Highly Energy‐Efficient Spin‐Orbit‐Torque Magnetoresistive Memory with Amorphous W─Ta─B Alloys

The spin Hall effect enables fast and reliable writing operations for next‐generation spin‐orbit‐torque magnetoresistive random‐access memories (SOT‐MRAMs). To develop SOT‐MRAMs; however, the spin Hall material should have a sufficiently low writing energy and high annealing stability for the semiconductor integration process. Thus far, none of the crystalline‐based spin Hall materials are able to satisfy these requirements. Here, a promising solution for SOT‐MRAMs is provided using amorphous W─Ta─B alloys. Even without a long‐range crystal order, W─Ta─B alloys exhibit both large effective spin Hall angles up to 40% derived from a Ta substitutional doping and superior annealing stability (up to 400 °C) due to the addition of B, enabling them to satisfy both requirements. Nanoscale three‐terminal SOT‐MRAM cells are fabricated, and these are demonstrated to have high magnetoresistance ratios (up to 130%) and extremely low intrinsic switching current densities (down to 4 × 106 A cm−2). These results show that amorphous spin Hall materials can provide the key for realizing high‐performance SOT‐MRAMs.


Introduction
[3][4][5] The current main-stream technology for MRAMs is the spin-transfer-torque MRAM (STT-MRAM), in which both the current paths for writing and reading information are passed through magnetic tunnel junctions (MTJs).Although STT-MRAMs have been commercialized as embedded nonvolatile memories, it is difficult to apply them to the high-speed cache memories of large-scale integrated (LSI) systems.This is because the large current density required for high-speed writing gradually damages the MTJ devices resulting in deterioration DOI: 10.1002/aelm.202300581 of the writing endurance.Moreover, high-speed read-out is also difficult due to interference between the read and write currents that pass through the same path.8][9][10][11][12][13][14][15] The memory cells of SOT-MRAMs have a three-terminal structure consisting of a MTJ and a wire of material with the spin Hall effect, here after spin Hall material, adjacent to the magnetic free layer of the MTJ.The spin current produced by the spin Hall effect in the spin Hall material is injected into the free layer of the MTJ and induces magnetization switching via spin-orbit torques.Due to its unique memory cell architecture, in which the current paths for readout and write operations are separated, the write current does not pass through the MTJ.Therefore, compared to STT-MRAMs, SOT-MRAMs have the advantage of faster write and read-out operations with higher endurance. [2,5]These features make SOT-MRAMs suitable for high-speed embedded memory applications such as cache memory.
The major challenge for developing SOT-MRAMs is the realization of spin Hall materials that can satisfy several essential requirements.The first requirement is reducing the switching current density.Depending on the resistance of the channel for the writing operation (spin Hall channel), and assuming the writing duration to be 10 ns, the intrinsic switching current density j c0 should be <1 × 10 7 A cm −2 to obtain writing energy <100 fJ per bit, which is equivalent to that of advanced STT-MRAMs. [5]To satisfy this requirement, the most important factor is that the spin Hall material should have a large spin Hall angle, which is defined as the ratio between the generated spin current density and the applied charge current density.Typical candidate materials with large spin Hall angles are Pt (or its binary alloys), [16][17][18] -phase W (-W), [19][20][21][22][23] -phase Ta (-Ta), [6] and topological insulators. [24,25]All these materials have polycrystalline (textured) or single-crystal structures, in which a large spin Hall effect is theoretically expected by an intrinsic mechanism based on the band structure of materials with long-range crystal order [26][27][28][29] .As a result of extensive experimental studies, some such materials exhibiting large spin Hall angles have been observed, and with these j c0 < 1 × 10 7 A cm −2 could be achieved [30][31][32][33][34][35] .The second essential requirement is the high annealing stability of the spin Hall material.To integrate SOT-MRAMs in LSI systems, the spin Hall channel and adjacent MTJs are annealed at 350-400 °C during the back-end-of-line process, which often degrades memory cells.The difficult point here is that the annealing stability of -W, which is a typical crystalline spin Hall material used in SOT-MRAMs, is relatively low (≈300 ˚C) due to its metastable structure.High temperature annealing transforms the structure into a bcc-structure also known as -phase W (-W), which results in a significantly reduced spin Hall angle. [19,20,23,36]In addition to the two requirements mentioned above, another important requirement is the compatibility with MgO-based MTJs in order that SOT-MRAMs to have a high magnetoresistance (MR) ratio (>100% for high-speed read-out) and a large thermal stability factor Δ ≥ 45k B T for high-speed cache memory (k B is the Boltzmann constant, and T is the system temperature assuming room temperature).Crystalline spin Hall materials with fcc or hcp structures, such as Pt, are not compatible with MgObased MTJs due to the mismatch in crystalline symmetry.Therefore, realizing spin Hall materials that simultaneously satisfy the above requirements is the biggest challenge in developing SOT-MRAMs, and no definitive solution has been obtained so far.
