Isotropic exchange-bias in twinned epitaxial Co/Co₃O₄ bilayer

After being discovered in 1956,¹ the exchange bias (EB) phenomenon was intensively investigated in experiment and theory for spintronic applications.² The most common EB system consists of a ferromagnet (FM) exchange-coupled to an antiferromagnet (AFM), which causes a horizontal shift of the hysteresis loop H_EB after field cooling (FC) through the Néel temperature T_N. This effect is often accompanied by a vertical loop shift and an asymmetry of the hysteresis loop.^3,4

The EB was first discovered in Co/CoO, which is still among the most frequently investigated oxide-based EB model systems.^5,6 Besides core-shell nanoparticles^7,8 and nanostructures,^2,9 Co/CoO is often examined in the form of thin-film systems. Co/Co₃O₄ is another typical model system for a ferromagnet combined with a diluted antiferromagnet, as described in the domain state model.^10,11 Co/Co-oxide thin-film systems can be grown in different orientations depending on the crystal orientation of the substrate, which results in different anisotropies and EB fields.⁵ The EB is also known to depend on the AFM and FM layer thickness.^9,12,13

Generally, Co/CoO and Co/Co₃O₄ thin film structures can exhibit EB fields in a broad range. Beschoten et al. found values ranging from ∼200 to 800 Oe after FC to 20 K depending on the thickness and oxygen dilution of the Co_1−yO layer.¹¹ Krivorotov et al. reported a similar value around 650 Oe for macroporous Co/CoO film systems.¹⁴ Smaller EB shifts of 60–180 Oe at 20 K were reported by Kumar et al. for polycrystalline Co/CoO bilayers with different interface roughness,¹⁵ while the AFM grain size was shown to increase the EB field from about 100 Oe to 200 Oe.¹⁶ Larger EB shifts than the ones mentioned are usually not found in Co/CoO or Co/Co₃O₄ thin film systems. Most recently, Co/Co₃O₄ nanorod arrays have been found to show a relatively small EB of 35–60 Oe,¹⁷ while Co/CoO core shell nanostructures exhibited much larger EB values of up to ∼600 Oe at low temperatures.^18,19

The blocking temperatures of the antiferromagnets CoO and Co₃O₄, above which no EB occurs, are significantly different, corresponding to Néel temperatures of T_N ≈ 295 K (CoO) and T_N ≈ 40 K (Co₃O₄), respectively, leading to vastly different temperature dependencies of such systems.²⁰ Nevertheless, some studies on Co/Co₃O₄ bilayer systems already showed that an EB can occur up to a temperature of 220 K²¹ or even higher.²² Systems with overoxidized Co_1−yO, where large EB shifts were found even near room temperature,^10,11 make it clear that it is important to investigate the precise crystal structure and composition of the AFM. (111)-oriented Co/Co_1−yO systems with their threefold crystallographic symmetry are expected to show less pronounced magnetic anisotropy compared to (110)-oriented systems with only twofold symmetry⁵ due to smaller angles to the nearest magnetic easy axis in the (111)-orientation.

Here, we report on the synthesis, detailed structural characterization, and magnetization measurements of a twinned Co(111)/Co₃O₄(111) bilayer system with sixfold crystallographic symmetry on an Al₂O₃(0001) (sapphire) substrate. Temperature- and angle-dependent magnetization measurements reveal a large isotropic EB field of more than 1000 Oe at 20 K, which is among the highest values reported for Co/Co_1−yO systems. While several studies have been published about different aspects of Co/Co₃O₄ EB systems, most of them only investigate one specific field-to-sample orientation. Nearly isotropic EB and coercive fields were, to the best of our knowledge, not yet reported for highly crystalline and thus anisotropic Co/Co₃O₄ EB systems. In addition, the EB described in this work is very large, making the system highly relevant for technical applications.

