Imaging the magnetic nanowire cross section and magnetic ordering within a suspended 3D artificial spin-ice

The study of three-dimensional (3D) magnetic nanostructures is a rapidly growing research area^1–4 due to their potential technological applications^5,6 and emerging fundamental physics driven by non-trivial geometry and topology,^7–9 as well as the frustration that can arise due to the interaction between 3D nanoelements.^10–14 Advances in theory have shown that curvature and torsion can yield additional energy terms in the form of effective Dzyaloshinskii–Moriya and effective anisotropy interactions.^15–17 Such interactions can lead to the stabilization of topological spin textures^18–20 without the need for multilayer systems and can also yield novel dynamic phenomena, such as domain wall automation.²¹ Advancements in complex fabrication processes have allowed the realization of many 3D geometries, including cylindrical magnetic nanowires,²² rolled up membranes,²³ helical structures,^24,25 and 3D artificial spin-ice structures,¹² by harnessing a range of self-assembly and lithographic methods. Next-generation magnetic racetrack memories⁵ rely upon 3D architectures, and before such technologies can be realized, the detailed physics of domain wall topology, propagation, and pinning needs to be understood in magnetic nanowires and elements of complex geometry. Advances in nanofabrication, theoretical modeling, and advanced characterization to validate such systems are a prerequisite in that regard.

Of the many fabrication methodologies available, direct-write technologies are particularly interesting since they provide the most freedom in the choice of geometry. Such technologies can harness either photons²⁶ or electrons^27,28 to generate the desired 3D geometry. For example, focused electron beam-induced deposition (FEBID) has been used to realize magnetic nanowires with transfer of domain walls from a planar substrate to an angled 3D nanowire²⁹ and exotic double helix structures that generate topological defects in the stray field.⁹ Two-photon lithography (TPL) is a methodology whereby 3D nanostructures can be written by design within a suitable negative-tone photoresist by scanning a tightly focused laser through the resist to achieve the desired geometry.²⁶ The technique has been used in combination with deposition technologies to create bucky ball structures,^30,31 magnetic nanowires,¹¹ 3D artificial spin-ice (3DASI) building blocks,³² and arrays.^10,12 To characterize 3DASI array systems, magnetic force microscopy (MFM) was used to directly image the propagation of monopoles across the lattice¹² and to image the demagnetized state, suggesting relaxation into an exotic magnetic charge crystal. However, a key limitation of MFM is the requirement to physically access the surface topography to obtain accurate magnetic information, a challenging feat for any 3D lattice. The resolution of MFM depends upon the lift height, which is usually limited to 50–100 nm. It is, therefore, expected that advanced characterization techniques using polarized x-ray spectro-microscopies for tomographic reconstructions, with sub-50 nm resolution, will become important in measuring next-generation 3D magnetic nanostructures.^1,2,9,33,34

In this article, we develop a new fabrication process that yields suspended 3DASI systems over an aperture. This allows measurement of full lattices with optical access to all nanowires without any need for tilt or rotation, while also eliminating any sheet film under the 3D nanostructure, in principle allowing the ordering to be determined. The element specific 3D geometric cross section of underlying nanowires is determined, and x-ray magnetic circular dichroism (XMCD) shows evidence of a magnetic signal, allowing the magnetic texture upon a single sub-lattice to be reconstructed.

TPL is used to realize a diamond-bond polymer lattice over the exposed corner of a 50 × 50 µm² Si aperture (Silson TEM-200-0.05-aperture). Crucially, the use of a dip-in TPL technique allows the structure to be firmly anchored upon a large region of a Si substrate, providing support for the suspended lattice region. Figure 1 shows a schematic of the fabrication process. The Si substrate is first cleaned in acetone and isopropyl alcohol (IPA) and then carefully dried using compressed air [Fig. 1(a)]. A negative-tone photoresist (IP dip, Nanoscribe) is drop-cast onto the Si substrate and loaded into a TPL apparatus [Fig. 1(b)]. A diamond-bond lattice, one unit cell in height (∼2.3 µm), is written across each corner of the aperture [Figs. 1(c) and 1(d)], after which the sample is developed in propylene glycol monomethyl ether acetate (PGMEA), rinsed in IPA, and then carefully dried using compressed air [Fig. 1(e)]. Figure 1(f) shows a schematic of a final structure with sub-lattice layers labeled SL1–SL4. The SL1 nanowire layer is the upper surface termination and consists of coordination two and coordination four vertices. Layers SL2 and SL3 are comprised entirely of coordination four vertices. The SL4 wire layer is akin to SL1 with alternating coordination two and coordination four vertices.

