Regulating Oxygen Ion Transport at the Nanoscale to Enable Highly Cyclable Magneto-Ionic Control of Magnetism

Magneto-ionics refers to the control of magnetic properties of materials through voltage-driven ion motion. To generate effective electric fields, either solid or liquid electrolytes are utilized, which also serve as ion reservoirs. Thin solid electrolytes have difficulties in (i) withstanding high electric fields without electric pinholes and (ii) maintaining stable ion transport during long-term actuation. In turn, the use of liquid electrolytes can result in poor cyclability, thus limiting their applicability. Here we propose a nanoscale-engineered magneto-ionic architecture (comprising a thin solid electrolyte in contact with a liquid electrolyte) that drastically enhances cyclability while preserving sufficiently high electric fields to trigger ion motion. Specifically, we show that the insertion of a highly nanostructured (amorphous-like) Ta layer (with suitable thickness and electric resistivity) between a magneto-ionic target material (i.e., Co3O4) and the liquid electrolyte increases magneto-ionic cyclability from <30 cycles (when no Ta is inserted) to more than 800 cycles. Transmission electron microscopy together with variable energy positron annihilation spectroscopy reveals the crucial role of the generated TaOx interlayer as a solid electrolyte (i.e., ionic conductor) that improves magneto-ionic endurance by proper tuning of the types of voltage-driven structural defects. The Ta layer is very effective in trapping oxygen and hindering O2– ions from moving into the liquid electrolyte, thus keeping O2– motion mainly restricted between Co3O4 and Ta when voltage of alternating polarity is applied. We demonstrate that this approach provides a suitable strategy to boost magneto-ionics by combining the benefits of solid and liquid electrolytes in a synergetic manner.

W ith the advent of global phenomena such as the Internet of Things, artificial intelligence, machine learning, or Big Data, the demand for highly functional and energy-efficient miniaturized microelectronic components is growing exponentially. 1,2 Spintronic systems 3,4 operated using electric currents through spin-transfer torque 5,6 or spin−orbit torque 7,8 effects are becoming key elements for next-generation nanoelectronics with enhanced memory and information processing capabilities. However, magnetization switching using electric current involves an undesirable Joule heating effect, which is detrimental to energy efficiency. 9,10 An interesting alternative is to modulate magnetic properties by applying electric fields instead of electric current, thus minimizing power dissipation. This has rapidly evolved into a whole area of research referred to as voltage control of magnetism (VCM).
Magneto-ionics refers to a particular mechanism for VCM in which voltage-driven ion transport (of, e.g., O 2− , 11−15 Li + , 16 F − , 17 H + , 18−21 or N 3−22−26 species) leads to a large and controllable modulation of magnetism without the need of strain transfer. This is different from voltage-controlled strainmediated multiferroic heterostructures, which are less con-venient for spintronics because repeated voltage actuation can lead to mechanical fatigue and eventual device failure. Archetypical magneto-ionic structures comprise a gate electrolyte (either solid or liquid) 27,28 in contact with a ferromagnetic (FM) target material. Voltage is applied across the electrolyte, using the FM film as a working electrode and causing a voltage polarity-dependent insertion/removal of ions into/from the target material. In this way, magnetic properties, such as saturation magnetization, 11,22 magnetic anisotropy and coercivity, 12,18 exchange bias field, 13,29 or skyrmion generation/ suppression, 20 among others, can be reversibly controlled, making magneto-ionic materials highly promising for ultra-lowpower magnetic devices. An extreme case is when voltage induces a complete reversible transition between FM and nonmagnetic states, leading to voltage-driven ON−OFF switching of ferromagnetism. This has been reported in a number of magneto-ionic systems, such as Co 3 O 4 , 11,14,30 Li +intercalated α-Fe 2 O 3 , 31 F − -intercalated La 2−2x Sr 1+2x Mn 2 O 7 , 17 CoN, 26,28 CoMnN, 25 or α-Co(OH) 2 . 32 ON−OFF switching of the ferromagnetic state has been also induced by electrostatic surface charging in FeS 2 . 33 In most magneto-ionic systems, the source of the moving ions is electrolytes (e.g., GdO x , HfO x , propylene carbonate with dissolved LiPF 6 or KI) that are in direct contact with pristine FM or ferrimagnetic layers (e.g., Co, Fe, Fe 2 O 3 ), whose properties are manipulated with voltage. 27,34−36 Magneto-ionic systems based on O 2− insertion from an external electrolyte to a target FM material often suffer from slow dynamics at room temperature and irreversible compositional/structural changes in the FM phase, eventually leading to degradation and limited cyclability. 12 In addition, for relatively thick films, it is challenging to achieve a fully OFF magnetic state with voltage due to the limited penetration of the external ions toward the interior of the FM target layers. 