In this article, we report a new promising solution, using an amorphous phased alloy as a spin Hall material, that satisfies all of the above-mentioned requirements.Here, the amorphous phase is defined as either a completely random atomic arrangement or a state with short-range ordering extending over several atomic distances (≈1-2 nm), which is often referred to as nanocrystalline.Although several amorphous spin Hall materials have been reported so far, [37][38][39][40] they have not received much attention due to the following two reasons.First, the intrinsic contribution of the spin Hall effect is often considered to be illdefined in materials without a long-range crystal order such as amorphous phased materials.Second, the annealing stability of amorphous alloys is usually poorer than that of crystalline alloys because crystallization of the amorphous structure takes place at relatively low temperatures.In this article, we show that amorphous phased W─Ta─B alloys with short-range order exhibit an effective spin Hall angle up to 40% and high thermal annealing stability up to 400 °C.We reveal that alloying with B and Ta plays important roles in stabilizing the amorphous phase with a shortrange crystal order and enhancing the intrinsic spin Hall effect in W via substitutional doping.Using these amorphous alloys, we fabricated nano-sized three-terminal SOT-MRAM cells with very low j c0 (down to 4 × 10 6 A cm −2 ), excellent annealing stability, high MR ratios (up to 130%) and sufficiently high thermal stability factors for cache-memory applications.

Thermally Robust Amorphous W─Ta─B Alloys
First, we studied W─Ta─B spin Hall materials by using the following multilayer films deposited on thermally oxidized Si substrates: Ta (1)/W 100-x (Ta 82 B 18 ) x (t)/Co─Fe─B(3)/MgO/Ta-O (thickness in nm), here W 100-x (Ta 82 B 18 ) x (hereafter designated W─Ta─B) is the spin Hall material layer, and Co─Fe─B corresponds to the magnetic free layer for that the composition is shown in Experimental Section.All films were post-annealed at 350 or 400 °C for an hour under vacuum.Figure 1a shows X-ray diffraction (XRD) patterns of t = 10 nm thick W (x = 0), W─Ta─B (x = 32.1), and Ta-B (x = 100) films post annealed at 350 °C.In the case of the W film, a clear XRD peak near 40°is observed that corresponds to the diffraction peak of -W (110).In the case of the W─Ta─B and Ta-B films, no distinct XRD peaks can be seen, however, a broad hump ≈40°can be seen (inset of Figure 1a).This behavior was observed even after post-annealing at 400 °C for the W─Ta─B film (Figure 1b).Using Scherrer's formula, the grain sizes of the W─Ta─B and Ta-B films were determined to be 1.3 ± 0.1 nm (Supporting Information).We interpret this behavior as indicating the presence of an amorphous phase with short-range order in the W─Ta─B layer extending over 4-5 interatomic distances.From the position of the broad humps, this short-range order is inferred to be a local atomic arrangement of either -W-type or -W-type.The cross-sectional high-resolution transmission-electron-microscopy (TEM) image (Figure 1c) shows a maze-like pattern and halo patterns of nanobeam diffraction from the W─Ta─B layer (Figure 1d) were observed.All these results provide further confirmation that the amorphous phase is dominant in the W─Ta─B layer.Energy Dispersive X-ray (EDX) mapping shows a uniform distribution of Ta within the W─Ta─B layer (Figure 1e).These results show that the amorphous structure of W─Ta─B is robust against thermal annealing, which is rather exceptional for amorphous alloys.
