Insight on hydrogen injection and GdO x / Co interface chemistry from in operando neutron reflectometry and secondary ion mass spectrometry

Ionic and redox control of magnetism can produce large changes to a variety of magnetic properties using a relatively small voltage. A model structure that continues to be of interest is Pt/Co/GdO x H y /Au, where the Co magnetic layer oxidation state and perpendicular magnetic anisotropy can be toggled using voltage control of proton transport through the GdO x H y electrolyte layer. The hydration of the oxide layer to form a hydroxide phase is the key to improve the speed of these magneto-ionic devices

V C 2023 Author(s). All article content, except where otherwise noted, is licensed under a Creative Commons Attribution (CC BY) license (http:// creativecommons.org/licenses/by/4.0/). https://doi.org/10.1063/5.0128835 Voltage control of magnetic films and interfaces using ionic transport or magnetoionics 1,2 introduces the possibility of inducing large changes in magnetic properties using straightforward lowvoltage gating. Magneto-ionic mechanisms have the potential to dramatically improve the energy efficiency of magnetic memories and spintronic devices. 3 A variety of ions have been used, most notably oxygen, 1,2,4,5 lithium, 6 hydrogen, 5,7-9 and nitrogen, 10 and large changes in properties, such as magnetic anisotropy, 2,7,11,12 magnetization, 2,5,6,11-13 exchange bias, 4,14 spin-orbit torque, 15,16 and control of magnetic textures, 17,18 have been observed. Pt/Co/oxide stacks are common magneto-ionic devices, where Co is the ferromagnetic layer, Pt provides the spin-orbit coupling needed to give the Co film perpendicular magnetic anisotropy (PMA), and the oxide is the ionic conductor and reservoir and also contributes to interfacial PMA through Co-O bonding. 19 Originally, oxygen ion migration alone was assumed to cause ionically induced changes in Co/GdO x . Previous works illuminate hydrogen's role in the redox chemistry of these devices and establish room temperature proton conduction as a mechanism for modulating the Co magnetism. 7,8 Other works have highlighted the importance of humidity and protonic defects in memristors and magneto-ionic devices. [20][21][22][23][24] Protonic defects (OH • O ) are created through the following reaction (given in Kroger-Vink notation) when oxides absorb water from the atmosphere: [25][26][27] Protons in oxides typically conduct through the Grotthuss mechanism, which consists of two steps: proton reorientation around the oxide ion and hopping to the next oxygen site. 26 Protonic defects are especially important in rare earth oxides, which are very hygroscopic 28 and react with water to form hydroxides such as Gd(OH) 3 . 29 This phase transition has been shown to change ionic 30 and electrical 29 properties. Prior works show that positively gating a Pt/Co/GdO x /Au stack (the Au electrode adjacent to air at high potential) splits water from the ambient water vapor and pumps protons into GdO x . The electric field drives protons to the Co/GdO x interface, interrupting the interfacial PMA and causing a reversible magnetic reorientation transition (MRT) to an in-plane state. 7 In devices with an initially oxidized Co layer (CoO), pumping in hydrogen reduces CoO to Co. 7,8 The reverse process happens under negative bias with H 2 O oxidizing Co. 7 Previous works used x-ray reflectivity (XRR) to show that during hydration of a GdO x film in a humid environment, a layer of Gd(OH) 3 grows from the surface, and without this hydration step, the device does not function. 8 Other works have used x-ray and neutron reflectivity to show similar protonated surface layers in protonconducting oxide films (e.g., In-doped BaZrO 3 ). 31 However, x-ray probes are relatively insensitive to hydrogen and instead probe changes in the electron density (primarily the Gd density). Unlike other elements in the system, hydrogen has a negative neutron scattering length, so significant protonation sharply suppresses the neutron scattering length density (SLD). Polarized neutron reflectometry (PNR) is also sensitive to the in-plane net magnetization depth profile. PNR is, therefore, an ideal way to examine hydrogen movement and magnetization changes. We first measured the hydrogen depth profile in a GdO x film using unpolarized neutron reflectometry (NR) and then used PNR to probe the hydrogen depth profile and the magnetism in the Pt/Co/GdO x /Au stack during in operando positive biasing. We verified the results with ex situ secondary ion mass spectrometry (SIMS) after positive and negative gating.
