Signature of gate-controlled magnetism and localization effects at Bi2Se3/EuS interface

Proximity of a topological insulator (TI) surface with a magnetic insulator (MI) can open an exchange gap at the Dirac point leading to exploration of surface quantum anomalous Hall effect. An important requirement to observe the above effect is to prevent the topological breakdown of the surface states (SSs) due to various interface coupling effects and to tune the Fermi level at the interface near the Dirac point. In this work, we demonstrate the growth of high-quality c-axis oriented strain-free layered films of TI, Bi2Se3, on amorphous SiO2 substrate in proximity to an MI, europium sulfide (EuS), that show stronger weak anti-localization response from the surface than previous studies with epitaxially interfaced heterostructures. Importantly, we find gate and magnetic field cooling modulated localization effects in the SSs, attributed to the position of interface Fermi level within the band gap that is also corroborated from our positron annihilation spectroscopy measurements. Furthermore, our experiments provide a direct evidence of gate-controlled enhanced interface magnetism in EuS arising from the carrier mediated Ruderman–Kittel–Kasuya–Yosida interactions across the Bi2Se3/EuS interface. These findings demonstrate the existence of complex interfacial phenomena affecting the localization response of the SSs that might be important in proximity engineering of the TI surface to observe surface quantum Hall effects.


INTRODUCTION
The process of breaking time reversal symmetry by introducing out-of-plane magnetism in the topologically protected surface states (SSs) of a 3-dimensional (3D) topological insulator (TI) 1,2 leads to opening of an exchange gap (EG) at the Dirac point 3 . Such a mechanism support several new quantum phenomena, such as quantum anomalous Hall effect (QAHE) 4,5 , half-integer quantum Hall effect (H-IQHE) 1,6 , topological magneto-electric effect 7,8 , and image magnetic monopoles 9 . In realizing these effects, the main challenge is to successfully demonstrate the presence of the EG with the Fermi level (E F ) lying within the gap. Currently, this is made possible in magnetic TIs such as in Cr-and V-doped compensated TIs, where perpendicular magnetism is developed in the bulk showing QAHE 5,10 . However, in comparison, inducing local magnetism at TI surface through the short range nature of magnetic proximity effect (MPE) can provide a number of advantages over bulk doping 3,5,11 ; these include lower bulk and surface defect density due to improved film crystallinity, better controllability of the SSs, and independently controlling the magnetic state at each surface of TI to build switchable topological devices 12 .
Conventionally, MPE can be achieved by using magnetic insulators (MIs). However, requirement of an out-of-plane magnetization at the TI surface makes this study difficult since most MIs naturally possess in-plane anisotropy. Also, new hybridization states 13 that may form at the TI/MI interface near the E F can contaminate the topological properties of the Dirac SSs [14][15][16] . As a result, observing these effects through transport studies can become nontrivial.
In this regard, MPE studies in the 3D TI, Bismuth Selenide (Bi 2 Se 3 ), with an MI, europium sulfide (EuS), is extensively researched [17][18][19] . Different research groups confirm the presence of an out-of-plane component or canted magnetization at the Bi 2 Se 3 /EuS interface with the magnetism existing at temperature much above the magnetic transition temperature (T c ) of EuS (~17 K) 18,20 . Our magnetometry studies on Bi 2 Se 3 (10 nm)/EuS (1 nm) also confirm the existence of enhanced interface magnetism with a second transition close to 232K (refer to Supplementary  Fig. 2 and Supplementary Note 2). Unlike other TI/MI interfaces 14,15,21,22 , recent theoretical calculation of the above interface also suggests induced magnetism, localized within the first quintuple layer (QL) of Bi 2 Se 3 23 , with the SS retaining most of the topological character compared to a pristine Bi 2 Se 3 SSs 20 . Furthermore, an induced EG of~9 meV in the surface Dirac cone is suggested 23 . Despite these reports, experimental observation of the EG opening at this surface has been elusive 24 . Possible reasons for the above include: (a) inherent intrinsic doping from large Se vacancies leading to bulk dominated conduction with E F pinning inside the conduction band (CB), and (b) loss of topological character/signal due to interface hybridization, surface-bulk coupling effects 25 , and/or due to reduction in thickness (in an effort to probe the surface contribution over the bulk conduction) that weaken the robustness of bulk bands to preserve the SSs.
