Spin–orbit torque driven perpendicular magnetization switching in Re/CoFeB/MgO with high thermal stability

Perpendicular magnetic tunnel junction (p-MTJ)-based magnetic random-access memory (MRAM) is the front-runner of the next generation non-volatile memory due to its robust magnetization states, which are undisturbed by external perturbation except for a strong magnetic field.^1–3 Manipulation of magnetization in p-MTJ using electric current [either spin transfer torque (STT) or spin–orbit torque (SOT)] has already pushed the rise of p-MTJ MRAMs into commercial productions.^2–5 Producing commercial MRAM products with high density requires the free layer of the MTJ to keep perpendicular magnetic anisotropy (PMA) after annealing up to a temperature of at least 400 °C, given that the back-end semiconductor processing is performed at a temperature of around 400 °C.^6,7 Thus, finding material systems with thermally stable PMA is of supreme importance. So far, heavy metal (HM)/CoFeB/MgO-based p-MTJs have been commonly used due to their sufficiently high tunneling magnetoresistance (TMR) value.^8–11 Although a range of materials, including Ta, Mo, W, Hf, and Pd, have been shown to have PMA, only Mo and β-W can still reveal PMA after annealing at 400 °C or higher temperature.^6,7 The origin of PMA in HM/CoFeB/MgO is believed to be mainly originating from the Fe–O orbital hybridization at the CoFeB/MgO interface and boron (B) out-diffusion.^6,7,12–15 Furthermore, the property of HM is another crucial factor affecting anisotropy and TMR, which needs to promote the CoFeB crystallization along MgO [002].^16–19 Thermal stability of CoFeB against the annealing temperature of 400 °C is shown to depend strongly on the crystal structure and oxygen affinity of HM metals.^6,7 It has been reported that in the case of Ta as the HM layer, the formation enthalpy of Ta oxide is comparable to that of MgO, which may lead to diffusion of Ta toward the CoFeB/MgO interface during high temperature annealing and thus causes severe dead layer formation.^6,7 Moreover, thin Ta is an amorphous layer, which is a high energy state as compared to the crystalline ones, leading to further enhancement of Ta diffusion into CoFeB. Severe Ta diffusion destroys the PMA.^6,7

Mo and β-W crystallize into a low energy BCC structure at the as-deposited state and also have lower formation enthalpy for their oxides as compared to that of Ta oxides. So, they facilitate the growth of MgO along the [002] direction during the deposition and allow CoFeB to crystallize along MgO [002] during the annealing process.^6,7 Because of their low-energy stable BCC phase and lower oxygen affinity than that of Mg, they can suppress interdiffusion during high temperature annealing and still keep strong PMA.^6,7 Moreover, W and Mo are well-known refractory metals with high melting points, which might be another factor to slow down the diffusion.

One of the critical factors for MRAM cells is thermal stability ∆ = K_effV/K_BT, where K_eff is the effective magnetic anisotropy energy density, V is the volume of the magnetic free layer, K_B is Boltzmann’s constant, and T is the absolute temperature.^20–22 In addition to enhancing K_eff by the underlayer, another straightforward way to increase ∆ is to increase the volume of the free layer, so this is where the concept of composite free layers comes in. Two CoFeB layers separated by a very thin HM layer have been reported to substantially increase the volume of the free layer and, thus, the thermal stability.^20–24 In addition to promoting perpendicular anisotropy, the HM may also need to play a role as the spin current source for writing in the SOT-MRAM.^25–27 Writing using the three terminal SOT, which utilizes the spin hall effect (SHE) and/or the Rashba effect, provides protection for the insulator against the thin oxide barrier (MgO) breakdown due to the separate write/read paths.^25–27 In addition, SOT writing may provide much faster switching than the STT writing.^25–27 So, having material systems, which can simultaneously provide SHE and thermally stable PMA against high temperature annealing, is important for the development of commercial SOT-MRAMs. Currently, β-W is the only HM serving well for these purposes.^28,29 The switching current density can be reduced to 1.1 × 10¹¹ A/m²²⁸ However, achieving a highly resistive β phase may pose a different challenge for mass production.^28,29 Moreover, very high resistivity of β-W (∼250 µΩ cm) may further lead to strong Joule heating during writing.