First, we investigated the hydration of an unpatterned GdO x film. A 1 cm 2 GdO x film was grown on a Si substrate with a thermally grown oxide layer. The film was hydrated at 50 C and 90% relative humidity for 20 h and was then vacuum annealed at 100, 200, and 250 C to drive the water out, all while NR scans were performed continuously. Each condition was sustained until no drift occurred between scans (approximately 2 h). After stabilizing at 100 C, the reflectivity did not return to its initial shape, so we stepped up the temperature until the reflectivity fringes more closely matched the initial value, at which point we assumed most water had been driven out. Figure 1 shows the neutron reflectivities and associated SLD profiles under each condition. Note that throughout this manuscript, error bars and uncertainties represent one standard error. The as-grown GdO x film thickness is 29.58 6 0.14 nm, including a 3.94 6 0.07 nm layer with reduced SLD at the surface, indicating some combination of oxygen deficiency and partial hydration from exposure to ambient humidity. Such surface regions have been seen in proton conducting perovskites. 31 After hydration, the neutron SLD near the air interface decreases further and the thickness of the low-SLD region increases. The hydrated GdO x layer is best fit by a bilayer oxide/hydroxide structure with a hydroxide layer thickness of 11.64 6 0.09 nm for a total   Fig. S1. This bilayer structure confirms that hydration occurs through a phase transition to Gd(OH) 3 with a front that extends deeper into the film with longer hydration time (see Fig. S2 for NR reflectivities displaying hydroxide expansion over time). The 13.8% thickness increase during 20-h hydration is consistent with the previously reported 15% thickness increase for 24-h hydration of a similar film measured with XRR. 8 Several factors simultaneously affect the hydroxide SLD. As the phase transition to Gd(OH) 3 occurs, the overall material density decreases from 7.41 (bulk Gd 2 O 3 ) to 6.0g/cm 3 [previously measured density of Gd(OH) 3 ]. 8 Although hydrogen incorporation and lattice expansion during the phase transformation are inherently connected, it is useful to consider their effects on the SLD separately. From the decreased number density of Gd and O atoms in the hydroxide phase alone, we expect an SLD decrease in 19% to a value of 1.98 Â 10 À4 nm À2 . Hydrogen incorporation suppresses the SLD further; SLD values significantly below 1.98 Â 10 À4 nm À2 in the hydrated sample, therefore, argue strongly for increased H concentration. The final hydroxide layer SLD is slightly higher than the expected value of 1.05 Â 10 À4 nm À2 [gray dotted line in Fig. 1(a)], meaning the Gd and O density is higher than that expected (perhaps due to some remaining Gd 2 O 3 ) or the H concentration is lower than that expected for stoichiometric Gd(OH) 3 .
When vacuum annealed at 100 C, the hydroxide layer SLD increases as hydrogen is driven out. The film thickness decreases above 200 C with further reduction in the hydrogen concentration in the hydrated layer. At 250 C, the GdO x SLD has returned to the as-grown value. Interestingly, the film thickness remains larger than its original value. The 250 C state does not have a low-SLD surface layer, meaning that between the as-grown and 250 C states, H in the surface layer of the as-grown film has been driven out and the oxygen content of that surface layer has increased to match the rest of the film. Oxygen incorporated as the film transforms into a hydroxide [according to Eq. (1)] may not be removed during vacuum annealing. The prior XRR results agree that after 6 days of hydration, the film expanded by 50% and decreased its SLD by 28%, giving a 10% increase in the total integrated x-ray SLD consistent with additional oxidation of the film. 8 The low SLD tail in each profile above Z ¼ 79 nm is probably due to adsorbed hydrocarbons, which desorb above 200 C. The 250 C profile, thus, shows a sharp interface instead of a tail.