37 Recently, smaller ions (e.g., H + , F − , or Li + ) have been introduced to achieve faster and more cyclable voltage-induced manipulation of magnetic properties. Fast, reversible, and cyclable tuning of perpendicular magnetic anisotropy, 18 Dzyaloshinskii−Moriya interaction, 38,39 or ferrimagnetic spin textures 40 has been achieved through chemisorbed O 2− or H + ion species. However, systems relying on H + are either sensitive to environmental conditions (e.g., humidity) or restricted by their incompatibility with traditional complementary metal-oxide semiconductor (CMOS)-based devices (i.e., standard fabrication processes for semiconductor devices, such as metal oxide field-effect transistors). 28 An alternative approach is to use structural oxygen or nitrogen, self-contained in the target materials, as the source of the moving ions. Examples of such materials are Co 3 O 4 , 11 CoN, 22,26 or CoMnN 25 films. These target materials, which are CMOS compatible, exhibit an initially fully OFF (i.e., paramagnetic) state and provide "ready-prepared" lattice sites for ion diffusion, allowing for net magneto-ionic generation of ferromagnetism by voltage-triggered O 2− or N 3− ion motion from the films toward a neighboring electrolyte. Unfortunately, achieving a high magneto-ionic cyclability in these materials (i.e., removing and reinserting the O 2− /N 3− ions many times by switching voltage polarity) remains a challenge.
To induce magneto-ionics, either solid or liquid electrolytes can be utilized. Solid electrolytes (with ultrathin dielectric layers) are preferred for solid-state spintronics. Ultrathin solid dielectric layers are needed to induce sufficiently large electric fields under moderate applied voltages. However, at such small thicknesses, difficulties arise to withstand high electric fields without electric pinholes. Moreover, thin solid electrolytes offer a limited ion buffering capability (i.e., they easily become saturated with ions and cannot sustain stable ion transport during long-term operation, especially when the magneto-ionic layer is thicker than the solid electrolyte layer). Liquid electrolytes are convenient for other magnetoelectric applications, such as magnetophoresis/microfluidics 41 or to emulate neuromorphic functionalities (the brain operates in a liquid environment). 15,26 Owing to the formation of the "electric double layer", whose thickness is <1 nm, liquid electrolytes are able to generate ultralarge electric fields (hundreds of MV cm −1 ) at the interface between the liquid and the target magneto-ionic layer, without electric pinholes. 27 Liquid electrolytes may also be good ion reservoirs. 11,14,22 However, when voltage is applied, the mobile ions released from the target layer into the liquid can travel long distances toward the counter electrode, making their reintroduction into the magneto-ionic layer (with voltage of opposite polarity) difficult. In addition, at the counter electrode, if sufficiently high voltage is applied, the dissolved ions (e.g., O 2− , H + , or N 3− ) can transform into the corresponding oxygen, hydrogen, or nitrogen gases and be released to the atmosphere in the form of bubbles. These effects are difficult to control and are highly detrimental for magneto-ionic reversibility and endurance. Restricting ion transport within the magneto-ionic layers could avoid these problems and is expected to improve the magneto-ionic cyclability.
Here, we propose an improved nanoscale-engineered magneto-ionic structure that results from inserting an amorphous Ta layer (which gets spontaneously passivated in air) between the Co 3 O 4 film (magneto-ionic target material) and propylene carbonate, PC (liquid electrolyte). Upon application of negative voltage, O 2− ions migrate from Co 3 O 4 to Ta, promoting the formation of TaO x (Ta is a good oxygen getter), which acts as a thin solid electrolyte with good ionic conductivity. This architecture preserves sufficiently high electric fields to trigger ion motion while allowing to repeatedly induce ON−OFF switching of ferromagnetism at room temperature. By optimizing the thickness of the Ta layer, a drastic increase of cyclability is achieved in Co 3 O 4 /Ta (25 nm)/PC (>800 cycles) compared to Co 3 O 4 /PC with no Ta insertion (<30 cycles). The Ta layer hinders oxygen ions from entering the liquid electrolyte and allows for O 2− redistribution inside the Co 3 O 4 and Ta layers, as assessed by high-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) and electron energy loss spectroscopy (EELS). Positron annihilation spectroscopy is used to precisely characterize the defect structure in Ta, which strongly contributes to the O 2− ion transport. An increase of the defect size during biasing is observed. The reduced electric conductivity of the passivated amorphous Ta layer is important to allow the penetration of the electric field inside the Co 3 O 4 layer and the concomitant O 2− ion transport. Since O 2− diffusion becomes mainly restricted within the Co 3 O 4 /TaO x structure (instead of O 2− being released to PC), this greatly improves the efficiency of ion transport in the material, resulting in largely enhanced magneto-ionic cyclability.

