Tailoring Ultrathin Magnetic Multilayers at Terraced Topologically Insulating Interfaces for Perpendicularly Magnetized Domains Benjamin A. Brereton, Soumyarup Hait, Ahmet Yagmur, Christy J. Kinane, Francesco Maccherozzi, Michele Conroy, Satoshi Sasaki, Thomas A. Moore, Sarnjeet S. Dhesi, Sean Langridge, Christopher H. Marrows Abstract Topological insulators and skyrmion-hosting, chiral magnetic multilayers are two well-explored areas of modern condensed matter physics, each offering unique advantages for spintronics applications. In this paper, we demonstrate the optimization process for the growth of a Bi 2 Se 3 /buffer/[Pt/CoB/Ru] ×N heterostructure that combines these two material classes: the Bi 2 Se 3 epilayer was grown by molecular beam epitaxy before transfer under ultrahigh vacuum to a separate growth chamber where the polycrystalline metallic multilayer was sputter deposited. The structure of the samples was characterized by co-fitted X-ray and polarized neutron reflectometry measurements and scanning transmission electron microscopy. Polarized neutron models and standard magnetometry show that a buffer layer exceeding a critical thickness is required to obtain the desired uniform, perpendicular magnetic anisotropy in every magnetic layer in the multilayer. Samples with both Ta and Mo buffers were used requiring thicknesses of 1.5 and 0.9 nm respectively. In minimizing the Bi 2 Se 3 terracing, buffered samples yield well-defined, out-of-plane, magnetic domains suitable for spin-orbit torque induced manipulation as determined by X-ray photoemission electron microscopy. I Introduction Topological insulators (TIs) have long been of interest in next-generation spintronic devices because of their highly efficient charge-to-spin conversion, analogous to the spin Hall angles of heavy metals: \theta_{\text{SH}}=J_{\mathrm{s}}/J_{\mathrm{c}} . In such systems \theta_{\text{SH}} can be much greater than unity 20 ; 6 , arising from spin-momentum locked surface states. Such devices utilize the large spin-orbit torques (SOTs) generated by this effect, and often are based on switching thin ferromagnetic films with perpendicular magnetic anisotropy (PMA) for low-current magnetic random access memory (MRAM) applications 42 ; 7 ; 32 ; 11 . Magnetic multilayers (MMLs) comprising ultrathin layers of magnetic and heavy metals, are known to host multitudes of exotic, chiral spin textures in systems of broken inversion symmetry due to a non-vanishing interfacial Dzyaloshinskii–Moriya interaction (DMI): most notably Néel-type magnetic skyrmions 13 ; 21 ; 34 ; 4 ; 29 ; 15 . Skyrmions are topologically non-trivial with a resultant quantized Berry phase for transport electrons: their topology granting them relative stability against perturbations and admitting SOT-driven motion and manipulation from spin currents such as those generated by the spin-Hall effect (SHE) in heavy metals 10 ; 34 ; 28 ; 16 ; 38 ; 3 . These properties have not gone overlooked for potential applications in energy efficient data storage and computation applications: most recently in reservoir computing 8 ; 9 ; 26 ; 36 . TI/chiral magnetic multilayer heterostructures present an opportunity to utilise the strengths of both materials to further reduce necessary current densities and enhance energy efficiency. Often chiral multilayers rely on the interfacial hybridization between Co and Pt. This yields both strong PMA and DMI effects. The total effective anisotropy, K_{\text{eff}} , can be expressed as a function of ferromagnet (FM) thickness, t_{\text{FM}} , as K_{\text{eff}}=\frac{K_{\text{i}}}{t_{\text{FM}}}+\frac{1}{2}\mu_{0}M_{\text{s}}H_{\text{D}}\,, (1) where K_{\text{i}} is the interfacial PMA term, M_{\mathrm{s}} the saturation magnetization, and H_{\mathrm{D}} the demagnetization field. In ultrathin films the demagnetization field arising from shape anisotropy, H_{\text{D}}=-M_{\text{s}} . The quality of the Pt/Co interface