We now turn to the Pt/Co/GdO x /Au patterned devices and examine how hydration and voltage gating affect their structure, hydrogen profile, and magnetism. The Pt/Co/GdO x /Au devices are susceptible to pinholes; so in order to have a device area large enough for PNR, we used an array of 30 1 mm 2 square crossbar devices (see the sample schematic in Fig. S3). We measured the PNR in the asgrown state, after hydration, and during in operando biasing at þ3 and þ10 V. The bias voltages were held constant, and PNR scans were done continuously until the reflectivities stabilized for the duration of one scan ($12 h). Figure 2 shows the PNR reflectivities and the SLD profiles for all conditions.
After scanning the as-grown state, the sample was hydrated for 6 days at 90 C, since previous XRR data found that the thickness change was complete within that time. 8 A saturated K 2 SO 4 solution maintained the relative humidity at 95%. Comparing the as-grown and hydrated SLD profiles in Fig. 2(b), we see that the GdO x layer becomes approximately 25% thicker with hydration. This change is not as dramatic as was seen in the XRR measurement, which could be explained by the Au electrode preventing hydrogen incorporation. Hydration would then have occurred through lateral diffusion and may have been slower or incomplete. This could also explain the lack of the hydrated layer at the top of the GdO x film, unlike what was observed after hydration of the unpatterned film. Lateral diffusion may play a larger role at the Co/GdO x interface, which could explain the decreased GdO x SLD next to the Co layer in the hydrated state. Alternatively, increased defect concentration near the bottom interface may cause preferential H occupation. The GdO x SLD starts near the expected bulk value for Gd 2 O 3 , 2.44 Â 10 À4 nm À2 , and approaches the expected Gd(OH) 3 SLD value of 1.05 Â 10 À4 nm À2 after hydration and gating. The low GdO x SLD near the Co layer after gating is likely due to the low hydrogen solubility in Pt, which makes hydrogen accumulate near the Co interface as it is pumped in. It could also be that a hydroxide layer forms at the bottom of the GdO x layer and gets thicker as more H is pumped in.
The oxide-to-hydroxide transformation is corroborated with SIMS, which shows increased signal from GdO x H y ions after gating compared to the as-grown state (Fig. 3). The SIMS devices were unhydrated with any initial hydrogen in GdO x coming from ambient water vapor. The presence of GdO x H y only after gating, therefore, confirms that significant additional hydrogen enters the oxide during positive biasing. The SIMS signal for H 2 O decreases with positive bias, likely since H 2 O on the surface (and also possibly in the oxide's grain boundaries) is split and incorporated into the hydroxide.
Au exhibits a slight thickness increase and a significant SLD decrease upon hydration, which is consistent with water and hydroxide accumulation on the surface. The Au layer roughens significantly and decreases further in thickness upon gating (corroborated with atomic force microscopy in Fig. S4), which is consistent with Au oxidation 32,33 or surface adsorbates. An alternative interpretation is that GdO x H y expands nonuniformly, causing a roughened interface that spreads Au out in Z, giving an apparently thicker, lower SLD film. The increase in diffuse scattering from the Au electrode after gating can be seen by eye, consistent with the significant increase in roughness in the fit. We note that a significant amount of water or hydrogen adhering to this layer may add apparent roughness unrelated to a change in the structure of the underlying gold. Thus, there is considerable uncertainty regarding the gated Au roughness values presented here.
