing some degree of Eu oxidation (Eu 2 O 3 phase) at the surface of the sample, which is also observed and estimated to be between 0.35-0.47 nm thick by XPS [29]. Eu 2 O 3 is not ferromagnetic, therefore it does not contribute to the XMCD signal [30]. XMCD is a powerful tool that yields the spin and orbital moments of the probed elements through the application of the magneto-optical sum rules [31,32]. For the rare earths, however, the sum rule analysis is less straightforward [33,34]. Since the XMCD intensity scales with the magnetization [34], values for the Eu 4f magnetic moments can be directly obtained by comparing the M 5 XMCD asymmetry with the expected theoretical value. This asymmetry A is the maximum value of (XMCD)/(XAS sum), where XMCD and XAS are the difference and the sum, respectively, of the right-and left-circularly polarized XAS spectra at the Eu M 5 edge. Table I lists the magnetic moments derived from A for each sample at 3 K (∼10 K for Bi 2 Se 3 /EuS(x ) with x = 1 and 5 nm samples). The calculated XMCD spectrum in Fig. 1(d) yields the theoretical asymmetry A theory = −0.497, which corresponds to the Hund's rule ground state (GS) value for Eu 2+ 4f 7 (S = 7/2, L = 0, J = 7/2), with an effective magnetic moment µ GS eff = 7.93 µ B . Because L = 0, this is a pure spin moment. Table I shows a reduction of µ eff of about 30-35% for all samples with respect to µ GS eff , which could be due to the presence of a small crystal field acting on the Eu 4f electrons, as observed for other rare earths [35]. These values are well in agreement with the magnetic moments of Eu in Bi 2 Se 3 /EuS calculated by density functional theory [18], also included in Table I. No significant enhancement in the Eu 4f magnetic moments of our TI/EuS samples is found compared to that of the isolated EuS layer. This observation is consistent with theoretical reports [17,18], but contrary to experimental findings using polarized neutron reflectometry [7].
Magnetic hysteresis loops were recorded by fixing the energy at the maximum of the Eu M 5 XMCD signal (1125.4 eV) while sweeping the field. Figure 2 shows loops measured for Bi 2 Te 3 /EuS, Sb 2 Te 3 /EuS/Se and Bi 2 Se 3 /EuS in grazing incidence. Values of A at saturation are very similar in all cases. This is consistent with Eu atoms having the same electronic and magnetic state regardless of the TI they are in contact with. The square shape of these hysteresis loops demonstrates that Eu 4f states are ferromagnetic at low temperatures (3 K) but they become paramagnetic at 20 K. This is in agreement with results of SQUID magnetometry, which indicate ferromagnetic ordering below T C ≈ 15 K [29]. Therefore, we do not observe signs of room-temperature ferromagnetic behavior in these TI/EuS systems, in contrast to previous reports [7].
The magnetic anisotropy of the Eu layer in the TI/EuS system was analyzed by comparing the XMCD spectra recorded in normal and grazing geometries at 2 T and at remanence (see Fig. S5 in Supplemental Material [29]). At 2 T, curves for normal and grazing geometries overlap for all TI/EuS samples. Without magnetic field, we observe remanence only in samples with the TI underlayer; the isolated EuS layer shows no XMCD signal at zero field. This behavior can be ascribed to the large spinorbit interaction that is intrinsic to the TI [18]. Moreover, there is a significantly larger remanence in the grazing geometry for all TI/EuS samples, which is indicative of a favorable magnetization axis in the plane of the sample. In-plane anisotropy is also observed macroscopically by SQUID magnetometry (see Fig. S4 in Supplemental Material [29]). Previous experiments on Bi 2 Se 3 /EuS demonstrated perpendicular anisotropy in EuS [7]. Theoretical calculations have shown that topological surface states and their large intrinsic spin-orbit coupling may explain this observation [18]. However, the orientation of the magnetization depends strongly on the strain at the interface. When the EuS lattice is relaxed, as is the case for our TI/EuS films (see Fig. S1 in Supplemental Material [29]), it is expected that the magnetization remains in plane [18].
Next we will focus on the magnetic proximity effects. XMCD measurements at accessible absorption edges of Bi, Te, Sb, and Se elements were performed in search of induced magnetism on these non-magnetic atoms. Figure  3 depicts XAS and XMCD spectra at the Bi M 4,5 and Se L 3 edges for Bi 2 Se 3 /EuS and at the Sb and Te M 5 edges for Sb 2 Te 3 /EuS. No XMCD signal is apparent in any of these absorption edges probed, remaining within the noise level.
TEY-detected XAS for soft x rays is a surface sensitive technique with the majority of electrons originating from within 3-5 nm [34,36]. This implies that the XAS signal not only arises from the uppermost region of the TI, which is in contact with EuS, but also from the bulk of the film with a contribution that decreases exponentially with depth. In order to verify the absence of proximityinduced magnetism, we performed XMCD measurements at the Se L 3 and Bi M 4,5 edges in a Sb 2 Te 3 /Bi 2 Se 3 /EuS heterostructure. The Bi 2 Se 3 layer is 1 nm thick, ensuring that the recorded XAS signal originates from the very Bi 2 Se 3 /EuS interface. XMCD at the Se L 3 for this sample is plotted in Fig. 3(b) (the Bi signal on this sample, not shown, is very weak). In agreement with the results on the Bi 2 Se 3 /EuS, no XMCD signal is found at this absorption edge.