The sample was grown epitaxially by molecular beam epitaxy (MBE) on an Al₂O₃(0001) substrate (dimensions 10 × 5 mm²; space group: $Fm3̄m$⁠) in the stacking order Al₂O₃/Co/Co₃O₄. The Co₃O₄ layer was grown at an oxygen partial pressure of p(O₂) = 5 × 10⁻⁶ mbar in the UHV MBE chamber, resulting in over-oxidation as compared to pure CoO. For comparison, Beschoten et al. found that a partial oxygen pressure of p(O₂) = 3.3 × 10⁻⁷ mbar during growth by MBE results in an unintentionally diluted CoO layer, while p(O₂) = 5 × 10⁻⁶ mbar led to optimal dilution regarding the EB shift.¹¹ 

X-ray diffractometry (XRD) and x-ray reflectometry (XRR) measurements were performed with a X’Pert Pro MPD PW3040-60 diffractometer (PANalytical) using Cu Kα radiation (⁠$λ=1.54056Å$⁠).

The surface chemistry of the samples was examined by x-ray photoelectron spectroscopy (XPS) in an Omicron Multiprobe Ultra-High Vacuum system (Scienta Omicron) at 7 × 10⁻¹¹ mbar using monochromatic Al Kα irradiation. The electron emission angle was 77° and the source to analyzer angle was 54.7°. The core-level spectra were recorded with an acquisition time of 0.8 s per 0.05 eV with a pass energy of 15 eV in a constant analyzer energy mode. The sample was grounded via a small piece of conductive carbon tape. The Fermi level cutoff was observed at −0.28 eV binding energy, and no referencing was performed. The spectrum contained a relatively strong carbon signal originating from adventitious carbon. No sputter etching was performed to clean the sample from adventitious carbon as Co₃O₄ can be reduced to CoO by ion bombardment.²³ The CasaXPS software (version 2.3.22PR1.0) was used for peak deconvolution. All Co 2p components were fitted on an E Tougaard background with symmetric Voigt line shapes GL(30) except for the metallic Co components, which were fitted using Lorentzian asymmetric line shapes LA(1,5,200). All spin–orbit-split peak pairs, including the satellite features, were constrained by the known ratio of 1:2 for p-orbitals. The Co²⁺ to Co³⁺ ratio was set to 1:2 according to the expected Co²⁺(Co³⁺)₂(O²⁻)₄ structure. The O 1 s components were fitted with GL(40) line shapes. The component ascribed to adsorbed species is broader than the others as C bonded O of adventitious carbon is known to be broadly distributed between 530 and 534 eV.

The microstructure of the sample has been investigated by transmission electron microscopy (TEM) employing a JEM 2200FS (JEOL) operated at 200 kV. The TEM is equipped with an omega filter for electron energy loss spectrometry (EELS) and an energy dispersive x-ray spectrometer (EDX) for elemental mappings. The cross-sectional lamella has been prepared using a FEI Helios dual beam FIB operated at 30 kV beam energy with a subsequent ion polishing step at 5 kV.

The in-plane (IP) and out-of-plane (OOP) magnetization measurements of the thin film sample were performed using the VSM mode of a MPMS 3 magnetometer (Quantum Design). The sample was cooled down from 350 K in the presence of 1 T magnetic field before starting the isothermal field loop measurements M(μ₀H) for each temperature.

The film growth during MBE was monitored by in situ reflection high-energy electron diffraction (RHEED) (patterns shown in Fig. S1 of the supplementary material). Its crystal structure was determined by XRD. Both RHEED and the specular θ − 2θ XRD scan confirm that Co (space group: $Fm3̄m$⁠) and Co₃O₄ (space group: $Fd3̄m$⁠) grow in the fcc(111) orientation consisting of two sublattices with 60° azimuthal difference. This structural domain twinning results from two incompatible stacking orders, ABC and ACB, as illustrated in Fig. 1(a). The degree of twinning in Co and Co₃O₄ was estimated using off-specular XRD texture mapping. The mappings were carried out for an IP sample rotation of Φ = 360° and an OOP sample tilt of Ψ = 〈50°, 60°〉. The coordinate system, as shown in Figs. 1(c) and 1(d), is chosen in such a way that the sapphire {$101̄2$} peak is located at Φ = 0° upon a sample tilt of Ψ = 57.60°.

The diffraction peaks of Co{002} at 2θ = 51.37° and Ψ = 54.7° (green and blue for the different phases) and of Co₃O₄{004} at 2θ = 44.82° and Ψ = 54.7° (light and dark purple) are shown in Figs. 1(c) and 1(d), respectively. The diffraction peaks of both layers are located at the same Φ-positions, indicating that the Co and Co₃O₄ layers share the same crystallographic orientation with respect to the substrate. Also shown in Fig. 1(c) are three sapphire {$202̄4$} peaks (red), which occur at Ψ = 57.6°.

