Voltage control of magnetism with magneto-ionic approaches: Beyond voltage-driven oxygen ion migration

The proliferation of modern electronic devices continues at a furious pace, integrated into everything from advanced scientific equipment to everyday household items. They form the backbone of technologies, such as the Internet of Things, Artificial Intelligence, and Big Data—technologies pushing current computer hardware and architectures to their limits, both in terms of functionality and energy efficiency. Functionally, the ubiquitous von Neumann architecture found in nearly all computers, while excellent at computing many sequential tasks very quickly, struggles to perform complex tasks, such as pattern recognition on large but imprecise or incomplete data sets. The biological brain, however, performs such task orders of magnitude more efficiently, inspiring neuromorphic computing,¹ a paradigm, which mimics the structure of the biological brain by de-centralizing duties away from single-task units and redistributing both memory and computation functions among a large number of neurons connected by synapses. Realizing such structures requires moving beyond volatile hardware units and relying instead on hardware able to offer analog tuning of properties depending on the stimulus provided, allowing for “learning” and “forgetting” functionalities, among others. In terms of energy efficiency, concerns regarding production and development portend the death of conventional hardware in the coming 10–15 years. Neural networks have applied neuromorphic principles to conventional computer hardware with tremendous effect,^2–4 but fundamental issues, such as manufacturing limitations, the so-called dark or dim silicon, and power and cooling restraints have already left current devices struggling to maintain efficiency at dimensions ≤10 nm,^5,6 sparking a search for life after Moore's Law.⁷ Spintronics^8,9 has enjoyed success using spin-polarized current to induce low-power magnetization reversal via spin-transfer torque^10,11 or spin–orbit torque,^12,13 but these still rely heavily on electric current to manipulate magnetic elements, leading to Joule heating. This concern also extends to, among other technologies, micro- and nanoelectromechanical systems (MEMS and NEMS) and hard disk drives, where current is used to manipulate magnetic elements. These challenges, coupled with expected worldwide electricity demands and environmental impacts of existing technology,^14–16 have led to an intense search for low-power, energy-efficient materials with flexible functionalities. One proposed solution to this challenge is the use of an applied external electric field, instead of current, to modulate the magnetic properties of a material. The expanding number of materials and mechanisms that produce this effect has grown into a field of research broadly classified as voltage control of magnetism (VCM).^17–22 

The intrinsic coupling between an applied magnetic (electric) field and a medium's electric polarization (magnetization) was first proposed by P. Curie when discussing the Wiedemann effect,²³ a coupling later coined the magnetoelectric effect (ME) by Debye.²⁴ Landau and Lifshitz postulated a linear ME effect in select magnetic crystal symmetry classes,²⁵ which Dzyaloshinskii extended to predict its existence in Cr₂O₃,²⁶ confirmed by experiment soon after.^27–30 This sets the stage for the pursuit of ME effects in single-phase multiferroics,^31,32 a class of material, which exhibits two or more types of ferroic order.³³ However, the fundamental requirements for a material to possess both ferroelectric and ferromagnetic order make multiferroics rare,³⁴ pushing research beyond single-phase materials. Composite materials^35–40 generate the ME effect by stitching together two ferroic phases, coupled by some mediating mechanism. Piezo-strain mediated ME effects utilize lattice changes in a ferroelectric material under bias to “strain” a magnetic material, inducing a change in its magnetic properties.^41–50 Carrier density-mediated ME effects rely on an applied electric field to tune surface charge density in magnetic semiconductors,^51–54 ultra-thin metallic films,^55–59 and complex magnetic oxides sensitive to carrier doping at the interface of complex-oxide heterostructures.^60–64 Exchange bias-mediated ME effects, where the interfacial exchange interaction between an antiferromagnetic ferroelectric and a ferromagnetic layer results in a biasing of the magnetization curve away from zero, can be controlled by applying voltage to the ferroelectric layer in several systems.^65–70 Orbital reconstruction ME effects⁷¹ occur when an applied field induces orbital reconstructions and hybridizations of atoms at the boundary between layers.^72–77 To these, we add a rapidly growing sub-field of VCM, the electrochemical effect, which has only emerged within the last decade. Dielectrics, such as ionic liquids, GdO_x, HfO₂, AlO_x, Y₂O₃-doped ZrO₂, and SrCoO_2.5,^78–86 have opened up the possibility to significantly alter the oxidation states of an adjacent material. The electrochemical effect is generally split into two mechanisms: electrostatic (charging) effects and magneto-ionics. The presence of both effects was shown in ultra-thin La_0.5Sr_0.5MnO_3-δ,⁸⁷ with negative bias, inducing reversible hole accumulation (electrostatic), and positive bias, inducing irreversible oxygen vacancy formation (magneto-ionics). Strictly electrostatic effects were demonstrated in the magnetic oxide semiconductor (Co,Ti)O₂,⁸⁸ while Walter et al.⁸⁹ very recently demonstrated the fascinating result that diamagnetic (zero-spin) FeS₂ could be biased into a ferromagnetic state by electrostatically induced band filling alone. The evolution of VCM and some important milestones are presented in the top timeline of Fig. 1(a), with selected works referenced.

