Beyond the Interface Limit: Structural and Magnetic Depth Profiles of Voltage-Controlled Magneto-Ionic Heterostructures

Electric-field control of magnetism provides a promising route towards ultralow power information storage and sensor technologies. The effects of magneto-ionic motion have so far been prominently featured in the direct modification of interface chemical and physical characteristics. Here we demonstrate magnetoelectric coupling moderated by voltage-driven oxygen migration beyond the interface limit in relatively thick AlOx/GdOx/Co (15 nm) films. Oxygen migration and its ramifications on the Co magnetization are quantitatively mapped with polarized neutron reflectometry under thermal and electro-thermal conditionings. The depth-resolved profiles uniquely identify interfacial and bulk behaviors and a semi-reversible suppression and recovery of the magnetization. Magnetometry measurements show that the conditioning changes the microstructure so as to disrupt long-range ferromagnetic ordering, resulting in an additional magnetically soft phase. X-ray spectroscopy confirms electric field induced changes in the Co oxidation state but not in the Gd, suggesting that the GdOx transmits oxygen but does not source or sink it. These results together provide crucial insight into controlling magnetic heterostructures via magneto-ionic motion, not only at the interface, but also throughout the bulk of the films.


Introduction
With exciting potentials of energy-efficiency and new functionalities in hybrid magnetoelectric devices, voltage control of magnetism is currently the focus of intense investigations. [1][2][3][4][5] Traditional spintronic architectures use electron spin for information storage and transmission, e.g., utilizing the giant 6,7 or tunneling 8,9 magnetoresistance, spin transfer torque, [10][11][12] or the spin Hall effects. 13,14 Recently, an alternative magneto-ionic approach has been demonstrated, e.g., in GdO x /Co Hall bar structures, that uses an electro-thermal conditioning sequence to drive oxygen from an oxide film into a neighboring ferromagnetic metal layer [15][16][17][18] and directly tailor the interface chemistry. In so doing, interfacial physical properties such as magnetic anisotropy and saturation magnetization are effectively controlled by the electric field, even reversibly. The most prominent effect of the magneto-ionic motion is the modification of the interface -indeed the ultra-thin Co films in these studies are less than five monolayers thick, and are consequently dominated by the interfacial properties. Considering the strong reduction potential of gadolinium 19 and relatively good chemical stability of cobalt oxide 20 , the observed reversibility cannot be explained by simple enthalpy considerations alone, suggesting that the bulk and interfacial behaviors may be different. However, probing the magneto-ionic motion in the bulk of the film is intrinsically challenging.
In this work we report direct mapping of structural and magnetic depth profiles in voltage-controlled magneto-ionic heterostructures with relatively thick (15 nm) films of Co, and demonstrate that the electrically-induced oxygen migration extends far beyond the interface.
Using polarized neutron reflectometry (PNR) the electrically-induced oxygen migration, manifest as increased nuclear scattering length density (SLD) and reduced magnetic SLD, is directly mapped. PNR depth profiles show that the electric field drives oxygen deep into the Co film (>10 nm), reducing the magnetization by more than 80% at the interface and 38% in the bulk. After reversing the polarity of the applied electric field, the magnetization recovers to 92% of the original value throughout the Co layer. Thus interface and 'bulk' behaviors are demonstrated to be different. Using the first order reversal curve (FORC) technique, we identify two distinct magnetic phases in the post-conditioned sample. This behavior differs significantly from that of the as-grown state, which exhibits a single magnetic phase, suggesting that the oxygen migration significantly alters the magnetic characteristics by impeding inter-granular coupling. Control experiments performed in the absence of an electric field reveal that annealing alone causes much smaller structural and magnetic changes to the Co film, distinguishing the roles of thermal conditioning from electro-thermal ones. X-ray absorption (XA) spectroscopy and X-ray magnetic circular dichroism (XMCD) confirm the role of the electric field in recovering the magnetization and cobalt oxidation states. XA performed on the GdO x shows that the Gd oxidation state is invariant, suggesting that the Gd transmits the oxygen rather than surrendering it.
PNR measurements of the as-grown sample and each conditioned state are shown in Fig. 1(a).
The R ++ and R -reflectivities show sensitivity to the nuclear and magnetic depth profiles, evident by spin-dependent oscillations. The difference in the R ++ and R -is approximately proportional to the quotient of the magnetization and the nuclear SLD (see Methods). Thus, the magnetic contribution to the data is highlighted by plotting the spin asymmetry [SA = (R ++ -R --)/(R ++ + R --)], as shown in Fig. 1(b). The oscillation amplitude first decreases after conditioning in +40 V then increases after conditioning in -40 V, suggesting a decrease of the saturation magnetization, M S , and/or a change in the structure, followed by a partial recovery towards the initial state. The nuclear and magnetic depth profiles from the converged model, shown in Fig. 1(c), confirm these trends.
The extracted depth profile of the as-grown sample accurately reproduces the designed structure, both in terms of thickness and nuclear SLD,  N . Our fits show excellent agreement between the measured and expected  N values of Co (2.27×10 -4 nm -2 ), Pd (4.02×10 -4 nm -2 ), and GdO x (2.74×10 -4 nm -2 ) 21,22 . However, the measured SLD of the thick Al 2 O 3 base layer is substantially lower than the expected bulk value (5.67×10 -4 nm -2 ), suggesting the presence of significant voids or an oxygen deficient stoichiometry. The GdO x -a neutron absorber -can be identified explicitly by the imaginary SLD in Fig. 1(c). After conditioning the sample at +40 V, the nuclear SLD of the Co layer,  N Co , increases by 34%, approaching that of CoO (4.29×10 -4 nm -2 ), which is consistent with incorporation of oxygen. Simultaneously, the GdO x /Co interface becomes much broader, increasing from a width of 3.3 nm to 10 nm and extending well into the Al 2 O 3 . After switching the voltage polarity to -40 V, the GdO x /Co interface width is reduced to 1.9 nm and  N Co decreases, demonstrating an induced migration of oxygen from the CoO x into and through the GdO x . The recovery of  N Co occurs predominantly within the 10 nm nearest the GdO x interface, while the top 5 nm, near the Co/Pd interface, remains unchanged from the +40 V conditioned state. Thus, we observe oxygen ion migration throughout the thickness of the Co layer, but reversibly only within the 10 nm closest to the GdO x /Co interface. However, at the GdO x interface after this second conditioning,  N Co is still 32% larger than the as-grown sample, demonstrating the oxygen migration is only semi-reversible.
Trends in the magnetic depth profile [dashed lines in Fig. 1(c)] agree with those observed in the nuclear profile. Specifically, the as-grown sample has a sharp step-function like GdO x /Co Control experiments were performed on a sample grown side-by-side with the electrothermally conditioned sample, following the same thermal treatment but without an electric field.
PNR measurements of the thermal-only sample, Fig. 2, show that the first thermal treatment also increases  N and reduces  M in the Co layer, but to a much lesser degree and with no significant changes in the GdO x /Co interface shape and extent. Quantitative comparison shows that  N Co increases by 17% and  M Co decreases by 12% after thermal treatment alone compared to the asgrown sample. This is much less than the electro-thermally treated sample, which showed an increase in  N Co of 34% and a decrease in  M Co of 38% after the +40 V treatment. After a second 15 min thermal conditioning the nuclear and magnetic profiles of the control sample do not change appreciably, suggesting a saturation effect or depletion of easily diffusible oxygen. These results confirm the role of the electric field in enhancing the oxidation of the Co layer during the +40 V conditioning and reducing it during the -40 V treatment.