RESULTS AND DISCUSSION
The basic building block of the investigated magneto-ionic system is a 40 nm thick Co 3 O 4 film grown by DC reactive sputtering at room temperature onto Cu (60 nm)/Ti (20 nm)/[100]Si (725 μm) substrates. The Co 3 O 4 films were coated with sputtered amorphous Ta protective layers of variable thickness (from 5 to 50 nm), which were left to passivate in air (see Experimental Section). Uncoated Co 3 O 4 films were also grown as a reference. Figure 1(a) shows a lowmagnification HAADF-STEM image of the as-prepared sample with a 25-nm-thick Ta protective capping layer. Clear interfaces between the various layers can be seen, and their thicknesses are in good agreement with the nominal ones. Note that the topmost Pt capping layer was grown only during the TEM lamella preparation (i.e., it was not present during magneto-ionic experiments). To investigate the Co, O, and Ta distribution, Co, O, and Ta EELS mappings from the area marked with an orange rectangle in Figure 1(a) were acquired. As shown in Figure 1(b), Co and O elements are homogeneously distributed within the Co 3 O 4 films. Additionally, some O signal is also detected inside the Ta layer, particularly on its upper part. This is ascribed to surface selfpassivation of Ta in air, which is known to result in the formation of a stable TaO x top layer. 42,43 Figure 1(c) shows the θ/2θ X-ray diffraction (XRD) pattern of an as-grown film. The XRD peak observed at 2θ ≈ 19.1°is consistent with (111) Co 3 O 4 (PDF 00-009-0418), whereas that at 2θ ≈ 36.7°could be ascribed to either (111) Co 3 O 4 or (111) CoO (PDF 00-001-1227). Peaks corresponding to the Cu buffer layer are also detected. Conversely, no traces of Ta are evidenced, suggesting that Ta is highly nanostructured or even amorphous-like. Growth of amorphous Ta by physical deposition methods has been previously reported, 44,45 and it is promoted by the existence of ions and particles with high kinetic energy when sputtering at sufficiently high gun powers. Amorphous Ta exhibits higher electric resistivity than its crystalline counterparts. 44 This is beneficial to keep sufficient electric field strength inside the actuated films while voltage is applied, thereby enabling magneto-ionics.
To further investigate the structure of the samples, highresolution transmission electron microscopy (HRTEM) images of the cross sections of as-prepared Co 3 O 4 and Ta layers were taken, as shown in Figure 1(d) and (e), respectively. The areas marked with red squares were chosen for fast Fourier transform (FFT) analyses, as shown in the insets. In Figure 1(d), the "1" spot corresponds to an interplanar distance of 0.248 nm and is consistent with (311) Co 3 O 4 (PDF 00-009-0418) or (111) CoO (PDF 00-001-1227) interplanar distances (0.244 and 0.245 nm, respectively), while spot "2" corresponds to an interplanar distance of 0.225 nm and is unambiguously ascribed to the (200) CoO (PDF 00-001-1227) interplanar distance (0.212 nm). The "3" spot corresponds to an interplanar distance of 0.448 nm, which is consistent with (311) Co 3 O 4 (0.467 nm, PDF 00-009-0418). This indicates that a mixture of CoO and Co 3 O 4 phases is plausible in the Co oxide film, in agreement with previously reported results. 15 For simplicity, in spite of the presence of CoO in particular close to the interface, the Co oxide film is labeled Co 3 O 4 throughout the article. Remarkably, no spots are observed for Ta, in agreement with its amorphous nature, as also evidenced by XRD (Figure 1(c)).