Magneto-ionics^83,84 first appeared in ultra-thin Co films adjacent to GdO_x, accomplished by inducing redox reactions at the interface under an applied bias, changing its magnetic state. A representative schematic of a typical magneto-ionic structure is shown in Fig. 1(b): a dielectric (either liquid or solid-state) is placed in contact with a target material (which maybe an oxide or a metal), with voltage applied across the film and dielectric, inducing ionic motion. More generally, magneto-ionics uses an applied electric field to insert ions into and remove them from a target material, undergoing changes in stoichiometry, oxidation state, crystalline structure, as well as large and cyclable changes in magnetic properties [Fig. 1(c)].^82,90–97 Magneto-ionic materials also offer several potential advantages for practical applications: They potentially require lower energy to actuate (as low ≈10 aJ/bit,^98,99 much lower than traditional current-using methods, ≈10 fJ/bit¹⁰⁰), no voltage to maintain a set state (nonvolatile) and offer the possibility of functional plasticity (i.e., the ability to undergo nonvolatile changes in its structure and properties in response to an external stimulus), making magneto-ionic materials intriguing candidates to be deployed in neuromorphic or stochastic computing, magnetic random access memory, domain-wall logic, and lab-on-a-chip devices.^101–103 Indeed, magneto-ionics has already been identified for potential use in several applications. Magnetic actuation is key in microelectromechanical systems,^104–107 where magneto-ionics may prove utile in fabricating multifunctional robotic materials.¹⁰⁸ The use of ionic transport to modify magnetic properties has already been observed in Pt/Co(Fe)/MgO, Ta/CoFeB/MgO, and Pt/Co/Ni/HfO₂ stacks,^79,109,110 representative structures for magnetic random-access memories^58,111 and envisioned to allow for low-power tunability of the switching elements via the insulating tunnel barrier. Spin textures, such as skyrmions,^112,113 chiral domain walls,^114,115 and spin spirals,^116,117 are stabilized under the Dzyaloshinskii–Moriya interaction (DMI), which have been shown to be tunable using magneto-ionics,^118–121 with possible applications in spin-orbitronics.¹²² Continued fundamental research has led to further exploration of different structures, materials, and mobile ions, with successful demonstration of magneto-ionic effects using hydrogen,^{118,123–125} lithium,^126–130 nitrogen,^131–133 and flourine.¹³⁴ The last two years have seen a flurry of exciting activity, a few of which are highlighted in Fig. 2. In 2019, Tan et al.^123,124 showed that insertion of H⁺ into the interface of Co/GdO_x stacks increased the speed of nondestructive toggling of magnetic anisotropy, demonstrating that the powerful contribution of H⁺ plays in altering Co oxidation state under voltage, while recent work by Huang et al. has shown voltage-control of the sublattice in ferrimagnetic GdCo and toggling magnetic order.¹³⁵ Yi et al.⁹¹ reported reversible and nonvolatile phase transitions at room temperature under voltage using both H⁺ and O²⁻ ions in digital complex oxides, showing that nondestructive structural changes under bias are possible [Fig. 2(a)]. Murray et al.¹³⁶ demonstrated that exchange bias can be controlled in Gd/Ni_1-xCo_xO thin films [Fig. 2(b)], where interfacial Ni_1-xCo_xO is reduced to magnetic NiCo to establish an exchange bias that can be modulated by up to 35% under voltage conditioning. Zehner et al. and Mustafa et al. also showed the ability to tune exchange bias under voltage by inducing redox reactions due to O²⁻ and Li⁺, respectively.^125,137,138 Chen et al.¹¹⁸ reported inducing Dzyaloshinskii-Moriya interaction by chemisorption and desorption of H⁺, seen in reversible and nonvolatile chirality transitions of magnetic domain walls [Fig. 2(c)]. Navarro-Senet et al.^139,140 and Robbennolt et al.^141,142 have produced dramatic electric field induced changes in the magnetic properties of nanoporous, mesoporous, and nanostructured materials with high surface-to-volume ratios, extending magneto-ionic phenomena beyond thin film structures [Fig. 2(d)].