Non-negligible Te and Sb XMCD signals have been re-ported for thin films of Cr-doped Sb 2 Te 3 [26,37] and bulk Cr-, V-, and Mn-doped Sb 2 Te 3 [26,37] and Crdoped (Bi,Sb) 2 Te 3 [25,38]. In all these cases, XMCD data was recorded in TEY for in-situ cleaved samples. TEY detected XAS and XMCD signal for our TI/EuS heterostructures have poor statistics compared to those reported for in-situ cleaved samples due to the 5 nm EuS and 2 nm Al on top of the TI. Despite the noise level of the data, application of the magneto-optical sum rules [31,32] to the XAS and XMCD signal in Fig. 3 can provide an upper limit for the induced magnetic moment on Bi, Sb, Se, and Te atoms in contact with EuS [29].
We found values of the spin moment, S z ≈ 10 −3 µ B per Bi and Se atom in the Bi 2 Se 3 /EuS samples and ∼ 10 −3 µ B per Sb and ∼ 10 −4 µ B per Te in the Sb 2 Te 3 /EuS/Se sample.
The negligible induced magnetic moment on Bi, Sb, Te, and Se atoms of (Bi,Sb) 2 (Se,Te) 3 in contact with EuS is a surprising finding. Recent observations suggest the presence of magnetism within the TI layer of Bi 2 Se 3 /EuS by magnetotransport measurements [9,39], polarized neutron reflectometry [7,11], magnetic secondharmonic generation [10] and low-energy muon spin rotation [12]. Katmis et al. [7] reported values of magnetization in the TI for a Bi 2 Se 3 (20 QL)/EuS(5 nm) sample between 250 and 300 emu/cm 3 , which correspond to a magnetic moment between 4.4 and 5.3 µ B per QL in a unit cell of Bi 2 Se 3 . Assuming that Bi and Se atoms become equally magnetic, and that magnetism develops over a unit cell closest to EuS, the magnetic moment amounts to ∼0.88-1.06 µ B per (Bi or Se) atom, which is about 2 to 3 orders of magnitude larger than those estimated for our Bi 2 Se 3 /EuS samples from XMCD measurements.
XMCD measurements have already been used to rule out reported proximity effects using other (non-element selective) experimental techniques. For example, the anisotropic magnetoresistance at Y 3 Fe 5 O 12 /Pt and ferrite/Pt interfaces was attributed to Pt atoms polarized by proximity effects. However, XMCD measurements at the Pt edges have shown no evidence of induced magnetism [40][41][42]. Theoretical studies using first-principles calculations show insignificant or no magnetism in the TI for Bi 2 Se 3 /EuS with ideally sharp interfaces [17,18,43], which is consistent with our XMCD results. Instead, Eu doping on Bi 2 Se 3 could result in local magnetic moments that are much larger than those induced by the adjacent magnetic layer, possibly explaining the experimental observations [17]. One Eu atom diffusing into the TI per unit cell would account for an observed magnetic moment of ∼1 µ B /at in a QL [17]. Intercalation of a single EuS layer has been theoretically predicted to be energetically favorable and leads to the observation of magnetism within the TI and, perhaps, of proximity effects, by suppressing the formation of trivial states [44]. However, our data does not show evidence of such an intercalation. Other interfacial effects, including charge transfer at the TI/MI interface and band bending could explain the absence of induced magnetism in the TI and the lack of observation of the QAHE [44][45][46][47][48]. Another possibility for the absence of proximity effects would be the presence of a magnetic dead layer at the TI/EuS interface. However, XAS and XMCD measurements of ultrathin EuS layers (≤ 1 nm) grown onto Bi 2 Se 3 reveal that EuS remains magnetic, which allowed us to exclude such scenario in our TI/EuS samples [29]. This conclusion is further supported by the increase of the in-plane magnetic remanence for the TI/EuS structures, compared to the isolated EuS layer, which can be ascribed to the large intrinsic spin-orbit interaction of the TI at a sharp TI/EuS interface [18].
In summary, we have performed element-selective magnetometry to study proximity effects at the interface of TI/EuS bilayers. A variety of TIs of the (Bi,Sb) 2 (Se,Te) 3 family were systematically investigated.
The easymagnetization axis of EuS remains in the plane of the sample and the magnetic remanence increases as a result of its contact with the TI. However, we found no evidence for an enhancement of Eu magnetic moments in the EuS layer, or an increase in the ferromagnetic ordering temperature. The magnetic signals on Bi, Sb, Te, and Se atoms of the TIs in contact with EuS were found to be negligible (below the detection limit of the XMCD technique) at temperatures and conditions where the EuS is magnetic and could polarize these non-magnetic atoms. Overall our XMCD measurements suggest that the observations in TI/EuS interfaces by magnetotransport and optical techniques are not due to proximity-induced magnetism but to a different mechanism, such as magnetic doping by Eu diffusion into the TI.