Abstract
We present a theoretical investigation of the magnetisation reversal process in CoFeBbased magnetic tunnel junctions (MTJs). We perform atomistic spin simulations of magnetisation dynamics induced by combination of spin orbit torque (SOT) and spin transfer torque (STT). Within the model the effect of SOT is introduced as a Slonczewski formalism, whereas the effect of STT is included via a spin accumulation model. We investigate a system of CoFeB/MgO/CoFeB coupled with a heavy metal layer where the charge current is injected into the plane of the heavy metal meanwhile the other charge current flows perpendicular into the MTJ structure. Our results reveal that SOT can assist the precessional switching induced by spin polarised current within a certain range of injected current densities yielding an efficient and fast reversal on the subnanosecond timescale. The combination of STT and SOT gives a promising pathway to improve high performance CoFeBbased devices with high speed and low power consumption.
Introduction
In today’s information based society enormous amounts of data are stored and processed daily. Current spintronic devices such as spin transfer torque magnetic random access memories (STTMRAMs) are among the most promising emerging memory technologies. STTMRAMs have been considered as candidates for universal memories thanks to their potential for fast read/write operations, low power consumption, nonvolatility and durability^{1,2,3,4,5,6}. STTMRAMs are based on magnetic tunnel junctions (MTJs), that is a trilayer structure consisting of two conductive ferromagnets and a thin insulator in the middle, usually CoFeBbased and MgO respectively^{7}. One ferromagnet, termed the reference layer (RL), has pinned magnetisation and polarises the spin current flowing into the layer. The other ferromagnet, whose magnetisation can be freely rotated, is known as the free layer (FL). It can be reversed by a large enough excitation of the spin polarised current. In the case of STTMRAM this is achieved by injecting a perpendicular current into the MTJ and exploiting the torque arising from the sd exchange interaction between the conduction electrons and the local magnetisation. The information is retrieved by sensing the magnetisation state of the latter via the tunnel magnetoresistance (TMR) effect. These devices can yield fast switching, hence fast writing operations, on the order of a few nanoseconds. However, to achieve faster switching large current densities are required^{2,3,5,8}. This undermines the potential for low power applications. Moreover, since the injected current and the magnetisation of the free layer share the same direction at rest, it is necessary to induce misalignment of the magnetisation in order to have a nonzero torque. This might cause long incubation times thereby reducing the speed. STTMRAM is a two terminal device, i.e. the same electrical path is used to write and read. This offers the opportunity of reducing the cell dimension of the memory, thus going beyond the current siliconbased technology. On the other hand, the thin oxide that serves as tunnelling barrier in the MTJ may suffer damage when large current densities or voltages are applied. This can cause the breakdown of the device and result in durability issues^{5,8}.
Apart from the conventional STT switching by injecting a current perpendicular to the MTJ, the currentinduced magnetisation switching can be achieved by spinorbit torque (SOT), in SOTMRAM devices^{8,9,10}. SOT can be generated by various mechanisms^{11} such as spin Hall effect (SHE), Rashba effect, inverse spin Galvanic effect, anomalous Hall effect and interfacial spinprecessional scattering. SOT has recently sparked interest in both theoretical and experimental research due to its application diversity. For example, SOT switching driven by mechanism of spinprocessional scattering and anomalous Hall effect have been experimentally demonstrated by Baek et al.^{12} and Ma et al.^{13} respectively. Interestingly, generation of unconventional spin current in symmetryreduced system such as CuPt^{14} and spinorbit torque generated by the broken magnetic symmetry in the antiferromagnetic IrMn\(_3\)^{15} have been reported. One of the most known and studied mechanisms is SHE, which arises when a current is injected into a nonmagnetic metal and spin up and spin down electrons are scattered to opposite directions. This creates an accumulation of spin density in the directions orthogonal to the current flow. This effect is caused by spinorbit coupling and thus is larger in heavy metals^{11}. SOTMRAMs are devices based on MTJs that utilise SOT for the writing process, while using the TMR effect to retrieve the information as in STTMRAM. SOT can be achieved by placing a nonmagnetic heavy metal (HM), Pt, Ta or W generally, in contact with the ferromagnetic layer and injecting an inplane electrical current into the HM. Spinorbit coupling is responsible for the generation of a pure spin current orthogonal to the current direction. This is absorbed in the ferromagnet and exerts a torque on the magnetisation of the ferromagnet that can reverse its polarisation^{11}. Due to the nature of the process, reversal occurs in a few hundreds of picoseconds. SOT does not suffer the incubation time issue, however it requires a way to break the symmetry so that the magnetisation ends in the desired final state. The traditional approach is to apply an inplane magnetic field in the direction of the injected current density. The application of a field has the downside of increasing the power consumption of the device and it might affect the magnetisation of the MTJ itself or nearby cells. Nonetheless, SOTMRAMs are characterised by a slightly lower power consumption than STTMRAM^{16,17}. SOTMRAM is a three terminal device in which the writing and reading paths are separated: the writing current is injected into the metallic contact and does not flow across the MTJ, whereas a weak current injected into the MTJ is used to read. This ensures that only weak currents cross the MTJ avoiding the risk of damaging the barrier, hence improving the endurance. Moreover, it increases the reliability by avoiding undesired writing while reading since the read and write path are separated^{3,5,11,16}. This comes at the cost of larger memory cells than STTMRAM. Nonetheless, their fast operation, on the subnanosecond timescale, make these devices suitable for CPU cache memories^{11,16,17}.
Various alterations to the standard STTMTJ stack^{5,6}, such as dual MgO MTJs^{18,19}, have been proposed. However, the major limitation affecting these solutions is the minimum MTJ size achievable. Voltage control of magnetic anisotropy (VCMA) MTJs^{20,21} offer the potential of low power magnetisation operations. In practice however, VCMA requires the assist from a charged current or a very precise control of the voltage pulses to yield a reliable deterministic switching^{22}. Shape anisotropy MTJs^{23,24,25}, whose free layer is patterned into a tall cylindrical ferromagnet, have been identified as pathway to scale the MTJ size down to or below 10 nm. However, efficient magnetisation dynamics and transport properties in such structures are yet to be fully demonstrated. To overcome the limitation of an external magnetic field in SOTMRAMs, fieldfree switching has been proposed by controlling the geometry of the device^{26}, introducing a tilted magnetic anisotropy^{27}, using inplane MTJs^{28} and Ttype structures^{29}, exploiting the exchange bias via coupling of the ferromagnetic layer with an antiferromagnet^{30}, combining VCMA with SHEinduced SOT in inplane MTJs^{31}, using a magnetic hard mask to shape the SOT track^{32}. However, downsides of these solutions are the scaling of the cell size and the limited efficiency^{26,30} and in some cases difficulties of integration in the current CMOS technology. Recently, Wu et al.^{33} proposed a voltagegateassisted SOTMRAM that combining VCMA and SOT offers low energy consumption, fast operations and a high storage density. However, the design requires MTJ inplane dimensions to be reduced and the writing still requires the application of anplane magnetic field.
Another possible solution to overcome the intrinsic limitations affecting STT and SOTMRAMs is to combine these two effects in a single device^{11,34,35,36,37,38,39,40,41,42,43}. This opens two scenarios: STT assisted by SOT and SOT assisted by STT. In the first case SOT acts by inducing an initial misalignment of the magnetisation of the ferromagnet. This would eliminate the incubation time seen in STTMRAMs leading to significant reduction of the switching time. When STT assists SOTdriven dynamics, STT can provide the necessary bias to break the symmetry and provide deterministic switching of the magnetisation. This removes the necessity of an external field or to resort to engineered magnetic structures^{8,11,28,44,45,46} reducing the required current density. Most of the studies on SOTMRAM assisted by STT focused on threeterminal devices^{11,37,45,46}, even though there are reports of twoterminal devices^{38,39}. SOTassisted STTMRAMs are investigated in Ref.^{34,35,36,38,41} showing improvements in the switching time and power consumption, however in these works long switching times are considered. Wang and collaborators^{40} propose a toggle spin torque MTJ where SOT and STT current pulses are sequentially applied, whilst Xu et al.^{42} propose a triple level cell combining STT and SOTbased MTJs where the writing mechanism is achieved in two steps.
