Abstract

We investigated spin transfer torque magnetization precession in a nanoscale pillar spin-valve under external magnetic fields using micromagnetic simulation. The phase diagram of the magnetization precession is calculated and categorized into four states according to their characteristics. Of the four states, the precessional state has two different modes: steady precession mode and substeady precession mode. The different modes originate from the dynamic balance between the spin transfer torque and the Gilbert damping torque. Furthermore, we reported the behavior of the temporal evolutions of magnetization components in steady precession mode at the condition of the applied magnetic field using the orbit projection method and explaining perfectly the magnetization components evolution behavior. In addition, a result of a nonuniform magnetization distribution is observed in the free layer due to the excitation of non-uniform mode.

1. Introduction

There has been intensive interest in spintronics for the past two decades [1]. An important example is giant magnetoresistance (GMR), proposed by Fert et al. [2] and Grünberg et al. [3], which has led to commercial products [4]. GMR effect has many potential applications such as magnetic sensor, magnetic read head, data memory and spin transistor, and so forth. Spin-valve [5], as a simple GMR device, also exhibits the switching effect. The merit of spin-valve is that the magnetization reversal can be realized under a small driving magnetic field. Besides being driven by the magnetic field, the magnetization may also be driven by a torque, spin transfer torque (STT), originated from the electron spins in the conducting current. The angular momentums may be transferred from the electrons of the spin-polarized current to the ferromagnet, giving rise to a switching of magnetization or stable precession of magnetizations [6] that leads to the generation of spin waves [7]. This mechanism of STT was initially proposed by Berger et al. [8] and Slonczewski [9] in 1996 and attracted significant interests because of the great potential for direct current-induced spintronic devices [1012]. The role of STT in magnetization switching has been verified by numerous experiments in spin-valve nanopillars [1315], magnetic nanowires [16, 17], point contact geometry [1820], and magnetic tunnel junctions [2124]. The most attractive application of current-induced magnetization switching is magnetic random-access memory (MRAM), which has the advantages of nonvolatile, high addressing speed, low-energy consumption, and avoidance of cross writing. Another important application is the current-induced spin transfer nanooscillators (STNO), in which the magnetization precesses, causing a spin wave in the microwave range. In previous researches, it was shown that the application of magnetic field affects the magnitudes of switching field and switching time [2528], opening new perspectives for information technology. Previous studies have shown that there are different magnetization types in the switching process depending on the magnetic field and the current [14]. Therefore, the understanding of magnetization precession and its mechanism under the external magnetic fields is very important for designing functional devices.

Many attempts have been made to increase the output power of STNOs; for example, an effective way to increase output power is magnetic-tunnel-junction- (MTJ-) based STNO with a larger linewidth [2932]. Compared with MTJ-STNO, there are several advantages that STNO with full metal GMR structure have, such as narrow oscillation linewidth, low device resistance, and good impedance matching. Therefore, it is very important to understand the magnetization precession state under the external magnetic field in the conventional 3d-FM-based GMR structure. The phase diagram of magnetization precession under in-plane external field has been studied by several authors [14, 26, 3335]. Kiselev et al. [36] showed that several different precessional modes exist, that is, small angle precession, large angle precession, and out-of-plane precession. Gmitra et al. [37] presented the phase diagram of magnetic field versus current density, illustrating that precessional state (PS) occurs at small current density and external magnetic fields. Although there have been a number of works on the phase diagram, the detailed analysis of the phase diagram is lacking. Xiao et al. [38] demonstrated that there were several precessional states in the magnetization reversal, that is, large-amplitude, in-plane, and out-of-plane precession from theory. In addition, they also presented the formation of phase diagram from the balance of STT and Gilbert damping torque. However, few published papers showed the detailed discussion of in-plane precession in the phase diagram.

