Spin orbit effects in CoFeB/MgO hetereostructures with heavy metal underlayers

We study effects originating from the strong spin orbit coupling in CoFeB/MgO heterostructures with heavy metal (HM) underlayers. The perpendicular magnetic anisotropy at the CoFeB/MgO interface, the spin Hall angle of the heavy metal layer, current induced torques and the Dzyaloshinskii-Moriya interaction at the HM/CoFeB interfaces are studied for films in which the early 5d transition metals are used as the HM underlayer. We show how the choice of the HM layer influences these intricate spin orbit effects that emerge within the bulk and at interfaces of the heterostructures.


Introduction
Spin orbit coupling plays an essential role in modern spintronics 1, 2 . The spin Hall effect 3 and the Rashba-Edelstein effect 4,5 , which enable spin current generation and/or spin accumulation, originate from the strong spin orbit coupling within the bulk of a heavy metal layer or at interfaces of different materials. The accumulated spins can diffuse into neighboring magnetic layer(s) to exert torque on the magnetic moments [6][7][8] .
At interfaces, spin orbit coupling is responsible for the perpendicular magnetic anisotropy 30,31 , which is essential for developing densely packed nanoscale magnetic elements. Technologically, the finding of perpendicular magnetic anisotropy at the CoFeB/MgO interface 30,31 was particularly critical owing to the large tunnel magnetoresistance [32][33][34][35] the structure exhibits. In addition, recent reports have shown that the exchange coupling can also be modified by strong spin orbit coupling at interfaces. The Dzyaloshinskii-Moriya interaction 36,37 , an anti-symmetric exchange interaction which favors orthogonal alignment of neighboring magnetic moments, develops if the interface contains materials with strong spin orbit coupling 38 . In addition to the broken structural inversion symmetry of the system, strong interfacial DMI are responsible for the emergence of chiral magnetic structures, such as chiral domain walls [27][28][29][39][40][41][42][43][44][45][46] , spin spirals 47 and skyrmions [48][49][50][51][52] .
Here we review and summarize our recent experimental results on the spin orbit effects in heavy metal (HM)/CoFeB/MgO heterostructures. Perpendicular magnetic anisotropy, the spin Hall effect, current induced torques and the Dzyaloshinskii-Moriya exchange interaction in heterostructures with different HM layers are studied. As these spin orbit effects are complementary and interrelated, we discuss individual effects as well as their correlations to provide a comprehensive view on this new frontier of Spintronics.

