Generation of localized strain in a thin film piezoelectric to control individual magnetoelectric heterostructures

Recent developments in the controlled movement of domain walls in magnetic nanowires by short pulses of spin-polarized current give promise of a nonvolatile memory device with the high performance and reliability of conventional solid-state memory but at the low cost of conventional magnetic disk drive storage. The racetrack memory described in this review comprises an array of magnetic nanowires arranged horizontally or vertically on a silicon chip. Individual spintronic reading and writing nanodevices are used to modify or read a train of approximately 10 to 100 domain walls, which store a series of data bits in each nanowire. This racetrack memory is an example of the move toward innately three-dimensional microelectronic devices.

Magnetoresistive random access memory (MRAM), because of its moderately fast access time and almost unlimited endurance, has offered a tantalizing application potential as next-generation non-volatile integrated memories1. However, mass production of MRAMs has long been hampered by their high writing energy from Ampère-current-induced magnetic field2 3. This would also cause severe cross-talk among neighbouring cells when miniaturizing the device size for higher storage capacity. One promising solution is to develop voltage-write MRAM devices4, that is, manipulating magnetization directly using electric voltage rather than current. Previous proposals include integrating a spin valve5 or magnetic tunnel junction (MTJ)6 unit onto a multiferroic layer (for example, Cr2O3 (ref. 7) or BiFeO3 (ref. 8)), and more recently using ferroelectric (FE) oxide as the tunnel barrier as in a Fe/BaTiO3/(La,Sr)MnO3 junction9 where significant voltage control of tunnel magnetoresistance6 was realized at rather low temperatures. Nonetheless, no existing proposals have so far achieved low voltage, high storage density and room temperature operation all at the same time. In this study, we illustrate a simple and new approach towards ultralow voltage-controlled MRAMs using phase-field simulations. We show that the proposed device is capable of achieving ultralow writing energy (less than 0.16 fJ per bit), room temperature (or even wider range) operation and high storage density (up to 88 Gb inch−2) all at the same time, associated with reliable thermal stability (quality factor over 40 at T=300 K) and comparable writing speed (sub-10 ns). Results Ultralow voltage-controlled MRAM device structure Figure 1a illustrates one possible layout of a memory cell array, which builds on planar complementary metal-oxide-semiconductor (CMOS) platforms. As in normal MRAMs, it has a typical 1-T(transistor)/1-magnetoresistive(MR) element cell architecture (Fig. 1b), where the readout is accomplished by sensing the resistance change of the cell. However, the most remarkable feature is its ideal zero current, gate-controlled-write process (see Methods section for the operation time sequence), which is fast and extremely power-efficient. There are virtually no cross-talks among neighbouring cells. It has a simple structure, which is easier to manufacture than conventional MRAMs, due to the absence of extra current-carrying lines. The simplicity offers a great potential for realizing high storage capacity. The basic building block, incorporating a spin valve or MTJ unit onto a (011)-oriented perovskite FE (for example, lead magnesium niobate-lead titanate, PMN-PT) layer (or maybe (001)-oriented rhombohedral FE layer with spontaneous polarization along 〈111〉 directions), is shown in Figure 1c. An external electric voltage is perpendicularly applied to the PMN-PT layer triggered by the connecting transistor. This produces large anisotropic in-plane piezostrains10, which would be further imparted to its upper magnetic-free layer and induces a 90° in-plane magnetization rotation in the latter via the strain-mediated converse magnetoelectric coupling11 12. By operating the (011)-oriented PMN-PT layer below its FE coercive field, bistable in-plane piezostrains can be obtained, that is, the strain states can be maintained even when the applied voltages are switched off, as demonstrated by a permanent magnetization switching observed recently in polycrystalline Ni thin film grown on (011) PMN-PT13. This interesting bistable magnetization switching should bring greater design flexibility to the present voltage-controlled MRAM devices, that is, the non-volatility can be achieved not only by virtue of high-quality epitaxial magnetic thin films whereby the magnetocrystalline anisotropy serves as the potential barrier3, but also in polycrystalline and/or amorphous thin films, despite a little lower