Introduction

Over the past five decades, the performance of electronic devices improved massively driven by the reduction in the size of transistors. It is increasingly clear that scientists need to develop novel mechanisms for the control of electron charge and spin degrees of freedom. An example of this could be based on the effect of spin torque due to spin polarized current. This mechanism involves the interaction of conduction-electron spins with localized moments leading an effectively voltage controlled spin valve magnetization states1,2,3,4. These results add new possibilities for realizing earlier proposed reconfigurable non-volatile logic elements utilizing, for instance, the magnetoresistance of two coupled magnetic layers with different coercive fields5. Finding suitable materials to effectively implement these ideas is one of main issues in this exciting field. Strongly correlated conducting oxides are an important class of materials in this mission6. These materials present a fascinating interplay between magnetic and electric properties, combined with charge-orbital ordering7. The importance of complex oxide systems for the development of novel device functions was emphasized in a recent study8. However, the study stopped short of reporting experimental work to substantiate the proposed device functionalities. One of most fascinating and invaluable oxides for revealing new physics and new phenomena involving spin, orbital and charge-orbital ordering is magnetite, Fe3O4. It is the archetype magnetic material with a narrow 3d band and therefore strong-correlation effects are expected. Indeed, it shows complex charge and orbital ordering and at Verwey temperature, TV = 120 K, the material undergoes a metal-insulator phase transition leading to a sharp drop in conductivity9. Recently electric field-induced switching between the low-and high-conductance states was found in magnetite10. On the other hand, one of the possible and perhaps least explored approaches for new functionalities of spin electronic devices is based on finding ways for the direct manipulation the interlayer exchange coupling (IEC). The discovery of the giant magnetoresistance and oscillatory IEC in fully metallic multilayer systems have enabled the development of new classes of spin valve based sensors and magnetic random access memories11,12. How to manipulate magnetization in these structures with nanoscale resolution is the key question13. The switching could also be implemented by means of multiferroic materials via a voltage switchable exchange bias14 or take advantage of a ferroelectric as the barrier material where the magnetization could be switched by an applied electric field through the switching of the ferroelectric order parameter15,16,17.

In this work, we present both experimental and theoretical results that for the first time demonstrate that the magnetization states (parallel or anti-parallel) of the spin valve can be directly manipulated by charge-orbital ordering of the coupled magnetic electrodes. We investigate the IEC between two epitaxial magnetite layers separated by a nonmagnetic epitaxial MgO dielectric. We show that the state of the charge-orbital ordering in magnetite defines the strength and even the very sign of the IEC. The IEC sign switching appears at the exact temperature of the Fe3O4 high-to-low conductance transition which changes the magnetization states of spin valve from an anti-parallel (parallel) to a parallel (anti-parallel) state. This opens new interesting avenues for the control of magnetization states of spin valve via the charge-orbital ordering in coupled ferromagnets.

Results

Our experimental study is performed on fully epitaxial Fe3O4/MgO/Fe3O4/NiO/MgO(100) stacks. Stacks with different MgO spacer thickness were prepared on the same MgO substrate using a shadow mask inside the molecular beam epitaxy system (Fig. S1 in the Supplementary). The purpose of the NiO (15 nm) under-layer is to enhance the coercivity of the bottom Fe3O4 pinned layer with respect to the top Fe3O4 free layer. The thickness of both Fe3O4 layers was 10 nm and that of the MgO spacer was varied between 0.8 nm and 3 nm. Reflection high energy electron diffraction (RHEED) was employed to confirm the epitaxial growth and establish the growth mode. The presence of the RHEED intensity oscillations confirms that the films grew in a layer-by-layer mode with a growth rate of 0.3 Å/s for Fe3O4 and 0.05 Å/s for MgO respectively18. Detailed RHEED characterization can be found in the Supplementary. One can see from Figs. S2c that MgO barrier layer was epitaxially grown on the Fe3O4/NiO bilayer and that the MgO barrier with a thickness of 0.8 nm (less than 4 monolayer) has the same quality as MgO substrate which indicate a sharp interface between MgO and bottom Fe3O4 and also guaranteed a sharp interface between MgO and top Fe3O4. Fig. S3 and the red curve in Fig. 1e show magnetization versus temperature for a 10 nm thick Fe3O4 and Fe3O4/MgO/Fe3O4/NiO with 0.8 nm thick MgO barrier layer during the heating cycle. The results shown in Fig. 1e and Fig. S3 confirm that the Fe3O4 thin films undergo the Verwey transition at 115 K which is close to well established bulk values. To further demonstrate the quality of Fe3O4 and the interface between the Fe3O4 and MgO, Fig. S4 and Fig. 1f show the transmission electron microscopy (TEM) of Fe3O4/MgO/Fe3O4 multilayer and the high angle angular dark field (HAADF) scanning transmission electron microscopy (STEM) image of the Fe3O4/MgO interface respectively. Sharp interfaces between MgO and Fe3O4 layers are observed.