In literature, the topological nature of SSs and its response to MPE are probed using magnetoconductance measurements by analyzing the weak anti-localization (WAL) or weak-localization (WL) signal from the spin-momentum locked SSs. In Bi 2 Se 3 , this signal is weak due to the dominant bulk contribution to conductivity, thereby making them harder to detect electronically. Reducing bulk conduction in Bi 2 Se 3 films is achieved either by studying thinner films, or by reducing the bulk conductivity. Previous studies with epitaxially interfaced Bi 2 Se 3 /EuS films 17 , including the study with thinner films (<8 nm), report anomalous Hall effect with no evidence of WAL or WL signal after EuS growth. Raman study of these interfaces also reveal strong charge transfer 26 that may suggest interface hybridization to weaken and destroy the topological signature of SSs. This may not be a surprise since lattice-matched interfaces can sometimes be detrimental for proximity engineering, as also observed in the study of proximity-induced superconductivity 27 .

Sample preparation
In this article, we adopt a different approach to reduce the bulk conductivity by growing out-of-plane textured Bi 2 Se 3 films on amorphous SiO 2 (a-SiO 2 ) substrate having a hexagonal surface atomic structure that is found essential for promoting the out-ofplane magnetization of EuS 18 . We achieve such a film order by depositing the films at 250°C in our molecular beam epitaxy system with a base pressure of 5 × 10 −10 mbar using a coevaporation technique. After deposition, films are cooled gradually under Se environment and a capping layer is grown at room temperature. X-ray diffraction (XRD) measurements in Fig. 1a reveal (003) family of peaks suggesting a highly c-axis oriented layered growth of Bi 2 Se 3 films that is found to be structurally as good as our epitaxially grown films on c-cut sapphire substrate. Cross-sectional transmission electron microscope (TEM) images ( Fig. 1c) also reveal a layered growth supporting the hexagonal structure of 0.95 nm thickness of the QL with no interfacial strain compared to the films grown on sapphire. Our reflective highenergy electron diffraction study suggested in-plane non-textured growth of the films. Similar growth findings were also noted in the work of Bansal et al. 28 . The structural phase of the Bi 2 Se 3 films was confirmed using Raman spectroscopy (see Supplementary Fig. 1 and Supplementary Note 1). For transport measurements, thicker Bi 2 Se 3 films in 10-16 nm range were grown and, subsequently, a 3 nm film of EuS was deposited in a sulfur-rich environment at room temperature, without breaking vacuum, followed by a 2-nmcapping layer of AlO x (x~1.3-1.5). Our XRD analysis showed no EuS peaks, suggesting its amorphous growth.
Metal-to-insulator transition (MIT) and weak anti-localization response Hall bar structures with length, L and width, W were patterned by mechanical etching or by in-situ shadow mask (Fig. 1d) as optical lithography techniques degraded the device mobility (see Fig. S3). Hall mobility (μ Hall ) of our 16 nm Bi 2 Se 3 films was found to bẽ 450 cm 2 /V s (inset of Fig. 2a). Figure 2a shows the temperature dependence of normalized longitudinal resistivity (ρ xx ) for 12 and 16 nm Bi 2 Se 3 films, grown under the same conditions, with two different capping layers, viz., 2 nm Se and 3 nm EuS. Both devices show a predominant metallic signature during cool down with the decrease in ρ xx for the EuS-capped devices being relatively gradual than the Se-capped devices, a trend observed in different samples with different W (200 μm to 2 mm). At lower temperatures, a weak upward trend in ρ xx is observed with a stronger  T for Se-capped (hollow symbols) and EuS-capped device (solid symbols) with 12 nm (triangles) and 16 nm (circles) thickness of Bi 2 Se 3 . Corresponding temperature value of the resistivity upturn is also indicated. Inset shows the temperature dependence of μ Hall of Se-and EuS-capped 16-nm-thick Bi 2 Se 3 devices. b Change in magnetoconductance, ΔG vs. B of Bi 2 Se 3 /Se (hollow square) and Bi 2 Se 3 /EuS (hollow circle) devices fitted with HLN formalism (solid line). c Temperature dependence of HLN fitting parameters,Ã (solid) and B ϕ (hollow) for Bi 2 Se 3 /Se (square) and Bi 2 Se 3 /EuS (circle) devices.
response in the EuS-capped devices. In the case of Se-capped films, the upturn in ρ xx occur at~18 K for 12 nm films and decrease to~15 K for 16 nm films. While, in the case of EuScapped films, a stronger upturn in ρ xx occurs at a temperature of 19 K, that remains unchanged in both the 12 and 16 nm Bi 2 Se 3 / EuS devices. In general, a weak upturn in film resistivity arises due to localization effects 29,30 in the 2D limit originating from the film disorder. However, observation of a stronger upturn in EuScapped devices at a higher temperature (near T c of EuS) than the reference films, while showing no dependence on the Bi 2 Se 3 film thickness, suggests an additional interface mechanism at play in our EuS-capped devices. Furthermore, our observation of similar μ Hall in the EuS-capped and Se-capped devices of Fig. 2a, also strengthen our above claim.