Figure 1(e) shows θ − 2θ scans of the Co₃O₄(004) and Co(002) diffraction peaks of the twinned phases A and B at Φ = 30° and Φ = 90°, respectively. The amount of twinning can be determined using the ratio of the integrated peak intensities corresponding to phases A and B. For the Co₃O₄ and Co layers, the amounts of twinning are estimated to be 100% (i.e., 50% phase A and 50% phase B) and 99% (i.e., 51% phase A and 49% phase B), respectively, within a margin of error of 2%. Both layers can therefore be regarded as fully twinned.

The XPS measurements, as shown in Fig. 2(a), are in good agreement with the expected Co₃O₄ structure. Beneath a layer of atmospheric carbon and an outer Co²⁺(OH⁻)₂ hydroxide layer, the lattice O²⁻ to Co^2+/3+ atomic ratio resulting from the fit model is 0.73, just below the nominal value of 0.75. The spectrum suggests the presence of trace metallic Co (about 2.5% of the total Co signal), which is unlikely to originate from the underlying metallic Co layer as the oxide layer is too thick. The O²⁻ to total Co atomic ratio resulting from the fit model is 0.75. Although these results somewhat depend on fit constraints and background type, the measurement confirms the Co₃O₄ stoichiometry within a reasonable margin of error.

Figure 2(b) shows XRR data with the layer thicknesses d, densities ρ, and roughnesses σ obtained as fit parameters. A satisfactory fit could only be achieved considering an additional top-layer of hydroxide and adventitious carbon as revealed by XPS. The resulting thicknesses are d = 7.5 nm and d = 22.2 nm for Co and Co₃O₄, respectively. The interface roughnesses steadily increase from the substrate to the top-layer (fit parameters of the top-layer: d = 0.5 nm, σ = 3.0 nm). The rather low density of the Co layer may indicate lattice defects at the grain boundaries as implied by the TEM images.

The TEM images in Fig. 2(c) show the sample cross section with multiple grain boundaries. The images are in good agreement with the XRR and XRD measurements. The corresponding electron energy loss spectrum (EELS) in Fig. 2(d) shows the characteristic Co L₃ and Co L₂ ionization edges. In the spectrum measured in the Co layer (EELS_1), the peaks occur at 782 and 798 eV. In the spectrum measured in the Co₃O₄ layer (EELS_2), the peaks are shifted by about 1.5 eV toward higher energy. The EDX mapping in Fig. 2(d) reveals that the boundaries between Co grains are oxidized down to the substrate. The comparatively lower oxygen concentration between the Co grains indicates but cannot with certainty be attributed to oxygen deficiency of Co₃O₄ (possibly in the form of CoO) as the grain boundaries may overlap with Co grains perpendicular to the plane of the TEM lamella.

Figure 2(e) shows representative SEM images of the sample surface. The structural domains (light regions) are about 50–100 nm in lateral dimensions. They are clearly distinguishable by randomly oriented grain boundaries (dark regions).

Figures 3(a)–3(c) show magnetic hysteresis loops measured at an angle between the external magnetic field and the [$101̄0$] orientation of the Al₂O₃ substrate of 0°, 45°, and 90°, corresponding to 0°, 15°, and 30° with respect to the nearest Co[002] magnetic easy axis in one of the twinned lattices; OOP magnetization curves are shown in Fig. 1(d). Generally, all IP measurements showed similar sets of hysteresis loops for the investigated temperatures between 10 and 300 K.

