Voltage-Controlled ON-OFF Ferromagnetism at Room Temperature in a Single Metal Oxide Film

: Electric-field controlled magnetism can boost energy-efficiency in widespread applications. However, technologically, this effect is facing important challenges: mechanical failure in strain-mediated piezoelectric/magnetostrictive devices, dearth of room-temperature multiferroics or stringent thickness limitations in electrically-charged metallic films. Voltage-driven ionic motion (magneto-ionics) circumvents most of these drawbacks while exhibiting interesting magnetoelectric phenomena. Nevertheless, magneto-ionics typically requires heat-treatments and multi-component heterostructures. Here we report on the electrolyte-gated and defect-mediated O and Co transport in a Co 3 O 4 single layer which allows for room-temperature voltage-controlled ON-OFF ferromagnetism (magnetic switch) via internal reduction/oxidation processes. Negative voltages partially reduce Co 3 O 4 to Co (ferromagnetism: ON), resulting in graded films including Co- and O-rich areas. Positive bias oxidizes Co back to Co 3 O 4 (paramagnetism: OFF). This electric-field-induced atomic-scale reconfiguration process is compositionally, structurally and magnetically reversible and self-sustained since no oxygen source other than the Co 3 O 4 itself is required. This process could lead to electric-field-controlled device concepts for spintronics.

. Voltage actuation and magnetic characterization by VSM. (a) Custom-made electrolytic cell used for the in-situ voltage-applied VSM measurements. (b) normalized magnetic moment (m) versus applied field VSM measurements of a pristine sample, after being treated at -10 V for 60 min and after recovery having applied +20 V for 60 min. (c) VSM measurements corresponding to a sample subjected to -10 V for 0, 15 min, 30 min and 60 min. (d) VSM measurements corresponding to a sample subjected to decreasing negative voltages of 0 V, -10 V, -25 V and -50 V, applied during 60 min each. The measurements shown in (d) were recorded on the same sample. That is, upon negative biasing, the sample was brought to its pristine state with a recovery process with positive voltages. Specifically, the sample was recovered for 120 min at +20, +50 and +100 V upon being treated for 60 min at -10 V, -25 V and -50 V, respectively.
As voltage is increased, ions in the electrolyte solution diffuse towards the surface of the electrode forming the EDL which leads to strong electric fields on the order of hundreds of MV/cm. 13 As can be seen in Figure 1b, the pristine Co3O4 sample shows no ferromagnetic behavior in agreement with its paramagnetic nature at room temperature. Conversely, after subjecting the sample to -10 V for 60 min, the measurements show a clear hysteresis loop, evidencing ferromagnetism. This generated ferromagnetism can be fully suppressed (and, thus, the process reversed) by applying a positive voltage of 20 V for 60 min. It might be anticipated that this reversible phenomenon could be explained in a magneto-ionic framework through voltage-driven O 2transport, giving rise to reduction and oxidation (i.e., redox) processes which are polarity dependent. For a given voltage (i.e., -10 V), Figure 1c shows the VSM measurements as a function of time. Further, Figure 1d shows the voltage-dependence of the hysteresis loops for a fixed biasing time (i.e., 60 min). The magnetic moment scales with time and, in a more pronounced manner, with voltage, indicating that the underlying atomic mechanisms probably rely on voltage-activated diffusion effects. The measured magnetic signal after applying -50 V is equivalent to a metallic 6 nm-thick Co film. A single sample was used to investigate the voltage-dependence of the magnetic properties when subjecting the system to negative biases (Figure 1d), aimed at shedding light on the endurance characteristics of the system. The sample was first subjected to -10 V for 60 min and, then, the voltage was switched off and, immediately after, a VSM measurement was recorded in-situ inside the electrolyte. To recover the pristine state, the sample was subjected to +20 V for 120 min. When treating the sample for 60 min at -25 and -50 V, the pristine state was recovered upon subjecting the sample for 120 min at +50 and +100 V, respectively. Further cycling was not accomplished due to the degradation of the electrolyte, which limited the endurance to three cycles. Electrolyte replacement without exposing the sample to air to assess the cyclability properties still remains challenging due to the limited dimensions of the home-made electrolytic cell (note that Eppendorf® tubes of 1.5 ml of volume are used as recipients). Further improvements of the cell, such as shape and volume capacity, will allow for a representative characterization of the endurance properties of the system. Treatments up to -200 V were also carried out to maximize the effects and to subsequently perform a detailed structural characterization of the samples. As shown in Figure S2, upon removal of the -200 V (applied for 30 min), the magnetic moment relaxes following an exponential decay but a clear ferromagnetic signal remains (even for several months). Note that the hysteresis loop recorded immediately after such a high negative applied voltage is open ( Figure S2a) since the magnetic moment progressively decreases during the time needed to acquire the loop. However, once the film is relaxed, the loops close, as expected. Note that this relaxation effect is considerably weaker for lower applied electric fields. For example, the loops obtained after applying lower voltages for 60 min close completely at high applied magnetic fields (Figure 1d).
