Recent developments and the future perspectives in magnetoelectric nanocomposites for memory applications

Microelectronics and high-density data storage are the two main driving forces for the development of modern information and communication technologies (ICT).¹ Both follow Moore’s law, and in the era of big data, they both bring significant energy problems.² At present, most data storage technologies are based on current read–write technologies. Even next-generation magnetic storage technology, i.e., spin transfer torque magnetic random-access memory (STT-MRAM) and spin–orbit torque magnetic random-access memory (SOT-MRAM), still depend on the passage of an electric current,³ which leads to Joule heating, and thereby increases power consumption. To further reduce the power consumption of such devices, it is important to strongly reduce or even stop the passage of a current across the device for yielding magnetization switching and instead use an electric field to modulate the magnetism. For a long while, magnetoelectrics have been proposed to achieve this effect. Here, the electric field applied across an insulating multiferroic (where ferroelectricity and ferromagnetism coexist) leads to a reversal of magnetism (the converse magnetoelectric effect, CME), giving the possibility for magnetoelectric random-access memory (MeRAM).⁴ The power consumption in such a device can be reduced by two orders of magnitude.⁵ 

A key problem for a simple magnetoelectric device is that single phase materials have some seemingly insurmountable problems, namely, low ferromagnetic (FM)/ferroelectric ordering temperatures and weak magnetoelectric coupling.⁶ One way to circumvent this problem is to create artificial magnetoelectrics, where a composite of two different materials is used. The materials are coupled together either by strain coupling (magnetostrictive and ferroelectric)^7,8 or exchange bias coupling [ferromagnetic and antiferromagnetic (AFM) ferroelectric]^9,10 or by charge gating.^11,12 Among them, the strain coupling has a much larger active length scale of the magnetic phase and has attracted the most attention.¹³ Different types of composites, including 2-2, 3-0, and 3-1, have been created to produce the magnetoelectric effect.¹⁴ 2-2 is a multilayer film structure, 3-0 is a nanoparticle in matrix structure, and 3-1 is a nanocolumn in matrix structure. Figure 1 schematically shows the mechanism of the magnetoelectric effect in strain-coupled composites and the current materials (2-2 and 3-1 composites) and device challenges and solutions toward practical applications. More details on the mechanisms involved in panel 3 for achieving a strong self-biased CME at room temperature are shown in Fig. 2 and are discussed below.

For the materials to have both effective coupling and be applicable for memory devices, all composite forms need to be grown in thin film form. Magnetic metals do not need to be grown epitaxially owing to their inherently high crystalline quality, ability to give strong functionality even when amorphous, and also their effective property coupling across interfaces, but they still need to be grown on piezoelectric substrates to yield a large ME effect. An example metallic system with good direct magnetoelectric effect (DME) coupling is Ta/Cu/Mn₇₀Ir₃₀/Fe₅₀Co₅₀ grown on piezoelectric AlN.¹⁵ This system is very promising for sensor or energy harvester applications but not for memory applications, considering that a large voltage is required to control the magnetism (i.e., CME), because of the large thickness of the piezoelectric substrate.^8,16 On the other hand, in 2-2 systems which use non-FE substrates [Fig. 1(b-i)], the lateral strain is largely clamped by the underlying substrate, and this strongly reduces the ME coupling. The 3-0 system is also problematic in that it is has isotropic growth of the 0-dimensional particles in the matrix in an uncontrollable manner. Moreover, the isotropic shape of the particles prevents the shape anisotropy from being present, which means that perpendicular magnetic anisotropy (PMA) is not preferred. This leaves the 3-1 composite as the only promising magnetoelectric system with the potential for spintronic applications. 3-1 systems are in the form of vertically aligned nanocomposite (VAN) films with, in theory, a strong magnetoelectric coupling coefficient, which is not limited by clamping from the substrate.¹⁷ 

There are scores of articles on 3-1 systems. It is interesting that there has been a relatively limited set of materials explored (most have centered on BaTiO₃,^18,19 BiFeO₃,^20–24 PbTiO₃,²⁵ and PZT²⁶ as the ferroelectric material)²⁷ and CoFe₂O₄,^{18,20,21,28,29} NiFe₂O₄,^19,30 and LSMO³¹ as the magnetostrictive materials. Moreover, most work has been shown for the direct magnetoelectric effect (DME) (magnetic field control of polarization).^29,32 We note that the DME is possible even when there is moderate leakage in the film, whereas for the CME, Joule heating from current leakage prevents sufficient voltage from being applied to change the magnetic anisotropy.³³ In oxide films, particularly of the 3-1 type, leakage is a very common problem. This problem has not been sufficiently well addressed, yet it is critical for achieving a sizable ME effect.