The changes occurring primarily by redox reactions at the interface of heterostructures have motivated studies of magneto-ionic effects in thin metal films. However, systems that utilize target ferromagnetic metal layers, such as Co, are limited to ultra-thin films (≲5 nm), where patterning on the nanoscale leaves them susceptible to superparamagnetic transitions and often lack nonvolatile states. In addition, oxygen ion migration has often required heat assistance to aide ion mobility,^{82,84,123,124,143} reducing potential energy efficiency. Effects beyond the interface have also been explored, although this brings the added complication of relatively large oxygen ions moving through the film, resulting in irreversible structural or compositional changes as the ions move along grain boundaries and fill vacancies, limiting reversible toggling to redox reactions at the interface.^{82,119,144,145} As a simple illustration, consider that in the transition from metallic Co to CoO, the change in relative cobalt positions while changing phase would approach 20%. To improve cyclability and durability, these kinds of stresses and distortions should be minimized, potentially by using the migrating ion in the as-deposited state. Oxide/metal heterostructures in thin films and islands have been explored using Fe/FeO_x^96,146 and Co/CoO_x structures, allowing the reversible removal and insertion of oxygen to toggle magnetic states. In particular, on-off switching of magnetism has been observed in iron-based hybrid transition metal oxide/metal,^90,140,147 transition metal oxide^148–151 films, and porous structures. Along this vein, we discuss our recent work on the structural-ion approach, where the target material possesses the migrating anions in the as-deposited state, in single layer oxide and nitride films, and so providing a crystal structure with more local vacancies and larger ionic pathways, facilitating ion movement in and out of the target material with reduced structural distortion.

We begin with magneto-ionics using the structural oxide approach in spinel Co₃O₄ films.^150,152 100 nm-thick Co₃O₄ films, prepared by plasma-enhanced atomic layer deposition onto SiO₂, were voltage biased using propylene carbonate (PC) with solvated Na⁺ ions. PC is a polar anhydrous liquid electrolyte, which generates a very large electric field with low voltage due to an electric double layer (EDL), allowing for magneto-ionic effects. Representative hysteresis loops of magnetization, M, as a function of applied magnetic field, H, measured by vibrating sample magnetometry (VSM) before and after the application of voltage, are shown in Fig. 3(a). The as-grown Co₃O₄ film initially shows no ferromagnetic signal, but after biasing under −10 V for 60 min, it shows an apparent hysteresis loop, clear evidence of a ferromagnetic response (on). This ferromagnetic signal is completely removed by subsequently applying a bias of +20 V for 60 min (off), i.e., demonstrating nonvolatile on-off control of ferromagnetism at room temperature. Additionally, the magneto-ionic response is found to scale with the increasing voltage; a series of biases (−10, −25, and −50 V) applied for 60 min results in increasing saturation magnetization, M_S, a result of increased voltage-driven ion migration and further reduction of Co₃O₄ to a metallic Co phase. The series was performed on the same film after applying a positive bias to return the sample to its original magnetic state (off), showing that cyclability is possible on the scale of 10² s, achievable without thermal assistance. On states are observed to have excellent stability, remaining steady and nonvolatile for months after on-off toggling. (Before and after M–H loops are provided in the supplementary material.) Interestingly, the redox process is found to penetrate deep into the film and, under high applied voltage (−200 V), introduces strong compositional and structural inhomogeneities, as observed by high-angle annular dark-field scanning transmission electron microscopy (HDAAF-STEM) and electron energy loss spectroscopy (EELS).