Here, we report another HM, Re, which provides both a thermally stable perpendicular CoFeB free layer with perpendicular anisotropy field H_K larger than 5000 Oe after 400 °C annealing and SOT switching of the perpendicular CoFeB with a critical switching current density of 1.36 × 10¹¹ A/m² under the in-plane field H_X of 50 Oe. Unlike W,^28,29 Re does not require a lot of tuning of deposition conditions to achieve a specific phase to have PMA and SHE. Moreover, a perpendicular composite free layer with thin Re as the spacer separating two CoFeB layers is demonstrated to have double volume of the ferromagnet (FM) and even higher H_K of 6000 Oe than that of a single FM. To further quantize the SHE in the Re/CoFeB/MgO structure, we employed a loop shift method^30–32 to determine the spin hall angle of Re to be 0.065 ± 0.003. Our results provide an additional material option for SOT-MRAM application.

Si/SiO₂/Re(5)/Co₂₀Fe₆₀B₂₀(0.9–1.5)/MgO(2)/Ta(3.5) (in nm) samples were deposited to study the dependence of PMA on CoFeB thickness. Si/SiO₂/Re(t)/CoFeB(1.1)/MgO(2)/Ta(3.5) samples with t = 1.1–10 nm were deposited to investigate how the Re thickness affects the PMA properties. Si/SiO₂/Ta(0.4)/Re(t)/CoFeB(1.07)/MgO(2)/Ta(3.5) samples with t = 1.1–10 nm were studied to check the effects of the Ta seed layer on the magnetic properties of CoFeB. Finally, the composite free layer structures Si/SiO₂/Ta(3.5)/MgO(1.5)/CoFeB(1.1)/Re(t)/CoFeB(1.1)/MgO(1.5)/Ta(3.5) were deposited. The base pressure of the sputtering system was 1.0 × 10⁻⁷ Torr, and the metallic layers were deposited using DC magnetron sputtering, and MgO was deposited using RF sputtering. All the samples were annealed at 300, 350, 400, and 425 °C under pressure lower than 1 × 10⁻⁵ Torr for 30 min. Magnetic properties were studied using the vibrating sample magnetometer (VSM). X-ray diffraction (XRD) analysis was performed with the x-ray wavelength of 0.254 nm. To study the SOT properties, we used photolithography and Ar ion milling to fabricate the typical Hall cross structure. The width and length of the current channel are 10 and 40 µm, respectively, while those of the voltage channel are 3 and 40 µm, respectively.

For annealing temperatures at 350 °C or higher, the samples Si/SiO₂/Re(5)/Co₂₀Fe₆₀B₂₀(t)/MgO(2)/Ta(3.5) (in nm) in Fig. 1(a) with the CoFeB thickness between 0.9 and 1.4 nm all exhibited perpendicular magnetic anisotropy. The variations in H_K with the CoFeB thickness after annealing at 400 °C are shown in Fig. 1(b). The initial increase in H_K with the thickness may be due to the existence of the dead layer. When the thickness further increases, H_K is reduced, reflecting the perpendicular anisotropy that originates from the interface. The VSM measurements for the samples Re(5 nm)/CoFeB(1.1)/MgO annealed at 400 °C are shown in Fig. 1(c). H_K can be as high as 5500 Oe after annealing at 350 °C and then gradually decreases with a further increase in the annealing temperature. Good PMA with H_K of 4500 Oe can still be obtained after annealing at 425 °C [Fig. 1(d)]. Other properties including the dead layer thickness (t_dead), saturation magnetization (M_S), and interfacial anisotropy (K_S) of samples annealed at 400 °C are shown in Figs. 2(a) and 2(b). t_dead and M_S were calculated from the plot M_St_CoFeB vs t_CoFeB, as indicated in Fig. 2(a) for 400 °C. M_S and t_dead for the as-deposited samples were 939 emu/cc and 0.51 nm, respectively. The higher M_S after annealing is due to the crystallization of CoFeB and B out-diffusion.