Next, we studied the longitudinal resistivity  xx of the W─Ta─B films.It is well known that both -W and amorphous-like W have high  xx (typically 100-300 μΩcm).On the other hand, -W shows low  xx (5-50 μΩcm). [19,20,23,36] xx of the W─Ta─B layers were estimated by measuring the sheet conductance G xx of the sample films with various W─Ta─B thicknesses.Figure 2a-c shows the thickness dependences of G xx for sample films with various compositions x (x = 0, 16, and 32.1) post-annealed at 350 °C.The slope of the linear fit corresponds to the longitudinal conductivity (inverse of  xx ) of the W─Ta─B layer.In the case of the W film (x = 0), G xx changes abruptly between 2 and 3 nm, above which -W is formed.Here, we define this thickness as "critical transformation thickness" t c , which is summarized as a function of composition x in Figure 2d.In the low alloying region (x < 32.1), t c emerges at ≈5-6 nm, where the resistivity drops <100 μΩcm (Figure 2b).This behavior is interpreted as a structural transition from -W to -W after thermal annealing, which is confirmed by XRD measurements (Supporting Information).Surprisingly, the high-resistive state of the W─Ta─B films is maintained up to 10 nm for x > 32.1, indicating t c > 10 nm in this region (open triangles in Figure 2d).The composition dependence of  xx for the W─Ta─B films in the low and high-resistive state are summarized in Figure 2e which shows that the value of  xx in the highresistive state is nearly constant (≈200 μΩcm) in all x region.To further check the annealing stability, the films were annealed at 400 °C, and it was confirmed that the W─Ta─B (x = 32.1)films can maintain their high resistivity even after post-annealing at 400 °C (Figure 2f).Now, we discuss that element, Ta or B, contributes to the stabilization of the amorphous structure in the present W─Ta─B alloy.To separate the contribution of each, we additionally prepared W 81 Ta 19 alloy (W-Ta) and W 90 B 10 alloy (W-B) films and conducted XRD and sheet conductance measurements (Supporting Information).In the case of the W-Ta film, we observed a clear resistivity drop as in the case of the W film that comes from the   structural transition to -W.In the case of the W─B film, on the other hand, the resistivity after annealing at 350 °C is almost constant at ≈200 μΩcm up to a thickness of 10 nm, where the amorphous phase was confirmed by XRD measurements.These results show that B rather than Ta stabilizes the amorphous phase, which is consistent with the case of Ta─B alloy. [39]A sudden increase in t c and drop in grain size (Supporting Information) for x ≧ 32.1 imply that there is a critical B composition, which is 5.8 at% in the case of x = 32.1, at which the amorphous phase is stabilized.Note that this composition is almost the same as the critical composition at which CoFeB becomes an amorphous structure. [41]We note that the crystallization of W─Ta─B in the low x region may be due to B diffusion into the 1 nm thick Ta buffer layer.

Spin-Orbit Torque Efficiency and Spin Hall Effect
Next, the spin-orbit torque (SOT) generated from the spin Hall effect in amorphous W─Ta─B alloys was investigated.The SOT acting on the adjacent ferromagnetic Co─Fe─B layer was evaluated using spin-torque ferromagnetic resonance (ST-FMR) with an additional injection of a DC bias current. [16,19,21]The measurement circuit and typical ST-FMR spectra under a DC bias current of ±2 mA are shown in Figure 3a,b, respectively.The linewidths of the spectra in Figure 3b are significantly modulated by the DC bias current, and the sign of the modulation slope varies with the angle of the applied in-plane magnetic field (Figure 3c).The linewidth modulation in the Co─Fe─B layer originates from the damping-like component of the SOT generated from the spin Hall effect in the W─Ta─B layer.