To analyze the influence of the above interface effect on the WAL response of the SSs, magnetoconductance (G(B), B is the magnetic field) study in the low-field range (B < 0.2 T, applied perpendicular to the film) was performed at low temperatures (<30 K). In Fig. 2b, a strong upward cusp in ΔG(B) = G(B) − G(0) is measured at zero field in both the EuS-capped and Se-capped devices with W/L~1, providing a clear signature of the presence of 2D WAL channel in our films. The above magnetoconductance plots are fitted with the Hikami-Larkin-Nagaoka (HLN) formalism 31 , given by ΔGðBÞ ¼Ãe 2 =2πh½lnðB ϕ =BÞ À ψð1=2 þ B ϕ =BÞ, where ψ(x) is the digamma function, e is electronic charge, and h is the Planck's constant. The temperature dependency of the fitting parameters, dephasing field (B ϕ ) and prefactorÃ, that denote the effective number of 2D conducting channels contributing to WAL is plotted in Fig. 2c. In this low-field range, the contribution of bulk states to the WAL response, determined by performing in-plane magnetoconductance measurements 32 , was found to be negligible without significantly affecting the parameters extracted from the HLN fitting. In TI, we have two 2D WAL channels corresponding to the bottom and top SSs. Proximity of one of the surface (top) with MI, which theoretically opens an exchange gap, can considerably weaken its WAL response. For the case of Se-capped devices, we findÃ $ 1:6, which indicates good decoupling between the top and bottom 2D WAL channels. However, in the EuS-capped devices, we notice a drop in WAL channel by~0.8 at 1.5 K suggesting additional localization response due to EuS capping (see Fig. 2c).
Our above findings of localization at the top Bi 2 Se 3 /EuS interface can arise due to different possible reasons. To rule out its origin to any trivial response from the EuS layer, temperature dependence of resistivity of a trivial metal, Pt (5 nm), was measured that showed no change in behavior due to its proximity with EuS (2 nm) (see Supplementary Fig. 11). These observations limit our explanation for the origin of the above response to possible electron-electron interactions (EEI) 33 or MPE-induced EG opening 19 . In the case of EEI, the large g-factor of the SSs can activate a large Zeeman splitting from the exchange field of EuS and contribute to corrections in the conductivity. In the following discussions, we refer to the above localization in EuS-capped films, causing an upturn in film resistivity, to an MIT response of the SSs.
Interface band-bending effects In TIs, the position of E F at the surface can greatly depend on the relative concentration of bulk and surface defect density 29,34 . A large mismatch in the surface and bulk E F can lead to charge transfer and bending of bulk bands near the surface, resulting in a depleted (an accumulated) surface region in films with lower (higher) surface defect density 35,36 (refer to Supplementary Note 4). Such surface band-bending response, creating interfacial electrostatic fields that shift the Dirac cone and position of the E F at the interface, has been observed using angle-resolved photoemission spectroscopy (ARPES) 37 and gated terahertz cyclotron resonance measurements 38 . In our films, the high bulk carrier density (>10 19 /cm 3 ) is expected to pin the bulk E F in the CB. However, understanding surface band-bending effects can help locate the position of E F at the interface and to qualitatively probe the origin of the above interfacial MIT response.