A key question is how to reduce leakage in 3-1 artificial magnetoelectric oxide films. With the aim of overcoming this problem, material level and structural level solutions have been tried, including doping of the FE materials, such as BiFeO₃, to make it less leaky^34,35 and designing quasi (0-3) systems to avoid the FM columns from contacting electrodes.³⁶ However, the in situ electric control of magnetism (CME effect) could still not be achieved in this system.

Recently, a 3-1 system showed strong improvement in performance (cf. earlier systems). Na_0.5Bi_0.5TiO₃ (NBT) was chosen as the ferroelectric material owing to its wide bandgap, which means less chance for the thermally generated electrons to jump to the conduction band and hence means lower leakage current. NBT was combined with the commonly studied CoFe₂O₄ (CFO) as the magnetostrictive material.³⁷ Hence, a 3-1 NBT–CFO film system was made. Leakage was lowered by a factor greater than 10² at the 750 kV/cm field, compared to the archetypal BFO–CFO system.³⁷ Furthermore, current blockage was ensured via a rectifying film–Nb:SrTiO₃ (Nb:STO) substrate interface. The vertical interfaces in the 3-1 film were also less leaky (by a factor of 10³) compared to the BFO–CFO system.³⁸ In the NBT–CFO system, for the first time, in situ electric field control of magnetic anisotropy was achieved at room temperature in a system applicable for MeRAM, showing a large magnetoelectric coefficient of 1.25 × 10⁻⁹ s m⁻¹, at an applied field of 2200 kV cm⁻¹.

Realizing the strong advantage of using NBT, in a follow-on work, a triple composite system, as depicted in Fig. 2(a), was created, where NBT was combined with an antiferromagnetic (AFM) material (NiO) as well with a soft magnetic Ni₂FeO₄ (NFO) material.³⁹ NFO is similar to CFO, albeit with a moderately lower magnetocrystalline anisotropy. The NFO and NiO phases form a core–shell structure, where NiO surrounds NFO and prevents NFO from contacting the Nb:STO substrate, as shown in Fig. 2(b). Hence, in this 3-1-1 system of Na_0.5Bi_0.5TiO₃–NiO–NiFe₂O₄, leakage was much further reduced by three orders compared to the 3-1 system Na_0.5Bi_0.5TiO₃–NiFe₂O₄, as shown in Fig. 2(c). The largely reduced leakage enabled a large room temperature magnetoelectric effect with a magnetoelectric coefficient of about 1.38 × 10⁻⁹ s m⁻¹. Moreover, with the exchange coupling between NFO and NiO, the system shows a self-biased magnetoelectric effect [Fig. 2(d)]. Self-biasing (i.e., the electric field change of the magnetism at zero applied magnetic field) was not shown before in any oxide 3-1 systems before and yet is critical for practical memory devices.

While the current research on oxide magnetoelectrics is still at the early stages of improving the basic materials properties, with the considerable performance promise shown by 3-1 composites, it is important to consider what could be the next stages toward achieving a practical device. In fact, some prerequisites for achieving practical devices based on 3-1 thin films have been realized in the past five years. This includes integration of the 3-1 nanocomposite films on silicon,^40–42 large-scale preparation of 3-1 films using magnetron sputtering,^43–45 robust perpendicular exchange bias at the vertical interfaces,^39,46,47 and long-range ordering of the magnetic pillars,^48–50 and the controllable lateral scaling of the magnetic pillars.⁵¹ 

As we show below, there are five main further challenges (and opportunities) relating to device structure and materials selection and control.

A magnetic tunnel junction (MTJ) is required for achieving MeRAM. In fact, not only are MTJs important device structures for achieving a practical MeRAM element, but also they have very important application prospects in magnetic sensing and neuromorphic computing.⁵² 

A prototype MeRAM device based on 3-1 nanocomposite films incorporating an MTJ is an important future direction. In Fig. 3, the structure of a MeRAM array based on two different 3-1 nanocomposite layers, created via self-assembled stacking, is schematically shown. In this device, the bottom 3-1 film has an AFM (matrix)–FM (column) structure and works as the reference layer, where the exchange bias can pin the magnetization of the magnetic column. The top 3-1 layer has the FE (matrix)–FM (column) structure, where the magnetization of the magnetic nanocolumns can be switched with the electric field through the magnetoelectric effect. The parallel and antiparallel alignment of the magnetization of the two layers will yield two different resistance states, i.e., “0” and “1” states.