To develop the performance of MTJ, it is important to understand the magnetisation dynamics of the magnetic material used in devices, since the critical current density for the reversal process, operating speed and power consumption are closely related to the dynamic behaviour. This work aims to develop a computational model which combines the atomistic spin model and a spin transport model for MTJ design. The majority of MTJ models employed to study magnetic systems are generally based on the macrospin and micromagnetic formalism^{3,34,35,38,41,47,48,49,50,51,52,53}. These face limitations when dealing with the miniaturisation and downscaling of the the device dimensions to few nanometre due to the continuum assumption that the magnetisation varies smoothly with position. In particular the macrospin approach assumes uniform magnetic properties over the whole ferromagnetic layer and therefore it is constrained to uniform magnetic configurations. On the other hand, the discrete nature of the atomistic model allows to have spatially varying magnetic properties over different lattice sites, something hard for micromagnetic simulations even with the finest discretisation, enabling the description of interfaces, surface and finite size effects. These features make possible to describe the anisotropy in CoFeB/MgO MTJs more accurately, one of the most important parameter for technological applications, that arises from and is localised at the CoFeBMgO interface or the increase of the magnetic damping at the same interface^{54}. For these reasons, in this work we investigate the influence of the combined STT and SOT phenomena on the magnetisation switching dynamics of CoFeB/MgObased MTJs at the atomistic level, whose details can be found in the “Methods” section. The results offer a new pathway to control the magnetisation in MTJs by injecting both an inplane current into the HM adjacent to the ferromagnetic layer and a current perpendicular to the MTJ. This is shown to give rise to higher efficiency of spin torque switching.
Results
In this work we simulate the magnetisation dynamics induced by a combination of STT and SOT in CoFeB/MgO/CoFeB MTJs by means of an atomistic spin model. Figure 1a presents a sketch of the investigated system. We parametrise CoFeB (1.0 nm, RL)/MgO (0.85 nm, Barrier)/CoFeB (1.3 nm, FL) as done in Ref.^{55}, where the number in brackets is the thickness, and FL and RL refer to free and reference layer, respectively. A list of the parameters used in the simulations is presented in Table 1 and details of the model and parameters are given in the “Methods” section. To simulate the magnetisation dynamics induced by the combination of SOT and STT, we consider a MTJ of diameter ranging from 5 to 30 nm and perform simulations at a temperature of 0 K. However, the discussion presented here focuses on the properties of MTJs with diameter of 20 nm, since this is the dimension most likely to be used in real applications. The protocol, sketched in Fig. 1b, is such that in the initial 0.5 ns both \(j_e^{\mathrm {SOT}}\) and \(j_e^{\mathrm {STT}}\) are injected. After 0.5 ns, \(j_e^{\mathrm {SOT}}\) is switched off while \(j_e^{\mathrm {STT}}\) continues to be injected into the system for the next 3.5 ns. As shown in Fig. 1a, the current density \(j_e^{\mathrm {SOT}}\) is injected into the HM along the positive ydirection. As a consequence the spin polarisation \(\varvec{\hat{\sigma }}\) is directed along \(x\). The transverse current density \(j_e^{\mathrm {STT}}\) used to induce STT is injected perpendicularly into the RL along the zaxis. The system is initialised in an antiparallel state with the RL magnetisation aligned along the positive zaxis and the FL polarised along the opposite direction. We would like to point out that the realisation of SOT and STTinduced magnetisation reversal in MTJs with current perpendicular to the stack where, in absence of the HM, SOT is solely provided by Rashba effect, is not a viable solution. In fact, to have a nonzero Rashba effect the current should be applied in an inplane configuration, as it has been reported in literature^{56,57}, due to symmetry requirements. Thus this solution cannot be applied to the MTJ stack considered in the present study. To evaluate the dynamical response of the combined SOT+STT protocol we perform three investigations triggered by: (A) pure spin transfer torque, (B) pure spin orbit torque and (C) the combination of both STT and SOT as detailed in the following sections.
Pure spin transfer torque
We initially investigate the effect of the transverse injected current density \(j_e^{\mathrm {STT}}\) in the absence of the inplane current \(j_e^{\mathrm {SOT}}\) injected into HM. To ensure a minimum torque acting on the system the FL magnetisation is tilted \({1}^{\circ }\) from the zaxis. Charge currents with different densities are perpendicularly injected into the MTJ structure. As demonstrated in Fig. 2, we observe the precessional motion of magnetisation in the xy plane before reaching the steady state in the z direction. Increasing charge current density results in higher spin transfer torque acting on the magnetisation and subsequently reduces the switching time. In the case of pure STT the system exhibits the expected dynamics with a slow rise of the precession until \(M_z\) changes sign, followed by a more strongly damped motion due to the STT symmetry. As discussed in Ref.^{58}, STTinduced magnetisation dynamics is characterised by a rotation of the inplane component and by either coherent reversal or noncollinear reversal of the outofplane component. The former is always observed and is caused by the STT symmetry, the latter depends on the size and magnitude of \(j_e^{\mathrm {STT}}\). The excitation modes, introduced in the Method section, show an overall dominant coherent character of the dynamics in these cases, with excitation of the vortex and antivortex modes^{58} for strong \(j_e^{\mathrm {STT}}\) and large diameters. It is worth mentioning that when STT is modelled via the spin accumulation, the reversal mechanism is coherent up to diameters of 20 nm. Only for large \(j_e^{\mathrm {STT}}\) noncollinear modes are weakly excited, with the usual transition from a vortex to an antivortex mode. This differs slightly from the case of a macrospinderived approach, as in Ref.^{58}, where for large \(j_e^{\mathrm {STT}}\) the reversal can exhibit strong excitations of the noncollinear modes at such dimensions. This can likely be ascribed to the missing inplane spacial dependence and angular dependence in the latter approach, which can lead to an overestimation of the torque acting on the system.
We can extract the critical current density for STT switching \(j_c^{\mathrm {STT}}\) as the minimum \(j_e^{\mathrm {STT}}\) that yields the reversal of the magnetisation within a fixed time. Since it is expected that \(j_c^{\mathrm {STT}}\) varies with the pulse length, we extract \(j_c^{\mathrm {STT}}\) for 0.5, 1.0, 2.0 and 3.0 ns and we plot it as a function of MTJ diameter in Fig. 3a. To obtain reversal in less than a nanosecond large current densities need to be injected into the MTJ, on the order of \({2.5\times 10^{12}} \; \text {Am}^{2}\). Such a requirement can be halved or more if the pulse is applied for 2 ns. Small diameters are characterised by longer incubation time, as observed in Ref.^{58}, resulting in larger critical current densities. Thus, STTinduced precessional switching sets a limitation to size scaling in a low temperature regime.
Pure spin orbit torque
We next investigate the effect of the inplane current density \(j_e^{\mathrm {SOT}}\) in the absence of transverse \(j_e^{\mathrm {STT}}\) to evaluate the contribution of SOT to the magnetisation dynamics. We vary \(j_e^{\mathrm {SOT}}\) and we study the switching dynamics for different strengths of the fieldlike component of SOT (\(A_{\mathrm {F}}\)) with respect to the dampinglike term (\(A_{\mathrm {D}}\)). The magnetisation response induced by SOT is shown in Fig. 3c,d for a low and high \(j_e^{\mathrm {SOT}}\), respectively. For \(j_e^{\mathrm {SOT}}\) less than \({1\times 10^{12}} \; \text {Am}^{2}\) there is no precession of the magnetisation in the FL. For increasing magnitudes of \(j_e^{\mathrm {SOT}}\) weak precession of the magnetisation can be observed, however no reversal occurs, as shown in Fig 3c for \(j_e^{\mathrm {SOT}} ={1\times 10^{12}} \; \text {Am}^{2}\). As \(j_e^{\mathrm {SOT}}\) increases further, the magnetisation tends to align along the direction of \(\varvec{\hat{\sigma }}\), \(x\) in our case, within less than 0.1 ns. Afterwards the magnetisation is in equilibrium and only slow changes in the magnetisation occur. In this case we do not observe the precession of the magnetisation as the system reaches equilibrium before one precessional cycle is completed. A comparison of the different curves in Fig. 3c,d shows that a nonzero \(A_{\mathrm {F}}\) hinders the precession of the magnetisation and for large \(j_e^{\mathrm {SOT}}\) results in a tilt of the magnetisation towards the zaxis. However, this misalignment from the inplane direction alone cannot yield the complete reversal of the magnetisation. Pure SOT dynamics is characterised by coherent rotation of the magnetisation, even for the largest size investigated. We attribute the coherent behaviour to the symmetry of SOT that tends to favour inplane configurations and to suppress higher excitation modes.