Figure 1
figure 1

Temperature dependent magnetic hysteresis loops measured along [100] direction and calculated IEC J1 as a function of temperature.

Magnetic hysteresis loops of Fe3O4/MgO/Fe3O4/NiO multilayers for different MgO thickness measured at 300 K (black line), 200 K (red line), 130 K (green line), 100 K (dark blue line) and 50 K (light blue line), respectively. The thickness of the MgO spacers are (a) 0.8 nm and (c) 3 nm respectively. Room temperature M (H) loops and the corresponding minor loops for samples with MgO thickness of (b) 0.8 nm and (d) 3 nm respectively. The in-plane magnetic field was applied along the [100] direction. (e) Calculated IEC J1 as a function of temperature for the sample with a 0.8 nm thick MgO spacer obtained from the experimental data and magnetization as a function of temperature during the warming cycle in the presence of a 200 Oe in-plane field (Red curve). (f) HAADF STEM image shows the sharp interfaces between Fe3O4 and MgO, where the darker columns and brighter columns in Fe3O4 correspond to Fe2+ and Fe3+ respectively.

Full size image

The magnetic properties of Fe3O4/MgO/Fe3O4/NiO stacks at different temperatures were examined by means of a Quantum Design Physical Properties Measurement System (PPMS). The in-plane magnetic field was applied along the [100] and [110] directions. Fig. 1c and Fig. S7b show the M (H) loops measured at different temperatures for the MgO layer thickness of 3 nm. At this large MgO thickness the exchange interaction between the two magnetite layers is very weak in comparison with magneto-static interactions. To confirm this, we show in Fig. 1d the minor loop. No shift was observed indicating that the top Fe3O4 can rotate freely and no IEC is present between two Fe3O4 layers. This makes the interpretation of the M (H) loops rather unambiguous. There are two characteristic field values, Hc1 and Hc2 which correspond to the coercivity of the layers of magnetite grown on the MgO and NiO respectively (See Fig. S8 in the Supplementary). The presence of the antiferromagnetic NiO layer increases the coercivity of the bottom magnetite layer due to the weak exchange bias effect which has been observed from both the M (H) measurements and planar hall effect measurements18. Fig. 1a shows the M (H) loops for the MgO thickness of 0.8 nm. Interestingly, above TV, the difference between the Hc1 and Hc2 fields for a 0.8 nm thick MgO spacer is greater than that for a 3 nm MgO spacer. In particular, the moment at zero field is almost zero which indicates a strong anti-ferromagnetic (AF) IEC between the two epitaxial Fe3O4 layers. To demonstrate further the nature of the AF IEC and rule out possible domain effects, we show in Fig. 1b the M (H) minor loop. A large positive shift of 150 Oe was observed which indicates a strong AF IEC between two magnetite films. Remarkably, the shape of the M (H) loops changes drastically for a 0.8 nm thick MgO spacer just below TV. At 50 K, one cannot distinguish clearly between Hc1 and Hc2. For a quantitative interpretation of the measured M (H) loops, we use micromagnetic theory19,20. A more detailed description about the simulation can be found elsewhere19,20. These results are summarized in Fig. 1e showing that, above TV, the two Fe3O4 layers are AF exchange coupled through the MgO spacer and its IEC strength slightly decreases with decreasing temperature. At 300 K, a good fit is obtained with an IEC value of −4.6 × 10−5 J/m2 for the MgO spacer thickness of 0.8 nm and the fitting value compares well with the values reported by several groups21,22. Surprisingly, below TV, the sign of the IEC changes from negative to positive. At 50 K, the IEC from fitting the M (H) loops is around 1 × 10−4 J/m2 which is close to the value calculated directly using the formula in Ref. 23. The IEC sign switching appears at the exact temperature of the Fe3O4 high-to-low conductance transition. A similar effect has been observed by applying the in-plane magnetic field along [110] direction (Fig. 2). Therefore, the change in the M (H) shape is due to the sign change of the IEC rather than any change in the in-plane magnetic anisotropy. To the best of our knowledge, our experimental results demonstrate for the first time that the IEC can be modulated by charge-orbital ordering of the coupled ferromagnetic electrodes. We would like to stress here the positive ferromagnetic (FM) IEC below TV is not due to the orange peel coupling24. Firstly, orange peel coupling is due to a rough interface while sharp interfaces were verified with RHEED and HAADF STEM characterization. Secondly, we observe a large negative IEC at room temperature while it is well established that the orange peel contribution has a positive sign. Thirdly, this sign change occurs very sharply at the Verwey transition temperature which is not expected for orange peel coupling.