ACS Nano www.acsnano.org
Article a sub-nm-thick electric double layer forms at the electrolyte side of the electrolyte/Ta interface, allowing for the generation of a high electric field. 47 This electric field is ultimately responsible for driving oxygen ions from the Co 3 O 4 to the Ta/ TaO x layers (as illustrated in Figure 2(b)). Voltage treatments were performed in situ, while hysteresis loops were recorded at room temperature by vibrating sample magnetometry (VSM), with an in-plane applied magnetic field. The total measured magnetic moment of the samples (in emu) was normalized to the area of the sample and the nominal thickness of Co 3 O 4 (to obtain emu cm −3 ). As seen in Figure 3, irrespective of the thickness of the Ta layer (i.e., 0, 5, 10, 25, or 50 nm), all films in the as-grown state show very little ferromagnetic response (<33 emu cm −3 , which is equivalent to ≈2.3% the magnetization of pure FCC-Co 48 ). This small ferromagnetic signal is common in sputtered Co 3 O 4 15 and is likely due to either a small fraction of residual Co clusters that do not become fully oxidized during the sputtering process or substrate contamination. The deposition of Ta onto Co 3 O 4 has a negligible effect on the initial ferromagnetic signal of the Co 3 O 4 films. Upon negative biasing at −25 V for 30 min or 1 h, clear hysteresis loops are observed. The generated saturation magnetization (M S ) is maximum for the sample without Ta (around 400 emu cm −3 , suggesting that 27% of the volume of Co 3 O 4 gets reduced to Co), and it slightly decreases when a thin Ta interlayer (5−10 nm) is inserted between the Co 3 O 4 film and the liquid electrolyte. M S further decreases for sufficiently thick Ta. Remarkably, magneto-ionic effects when Pt (instead of Ta) is grown onto Co 3 O 4 are strongly reduced: a small hysteresis loop is observed only for 5 nm Pt (compare Figure 3(f) with (b)), and no ferromagnetic response is obtained for thicker Pt. In all cases, the initial virtually nonmagnetic state can be recovered by applying +25 V for 30 min. Figure 3 shows that, for all samples, most of the ferromagnetic response is induced in less than 30 min, since M S does not further increase when applying voltage for 1 h. The time evolution of M S for all samples under −25 V is shown in Figure 2(c). An external magnetic field of 10 kOe (i.e., above the anisotropy field of the generated ferromagnetic counterpart) was applied during these measurements to ensure magnetic saturation. In all films, an immediate increase of M S is observed in response to the applied ΔV, evidencing a quick onset of the oxygen ionic motion, which leaves metallic ferromagnetic Co behind. 14,15 The obtained steady-state value of M S decreases with the Ta thickness, from 401 emu cm −3 to 19 emu cm −3 (in agreement with the magnetic hysteresis loops shown in Figure 3(a)−(e)). One important parameter of large significance for device applications is the minimum threshold voltage required to trigger magneto-ionic effects. 22,25 The onset voltage for Co 3 O 4 films without and with a 25-nm-thick Ta capping layer was evaluated by subjecting the films to increasing negative voltage steps of −2 V, until M S started to increase, as shown in Figure 2(d). The results reveal that the onset voltage is approximately −6 V for both samples, whereas a voltage of +8 V leads to complete recovery in both cases, which agrees with previous works on similar systems. 11 Figure 2(c) and (d) also show that the rate at which M S increases under voltage application is larger for a smaller Ta layer thickness. This, together with the reduction of the steadystate M S for thicker Ta, indicates that the effective electric field acting on the Co 3 O 4 layer becomes lower for larger Ta thickness. This suggests that, for thicker Ta, there is a more severe dissipation of space charge density, while O 2− is migrating from Co 3 O 4 to form TaO x . 49

ACS Nano www.acsnano.org Article
In a first approximation, this effect can be modeled considering the system as a simple voltage divider. The voltage drop at the Co 3 O 4 /Ta interface can be estimated assuming that, during the magneto-ionic process, the Co 3 O 4 and the newly formed TaO x layers act as two resistances connected in series. In this case, the voltage at the Co 3 . Considering that R = ρδ/A (where ρ is the resistivity of each material, δ is the film thickness, and A is the lateral area of the films), it is straightforward to estimate ΔV int as a function of the layers' thicknesses and the electric conductivities of Co 3 O 4 and TaO x . The electrical conductivity of Co 3 O 4 , measured by the Van der Pauw method, is approximately σ Co3O4 = 20 S m −1 . 14 The conductivity of TaO x strongly depends on the oxygen content, and it varies by several orders of magnitude, from 10 5 S m −1 in amorphous metallic Ta 44 to 10 −6 S m −1 for highly oxidized Ta. 50 The presence of oxygen in the naturally passivated Ta ( Figure  1(b)) can easily bring ρ to values in the range 10 −1 −10 2 S m −1 before any voltage is applied. 