To shed light on the atomic mechanisms which take place under voltage actuation, compositional characterization of the top sample surface through X-ray photoelectron spectroscopy (XPS) was carried out ( Figure 2). As shown in Figure 2a, the XPS spectrum of the pristine sample is consistent with a rather pure Co3O4 phase with traces of CoO, while, upon treatment at -10 V for 60 min, metallic Co or a Co-rich phase is present at least at the sample surface. Hence, negative biases promote reduction from Co3O4 to CoO and from CoO to ferromagnetic Co (Figure 2b), in agreement with the magnetometry results. It should be mentioned that CoO may partially come from natural oxidation of Co since the XPS is performed ex-situ and, thus, the sample is exposed to air. In contrast to negative biases, positive voltages allow for the recovery of the Co3O4 phase, evidencing that Co tends to re-oxidize ( Figure 2c). This demonstrates the high reversibility of the process, which not only takes place magnetically but also compositionally, and further indicates that voltage-driven redox processes could be a plausible scenario. Figure 2d  From a compositional viewpoint, it is clear that this ON-OFF ferromagnetism can be electricallymodulated by voltage-driven ion migration. However, further information on morphological and structural aspects is essential to determine the mechanisms which govern this ionic transport.
Actually, Co3O4 is prone to exhibit vacancies 33 and, in case of bulk Co3O4, Co migration is vacancy-mediated (i.e., via Co 3+ vacancies). 34,35 For a more detailed structural characterization, three samples were investigated: an as-deposited film, a film treated at -200 V for 30 min (where the effects of voltage are maximized) and a sample treated at -50 V for 30 min and subsequently fully recovered (i.e., brought back to the paramagnetic state) by applying +100 V for 60 min. Full recovery from the sample treated at -200 V would require positive voltages higher than the ones available in our setup, hence the recovery was investigated in a sample treated at -50 V. Remarkably, scanning electron microscopy ( Figure S3) and θ/2θ X-ray diffraction ( Figure S4 and Table S1) indicate that the ion migration mechanism is, at least partially, grain boundary-mediated (see Supporting Information).  (Table S2). Conversely, the sample treated at -200 V for 11 agreement with the XPS characterization (green square in Figure S5b, Figure 3c,d and Table S3).