In the case of pure SOT, since we cannot use the switching time to define the critical current density \(j_c^{\mathrm {SOT}}\), we consider the minimal \(j_e^{\mathrm {SOT}}\) that yields a reduced magnetisation \(m_z=M_z/M_{\mathrm {s}}\) larger than 0.7, as proposed in Ref.^{41}. The plot of the size dependence of \(j_c^{\mathrm {SOT}}\) as a function of different \(A_{\mathrm {F}}\) strengths is presented in Fig. 3b. \(j_c^{\mathrm {SOT}}\) decreases for increasing diameters, in a similar trend to \(j_c^{\mathrm {STT}}\), due to the magnetostatic contribution. However, we cannot extract \(j_c^{\mathrm {SOT}}\) for \(A_{\mathrm {F}} ={0}{\%}\) and \({10}{\%}A_{\mathrm {D}}\) since the magnetisation lies inplane at the end of the pulse, as clear from Fig. 3d.
Effect of combined spin transfer torque and spin orbit torque
Having fully characterised the dynamical response to STT and SOT individually, we now study the combination of both the effects using the protocol shown in Fig 1b. Our results show that STT can yield reversal, but this occurs in a few nanoseconds. On the other hand, a rapid magnetisation response of a few hundreds of picoseconds can be achieved through the SOT, with the caveat that an additional contribution is required to obtain a full and deterministic reversal of the magnetisation. Hence, using the SOT rapidly to develop a transverse magnetisation to assist the STT might be expected to yield a more efficient switching dynamics than for the two cases separately. Figure 4a shows the time evolution of the reduced zcomponent of the magnetisation as a function of \(j_e^{\mathrm {SOT}}\) for \(j_e^{\mathrm {STT}}\) = \({1\times 10^{11}, 1\times 10^{12}\text { and }5\times 10^{12} \; \text {Am}^{2}}\) with \(A_{\mathrm {F}} =A_{\mathrm {D}}\). Similarly to the results in presented in Ref.^{41}, we can distinguish different scenarios depending on the combination of \(j_e^{\mathrm {STT}}\) and \(j_e^{\mathrm {SOT}}\):

(1)
Small \(j_e^{\mathrm {STT}}\) and large \(j_e^{\mathrm {SOT}}\): in this case STT aids SOT acting as external field to bias the final magnetisation direction;

(2)
Intermediate \(j_e^{\mathrm {STT}}\) and large \(j_e^{\mathrm {SOT}}\): the system behaves as described in the previous point

(3)
Intermediate \(j_e^{\mathrm {STT}}\) and small \(j_e^{\mathrm {SOT}}\): in this case SOT alone cannot complete the reversal. Here SOT provides the initial tilt of the magnetisation to initiate the precessional dynamics during the 1 ns it is on. Once \(j_e^{\mathrm {SOT}}\) is switched off, the magnetisation follows the spin polarised induced magnetisation dynamics.

(4)
Large \(j_e^{\mathrm {STT}}\): here the reversal occurs in less than a nanosecond with SOT due to reduction of the incubation time through the SOTinduced initial torque assistance to the STT.
The following analysis of the magnitudes of the current densities shows that both conditions 1 and 4 are not efficient in terms of energy requirements. Those could be employed in high speed devices where the power consumption is not an issue. On the other hand, the scenario depicted in 3 would reduce the total switching time without increasing the energy requirements. Spin transfer torque switching assisted by spin orbit torque could prove useful for applications, in particular those where the balance between the efficiency and the power consumption is essential, such as in portable and smart devices.
We compute the write energy consumption \(E_{\mathrm {w}}\), i.e. the energy required to reverse the magnetisation of the MTJ. \(E_{\mathrm {w}}\) is obtained by summing the contributions from the MTJ and the HM:
where the integration is carried over the whole pulse duration \(t_f  t_i\). \(I_\mathrm {MTJ}(t)\) and \(I_\mathrm {HM}\) are the electrical currents in the MTJ and in the HM at time t, respectively; similarly \(R_\mathrm {MTJ}(t)\), \(R_\mathrm {HM}(t)\) are the resistances of the MTJ and the HM. \(R_\mathrm {MTJ}(t)\) is related to the gradient of the spin accumulation and spin current, and can be obtained from the resistancearea product \(RA_\mathrm {MTJ}\)^{59,60}:
\(RA_\mathrm {MTJ}^i\) is the resistancearea product at position i, \(\Delta m\) is the difference of spin accumulation across the layers, \(j_m ^\mathrm {tot}\) is the total spin current within the system, \(a_0\) is the lattice constant of the material, t is the thickness of the regions the system is discretised into, e is the electron charge and \(k_{\mathrm {B}} T = {10}\text { meV}\) is the electronic temperature. This temperature is the electronic temperature at which ab initio calculations of the density of states of minority and majority spin populations, used to determine the equilibrium spin accumulation \(m_{\infty }\), are performed^{60,61}. \(j_m ^\mathrm {tot}\) accounts for both STT and SOT, hence during the initial 0.5 ns where both currents are applied \(j_m ^\mathrm {tot}\) includes the contribution \(j_m ^\mathrm {SOT}=\vartheta _{\mathrm {SH}} j_e^{\mathrm {SOT}}\).
The resistance of the HM contact \(R_\mathrm {HM}(t)\) can be obtained from the resistivity \(\rho _\mathrm {HM}\) of the HM as \(R_\mathrm {HM} = \rho _\mathrm {HM} L_y / (L_x L_z)\), where \(L_y\) is the length of the contact for \(j_e^{\mathrm {SOT}}\) injected along y and \(L_x L_z\) is the crosssectional surface crossed by the electrons^{35}. For Ta if we assume \(\rho _\mathrm {HM} \sim {200} \; \upmu \Omega \text {cm}\)^{62} and take a HM contact of dimensions \({50\times 50\times 4} \; \text { nm}\), we obtain \(R_\mathrm {HM} \sim {500} \; {\Omega }\). In Fig. 4b we present the write energy for a 20 nm MTJ with \(A_{\mathrm {F}} =A_{\mathrm {D}}\), corresponding to the dynamics shown in Fig. 4a. With the assistance of the SOT generated by the inplane current \(j_e^{\mathrm {SOT}}\), the magnetisation of the free layer can be reversed faster, but the average power dissipation of MTJ is greatly increased. However, when we consider the energy consumption of the MTJ related to the magnetisation reversal, i.e. neglecting other effects such as Joule heating, it can be clearly seen that low write energy can be achieved by optimising \(j_e^{\mathrm {STT}}\) and \(j_e^{\mathrm {SOT}}\). The obtained write energy is of the order of hundreds of fJ, in agreement with the results reported in literature for this kind of devices and also with technological requirements^{9,10,34,35,42,45,46,63,64}. Increasing \(j_e^{\mathrm {SOT}}\) results in high energy consumption, whereas the combination of SOT and STT currents assists in the reduction of write energy. Interestingly, for magnetisation switching solely driven by STT current, low energy consumption can be observed at high current density. This can be explained by considering that high current density significantly reduces the switching time yielding low writing energy. However, in practice it is difficult to generate high current density in a dot of few nanometre diameter without causing large Joule heating. The results suggest that the combination of low current density of SOT and STT can be used for a subnanosecond switching with low writing energy. This can be seen in panel c) of Fig. 4, where the product of write energy with switching time is plotted versus \(j_e^{\mathrm {STT}}\) and \(j_e^{\mathrm {SOT}}\). The diagram clearly shows how \(j_e^{\mathrm {STT}} \sim {\sim 1.2\times 10^{12} \; \text {Am}^{2}}\) and \(j_e^{\mathrm {SOT}} \lesssim {1\times 10^{12}} \; \text {Am}^{2}\) yield a switching that consumes 100fJ or less and it occurs in less than a nanosecond. More generally, such a diagram can prove useful to choose the most suitable switching schemes. Further, we consider the effect of different strengths of \(A_{\mathrm {F}}\). To analyse this, we consider \(j_e^{\mathrm {SOT}} =j_e^{\mathrm {STT}} ={1\times 10^{12}} \; \text {Am}^{2}\), presented in Fig. 5a. As discussed for the case of pure SOT and taking into consideration Eq. (11), \(A_{\mathrm {F}}\) acts on both precession and relaxation. Accordingly, we see a reduction in the switching time for weaker contributions of \(A_{\mathrm {F}}\), with the fastest dynamics occurring for zero \(A_{\mathrm {F}}\). The amplitude of the inplane components of the magnetisation decreases as \(A_{\mathrm {F}}\) increases and when SOT is turned off the reversal is delayed. Thus, weak SOT contribution to the precessional motion can yield better performance, with the limit case of zero \(A_{\mathrm {F}}\). However, this condition does not represent the most physical case for a structure such as the MTJ investigated here, where interfaces are not equivalent and the stack is not symmetric. Instead, it is reasonable to expect a nonzero fieldlike contribution. In Fig. 5b–e we present the total switching time as a function of both \(j_e^{\mathrm {STT}}\) and \(j_e^{\mathrm {SOT}}\) for different \(A_{\mathrm {F}}\) strength. From the phase plot we can see a large spectrum of combination of \(j_e^{\mathrm {STT}}\) and \(j_e^{\mathrm {SOT}}\) to yield the desired switching time. By comparing the contour plots for different \(A_{\mathrm {F}}\) we can see how, as \(A_{\mathrm {F}}\) increases, faster switching is achieved in the region 2 to \({4\times 10^{12}} \; \text {Am}^{2}\). In particular we see that the fastest switching time achievable by applying both \(j_e^{\mathrm {SOT}}\) and \(j_e^{\mathrm {STT}}\) below \({1\times 10^{12}} \; \text {Am}^{2}\) is 1 ns, independently of \(A_{\mathrm {F}}\).
