Tailoring Ultrathin Magnetic Multilayers at Terraced Topologically Insulating Interfaces for Perpendicularly Magnetized Domains

TL;DR

This work demonstrates how inserting a thin buffer layer between a terraced Bi$_2$Se$_3$ topological insulator and an ultrathin chiral magnetic multilayer recovers uniform perpendicular magnetic anisotropy across all FM layers, enabling potential spin–orbit torque manipulation of spin textures. The authors grow Bi$_2$Se$_3$ by MBE and couple it to a Pt/CoB/Ru multilayer via Ta or Mo buffers, then characterize structure with XRR, XRD, AFM, and HAADF-STEM, and magnetism with SQUID-VSM and polarized neutron reflectometry (PNR). They find buffer-thickness thresholds (Ta ≈ 1.5 nm, Mo ≈ 0.9 nm) that restore full PMA, with PNR confirming depth-uniform magnetization and a near-zero in-plane remanence, while unbuffered interfaces exhibit bottom-layer PMA loss and in-plane remnants. XMCD-PEEM shows domain textures are strongly influenced by Bi$_2$Se$_3$ terrace width, yielding well-defined labyrinthine domains on wide terraces, especially with Ta buffers, highlighting the importance of surface morphology alongside buffering for TI-based spintronic devices.

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]$_{\times N}$ 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.

Figures (8)

• 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.
• 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).
• 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)]$_{\times 6}$ 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 \ref{['fig:PNR_example']} & \ref{['fig:PNR']}. (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.
• 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. \ref{['fig:PNR']}.
• 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 ($n/6$).