¹⁵⁰ In the top figures of Fig. 3(b), it is clear that the Co and O distributions are isotropic across the film, consistent with Co₃O₄. In the bottom figures of Fig. 3(b), biasing at −200 V clearly reveals a mixture of Co and O concentrations. The ion flow is funneled through irregular O-rich diffusion channels, which are highly nanostructured or partially amorphous, leaving Co-rich areas in its wake. Fascinatingly, the Co-rich and O-rich regions are formed by a combination of redistribution of cobalt and oxygen within the film, as well as oxygen moving toward the liquid electrolyte. Structural-ion systems also possess nonvolatile properties, which can be applied to neuromorphic architectures, where tuning of oxidation states as a function of depth in the film leads to continuous, analog changes in M_S, coercive field, H_c, or the switching field distribution (SFD), allowing for hysteretic “learning” or “forgetting” processes depending on the strength and duration of the applied field. To further extend the utility of cobalt oxide, the effects of the gating configuration on magneto-ionic speed were investigated by comparing transistor-like with capacitor-like configurations on two identically sputtered Co₃O₄ films. To understand the effects of device configuration, numerical simulations, modeling the initial voltage distributions for each configuration upon electrolyte-gating, were performed. The results show a stark difference in the electric potential of each configuration [see Fig. 3(c)]. The transistor-like configuration, due to the limited conductivity of Co₃O₄ and the insulating nature of the SiO₂ buffer layer, results in an inhomogeneous electric field along the film's length and width, weakening further away from the electrical TiN contact. The capacitor configuration, on the other hand, forms a nearly uniform electric field distribution along the length/width of the film (due to the electrically conducing TiN layer), resulting in a fully activated Co₃O₄ layer. The two Co₃O₄ systems were biased at −50 V as magnetic hysteresis loops were continuously recorded. In Fig. 3(d), the evolution of the saturation magnetizations, M_S, for both systems is plotted as a function of time. While both systems show an increase in M_S, the capacitor configuration shows a near sixfold improvement in total magnetization (119 vs 700 emu cm⁻³), indicating a much stronger magneto-ionic effect. Indeed, the on state in the capacitor configuration is activated faster than the transistor-like structure, with over a 35-fold improvement in magnetization rate (9 vs 325 memu cm⁻³ s⁻¹), showing that the biasing configuration plays a key role in the magneto-ionic performance of structural ion films. Aside from a much-improved magneto-ionic rate, the on state possesses a narrower H_C distribution [Fig. 3(e)], and the minimal onset threshold voltage is reduced (−4 vs −10 V), a direct consequence of the chosen configuration. The observed TEM and EELS images taken of lamellae cut from regions distant from the contact are in agreement,¹⁵² showing a weakened ionic effect in the transistor-like configuration, although further examination would further illuminate the mechanisms at play, as analysis until now has not attempted to include dynamics. For example, TEM/EELS or VEPALS data taken for a series of exposure times, both near and far from the electrical contact, could provide further insight on how ions are distributed under bias using these configurations. While possessing mild activation rates, applications in neuromorphic and stochastic computing or magnetic MEMS are possible.⁹⁵ In addition, transition metal oxides, such as Co₃O₄, are relatively easy to grow in via reactive sputtering, making them well-suited for industrial applications.