plays a key role in determining the magnitudes of the DMI vector and K_{\text{i}} 2 ; 33 ; 14 , which increasingly dominates over the other volumetric contributions at low t_{\text{FM}} . The DMI contribution competes against a standard Heisenberg-like exchange interaction producing non-collinear and incommensurate Néel-type spin-canting over a characteristic lengthscale, \Delta . Previous PMA TI/FM switching studies 25 ; 5 note their use of buffer layers deposited on the TI surface before the MML is grown for out-of-plane (OOP) switching geometries and, when buffer layers are removed, FM easy axes tend to lie in-plane (IP). Inserting a buffer layer over a TI will introduce spin-flip scattering causing the spin-signal to be lost over a material dependent spin-diffusion length l_{\text{SD}} according to J_{s}(t)=J_{s,0}\,e^{-t/l_{SD}} ; this is in addition to the additional scattering introduced as a result of introducing a second interface 27 . This problem is particularly acute for heavy, refractory metals such as Ta, which possess large spin-orbit coupling (SOC) and hence spin diffusion lengths are short: l^{\text{Ta}}_{\text{SD}}\sim 2 nm 37 – yet are the standard materials for buffers for this type of MML stack. Typical TIs belonging to the commonly used Bi 2 X 3 class of materials are known to form in quintuple layers weakly bound by van der Waals forces 12 ; 1 ; 39 . This layering forms distinct triangular terracing at the surface of an TI surface, which raises concerns for magnetic pinning sites in TI/MML heterostructures. As such this study aims to tailor a buffer to ensure a high-quality multilayer structure and optimize for magnetic domains decoupled from the TI surface topography whilst minimising the loss of spin current between the TI and MML. II Heterostructure Growth The heterostructures that we studied combine 20 nm epitaxial layers of Bi 2 Se 3 , a well-known TI 40 ; 35 ; 39 ; 17 ; 31 , and a [Pt(0.8 nm)/CoB(0.6 nm)/Ru(0.5 nm)] \times 6 MML on a Ta or Mo buffer layer as shown in Fig. 1 . The TI was grown by molecular beam epitaxy (MBE) and the multilayer was grown by dc magnetron sputtering at room temperature. Figure 1: Schematic representing standard sample structure grown on Bi 2 Se 3 with illustrated skyrmion spin-texture in the CoB layers where m_{z} lies along the out-of-plane axis. The Bi 2 Se 3 (20 nm) epilayers were prepared in a solid-source molecular beam epitaxy (MBE) system that operated at a base pressure of 7×10 -10 mbar. The films were grown by co-depositing evaporated bismuth and selenium, both supplied from standard dual-filament Knudsen cells, on c-plane (0001) sapphire substrates. To effectively suppress the formation of selenium vacancies, the growth was carried out under Se-rich conditions, with the selenium flux at least 20 times higher than the bismuth flux. Two different samples were prepared. The first sample, characterized by narrow surface terraces \sim 50 nm wide, was grown at a temperature of 235 ∘ C. The second, wide-terrace sample was deposited using a two-step method, with a nucleation layer of \sim 2 nm grown at 130 ∘ C and the remainder of the layer grown at 300 ∘ C increasing average terrace width to \sim 100 nm. Connected MBE and sputtering chambers allowed the entire heterostructure to be deposited without breaking ultra-high vacuum, thereby minimizing degradation of the TI surface and keeping the interface between the materials as clean as possible. After transfer to the sputtering chamber, any buffer layer, followed by the MML with a 4.0 nm Pt capping layer, was deposited at room temperature. Here we have studied the buffer layer materials Ta and Mo, a 5d and 4d refractory metal, respectively. On reference MML samples grown on thermally oxidized Si(100)/SiO 2 (100 nm) substrates and 5.0 nm Ta buffers, PMA could be reliably obtained down to a minimum nominal Pt thickness of 0.8 nm - however the PMA strength is noticeably reduced when