Magnetometry
Magnetic hysteresis loops of the samples as-grown, after both +/-40 V treatment (E+T), and after two thermal-only treatments are shown in Fig. 3(a). The Co M S in the as-grown sample was measured to be 1230 emu cm -3 , in good agreement with the PNR value of 1180 emu cm -3 .
Further, M S is decreased by 10% in the E+T treated sample and 7% in the thermal-only sample compared to the as-grown sample, similar to the PNR data which show a reduction of 10%. The good agreement between the magnetometry and PNR results supports the validity of the model used to fit the data. Sample coercivity and remanent magnetization are also decreased compared to the as-grown sample by 68% and 55% in the E+T sample and 54% and 11% in the thermal sample, respectively, indicating significant changes in the magnetic characteristics.
Details of the magnetization reversal have been investigated by the FORC method. The mT,   H B = 0.17 mT) and is no-longer circularly symmetric, but rather has a 90° bend with symmetries along the +H and -H R axes, typical of a domain nucleation/growth reversal mechanism. 28,29 The FORC distribution for the thermal-only sample shows the same negative feature that again identifies magnetic coupling, and a new feature associated with the observed major loop protrusion. Integrating the FORC features gives a magnetic phase fraction: 24 the reversible phase contributes to 0%, 31%, and 24% of the magnetization in the as-grown, E+T, and thermal-only samples, respectively.
Reversible phases exhibit no hysteresis, and therefore are manifested in the FORC distribution along the   H C = 0 axis, e.g., when the phase has essentially zero coercivity or when the magnetic field is applied along the magnetic hard axis. Major hysteresis loops measured in the out-of-plane direction (see Supplemental Material) for these samples display little hysteresis, implying that the out-of-plane direction remains the hard axis. These results suggest that oxygen migrates in the film after E+T or thermal-only conditioning and segregates to grain boundaries in the Co layer, thus disrupting long-range magnetic correlations and effectively breaking down the affected Co films into isolated grains. Once the coupling between the grains becomes weaker, their respective magnetocrystalline anisotropies in confined grains play a large role in determining the magnetic orientation resulting in much reduced coercivity and remanent magnetization.