51 Figure 2(e) shows the evolution of the calculated interface voltage as a function of σ TaOx for the different Ta film thicknesses. For highly conductive Ta (or TaO x ) spacer layers, ΔV int becomes independent of the Ta thickness, which means that any eventual magneto-ionic effect would be independent of the Ta thickness (and ΔV int would be always equal to −25 V). However, in this case, the electric field would be highly screened at the interface and would not penetrate inside Ta (since, in metals, the electric field is confined within the Thomas−Fermi screening length, which is typically <0.5 nm). 27 Thus, for highly metallic interlayers (e.g., noble metals like Pt, as in Figure 3(f)), little magneto-ionic effects are expected since there is no driving force (no electric field) to induce ion motion inside the metal. 24 This is opposite to what happens in semiconductors (e.g., TaO x ), where the electric field will penetrate deeper into the layer. Interestingly, for low electrically conductive TaO x (i.e., σ TaOx < 100 S m −1 ), ΔV int becomes clearly lower as the Ta layer thickness increases (Figure 2(e)). For δ Ta = 50 nm, if σ TaOx is lower than 5 S m −1 , ΔV int eventually drops below the threshold voltage (region highlighted in cyan), meaning that no magneto-ionic effects will be induced in this case (while, for thinner Ta layers, for the same σ TaOx value, ΔV i will still be above the threshold). Note that during magneto-ionic motion of O 2− from Co 3 O 4 to TaO x , the values of σ TaOx will rapidly decrease, thus reducing the interface voltage and slowing down the magneto-ionic process. Another interesting effect is that when Co 3 O 4 transforms to metallic Co, the conductivity of the magnetoionic layer (σ Co3O4 ) increases. This, in turn, has an effect on ΔV int , which is shown in Figure 2(f). In bulk metallic Co, σ Co can reach values >10 7 S m −1 . 52 As plotted in Figure 2(f), when σ Co3O4 transforms to Co and σ Co3O4 increases, the interface voltage drastically drops, and ΔV imt easily falls below the threshold voltage. Then any magneto-ionic response will tend to stop. As an example, for the particular case of δ Ta = 50 nm, any eventual O 2− ion motion triggered while σ TaOx is sufficiently large will tend to stop as soon as σ Co3O4 increases above 10 3 S m −1 (see Figure 2(f)) or σ TaOx decreases below 5 S m −1 (Figure 2(e)). It is noteworthy that, although this intuitive picture is a simplified representation of the reality, it already provides some basic understanding of the role of the TaO x layer thickness and resistivity on the induced magneto-ionic effects.
Notwithstanding the decrease of the steady-state M S with the deposition of a Ta capping layer (by approximately 25% for δ Ta = 25 nm), the formation of TaO x (which acts as a solid electrolyte) has one very beneficial effect: it drastically enhances magneto-ionic cyclability. This is evidenced in Figure  4, which shows the cyclability results (i.e., repeated increase/ decrease of M) in a Co 3 O 4 thin film and Co 3 O 4 /Ta (25 nm) bilayer films upon application of −25 V/+25 V voltage pulses with a duration of 80 s. As shown in Figure 4(a) and (b), a very stable and reversible behavior is observed during the first 10 cycles for both samples. However, the Co 3 O 4 /Ta (25 nm) bilayer sample maintains a very stable cyclability even after 800 cycles (Figure 4(a)), whereas no sign of magneto-ionic effect is detected for Co 3 O 4 films after 30 cycles (Figure 4(b)). This demonstrates a significant enhancement of the endurance of the system by the deposition of a Ta overlayer with appropriate thickness.
The improved cyclability in the Co 3 O 4 /Ta (25 nm) bilayer is ascribed to the role of Ta in allowing diffusion of oxygen ions. In this sample, when negative voltage is applied, the O 2− ions exiting Co 3 O 4 are captured by Ta, forming a TaO x solid electrolyte, instead of being directly released to the liquid electrolyte. Ta is a well-known oxygen getter, 53 and TaO x is a good ionic conductor. 54 In unprotected Co 3 O 4 films, O 2− ions are directly dissolved in the propylene carbonate, and once solvated with the PC chains, they can travel long distances toward the positively charged counter electrode (i.e., the Pt wire), where they eventually form O 2 gas bubbles and are released to the atmosphere at sufficiently high voltages. This long-distance transport of ions in the liquid is the main reason for the poor cyclability of uncapped Co 3 O 4 films. However, when a Ta interlayer is grown adjacent to Co 3 O 4 , O 2− ions move back and forth along relatively short distances (for alternating applied voltages of opposite polarities), and the process is confined between Co 3 O 4 and the TaO x solid electrolyte, thus drastically improving cyclability.