In our above effort, we performed depth-resolved positron annihilation spectroscopy (PAS) measurements at room temperature using variable low-energy positron beam that allowed profiling the vacancy-type defects 39,40 , i.e., primarily the Se vacancies (see "Methods"). From the Doppler-broadened positron annihilation spectrum, S-parameter, defined as the ratio of the counts in 511 ± 1 keV peak region to the total counts in 511 ± 10 keV region, is deduced. The S-parameter characterizes the annihilation of positrons with low-momentum valence electrons, which is directly related to the concentration of vacancy defects. In Fig. 3a, we plot the S-parameter, extracted from the positron annihilation spectrum, vs. positron beam energy (E p )/mean implantation depth for the reference Bi 2 Se 3 and the Bi 2 Se 3 /EuS films having a capping layer of AlO x . Here, the effect of variation in the capping layer thickness for the two films has been included in the PAS analysis (see Supplementary Note 7). We observe that for very low positron energies, the S-parameter is higher for the reference Bi 2 Se 3 films than the EuS-capped films. As the energy increases beyond~1.25 keV, corresponding to an implantation depth of~9 nm, the S-parameter for the two samples merge suggesting similar bulk response of the Bi 2 Se 3 films (see  Supplementary Fig. 6). The plots are shifted vertically for clarity. Supplementary Note 7 for detailed analysis). Variable energy positron fit (VEPFIT) software analysis 41 was carried out by assuming a two-layer fitting, comprising of the Bi 2 Se 3 film and the underneath SiO 2 substrate layer (see Supplementary Note 7). The above analysis to the experimental S vs. E p curve yielded best fits by invoking a mean internal electric field in the Bi 2 Se 3 film within the depth of 9 nm from the top TI surface. In the reference Bi 2 Se 3 samples, a large positive interfacial electric field of 6.1 × 10 −4 V/nm was extracted supporting the formation of a downward band bending (top right inset of Fig. 3a), as also observed in ARPES study of Bi 2 Se 3 surfaces 37 . On the contrary, in the EuScapped films, an order of magnitude drop in the mean electric field is observed (~0.9 × 10 −4 V/nm). Considering the interfacial electric field to be proportional to the carrier density, we find that the surface carrier density at the Bi 2 Se 3 /EuS interface (compared to bulk value of~4 × 10 13 /cm 2 ) can fall by nearly an order of magnitude to~10 12 /cm 2 . These interpretations are in agreement with density functional theory (DFT) studies that reveal Eu-Se bond formation at the Bi 2 Se 3 /EuS interface to be more favorable 23 leading to a lower surface defect density. Such a condition support a nearly flat-band condition moving the interface E F closer to the band edge (bottom inset of Fig. 3a, also see Supplementary Note 4).
In Fig. 3b, we show temperature dependence of longitudinal resistance (R xx ) vs. back-gate voltage (V g ) of the Bi 2 Se 3 /EuS devices. Interestingly, a hysteresis response in R xx is observed at all temperatures below 125 K, with similar response observed in four different sets of samples and with different film thickness (see Supplementary Note 6 and Supplementary Fig. 6). In comparison, the Se-capped devices showed no such behavior (see Supplementary Note 5 and Supplementary Fig. 5). Presence of such hysteresis response observed only in Bi 2 Se 3 /EuS devices at low temperatures further supports our PAS analysis that the interface E F lies near the band edge and drops into the band gap with cool down due to freezing of charge carriers. With the position of interface E F in the band gap, V g modulation leads to pinning/depinning effects at the band edge, responsible for the observed hysteresis in Fig. 3b. These findings may therefore suggest the origin of our interfacial MIT response to the localization affects around the Dirac point of the top SSs.
Effect of gate voltage on interface magnetism The MIT response in our Bi 2 Se 3 /EuS devices was further investigated in Fig. 4a by measuring R xx vs. T at different V g s. Here, at negative V g , an insulating transition was observed at temperature, T MIT~1 9 K, which was enhanced to~32 K at positive V g with a corresponding increase in the carrier density (top left inset of Fig. 4a). In comparison, the Se-capped devices showed no such response (Fig. 4b). Such an increase in T MIT at positive V g , which also makes the device conducting (seen by the downward shift of plots), at first seems counter-intuitive. However, this observation may be suggestive of conduction-electron-mediated ferromagnetic exchange between the Eu atoms strengthening the interface magnetism, with an enhancement in T c of the EuS "interface layer". This enhanced interface magnetism in our devices is resulting in a larger increase in resistance below T MIT , peaking at V g~5 0 V (bottom-right inset of Fig. 4a). Comparing with the warm-up data of R xx vs. T (Fig. 4c), we observe a clear hysteresis in the low-and high-field range that supports the existence of different magnetic transition temperature of bulk (T bulk c ) and interface (T interface c ) EuS layer. This feature is also revealed in resistance plots (V g > 15V) in Fig. 4a with a point of inflection at 19 K, below which, the magnetic transition of EuS layers away from the interface causes the resistance to increases further. Similar increase in T c of EuS, however, due to bulk electron doping has been previously reported 42,43 . It is interesting to note that at large negative V g (depleted interface), we observe a higher value of T MIT~1 9 K, i.e., 2 K above the bulk T c of EuS (17 K) which may indicate the role of high spin-orbit strength of the Bi 2 Se 3 SSs 20 .