Two sample sets were produced to investigate the extent to which a lattice can be suspended over an aperture. In sample set A, TPL was used with piezoelectric stages, scanning the sample with respect to the point of focus and only with an exposure of sub-lattices SL1 and SL2. In the second sample set, galvanometric mirrors were used to steer the laser focus with respect to the sample, with full exposure of sub-lattices L1–L4. Both sample sets were then subject to a tri-layer deposition, Al (4 nm)/Ni₈₁Fe₁₉ (43 nm)/Al (3 nm), to place a protected magnetic coating onto the polymer nanowires within the lattice [Fig. 1(f)]. Deposition of the stack was performed using a thermal evaporator operating at a pressure of 3 × 10⁻⁷ mbar. A crystal quartz monitor present during evaporation measured the deposition thickness, which was later confirmed with atomic force microscopy measurements. Sample set B was subject to an oxygen plasma step to further reduce the thickness of the underlying polymer. As shown previously,^10–12,26 the realized diamond lattice yields an array of single-domain magnetic nanowires that exhibit an Ising-like behavior.

Sample set A was fabricated using a piezo-mode, to provide longer exposures and aid in adhering the written polymer to the substrate surrounding the aperture. Such samples provide maximum stability of the suspended lattice, providing initial feasibility of the transmission-based experiments. Figure 2(a) shows a scanning electron microscopy (SEM) image of the 3DASI lattice, with Fig. 2(b) showing an image taken at 45° tilt. The individual nanowires are found to have widths of 301 ± 7 nm, a length of 1 µm, and an average thickness of ∼1 µm. Representative XMCD images taken at the Ni L₂ and L₃ edges are shown in Figs. S1(a) and S1(b), respectively, for the region outlined in Fig. 2(a) (dashed lines). Bands of contrast can be seen upon individual nanowires. However, we note that the apparent contrast does not invert between the L₂ image and L₃ image, suggesting that this contrast is not magnetic in origin but likely an artifact. Although XMCD contrast is absent, transmission-based experiments allow the intriguing possibility to explore the 3D cross section of the nanowires in these samples. Previous work has hinted that the wires have a novel crescent-shaped cross section.^10, Figures 2(c) and 2(d) show the thickness maps of the Ni and Fe components of the magnetic material upon the lattice, respectively; both show that the magnetic material is uniform across much of the lattice with a region of decreased thickness found at the intersection of SL1 and SL2. In this location, the measured optical density drops by 40%. This can be explained with SEM measurements, which show that the suspended lattice tilts slightly into the aperture, which combined with the evaporation at a normal incidence to the sample will result in a change in deposition thickness transverse to the tilt axis. Extracting line profiles across the wire width allows the cross-sectional thickness variation to be determined, as shown in Fig. 2(e). The profile shows a graded thickness as previously hinted by SEM imaging.¹⁰ When considering the cross-sectional geometry of a TPL voxel, which has an elliptical shape,²⁶ the superposition of a graded NiFe thickness yields wires of crescent-shaped cross section, as shown in the inset of Fig. 2(e). Magnetic nanowires with a crescent-shaped cross section are expected to host domain walls with perturbed spin textures due to the curvature yielding effective anisotropy and Dzyaloshinskii–Moriya energies.¹⁶ Overall, initial STXM studies conducted at Diamond Light Source on sample set A provided feasibility of measurement and gave insight into the physical cross section of the nanowires.

Sample set B was fabricated with galvanometric mirrors, which reduced the exposure dose of the lattice. A key advantage of this approach is a reduced thickness of the polymer nanowires. Exposure to oxygen plasma further reduces the polymer thickness. A SEM image of a suspended 3DASI from sample set B is shown in Fig. 3(a), with high magnification images shown in Figs. 3(b) and 3(c). Nanowires were found to have widths of ∼220 ± 6 nm and lengths of 1 µm. Atomic force microscopy (AFM) was used to measure surface topography, as shown in Fig. 3(d). The 3DASI in this sample set comprises four sub-lattice layers designated SL1 through to SL4, as depicted in Fig. 3(c). The average thickness of the nanowires for this sample set was found to be ∼810 ± 18 nm, which is lower than that of sample set A.