In this paper, we present a systematic study on magnetization precession under spin-polarized currents and magnetic fields using micromagnetic simulation. The magnetization precession is grouped into several types according to their characteristics, which are unchanging state (UN), precessional state (PS), multidomain state (MS), and bistable state (BS). Among them, PS under magnetic field has two precession modes. Furthermore, the orbit projection method is applied to explain the magnetization component oscillation, and the snapshots of magnetization distribution show nonuniform magnetization precession. The magnetization precession phase diagram is helpful for understanding the magnetization precession mode and affecting the oscillation linewidth during the design of spin transfer nanooscillators.

2. Theoretical Model

The magnetization precession of a nanoscale pillar structure is investigated under the influence of spin-polarized currents and magnetic fields by micromagnetic simulations based on the Landau-Lifshitz-Gilbert (LLG) equation incorporating STT effect. The nanopillar structure is schematically shown in Figure 1. The thicknesses of each layer are Co (2 nm)/Cu (4 nm)/Co (10 nm) from top to bottom, and the size of the nanopillar is 16×64×16nm3. The two Co layers are separated by a thin Cu layer; the top Co layer is the free layer whose magnetization precession is triggered by a spin-polarized current. The bottom Co layer is the pinned layer with its magnetization vector P fixed in the direction along the positive x-axis. The initial magnetization vector M of the layer is along the negative or positive x-axis. The middle Cu layer is a space layer whose function is to avoid the exchange coupling between the two Co layers. The thickness of the spacer layer (4 nm) is much smaller than the spin diffusion length to conserve the spin momentum. The positive current is defined as the current flow from the free layer to the pinned layer.

The dynamics could be described by the temporal evolution of the magnetization M, and M obeys the generalized Landau-Lifshitz-Gilbert-Slonczewski (LLGS) equation: 𝑑𝐌𝑑𝑡=𝛾𝐌×𝐇e𝛼𝛾𝑀𝑠𝐌×𝐌×𝐇e2𝜇𝐵𝐽1+𝛼2𝑒𝑑𝑀3𝑠+𝑔(𝐌,𝐏)𝐌×(𝐌,𝐏)2𝜇𝐵𝛼𝐽1+𝛼2𝑒𝑑𝑀2𝑠𝑔(𝐌,𝐏)(𝐌,𝐏).(1)

The first term on the right-hand side of (1) is a conventional magnetic torque with gyromagnetic ratio γ, where 𝛾=𝛾/(1+𝛼2). This torque is driven by an effective field, 𝐇e1=𝜇0𝛿𝐸.𝛿𝐌(2)

The effective field of the LLGS includes the anisotropy, demagnetization, external, and the exchange fields, namely, 𝐇e=𝐇exch+𝐇ani+𝐇ms+𝐇ext. Taking account of exchange energy, anisotropy energy, magnetostatic energy, and Zeeman energy, the total energy E is given by 𝐸=𝐸ani+𝐸exch+𝐸ms+𝐸ext.(3) The anisotropy energy of Co in a Co/Cu/Co nanopillar is 𝐸ani=𝐾11𝑚2𝑠(𝑟)+𝐾21𝑚2𝑠(𝑟)2𝑑3𝑟,(4) where 𝐾1 and 𝐾2 are the anisotropy constants, 𝑚𝑠(𝑟) is the unit magnetization vector at each grid, 𝑚𝑠(𝑟)=𝑀𝑠(𝑟)/𝑀𝑠, and the easy axis is along the x-axis. Considering the fact that 𝐾1 is far greater than 𝐾2, we ignore the second term in the simulations.

The exchange energy is 𝐸exch=𝐴𝑀2𝑠||||𝑀(𝑟)2𝑑3𝑟,(5) where A is the exchange stiffness constant.

The magnetostatic energy can be presented as a sum of energies of interacting magnetic dipoles as follows: 𝐸ms=12𝑀𝑖𝛿(𝑟)𝑖𝑗||𝑟𝑟||33𝑟𝑖𝑟𝑖𝑟𝑗𝑟𝑗||𝑟𝑟||5×𝑀𝑗𝑑(𝑟)3𝑟𝑑3𝑟.(6) We utilize the fast Fourier transform (FFT) technique for obtaining the magnetostatic energy as it involves double integrals in real space.