Figure 3d summarizes the composition x dependence of the spin-Hall-effect-driven SOT efficiency per unit current density (electric field),  j SHE (  E SHE =  j SHE ∕ xx ).A 5-nm-thick W─Ta─B layer annealed at 350 °C was used for evaluation.All samples were confirmed to be amorphous from XRD measurements (Supporting Information).| j SHE | shows the maximum value to be ≈40% with x = 32.1,whereas | E SHE | monotonically decreases from 2150 Ω −1 cm −1 with x.Note that the field-like component of the SOT in the W─Ta─B/Co─Fe─B system was also investigated and confirmed that its efficiency per unit current density is much smaller than that of  j SHE (Supporting Information).To estimate the spin Hall angle of W─Ta─B, we measured  j SHE for various thicknesses t of W─Ta─B with x = 32.1, as shown in Figure 3e.According to the spin drift diffusion model, [14,15,42]  j SHE can be expressed as where,  eff SHE is the effective spin Hall angle, which is a product of the spin Hall angle  SHE and the interfacial spin transparency T int (  eff SHE = T int  SHE ), and  sf is the spin-diffusion length.Using Equation (1) to fit curves to the data in Figure 3e, we estimated  eff SHE and  sf to be − 0.36 ± 0.01 and 1.3 ± 0.3 nm, respectively.Moreover,  eff SHE and  sf increase up to −0.43± 0.01 and 1.9 ± 0.1 nm after annealing at 400 °C (red curve in Figure 3e), which clearly shows that the spin Hall effect in amorphous W─Ta─B alloy is robust against high-temperature annealing.The estimated values of  eff SHE (−0.36 and −0.43) correspond to a lower limit of  SHE taking account that the spin current fully transparent at the W─Ta─B/Co─Fe─B interface (T int = 1).[44][45] The value of T int in W─Ta─B (x = 32.1)/Co─Fe─Bannealed at 350 °C is estimated to be 0.67 ± 0.04 (Supporting Information) which is slightly lower than that of the crystalline W/CoFeB system (T int = 0.8). [45]Taking the value of T int into account, we evaluated  SHE of W─Ta─B annealed at 350 °C to be − 0.53 ± 0.04.This value is comparable to  SHE of oxygen-incorporated W, [21] which shows the largest spin Hall effect among W-based materials.The spin Hall effect in amorphous alloys has been reported for W─Hf, [37,38] Ta─B, [39] and Y─Pt. [40]The absolute values of  SHE in these materials are which is of those obtained in the case of the present amorphous W─Ta─B alloys, which have | SHE | > 40%.
Hereafter, the origin of the spin Hall effect in amorphous W─Ta─B alloys is discussed.First, we discuss the effect of B on the resistivity and the spin Hall effect.The resistivity of W in the high-resistive state is ≈200 μΩcm, which is comparable to those of W─Ta─B alloys in the high-resistive state and W-B alloy (Supporting Information).This indicates that B concentration used in this work (up to 18 at% in x = 100) has a negligible effect on resistivity.Next, we investigated the effect of B on the spin Hall effect by measuring the SOT efficiency in the system with 5-nm thick W-B alloy as spin Hall material.3d) are almost the same with those of as-deposited -W (dotted horizontal line).Therefore, the addition of B does not directly contribute to the spin Hall effect.Next, we consider the extrinsic mechanism for the spin Hall effect.In the case of disordered binary alloys, it has been reported that extrinsic spin-orbit scattering significantly enhances the spin Hall effect. [46]The presence of an extrinsic mechanism can be verified by scaling the spin Hall angle or | j SHE | to  xx with a scaling factor of 1.In the case of amorphous W─Ta─B alloys, | j SHE | decreases monotonically with increasing x ≥ 22.9, although  xx is maintained at ≈200 μΩcm.This result indicates that the intrinsic mechanism is more dominant than the extrinsic mechanism in the amorphous W─Ta─B.Recently, it has been revealed that the unique phenomena occurring in topological materials, such as the spin-momentum locking state and the giant anomalous Hall and Nernst effects, can be preserved in an amorphous structure [47,48] and these are attributed to the short-range crystal order in the amorphous structure.29] Substitutional doping of Ta into W corresponds to effective hole doping.Thus, the addition of Ta can enhance the intrinsic spin Hall effect in -W.This scenario can also explain the observed increases in | E SHE | and | j SHE | of W─Ta─B (x ≤ 32.1).