In EuS, magnetism arises from the overlap of the Eu 4f orbitals with the valence band below the E F that leads to a weaker exchange coupling 44 . DFT studies by Kim et al. 20 show that the SSs that weakly interconnect the conduction and valance bands can mediate exchange coupling between Eu atoms through the Ruderman-Kittel-Kasuya-Yosida (RKKY) type indirect interactions [45][46][47] , with the exchange constant being sensitive to the position of E F near the Dirac point. The observed gate-controlled tuning of T MIT in our devices provides evidence for such an indirect exchange mechanism existing at the Bi 2 Se 3 /EuS interface. Also, cooling the device in perpendicular magnetic fields (H FC ), at V g = 50 V showed a surprising decrease in T MIT (or T interface c ), with the increasing strength of H FC (see Fig. 5). Such a response may arise because, here, the external magnetic field interacts not only with the localized 4f spins of Eu by direct interaction, but also through the conduction electrons at the interface. Such a theoretical framework of RKKY interactions that include the sign and value of the interaction potential between the localized spins and the polarized conduction electrons in external magnetic field has not been extensively explored 48 . In addition, EEIs in presence of Zeeman field can also greatly affect the above mentioned indirect interactions causing a reduction in T interface c 49 . Gate voltage dependence of Hall signal Apart from the enhancement in T MIT for V g > 15 V in Fig. 4a, we also observe a prominent and reproducible drop in device resistance preceding MIT. This sudden drop in resistance shows a maximum of~10 Ω at V g = 50 V. Also, when cooled in higher magnetic fields, the resistance drop preceding T MIT starts to vanish (Fig. 5). Although the increase in R xx due to magnetic proximity is in agreement with the localization effects observed from our WAL studies, the concomitant drop in R xx at the onset of interface magnetism is not well explained based on any spin disorder or EEI models supporting WL mechanisms 16,33,50 . In Fig. 6, we plot Hall resistance (R xy ) vs. V g of the device, measured at 1.5 K, that showed a hysteresis response with the value rising above V g~1 5 V in the positive sweep cycle reaching~70 mΩ and dropping down at −27 V in the negative sweep cycle of V g . Comparing with the data of Fig. 4a where the resistance drop in R xx preceding T MIT is observed only for V g > 15 V, we notice an increase in R xy in the same V g window (during the positive V g sweep cycle) in Fig. 6 that might suggest a common phenomenon responsible for the above two responses. We have analyzed the possibility to associate the above origin to a weak signature of H-IQHE driven by the movement of E F in the EG (see Supplementary Note 8). However, the weak experimental signal, possibly due to the large bulk conductivity, canted magnetism of EuS and finite temperature makes this analysis inconclusive and open for future exploration.
In conclusion, we report favorable growth conditions of EuSproximitized Bi 2 Se 3 films on a-SiO 2 substrate providing evidence of lower vacancy-type surface defect density that support the movement of the interface E F toward the Dirac point. Also, we observe localization response in the proximitized SSs below the T c of EuS with a possible topological origin. Most importantly, we demonstrate the possibility to modulate interface magnetism using V g , driven by an indirect RKKY interaction at the Bi 2 Se 3 /EuS interface. These findings offer us the possibility to design a more robust surface quantum Hall system in our quest to develop switchable topological electronic devices.

METHODS
Growth, device fabrication, and transport measurement Film growth was performed in a custom designed molecular beam epitaxy system with a base pressure of 5 × 10 −10 mbar with all crucial interfaces prepared without breaking vacuum. Hall bar devices were mechanically patterned under a microscope and contacts were made using wire bonding on a 44 leadless chip carrier (refer to Supplementary Note 3). In addition, Hall bar geometry devices were also fabricated using in-situ shadow mask technique. Low-temperature magneto-transport measurements were performed in a variable temperature insert cryostat with a base temperature of 1.5 K.

Positron annihilation spectroscopy
Depth-resolved Doppler broadening measurements were carried out on the samples using the variable low-energy positron beam. Using this approach, vacancy-type defects were studied in our films starting from the surface up to a depth of a few hundreds of nm by implanting positron beam of varying energy ranging from 235 eV to 2.735 keV in steps of 0.1 keV. The mean implantation depth follows an empirical relation: z (nm) ¼ ð40 E 1:6 p Þ=ρ where E p is the positron beam energy in keV and ρ is the density of the sample in g/cm 3 . For the PAS study, Doppler broadened 511 keV annihilation energy spectrum was obtained for each positron beam energy using HPGe detector having an energy resolution of 1.5 keV for 662 keV γ-ray of 137 Cs. From each annihilation spectrum, the S-parameter was deduced.

DATA AVAILABILITY
The data that support the findings of this study are available from the corresponding author upon reasonable request.