A small dent in the magnetization reversal from positive to negative saturation after crossing zero magnetic field (indicated by green arrows) is clearly visible at the 15° field-to-easy axis orientation in Fig. 3(b), less so at 0° in Fig. 3(a), and not at all at 30° in Fig. 3(c). Such an asymmetry is often recognized in EB systems and commonly attributed to different magnetization reversal processes on either side of the hysteresis curve.²⁴ In thin-film systems, it might be caused by the relaxation of the magnetization toward the nearest easy axis near the zero external magnetic field. This effect is especially well known from systems such as Fe/MnF₂ where the FM magnetization relaxes toward the nearest easy axis at vanishing external magnetic field.²⁵ However, in Co/Co_y−1O systems, it is usually less pronounced than in systems with strong magnetic anisotropy of the AFM.²⁵ Here, the dent is correspondingly strongest for a field-to-sample orientation of 45°, as seen in Fig. 3(b), i.e., only 15° to the nearest easy axis, so that relaxation can easily occur at remanence. This becomes visible as rotation of the magnetization away from the measurement axis and thus as a dent. On the other hand, no such dent is visible for the 90° field-to-sample orientation, where the external magnetic field is oriented parallel to an easy axis, as seen in Fig. 3(c), so that no rotation of the magnetization occurs at remanence.

Although the saturation magnetization is consistent for all IP measurements, it is smaller by three orders of magnitude in the OOP magnetization curves. It is well known that only a small OOP component of magnetization can be expected in Co/CoO(111) systems because the AFM moments are oriented at 8° OOP.²⁶ Typically, OOP magnetization in Co/CoO or Co/Co₃O₄ EB systems can be triggered by a nanostructure of the sample^27,28 by using Co/Pt²⁹ or Co/CoO multilayers,³⁰ but it has also been reported for oxygen ion implantation in Co thin films where a strong OOP magnetization of the top surface was found, while the magnetization in the lower layers was oriented IP.³¹ For the case of such buried uncompensated spins, a micromagnetic simulation by Menéndez et al. attributed the training effect (i.e., the reduction of the EB for subsequent hysteresis loops) and asymmetric magnetization reversal to the perpendicular anisotropy.³² 

Only a few investigations of the OOP EB in Co/Co_0031−yCo thin-film systems can be found in the literature. Wu et al. prepared granular Co films with a naturally oxidized CoO layer on top and reported much smaller EB and coercive fields than those in the IP orientation, as well as a positive EB for a 3.6 nm thick film, of which ∼2 nm were oxidized.³³ The FM and AFM layers in our study were much thicker and epitaxially grown; therefore, such effects were not observed. Ovejero et al. investigated thick magnetron-sputtered Co/CoO layers and found larger coercive fields in OOP than in IP magnetization measurements.³⁴ Similar to our results, they reported a two-step magnetization reversal for OOP magnetization measurements. This finding was attributed to a preferential EB direction for an intermediate cooling field, as no two-step process was observed when the AFM surface was fully oriented at higher cooling fields. Comparing these samples with either much higher or much lower layer thicknesses than in this study, we found smaller coercive fields and a two-step magnetization reversal process in OOP measurements, combining effects of both aforementioned studies.

Unexpectedly, the EB, as depicted in Fig. 3(e), is virtually identical in all IP field orientations and very similar to the OOP hysteresis loops. The EB values are also larger compared to those previously reported in the literature for similar samples and temperatures.^11,14–16 The blocking temperature is ∼260 K, i.e., slightly below T_N of CoO (293 K), but significantly higher than T_N of Co₃O₄ (∼40 K). The occurrence of an EB above T_N of Co₃O₄ in Co/Co₃O₄ samples has been reported in the literature and is usually attributed to the presence of CoO at the FM–AFM interface.^21,22 If that is the case, its amount is insufficient to be detected by XRD. However, the apparent disorder and oxygen deficiency at the grain boundaries, as revealed by TEM measurements, may indeed suggest the presence of CoO.

The coercivity (see Fig. S3 in the supplementary material) is very similar for all IP measurements. From Brillouin light scattering measurements of Co(111)/CoO(111) samples at different temperatures, it is well-known that a weak sixfold magnetic anisotropy occurs below and above the blocking temperature.⁵ While epitaxial Co/CoO and Co/Co₃O₄ grown on MgO(110) substrates with twofold anisotropy was reported to result in angle-dependent EB and coercivity,⁵ no such reports were found for the angle-dependence in Co(111)/CoO(111) or Co(111)/Co₃O₄(111).