The Zeeman energy is 𝐸ext=𝜇0𝐻ext𝑑𝑀(𝑟)3𝑟,(7) where 𝐻ext is the externally applied magnetic field.

The second term in (1) is the “Gilbert damping” which takes into account energy dissipation, such as coupling to lattice vibrations [39] and spin-flip scattering [40]. This is used by most practitioners, although there is an active debate whether this form of the damping is correct [4143].

The last two terms on the right side of (1) describe STT which tends to drag the magnetization away from its initial state to its final state and drive the magnetization precession around the effective field. The scalar function is given by [9] 𝑔(𝐌,𝐏)=4+(1+𝜂)33+𝐌𝐏/𝑀2𝑠4𝜂3/21.(8) where 𝐇STT is the corresponding effective field given by 𝐇STT=2𝜇𝐵𝐽𝑔(𝐌,𝐏)𝐌×𝐏𝛾𝑒𝑑𝑀3𝑠,(9) where μB, J, d, e, and 𝑀𝑠, are the Bohr magneton, current density, thickness of the free layer, electron charge, and saturation magnetization, respectively.

Analysis and micromagnetic calculations based on this model show that magnetization reversal in the layer requires an initial deviation from strictly parallel or antiparallel alignment. If M is parallel or antiparallel to 𝐏(𝜃=0or𝜃=180), STT is zero as 𝐌×𝐏=𝟎. To initiate the simulation, a small angle deviation from the parallel or antiparallel configuration between M and P was assigned. The classical Oersted field scales as 1/r generated by conduction electrons, where r is the lateral size of the free layer, while STT scales as 1/𝑟2. Therefore, for small sizes like a nanopillar with the cross-section 64×64 nm2, STT dominates over the classical Oersted field. We ignore the Oersted field and thermal activation in this work for studying a “minimal” model of the spin torque-driven magnetization precession. In our simulations, we assumed 𝜃=0.1 for parallel and 𝜃=179.9 for antiparallel configurations at the beginning. The parameters of Co are adopted from the values at helium temperature (4.2 K) [26]: 𝜂=0.35,𝑀𝑠=1.446×106A/m, 𝛾=2.3245×105m/(As), 𝐾1=6.86×105J/m3, 𝛼=0.01, 𝐴=2.0×1011J/m, and 𝑑=2nm. The dynamics of magnetization is investigated by numerically solving the time-dependent LLGS equation using the Gauss-Seidel projection method [44, 45] with a constant time step Δ𝑡=14.875fs. Simulation performed with shorter time step gave the same results. The free layer is discrete in computational cells of 2×2×2nm3 [46].

3. Results

Figure 2 shows the phase diagram of magnetization precession under perpendicular magnetic fields along z-axis direction. Each precessional state corresponds to a certain current density range under external field. The case of positive current results in magnetic switching from antiparallel state to parallel state, like the situation without magnetic field [26]. The magnetization process can be grouped into four types: unchanging state (UN), precessional state (PS), multidomain state (MS), and bistable state (BS). In UN, the magnetization keeps in the initial antiparallel state. The magnetizations in PS precess around the effective field with steady motion in the film plane. In MS, the magnetization lies in the multidomain state. Only in BS, the magnetization can be switched from the initial antiparallel state to the final parallel state. However, different from the situation of the uniform state in PS under zero field [26], when the external magnetic field is applied, two precession modes were found in this state: steady and substeady precession modes.

A negative current favors antiparallel state, resulting in the magnetic reversal from the initial parallel to final antiparallel state, as shown in Figure 2(b). Different from the positive current case, only three types of states were observed in the reverse process with PS missing.

To illustrate the feature and mechanism of magnetization precession under perpendicular magnetic fields, we take 𝐻𝑧ext=0.10𝑀𝑠(1.82KOe) as an example. In this condition, if the current density varies in the range of 0𝐽<1.50×107A/cm2, the magnetization dynamics lies in the US. In this state, STT energy is less than the Gilbert energy dissipation. Any finite fluctuation is damped down, and the magnetization keeps in the initial state. Figures 3(a) and 3(b) show that the temporal evolutions of magnetization components are heavily damped under 𝐽=1.00×107A/cm2.