To further extend the potential of the structural ion approach, an examination of the functionality of ions beyond oxygen in magneto-ionic target materials would likely prove fruitful. Indeed, we have shown that nitrogen^132,133 has significant advantages, both energetically and functionally, over their oxygen-based counterparts. Nitrogen¹³³ magneto-ionics was first observed in paramagnetic, polycrystalline CoN sputtered onto a Cu/Ti buffer layer that was biased much in the same way as the above Co₃O₄, using PC with solvated Na⁺ ions to electrolyte-gate the films at -50 V while measuring M vs H loops. In Fig. 4(a), M_S is shown as a function of time, where the magneto-ionic rate of CoN is clearly much higher than Co₃O₄ films of the same thickness sputtered on identical substrates. M_S increases with time, asymptotically, reaching 588 emu cm⁻³ for Co₃O₄ and 637 emu cm⁻³ for CoN. The linear magnetization rate of CoN is calculated to be 722 memu cm⁻³ s⁻¹, a fivefold enhancement over sputtered Co₃O₄. However, it should be emphasized that this value may be considerably underestimated; the measurement device (VSM) is limited to measurements on the order of 10⁻⁷emu and 10⁻¹ s, so any assessment of set and reset events should be taken as a rough approximation. These promising results suggest that perhaps other transition metal nitrides,¹⁵³ such as the Fe–N system,¹⁵⁴ are also potential magneto-ionic candidates with built-in industrial qualities, including high hardness, melting point, and tunability of properties with nitrogen concentration. Indeed, a study on polycrystalline iron nitrides¹³² reveals the viability of FeN; M_S as a function of time is presented in Fig. 4(a) alongside CoN. Interestingly, FeN exhibits a distinct multistep process, with M_S initially increasing at a rate slower than CoN but faster than Co₃O₄, attributed to the appearance of a Fe₃N phase, and then slowing down and reaching a maximum M_S of 898 emu cm⁻³, i.e., higher than CoN. Perhaps the most interesting feature of nitrogen magneto-ionics, beyond the increased magneto-ionic rates, is the exact way in which the ions are transported from the film. An examination of the films after biasing at -50 V using TEM and EELS gave a fascinating insight as to how the different anions move through the films [Fig. 4(b)]. After biasing at −50 V, the Co₃O₄ film redistributes oxygen into irregular O-rich channels, similar to that seen in Fig. 3(b). However, in CoN, a well-defined, planar interface parallel to the surface is observed, dividing the film in two sublayers with two distinct microstructures. No nitrogen is detected in either of the two sublayers, escaping beyond the film and into the electrolyte. Sublayer 1 (near the electrolyte) is found to be amorphous-like and less dense, while sublayer 2 (near the electrode) is highly nanocrystalline. A similar, although somewhat rougher, sublayer structure is observed in FeN, reflected in the homogeneous distribution of atomic Fe and N near the working electrode and a reduced amount of N atoms near the electrolyte side, consistent with the formation of Fe₃N. To understand if this planar migration front is consistent through the voltage range and perhaps correlated with defect structure, variable energy positron annihilation lifetime spectroscopy experiments for CoN were performed and results are shown in Fig. 4(c).^132,133 For simplicity, only the intensity (I₂) of lifetime τ₂, corresponding to the density of defects in the form of surface states and grain boundaries, is shown. A monotonic increase in relative intensity I₂ is found in the as-prepared CoN, whereas relative jumps are found for the treated CoN films (highlighted in yellow). The depth position of these relative maxima increases with the applied voltage (−20 to −50 V), confirming the occurrence of a planar migration front, in agreement with TEM observations. The ionic transport upon electrolyte-gating in CoN is, thus, consistent with a uniform nitrogen migration through vacancies and grain boundaries, leaving behind a larger density of grain boundaries. Although not shown here, a migration front is also observed in FeN,¹³² even under the minimal bias needed to activate ionic motion (onset voltage). It must be noted here that this planar front is a unique trait, one which can be imagined to allow the imprinting of magnetic shapes parallel to the applied field, and potentially allowing for the toggling of related magnetic and stoichiometrically controlled crystalline states. This planar front is currently attributed¹³¹ to the highly nanostructured nature of the films, isotropic grains, and a defect gradient along the depth of the film, with more defects near the top and fewer near the working electrode, resulting in faster ionic motion occurring near the surface under voltage, while relatively fewer defects result in lower ionic rates deeper in the film. This, along with the smaller, isotropic grains compared with Co₃O₄ and lower electrical resistivity effectively makes the ionic front penetration depth a voltage dependent quantity. While CoN does generate magnetic phase faster than FeN and with a more clearly defined planar front, the generated