Magnetization Switching Operation in Three-Terminal MTJ Cells for a SOT-MRAM
The amorphous W─Ta─B alloy was integrated into a threeterminal MTJ (3T-MTJ) cell for SOT-MRAM and the write/read performance was demonstrated.MTJ multilayer films (hereafter referred to as "A" and "B") with the following structure were prepared: Substrate / W 67.9 (Ta 82 B 18 ) 32.1 (8)/Co 20 Fe 60 B 20 (1.8)/MgO/ip-SAF/capping layer.Here, the Co 20 Fe 60 B 20 layer acts as a magnetic free layer, and the ip-SAF layer, which stands for in-plane magnetized synthetic antiferromagnet, acts as a reference layer for the MTJ.The SOT-MTJ films were post-annealed for an hour at 360 °C (film A) and at 400 °C (film B) under an inplane magnetic field of 1 T to enhance MR ratio as well as to give an exchange-bias to the ip-SAF of film A. 3T-MTJ devices (Figure 4a) were fabricated from the MTJ films.Devices A and B were fabricated from the films A and B, respectively.For both devices, A and B, rectangular-shaped MTJ nano-pillars were fabricated.Both devices exhibited MR ratios of over 100%.The long axis of the nano-pillars was placed orthogonal to the write current path.
The magnetoresistance curve of a 50 nm × 200 nm MTJ in device A under an in-plane magnetic field along the y-axis is shown in Figure 4b.The MTJ resistance (R MTJ ) jumps correspond to magnetization switching of the free layer.The SOT-induced magnetization switching curve for the same device for a pulse width  of 1 ms is shown in Figure 4c.The switching current density j sw from the AP to the P state (P to the AP state) is −2.5 × 10 6 A cm −2 (−3.1 × 10 6 A cm −2 ).Here, we assumed that the shunting current via the Co─Fe─B layer is negligibly small because the width of the spin Hall channel is much larger than the size of the MTJ.To investigate the thermal stability and intrinsic switching current density, j C0 , of the devices, the pulse-width dependence of j sw was investigated (Figure 4f).The pulse width was varied between 100 μs and 1 s.The controlled pulse duration corresponds to the thermally activated region, where j sw can be expressed as where Δ and  0 correspond to the thermal stability factor and the inverse of the attempt frequency ( 0 = 1 ns), respectively.Using Equation ( 2) and the data shown in Figure 4d, we obtained j c0 = (4.0 ± 0.3) × 10 6 A cm −2 and Δ/k B T = 45 ± 3. Note that j c0 and Δ for many devices with different MTJ dimensions were measured and the measurement results confirmed that all devices show j c0 values between 4.0 and 5.0 × 10 6 A cm −2 and Δ/k B T> 40.(Supporting Information).It should be noted that the annealing stability of device A is limited to 360 °C because the diffusion of Mn atoms from the IrMn layer in the ip-SAF degrades the MTJ film.To confirm the higher annealing stability of the W─Ta─B alloy integrated into the MTJs, we fabricated device B in which the IrMn layer was excluded and annealed it at 400 °C.Even without the exchange bias from the Ir-Mn layer, device B exhibits an antiparallel magnetic configuration, as shown in Figure 4d, because the ip-SAF layer has shape-induced uniaxial magnetic anisotropy.As shown in Figure 4e, device B showed SOT-induced magnetization switching, and j c0 and Δ/k B T were estimated to be 6.6 ± 0.2 × 10 6 A cm −2 and 47 ± 3, respectively (shown by the red plots in Figure 4f).It is worth noting that j c0 of device B is 30% larger than of device A. The slight increase in j c0 may be caused by inter-diffusion between the W─Ta─B layer and the magnetic free layer during the high-temperature annealing process.Such inter-diffusion could be suppressed by optimizing the stacking structure of the MTJ multilayer, for example, by inserting a sub-nanometer-thick additional spacer layer such as Hf [31,51] or by depositing a magnetic free layer at cryogenic temperature. [52]

Significance of Amorphous W─Ta─B Alloy for SOT-MRAMs
In Table 1, j c0 , Δ, the MR ratio and the annealing stability T ann of the amorphous W─Ta─B alloy are compared with the values reported for various in-plane magnetized 3T-MTJ devices with various spin Hall materials such as -Ta, -W, and Pt-based materials.In addition to the very low j C0 due to the high  j SHE , devices A and B have high MR ratios (≈130%), relatively high thermal stability factors (Δ/k B T > 45), and excellent annealing stabilities (up to 400 °C), which satisfy all the major requirements for SOT-MRAMs.To evaluate the energy efficiency of 3T-MTJs, we estimated the writing power consumption of 3T-MTJs with various spin Hall materials.The SOT-MRAM cells were rescaled by assuming a self-aligned fabrication method [39] , in which the widths of the MTJ nano-pillars are equal to the widths of the spin Hall channels.For device A and similar sized MTJs and under the assumption of 10-ns switching, the writing power consumption was estimated to be 56 fJ per bit at minimum (Supporting Information).This writing energy is comparable to that of stateof-the-art STT-MRAMs. [5]n addition to the significance of low-power operation, it is noteworthy that the developed spin Hall material enables a high degree of freedom in designing SOT-MRAM devices.4] We confirmed that a perpendicularly magnetized free layer can be formed by using an amorphous W─Ta─B alloy as a spin Hall channel and demonstrated a operation (Supporting Information).