Experimental results
Films are deposited at room temperature using magnetron sputtering on Si (001) substrates coated with ~100 nm thick SiO 2 . The film stack is: Si-sub|d SEED Ta|d N HM|t F Co 20 Fe 60 B 20 |2 MgO|1 Ta (units in nanometers) where HM is one of the following transition metals: Hf, Ta, TaN, W and Re. To promote smooth growth of the HM layer, a Ta seed layer (thickness d SEED ) is occasionally formed before deposition of the HM layer.
Reactive sputtering is used to form TaN: N 2 gas is added into the Ar gas atmosphere during the sputtering of Ta. The ratio (Q) of N 2 and Ar gas flows, measured using a mass flow meter attached to each gas line, is used to represent the N 2 concentration of the sputtering gas atmosphere. Q is varied from 0, corresponding to pure Ta, to ~9% (see Ref. [ 53 ] for details). The resulting nitrogen composition of the film, evaluated using Rutherford backscattering spectroscopy (RBS), ranges between ~52 at% (Q~0.7%) to 62 at% (Q~9%). All films are post-annealed at 300 ˚C for one hour in vacuum. No magnetic field is applied during the annealing process.
Magnetic moment (M) at saturation and the effective magnetic anisotropy energy (K EFF ) are measured at room temperature using vibrating sample magnetometry (VSM).
K EFF is estimated from the areal difference between the out of plane and in-plane magnetization hysteresis loops. Positive K EFF corresponds to magnetic easy axis directed along the film normal.
Throughout this paper, we use the coordinate system defined as the following.
Positive (negative) current flows along +x (−x), the film normal is parallel to the z axis and a right handed coordinate system is employed.
Volume (V) of the film is calculated using the nominal thickness of the CoFeB layer (t F ) and the deposition area (A) of the film defined by the opening of the shadow mask used during the deposition process, i.e. V=t F ·A. Note that M/V differs from the saturation magnetization (M S ) when a magnetic dead layer [53][54][55] forms within the magnetic (CoFeB) layer or when proximity induced magnetization (PIM) appears in the neighboring layer [56][57][58] . For the films studied here, we find evidence of non-zero magnetic dead layer thickness in many film structures but no indication of the PIM. A similar phase transition is also found for HM=Re (Ref. [ 62 ]). Re undergoes a transition from a thin amorphous-like phase to a thick highly textured hexagonal closed packed (hcp) phase at d N~6 nm where changes in M/V and K EFF are found. The highly textured hcp phase appears at d N~6 nm for films with and without the Ta seed layer.
From transmission electron microscopy studies 62 , we also find that the surface roughness of the Re underlayer is much larger than the other elements: this is likely to do with the poor wetting of Re on SiO 2 . It turns out that inserting the Ta seed layer can significantly improve the surface roughness 62 , thus improving values of M/V and K EFF at small d N . For the thick Re underlayer films, the difference in M/V and K EFF between the films with and without the Ta seed layer becomes negligible. With regard to the size of PMA, we find the Re underlayer films with the Ta seed layer shows larger K EFF than that of the thicker Re underlayer films, indicating that the thin amorphous-like phase serves as a better underlayer in promoting large K EFF against the highly textured hcp phase. This trend is similar to what has been observed for the W underlayer films.
The parameters that characterize the magnetic properties of the films are summarized in Fig. 2. We fit the t F dependence of M/A with a linear function to obtain the saturation magnetization M S and the magnetic dead layer thickness t D from the slope and the x-axis intercept of the linear function, respectively 53  becomes the largest when HM=Ta. This is consistent with earlier studies in which Ta is found to absorb boron from CoFeB [64][65][66][67] . There is also a possibility that the CoFeB layer intermixes with the neighboring HM layer, which will likely result in degradation of M S .
However, for most cases, M S does not fall below that of bulk CoFeB, indicating that the degree of intermixing is not large. This is also supported by elemental mapping studies using transmission electron microscopy (unpublished results).
The magnetic dead layer thickness (Figs. 2(f)-2(j)) shows a relatively large variation with d N . We infer that the dead layer forms due to structural and/or chemical intermixing that occur at the HM/CoFeB interface (if this intermixing extends into the CoFeB layer, one would expect to see a reduced M S ). As the thickness of the dead layer is less than half a nanometer in most cases, it is difficult to identify the constituting elements. Recent studies 66 have shown that a thin TaB layer forms at the interface between Ta and CoFeB.
It is yet to be confirmed whether the diffusion of boron is responsible for the formation of the dead layer for films with other HM underlayers. The largest K I of ~1.8 erg/cm 2 is found when HM=TaN 53 . To obtain a large K I in CoFeB/MgO based structures, it is considered important to remove the boron from the CoFeB/MgO interface and promote the Fe-oxygen bonding that is essential for establishing perpendicular magnetic anisotropy 58,[68][69][70] . The HM underlayer has to enable boron diffusion from the CoFeB layer; however too much diffusion may promote intermixing of HM and CoFeB, which will result in an increase of t D (and possible degradation of M S ). We infer that Ta 48 N 52 (Q~0.7%) is at the right balance of promoting boron diffusion and simultaneously limiting the intermixing. Note that when more nitrogen is added to TaN, t D becomes nearly zero owing to the limited diffusion process (TaN is known to be a good diffusion barrier 71 ). However, since the boron diffusion process also becomes limited, K I is reduced to nearly half of that shown in Fig. 2(m) (see Ref. 53 ). For HM=W and Re, Fig. 2(d) and 2(e) indicate that boron mostly remains within the CoFeB layer as M S is close to that of bulk CoFeB. As a consequence, K I is smaller than that of Ta and TaN.

Spin Hall effect
We use the spin Hall magnetoresistance 72 we find that an amorphous(-like) phase is the dominant phase when the HM layer is thin.
The resistivity of the amorphous-like phase for the four elements is shown in Fig. 4(a).
To estimate the spin Hall angle, we using a drift-diffusion model 75 to account for the SMR: 4(c) for the amorphous-like phase. If we were to take into account the effect of the longitudinal spin absorption 78 , the spin Hall angle estimation will increase by ~10-20%.
The spin Hall conductivity ( SH ) is calculated using the following phenomenological relation:  SH is plotted for the four elements in Fig. 4(d), which indicate that  SH is strongly influenced by the number of 5d electrons of the HM layer. The largest  SH is found in W albeit its resistivity is the smallest among the four elements studied. The size of  SH is similar to that of Pt reported previously 79 . The underlying mechanism of the spin Hall effect [80][81][82][83][84][85][86] in the amorphous 5d transition metals, whether its origin is intrinsic or extrinsic, remains elusive. These results, however, indicate that the element dependent spin orbit coupling plays an important role in defining spin dependent transport and the spin Hall effect.