magnetoresistance ratio due to structural imperfections in the latter. Compared with previous MRAM proposals7 8 9, the present voltage-write MRAM device is actuated at an ultralow voltage, as discussed below. Second, the strain-mediated magnetoelectric coupling offers a straightforward way to electrically manipulate the magnetization at room temperature, whereas the promising Fe/BaTiO3/(La,Sr)MnO3 junction9 is limited to low temperature. Moreover, our proposed voltage-write MRAM device has a much wider operation temperature range than the voltage-controlled MRAM that uses Cr2O3 with its Néel temperature (T N~307 K) close to room temperature7. Third, the perpendicular writing voltage bias (Fig. 1c) makes it easy to integrate it with existing CMOS platforms (Fig. 1a) to fabricate nanoscale high-density memory devices. In contrast, multiferroic BiFeO3-based MRAMs demonstrated so far8 require an in-plane actuation electric field. Such a transverse actuation voltage along the in-plane direction usually resorts to an interdigitized transducer patterning, which is typically of a micron size14, and intrinsically prohibits high storage capacity. This is also the case for the voltage-controlled MRAM using (001)-oriented FE layer15 16 17. Working and design principles of the device For illustration, we perform phase-field simulations of this electric-field-controlled MRAM device by choosing a (001) Ni film as the free layer. A spin valve comprising a synthetic antiferromagnetic layer (Fig. 1c) is considered as the upper MR element while an MTJ is also available. An initial uniform [010] magnetization is assumed for the Ni-free layer, which is exchange coupled to its upper reference layer. This direction can be set by annealing the device in a magnetic field. As described above, a perpendicular electric field E is applied to the bottom (011) PMN-PT layer to generate the anisotropic in-plane piezostrains13, subsequently inducing the in-plane magnetization rotation. By solving the Landau–Lifshitz—Gilbert (LLG) equation, both the macroscopic magnetization states and microscopic magnetic domain structure can be obtained (Methods section). We also performed a preliminary simulation to study the electric-field-induced magnetization switching in Ni/PMN-PT bilayer instead of the whole device. The simulation results agree well with corresponding experimental results13 (Supplementary Fig. S1 and Supplementary Note 1), demonstrating that this micromagnetic phase-field approach is valid. By further introducing the dipole coupling between the Ni-free layer and its upper reference layer (equation (4) in Methods section), the thickness-dependent E-induced magnetization switching in the Ni-free layers are plotted in Figure 2a. As shown, bistable magnetization states are exhibited at E=0, for example, m x =0.9538 and 0.0364 for the 64×64×35 nm3 Ni-free layer, corresponding to a magnetization rotation of about 70°. This is also directly illustrated by their magnetic vector diagrams shown in Figure 2b. Moreover, the magnetization–E curves become gradually flatter with the film thickness decreasing from 35 to 5 nm, along with a small increase in the maximum magnetization rotation around 5°. This can be attributed to the suppression of the out-of-plane magnetization component due to enhanced demagnetization in thinner magnetic thin films, which would in turn facilitate the in-plane magnetization rotation. The larger magnetization rotation further leads to a broader resistance change of the spin valve, as the device resistance is determined by the relative magnetization alignment, analytically expressed as15, 1 Here R 0 denotes the device resistance with parallel magnetization configuration in the reference and free layer, and R MR the maximum magnetoresistance ratio, that is, the relative device resistance change between the parallel and antiparallel magnetization alignments. For illustration, R MR is taken as 7% for a normal spin valve18 with giant magnetoresistance effect. Corresponding to the greater magnetization switching, increased maximum resistance changes are exhibited as indicated in Figure 2c. This may suggest the first design principle of such voltage-write MRAM device: the thinner, the better. However, growing ultrathin films requires more sophisticated processing control. Besides, the interface magnetic anisotropy contribution may lead to additional effect on the magnetization switching19, especially in ultrathin (several monolayers) films20, but is not considered herein regarding the robust magnetoelastic coupling of Ni as well as the large piezoelectric response of PMN-PT, which should in principle produce a large strain-mediated magnetoelectric response19. Nonetheless, based on the results in Figure 