Conclusion
We have shown that an amorphous spin Hall material, specifically, a W─Ta─B alloy, satisfies the requirements of having a large spin Hall angle and high annealing stability.Integrating this spin Hall material into the spin Hall channel of a SOT-MRAM device can produce a highly energy-efficient writing operation comparable to state-of-the-art MRAM technology, in addition to high MTJ performance.The developed W─Ta─B alloy provides an ideal solution as the spin Hall material for SOT-MRAMs, which can be prepared by a relatively simple deposition process.Our results show that exploration of amorphous spin Hall materials can be a new route to promoting the practical application of nextgeneration nonvolatile memory devices.

Experimental Section
Sample and Device preparation: All the multilayer films including the 3T-MTJs were grown on thermally oxidized Si substrates in an ultra-highvacuum magnetron sputtering system (Cannon Anelva C7100).Multilayer films for investigating the crystal structure, the electrical properties, and the spin-orbit torque efficiency consisted of, from the substrate upward: Ta (1)/W 100-x (Ta 82 B 18 ) x (t)/Co─Fe─B (3)/MgO/Ta-O (thickness in nm).The W 100-x (Ta 82 B 18 ) x layer (W─Ta─B layer) was deposited by co-sputtering from W and Ta 80 B 20 alloy targets, and the composition ratio x and composition ratio between Ta and B were determined by inductively coupled plasma mass spectroscopy using 50 nm thick W─Ta─B and Ta-B films.The thickness of the W─Ta─B layer was varied from 2 to 10 nm.The composition of the Co─Fe─B layer was Co 19 Fe 56 B 25 .The 1 nm-thick Ta layer acts as a seed layer to ensure the flatness of the system.The top Ta-O capping layer was formed by naturally oxidizing a 2 nm thick Ta layer under ambient conditions.The multilayer films were post-annealed at 350 or 400 °C for an hour under vacuum.The structure of the amorphous phase was confirmed by XRD measurements (Rigaku, SmartLab) using Cu-Ka radiation ( = 0.154 nm).Note that the background signal from the substrate has been subtracted.For the ST-FMR measurements, the multilayer films were fabricated into 3 μm-wide, 30 μm-long microstrips by optical lithography and Ar ion milling.Electrical contacts were made by depositing Cr(5)/Au(100) or Ru(5)/Pt(100) (thicknesses in nm).