reducing Pt thickness beneath 1 nm 24 ; 27 ; 22 ; 41 . To maximise the effects of interfacial PMA against shape anisotropy, the FM CoB layer was thinned to the minimum thickness to be reliably continuous, which was found here to be 0.6 nm. This optimization yields a multilayer structure with strong PMA through minimized use of heavy metals between the FM and TI layers. Samples which were not required for imaging were grown with an additional 4 nm Pt cap to prevent oxidation. III Multilayer Structure on Terraced Bi 2 Se 3 Figure 2: 2 \upmu m \times 2 \upmu m AFM images of Bi 2 Se 3 /Ta/MML samples grown with recipes leading to different terrace widths: narrow (a) and wide (b). The terrace widths are detected using watershed grain detection, the areas were converted into equivalent disk radii r_{\mathrm{T}} and fitted with Gaussian distributions shown in (c). Resultant distributions in sample height is shown in (d). Epilayers of Bi 2 Se 3 grow with a terraced surface as a result of the lateral expansion and step flow growth of quintuple layers from defects during MBE growth. The terracing represents a unique challenge for Bi 2 Se 3 -based devices featuring such thin layers since a single quintuple layer, determined to be 0.96\pm 0.03 nm from X-ray diffraction (XRD) measurements, is equivalent to half a single trilayer thickness of the MML (1.9 nm). As such the average terrace width is an important parameter to optimise for in film growth: wider terraces will mean fewer potentially disruptive terrace edges. Figure 3: X-ray characterization of sample structure. (a-c) XRR curves of the main [Pt(0.8 nm)/CoB(0.6 nm)/Ru(0.5 nm)] ×6 MML stack as grown on Bi 2 Se 3 with (a) Mo and (b) Ta buffer layers as well as (c) on a thermally oxidized silicon substrate with 5.0 nm Ta buffer and 4.0 nm Pt cap. Multilayer and Bi 2 Se 3 peaks are marked with grey and blue dashed lines, respectively. (a-b) show curves using narrow and wide terraced Bi 2 Se 3 however only the wide terraced samples are fitted. The range over which the slab model is fitted is shown in a solid line. Axes beneath (d-f) show the fitted SLD distribution corresponding to these fits. The underlying slab models from (d) and (e) are shared with those later shown in Figs 6 & 7 . (g) XRD patterns of wide-terrace Bi 2 Se 3 grown both with and without overlying MMLs using both buffers. Relevant peaks are labelled. (h) Zoomed axes of (g) showing the Pt (111) peak. All intensity scales in this figure are logarithmic. From the atomic force microscopy (AFM) of complete Bi 2 Se 3 /Ta/MML heterostructures shown in Fig. 2 (a-b), the Bi 2 Se 3 surface topography can be seen to imprint through the overlying multilayer. The result of different Bi 2 Se 3 growth recipes on average island size and terrace width is evident. The terrace widths were detected using watershed grain detection, using the open-source software Gwyddion 23 . The distribution in equivalent disk radii, r_{\mathrm{T}} where \pi r^{2}_{\mathrm{T}}=A_{\mathrm{T}} for terrace area A_{\mathrm{T}} can be shown to be well approximated by Gaussian distributions with mean radii of \langle r_{\mathrm{T}}\rangle=51\pm 1 and 120\pm 4 nm for narrow and wide terraced Bi 2 Se 3 respectively as shown in Fig. 2 (c). The fitted standard deviation of distributions is higher for the wide terrace recipe at 100 nm versus 20 nm for narrower terraces indicating a greater variety of island sizes and terrace widths. The terracing is evident in the sample height distributions in Fig. 2 (d) with most of both sample surfaces being found within \pm 1 terrace steps of the dominant terrace height. Through slab-model fitting of X-ray reflectometry (XRR) measurements, our MMLs exhibit monolayer-level precision with interfacial roughnesses on the order of 2-3 Å in corresponding fitted slab models shown in Fig 3 (a-f). Measurements were taken using a Cu K \alpha source of \lambda=1.541 Å with a coherence length of