Sample set B was measured using in-house magneto-optical Kerr effect (MOKE) magnetometry prior to x-ray microscopy measurements. MOKE loops upon the Si substrate that surrounds the aperture yielded a hysteresis loop with a coercive field of μ₀H_C = 0.6 mT as-deposited and 0.5 mT after a 30 min plasma exposure, as shown in Fig. S2 of the supplementary material. The slightly increased coercivity is likely due to the substrate roughness or slight oxidation but is still consistent with the range measured previously³⁸ for evaporated Ni₈₁Fe₁₉. Figure 4(a) shows the XAS taken at the Fe L₃ edge upon samples with varying oxygen plasma exposure, with the full spectra shown in Fig. S3 of the supplementary material. The sample with no oxygen plasma exposure shows a peak at 707.5 eV, indicating metallic Ni₈₁Fe₁₉. The presence of a shoulder at ∼709 eV suggests some oxidation,⁴⁰ despite a capping layer. Measurement of samples with increasing oxygen exposure yields more evidence of a multiplet structure with increasing peak intensity at ∼709 eV. The inset of Fig. 4(a) shows that the ratio of Fe peak to Fe₃O₄ peak decreases with increasing oxygen plasma exposure time. Overall, these results suggest that samples with an intermediate oxygen plasma exposure time will have the best balance of a reduced polymer thickness but with a large volume fraction of metallic Ni₈₁Fe₁₉. Specifically, for the 60-min oxygen plasma exposure, the polymer thickness is reduced by ∼10%. This should yield an improved transmission while maintaining a Fe/Fe₃O₄ L₃ peak ratio of 0.95.

Figure 4(b) shows an XAS optical density image taken at the Fe L₃ edge upon a sample that was exposed to oxygen plasma for the optimal 60 min. A crucial uncertainty with respect to such samples was the extent to which the 3DASI structure, which is suspended, fabricated using Galvanometric mirrors with reduced anchoring, and subject to oxygen plasma, would remain stable. Analysis of multiple images taken over several hours, as shown in Supplementary Video 1, shows little movement or distortion of the lattice. Some regions within Fig. 4(b) can be clearly seen to have higher x-ray absorption. In many cases, these areas are correlated with overlap in lower structures.

Figure 5 shows the XMCD images using TXM, of a lattice subject to a 60-min oxygen plasma exposure, in an as-deposited state [Fig. 5(a)] and after the application of a 100 mT field [Fig. 5(b)]. An apparent contrast is present in the images, with dark regions particularly noticeable at the vertices. A close inspection of these regions suggests that this contrast is not of magnetic origin but due to overlapping features. To carefully inspect the measurement for evidence of magnetism, non-overlapping regions were found and chains, connected wires on a specific sub-lattice, were identified for SL1, SL2, and SL3, as shown in Fig. S4 of the supplementary material. The average XMCD signal for Fe L₃ across the central region of the wire, where there is a negligible structural background, was measured. This was then plotted for each of the three wire sub-lattices for the as-deposited state and after the application of a 100 mT magnetic field, as shown in Figs. 5(c) and 5(d). Previous studies have shown the magnetic configuration that would be expected after the application of a saturating field.¹² In-plane saturation along each of the sub-lattices will yield a net magnetization component parallel to the field. However, since each wire along a given sub-lattice has an alternating ±35.25° angle, out of the substrate plane, and since XMCD measurements are sensitive to the magnetization component along the photon propagation direction, for a saturated sample, we expect to see alternating contrast as depicted in Fig. S5 of the supplementary material. An examination of the SL1 chain contrast, as shown in Fig. 5(c), yields inconclusive results. Since the uppermost layer is most susceptible to artifacts due to overlap, only short chains can be identified, and these do not show a difference in trend when comparing configurations before and after field application. This suggests that any XMCD signal upon the SL1 sub-lattice is below the noise floor of the measurement. Analysis of data upon SL2, shown in Fig. 5(d), is more intriguing. The relative contrast, wire-to-wire, is different before and after the application of the magnetic field. Notably, the chain data after the application of a magnetic field take the expected form with alternating contrast, as depicted in Fig. S5 of the supplementary material. The XMCD signal is weak, at around 4%, likely due to a reduced magnetic moment driven by oxidation. An alternative analysis that can be performed is to inspect the magnitude of the contrast for wires in adjacent chains. Here, we expect wires with positive and negative magnetization projections onto k to have distinct values. Figure 5(e) shows the XMCD signal for the two sets of wires before and after the application of a field. Before the application of the field, the average XMCD value for both sets of wires are found to be within 0.5%, suggesting a net out-of-plane magnetization. After the application of the field, both SL2 and SL3 had a difference of ∼4%, consistent with the previous analysis, suggesting the presence of a weak magnetic signal. However, only SL2 has distinct populations of the XMCD signal, for which there is no overlap within 2σ, suggesting high confidence in magnetic contrast. Plotting the magnetization of the SL2, as shown in Fig. 5(f), also shows that the measured configuration is consistent with the direction of the external field, providing additional confidence. We note that similar analysis on SL3 shows weaker evidence of magnetic contrast, as shown in Figs. S6 and S7 of the supplementary material. Although there is a difference in the mean XMCD signal between upward and downward wires, there is a strong overlap in the two populations. We note that it is possible for a small drift in focus to account for the small XMCD signals observed on the SL2 and SL3 sub-lattices, but that this would impact all sub-lattices, whereas for SL1 [Fig. 5(a)], there is clearly no change in the trend before and after field application. Additional comparative analysis of wire widths (SL1, SL2, and SL3), shown in Fig. S8 of the supplementary material, between XAS images of pre-field and post-field applications revealed consistent measurements within the margin of error, suggesting a negligible drift in focus.