This confirms that, for negative applied voltages, O migrates from Co3O4 and leaves behind metallic Co areas. Similar results were also obtained from regions closer to the SiO2 interface (see Figure S6, which correspond to the red squares in Figure S5), evidencing that, at such voltages, the process not only reduces the surface but also affects the whole Co3O4 film (Tables   S4-S7). In Figure S7, a HCP-Co nanocrystallite is identified and highlighted in red. The nanocrystallinity of the generated HCP-Co is consistent with the absence of clear reflections in the θ/2θ XRD pattern of the treated sample (red curve in Figure S4). Finally, HRTEM further confirms that, upon positive biasing, the initial state is recovered ( Figure S8).  Table S2 and S3). In green, blue and red, reflections corresponding to Co3O4 (JCPDF 42-  Figure 5d evidence that, upon negative biasing, oxygen ion migration is promoted along channels which act as diffusion paths, leaving aside Co-rich areas. This is in agreement with the EFTEM characterization of Figure 4c. These channels are about 40 nm-wide and appear sporadically (for instance, along an approximately 400 nm section of the sample, only one diffusion channel is observed in Figure 4c). As can be seen in Figure S10, the pristine sample sometimes exhibits a columnar-shaped morphology along its width suggesting that grain boundaries, where diffusion is usually enhanced with respect to the inner parts of the grains, may sometimes contribute to the formation of these channels which act as diffusion paths for O. As can be seen in Figure S11, which shows a high resolution TEM image of one of these O-rich channels, these paths are of highly nanostructured nature or even partly amorphous-like, allowing for an enhanced O diffusion and a large incorporation of O as it happens in amorphous cobalt oxide phases which exhibit a high catalytic response. 36,37 Additionally, Figure 5g  following the electric field which is applied perpendicularly to the sample surface (see the way voltage is applied in Figure S1), while Co cations diffuse in opposite sense, resulting in a Co-rich top part and an O-rich bottom layer. This is in agreement with the EFTEM characterization of Figure 4b. These results indicate that formation of HCP-Co is accompanied with an increase of the O content in Co3O4 (i.e., formation of cationic vacancies in the structure of Co3O4). Actually, it is known that Co3O4 can accommodate more oxygen than the stoichiometric composition in its structure, 38 particularly under certain conditions of temperature and oxygen partial pressure. 39 Here, such an effect is electric-field-induced. Similar results have been reported in electrolytegated VO2, where oxygen concentrates forming chains of edge-sharing VO6 octahedra. 24 Analogous chains of dimeric-sharing CoO6 octahedra have also been reported in CoOx nanoparticles. 36 Remarkably, a reduction of the interplanar distances has been reported in O-rich Co3O4 fibers. 38 The coexistence of stoichiometric with off-stoichiometric Co3-xO4+x regions could be correlated with the asymmetry in the XRD peaks of the negatively biased sample ( Figure S4). Anyhow, when negatively gating the system, no oxygen bubbling is observed, ruling out O2 formation. Assuming that no oxygen dissolves in the liquid electrolyte, O must redistribute along the sample while preserving charge neutrality. In contrast to crystalline materials, 40   To rule out spurious effects arising from Na + from the electrolyte or Si 4+ from the substrate, energy-dispersive X-ray (EDX) spectroscopy and EELS spectra were acquired on the sample treated at -200 V for 30 min. Neither Na nor Si was detected (Figure S12, S13 and S14). The mechanism reported herein is thus different from previous works on lithiation of ZnFe2O4, CuFe2O4 and γ-Fe2O3 aiming at partially tuning their magnetization by electrochemical treatment in 1 M LiPF6 in ethylene carbonate and dimethyl carbonate solutions. 28,41 Finally, Co and O EELS mappings were also performed for the fully recovered sample and, as expected, a homogeneous composition throughout the film close to stoichiometric Co3O4 was obtained ( Figure S15).
Thus, the structural characterization clearly demonstrates that, in contrast to conventional magneto-ionic systems where there is a source/reservoir of oxygen to trigger the redox processes, [15][16][17][18] in the here-prepared Co3O4 films the magnetic switch is accomplished by a reversible atomic scale reconfiguration of the O and Co ions within the film itself. keV and normalized to a Co defect-free reference spectrum, are presented in Figure 6d for the asprepared, treated and recovered samples, where annihilation events come from the film region only (see Figure S18). The low momentum part of the spectra, pL<8×10 -3 m0c, is simply another representation of S (Ep=3 keV) from Figure S16. The high electron momentum part of the ratio plots, pL>8×10 -3 m0c, shows a minimum at pL≈16×10 -3 m0c for all the samples, which is in qualitative agreement