In this device, the electric field and current are used in the write and read operations, respectively. This means that in the write operation, the leakage of the device is small (resistance is large), while in the read operation, a current can be easily applied (the resistance is small enough), which seems to be a contradiction. In fact, there is no contradiction owing to a rectification effect, which allows us to write and read only by changing the sign of the voltage, i.e., a write operation can be performed when a negative voltage (−V) is applied, and a read operation can be performed when a positive voltage (+V) is applied. Rectifying interfaces have been realized in ME composite systems such as NBT–CFO (3-1) and NBT–NiO–NFO (3-1-1) films via a p-n junction at the film–substrate interface.^37,39

A strain-mediated 180° switching of the magnetization is a challenge for the 3-1 MeRAM structure since the electric field alone cannot break the time-reversal symmetry. However, some strategies have been proposed to achieve 180° magnetization switching.^53,54 For example, it can be realized in the 2-2 structured films by the pulsed voltage induced precession of the magnetization and with an in-plane external magnetic field. This should also apply for the 3-1 MeRAM device. The magnetization switching process is as follows: In the MeRAM cell, the pillar is perpendicularly magnetized. The voltage pulse is applied to the cell to remove the magnetic anisotropy and induce the precession of the magnetic moments around an external magnetic field. If the voltage is turned off at one-half period of precession, the magnetization will be switched. The external magnetic field can be replaced by the exchange bias effect to achieve a field-free switching, which has been employed in spin–orbit torque (SOT) switching techniques.⁵⁵ 

Compared with MTJ devices based on metallic multilayer films, such as CoFeB/MgO/CoFeB,⁵⁶ which are currently used in MRAM, MTJ devices based on 3-1 oxide nanocomposites have the following potential advantages: First, the nano-pillars in this type of material are separated by insulating film matrices, and hence, each pillar is electrically isolated from others and can work as a data bit independently. Thus, micro-processing is not needed for the device fabrication as for conventional MTJs, potentially reducing the cost significantly. Second, the physical properties of the materials in the 3-1 films can be enhanced by three-dimensional strain and interface effects, e.g., ferroelectric Curie temperature,⁵⁷ or perpendicular magnetic anisotropy (PMA).¹⁸ Third, the size of magnetic nanopillars in the 3-1 films can be as low as < 10 nm,⁵⁸ and thus, the areal density can exceed 1 Tbit in.⁻², which is required by ultra-high-density data storage.

The magnetic material in MTJs needs to have high spin polarization, low resistivity, a small switching magnetic field simultaneously, and a high magnetic ordering temperature. Magnetic oxides, such as Fe₃O₄, NiFe₂O₄, and La_0.7Sr_0.3MnO₃, can meet the above requirements. Among them, Fe₃O₄ and La_0.7Sr_0.3MnO₃, and Sr₂FeMoO₆⁵⁹ and other double perovskites^60,61 are semi-metals with 100% theoretical spin polarization. Although it is difficult to prepare nanocomposites containing Fe₃O₄ experimentally, because the more oxidized phase Fe₂O₃ tends to be more stable than Fe₃O₄, SrTiO₃–Fe₃O₄ 3-1 films (Fig. 4) as well as BiFeO₃–Fe₃O₄ films have been grown under well-controlled conditions.⁴⁷ 

The tunneling layer can be chosen according to the crystalline structure of the magnetic materials: if the magnetic material is a spinel structure, NiMn₂O₄,⁶² CoCr₂O₄,⁶³ MnCr₂O₄,⁶⁴ and FeGa₂O₄⁶⁵ can be used as the tunneling layer; if the magnetic material is a perovskite structure, SrTiO₃⁶⁶ can serve as the tunneling layer. However, the rule for the choice is not absolute, that is, the magnetic material of the spinel can also work with a perovskite tunneling layer and vice versa. In addition, traditional tunneling materials, such as MgO and Al₂O₃, can also be used as the tunneling layer in the new “3-1” MTJs.