In the range of 1.50×107A/cm2𝐽<1.95×107A/cm2, the magnetization dynamics lies in the PS, where the Gilbert energy dissipation equals the STT energy input. PS corresponds to the in-plane precession (IPP) in the film plane; the out-of-plane precession (OPP) is not found in our micromagnetic simulations, which means the transition from the precessional state to multidomain state occurs directly in the presence of magnetic fields. The absence of OPP was identified as a consequence of the underestimation of the exchange constant and the role of the initial self-magnetostatic field, which was demonstrated by Jaromirska et al. [47]. The two modes in the PS, that is, steady precession mode, and substeady precession mode, are in the ranges of 1.50×107A/cm2𝐽<1.65×107A/cm2 and 1.65×107A/cm2𝐽1.95×107A/cm2, respectively.

Figure 4 shows the temporal evolution of the magnetization components in the steady precession mode. The temporal evolution of 𝑀𝑥/𝑀𝑠 shows two stages: initial and stable stages. In the initial stage, the Gilbert damping torque is smaller than the STT, therefore, the amplitude of 𝑀𝑥/𝑀𝑠 increases with the time evolution. Then the Gilbert energy dissipation equals the STT energy input in the stable stage; the magnetic moments precess around the effective field, with oscillations at constant frequency and amplitude. Different from the precession without magnetic field, the temporal evolution of the magnetization component 𝑀𝑥/𝑀𝑠 has a “sag” in the oscillation crest at a perpendicular magnetic field. Although this phenomenon had also been reported by Siracusano et al. [48] and Horley et al. [49, 50], the explanation is not found. We shall give a systematic explanation of this phenomenon later.

In the range of 1.65×107A/cm2𝐽1.70×107A/cm2, the magnetization lies in the substeady precession mode. Figure 5 depicts the temporal evolutions of the magnetization components under 𝐽=1.70×107A/cm2. The amplitude is modulated by a low-frequency sinusoidal contour, showing the feature of “beat” phenomenon. The beat frequency equals to 4.03 GHz and this phenomenon does not appear in the case of zero applied fields. Though Finocchio et al. [51] and Berkov and Gorn [52] reported this mode, the mechanism is not presented. We will show the discussion of the mechanism.

Figure 6 shows the temporal evolution of the magnetization components. In the range of 1.95×107A/cm2𝐽<5.5×107A/cm2, the magnetization dynamics lies in the MS, which corresponds to the region “W” described in [14]. The mechanism is similar to that of PS. The differences are that the free layer is not a monodomain and the amplitudes of magnetization oscillation are not constant. This multidomains evolution process could be explained by the large current input energy. The energy per unit time pumped into the nanopillar by the current is so large, that the formation of magnetic excitations with the wavelength much shorter than the element size becomes possible, leading to the formation of multidomains.

When <𝐽10.0×107A/cm2 and5.5×107A/cm2𝐽<, the magnetization can be switched by the STT. As shown in Figure 7, the magnetization dynamics lies in the BS. In this state, the magnetization can be switched back and forth between two stable states: parallel or antiparallel configurations. The threshold current densities are 𝐽𝑐=5.5×107A/cm2 from antiparallel state to parallel state and 𝐽𝑐=10.0×107A/cm2 from parallel state to antiparallel state. The magnetization switching is accompanied by drastic magnetization oscillation. In fact, as soon as the current is applied, the magnetization oscillation occurs until the final static state is reached. Comparing with the critical current density of the zero fields, the threshold switching current under perpendicular magnetic fields is lower. This result can explain that the applied field changes the direction of effective field and stimulates the magnetization precession around it.

4. Discussion

To understand magnetization precession phase diagram of spin-valve nanopillar under perpendicular external magnetic fields, one needs to study the characteristics of STT and Gilbert damping torque. The balance between them determines the final states of the system.