Figure 2
figure 2

Temperature dependent magnetic hysteresis loops measured along [110] direction.

Magnetic hysteresis loops of Fe3O4/MgO/Fe3O4/NiO multilayers for different MgO thickness measured at different temperatures. The thickness of MgO spacers are (a) 0.8 nm and (b) 3 nm respectively. (c) and (d) are magnetic hysteresis loops of Fe3O4/MgO/Fe3O4/NiO multilayers for different MgO thickness measured at 300 K and 50 K. The in-plane magnetic field is applied along [110] direction.

Full size image

It is well-known that local Coulomb interactions (between Fe 3d) and electron-lattice coupling play crucial roles in the Fe3O4 Verwey transition both in the bulk and on the surface25,26,27. Recently, the charge ordering was found to be accompanied by structural distortion25 and was described well by introducing the Coulomb on-site interaction U and the Hunds exchange J between B-site Fe 3d states using the LDA + U method26. The Vienna Ab Initio Simulation Package (VASP) implementation of the LDA + U method was used by Lodziana to investigate both the effects of local Coulomb correlation and electron-lattice coupling on the electronic states of the Fe3O4 (001) surface27. In both cases of bulk and surface Fe3O4, model yields are consistent with experiment results. To summarize, above TV basic electronic, magnetic and structural properties of Fe3O4 can be well explained by DFT-LDA while properties observed below TV can be described within the LDA + U method with an effective Coulomb potential U. Below we use this model to gain an understanding of the role of charge-orbital ordering on IEC to interpret our observations. Fig. 3a shows the positions of some atoms of the unit cell used in the calculations, the atomic structure (Fig. 3b) was obtained through self-consistent total energy minimization for electronic and ionic degrees of freedom using the LDA + U method. The charge ordered pattern obtained from the calculations is similar to that proposed by Attfield et al.28 and is also consistent with direct experimental observation29. The calculated charges (q) and magnetic moments (M) for B-site Fe atoms (the positions are marked in Fig. 3a) below TV and above TV are summarized in Table 1. From Table 1, one can clearly see that charge and orbital ordering indeed emerge self-consistently within the model used. The 3d orbital average δq and average δM between Fe3+ and Fe2+ cations are 0.191 e and 0.585 μB respectively, which are consistent with earlier reports26. The total charge difference between Fe2+ and Fe3+ ions (δq in Table 1) is mainly due to the t2g d-orbital. In Fig. 3c we show the calculated IEC J1 as a function of vacuum spacer thickness of Fe3O4/vacuum/Fe3O4 junctions below TV (U = 4.5 eV and U = 5.0 eV) and above TV. The vacuum spacer is chosen to represent a generic tunneling barrier. One can clearly see from Fig. 3c that the charge-orbital ordering changes the strength and even the sign of the long-range IEC. For Fe3O4/vacuum/Fe3O4 junctions with a 4.2 nm thick vacuum spacer, the IEC is positive at T > Tv and changes to negative at T < TV for U = 4.5 eV. In Fig. 3b one can see that the charge and orbital ordering in Fe3O4 leads to a complex lattice distortion. The Fe-O bonds formed by surface oxygen atoms with B-site Fe atoms (marked as green in Figs. 3a and 3b) have been dramatically modified due to charge-orbital ordering. The apical O atoms move toward the Fe atoms and result in Jahn-Teller deformations on the (001) surface below TV (Fig. 3d). To further demonstrate the origin of the charge orbital ordering driven IEC sign change, we plotted in Fig. 4 the density of state (DOS) of the Fe3O4/vacuum (0.42 nm)/Fe3O4 junction and the DOS of its surface oxygen atoms at T > TV and T < TV for U = 4.5 eV. One can see that for junctions with a 0.42 nm vacuum spacer at T > TV, when the magnetization of two Fe3O4 layers changes from anti-parallel to parallel configuration, the total DOS of the junction and its surface oxygen shifts to the low energy state indicating a FM IEC. In contrast, at T < TV for U = 4.5 eV, when the magnetization of two Fe3O4 layers changes from anti-parallel to parallel configuration, the total DOS of the junction and its surface oxygen shifts to the high energy state which means a AF IEC. We also plotted the spin density distribution in Fe3O4/vacuum (1.68 nm)/Fe3O4 junctions for cases above TV and below TV in Figs. 4e and 4f respectively30. One can clearly see from Figs. 4e and 4f that the changes in the electronic states at the interface modify the propagating wave functions in the barrier. Therefore, the origin of the charge ordering driven IEC sigh change can be attributed to that the charge-orbital ordering modifies the surface states, particularly the spin polarization of surface oxygen atoms. The changes in the electronic states at the interface further result in a significantly large phase shift of the propagating wave functions in the barrier, which defines the sign and strength of the IEC (See Fig. 3c and the following Fig. 5c).