Finally we investigate the reversal mechanism resulting from the combined application of STT and SOT to the system to evaluate the differences from the individual cases. The analysis involves calculation of the lowestfrequency modes (\(w=0,1,1\)) defined in Eq. (12). When we combine the two effects, we observe different dynamics depending on the relative strengths of \(j_e^{\mathrm {STT}}\) and \(j_e^{\mathrm {SOT}}\). As expected, for diameters up to 20 nm the reversal is coherent with rotation of the inplane components of the FL magnetisation. This is confirmed by the coherent mode \(m_0\) being the only nonzero excitation mode. This is shown in Fig. 6a where the time evolution of the \(m_0\) amplitude is plotted for \(A_{\mathrm {F}} =0.5 A_{\mathrm {D}}\) as a function of different \(j_e^{\mathrm {STT}}\) and \(j_e^{\mathrm {SOT}}\). For large \(j_e^{\mathrm {SOT}}\) \(m_0^2\) reaches the maximum in about 100ps, subsequently it retains a finite value. The height of these plateaus increases with \(j_e^{\mathrm {SOT}}\) followed by a decrease to zero after 0.5 ns, when \(j_e^{\mathrm {SOT}}\) is turned off. The excitation modes we consider are deviations from the uniform state, i.e. magnetisation aligned along the easy axis (\(\hat{z}\)). Thus, a finite value is indicative of the incomplete switching, with the magnetisation uniformly aligned in a direction with an angle from the zaxis. Conversely, increasing the magnitude of \(j_e^{\mathrm {STT}}\) for a fixed \(j_e^{\mathrm {SOT}}\) corresponds to a decrease in the height of these constant regions. Thus, a stronger \(j_e^{\mathrm {STT}}\) aids the reversal of the magnetisation, as also visible from Fig. 4a. The persistence of \(m_0\) during the whole STT + SOT pulse shows how SOT is able to keep the system in an outofequilibrium excited state when \(j_e^{\mathrm {SOT}}\) is large. Hence, to increase the energy efficiency, the pulse should be kept as short as possible for large \(j_e^{\mathrm {SOT}}\). For a diameter of 30 nm, in the absence of \(j_e^{\mathrm {SOT}}\) the reversal becomes noncollinear. However, the curves in Fig. 6a remain substantially unchanged for a diameter of 30 nm and hence this is not shown. Interestingly \(j_e^{\mathrm {SOT}}\) tends to counteract the formation of noncollinear configurations. This effect is proportional to the magnitude of \(j_e^{\mathrm {SOT}}\) and can be understood considering that SOT tends to favour inplane magnetic configurations. Let us consider \(j_e^{\mathrm {STT}} ={5\times 10^{12}} \; \text {Am}^{2}\). For \(j_e^{\mathrm {SOT}} ={0} \; \text {Am}^{2}\) the reversal is noncollinear despite \(m_0\) being the largest excited mode, as can be seen in Fig. 6b. As \(j_e^{\mathrm {SOT}}\) increases \(m_{1,1}\) modes are suppressed and the reversal becomes more coherent. For \(j_e^{\mathrm {SOT}} \ge {1\times 10^{12}} \; \text {Am}^{2}\) \(m_0\) is the only mode with nonzero amplitude, as shown in Fig. 6c. The change in reversal mechanism is also observed in the snapshots of the zcomponent of the FL magnetisation in these same figures. While for pure STT we can clearly observe the formation of a domain at the edge of the disk that propagates through the system accompanied by rotation around the zaxis, the magnetic texture becomes uniform for \(j_e^{\mathrm {SOT}} ={1\times 10^{12}} \; \text {Am}^{2}\). Similarly to what was already observed for the 20 nm diameter, further increase in \(j_e^{\mathrm {SOT}}\) results in \(m_0\) remaining excited until \(j_e^{\mathrm {SOT}}\) is turned off. Different \(A_{\mathrm {F}}\) strengths do not affect the reversal mechanism significantly and this is true for all diameters investigated. The main effect is the overdamped dynamics that quenches the precession of the inplane components and causes a different switching time, as already observed above. Our results suggest that for small MTJ diameters even a weak \(j_e^{\mathrm {SOT}}\) is able to make the reversal coherent. To induce nonuniformities such as nucleation processes in this combined STTSOT system larger diameters^{39,45,46} and/or strong current densities would be required. As such, our assumption of spatially uniform current densities should not impact the results presented here.
In Ref.^{35} switching around 1 ns for \(j_e^{\mathrm {SOT}} \sim 2.5\times 10^{11} \; \text {Am}^{2}\) is achieved in STTMTJs via SHE assist. The simulations are performed by means of a macrospin approximation and a relatively low anisotropy field \(\sim {1}\text { T}\) is used, thus allowing for switching at lower current densities. Wang et al.^{38} experimentally obtain fieldfree switching in three terminal CoFeB/MgObased perpendicular MTJs via interplay of STT and SOT. They also perform macrospin simulations including damping and fieldlike contributions for both STT and SOT. They find that \(A_{\mathrm {F}}\) results in a reduction in \(j_e^{\mathrm {SOT}}\) required to achieve reversal in contrast with our result where, depending on the relative magnitude between \(j_e^{\mathrm {STT}}\) and \(j_e^{\mathrm {SOT}}\), \(A_{\mathrm {F}}\) contributions can either hinder or assist the reversal. We attribute the difference in the behaviour to the sign, and thus symmetry, of the dampinglike torque (\(\varvec{m} \times \varvec{m} \times \varvec{\hat{\sigma }}\)) and fieldlike torque (\(\varvec{m} \times \varvec{\hat{\sigma }}\)) within the macrospin model used by Wang et al. instead of the conventional formalism^{11}. An analogous argument can be applied to the simulations performed in Ref.^{40} where a toggle MRAM is proposed. Our results exhibit similar features to those presented in Ref.^{41} by Pathak et al. However, we have investigated the effect of fieldlike contributions in SOT which were not considered in Ref.^{41} and we have found that a system where the fieldlike torque is minimised yields faster switching, in particular for larger \(j_e^{\mathrm {SOT}}\). Moreover, the spin accumulation includes these contributions naturally for STT, and thus it allows for a more complete treatment of the phenomenon. In addition, Pathak et al. attribute the initial tilt of the magnetisation required to provide a nonzero torque to the presence of an Oersted field. While this can justify the initial tilt of the magnetisation in case of pure STT, we argue that the Oersted contribution should be negligible in case of SOT. In our approach we do not include such a contribution and in case of STT at 0K we initialise the FL magnetisation at an angle of \({1}^{\circ }\) from the zaxis to ensure a nonzero torque acts on the magnetisation. However, SOT symmetry (\(\hat{\mathbf{S}}\times \varvec{\hat{\sigma }}\)) is such that when the FL magnetisation has an outofplane component there is always a nonvanishing torque^{38}, even for \(A_{\mathrm {F}} =0\). This is evident from Eq. (8) and it also represents one of the main advantages of SOT. We can estimate the Oersted contribution due to \(j_e^{\mathrm {SOT}} ={1\times 10^{12}} \; \text {Am}^{2}\) from the BiotSavart law to be \(\lesssim {0.01}\text { T}\) for our geometry, although this varies both in magnitude and direction; for the same \(j_e^{\mathrm {SOT}}\) the SOT fields are about an order of magnitude larger. Thus, importantly we expect the Oersted field contribution to be negligible in a combined STT–SOT dynamics since SOT provides the initial tilt of the magnetisation avoiding long incubation times. It is worth noting that at operational temperatures this contribution could be a source of Joule heating, as suggested in Ref.^{65}. In Ref.^{43} Zhang et al. apply a combination of SOT and STT in a currentinplane configuration with an inplane MTJ, whereas we aim to exploit perpendicular MTJ since it is already working for pure STT dynamics. Sato et al.^{39} report a twoterminal SOTMRAM cell based on a CoFeB/MgO MTJ. In this device the authors find that the switching mechanism is dominated by SOT and that STT offers little assist. Moreover, the setup requires an external inplane magnetic field and the MTJ diameter is 110 nm. Such a MTJ dimension is characterised by noncoherent magnetisation dynamics, something undesirable in such applications, and is not feasible for high storage applications. Grimaldi et al.^{45} and Krizakova et al.^{46} demonstrate very promising SOTinduced switching assisted by STT and VCMA in three terminal CoFeBbased MTJs. However, to achieve a deterministic reversal with 100% switching probability an external inplane magnetic field is applied in the case of SOTinduced dynamics. This is avoided in Ref.^{46} thanks to the stray field generated by an inplane ferromagnetic layer embedded in the hard mask used for the SOT line, using a similar expedient to the one employed in Ref.^{32}. In both studies however the relatively large dimension (\(>{50}\;\text {nm}\)) of the MTJ comports domainwall mediated magnetisation dynamics, and this appear as incubation time even for pure SOT dynamics affecting the speed and energy consumption of the device. Moreover, if on the one hand having the interplay of SOT, STT and VCMA allows for more available configurations and solutions, on the other it makes the setup and control of these structures more complex than in the MTJ proposed in the our work.