STT can be described by the Slonczewski’s model, which takes into account the interface spin-flip scattering effect. STT versus θ is described by (9), and Gilbert damping torque is proportional to the function sin θ [53, 54]. They are plotted in Figure 8, where the current density is 9.0×107A/cm2, and the perpendicular external magnetic field of 0.10𝑀𝑠 is applied. STT reaches its maximum at 𝜃=127.17, which leads to easier reversal from the antiparallel to the parallel state than that of the reverse process. Thus |𝐽𝑐|(5.5×107A/cm2) is smaller than |𝐽𝑐|(|10.0×107A/cm2|).

As shown in Figure 9, in UN, the total torque is negative in all ranges of θ since STT is less than the Gilbert damping torque in this state. Therefore, any finite oscillation is damped and the magnetization keeps in the initial state.

There are two curves of total torque for PS in Figure 9, corresponding to two precession modes. Stable precession mode occurs at angles where the total torque changes from positive to negative, which means the Gilbert energy dissipation equals STT energy input in this case. In the steady precession mode, the total torque is always equal to zero with θ in the range from 178° to 179°. The magnetization trajectory is precessing around the effective field, showing a limiting cycle of saddle-shaped orbit which is plotted in Figure 10(a). The color curves stand for the projections on the planes Oxy (red), Oxz (green), and Oyz (blue), respectively. The simulated magnetization dynamics has a steady-state character, as can be observed from the temporal evolutions of magnetization components in Figure 4. With the increase of current density, the temporal evolutions of magnetization components lie in substeady precession mode, and the modulation oscillation can be observed in Figure 5. The total torque equals to zero at the range of 166𝜃179, and the magnetization trajectory precesses with the effective field in this range from the limiting cycle to a band of cycles as shown in Figure 10(b). The quasiperiodic magnetization trajectory filling a bent torus was also shown in [46, 47], but we give the explanation of this mode formation. The explanation of the “beat” phenomenon is that since the range of θ corresponding to the balance between STT and the Gilbert damping torque increases from 178𝜃179 to 166𝜃179 with the increase of current density, that is, the range of the precession angle is enlarged, therefore, the transition of the trajectory from the limiting cycle to the band of cycles occurs. In addition, there is no PS in the case of negative current, since the total torque is negative at small θ. This explains the absence of PS in the case of negative current density, as shown in magnetic dynamics phase diagram of Figure 2.

In MS, the total torque becomes positive at the range of 70𝜃179. It is impossible for the magnetization dynamics to have a steady precessional state at 𝜃=70 because micromagnetic simulation divides the pillar into many cell grids and the magnetization in each domain will precess around the local effective field. It is different from the macrospin model which describes the magnetic particle as a macroscopic magnetic moment [47, 52, 53].

In BS, STT increases with θ and eventually destabilizes the initial antiparallel state. When the total torque becomes positive in all ranges of θ, the system abruptly switches to the parallel state. In this state, STT is always bigger than the Gilbert damping torque in the whole switching process so that the total torque may drive the magnetization switching. In addition, the balance between STT and Gilbert damping torque for the negative current has similar behavior except in PS.

We show the sketch of magnetization precession trajectories in Figure 11(a) for explaining the “sag” in the oscillation crest of Figure 4. The magnetization precesses around the effective field 𝐻e along the saddle-shaped orbit [14]. The temporal evolution of three components 𝑚𝑥, 𝑚𝑦, and 𝑚𝑧 of magnetization 𝐦(𝐦=𝐌/𝑀𝑠) may be obtained by projecting the magnetization trajectory onto three axes. For example, the magnetization trajectory under the x-axis magnetic field can be projected to the plane Oxy and then to the x-axis to get the x-component of magnetization 𝑚𝑥, as shown in Figure 11(b). There is a curve with two overlapping minor arcs symmetric to the x-axis. When the magnetization is precessing around the x-axis, the 𝑚𝑥 component moves back and forth twice during one complete precession cycle, and hence the oscillation frequency of 𝑚𝑥 is approximately twice as large as that for 𝐦 oscillation. The projection under the y-axis field is a curve asymmetrical to the x-axis and inclining towards +y-axis. Thus, the temporal evolution of 𝑚𝑥 has a “salient” in the trough, which results from the asymmetrical projection to the x-axis on the plane Oxy. The projection under the z-axis field is a loop with two nonoverlapping minor arcs symmetrical to the x-axis, which leads to the result of magnetization oscillation having a “sag” in the crest. For the field along xyz-axis, the oscillation has two maximums and two minimums in one cycle of precession, since the projection is a loop asymmetrical to the x-axis and inclining towards +y-axis.