L_{\mathrm{c}}\sim{\lambda^{2}}/{\Delta\lambda}\approx 300 nm. This characteristic lengthscale is sufficiently similar to the Bi 2 Se 3 terracing for the effect of widening the average terrace width to be apparent on the reflectivity. The extracted multilayer periodicity of 1.9 nm corresponds to a multilayer Bragg peak at 2\theta\sim 4.8^{\circ} . For the narrow terraces where \left\ll L_{\mathrm{c}} , the effective multilayer scattering length density (SLD) is averaged over multiple terraces and so multilayer features are suppressed. When the terraces become wider such that \left\lesssim L_{\mathrm{c}} , the effective medium approximation (EMA) begins to break down and coherent reflection from wider terraces allows for a broad characteristic Bragg peak to emerge. This is reproducible on both Mo and Ta buffer layers shown in Fig. 3 (a-b) and confirms that the trilayer structure grown on Bi 2 Se 3 is resolvable. This dependence of the multilayer Bragg peak on the average terrace width is attributed to difficulties fitting high-angle reflectivity data for samples grown on the Bi 2 Se 3 . This is addition to the Laue oscillations of the lowest order Bi 2 Se 3 XRD peak, which appears at 2\theta=9.3^{\circ} . Hence these XRR curves are only fitted in the low angle regime. Figure 4: Cross sections prepared by FIB of Bi 2 Se 3 /buffer/MML/Pt(4.0 nm) heterostructures as seen under HAADF-STEM imaging. (a-b) Cross sections showing MML overlaying multiple terrace edges (a) with and (b) without a Ta (1.5 nm) buffer on narrow terrace Bi 2 Se 3 ; (c) Zoomed in section of (b) showing terrace edge with resultant layer intermixing; (d) observed quintuple layering of Bi 2 Se 3 with corresponding schematic of a single quintuple layer Bi 2 Se 3 shown in (e); (f-g) integrated line profiles of HAADF grey count at and far away from the terrace edge with (f) and without (g) buffer. Overlaid in (f-g) are the respective, co-fitted X-ray SLDs corresponding to models used in Fig. 7 . The full XRD patterns shown in Fig. 3 (g) show distinct diffraction peaks matching the Bi 2 Se 3 crystal structure. The large number of Laue oscillations are seen indicating good sample quality. Notably, these are seen before and after MML deposition, indicating little to no damage to the TI throughout the sputtering process. The choice between Mo and Ta buffers creates a noticeable difference in the XRD spectra. When grown without the 4 nm Pt capping layer, multilayers grown upon a Ta (2.0 nm) buffer display a diffraction peak seen at 39.88 ∘ , characteristic of Pt (111) ordering within the multilayer structure as shown in Fig. 3 (h). Pt grown on Mo (1.0 nm) does not show evidence of such Pt(111) ordering in XRD above the noise floor. This is potentially supported by subtle differences in X-ray SLD distributions where the fitted Bi 2 Se 3 /Ta/MML profile is the only sample to show three distinct metallic layers in the multilayer repeat unit, indicating sharper interfaces. We can directly observe the structural effect of TI surface terracing on the MMLs using high angle annular dark field scanning transmission electron microscopy (HAADF-STEM), shown in Fig. 4 . The multilayer structure over the terrace edge is observed both (a) with and (b) without a Ta (1.5 nm) buffer, and in both cases can be seen to closely follow the profile of the underlying Bi 2 Se 3 surface. The quintuple layering of Bi 2 Se 3 which leads to terracing is evident under closer imaging as shown in (d). Notably no Se vacancies are found on the atomic scale. Taking an averaged vertical line profile, the layers in both samples can be seen to be well defined when away from terrace edges shown in Fig. 4 (f-g). The features in the HAADF-count broadly match the fitted X-ray SLD distributions used throughout. Some slight modulation in stack thickness is observed in the non-buffered sampled in Fig. 4 (f) with layers being thicker towards the top of the stack, however this is attributed