In this study, FC was performed in an intermediate cooling field. Taking into account the suggestion of Ovejero et al.³⁴ that intermediate cooling fields only partially align the AFM domains, it could be speculated that in the case of epitaxially grown films with more than two energy minima, i.e., higher than twofold structural anisotropy, the orientation of the AFM domains is distributed in orientations parallel to the easy axis, which are nearest to the cooling field direction. In other words, FC along 90°, which is an easy axis, would result in all AFM domains being oriented in the 90° direction, resulting in the EB being oriented in the 90° direction. FC along 0°, on the other hand, would result in the AFM domains being oriented along −30° or +30°, i.e., along the easy axes nearest to the cooling field direction. This would result in the EB being oriented in the 0° direction, as expected, with a similar value as for FC along 90°. In EB systems with twofold symmetry, on the other hand, FC along the hard axis should result in a much weaker EB compared to FC along the easy axis, as it was, indeed, previously reported.⁵ 

The random distribution of grain boundaries and correspondingly of magnetic domain wall orientations, as seen in the SEM images in Fig. 2(e), could further contribute to the isotropy of the EB. While such a fractal domain structure was also found for AFMs with different properties in Monte Carlo simulations, serving as the theoretical basis for the domain state model,³⁵ it was not yet used to calculate the impact on the angle-dependence of the EB.

Finally, Fig. 3(f) shows the training effect for the first four hysteresis loops measured after FC to the respective temperatures. The training effect is often attributed to rotatable uncompensated interface spins.^36–38 For Co/CoO systems, other explanations include the formation of a stable AFM multidomain state after the first hysteresis measurement,³⁹ generally irreversible changes of the AFM domain upon measurements,⁴⁰ or even a change of the reversal mechanism in the descending branch after the first loop.^41,42 The latter was shown to depend on the creation of interfacial domains during the first reversal.⁴² Especially at lower temperatures, a strong reduction of the EB from the first to the second field cycle was observed, which was often interpreted as an athermal effect.^13,43,44 Binek et al. showed that this behavior can be better modeled by their equation based on the discretized Landau–Khalatnikov equation than by a power law and can be attributed to the spin configurational relaxation at the interface.^45,46 This model was fitted to the training effect data, as shown by the straight lines in Fig. 3(f).

We examined the field-direction-dependent exchange bias in a twinned epitaxial Co(111)/Co₃O₄(111) bilayer system grown on a sapphire(0001) substrate using molecular beam epitaxy. The measured EB of over 1000 Oe is among the highest values reported for Co/Co_1−yO systems. Unexpectedly, field cooling along a hard, an easy, and an intermediate axis resulted in virtually isotropic EB fields below the blocking temperature of ∼260 K. We attribute this unexpected isotropy to a splitting of the exchange bias direction into the Co[002] easy axes nearest to the cooling field direction when field cooling is performed along a hard axis.

The supplementary material contains the following: 1. RHEED images measured in situ during MBE-deposition, 2. training effect data for 45° and 90° sample-to-field orientations, and 3. coercivity over temperature.

The authors acknowledge the support for the publication costs by the Open Access Publication Fund of Bielefeld University and the Deutsche Forschungsgemeinschaft (DFG). This work was carried out in the framework of the Joint Lab BiBer. The authors would like to thank Prof. Dr. Armin Gölzhäuser and Prof. Dr. Günter Reiss from Bielefeld University for providing access to their laboratories. Furthermore, the authors would like to thank Dr. Karsten Rott and Laila Bondzio for the sample preparation for TEM.

The authors have no conflicts to disclose.

Martin Wortmann: Conceptualization (equal); Formal analysis (equal); Investigation (equal); Visualization (lead); Writing – original draft (lead); Writing – review & editing (lead). Tapas Samanta: Formal analysis (equal); Investigation (equal). Maik Gaerner: Formal analysis (equal); Investigation (equal). Michael Westphal: Investigation (equal). Johannes Fiedler: Formal analysis (equal); Software (equal). Inga Ennen: Investigation (equal). Andreas Hütten: Funding acquisition (equal); Resources (equal); Supervision (equal); Writing – review & editing (equal). Tomasz Blachowicz: Formal analysis (equal); Writing – review & editing (equal). Luana Caron: Funding acquisition (equal); Resources (equal); Supervision (equal); Writing – review & editing (equal). Andrea Ehrmann: Conceptualization (equal); Formal analysis (equal); Investigation (equal); Software (equal); Writing – original draft (equal); Writing – review & editing (equal).

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