K_S was calculated using the equation K_eff = (K_b − 2πM_S²) + K_S/t_CoFeB, where volume anisotropy (K_b) is assumed to be negligible, and K_S at different temperatures was calculated from the plot of K_eff × t_CoFeB vs t_CoFeB, as shown in Fig. 2(b) for 400 °C annealing as an example. Finally, the temperature dependences of t_dead, M_S, and K_S are shown in Figs. 2(c)–2(e), respectively. K_S is about 1.98, 1.73, and 1.62 erg/cm² for samples annealed at 350, 400, and 425 °C, respectively [Fig. 2(e)]. K_S is reported to be 1.7 and 2.05 erg/cm² for W and Mo buffers, respectively, after 400 °C annealing.^6,7 Our results of H_K, M_S, and t_dead are comparable to those reported in the case of W/CoFeB/MgO after 400 °C annealing.⁶ 

Next, the Re thickness effect on the magnetic properties was studied. The hysteresis loops measured by the VSM of the samples Si/SiO₂/Re(t)/CoFeB(1.1)/MgO(2)/Ta(3.5) in Fig. 3(a) revealed that the PMA only existed with the Re thickness ranging between 4 and 8 nm after annealing. For very thin Re (3 nm or less) samples, both in-plane and out-of-plane hysteresis curves look like S-shaped hard axes; the MH loop for 400 °C annealed Re(1.1)/CoFeB(1.1)/MgO(2)/Ta(5) sample is shown in Fig. 3(b) as an example. H_K reduced significantly for samples with the Re thickness of 9 nm or higher. The H_K dependence on the Re thickness is shown in Fig. 3(c) 

To further understand why the thin Re layer cannot provide good PMA, we deposited a very thin Ta with the thickness of 0.4 nm as a seed layer for Re and revealed that even very thin Re (1.1–3 nm) can have thermally stable PMA with H_K close to 5000 Oe after 400 °C annealing. The MH curve for 1.1 nm Re with the 0.4 nm Ta seed layer is shown in Fig. 3(d). However, for the samples with thicker Re (9 nm or higher), the perpendicular anisotropy is still poor even with the Ta seed layer.

To test if Re can be used for the interlayer of the composite free layer, we prepared a composite free layer with the structures Si/SiO₂/Ta(3.5)/MgO(1.5)/CoFeB(1.1)/Re(t)/CoFeB(1.1)/MgO(1.5)/Ta(3.5) with t = 0.3–1 nm, as shown in Fig. 4(a). All samples show PMA and ferromagnetic coupling between the two CoFeB layers in this Re interlayer thickness range after 400 °C annealing. We obtained maximum net magnetization for the Re thickness of 0.47 nm [Fig. 4(b)] with nearly double net magnetization and slightly higher H_K of ∼6000 Oe than a single CoFeB. The higher H_K is possibly due to the double Fe–O interface.

Next, SOT studies were done by fabricating the typical hall cross geometry from the samples Re(t)/CoFeB(1.1)/MgO(2)/Ta(5), t = 1.1–5 nm, and a Ta seed layer of 0.4 nm was used to promote PMA for t = 1.1–3 nm. The SOT measurement setup and spin current generation due to SHE are shown in Fig. 5(a) with 5 nm thick Re. We applied 0.1 ms current pulses with an increase in the magnitude to achieve SOT switching. As shown in Fig. 5(b), we achieved full magnetization switching under H_X = 50 Oe with a critical current density of 3 × 10¹¹ A/m² for Re(5)/CoFeB(1.1)/MgO(2)/Ta(5). The critical current density was calculated by considering the proportion of the current flowing into Re. We also performed the H_X field dependence of SOT switching by applying H_X up to 800 Oe, as shown in Fig. 5(c). Under this field range, the critical current density is slightly reduced with an increase in H_X, which is a typical SOT switching behavior.

Based on the switching polarity, we may know that the spin hall angle of Re has the same sign as W, that is, it is negative. Finally, we investigated how the Re thickness affected SOT switching. We found that even 1.1 nm Re could switch CoFeB. The dependence of the critical current density J_C on the Re thickness is shown in Fig. 5(d). J_C is slightly increased with an increase in the Re thickness and ranges from 1.36 × 10¹¹ to 3 × 10¹¹ A/m².