Stacking structures of the 3T-MTJs were as follows (from the substrate upward): W 67.9 (Ta 82 B 18 ) 32.1 (8)/Co 20 Fe 60 B 20 (1.8)/MgO/ip-SAF / capping layer.Note that IP-SAF stands for in-plane magnetized synthetic antiferromagnet, which consists of Co─Fe─B(2)/W(0.15)/Co-Fe(2)/Ru(0.85)/Co-Fe(2.5), and acts as a reference layer.Two types of MTJ films, "A" and "B" were prepared.In film A, a 7 nm thick IrMn was inserted between the ip-SAF layer and the capping layer.For film B, the IrMn layer was not inserted.The 3T-MTJ films were post-annealed for an hour at 360 °C for film A and at 400 °C for film B under an in-plane field of 1 T.Both films exhibit magnetoresistance ratio of >100% and a resistance-area (RA) products of ≈190 Ωμm 2 .3T-MTJ devices were fabricated from films A and B by the following process.First, the MTJ films were patterned into rectangular-shaped MTJ pillars by using electron beam lithography and Ar ion beam milling.The size of the MTJs was varied from 50 nm × 200 nm to 100 nm × 300 nm.The milling process was stopped immediately after the magnetic free layer was etched away, which was controlled by monitoring the secondary ion mass spectra.After the etching process, a Si-N/Si-O passivation layer was deposited by ion beam sputtering without breaking the vacuum.Next, a spin Hall channel having a width of 2.4 or 4 μm was patterned by optical lithography, and Ar ion milling, followed by deposition of an Al-O/Si-O passivation layer and a lift-off process.Finally, electrode pads consisting of Ru (5)/Pt (100) or Cr (5)/Au (100) (thickness in nm) were formed by optical lithography and a lift-off process.
Transport Measurements: The setup for the ST-FMR measurements is shown in Figure 2a.A GHz RF current with frequency f and a DC bias current was injected into the microstrip devices.The RF current excites FMR in the Co─Fe─B layer via a combination of the SOT and the Oersted field.The FMR spectra were electrically measured through the rectified voltage V FMR due to the applied RF current and oscillation of anisotropic magnetoresistance (or spin Hall magnetoresistance).A lock-in amplifier was used to precisely measure V FMR by modulating the RF current at a low frequency of 1-3 kHz.Additionally, the DC-bias-current injection modulates the effective magnetic damping or resonance linewidth (W) due to the damping-like component of the SOT acting on the Co─Fe─B layer.Considering that the modulation originates from the SOT driven by the spin Hall effect, the modulation slope (dW/dI DC ) can be expressed as where  is gyromagnetic ratio,  is the in-plane magnetic field angle, H demag is the effective demagnetization field, H res is the resonance field, M s is the saturation magnetization of the Co─Fe─B layer, d (t) is the thickness of the Co─Fe─B (W─Ta─B) layer, and w is the width of the microstrip devices.μ 0 is the permeability in a vacuum, ℏ is the Dirac constant, and e is the elementary charge. is the shunting current ratio of the W─Ta─B layer and was estimated from the  xx of the Ta, W─Ta─B, and Co─Fe─B layers (Supporting Information).Using Equation (3), we determined the  j SHE and effective spin Hall conductivity (  E SHE =  j SHE ∕ xx ).Note that H demag and  were obtained from the frequency dependence of H res and W (Supporting Information).
For SOT-MRAM device operation, a DC current source measure unit (Keysight B2912A) was used to supply the writing current pulses into the spin Hall channel and to measure the MTJ resistance.The current pulse duration for switching was varied between 100 μs and 1 s.To measure the MTJ resistance, a DC voltage with a magnitude of 10-20 mV was applied.A mechanical relay switch was used to change the electrical paths of the writing and reading operations.J sw was determined by repeating the switching curve >100 times for each pulse width.

Figure 1 .
Figure 1.Crystal structure of amorphous W─Ta─B alloy films.a) X-ray diffraction (XRD) patterns of W (black), W─Ta─B (red), and Ta-B (blue) films post-annealed at 350 °C.The composition x of the W─Ta─B layer is x = 32.1.For clarity, an offset has been introduced to separate each XRD pattern.The vertical dotted lines show diffraction peaks from -W and -W.The inset shows an extended view of W─Ta─B and Ta-B films.b) XRD patterns for W─Ta─B films (x = 32.1) in as-deposited (black), 350 °C annealed (orange), and 400 °C annealed (red) conditions.c) Cross-sectional high-resolution TEM image of the W─Ta─B film.d) Nano-beam diffraction patterns of the W─Ta─B layer.The two measurement positions are shown as red spots in the TEM image in (c).e) High-angle annular dark field scanning transmission electron microscopy (HAADF-STEM) image (left) and energy dispersive X-ray (EDX) mapping of Ta (middle) and W (right) in the W─Ta─B film.The Ta signal in the upper and lower regions (thicknesses of ≈2 and ≈1 nm, respectively) corresponds to the Ta-O capping layer and the Ta seed layer.