X-ray Absorption and Circular Dichroism
Oxidation of the cobalt after both +/-40 V electro-thermal (E+T) and thermal-only treatments is confirmed in the XA and XMCD measurements (see Methods) shown in Fig. 4(a).
Oxidation of the Co layer 30 is identified directly by the emergence of peaks at E = 779.2 eV and 776.8 eV, which are not present in the as-grown profile. The peak at 779.2 eV is largest in the thermal-only sample, indicating significantly increased oxidation relative to the E+T sample.
This trend is supported by the XMCD spectra, which shows that the as-grown sample has the largest dichroism, indicating the largest magnetization. The E+T sample has the second largest dichroism, and the thermal-only sample has the smallest. The different ordering in the dichroism compared to the bulk magnetometry may identify variation in the depth-dependent oxygen binding behavior. This is consistent with the XA results, which showed a larger oxidation peak in the thermal-only treatment than the E+T sample. XMCD signal from the Gd, shown in Fig.   4(b), shows no dichroism for all three samples, indicating a negligible contribution to the magnetization. Interestingly, the XA signal for the Gd shows no significant change for any of the samples, suggesting a relatively constant Gd oxidation state, regardless of oxygen migration into or out of the Co.

Discussion
The PNR, magnetometry and X-ray results clearly demonstrate that electro-thermal conditioning can drive oxygen semi-reversibly into a thick (15 nm) Co film, profoundly changing its magnetic properties. Depth profiling with PNR indicates that while these effects are most prominent at the GdO/Co interface, they also extend throughout the entire 15 nm thick Co film. Reversing the polarity of the applied voltage drives oxygen out of the Co, partly restoring  N Co and  M Co to their original values at the GdO x /Co interface, but leaving  N Co and  M Co unchanged near the Co/Pd interface. Thermal conditioning of the control sample also promotes oxidation of the Co layer, but the supply of highly mobile oxygen that can be moved by entropydriven diffusion is clearly limited. In the following discussion we determine the oxygen stoichiometry from the nuclear scattering profile and consider the underlying mechanics of the oxygen migration.

Oxygen Depth Profile
The role of the electric field and entropy-driven oxygen migration is seen qualitatively by comparing the profiles for the +40 V electro-thermally treated sample with the thermally treated sample, Figs. 1(c) and 2(c) respectively. The thermal treatment is shown to scale the magnetic depth profile relative to the as-grown sample, but not change its shape. In comparison, the magnetic depth profile for the +40 V sample strongly deviates from that in the as-grown sample, suggesting that the electric field drives oxygen into the film, while the thermally activated, entropy-driven oxygen migration is relatively uniformly distributed.
Using the neutron coherent scattering length, b, for cobalt (2.49 fm) and oxygen (5.81 fm), 31 the CoO x stoichiometry can be directly calculated. Specifically, the nuclear SLD is calculated as: SLD = ( ) where N Co(O) is the total number of cobalt (oxygen) atoms within the volume, V, of the Co film. Assuming the as-grown film is pristine Co, which is supported by the good agreement with the referenced bulk  N Co value, and that the cobalt number density remains constant during treatment, a lower-limit to the oxygen profile can be calculated: . Interestingly, the depth profiles (especially the magnetic profile) indicate that the oxygen is semi-reversibly driven out of the GdO x /Co (0-10 nm) interface after treatment in -40 V, while the remaining oxygen is left trapped deeper within the Co. We suggest that as the GdO x /Co interface becomes depleted of oxygen, becoming more metallic. The oxidized region deeper within the film gets surrounded by conductive layers above and below, screening the electric field; without an electric field the oxygen does not migrate, resulting in the observed trapping effect. Thus, we suggest that the observed 10 nm thickness of the Co presents a practical limit for the electrically-driven oxygen migration within these otherwise metallic films.