In the as-grown and hydrated fits, the Co layer SLD is 4.280 6 0.017 Â 10 À4 and 4.282 6 0.013 Â 10 À4 nm À2 , respectively, closely matching that of bulk CoO. Partial or complete Co oxidation is unsurprising, as it was exposed to oxygen during reactive sputtering of GdO x . Upon gating, the Co layer SLD decreases to 1.44 6 0.10 Â 10 À4 nm À2 . While we expect both reduction of CoO and incorporation of hydrogen to decrease the Co layer SLD, the optimized value is significantly below the bulk Co value of 2.27 Â 10 À4 nm À2 , which is consistent with significant hydrogenation. For additional context, we calculate that the bulk SLD of cobalt hydroxide [Co(OH) 2 ] is 1.54 Â 10 À4 nm À2 while the optimized values for the þ3 and þ10 V gated conditions are 1.44 6 0.10 Â 10 À4 nm À2 and 1.111 6 0.071 Â 10 À4 nm À2 , respectively. Co reduction under positive bias was previously reported in this system, and H incorporation is what depressed the SLD of the surrounding GdO x , so both processes likely contributed in reducing the Co layer SLD. Increased porosity in Co upon reduction Applied Physics Letters ARTICLE scitation.org/journal/apl could also explain the lowered SLD. However, the SIMS data in Fig. 3 show a small Co À signal that decreases upon gating by 61%, while the CoOH À and CoOH 2 À signals increase by 217% and 328%, respectively. Furthermore, while SIMS confirms the presence of CoO compounds in the device, the gating does not result in a significant change in the integrated CoO intensity. Instead, the largest variations in the SIMS intensity are confined to hydrogen-containing species, supporting hydration as the major cause of SLD reduction over increased porosity, although likely both play a role. Note that we looked extensively in the SIMS data for evidence of cobalt hydrides (CoH x ) and found none. CoH x compounds form only at extremely high pressure, so while we see evidence of H in the Co layer, a CoH x phase is unlikely. 34 We found several unexpected results in the magnetic portion of the PNR and in the SIMS data, which lead us to a more complex model of the chemistry of the Co magnetic layer. Our first unexpected finding is that in all states, as-grown, hydrated, and gated, the Co layer had much smaller magnetic SLD than expected-only 15% the bulk

ARTICLE
scitation.org/journal/apl Co value of 1400 emu/cm 3 (1 emu/cm 3 ¼ 1 kA/m), which corresponds to an expected magnetic SLD of 4.12 Â 10 À4 nm À2 . An in-plane field of 700 mT was applied to pull the Co magnetization in-plane, so any magnetization should be visible in the magnetic SLD profile [ Fig. 2(b)]. One possible explanation is that only a few devices contained unoxidized Co. However, constraining the fit to have a very small "active device" area with metallic ferromagnetic Co worsens the fit significantly, so this is not likely. Enough Co was present in all as-grown devices to measure a polar MOKE signal displaying PMA (see Fig. S5). Thus, in the as-grown state, the diminished Co magnetization can be explained by a mixed phase of CoO and Co in accordance with the structural analysis. The second unexpected result is that the Co magnetization grows after hydration and decreases during biasing. To understand why the magnetic SLD is fit this way, we examine the Fresnel spin difference, or the difference of the non-spin-flip reflectivities ("" minus ##), normalized by the theoretical reflectivity of the bare Si substrate. Figure 2(a) shows the spin-dependent neutron reflectivities for the four measurement conditions, and Fig. 4 shows the Fresnel difference of those reflectivities. Figures 4(b)-4(e) show detailed views of the lowq regions. The amplitude of the oscillations in the Fresnel difference is proportional to the total magnetization in the film. The Fresnel difference oscillates in all conditions, and for the first four (lowest q) peaks, the amplitude of the oscillation in the positive biased state is smaller than the amplitude in the hydrated state, implying that the hydrated state must have a larger magnetic SLD than the as-grown and biased states. Figure S6 gives the statistical analysis of these data. Based on previously reported MOKE measurements, we would expect CoO to be reduced to Co with the presence of H 2 O under positive bias, which (absent any other chemical or structural changes) should increase the net magnetization. It is possible that some CoO is in fact reduced, but that other effects, discussed below, obscure the effect on the measured magnetization.
The third unexpected result is that the Fresnel spin difference below the critical edge (Q crit ¼ 0.1 nm À1 ) is nonzero [shown clearly in Fig. S6(a)], indicating that some magnetization is co-located with a neutron absorber (Gd) in the film. Co and GdO x were thought to be distinct layers, while GdO x itself is not magnetic. The best description of this feature appears when a layer of GdO x near the Co/ GdO x interface is allowed to express a net magnetization in the model. This interfacial magnetization is notably reduced upon gating. One explanation for magnetized Gd is an intermixed sublayer of Co and GdO x (possibly also including Pt). Paramagnetic behavior in Co-doped nanocrystalline GdO x at room temperature has been observed. 35 The intermixing could have occurred during sputtering, which is common especially for heavy elements, 36 or possibly during hydration, when the sample was held at 90 C for 6 days. Intermixing can explain the suppression of Co magnetization below the bulk value in the as-grown and hydrated states, especially if the Co and Gd develop antiparallel moments. It could similarly explain the increase in magnetization upon hydration and the suppression during gating. If indeed Co and Gd intermixing suppresses magnetization in the as-grown state, deconvoluting the chemistry controlling the net moment becomes nontrivial with Co and Gd reacting with O and H in ways which may have offsetting effects.