We would like to stress that the simulations performed in the aforementioned works are obtained via either macrospin or micromagnetic approaches. Differently from these, the atomistic approach allows to account for the localised nature of the anisotropy, which arises at the interface in CoFe/MgO systems, as well as of the damping, as discussed in Section and shown in Table 1. In addition, since the thickness of the ferromagnetic layers in state of the art devices is of the order of a nanometre, even fine micromagnetic discretisation cells of \({1} \;\text {nm}^{3}\) are not able to resolve the dynamics of both magnetisation and, where considered, spin accumulation across the thickness accurately. In particular a macrospin approach assumes uniform magnetisation in the whole ferromagnetic layer, and as a consequence it cannot resolve non uniform magnetisation configurations such as those shown in Fig. 6, which can occur in device dimensions of technological relevance. Atomistic approaches can instead provide this level of detail and become the most appropriate choice to deal with such complex and miniaturised systems. Furthermore, we have investigated systematically the impact of combined STT and SOT on the reversal mechanism. SOT suppresses the nonuniform excitation modes at large dimensions and even for a small \(j_e^{\mathrm {SOT}}\) the reversal becomes coherent. However this comes at the cost of maintaining the FL magnetisation in a nonequilibrium state until the SOT pulse is turned off.
Discussion and conclusions
We have developed a model that combines atomistic spin dynamics selfconsistently with spin transport models to investigate the miniaturisation necessary for novel MTJ design. By means of this approach we have modelled and investigated the magnetisation dynamics resulting from the combined action of spin orbit torque and spin transfer torque in CoFeB/MgO MTJs. By controlling \(j_e^{\mathrm {STT}}\) and \(j_e^{\mathrm {SOT}}\) the switching of the free layer magnetisation can be achieved in different regimes. Large \(j_e^{\mathrm {SOT}}\) yields fast switching dynamics, where the SOT induced magnetisation dynamics is aided by the STT and fieldfree switching is achieved. However, in this case there is the downside of a high power consumption. For low \(j_e^{\mathrm {SOT}}\) the SOT provides the initial tilt of the magnetisation assisting the STT dynamics with the possibility of reducing incubation time issues. Depending on the relative strength of \(j_e^{\mathrm {SOT}}\) and \(j_e^{\mathrm {STT}}\), the magnetisation follows a dynamical path which is closer to either that of STT or SOT or to a hybrid of both. Moreover, the reversal is characterised by coherent dynamics in presence of a nonzero \(j_e^{\mathrm {SOT}}\) up to MTJ diameters of 30 nm, attributed to the symmetry of SOT that tends to suppress higher excitation modes. This limits the possibility of pinning of the magnetic texture and avoid stochasticity of the switching process. Our results highlight an intermediate regime with moderate \(j_e^{\mathrm {STT}}\) and low \(j_e^{\mathrm {SOT}}\) where the SOT assist the STT dynamics yielding switching below the nanosecond by reducing the transient time. Such a configuration of STT+SOT could result in improved device performances without affecting the energy consumption.
Methods
Atomistic spin model
In this work we use an atomistic spin model as implemented in the open source software package vampire^{66,67}. The magnetisation dynamics of the system is modelled by integrating the stochastic LandauLifshitzGilbert (sLLG) equation of motion^{68}:
where \(\hat{\mathbf{S}}_i\) is the normalised unit spin vector on site i, \(\gamma =1.76086\times 10^{11}\text {s}^{1}\text {T}^{1}\) is the gyromagnetic ratio of the electron and \(\mathbf {B^i_{{eff}}}\) is the net effective field acting on the spin i. \(\mathbf {B^i_{{eff}}}\) is obtained as:
The first term of Eq. (4) accounts for all the contributions in the localised extended Heisenberg Hamiltonian (\(\mathcal {H}\)):
where \(J_{ij}\) is the exchange coupling constant between spins i and j, \(k_{\mathrm {u}} ^i\) is the uniaxial energy constant on site i along the easyaxis \(\hat{\varvec{e}}\), \(\mu _{\varvec{s}} ^i\) is the atomic spin moment on the atomic site i and \(\mathbf {B_{{app}}}\) is the external magnetic field. The magnetostatic contribution \(\mathcal {H} _{\mathrm {dmg}}\) is calculated using a modified macrocell approach^{69} based on the work of Bowden et al.^{70}. The second term of Eq. (4) couples the spin system with the heat bath via the the atomistic Gilbert damping parameter \(\lambda\). For processes occurring on the nanosecond timescale the thermal field \(\varvec{\xi }\) can be described as a white noise term represented by a 3D Gaussian distribution whose width depends on temperature and \(\lambda\)^{71}.
The spin transfer torque induced by the injection of spin polarised electrons into a ferromagnet is described following the spinaccumulation model including the contribution of adiabatic and nonadiabatic torques^{60,61,72}. A spin polarised current builds a spin accumulation (\(\mathbf {m}\)), an imbalance between the population of spinup and spindown electrons, when it crosses a ferromagnetic material. The exchange interaction between the s spin polarised electrons and the d electrons of the ferromagnet results in a transfer of angular momentum. This interaction can be described as:
where \(J_{\mathrm {sd}}\) is the s–d exchange interaction, \(\mathbf {m}\) is the spin accumulation and \(\hat{\mathbf{M}}\) is the unit vector magnetisation of the ferromagnet. The evolution of \(\mathbf {m}\) occurs on a faster timescale, about 1ps, than the magnetisation, around 100ps, for the processes we are interested in. This allows to decouple the dynamics of the two: first solving for \(\mathbf {m}\) assuming the magnetisation texture constant, and then calculating the torque via \(\mathcal {H} _{\mathrm {STT}}\), which can be included in the Hamiltonian of Eq. (5) and as such it enters the LLG equation via \(\mathbf {B^i_{{eff}}}\). The dynamics of the system drives \(\mathbf {m}\) towards an equilibrium value \(m_{\infty }\), defined as the difference between the equilibrium populations of spinup and spindown electrons^{60}. \(m_{\infty }\) is a characteristics of the materials and can be obtained from the density of states of majority and minority spin populations (\(\propto N^{\uparrow }  N^{\downarrow }\))^{60,73}, for instance by means of ab initio calculations. These calculations, as those reported in^{60}, are often performed at an electronic temperature of 10 meV. The spin accumulation can be conveniently calculated in a rotated coordinate system where the unit vector components \(\hat{\varvec{b}}_1\) and \(\hat{\varvec{b}}_{2,3}\) are aligned parallel and perpendicular to the magnetisation direction respectively. The general solution of spin accumulation can be split into two parts: the longitudinal (\({\varvec{m}_{\parallel }}\)) and transverse (\({\varvec{m}_{\bot }}\)) components given by:
where \((k_1 \pm ik_2) = \sqrt{\lambda _\mathrm {sf}^{2} \pm i \lambda _{J}^{2}}\), \({\lambda _\mathrm {sdl}}\) is the spin diffusion length, \({m}_{\parallel }(\infty )\) is the equilibrium value of spin accumulation, the spinflip length is defined as \({\lambda _\mathrm {sf}} = \sqrt{2{D_0}{\tau _{sf}}}\) and \({\lambda _\mathrm {J}}\) is the spinprecession length. The unknown variables u, v and \(m_{\Arrowvert }(0)\) can be evaluated from the boundary condition of the continuity of the spin current \(j_m\).