A key question is why only specific sub-lattices within the structure show evidence of magnetic contrast. The oxygen plasma exposure is expected to be uniform across ∼100 mm diameter. However, the extent to which the plasma can penetrate a 50 × 50 µm² aperture is less certain. Furthermore, the stray field lines from the NiFe coating may partially shield the lattice, particularly the lower layers. This may impact both the extent to which different sub-lattices have oxidized NiFe and the resulting gradient in polymer thickness when going from SL1 to SL4 sub-lattices. Overall, this is likely to lead to a trend whereby SL1 has the most polymer removed but also has the most prominent reduction in magnetization due to oxidation. At the other extreme, SL3 and SL4 are likely to have a higher magnetization but the least polymer removed, yielding a lower signal-to-noise ratio in XMCD measurements. Then, the intermediate SL2 sub-lattice is likely to be optimal in terms of magnetization and polymer removal.

In conclusion, a 3DASI lattice has been fabricated over an aperture using two-photon lithography, deposition, and oxygen plasma exposure. The 3D lattice is found to be stable when subject to repeated soft x-ray exposure. The plasma exposure of the polymer scaffold improved x-ray transmission through the structure, yielding a greater signal than would otherwise be seen. XAS data are used to extract the cross section of the magnetic nanowire, which is found to take a novel crescent-shaped geometry. XMCD measurements suggest that a signal can be measured on the buried sub-lattices, hinting at the expected configuration after an applied field. Minor adjustments in the fabrication procedure, by using a shorter wavelength laser,⁴¹ are expected to reduce the need for oxygen plasma and, when combined with optimized beamline measurement protocols, are expected to provide uniform magnetic contrast across all lattice layers and an improved signal to noise ratio.

S.L. acknowledges the funding from the Engineering and Physics Research Council (Grant Nos. EP/X012735/1 and EP/R009147/1) and the Leverhulme Trust (Grant No. RPG-2021-139). D.R. and P.F. were supported by U.S. Department of Energy, Office of Science, Office of Basic Energy Sciences, Materials Sciences and Engineering Division under Contract No. DE-AC02-05-CH11231 (NEMM program MSMAG).

The authors have no conflicts to disclose.

Edward Harding: Conceptualization (equal); Data curation (equal); Formal analysis (equal); Funding acquisition (equal); Investigation (equal); Methodology (equal); Project administration (equal); Supervision (equal); Writing – original draft (equal). Tohru Araki: Data curation (equal); Formal analysis (equal); Investigation (equal); Methodology (equal); Writing – original draft (equal); Writing – review & editing (equal). Joseph Askey: Data curation (equal); Formal analysis (equal); Investigation (equal); Methodology (equal); Writing – review & editing (equal). Matthew Hunt: Conceptualization (equal); Formal analysis (equal); Funding acquisition (equal); Investigation (equal); Writing – review & editing (equal). Arjen Van Den Berg: Data curation (equal); Formal analysis (equal); Investigation (equal); Methodology (equal); Writing – review & editing (equal). David Raftrey: Formal analysis (equal); Investigation (equal); Writing – review & editing (equal). Lucia Aballe: Data curation (equal); Formal analysis (equal); Investigation (lead); Methodology (equal); Writing – review & editing (equal). Burkhard Kaulich: Funding acquisition (equal); Investigation (equal); Methodology (equal); Writing – review & editing (equal). Emyr MacDonald: Formal analysis (equal); Investigation (equal). Peter Fischer: Conceptualization (equal); Formal analysis (equal); Funding acquisition (equal); Investigation (equal); Writing – review & editing (equal). Sam Ladak: Conceptualization (equal); Funding acquisition (equal); Investigation (equal); Methodology (equal); Project administration (equal); Supervision (equal).

The data that support the findings of this study are openly available in the Cardiff University data catalog at http://doi.org/10.17035/d.2024.0305551299.