2, 5 nm could be a well-compromised thickness for metallic Ni-free layer deposited on the PMN-PT. In particular, the actuation electric field required for the MRAM device is ultralow, that is, |E| 10 V μm−1 (ref. 24)), is applied only on the PMN-PT layer. This structural feature of the present device is different from that in STT-RAMs, where the write and read current flow across the whole device and share the same tunnel. Writing speed and storage capacity Another important performance attribute of the voltage-controlled MRAM device is the writing speed, which typically takes less than 10 nanoseconds (ns) or with a possibility of sub-1 ns (see the write-time estimation and corresponding electric-field-induced dynamic magnetization switching in the Methods section). This operation speed is significantly higher than conventional magnetic-write MRAM (around 20 ns (ref. 1)), while comparable with the STT-MRAM devices (~3 to 10 ns (ref. 25)). Finally, the fully gate voltage-controlled operation allows a good compatibility of the present device with current CMOS platforms, opening up possibilities for ultrahigh densities. For instance, assuming a typical channel length of 45 nm for the bottom transistor (Fig. 1a) and the lateral size of 64 nm for the upper ME spin valve (or MTJ)/PMN-PT heterostructure, a storage density of around 88 Gb inch−2 can in principle be realized, challenging traditional NAND Flash memory on mass data storage applications26. For comparison, the properties obtained in the present device prototype are summarized in Table 1, including several other existing room temperature non-volatile memory technologies. Discussion We have demonstrated a simple and new approach towards ultralow voltage-controlled MRAMs. The present voltage-controlled MRAM device can be fabricated on CMOS platforms, wherein the basic building block comprises of a spin valve or MTJ MR element grown on a (011) PMN-PT layer. Without necessarily requiring epitaxial growth of the magnetically free layer, simply utilizing a specifically engineered FE layer with voltage-induced bistable anisotropic in-plane piezostrains, one can realize most important attributes required by modern MRAM technology at the same time. Therefore, we expect our finding will stimulate future experimental and engineering efforts on developing such an electric-voltage-controlled device, which has superiorities over the previous device paradigms7 8 9 14 22. The present device proposal represents a simple pathway to realizing high-density, low voltage-controlled MRAM applications at room temperature, either as standalone/embedded memories or logic circuits. Methods Operation time sequence of the device As seen in Figure 5a, the write voltage bias is applied between the Bitline (BL2) and the plateline (PL), which is controlled by the Wordline (WL) connecting to the gate of transistor. Corresponding writing time sequence is shown in Figure 5b. While to read, the sensing current (I sen) travels through the transistor and the MR element from BL2 to BL1, to generate the high and low output voltages, that is, 2 where R MOS is the resistance of the transistor; R high and R low denote the bistable resistance states of the MR element resulting from the voltage-induced magnetization switching in the magnetic-free layer. The reference current (I ref) is set to a specific value so that the corresponding reference voltage (V R) is ideally between V high and V low. Introduction to the phase-field method In the phase-field approach, the magnetic domain structure is described by the spatial distribution of the local magnetization vectors M=M s m=M s (m x , m y , m z ), where M s and m i (i=x, y, z) represent the saturation magnetization and the direction cosine, respectively. The temporal evolution of the magnetization configuration and thus the domain structure is governed by the LLG equation, that is, 3 where γ 0 and α are the gyromagnetic ratio and the damping constant, respectively, and Heff the effective magnetic field, given as Here μ 0 denotes the vacuum permeability, and F tot is the total free energy of the magnetic-free layer, 4 where F mc, F ms, F ex and F elas are the magnetocrystalline anisotropy, magnetostatic, magnetic exchange and elastic energy, respectively. F dip describes the ferromagnetic-type dipole coupling between the free layer and its upper reference layer, given as for an initial [010] magnetization; H dip is the effective interlayer dipole coupling field taken as 40 Oe for illustration (see more discussions on the influence of H dip in Supplementary Fig. S2 and Supplementary Note 2). The magnetostatic energy F ms can be written as, 5 where Hd denotes the stray field arising from the long-range magnetization interactions in