ferromagnetic phases possess different hysteresis loop shapes. In general, iron nitride films show a larger increase in coercive field, H_C, up to ≈10² Oe, compared with ≈10¹ Oe in cobalt nitride, indicative of relatively smaller activated magnetic regions, comprising single- or few-domain structures in mostly denitrided regions. This suggests that FeN, while slower, may be more useful for magnetic memory applications. The energy required to set a magnetic state was evaluated by an onset and reset process, shown in Fig. 4(d). Not only is the onset voltage for CoN (−4 V) found to be lower than that for Co₃O₄, but it also recovers the paramagnetic state under biasing at +4 V (for Co₃O₄ a larger positive voltage is required). The insets of Fig. 4(d) show EELS images of the CoN film during the onset and recovery processes, revealing that under −4 V biasing, CoN develops a slightly nanoporous structure and a decrease in the nitrogen content near the electrolyte, while, after +4 V biasing, the film is nearly fully recovered with very small nitrogen-deficient regions. This polarization symmetry is a distinct advantage of N³⁻ over O²⁻, which requires asymmetric biasing to recover the original state (i.e., a positive bias of +20 V must be applied to recover a sample biased at −6 V). Similar results are observed in FeN, where complete recovery is achieved under equivalent positive biasing. To understand this difference between oxygen and nitrogen, formation energies of Co–O, Co–N, and Fe–N, shown in Fig. 4(e), were calculated by determining minimum energy paths of N inserted into a transition metal slab, plotted as a function of the position of the oxygen or nitrogen relative to the surface (reflective of an on-off transition). The critical electric field needed to induce oxygen or nitrogen motion can be estimated by E_C = ΔV/Δz, where ΔV is the electric potential per atom to overcome the energy barrier and Δz is the distance between minima. This critical field is found to be ≈8.1 V nm⁻¹ for Co₃O₄, ≈6.6 V nm⁻¹ for FeN, and ≈5.3 V nm⁻¹ CoN, in agreement with the observed onset voltages. Although the crystallographic structures of CoN, FeN, and Co₃O₄ were not exactly reproduced in the modeling, it is still a reasonable approximation of an off-on process in a polycrystalline film. These results complement recent calculations, which show that CoN has a lower energy barrier compared to Co₃O₄ as well as previous simulations where CoN was found to have a lower cohesive energy than FeN.^155,156 Nitrogen magneto-ionics, thus far, presents a desirable blend of qualities as a magneto-ionic anion, such as requiring lower activation voltages, higher cyclability (under symmetric bias), much faster rates of motion and, perhaps most interestingly, a planar ionic motion front that generates a voltage-programmable ion depth profile, potentially allowing for the nonvolatile imprinting and destruction of magnetic sublayers. These features, such as the voltage-tunable analog changes in M_S [Fig. 4d)], may open the door to an eventual utilization of magneto-ionics in neuromorphic applications. Remarkably, the magnetic signals described here rely on voltage-driven oxygen or nitrogen ion migration, much like ion channels activated with electric stimuli in neurons' membranes in the brain.

Extending magneto-ionic targets with structural-ions toward device viability require tackling several hurdles, including the effect of dimensional reduction and patterning. As a preview, initial steps toward “imprinting” a magnetic region into a structural-ion target under biasing are shown for a 100 nm thick, polycrystalline Co₃O₄ film sputtered onto a Cu/Ti/SiO₂ substrate. Cu stripes were then patterned atop the Co₃O₄ film, leaving only the exposed (uncovered) Co₃O₄ regions in contact with PC during biasing. The Cu stripes were fabricated by photolithography using a photoresist of stripes with dimensions of 2.4 mm × 5 μm and a pitch of 12 μm, atop which a 100 nm thick Cu layer was then sputtered. Once the photoresist was lifted off, alternating Cu and exposed Co₃O₄ stripes were left behind, constituting 45% of the area, as shown in Fig. 5(a). We have assigned directionality to the stripes on the film, with 0° corresponding to the direction parallel to the lines, and 90° corresponding to the direction perpendicular to the lines. Once again using PC, the samples were biased at −50 V for 75 min, with hysteresis loops measured by VSM post-biasing. Biasing not only results in magneto-ionic effects in the exposed Co₃O₄ stripes but, to a lesser extent, also below the Cu lines, results in a well-defined patterned magnetic composite. The patterns even exhibit a magnetic shape anisotropy depending on the measurement direction, as seen in Fig. 5(b), a feature not seen in treated samples without patterned Cu stripes. This confirms a magneto-ionic generation of magnetic anisotropy. Here, the hard-axis is found to be along the stripe short axis (90°), consistent with a hard axis due to shape anisotropy perpendicular to the stripe long axis, while the easy axis direction, 0°, is consistent with shape anisotropy observed parallel to the stripe long axis, suggesting