Table 1 Calculated charges (q) and magnetic moments (M) for B-site Fe atoms. Total and 3d orbital charges (q) and magnetic moments (M) for B-site Fe atoms (the positions are marked in Fig. 2a) calculated within the model prohibiting (T > TV) and allowing (T < TV) charge-orbital ordering. The vacuum thickness is 1.68 nm and U is set to be 4.5 eV
Full size table
Figure 3
figure 3

Calculated atomic structure.

Atomic structure and ionic charge distributions at (a) T > TV and (b) T < TV. Arrows [001] in Figs. (a) and (b) indicate the same direction perpendicular to the surface. (c) Calculated IEC J1 as a function of vacuum spacer thickness of Fe3O4/vacuum/Fe3O4 at T > Tv (black line) and T < TV for U = 4.5 eV (red line) and U = 5 eV (green line). (d) Atomic structure of (001) surface at T < TV. The blue arrows indicate the apical O atoms move toward Fe atoms.

Full size image
Figure 4
figure 4

Calculated density of state and electron charge density distributions.

DOS of the Fe3O4/vacuum (0.42 nm)/Fe3O4 junction at T > TV (a) and T < TV for U = 4.5 eV (c) and the DOS of its surface atoms at T > TV (b) and T < TV for U = 4.5 eV (d). Calculated electron charge density distributions for spin up (blue) and spin down states (yellow) (e) without and (f) with CO. The spin density distributions are plotted for the sample with a vacuum thickness of 1.68 nm and the net magnetization of two Fe3O4 layers are in the antiparallel configuration.

Full size image
Figure 5
figure 5

Conceptual illustration of MR element utilizing effect discovered in this study.

Mutual orientation of two Fe3O4 layers in magnetizations (a) parallel and (b) anti-parallel states controlled by the sign of IEC. This concept is further illustrated with calculated (c) IEC J1 and (d) logic output for generic Fe3O4/vacuum/Fe3O4 junctions as a function of screened Coulomb interaction U.