Perovskite–perovskite 3-1 nanocomposite films, such as BaTiO₃–La_0.7Sr_0.3MnO₃, tend to have flat film surfaces.³¹ This is because these materials have the same crystalline structures. However, perovskite–spinel composites, which are the most commonly studied 3-1 ME systems and which exhibit superior performance, have rougher surfaces. This represents a serious challenge for making a high-performance MTJ structure. Since the thickness of the barrier layer in MTJs is often only a few nanometers, it is particularly important to control the surface roughness of the bottom 3-1 film. However, the microstructure of those systems is determined by the surface and interface energy of the system.⁶⁷ This means that a different crystalline orientation of the phases can achieve a relatively smooth surface. As shown in Fig. 5, for the example of the Na_0.5Bi_0.5TiO₃–NiO–NiFe₂O₄ 3-1 system, for films grown on SrTiO₃(001) substrates, a considerably rough surface is achieved. On the other hand, films of the same composition grown on SrTiO₃(111) substrates are flatter by around an order of magnitude.³⁹ 

Traditional MTJs are usually fabricated by the top-down method from a continuous multilayer film, so there is no alignment problem of the upper and lower magnetic layers. However, MTJs based on “3-1” nanocomposite films (Fig. 3) are formed by self-assembly in a bottom-up manner, and the precise stacking of magnetic pillars is the key to the successful preparation of the device. Fortunately, studies on other 3-1 systems have shown that in 3-1 multilayer films, the nano-pillars in one layer tend to grow on top of nano-pillars in the layers below (Fig. 6).^68,69 This growth relationship can be further strengthened by appropriate materials selection. For example, perovskite–spinel combinations can be used for both the upper and lower 3-1 films. In addition, although likely not essentially, a way to assure excellent registry between layers is to employ a barrier layer formed of perovskite-spinel materials (both being insulating).

There are two sources of the crosstalk between adjacent magnetic columns in the 3-1 MTJ. The first source is the magnetostatic coupling. Ideally, we expect that the magnetization of the magnetic nano-pillars can be maintained after being written and that they can be switched independently without disturbing the other nano-pillars. However, the magnetostatic coupling between the magnetic nano-pillars makes this difficult. The magnetostatic coupling, which has so far been ignored,^58,70 can reduce the PMA, causing the anisotropy axis to move toward the in-plane direction, which further reduces the remnant magnetization (M_r) in the out-of-plane direction. In order to solve this problem, one can fabricate 3-1 films with well-designed pillar positions as mentioned above^48,50 so that they are well separated from one other. However, this will be at the expense of storage density. Another approach to solve the problem is to use antiferromagnets, where there is no magnetostatic coupling, since there is no stray field.⁷¹ Hence, the development of antiferromagnetic spintronic 3-1 devices represents a promising future direction.

Another source of the crosstalk is the long active length of the strain coupling. Although the strain is a long-range order parameter, interface-mediated strain coupling in oxide thin films is typically fully relaxed in less than one hundred nanometers of film thickness.^72,73 For example, it has been shown that strain induced by a PMN-PT substrate on a SrRuO₃ film grown on top of it has a working distance of 18 nm.⁷⁴ In this system, the active component (FE, PMN-PT substrate) comprises, by far, the largest volume fraction and the passive component (FM SrRuO₃ film) is the minor fraction. On the other hand, in the 3-1 system, the active component is only the matrix under the electrode, which takes a very small volume fraction compared to the passive component, which is the other part of the matrix that is not covered by the electrode, plus all the FM columns. Thus, in the 3-1 film based device, the active length will be much smaller than that in the FE (substrate)–FM (film) system, and hence, the crosstalk problem induced by the strain coupling should be minor. Furthermore, if there is any crosstalk problem, it can be reduced by adjusting the distances between the nano-pillars and the sizes of the electrodes and by optimizing the applied voltages.

In conclusion, 3-1 type magnetoelectric oxide composite films have considerable promise for low-power spintronic devices. They have been researched steadily for more than 15 years. Continuous research efforts have made remarkable progress in the basic physical understanding, and there has also been moderate progress toward practical applications. This Perspective summarizes recent material developments and serves as a reference for exploring future research directions in this field. It also highlights the challenges and opportunities for achieving prototype MeRAM devices based on magnetic tunnel junction (MTJ) structures formed from 3-1 systems. Finally, it proposes ways forward for achieving optimum MTJ performance through the correct materials selection and growth.

We acknowledge funding from the National Natural Science Foundation of China (NSFC Grant No. 12104052), the Leverhulme Trust (Grant No. RPG-2015-017), and the Royal Academy of Engineering (Grant No. CIET1819_24).

The authors declare no competing interests.

The data that support the findings of this study are available from the corresponding author upon reasonable request.