that the observed hard and easy axes are the result of selectively activating Co₃O₄ regions due to the Cu mask. To further explore the magnetic imprinting achieved in the film, atomic force microscopy (AFM) and magnetic force microscopy (MFM) measurements were performed, as shown in Fig. 5(c). The topography by AFM (top figure) shows the masked film, with the broader, lighter color corresponding to the Cu mask, and the darker regions corresponding to the exposed, underlying Co₃O₄. The magnetic profiling by MFM (bottom figure) shows the MFM signal as measured at (non-saturated) remanence after applying an external magnetic field of 100 Oe to the film. A stripe pattern can be identified along the 0° direction, with clear magnetic domain structure whose sizes match the width of the exposed vs Cu-masked regions, confirming the generation of a well-defined patterned magnetic composite (due to the dissimilar magneto-ionic strengths in the non-masked and masked regions of the Co₃O₄ film). These stripes are even more pronounced under higher applied field (−450 and 450 Oe), shown in Fig. 5(d), with excellent demarcation of the exposed Co₃O₄. To aid in visualizing the evolution of the magnetic domains under an applied field, a movie is provided in the supplementary material. Along the same vein, nitrogen magneto-ionics has also been demonstrated on 125 nm thick polycrystalline FeN reactively sputtered onto a Ti/Cu substrate, subsequently masked using electron beam lithography patterning to obtain dots of 2 μm in radius with a pitch of 10 μm. Figure 5(e) shows the resulting FeN dots, which exhibit a clear ferromagnetic-on state when the dots were biased at −50 V for 75 min, with hysteresis loops measured by VSM after biasing was complete [Fig. 5(f)] the first demonstration of magneto-ionics in a patterned nitrogen magneto-ionic target.

Magneto-ionics research in the past few recent years has opened a plethora of possible materials chemistry, structure types, and magnetic states to be explored. Perhaps the most glaring involves operational speeds, durability, industrial deployment, and establishing functionalities (such as plasticity in neuromorphic computing). Many systems suffer from slow room-temperature ionic response (≈10²–10³ s) and are typically boosted via heating to induce ionic motion (at the expense of energy efficiency) in the range of 10⁻¹ s or below,^{82,84,143,157} which is much slower than current transistor flip speeds (≲10⁻¹⁰ s), although proton-based systems are pushing switching times down to the order of 1 ms or lower.^{81,123,124,135} In the structural ion systems discussed so far in this Perspective, switching speeds and cyclability must be addressed before they can be considered competitive with other magneto-ionic approaches. Switching states in structural ion systems require 10²–10³ s at room temperature, well behind H⁺ or heat-assisted O²⁻ systems. Due to the reduction of energy efficiency when applying heat treatment to improve ionic motion, alternate approaches have been considered to reduce switching speeds. The structural ion films studied so far have been relatively thick when compared to other approaches (≈50–100 nm). Very recent work on oxide thin films⁹⁷ and an upcoming work on nitride thin films¹⁵⁸ with reduced thicknesses (≤25 nm) have demonstrated switching speeds <10 s for oxygen ion motion and <1 s for nitrogen ion motion. To achieve speeds of 1 ms or faster, however, additional work must be done, such as further microstructure tuning by, for instance, defect engineering through ion irradiation¹⁵⁹ and/or miniaturization by micro-/nanofabrication to promote surface diffusion over bulk diffusion since the latter results in faster ion mobility.¹⁶⁰ Moreover, in order to expand technological applications, other forms of the dielectric layer used to bias the films should be considered. Up to now, the structural-ion films discussed have been biased using a liquid electrolyte, a common and convenient biasing tool. The massive electric fields achieved via EDL formation are excellent for fundamental research, as is the relative ease of use (compared with sputtering a solid-state ionic conductor) and ability to apply biasing in exotic motifs outside thin films (foams, nanoporous materials, pillars, etc.). However, in order to move toward device applicability, solid state heterostructures using a dielectric sublayer to gate the films will be necessary. The fastest room temperature switching in oxygen-based systems has reached speeds up to 0.2 ms,⁸⁰ utilizing the very high oxygen mobility in SrCoO_2.5 to produce limited but measurable interfacial interactions, an indication that properly tuned dielectrics can have a huge impact on ionic migration. Generally speaking, progress via hydrogen boosting, as seen in GdO_x,¹²³ and battery-like intercalation can be promising,^130,134 while in the specific case of nitrogen magneto-ionics solid state structures will require exploring possible nitrogen reservoir sources¹⁶¹ (such as zirconium oxide-, tantalum oxide-, or mayenite-based nitrides, or GdN_x) that can assist the insertion and extraction of nitrogen ions from transition metal nitride layers.