To further quantify the SOT behavior of Re, we performed the loop shift measurement to calculate the damping-like (DL) efficiency (ξ_DL) of Re.^30–32 The measurement setup is shown in Fig. 6(a). The damping-like (DL) effective field generated due to SHE in heavy metals can be obtained by applying a series of different magnitudes of DC current pulses while sweeping H_Z under different H_X fields [Fig. 6(b)]. From the loop shift, we can calculate the DL efficiency using the formula

where $s=tFMρHMtHMρFM$⁠. Here, e is the elementary charge, $ℏ$ is reduced Planck’s constant, u₀ is the permeability of vacuum, M_S is the saturation magnetization of the ferromagnet, t_FM is the thickness of the ferromagnet, w is the width of the current channel, t_HM is the thickness of the HM, and $HzEffI$ is the saturated effective field per unit current calculated from the loop shift measurement. Figure 6(b) shows the loop shift corresponding to I_DC = ±9 mA for H_X = 680 Oe, and Fig. 6(c) shows the effective field (net shift) generated by different I_DC applied for H_X = 680 Oe. From the plot in Fig. 6(c), we can calculate the effective field per unit current under different H_X$HzEffI=4Oe/mAatHX=680Oe$⁠. Finally, $HzEffI$ calculated in Fig. 6(c) is plotted as a function of H_X, as shown in Fig. 6(d). The slope will saturate at a given H_X = H_DMI and the saturated value of $HzEffI$ can be used to calculate ξ_DL using the above equation. The calculated ξ_DL was −0.065 ± 0.003. The DL efficiency is comparable to Pt with ξ_DL + 0.07 − +0.12^33,34 but less than that of Ta (∼−0.15)^34,35 and β-W (∼−0.21).²⁸ H_DMI in the Re/CoFeB structure is about 400 Oe based on the measurement shown in Fig. 6(d). ϴ_SH can be calculated from ξ_DL using the equation ξ_DL = T_intϴ_SH, where T_int is the interfacial spin transparency. When T_int is assumed to be unity, the ϴ_SH is equal to −0.065 ± 0.003. In reality, the T_int is typically less than 1, so this ϴ_SH value can be underestimated.

We believe that the origin of PMA in Re/CoFeB/MgO still mainly comes from the Fe–O orbital hybridization at the CoFeB/MgO interface. The magnetic properties of CoFeB (Ms, dead layer, and H_K) are affected by the interdiffusion between CoFeB and Re. The wide angle XRD and grazing incidence x-ray diffraction (GIXD) analysis of Re, as shown in Figs. 7(a) and 7(b), respectively, reveal that the as-deposited 5 nm Re is amorphous; therefore, during annealing, CoFeB can easily crystallize along the MgO [002], and boron can be absorbed by the Re underlayer. The GIXD measurements shown in Fig. 7(b) reveal that after annealing, 5 nm Re is partially crystallized into the hexagonal close-packed (HCP) crystal structure, which is a stable and lower energy state than the as-deposited amorphous state. Thus, we believe that this crystallization suppresses the interdiffusion between Re and CoFeB during high temperature annealing.

Moreover, after W, Re is the second highest melting-point metal, which could further suppress interdiffusion of Re into CoFeB during high temperature annealing. In fact, a recent report also revealed good thermal stability for Co/Re multilayers.³⁶ Consequently, we can get high M_S and low dead layers after high temperature annealing for Re/CoFeB/MgO. In addition, the affinity between HM and oxygen is also critical. If the affinity of HM-O is high, the HM may have the tendency to diffuse to the MgO side and react with oxygen; therefore, the bonding of Fe–O is reduced, and the PMA is destroyed during high temperature annealing, as observed in the case of Ta.⁷ Regarding the oxygen affinity of Re to form oxides, the enthalpy of the formation of Re oxide is −144 kJ/mol, the least among Mo oxide (−196 kJ/mol), Ta oxide (−292 kJ/mol), and W oxide (−197 kJ/mol).³⁷ Thus, among all these heavy metals, Re is least expected to “steal” oxygen from the CoFeB/MgO interface and least expected to diffuse toward the Fe–O interface, which makes CoFeB sustain its PMA after high temperature annealing.