The SOT contribution is introduced as an additional term to in LLG equation, following a similar approach to Slonczewski^{74} for STT. If we assume a bulk origin of the phenomenon, hence SOT induced by spin Hall effect (SHE), we can write:
where \(\varvec{\hat{\sigma }}\) is the spin polarisation unit vector, \(A_{\mathrm {D}}\) and \(A_{\mathrm {F}}\) are the SHE spin orbit torque parameters that determine the strength of the damping and fieldlike torque terms respectively. In writing Eq. (8), we have assumed that SOT can affect both the precessional dynamics and the relaxation process. It is fact not simple in structures such as an MTJ to discriminate and clearly separate these effects^{11}. \(\varvec{\hat{\sigma }}\) represents the direction of the polarisation of the spin current induced by the flow of electrons in the non magnet. \(\varvec{\hat{\sigma }}\) is perpendicular to the electron flow, hence for a \(j_e^{\mathrm {SOT}}\) flowing along the yaxis \(\varvec{\hat{\sigma }}\) is directed both along x and zdirection. If the stack is piled along the zdirection, in first instance we can neglect the effect of \(\varvec{\hat{\sigma }}\) along the same direction as there are not relevant interfaces. In this case \(\varvec{\hat{\sigma }}\) can be expressed as:
where \(\hat{\mathbf{z}}\) is unit vector normal to the surface and \(\hat{\mathbf{E}}\) is the electric field associated with \(j_e^{\mathrm {SOT}}\). If we assume that the main origin of SOT is the spin Hall effect (SHE), the SOT parameters are given by:
Here, \(\vartheta _{\mathrm {SH}}\) is the spin Hall angle which gives the conversion efficiency of electrical current into spin current, \(d\) is the ferromagnet thickness, \(V_{\mathrm {at}}\) is the atomic volume and \(M_{\mathrm {s}} =\mu _{\varvec{s}}/V_{\mathrm {at}}\) is the magnetisation of the ferromagnet. This is not true generally, in particular for structures with structural inversion asymmetry^{11}, where the Rashba or inverse spin galvanic effect (iSGE) should be accounted for. However, due to the complexity in determining the strength of the Rashba effect, we have decided to apply the aforementioned approximation. In presence of such SOT, the solved LLG equation can be expressed explicitly as:
Even though this form of the LLG equation is not computationally efficient, it demonstrates that if both dampinglike and fieldlike contributions are included they intermix. If we consider an effective field \(\mathbf {\tilde{B}^i_{{eff}}} = \mathbf {B^i_{{eff}}} + \mathbf {B^i_{{SOT}}}\), where \(\mathbf {B^i_{{SOT}}} =A_{\mathrm {D}} \left( \hat{\mathbf{S}}_i \times \varvec{\hat{\sigma }} \right) + A_{\mathrm {F}} \varvec{\hat{\sigma }}\), the LLG equation with SOT can be written as Eq. (3) where \(\mathbf {B^i_{{eff}}}\) is replaced by \(\mathbf {\tilde{B}^i_{{eff}}}\).
Investigated system
We model CoFeB as cylindrical alloy with a bodycentred cubic (bcc) crystal structure with lattice constant 2.86Å. CoFeB/MgO is characterised by a large perpendicular singleion uniaxial interfacial anisotropy with zero anisotropy in the bulk^{75}. The CoFeB atomic layers opposite to MgO are assumed to have no particular interfacial properties and we neglect the small bulk anisotropy of CoFeB^{7}. We use a mean field approximation that depends on the Curie temperature \(T_{\mathrm {c}}\) to extract the atomistic exchange parameters \(J_{ij}\). To extract interfacial and bulk \(J_{ij}\) values we impose a unique \(T_{\mathrm {c}}\) in the whole CoFeB, where \(T_{\mathrm {c}}\) is taken from CoFeB/MgO thin film measurements^{75}. We describe CoFeB as having an atomic spin moment \(\mu _{\varvec{s}}\) of 1.6 \(\mu _{\mathrm {B}}\), in agreement with experimental reports on similar systems^{7}. The CoFeB/MgO interface has a larger damping than the rest of the CoFeB, in agreement with both experimental^{7,76} and theoretical^{54} works. Boron is not directly included in the simulations because it is not magnetic and because tends to diffuse out of the CoFe layer during annealing. The FL is capped by a heavy metal (HM) layer, which we take to be Ta. The role of HM is to convert electrical current into spin current that flows in the adjacent ferromagnet, whilst no magnetisation dynamics is assumed occurring within the material itself. Hence, we account for the effect of HM via the spin Hall angle \(\vartheta _{\mathrm {SH}}\), that we assume 0.11 for Ta^{34,77,78,79}, and we do not include it in the simulations explicitly.
Normal modes analysis
In order to characterise the reversal process we employ the method of Visscher et al.^{48,50} based on the stability of the lowest frequency normal modes, which are a small perturbation of the equilibrium magnetic state. They classify these modes by the winding number w, an integer index that can be understood as the number of times the magnetisation of the disk winds around a symmetry axis perpendicular to the plane of the disk when the disk is revolved once. The amplitudes of the three lowestfrequency modes (\(w=0,1,1\)) can be expressed as^{48,50}:
where x, y, \(s_x\) and \(s_y\) are the spatial and spin coordinates of the atoms within the system, respectively, i is the imaginary unit, w is the surface of the disk and dxdy is the surface element over which the integration is carried out. For the \(w=0\) mode the magnetisation in nearly independent of position, hence we refer to this as the uniform mode. This is the lowest energy mode and is defined as the “coherent mode”. The other two modes represent the first excited states and are defined as “vortex” (\(w=1\)) and “antivortex” (\(w=1\)).
Data availability
The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.
References
Ikeda, S. et al. Magnetic tunnel junctions for spintronic memories and beyond. IEEE Trans. Electron Devices 54, 991–1002. https://doi.org/10.1109/TED.2007.894617 (2007).
Apalkov, D. et al. Spintransfer torque magnetic random access memory (STTMRAM). J. Emerg. Technol. Comput. Syst. 9, 1–35. https://doi.org/10.1145/2463585.2463589 (2013).
Khvalkovskiy, A. V. et al. Erratum: Basic principles of STTMRAM cell operation in memory arrays. J. Phys. D Appl. Phys. 46, 139601 (2013).
Kent, A. D. & Worledge, D. C. A new spin on magnetic memories. Nat. Nanotechnol. 10, 187–191. https://doi.org/10.1038/nnano.2015.24 (2015).
Dieny, B., Goldfarb, R. B. & Lee, K. J. Introduction to Magnetic RandomAccess Memory (Wiley, 2017).
Bhatti, S. et al. Spintronics based random access memory: A review. Mater. Today 20, 530–548. https://doi.org/10.1016/j.mattod.2017.07.007 (2017).
Ikeda, S. et al. A perpendicularanisotropy CoFeB–MgO magnetic tunnel junction. Nat. Mater. 9, 721–724 (2010).
Endoh, T., Honjo, H., Nishioka, K. & Ikeda, S. Recent Progresses in STTMRAM and SOTMRAM for Next Generation MRAM. In 2020 IEEE Symposium on VLSI Technology, Vol. 2020June, 1–2, https://doi.org/10.1109/VLSITechnology18217.2020.9265042 (IEEE, 2020).
Garello, K. et al. SOTMRAM 300MM Integration for Low Power and Ultrafast Embedded Memories. In 2018 IEEE Symposium on VLSI Circuits, Vol. 54, 81–82, https://doi.org/10.1109/VLSIC.2018.8502269 (IEEE, 2018).
Cubukcu, M. et al. Ultrafast perpendicular spinorbit torque MRAM. IEEE Trans. Magn. 54, 1–4. https://doi.org/10.1109/TMAG.2017.2772185 (2018).
Manchon, A. et al. Currentinduced spinorbit torques in ferromagnetic and antiferromagnetic systems. Rev. Mod. Phys. 91, 035004. https://doi.org/10.1103/RevModPhys.91.035004 (2019).
Baek, S.H.C. et al. Spin currents and spin orbit torques in ferromagnetic trilayers. Nat. Mater. 17, 509–513. https://doi.org/10.1038/s4156301800415 (2018).
Ma, T. Y. et al. Evidence of magnetization switching by anomalous spin hall torque in NiFe. Phys. Rev. B 101, 134417 https://doi.org/10.1103/PhysRevB.101.134417 (2020).