to the sputter targets heating up during growth leading to growth rate drift. At the terrace edge, the multilayer without any buffer is seen to break down due to increased layer-to-layer intermixing which worsens closer to the Bi 2 Se 3 interface as shown in Fig. 4 (c) & (g). When the Ta buffer is inserted however, this effect is not seen and the line profile is unchanged at the terrace edge with all layers remaining resolvable as shown in Fig. 4 (a) & (f). This continuous multilayer structure can be expected to reduce the effect of pinning of domain walls and topological spin textures at terrace edges, essential for any future device applications. IV Achieving Full Perpendicular Magnetic Anisotropy via Buffer Insertion IV.1 Magnetometry Figure 5: Buffer thickness dependence of MML magnetic properties, extracted from SQUID-VSM measurements for Ta and Mo buffer thickness series on Bi 2 Se 3 : (a) PMA fraction taken as the fraction of M {}_{\text{s}} ; (b) the effective anisotropy field, \mu_{0}H_{\text{K}} or corresponding K_{\text{eff}} assuming constant M_{\mathrm{s}} . Both (a) & (b) include fitted sigmoid + Gaussian functions included to highlight the trend shape as well as a reference series on Si/SiO 2 /Ta(x). (c) Moment per unit area versus total t_{\text{CoB}} in the stack. A linear relation is fitted to simple Pt/CoB(x)/Ru trilayers to calculate M_{\mathrm{s}}=600\pm 20 emu/cc; a second is included to account for 6 proximity magnetized Pt interfaces in the MML stacks. Included are the averaged M_{\mathrm{s}} values for Mo and Ta buffer thickness series shown in (a-b); (d-e) SQUID-VSM loops MMLs on Bi 2 Se 3 on a Ta(10 nm) buffer (d) or no buffer (e) taken in both IP and OOP geometries. Easily magnetizable fractions in both geometries are highlighted corresponding to a signal equivalent to an integer number of layers magnetized ( n/6 ). To determine the magnetic properties, the samples were measured by superconducting quantum interference device vibrating sample magnetometry (SQUID-VSM) from which key magnetic parameters such as the stack-averaged, effective anisotropy field H_{\text{K}} , the here-named “PMA ratio”, and saturation magnetization M_{\text{s}} - could be extracted for each sample. The anisotropy field is defined as H_{\text{K}}=2K_{\text{eff}}/\mu_{0}M_{\text{s}} and where K_{\text{eff}}>0 is corresponds to an out-of-plane easy axis. The stack-averaged effective anisotropy K_{\text{eff}} was calculated by integrating the difference between in-plane and out-of-plane M(H) curves. We define the PMA ratio as M_{\text{PMA}}/M_{\text{s}} , the magnetization fraction of the sample magnetized under a small 100 Oe bias field out-of-plane as a fraction of the total saturation magnetization M_{\mathrm{s}} . This small field is needed to saturate out any domains that exist at zero field, and is used in place of the more usual squareness ratio which would be near-zero for samples with wasp-waisted hysteresis loops, despite possessing strong PMA. As shown in Fig. 5 (f), samples grown without a buffer between the TI and MML show easily magnetisable fractions of 5/6 and 1/6 for the out-of-plane and in-plane directions, respectively, as well as a reduced H_{\text{K}} . This suggests that five of the six CoB layers in the MML possess PMA, but one does not and is instead in-plane magnetized at low fields. Samples were then grown with a range of Mo and Ta buffer layers thicknesses x to find the minimum thickness needed to recover full PMA in every layer. K_{\text{eff}} , the PMA fraction and M_{s} were then plotted as a function of buffer material and thickness. These values are shown in Fig. 5 (a), (b), and (c) respectively. Inserting a Ta buffer between the TI and the multilayer recovers the PMA fraction to >90 % once a thickness of x\sim 1.5 nm is exceeded, as shown in Fig. 5 (a), with H_{\text{K}} continuing to rise up to a thickness of 2 nm where H_{\mathrm{K}} reaches a plateau