the system, given as being the average magnetization field). To incorporate the influence of the specific sample geometric size, a finite-size magnetostatic boundary condition is used for obtaining F ms, with the demagnetization factors N in the magnetostatic energy calculated numerically27. The mathematical expressions of F mc and F ex for cubic ferromagnets are the same as those given in refs 11 and 27. The elastic energy F elas can be written as, 6 where e ij is elastic strain, c ijkl the elastic stiffness tensor and ɛ ij the total strain that can be represented as the sum of homogenous and heterogeneous strains following Khachaturyan's mesoscopic elastic theory28, 7 The heterogeneous strain η ij is defined in such a way so that , and the homogeneous strain describes the macroscopic shape change of the magnetic thin film. Specifically, and , namely, the homogeneous strains along the two in-plane crystal axes, can be expressed as, 8 Here and denote the contribution from the piezoelectric transformation of the bottom PMN-PT layer, which we directly use the experimental data reported in ref.13 as the input. We also assume that the PMN-PT layer is much thicker than the upper magnetic film to ensure a full strain transfer29. ɛ 0 is the biaxial residual strain resulting from the lattice and/or thermal mismatch. For a polycrystalline magnetic film deposited at room temperature, ɛ 0 almost equals zero; while for an epitaxial film, it depends on the specific substrate as well as the growth condition30. Note that ɛ 0 should exert no influence on the in-plane magnetization rotation (Supplementary Fig. S3 and Supplementary Note 3), although it does have effects on the specific magnetic domain structure in the free layer31. Here it is taken as −0.5% for illustration. represents the stress-free strain describing the stress-free deformation of a cubic ferromagnet associated with the local magnetization change, 9 where λ 100 and λ 111 are the magnetostrictive constants. By further considering a mixed boundary condition for the film-substrate system32, F elas can be calculated by combining Khachaturyan's theory28 with the Stroh's formalism of anisotropic elasticity33. The temporal evolution of the local magnetization and thus the domain structures are obtained by numerically solving the LLG equation using the semi-implicit Fourier spectral method34. To illustrate the Ni-free layer with a size of 64×64×35 nm3, the magnetic film is discretized into a three-dimensional array of cubic cells of 64Δx×64Δy×35Δz, with the grid size Δx=Δy=Δz=1 nm in real space accordingly. A three-dimensional finite-size boundary condition is considered to calculate the geometric-size-related magnetostatic anisotropy energy as mentioned above. By varying the simulation zone size and/or the grid size, magnetic-free layers with different geometric sizes (for example, thickness and lateral size, see Figs 2 and 3) can then be treated. Each simulation starts with an initial [010] magnetization distribution accordingly and proceeds for long enough time to ensure a stabilized magnetization distribution, with a normalized time step Δτ=0.01. The Landau, elastic and magnetic parameters used for simulation are described in the literature11 35. Write-time estimation Such electric-field-controlled MRAM device can at least allow a sub-10 ns writing, which is much faster than conventional magnetic-field-switched MRAM (about 20 ns (ref. 1)) and comparable with practical STT-MRAM devices (about 3–10 ns (ref. 25). First, based on LLG equation, the average magnetization switching time is estimated as about 0.95 ns for magnetic-free layers with various lateral sizes (the thickness is set as 5 nm likewise), typically allows a GHz operation frequency. The details about such electric-field-induced dynamic switching will be shown below. The dynamic parameters used in LLG equation are: Damping constant: α=0.038 (experimental measurement for 5 nm Ni films36). Gyromagnetic ratio: γ 0 =−2.42×105 m (A s)−1 (using g=2.21 from ref. 36). However, there are other factors that can affect practical writing speed such as the time the PMN-PT layer produces the piezostrain (that is, the electric polarization switching time with the associated strain), and the time for the strain transfer across the Ni/PMN-PT interface (we tentatively do not consider the influence of buffer layer that might be required for epitaxial growth) and getting its mechanical equilibrium. For the first part, note that polarization normally switches much faster than the magnetization (the switching time is usually below 10 ns (ref. 37). That is also the reason why a FeRAM takes less time to write information compared with common MRAMs (see Table 1). Moreover, it is worth noting that there is