Mechanics of Oxygen Migration
Brief considerations of the underlying mechanisms quickly reveal that this effect cannot be justified with bulk chemistry alone. First considering the thermally treated sample, the initial The chemical potential within each of the Al 2 O 3 , GdO x and Co films is expected to be uniform, and thus O 2defects are expected to be highly mobile within each. 17 The application of a static electric field then causes field-induced ion migration. 35 However, at the boundary between two layers there will exist a difference in the chemical potential, which may cause an accumulation of oxygen at e.g. the GdO x /Co interface. Considering this issue, the enthalpy of formation is calculated in each layer and compared to the electric potential energy available to overcome this interfacial barrier. Assuming a lattice-site hopping model, 36 occur along the forward reaction, but will not be reversible as it again costs too much energy.
Other considered reactions are presented in the supplemental materials, but no bulk energy calculation was able to support the observed results.
We present an alternative consideration (i.e. treating the grains as nanoscale clusters) that provides a reasonable energy landscape for the observed migration. In this picture, illustrated in

Conclusion
In summary, we report the first direct depth profile mapping of voltage-moderated oxygen migration in magnetic Co thin-films. Using X-ray and polarized neutron reflectometry we observed changes in the structural and magnetic profiles, consistent with partial oxidation of the Co layer. The oxidation and corresponding magnetic changes associated with application of a +40 V treatment were found to be strongest at the GdO x /Co interface. We showed that the interfacial oxidation was largely reversible, and could be driven back to the Co layer with a reversed electric field, while the oxygen deeper in the film remained trapped. The effects of the electric field and thermal treatments were separated by comparing samples conditioned with and without an electric field. Magnetometry using the FORC technique revealed that the treatment altered the magnetic properties of the film, resulting in two distinct magnetic phases. X-ray spectroscopy revealed increased oxidation in the Co film after any conditioning, but less after the electro-thermal treatment with the reversed polarity. Thermal and electric cycling dramatically change the granular structure of the system, providing a means by which the GdO x can easily transport the oxygen. These results provide a depth-resolved view of magneto-ionic motion beyond the conventional interface limit, opening a new avenue to explore their applications in future device concepts. edge, following previously outlined procedures. [48][49][50] Measurements were performed using a constant beam polarization and an alternating in-plane magnetic field of ±200 mT. Signal was detected by fluorescence yield.           To further demonstrate confidence in the presented model, we attempted to fit the data with models with fixed magnetic and nuclear profiles. That is, the nuclear (magnetic) profile is constrained to be constant throughout all of the models, probing whether these results can be the result of only magnetic (nuclear) changes. In both cases the models would not converge to any reasonable value (e.g. physically unreasonable nuclear SLDs and interface widths and thicknesses significantly different from the designed structure).

Other Possible Energies of Formation
Below is a list of potential enthalpy of reactions for bulk gadolinium, cobalt and oxygen, with the calculated enthalpy of reaction. 32,33 Positive enthalpy of reaction indicate the reaction is endothermic and thus must consume at least that much energy from the system. Negative enthalpy of reaction indicates exothermic. In order to be consistent with the observed resultswhich saw reversibility in the interfacial cobalt oxidation, by reversing the electric field -the enthalpy must be on a similar scale to the thermal + electric potential energy (67 meV).
GdO+CoCoO+Gd, H = -2.3 eV All the above reactions -in addition to the ones presented in the paper -suggest that the chemical energies are much larger than the thermal and electric potential energy.

Out-of-Plane Magnetometry
Out-of-plane major hysteresis loops for the samples as-grown, after sequential +/-40 V and thermal-only treatments are shown in Fig. S1. The plot shows no significant hysteresis, and exceedingly small remanence and coercivity for all samples. This behavior identifies that the outof-plane direction is the magnetic hard axis, as expected from the shape anisotropy. The largest hysteresis is observed in the as-grown sample, and may be the result of either residual magnetocrystalline anisotropy, or sample misalignment during measurement. Figure S1. Out-of-plane hysteresis loops for the samples as-grown (black), after sequential +/-40 V (E+T, Red) and thermal-only treatments (Blue)

Depth-Resolved Oxygen Profile
In the main text, the oxygen content was calculated from the fitted SLD of the Co layer and the nuclear scattering lengths of Co and O. Two assumptions were made: (1) the as-grown Co-film was pristine, and (2) the number of Co atoms in the layer volume remained constant.
Both of these are expected to be accurate based on the fitted SLD of the as-grown film and the fitted thicknesses. Expanding on (2) and assuming that the Co atoms do not migrate within the layer, the same approach can be used to calculate the depth-resolved oxygen profile, shown in