Furthermore, the “sag” in the crest will disappear as the magnetic field increases from zero to 0.30𝑀𝑠𝑧(𝐻𝑧=0.30𝑀𝑠), as shown in Figure 12. In the zero field, the projection on the plane Oxy is a curve with two overlapping minor arcs. When the 𝐦 precesses around the effective field for one cycle, the 𝑚𝑥 has two oscillations. There is no “sag” in the crest in this case. But the applied field increases to 0.05𝑀𝑠𝑧, there are “sags” in the crests of the temporal evolution of magnetization component 𝑚𝑥, because of the projection of a loop with two nonoverlapping minor arcs. As the applied field increases to 0.20𝑀𝑠𝑧, the “sag” in the crest becomes smaller than that for the field of 0.05𝑀𝑠𝑧. When the applied field increases to 0.30𝑀𝑠, the projection on the plane Oxy becomes a loop with a straight line and an arc from two nonoverlapping minor arcs and the “sag” disappears. Similar behavior can be found in the fields along y-axis and xyz-axis.

Figure 13 shows the temporal evolution of 𝑚𝑥, trajectory, and snapshots of the magnetization distribution under 𝐻𝑦=0.04𝑀𝑠 and 𝐽=1.60×107A/cm2. The 𝑚𝑥 has a maximum −0.930 at the point a, a second maximum −0.973 at the point b, and two minimums −0.983 at the points b and d. The five snapshots of magnetization distributions correspond to the five points in the temporal evolution of magnetization. The colors represent the magnitude of the magnetization component 𝑚𝑧 (−0.3~0.3), while the arrows indicate the magnetization projection on the plane Oxy, through which the magnetization components 𝑚𝑥 and 𝑚𝑦 were shown. The magnetization almost entirely lies in the plane Oxy at the point a, where 𝑚𝑧 is nearly zero. At the point b, the average magnetization component 𝑚𝑧 is about −0.2. The magnetization is first excited at the boundaries of the structure via the nucleation process because of spatially nonuniform local demagnetization fields [33] and becomes broader and broader with time, showing nonuniform magnetization precession. In the point c, the magnetization is also on the plane Oxy. However, the arrows rotate toward a different direction. The average magnetization component 𝑚𝑧 is 0.2 at the point d. At the end of the period, the oscillation returns to its original state.

5. Conclusions

Based on our micromagnetic simulation of spin transfer torque magnetization precession under magnetic fields in Co/Cu/Co nanopillars, the magnetization process can be grouped into four types: unchanging state (UN), precessional state (PS), multidomain state (MS), and bistable state (BS). Of the four states, PS could be further divided into two modes: steady precession mode and substeady precession mode. Balance of the spin transfer torque and Gilbert damping torque can lead to the three modes; the magnetization dynamics depends on the total torque of STT and Gilbert damping torque. Only in the case of positive current, the PS state can present. In addition, the temporal evolutions of magnetization components in steady precession mode are explained by the method of the orbit projection, and the “sag” in the crest will disappear as the magnetic field increases from zero to 0.30𝑀𝑠𝑧. Furthermore, the magnetization distributions together with its trajectory and temporal evolution are shown to elucidate the nonuniform magnetization precession.

Acknowledgments

This work was sponsored by the National Science Foundation of China (11174030), by the US National Science Foundation under Grant no. DMR-1006541 (Chen), and in part by the China Scholarship Council. The computer simulations were carried out on the LION and Cyberstar clusters at The Pennsylvania State University.