Liu, L. et al. Symmetrydependent fieldfree switching of perpendicular magnetization. Nat. Nanotechnol. 16, 277–282. https://doi.org/10.1038/s41565020008268 (2021).
Zhou, J. et al. Magnetic asymmetry induced anomalous spinorbit torque in IrMn. Phys. Rev. B 101, 184403 https://doi.org/10.1103/PhysRevB.101.184403 (2020).
Prenat, G. et al. Ultrafast and highreliability SOTMRAM: From cache replacement to normallyoff computing. IEEE Trans. MultiScale Comput. Syst. 2, 49–60. https://doi.org/10.1109/TMSCS.2015.2509963 (2016).
Jabeur, K., Pendina, G. & Prenat, G. Study of spin transfer torque (STT) and spin orbit torque (SOT) magnetic tunnel junctions (MTJS) at advanced CMOS technology nodes. Electr. Electron. Eng. Int. J. 6, 01–09, https://doi.org/10.14810/elelij.2017.6101 (2017).
Diao, Z. et al. Spin transfer switching in dual MgO magnetic tunnel junctions. Appl. Phys. Lett. 90, 132508 https://doi.org/10.1063/1.2717556 (2007).
Lee, S.E., Takemura, Y. & Park, J.G. Effect of double MgO tunneling barrier on thermal stability and TMR ratio for perpendicular MTJ spinvalve with tungsten layers. Appl. Phys. Lett. 109, 182405 https://doi.org/10.1063/1.4967172 (2016).
Weisheit, M. et al. Electric fieldinduced modification of magnetism in thinfilm ferromagnets. Science 315, 349–351. https://doi.org/10.1126/science.1136629 (2007).
Maruyama, T. et al. Large voltageinduced magnetic anisotropy change in a few atomic layers of iron. Nat. Nanotechnol. 4, 158–161. https://doi.org/10.1038/nnano.2008.406 (2009).
Kang, W., Ran, Y., Zhang, Y., Lv, W. & Zhao, W. Modeling and exploration of the voltagecontrolled magnetic anisotropy effect for the nextgeneration lowpower and highspeed MRAM applications. IEEE Trans. Nanotechnol. 16, 387–395. https://doi.org/10.1109/TNANO.2017.2660530 (2017).
Watanabe, K., Jinnai, B., Fukami, S., Sato, H. & Ohno, H. Shape anisotropy revisited in singledigit nanometer magnetic tunnel junctions. Nat. Commun. 9, 663. https://doi.org/10.1038/s41467018030037 (2018).
Perrissin, N. et al. Perpendicular shape anisotropy spin transfer torque magnetic randomaccess memory: Towards sub10 nm devices. J. Phys. D Appl. Phys. 52, 234001 https://doi.org/10.1088/13616463/ab0de4 (2019).
Igarashi, J. et al. Temperature dependence of the energy barrier in X/1X nm shapeanisotropy magnetic tunnel junctions. Appl. Phys. Lett. 118, 012409 https://doi.org/10.1063/5.0029031 (2021).
Safeer, C. K. et al. Spinorbit torque magnetization switching controlled by geometry. Nat. Nanotechnol. 11, 143–146. https://doi.org/10.1038/nnano.2015.252 (2016).
You, L. et al. Switching of perpendicularly polarized nanomagnets with spin orbit torque without an external magnetic field by engineering a tilted anisotropy. Proc. Natl. Acad. Sci. 112, 10310–10315. https://doi.org/10.1073/pnas.1507474112 (2015).
Aradhya, S. V., Rowlands, G. E., Oh, J., Ralph, D. C. & Buhrman, R. A. Nanosecondtimescale low energy switching of inplane magnetic tunnel junctions through dynamic Oerstedfieldassisted spin Hall effect. Nano Lett. 16, 5987–5992. https://doi.org/10.1021/acs.nanolett.6b01443 (2016).
Kong, W. J. et al. Allelectrical manipulation of magnetization in magnetic tunnel junction via spinorbit torque. Appl. Phys. Lett. 116, 162401. https://doi.org/10.1063/5.0001758 (2020).
Fukami, S., Zhang, C., DuttaGupta, S., Kurenkov, A. & Ohno, H. Magnetization switching by spinorbit torque in an antiferromagnetferromagnet bilayer system. Nat. Mater. 15, 535–541. https://doi.org/10.1038/nmat4566 (2016).
Inokuchi, T. et al. Improved read disturb and write error rates in voltagecontrol spintronics memory (VoCSM) by controlling energy barrier height. Appl. Phys. Lett. 110, 252404 https://doi.org/10.1063/1.4986923 (2017).
Garello, K. et al. Manufacturable 300 mm platform solution for FieldFree Switching SOTMRAM. In 2019 Symposium on VLSI Circuits, T194–T195, https://doi.org/10.23919/VLSIC.2019.8778100 (IEEE, 2019).
Wu, Y. et al. Voltagegateassisted spinorbittorque magnetic randomaccess memory for highdensity and lowpower embedded applications. Phys. Rev. Appl. 15, 1–10. https://doi.org/10.1103/physrevapplied.15.064015 (2021).
van den Brink, A. et al. SpinHallassisted magnetic random access memory. Appl. Phys. Lett. 104, 012403. https://doi.org/10.1063/1.4858465 (2014).
Wang, Z., Zhao, W., Deng, E., Klein, J.O. & Chappert, C. Perpendicularanisotropy magnetic tunnel junction switched by spinHallassisted spintransfer torque. J. Phys. D Appl. Phys. 48, 065001 https://doi.org/10.1088/00223727/48/6/065001 (2015).
Chang, L. et al. Evaluation of spinHallassisted STTMRAM for cache replacement. Proceedings of the 2016 IEEE/ACM International Symposium on Nanoscale Architectures, NANOARCH 2016 73–78, https://doi.org/10.1145/2950067.2950107 (2016).
Zeinali, B., Madsen, J. K., Raghavan, P. & Moradi, F. UltraFast SOTMRAM Cell with STT Current for Deterministic Switching. In 2017 IEEE International Conference on Computer Design (ICCD), 463–468, https://doi.org/10.1109/ICCD.2017.81 (IEEE, 2017).
Wang, M. et al. Fieldfree switching of a perpendicular magnetic tunnel junction through the interplay of spinorbit and spintransfer torques. Nat. Electron. 1, 582–588. https://doi.org/10.1038/s4192801801607 (2018).
Sato, N., Xue, F., White, R. M., Bi, C. & Wang, S. X. Twoterminal spinorbit torque magnetoresistive random access memory. Nat. Electron. 1, 508–511. https://doi.org/10.1038/s419280180131z (2018).
Wang, Z. et al. Proposal of toggle spin torques magnetic RAM for ultrafast computing. IEEE Electron Device Lett. 40, 726–729. https://doi.org/10.1109/LED.2019.2907063 (2019).
Pathak, S., Youm, C. & Hong, J. Impact of spinorbit torque on spintransfer torque switching in magnetic tunnel junctions. Sci. Rep. 10, 1–8. https://doi.org/10.1038/s4159802059533y (2020).
Xu, Y. et al. Writeefficient STT/SOT hybrid triplelevel cell for highdensity MRAM. IEEE Trans. Electron Devices 67, 1460–1465. https://doi.org/10.1109/TED.2019.2963421 (2020).
Zhang, C., Takeuchi, Y., Fukami, S. & Ohno, H. Fieldfree and subns magnetization switching of magnetic tunnel junctions by combining spintransfer torque and spinorbit torque. Appl. Phys. Lett. 118, 092406 https://doi.org/10.1063/5.0039061 (2021).
Liu, Y., Zhou, B. & Zhu, J.G. Fieldfree magnetization switching by utilizing the spin Hall effect and interlayer exchange coupling of iridium. Sci. Rep. 9, 325. https://doi.org/10.1038/s41598018375864 (2019).
Grimaldi, E. et al. Singleshot dynamics of spinorbit torque and spin transfer torque switching in threeterminal magnetic tunnel junctions. Nat. Nanotechnol. 15, 111–117. https://doi.org/10.1038/s4156501906077 (2020).
Krizakova, V. et al. Interplay of voltage control of magnetic anisotropy, spintransfer torque, and heat in the spinorbittorque switching of threeterminal magnetic tunnel junctions. Phys. Rev. Appl. 15, 054055. https://doi.org/10.1103/PhysRevApplied.15.054055 (2021).
Jang, P. H., Song, K., Lee, S. J., Lee, S. W. & Lee, K. J. Detrimental effect of interfacial DzyaloshinskiiMoriya interaction on perpendicular spintransfertorque magnetic random access memory. Appl. Phys. Lett. 107, 202401. https://doi.org/10.1063/1.4936089 (2015).