To further understand the effect of the top Ta layer on the induced magnetic properties from the perspective of the ion transport mechanism, cross-sectional lamellae of a bilayer sample (40-nm-thick Co 3 O 4 plus 25-nm-thick Ta overlayer) electrolyte-gated at −25 V for 1 h were studied by HAADF-STEM and EELS. As shown in Figure 5(a)−(c), an inhomogeneous microstructure inside the Co 3 O 4 layer is generated upon voltage application, with Co-rich and O-rich regions, leading to void-like morphologies, in agreement with previous studies on this kind of magneto-ionic material. 11 In turn, an enrichment in O is observed in the upper part of the Ta layer. This is clearly evidenced by quantitative EELS analysis ( Figure 5(d)) and high-resolution TEM (Figure 5(e)).   tends to become more amorphous. This is corroborated by θ/ 2θ X-ray diffraction (see Figure 6(a)), where the peak from Co 3 O 4 (111) planes disappears upon voltage treatment. In addition, the FFT spots obtained from high-resolution TEM prove the existence of metallic Co in the treated sample, as can be seen in Figure 6(b). The formation of metallic Co is responsible for the M S increase after voltage treatment ( Figure  2(c) and Figure 3). To assess the electrochemical behavior of the system, cyclic voltammetry (CV) curves were recorded for all the surfaces involved in the main magneto-ionic setup (see Figure S1 in the Supporting Information). When the Cu grown on Ti is exposed to the electrolyte, a cathodic wide peak is observed, centered at −0.9 V vs Pt reference electrode, while further anodic runs evidence a flat peak. This is consistent, in principle, with the reduction of Cu 2 O (native oxide layer formed on Cu when exposed to air) to Cu (first wave) and consequent oxidation of Cu to Cu 2 O. The CV curve of Co 3 O 4 grown onto the Ti/Cu does not show the Cu waves, confirming that the cobalt oxide layer does cover completely the Cu phase and prevents its oxidation/reduction. Once Ta is grown on Co 3 O 4 , a tiny wide oxidation wave is observed, centered at 0.6 V vs Pt, in agreement with the expected oxidation of Ta. Since both coatings (Co 3 O 4 and TaO x ) are less conductive than Cu, the redox processes are hindered compared to the case of Cu.
To shed further light on the microstructure of the films upon magneto-ionic actuation, Doppler broadening variable energy positron annihilation spectroscopy (DB-VEPAS) 55 and variable energy positron annihilation lifetime spectroscopy (VEPALS) 56 experiments were performed (see Experimental Section). PAS is a sensitive probe to open volume defects on atomic scales and across depth due to positrons' preferential localization and subsequent annihilation with electrons in crystal empty spaces, i.e., vacancies and their agglomerates. The energy distribution and time necessary for annihilation depends on the local electron density. 57 As shown in Figure  7(a), after negative biasing at −25 V, the low electron momentum fraction, S (directly proportional to the size and concentration of defects), 58 increases in the Ta region but slightly drops in the Co 3 O 4 layer, suggesting a substantial raise of defect density in Ta and a slight drop of defect concentration in Co 3 O 4 . The increase of defect density in Ta is probably a consequence of the O 2− interdiffusion and the formation of TaO x . Additionally, the average defect size (which is proportional to the average positron lifetime, τ av ), strongly increases in both TaO x and Co 3 O 4 layers after voltage actuation (Figure 7(b)). The expected average size is in the

ACS Nano
www.acsnano.org Article range of large vacancy agglomerates (8−10 vacancies). 59 Figure 7(b) and (c) show the existence of three discrete lifetime components (τ 1, τ 2 , and τ 3 ), 59 which correspond to three different average defect sizes. 11 They were obtained by deconvolution of PALS spectra using a nonlinear least-squared fitting method (package PALSfit software). 60 The corresponding relative intensities (I 1 , I 2 , and I 3 ) relate to the concentration of each defect type (size). The larger τ i is, the larger the defect size is since it takes longer for positrons to be annihilated with electrons. 57 In the as-grown sample, only τ 1 and τ 2 lifetime components are detected, meaning an absence of large voidlike structures (no τ 3 ) in the prebiasing state. The most abundant defect size (τ 1 ) contributes as I 1 = 62−77% of the overall signal and is in the range of bivacancy clusters in the upper subsurface Ta region (i.e., passivation layer) and single vacancy in the underneath Ta sublayer. 59

CONCLUSIONS
In summary, this work demonstrates the beneficial effect of adding a thin solid ionic conductor (in this case, a naturally passivated amorphous Ta interlayer) on the magneto-ionic response of Co 3 O 4 films under the action of an electric field when immersed in liquid electrolytes. A priori, adding a capping layer between the Co 3 O 4 film and the liquid electrolyte could be thought of as simply hindering magnetoionics and reducing (or eventually suppressing) any oxygen ion migration triggered by the externally applied electrical voltage. This is indeed observed for thin metallic capping layers (e.g., 5 nm Pt) or thick highly resistive Ta layers (with a thickness larger than 50 nm). However, for thinner Ta interlayers, in spite of a moderate reduction of the steady-state M S (by a factor of 25% for 25-nm-thick Ta), a drastic enhancement of magneto-ionic cyclability is observed, from less than 30 cycles in uncoated Co 3 O 4 to more than 800 cycles for a Ta thickness of 25 nm. Such enhancement of endurance is ascribed to the key role of the generated TaO x layer in preventing O 2− from being released to the liquid electrolyte, i.e., limiting oxygen ion transport within the Co 3 O 4 /TaO x layers. This is confirmed by compositional/structural characterization using HAADF-STEM and EELS as well as positron spectroscopy experiments. Beyond magneto-ionics, the reported strategy to enhance cyclability can be easily extrapolated to other systems relying on ion transport mechanisms, such as iontronics, sensors, or neuromorphic computing.