Current induced torque
Current induced torque that arises in thin film heterostructures due to application of in-plane current can be modeled using the framework developed for describing spin transfer torque 77, 87-89 in spin valve nanopillars and magnetic tunnel junctions. The Landau-Lifshitz-Gilbert (LLG) equation that includes both the damping-like and fieldlike spin transfer torques reads: where  is the Gilbert damping constant,  is the gyromagnetic ratio,  Results from the spin Hall magnetoresistance measurements shown in Fig. 3(a) indicate that the spin Hall angle of the amorphous-like and hcp phases of Hf is similar. However, the effective field is smaller for the amorphous-like phase than that of the hcp phase: Fig.   5(a) and 5(e) shows that the effective field is nearly zero for the amorphous-like phase (d N ≲ 2 nm). This is partly due to the difference in the CoFeB magnetization ( Fig. 1(a) Finally we note that there remains an uncertainty in obtaining the current induced effective field using the harmonic Hall voltage measurements, which has been discussed in Ref. 29 . In general, when the planar Hall-like resistance becomes comparable to the anomalous Hall resistance due to the large contribution from the SMR, we find anomalies in the effective field. Such anomaly has been found in W underlayer films. To date, the origin of the anomaly is not clear. We have thus limited discussion of the effective field in systems which possess small planar Hall resistance. (The damping-like effective field can be estimated using the spin Hall angle, obtained for example by the SMR measurements, and the saturation magnetization; see Refs. 10, 103 ). Since ̂ is orthogonal to the current flow, must be pointing along the current flow to achieve non-zero efficiency. For wide wires (i.e. typically the width larger than 100 nm) the magneto-statically preferred domain wall magnetization in perpendicularly magnetized films is the Bloch type 106  We have studied the magnitude and sign of the DM interaction at the HM/CoFeB interface using current induced motion of domain walls. In Fig. 6, we show representative results from HM=W. The CoFeB thickness dependence of the DM interaction is studied here. The magnetization per unit volume M/V and the effective magnetic anisotropy energy K EFF are plotted as a function of the CoFeB layer thickness (t F ) in Figs. 6(a) and

The interface Dzyaloshinskii-Moriya interaction
6(b). The magnetic easy axis points along the film normal when 0.8 nm ≲ t F ≲ 1.2 nm.
The t F dependence of the propagation field (H P ) needed to induce motion of domain walls with the out of plane field is shown in Fig. 6(c). Although the variation of H P with t F is not large, the trend is similar to that of K EFF , indicating that the domain wall width meV. This value is smaller than that reported 39 for W(001)/Fe, D I =1.4 meV.
We find that the DM exchange constant strongly depends on the HM material used to interface the CoFeB layer. Figure 7 shows the average DM exchange constant D of HM/CoFeB/MgO heterostructures with different HM layers 29,114 . D is negative for Hf, nearly zero for Ta, becomes positive for TaN and W. Note that the D shown in Fig. 7 is for the thicker hcp phase of Hf, whereas the other three materials (Ta, TaN and W) are predominantly amorphous. We have also studied current induced motion of domain walls in Re/CoFeB with the Ta seed layer, however, we find domain walls hardly move with current, suggesting that D is nearly zero since the damping-like torque is similar in magnitude with that of films with HM=Ta.
The results shown in Fig. 7 indicate that the DM exchange constant D depends on the number of 5d electrons, suggesting an electronic origin. As the change in the electronegativity with the number of 5d electrons is similar to that of D shown in Fig. 7, we infer that the relative difference between the electronegativity of the HM and CoFeB layer may play a role in defining the DM interaction. With regard to the origin of DMI 115,116 , it has been suggested that proximity induced magnetism (PIM) is responsible for the large D at the Pt/Co interface 117 . Here we find little evidence of PIM in the HM/CoFe/MgO (HM=Hf, Ta, TaN, W, Re) heterostructures studied. As the size of D of Pt/Co interface is at least a few times larger than that of HM/CoFeB, it remains to be seen whether PIM enhances D of other systems.
It is well established that the DM exchange constant can be significantly modified by the structure (texture) of the interface 39,118 . Whereas Pt/Co films with large D typically possesses the fcc structure with (111) plane facing the interface 44, 119 , the CoFeB in HM/CoFeB/MgO heterostructures is predominantly amorphous and many of the HM layers, including W which shows the largest D, is amorphous-like (the only exception is the thicker hcp Hf with which we find non-zero D), suggesting that contribution from the structure on D may be small in this system. It should be noted that we find a non-zero thickness of a magnetic dead layer for the heterostructures with HM=W (see Fig. 2(i)).
Since the constituents of the dead layer are not known at the moment, it is difficult to evaluate the influence of the dead layer on the DM interaction.
Finally, it has been reported recently that in Ta/CoFeB/MgO, the amount of boron present at the interface may influence D via modification of local atomic configuration 120 .
As the boron diffusion significantly depends on the material used for the HM layer (unpublished results), it is difficult to determine the degree of such contribution to the DMI for all the structures studied here. Further investigation is required to clarify quantitatively the origin of DMI at the HM/CoFeB interface.

Summary
We Identifying the chemical and structural phase of the interface, in particular, its relation to the formation of the magnetic dead layer, is particularly important to describe the origin of the DM interaction in this system.
With the engineering of the film stack and materials innovation, it is very likely that new insights into spin orbit effects will continue to unveil for the coming years.