no full polarization reversal in the PMN-PT layer in our case, as the actuation electric field (−0.13 to 0.1 V μm−1) remains below the FE coercive field (approximately ±0.2 V μm−1 (ref. 13)), which further reduces the switching time. Also for a given polarization distribution, one can assume that the FE layer undergoes piezoelectric deformation and reaches its mechanical equilibrium almost instantaneously, as the mechanical relaxation of an elastic field is even faster than that of a polarization38. Moreover, it should also be noted that although higher actuation electric field bias could lead to faster electric polarization switching in the FE layer to some extent, the overall device switching speed would not be significantly enhanced due to the inherent piezostrain-induced magnetization switching. Specifically, the maximal remnant piezostrain at zero electric field (E=0) almost remains stable when the write E-field is increased from 0.13 to 0.2 V μm−1 (ref. 13). This feature is quite different from normal STT-RAM devices where faster switching could be facilitated by imposing higher operating voltage, while unavoidably exhibits higher vulnerability to dielectric break down as well. As for the pizeostrain transfer time τ, we give a rough estimation on this short time span39, 10 where d 1 and d 2 are the thicknesses; and v 1 and v 2 the velocity of sound in the FM and FE layer, respectively. For illustration, assuming d 1=5 nm (the Ni film thickness), d 2=2 μm for the PMN-PT layer and v 1≈v 2=3,000 m s−1 (it is well known that the speed of sound varies from substance to substance, but generally it travels very fast in solids, say, about 5,120 m s−1 in iron at 293 K), then we have τ≈0.67 ns. Note that we have considered the stress wave that covered the whole device system, but the effective mechanical coupling is more likely to be confined in a closer vicinity of the interface, which means the strain transfer time could be even shorter. Overall, although the practical strain mediation situation could be far more complex than the simple model described above (such as the influences of the mechanical resonance frequency, and the Resistor-Capacitor time constant associated with charging the FE capacitor, see related discussion in ref. 40), it is rational to consider the stress to be applied almost instantaneously to the FM film40. Thus in total, the theoretical writing time of such electric-field-controlled MRAM device can be estimated to be about 1.62 (=0.95+0.67) ns, well within a 10 ns range required by future applicable MRAM technologies41. This result is comparable with the results reported for other related device proposals with similar strain-mediated electric-write procedure, such as multiferroic logic40 42 or nanowire memory43 devices, which use either Terfenol-D40 (Ni (ref. 42)) particles, or FeGa (ref. 43) nanowires attached to a commercial Pb(Zr,Ti)O3 layer, all with an estimated operation time less than 10 ns. Electric-field-induced dynamic magnetization switching The combination of phase-field model with micromagnetic simulation here in this paper allows us to go beyond the previous macro-spin model used for single-domain magnets40 42. Indeed, the free layer does not always behave as a single macro-spin when the lateral sizes exceed a certain critical value (possibly around 128 nm, see Fig. 3b). For illustration, Figure 6a shows the time-dependent electric-field-induced magnetization switching processes in 64×64×5 and 224×224×5 nm3 Ni-free layers, where the films are in single-domain and multi-domain states, respectively. As seen, the dynamic magnetization switching process exhibit a typical processional rotation feature (see the fluctuations of m x in Fig. 6a after t=0.77 ns, governed by LLG equation) in single-domain 64×64×5 nm3 Ni films. By contrast, the 224×224×5 nm3 Ni films present a much flatter magnetization switching by virtue of domain-wall motion (see the illustrations in Fig. 6b), which is common in multi-domain magnetic thin films44. Note that the normalized magnetization m x does not eventually turn to unity due to the cancelling out of the magnetic vectors along the [100] direction. Author contributions J.M.H. conceived and performed the simulations. J.M.H., Z.L. and C.W.N. designed the device structure. C.W.N. and L.Q.C. directed the work. J.M.H., L.Q.C. and C.W.N. co-wrote the paper. All contributed to the analysis. Additional information How to cite this article: Hu, J.-M. et al. High-density magnetoresistive random access memory operating at ultralow voltage at room temperature. Nat. Commun. 2:553 doi: 10.1038/ncomms1564 (2011). Supplementary Material Supplementary Information Supplementary Figures S1-S3, Supplementary Notes 1-3 and Supplementary References