A thin Re (≦3 nm) layer without the Ta seed layer shows the isotropic behavior of CoFeB for both in-plane and out-of-plane measurements, as shown in Fig. 3(b), but a thin Re layer with the Ta seed layer shows good PMA after high temperature annealing, as shown in Fig. 3(d). It is possible that a very thin Re film is not continuous, leading to loss of PMA similar to the results observed for thin W.⁶ On the other hand, when we use the Ta seed layer, the film becomes rather continuous³⁸ probably because of much better adhesion of Ta on SiO₂, so the PMA can be achieved. Thicker Re does not have good perpendicular anisotropy no matter if the Ta seed layer is used or not. The XRD analysis of the thick Re (10 nm), as shown in Fig. 7, shows that both as-deposited and annealed 10 nm Re films are well crystallized into the HCP structure. The well crystalized Re layer may hinder the B out-diffusion and suppress the crystallization of CoFeB along MgO (002). Another possible reason could be the increase in the roughness of Re with the increase in the thickness, which could further suppress the PMA.

The SOT in Re/CoFeB might be coming from the resistive amorphous/amorphous + HCP state of Re, as it has been shown that the highly resistive phase of HMs such as Ta and β-W possesses higher DL SOT efficiency.^28,29 Moreover, when the Pt resistivity is tuned to a higher value by alloying or tuning deposition conditions, it can show larger ξ_DL^32,39 Our critical current density of 1.34–3 × 10¹¹ A/m² and ξ_DL of −0.065 ± 0.003 are similar to Pt, but ξ_DL is smaller than Ta or W. Slightly lower J_C was observed for a very thin Re that maybe ascribed to larger resistivity of Re at a very thin Re thickness. On the other hand, note that the requirement of H_X in Re/CoFeB/MgO for switching is quite small, which can be ascribed to small H_DMI of ∼400 Oe in the Re/CoFeB structure as compared to Pt/Co or Pt/CoFeB, which have H_DMI of 5000 and 2500 Oe, respectively.³⁰ The H_DMI in Re/CoFeB/MgO is similar to that of typical β-W/CoFeB/MgO (∼300 Oe)⁴⁰ and Ta/CoFeB/MgO (∼250–300 Oe),³⁰ which can also be switched under rather small H_X. Finally, the lower resistivity of Re (less than 100 µemu cm) as compared to that of Ta (∼200–270 µemu cm) and β-W (∼250 µemu cm) may relieve the concern of the Joule heating during writing.

In this paper, we present a new HM rhenium that could generate PMA in CoFeB/MgO systems after high temperature annealing up to 425 °C. With a thin Ta seed layer, even 1.1 nm Re can support the perpendicular CoFeB with H_K close to 5000 Oe. The as-deposited thin Re layer is amorphous, which becomes partially crystallized during annealing. The initial amorphous Re may promote the B out-diffusion, and CoFeB crystallizes along MgO [002]. Later on, the partially crystallized Re and low oxygen affinity of Re may suppress the interdiffusion between CoFeB and Re, which sustains the PMA after high temperature annealing. Furthermore, we demonstrate that Re can be a spin current source for SOT switching with similar damping-like torque efficiency to that of Pt. Unlike β-W, Re does not need to form a specific phase, and it has lower resistivity to avoid Joule heating. So far, only β-W and Re have been demonstrated to simultaneously possess high thermal stability of perpendicular CoFeB up to 400 °C and generate the spin current for SOT switching. It may pave an alternative avenue for the material design of SOT-MRAM.

This work was supported by Solar Applied Materials Technology Corporation on sputtering targets and partially supported by the Ministry of Science and Technology of Taiwan (Grant Nos. 109-2218-E-007-018 and 110-2218-E-007-034). We also acknowledge the help of Professor Chi-Feng Pai, National Taiwan University, Taipei, Taiwan, and his Ph.D. student Yu-Hu Chen for helping out with the loop shift measurements.

The authors declare no conflicts of interest to disclose.

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