at \mu_{0}H_{\mathrm{K}}=0.65\pm 0.02 T. This is a similar process to that of Ta on natively oxidized Si, where the PMA fraction increases from \sim 5/6 without any buffer, plateauing once a critical thickness is surpassed. Differences to note are that on Si, the PMA ratio and H_{\mathrm{K}} recover faster, plateauing at the same point at x\sim 0.8 nm and to a higher H_{\mathrm{K}} value of \mu_{0}H_{\mathrm{K}}=0.80\pm 0.01 T or the equivalent of K_{\text{eff}}=0.66\pm 0.01 MJ/m 3 . To obtain a H_{\text{K}} value comparable to multilayers grown on natively oxidized Si/SiO 2 on Bi 2 Se 3 , the buffer must be increased further, seen in Fig. 5 (b), to agree with Si at x=10 nm. At this large thickness of Ta we can expect that the multilayer would be effectively shielded from any spin-signal from the Bi 2 Se 3 leaving it redundant. Repeating the same experiment with Mo instead of Ta, a different behavior was observed, with the both H_{\text{K}} and the PMA ratio peaking at a low thickness of 0.94\pm 0.06 nm to an increased \mu_{0}H_{\mathrm{K}}=0.72\pm 0.05 T. Increasing the buffer thickness further leads to a sharp decrease in H_{\text{K}} and small decrease in the PMA ratio. This shows there is sufficient Pt (111) /Co hybridization for PMA despite the Pt (111) peak not being visible in XRD measurements in Fig. 3 (d). We can hypothesise that there is increased intermixing between Pt and CoB when grown on Mo versus Ta leading to insufficient long range ordering for the Pt (111) peak to be visible. The drop observed in H_{\mathrm{K}} could be explained as this effect worsening with increased Mo thickness. We can therefore conclude that depositing at least 1.5 nm of Ta or \sim 0.9 nm of Mo on the Bi 2 Se 3 epilayer surface as a buffer layer recovers the PMA in the layer that is in-plane magnetized with no buffer. The choice of buffer comes with additional subtle changes in the MML’s magnetic characteristics. Multilayers grown on both Ta and Mo buffers show M_{\mathrm{s}} values consistent with a trilayer Pt(1.5 nm)/CoB( x )/Ru thickness series as grown on Ta (5 nm) on natively oxidized Si, as shown in Fig. 5 (c), where M_{\mathrm{s}} for Co 68 B 32 is 600\pm 20 emu/cc. This indicates no substantial magnetic dead layers forming as a result of the Bi 2 Se 3 or of the reduced Pt thickness. Proximity induced magnetism (PIM) in Pt is indicated by a non-zero intercept of (m/A)_{\text{Pt}}=8\pm 3 \upmu emu/cm 2 per Pt interface. IV.2 Polarized neutron reflectometry Figure 6: Example polarized neutron reflectometry data: (a) shows Fresnel reflectivity curves for both neutron polarizations: the resultant SA is plotted in (b). Simulated results from a fitted slab model are shown in (a) and (b): the corresponding model’s 1\sigma confidence intervals for magnetic and non-magnetic neutron scattering distributions are shown in (c) as well as the co-fitted X-ray SLD from 3 (e). The equivalent \rho_{\mathrm{M}} from the calculated CoB M_{\mathrm{s}} value taken from Fig. 5 (c) is shown as reference; (d) shows the sample geometry with respect to the neutron beam and applied field. Whilst the conventional magnetometry measurements show that full PMA can be recovered in these ultrathin films on Bi 2 Se 3 , they still highlight that the magnetic anisotropy can vary greatly throughout the stack. This could be problematic when spin-torques are predominantly exerted on the bottom-most FM layer where we cannot easily directly image. A solution to this is to verify the multilayer stack is magnetically homogenous such that we can be confident magnetic textures seen through surface-sensitive measurements correspond to the full texture throughout the stack through magnetostatic coupling. Polarized neutron reflectivity (PNR) was used to obtain a depth resolved magnetization profile of multilayers grown on the various buffers. In this way we can identify the layer that lacks PMA when no buffer is present as well as any