Munira, K. & Visscher, P. B. Calculation of energybarrier lowering by incoherent switching in spintransfer torque magnetoresistive randomaccess memory. J. Appl. Phys. 117, 17B710. https://doi.org/10.1063/1.4908153 (2015).
Sampaio, J. et al. Disruptive effect of DzyaloshinskiiMoriya interaction on the magnetic memory cell performance. Appl. Phys. Lett. 108, 112403. https://doi.org/10.1063/1.4944419 (2016).
Visscher, P. B., Munira, K. & Rosati, R. J. Instability Mechanism for STTMRAM switching. Arxiv. CondMat 3–6 (2016).
Pathak, S., Cha, J., Jo, K., Yoon, H. & Hong, J. Fast and efficient STT switching in MTJ using additional transient pulse current. Appl. Phys. Lett. 110, 232401. https://doi.org/10.1063/1.4985129 (2017).
Bouquin, P., Rao, S., Kar, G. S. & Devolder, T. Size dependence of spintorque switching in perpendicular magnetic tunnel junctions. Appl. Phys. Lett. 113, 222408. https://doi.org/10.1063/1.5055741 (2018).
Beik Mohammadi, J. & Kent, A. D. Spintorque switching mechanisms of perpendicular magnetic tunnel junction nanopillars. Appl. Phys. Lett. 118, 132407. https://doi.org/10.1063/5.0046596 (2021).
SampanAPai, S. et al. Temperature and thickness dependence of statistical fluctuations of the Gilbert damping in CoFeB/MgO bilayers. Phys. Rev. Appl. 11, 044001. https://doi.org/10.1103/PhysRevApplied.11.044001 (2019).
Meo, A. et al. Thermally nucleated magnetic reversal in CoFeB/MgO nanodots. Sci. Rep. 7, 16729. https://doi.org/10.1038/s41598017169113 (2017).
Mihai Miron, I. et al. Currentdriven spin torque induced by the Rashba effect in a ferromagnetic metal layer. Nat. Mater. 9, 230–234. https://doi.org/10.1038/nmat2613 (2010).
Dieny, B. & Chshiev, M. Perpendicular magnetic anisotropy at transition metal/oxide interfaces and applications. Rev. Mod. Phys. 89, 025008 https://doi.org/10.1103/RevModPhys.89.025008 (2017).
Meo, A. et al. Spin transfer torque switching dynamics in CoFeB/MgO magnetic tunnel junctions. Phys. Rev. B 103, 054426. https://doi.org/10.1103/PhysRevB.103.054426 (2021).
Chung, N. L., Jalil, M. B. A., Tan, S. G. & Bala Kumar, S. Interfacial resistance and spin flip effects on the magnetoresistance of a current perpendicular to plane spin valve. J. Appl. Phys. 103, 07F308. https://doi.org/10.1063/1.2831368 (2008).
Chureemart, P., Cuadrado, R., D’Amico, I. & Chantrell, R. W. Modeling spin injection across diffuse interfaces. Phys. Rev. B 87, 195310. https://doi.org/10.1103/PhysRevB.87.195310 (2013).
Saenphum, N., Chureemart, J., Chantrell, R. & Chureemart, P. Model of spin transport in noncollinear magnetic systems: Effect of diffuse interfaces. J. Magn. Magn. Mater. 484, 238–244 (2019).
Cecot, M. et al. Influence of intermixing at the Ta/CoFeB interface on spin Hall angle in Ta/CoFeB/MgO heterostructures. Sci. Rep. 7, 968. https://doi.org/10.1038/s4159801700994z (2017).
Jan, G. et al. Demonstration of UltraLow Voltage and Ultra Low Power STTMRAM designed for compatibility with 0x node embedded LLC applications. In 2018 IEEE Symposium on VLSI Technology, Vol. 2018June, 65–66, https://doi.org/10.1109/VLSIT.2018.8510672 (IEEE, 2018).
Sharma, A., Tulapurkar, A. A. & Muralidharan, B. Proposal for energy efficient spin transfer torquemagnetoresistive random access memory device. J. Appl. Phys. 129, 233901. https://doi.org/10.1063/5.0052693 (2021).
Ho, P., Zhang, J., Bono, D. C., Chen, J. & Ross, C. A. Oersted field and spin current effects on magnetic domains in [Co/Pd] 15 nanowires. IEEE Trans. Magn. 52, 1–6. https://doi.org/10.1109/TMAG.2016.2526972 (2016).
Computer code vampire 5.0.
Evans, R. F. L. et al. Atomistic spin model simulations of magnetic nanomaterials. J. Phys. Condens. Matter 26. https://doi.org/10.1088/09538984/26/10/103202 (2014).
Ellis, M. O. A. et al. The LandauLifshitz equation in atomistic models. Low Temp. Phys. 41, 705–712. https://doi.org/10.1063/1.4930971 (2015).
Jenkins, S. et al. Magnetic stray fields in nanoscale magnetic tunnel junctions. J. Phys. D Appl. Phys.https://doi.org/10.1088/13616463/ab4fbf (2019).
Bowden, G. J., Stenning, G. B. G. & van der Laan, G. Inter and intra macrocell model for point dipoledipole energy calculations. J. Phys. Condens. Matter 28, 066001 https://doi.org/10.1088/09538984/28/6/066001 (2016).
Lyberatos, A., Berkov, D. & Chantrell, R. A method for the numericalsimulation of the thermal magnetization fluctuations in micromagnetics. J. Phys. Condens. Matter 5, 8911–8920 (1993).
Chureemart, P., D’Amico, I. & Chantrell, R. W. Model of spin accumulation and spin torque in spatially varying magnetisation structures: Limitations of the micromagnetic approach. J. Phys. C 27, 146004 https://doi.org/10.1088/09538984/27/14/146004 (2015).
Zhang, S., Levy, P. M. & Fert, A. Mechanisms of spinpolarized currentdriven magnetization switching. Phys. Rev. Lett. 88, 236601 https://doi.org/10.1103/PhysRevLett.88.236601 (2002).
Slonczewski, J. Currentdriven excitation of magnetic multilayers. J. Magn. Magn. Mater. 159, L1–L7. https://doi.org/10.1016/03048853(96)000625 (1996).
Sato, H. et al. Temperaturedependent properties of CoFeB/MgO thin films: Experiments versus simulations. Phys. Rev. B 98, 214428 https://doi.org/10.1103/PhysRevB.98.214428 (2018).
Devolder, T. et al. Damping of CoxFe80xB20 ultrathin films with perpendicular magnetic anisotropy. Appl. Phys. Lett. 102, 022407 https://doi.org/10.1063/1.4775684 (2013).
Liu, L. et al. Spintorque switching with the giant spin Hall effect of tantalum. Science 336, 555–558. https://doi.org/10.1126/science.1218197 (2012).
Althammer, M. et al. Quantitative study of the spin Hall magnetoresistance in ferromagnetic insulator/normal metal hybrids. Phys. Rev. B Condens. Matter Mater. Phys. 87, 1–15. https://doi.org/10.1103/PhysRevB.87.224401 (2013).
Weiler, M. et al. Experimental test of the spin mixing interface conductivity concept. Phys. Rev. Lett. 111, 1–5. https://doi.org/10.1103/PhysRevLett.111.176601 (2013).
Acknowledgements
P.C. and J.C. gratefully acknowledge the financial support from Thailand Science Research and Innovation (TSRI). This project was undertaken on the Viking Cluster, which is a high performance compute facility provided by the University of York. We are grateful for computational support from the University of York High Performance Computing service, Viking and the Research Computing team.
Author information
Authors and Affiliations
Contributions
A.M., J.C., R.W.C. and P.C. conceived the project, A.M. performed the simulations and analysed the results. All authors contributed to the writing of the manuscript.
Corresponding authors
Ethics declarations
Competing interests
The authors declare no competing interests.
Additional information
Publisher's note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Meo, A., Chureemart, J., Chantrell, R.W. et al. Magnetisation switching dynamics induced by combination of spin transfer torque and spin orbit torque. Sci Rep 12, 3380 (2022). https://doi.org/10.1038/s41598022072772
Received:
Accepted:
Published:
DOI: https://doi.org/10.1038/s41598022072772
This article is cited by

Magnetostrain effects in 2D ferromagnetic van der Waal material CrGeTe$$_3$$
Scientific Reports (2023)

Probing spin dynamics of ultrathin van der Waals magnets via photonmagnon coupling
Nature Communications (2023)
Comments
By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.