EXPERIMENTAL SECTION
Sample Fabrication. Co 3 O 4 thin films of 40 nm thickness were grown at room temperature by reactive sputtering in a high-vacuum chamber (with a base pressure of <8 × 10 −8 Torr) on nondoped ⟨100⟩-oriented Si wafers previously coated with a 20-nm-thick titanium adhesion layer and 60-nm-thick copper seed layer. Prior to growing Co 3 O 4 , the Cu seed layers were partly masked to leave enough space for the electric contact (i.e., to later serve as a working electrode). The Co 3 O 4 films were subsequently coated with Ta protective layers of variable thickness, ranging from 5 to 50 nm, which were left unprotected to self-passivate in air. Two reference films were also prepared to serve as references: uncoated Co 3 O 4 , and Co 3 O 4 coated with Pt with the same range of thicknesses as Ta. The growth of Co 3 O 4 was carried out in a mixed Ar and O 2 atmosphere using an oxygen partial pressure of 50% and a total working pressure of 3 × 10 −3 Torr. The distance between the substrate and targets was around 10 cm, and the growth rate was approximately 0.6 Å s −1 . Ti, Cu, Ta, and Pt layers were grown under 3 × 10 −3 Torr Ar. The gun power to grow Ta was 100 W.
Magneto-Ionic Characterization and Cyclic Voltammetry Curves. Room-temperature magnetoelectric measurements were performed using a commercial vibrating sample magnetometer from Micro Sense (LOT, Quantum Design), with a maximum applied inplane magnetic field of 2 T. The samples were electrolyte-gated using an external Agilent B2902A power supply, applying voltage between the counter electrode (a Pt wire) and the working electrode (e.g., the investigated Si/Ti/Cu/Co 3 O 4 /Ta thin films) in a homemade electrolytic cell (see Figure 2(b)). The electrolyte consisted of anhydrous propylene carbonate with Na + and OH − solvated species (10−25 ppm), formed by immersing small pieces of metallic sodium that were able to react with any possible traces of water. 11 Negative voltages in this work indicate the accumulation of negative charges at the working electrode (and vice versa for positive voltages). The magnetization (M) was obtained by normalizing the magnetic moment to the sample volume exposed to the electrolyte. Note that the linear slopes in the hysteresis loops at high fields (arising from diamagnetic or paramagnetic contributions) were subtracted by correcting the background signal (i.e., at fields always significantly larger than the saturation fields).
Cyclic voltammetry curves for all the surfaces involved in the main magneto-ionic setup were also recorded. For these experiments, a scan speed of 10 mV/s and a potential sweep range from 0 V to −2 V to +1.4 V and to 0 V vs Pt were selected. The exposed surface was 0.5 cm 2 , and the distance between the reference and working electrode was 4 mm, thus mimicking the conditions utilized during the magneto-ionic experiments. Here, however, we used three electrodes (including a reference electrode), whereas magneto-ionic experiments were performed in a two-electrode configuration. Thus, all voltages in the CV curves are given with respect to a reference electrode, in this case Pt (99.99% Goodfellow), known to act well in organic solvents. The global cell (two-electrode) potential has been measured during these CV experiments to be on the order of 4 to 6 V depending on the system. Further potential could not be applied during the CV curves due to the limitations of the utilized potentiostat (VSP Biologic). However, the curves obtained in this potential range are already representative of the electrochemical behavior of the system.