large changes in anisotropy. Spin-polarized neutron reflectivities R_{+} and R_{-} were taken as a function of wavevector transfer Q through time-of-flight detection on the POLREF instrument at ISIS, and the spin asymmetry (SA), defined as SA =(R_{+}-R_{-})/(R_{+}+R_{-}) , between them was calculated. An example PNR measurement is shown in Fig. 6 , including the measurement geometry. SAs were measured at applied in-plane fields from full or near saturation ( \mu_{0}H=700 mT) down to the minimum applicable field to prevent beam de-polarization ( \mu_{0}H=10 mT). Since PNR in the geometry we used, see Fig. 6 (d), is sensitive to the in-plane component of magnetization, our experiment was designed to measure that component as it increases whilst the in-plane applied field forces the magnetization into the magnetically hard film plane. When the PNR data were fitted (Fig. 6 (a) & (b)) using the Refl1D code 19 , the resulting magnetic SLD distribution appears as a spike for each of the six magnetic CoB layers (Fig. 6 (c) with an amplitude proportional to that in-plane component. The PNR confirms the SQUID-VSM result that both Ta (2 nm) and Mo (1 nm) buffers yield multilayers with zero in-plane remanance, demonstrated by a vanishing spin-asymmetry under a 10 mT field, which shows that there is no in-plane component of magnetization present. The lack of any notable change in shape of the spin asymmetry with field, shown in Fig. 7 (e) and (f) suggest magnetic profiles changing in magnitude and not substantially in shape. We can conclude that magnetic anisotropy is uniform throughout the multilayer stack, which possesses full PMA. Figure 7: PNR as a function of external applied field. (a-b) Fitted magnetic scattering length density profiles, \rho_{\text{M}}(z) for [Pt/CoB/Ru] ×6 multilayers grown on Ta (2 nm) (a) and Mo (1 nm) (b) buffer layers on wide-terrace Bi 2 Se 3 . Opaque and translucent profiles show 1 and 2\sigma confidence ranges, respectively. (c-d) Corresponding magnetometry data for MMLs on Ta (2 nm) and Mo (1 nm) buffers respectively. Out-of-plane and in-plane SQUID-VSM measurements are shown over the integrated \rho_{\text{M}} profiles from (a-b). (e) and (f) Fitted spin-asymmetry profiles for Ta and Mo buffers, respectively, with simulated data corresponding to the \rho_{\text{M}}(z) profiles shown in (a-b). (g) Fitted SLD distributions for various buffers taken at the minimum 10 mT alignment field. (h) Individual, layer-resolved H_{\text{K},n} values obtained from the model fittings for various buffer types. Stack averaged H_{\text{K}} values extracted from SQUID measurements in Fig. 1 (b) are shown for reference. (i) Fitted low field spin asymmetries corresponding to the slab models in (g). Measured neutron reflectivity curves were fitted to nuclear and magnetic SLD profiles ( \rho_{\text{N}} and \rho_{\text{M}} ) using slab model fitting. Due to the low thickness of the individual layers, neutron data was co-fitted with the X-ray data and models shown in Fig. 3 (a-b,d-e) to extend our range in Q . Our magnetic slab models assume a phenomenological, Langevin-type M(H) dependence and fit each CoB with a individual effective anisotropy, K_{\text{eff},n} where the magnetic SLD in the n^{\text{th}} CoB layer follows: \rho_{M,n}(H)={\mu_{0}}{\alpha}M_{\text{s}}{\cal L}(a_{n}\mu_{0}H) (2) where {\cal L}(a_{n}H) is the Langevin function and \alpha=2.32 Å -2 T -1 is a conversion factor 18 . This allows for field-dependent simulations based on a single scaling parameter a_{n} per layer which can be related to K_{\text{eff},n} and has units of T -1 . The final fits are shown in Fig. 7 (a-b) and were obtained using the open source library Refl1D 19 . Integrating \frac{1}{\mu_{0}\alpha}\int\rho_{M}\,\text{d}z yields the moment per unit area which can be shown to agree with the measured SQUID-VSM data shown in Fig. 7 (c-d). In fitted \rho_{\text{M}} distributions at saturation where H_{\text{K}}