with the calculated curves from Figure 6e. The high momentum part is a fingerprint of the defect site atomic surrounding and, interestingly, it does not consist of pure HCP-Co phase but, instead, is in agreement with the calculations for mixed vacancies involving different amounts of O and Co atoms (Figure 6e). This does not exclude Co segregation, since positrons preferentially annihilate at the oxide phase (due to their higher affinity to oxides). Table S8 shows the calculated positron lifetime values for different types of vacancy defects in HCP-Co and Co3O4. The results indicate that existence of small vacancy clusters, probably in the form of cobalt vacancy dimers (two vacancies), V2xCo, or trimers (three vacancies), V3xCo, coupled with an oxygen monovacancy (single vacancy), VO (where a lifetime higher than 0.2 ns is obtained). These configurations, compatible with our experiment, are indicated in bold in Table S8 (further details can be found in the Supporting Information). The experimental values of τ1 corresponding to the treated samples are still larger than the calculated value of 0.2526 ns which corresponds to 6 vacancies (see Figure 6a and Table S8) upon the assumption of a stoichiometric and crystalline Co3O4 phase in the simulations. This discrepancy is in concordance with the formation of highly nanostructured or even amorphous-like phases upon negative biasing. In conclusion, voltage-driven O and Co redistribution has been demonstrated in 100 nm-thick Co3O4 films through electrolyte-gating, allowing for the controlled generation and suppression of ferromagnetism. A negative voltage reduces Co3O4 to Co (ferromagnetism: ON), whereas the process is reversed by applying a positive bias, aimed at oxidizing Co back to Co3O4 (paramagnetism: OFF). These gate-induced O and Co migrations are driven by mixed vacancy clusters as evidenced by positron annihilation spectroscopy. Ionic transport seems to be promoted at grain boundaries and is further assisted by the formation of diffusion channels that incorporate large amounts of O. Part of the generated ferromagnetism is not transient but stable, although it can be easily erased by applying adequate positive voltage values. The process is selfsustained in the sense that no external source/sink of oxygen is required. Our approach also circumvents the need of thermally-assisted ionic migration, i.e., voltage-driven oxygen motion takes place at room temperature. Furthermore, our procedure could in principle be extended to room temperature oxide-based antiferromagnets (e.g., NiO), allowing for an extra degree of freedom taking advantage of exchange bias effects.

Sample preparation
Co3O4 films were grown on thermally-oxidized Si wafers (i.e., SiO2 (100 nm)/ <100> Si (1 mm)) by plasma-enhanced atomic layer deposition (PE-ALD) using bis(cyclopentadienyl)cobalt (CoCp2, STREM, min. 98%) as Co precursor and O 2 plasma as oxygen source. 30 The depositions were performed in a home-built ALD reactor with a base pressure of 10 -6 mbar. 30,31 The container with the solid CoCp2 precursor and the tube to the reactor were heated to 80°C and 100°C, respectively, to prevent deposition of the precursor in the tubes or valves. The Co3O4 layers were produced by alternate pulsing of CoCp2 and O2 plasma at a deposition temperature of 200°C. Argon was used as a carrier gas for the Co precursor. The ALD cycle consisted of 5 s CoCp2 exposure, 5 s of pumping, 5 s O2 plasma exposure and 5 s of pumping. During the exposures, the pressure in the ALD chamber was in the 10 -3 mbar range. The growth rate was 0.5 Å/cycle, and a total of 2000 ALD cycles were applied to obtain 100 nm thick Co3O4 films.

Structural and compositional measurements
Scanning electron microscopy (SEM) images were acquired using secondary electrons in a FEI Magellan 400L microscope operated at 20 kV. Transmission electron microscopy (TEM), scanning transmission electron microscopy/energy-dispersive X-ray spectroscopy (STEM/EDX) and electron energy loss spectroscopy (EELS) were performed on a TECNAI F20 HRTEM /STEM microscope operated at 200 kV. Cross sectional lamellae from the as-grown, treated (−200 V for 30 min) and recovered (−50 V for 30 min / +100 V for 1 h) samples were prepared by focused ion beam (FIB) and placed onto a Cu TEM grid. θ/2θ X-ray diffraction (XRD) patterns of the different samples were recorded on a Philips X'Pert Powder diffractometer with a Pixel1D detector in the 17°-40° 2θ range using Cu Kα radiation with intensity (Cu Kα2)/intensity (Cu Kα1)=0.5. The structural parameters of Co3O4, such as crystallite size (i.e., average coherently diffracting domain) or lattice parameter were evaluated by fitting the XRD patterns in the 18°-20° 2θ range using the MAUD Rietveld refinement program. 50 Variable Energy Positron Annihilation Spectroscopy (VE-PAS) 42 beamline, which provides mono-energetic but variable energy positron beam. Further details on this type of experiments, as well as on positron annihilation lifetime spectroscopy (PALS) and