Structural and Compositional Measurements. θ/2θ XRD patterns were collected on a Materials Research diffractometer from Malvern PANalytical Company, equipped with a PIXcel 1D detector, using Cu Kα radiation. HRTEM, HAADF-STEM, and EELS were carried out on a TECNAI F20 HRTEM/STEM microscope operated at 200 kV. Cross-sectional lamellae were prepared by focused ion beam, placed onto a copper transmission electron microscopy grid, and topped with a protective platinum layer.

Doppler Broadening Variable Energy Positron Annihilation Spectroscopy and Variable Energy Positron Annihilation
Lifetime Spectroscopy. DB-VEPAS measurements were conducted at the setup for in situ defect analysis (AIDA) 55 of the slow positron beamline (SPONSOR). 62 Positrons were accelerated and monoenergetically implanted into samples in the range of E p = 0.05−35 keV, which allows for depth-sensitive analysis. The mean positron implantation depth was approximated using a simple material density (ρ)-dependent formula: ⟨z⟩ = 36/ρ·E p 1. 62 . 63 Since at the annihilation site thermalized positrons have very small momentum compared to the electrons, a broadening of the 511 keV line is observed mostly due to momentum of the electrons, which is measured with a high-purity Ge detector (overall energy resolution of 1.09 ± 0.01 at 511 keV). This broadening is characterized by a parameter S defined as a fraction of the annihilation distribution in the middle (511 ± 0.93 keV). The S-parameter is a fraction of positrons annihilating with lowmomentum valence electrons and represents vacancy-type defects and their concentration. 58 VEPALS measurements were conducted at the monoenergetic positron source (MePS) beamline, which is an end station of the radiation source ELBE (Electron Linac for beams with high Brilliance and low Emittance) at Helmholtz-Zentrum Dresden-Rossendorf (Germany). 56 A digital lifetime CeBr 3 scintillator detector was used, with a homemade software employing an SPDevices ADQ14DC-2X with 14-bit vertical resolution and 2 GS s −1 (gigasamples per second) horizontal resolution and with a time resolution function down to about 0.230 ns. 64 The resolution function required for spectrum analysis uses two Gaussian functions with distinct intensities depending on the positron implantation energy, E p , and appropriate relative shifts. All spectra contained at least 1 × 10 7 counts. The spectra were deconvoluted using the PALSfit fitting software into discrete lifetime components, which directly confirm different defect types (i.e., sizes). 60 Typical lifetime spectrum N(t) is described by N(t) = ∑(1/τ i )I i exp(−t/τ i ), where τ i and I i are the positron lifetime and intensity of the ith component, respectively (∑I i = 1).
The corresponding relative intensities (I i ) reflect to a large extent the concentration of each defect type, and positron lifetimes (τ i ) are directly proportional to defect size (i.e., the larger the open volume, the lower the probability and the longer it takes for positrons to be annihilated with electrons). The positron lifetime and its intensity were probed as a function of positron implantation energy E p or, in other words, implantation depth (thickness). The average positron lifetime τ av is defined as τ av = ∑ i τ i ·I i , which is the weighted average of the defect size. The shortest lifetime component (τ 1 < 0.32 ns) represents positron annihilation inside vacancy clusters (likely within grains) and/or at the grain boundaries, depending on the film's microstructure. The intermediate lifetime (0.35 < τ 2 < 0.90 ns) accounts for annihilation at larger vacancy clusters (linked to grain boundaries and their intersections), surface states, and small voids/ pores (0.28−0.37 nm in diameter, calculated based on the shape-free model for pore-size estimation of Wada et al.; 59
Cyclic voltammetry curves of the basic investigated layer structures (PDF)

AUTHOR INFORMATION
Corresponding Authors the Generalitat de Catalunya (2021-SGR-00651) is acknowledged. J.S. thanks the Spanish "Fábrica Nacional de Moneda y Timbre" (FNMT) for fruitful discussions. E.M. is a Serra Huńter Fellow. Parts of this research were carried out at ELBE at the Helmholtz-Zentrum Dresden -Rossendorf e. V., a member of the Helmholtz Association. We would like to thank the facility staff for assistance. This work was partially supported by the Impulse-und Net-working fund of the Helmholtz Association (FKZ VH-VI-442 Memriox) and the Helmholtz Energy Materials Characterization Platform (03ET7015).