The SIMS results lend credence to this picture. The SIMS shows that CoO x H y ions are present and the total signal from CoOH À and CoOH 2 À increases after positive gating and decreases back to initial values upon negative gating. These results imply that some oxygen remains in the Co layer (or an intermixed layer) and the hydrogen driven in during positive gating penetrates this layer and forms a cobalt hydroxide phase. We conclude that the model of toggling between completely metallic and completely oxidized states is not accurate. In reality, CoO, Co, and CoO x H y phases may co-exist with the relative prevalence of each phase changing under different hydration and gating conditions. These complexities would not be observable in MOKE measurements, which primarily probe the metallic Co. The Pt/Co/GdO x /Au devices in which CoO/Co toggling was previously measured using MOKE 7,8 may or may not have displayed Pt, Co, and Gd intermixing and mixed Co, CoO, and CoO x H y phases. As long as some metallic Co was present, MOKE could still show transitions between a CoO/CoO x H y nonmagnetic state and magnetic states containing metallic Co.
The best fit also includes slight paramagnetism in the rest of Gd, including regions that are not next to Co, for all conditions. The hydrated condition has the highest fit value, 1.82 6 1.19 emu/cm 3 . GdO x nanoparticles have been shown to be paramagnetic at room temperature, so slight paramagnetism in the nanocrystalline GdO x film is reasonable. 37 We set out to investigate the effects of hydration and voltage gating on the structure and magnetization of the Pt/Co/GdO x H y /Au system. The structural PNR results showing depressed SLD due to hydrogen and the SIMS result showing signal from GdO x H y ions are the first direct measurements of hydrogen in this system; along with the NR results, they support the models for phase transformation from oxide to hydroxide during hydration of GdO x and for water splitting and proton incorporation during positive biasing. NR results show an increase in the total integrated GdO x layer SLD after hydration and vacuum annealing, which we attribute to irreversible uptake of oxygen. During voltage gating, H accumulates in the Co layer and in the GdO x layer near the Co/GdO x interface. The hydrogen concentration, the depth of the phase transformation front, and the amount of GdO x H y compounds all increased further with gating, indicating that incomplete hydration can be compensated by adding hydrogen during biasing.
The prevailing model of this system suggested that voltage gating induced complete transformations among a nonmagnetic CoO state, an in-plane Co state with H loaded at the GdO x H y interface, and a PMA Co state with no H loaded. This work complicates that model. The PNR and SIMS results reveal the presence of CoO, CoO x H y , and a smallerthan-expected amount of Co as well as a possibly intermixed layer of Pt, Co, and Gd that suppresses the net magnetization. Lower temperatures during hydration or more optimized deposition could reduce intermixing and the evolution of unwanted species. Furthermore, it may be possible to substantially tune the switching behavior and magnetization changes through judicious selection of the initial Co state. A partially oxidized CoO layer, for example, appears to evolve some fraction of Co hydroxide species, while such reactions may be suppressed in completely metallic Co. This work sheds light on the complicated interface chemistry and chemical irreversibility crucial to understand and tune voltage-controlled magneto-ionic devices.
See the supplementary material for figures illustrating the alternate gradient NR fit for the unpatterned film, the expansion of the unpatterned film over time during hydration, a schematic and initial MOKE loop for the sample used for the PNR experiments, a statistical analysis of the Fresnel spin difference of the PNR reflectivities, and the SLD profiles for all nonactive regions on the PNR sample. It also includes the experimental details for all measurements performed in this work, including a detailed discussion of the fitting parameters and constraints used in the PNR analysis of this multi-region, multi-condition system.