In terms of durability, state-of-the-art magneto-ionic materials have shown high quality repeatability well over 1000 cycles,^81,124 while O²⁻ systems, including structural ion, have demonstrated 100 cycles or less. This can partly be attributed to the deformation or changes introduced when inducing ionic motion beyond the interface and may point toward three-dimensional, high surface-to-volume ratio structures focusing on interfacial effects being a more practical option. However, tailoring ionic pathways by, for example, ion irradiation of the target films may allow for more facile insertion and extraction of ions, improving durability and ionic rates. This can be done not only with He, Ar, or Xe ions (inert ions) but also by the same magneto-ionic ion species, such as O²⁻ or N³⁻. This process is known to allow for tuning of structural defects and may affect ionic pathways and the electric field generated inside the target material.¹³¹ The structural ion approach allows for tuning ion vacancies or enriching regions of the target material and consequently modifying magneto-ionic performance, as recently demonstrated by Jensen et al. in Gd/NiCoO.¹⁶² Indeed, structural ion materials are still far from being fully optimized. The threshold voltage needed to actuate nonvolatile on-off behavior is ±4 V in CoN and ±8 V in Co₃O₄, comparable with voltages applied in other systems as can be found in a recent Perspective by Gu et al.,¹⁶³ although −50 V was needed to extract nitrogen up to a depth of 40 nm in relatively thick films. However, while several studied systems have required an asymmetric voltage to recover the pristine state and, in many cases, the phase transformations are irreversible, our work has demonstrated that symmetric and recoverable states can be achieved in transition metal-nitride systems, suggesting that there is still more to learn about these systems. In particular, recent works have shown that DC voltage pulsing at frequencies in the range of 1–100 Hz results in cumulative magnet-ionic effects,^97,158 allowing for the frequency-dependent tuning of magnetic parameters, a path toward the neuromorphic applications of magneto-ionics, still in its infancy.^18,93 The observed dynamics under DC pulsing include a cumulative increase and partial recovery of magnetization (analogous to potentiation and depression), threshold activation, and spike-rate-dependent (i.e., frequency-dependent) magnetic plasticity for learning functionalities, giving access to functions seen in the biological synapse. Beyond ferromagnetic behavior, antiferromagnetic systems, such as or Mn_xCo_1−xN or NiCoO, have yet to be exploited under magneto-ionic control at room temperature, and achieving on-off AFM toggling would represent a major step forward. Further dimensional scaling, as seen in the patterned lines and dots described above, or more exotic electrodeposited motifs, may provide a means by which to tune magnetic properties by the large-scale geometry of the system. Finally, considering the exotic phenomena exhibited by perovskite structures, the possibility of as-yet-unseen perovskite nitrides may open the possibility of complex nitride heterostructures with novel magnetic properties.¹⁶⁴ 

In summary, recent research progress in magneto-ionics is overviewed, expanding to include a host of new magnetic materials, ions, and dielectrics in this rapidly-growing field. We presented our Perspective on what role structural-ion magneto-ionics and ionics beyond oxygen can play and highlighted the intriguing value of robust, fast, low-energy room-temperature on-off ferromagnetism using nitrogen magneto-ionics. Nitrogen magneto-ionics present a robust alternative for efficient voltage-driven effects and may enable the use of magneto-ionics in devices that require endurance and moderate speeds of operation, such as neuromorphic computing, magnetic MEMS, and MRAM, and perhaps promote the use of nitride semiconductors in fields, such as electrochemical sensors, catalysis, batteries, spintronics, or iontronics.