Figure 2 .
Figure 2. Electrical properties of amorphous W─Ta─B alloy films.a-c) W─Ta─B thickness dependence of sheet conductance with composition a) x = 0, b) x = 16, and c) x = 32.1.The vertical arrows correspond to the critical transformation thickness, t c .The solid line and dotted lines show linear fitting results.d) Composition x dependence of the critical thickness t c .The open triangles correspond to t c > 10 nm.The green (orange) area indicates the low (high) resistive state of the W─Ta─B layer, and the boundary (solid line) is a guide for the eyes.e) Composition x dependence of longitudinal resistivity  xx .The solid (open) symbols show  xx calculated below (above) t c .f) W─Ta─B thickness dependence of sheet conductance for x = 32.1 after thermal annealing at 350 °C (black) and 400 °C (red).

Figure 3 .
Figure 3. Evaluation of spin Hall effect in W─Ta─B alloy films.a) Schematic of spin-torque ferromagnetic resonance (ST-FMR) measurement setup.b) ST-FMR spectra for the sample with x = 32.1 under an in-plane field.μ 0 is the permeability in a vacuum.The in-plane field angle  is set to  = 45 degrees.The red (blue) data points correspond to the spectrum under a DC bias current of +2 (−2) mA.The solid lines represent the Lorentzian curves fitted for each spectrum.c) Bias current I DC dependence of the spectral linewidth μ 0 W for two applied field angles ( = 45 and −135°).The solid lines represent linear fitting results.d) Composition x dependence of the SOT efficiency per unit current density  j DL (left) and the SOT efficiency per unit electric field  E DL (right).The open triangles and the horizontal dotted lines correspond to  j DL and  E DL of W-B alloy and as-deposited -W, respectively.e)  j DL as a function of the W─Ta─B (x = 32.1)layer thickness after thermal annealing at 350 °C (black) and 400 °C (red).The solid lines are the curves fitted according to Equation (1).
| j SHE | (| E SHE |) of the W-B alloy was estimated to be 0.31 (1.54 × 10 3 Ω −1 cm −1 ).The magnitudes of | j SHE | and | E SHE | in W-B (open triangles in Figure

j
SHE |) by up to ≈45% (30%) for 0 < x < 40 compared to that of W-B and as-deposited -W.With further increases in x (x > 40), both | E SHE | and | j SHE | decrease monotonically.Although B stabilizes the amorphous phase, it has no effect on the spin Hall effect as shown in Figure 3d, and therefore, the enhancement of the spin Hall effect is attributed to the addition of Ta into W. Since the spin Hall angle (or spin Hall conductivity) of -Ta is smaller than that of -W, from simple interpolation between -Ta and -W, | E SHE | and | j SHE | are expected to monotonically decrease with respect to x with the maximum value at x = 0 (-W).However, such a simple alloying effect cannot explain the clear increase in | E SHE | and | j SHE | for 0 < x < 40.

Figure 4 .
Figure 4. Energy-efficient spin-orbit-torque magnetoresistive random-access memory (SOT-MRAM) operation.a) Schematic of a 3T-MTJ for a SOT-MRAM.b-e) Magnetoresistance curves under an in-plane magnetic field [(b) and (d)]and SOT-induced switching curves [(c) and (e)] in devices A and B (black and red data, respectively).The size of the MTJ is 50 nm × 200 nm for device A and 80 nm × 240 nm for device B. The pulse width t was 1 ms, and no external field was applied.f) Pulse-duration dependence of the switching current density for AP-P switching (solid symbols) and P-AP switching (open symbols) in devices A (black) and B (red).

Table 1 .
Comparison of j C0 , Δ/k B T, the MR ratio, and annealed temperature (T ann ) of present W─Ta─B based 3T-MTJs and various in-plane magnetized 3T-MTJs with different spin Hall materials.The bottom line shows the requirements for high-speed SOT-MRAM cache memories.The thickness in nm of the spin Hall material used for device fabrication is shown in parentheses.The MTJ dimensions for devices A and B are 50 nm × 200 nm and 80 nm × 240 nm, respectively..