Full size image

Discussion

The IEC sign change with temperature has been reported in metallic tri-layer systems31 and the mechanism behind it is associated with a slow functional dependence of the Ruderman-Kittel-Kasuya-Yosida (RKKY) IEC on temperature due to the Fermi distribution factor32. For Fe3O4, the IEC oscillating with spacer thickness has been reported in TiN/Fe3O4 superlattices and it is attributed to the quantum confinement of the spin polarized electrons33. However, those mechanisms are entirely unsuitable for manipulating the magnetization state of spin valves at the nano-scale. In contrast, we report an abrupt change of IEC at the Verwey transition due to charge-orbital ordering of the magnetic electrodes. The mechanism put forward in this work is promising for voltage control of spin-valve magnetization state at the nano-scale since the charge-orbital ordering can be controlled by an electric field34. To highlight this, Figs. 5a and 5b show a schematic MR element where two Fe3O4 layers are in magnetization parallel or anti-parallel states depending on the sign of the IEC. In the proposed MR element, the sign and strength of the IEC can be controlled by the top gate voltage (VG). For a FM IEC, without an external field, the magnetization of the two FM layers prefers a parallel configuration (Fig. 5a) and the resistance is relatively low (logic output 0). A spin valve with an AF IEC has high resistance (logic output 1) as the two FM layers are in an antiparallel magnetization state (Fig. 5b). Figs. 5c and 5d show the calculated IEC and logic output as a function of the screened Coulomb interaction U for Fe3O4/vacuum/Fe3O4 junctions with a 4.2 nm thick vacuum spacer. This dependence represents changes in Coulomb correlations due to charge-orbital ordering tuned by VG. One can see that the IEC oscillates as a function of U. For U = 4.3 eV, the IEC is positive and the logic output is 0. Increasing U to 4.5 eV, the sign of the IEC changes from positive to negative and the corresponding logic output changes from 0 to 1. Further increasing U to 5.1 eV, the IEC becomes positive again.

In summary, we have experimentally and theoretically demonstrated that the magnetic exchange coupling of all oxide thin film layered stacks of (001) NiO/Fe3O4/MgO/Fe3O4 can be manipulated by charge-orbital ordering in magnetic electrodes. We showed that the state of the charge orbital ordering in magnetite defines the strength and even the very sign of the IEC. The IEC sign switching appears at the exact temperature of the Fe3O4 high-to-low conductance transition which changes the magnetization states of the spin valve from an anti-parallel (parallel) to a parallel (anti-parallel) state. This finding opens new possible ways for external stimuli (electrical field) control of magnetization states of spin valves via direct IEC manipulation by charge-orbital ordering. The mechanism of direct manipulation IEC should be comparable with the widely reported current driven magnetization switch and theoretically suggested IEC change due to the switching of ferroelectric polarization.

Methods

Sample preparation

The Fe3O4/MgO/Fe3O4/NiO/MgO (100) stacks were grown on the same MgO (001) single crystal substrates using a shadow mask inside an oxygen plasma assisted molecular beam epitaxy system. The substrates were chemically cleaned prior to their insertion into the growth chamber and then cleaned in situ at 600°C in UHV for 1/2 hour followed by annealing in 1 × 10−5 Torr oxygen for 2 hours duration. The NiO epilayer (15 nm) was deposited at a substrate temperature of 600°C with an oxygen partial pressure 5 × 10−6 Torr. After the NiO growth, the sample was cooled to 250°C and a 10 nm Fe3O4 film was grown which serves as the bottom FM layer18,19. Then a single crystalline MgO (001) was formed at 250°C with a rather low deposition rate 0.05 Å/s and annealed at 250°C for 20 min in a vacuum of 5 × 10−9 Torr. The thickness of MgO spacer is controlled by placing the shadow mask at different positions. Subsequently, a second (top) Fe3O4 layer was deposited using the same growth conditions as those for the first Fe3O4 layer.

Magnetic and transport measurements

The magnetic and transport properties of Fe3O4/MgO/Fe3O4/NiO spin valves at different temperatures were examined by means of a Quantum Design Physical Properties Measurement System (14 T, Quantum Design). The minor loops are taken after a positive saturation of the sample.

Micromagnetic simulation

The micromagnetic calculations were performed with OOMMF code20. The system we considered consists of two magnetic thin films with the same thickness but different coercive fields (Hc1 and Hc2).

First-principles calculation

For the LDA + U simulations in this work, we use J = 0.89 eV and U was varied between 4.3 eV and 5.5 eV26. We employed the projected augmented plane wave method (PAW)35, implemented within VASP36. The valence configuration 3d74s1 for Fe and 2s22p4 for O were used. The generalized gradient approximation37 is used for the exchange-correlation energy. A monkhorst-pack 6 × 6 × 2 k-point grid was applied to fully optimize the atomic configurations. To model the IEC above TV, we consider Fe3O4 without lattice distortion in accordance with the measured lattice structure as shown in Fig. 2a.