betaFeedback

WorksPhysics and Astronomy

Generating unconventional spin-orbit torques with patterned phase gradients in tungsten thin films

Abstract

A key aim in spintronics is to achieve current-induced magnetization switching via spin-orbit torques without external magnetic fields. For this, the focus of recent work has been on introducing controlled lateral gradients across ferromagnet/heavy-metal devices, giving variations in thickness, composition, or interface quality. However, the small gradients achievable with common growth techniques limit both the impact of this approach and understanding of the underlying physical mechanisms. Here, spin-orbit torques are patterned on a mesoscopic length scale in tungsten thin films using direct-write laser annealing. Through transmission electron microscopy, resistivity, and second harmonic measurements, the continuous transformation of the crystalline phase of W films from the highly spin-orbit coupled, high resistivity ββ phase to the minimally spin-orbit coupled, low resistivity αα phase is tracked with increasing laser fluence. Gradients with different steepness are patterned in the tungsten phase to create spin-orbit torque channels and, when interfaced with CoFeB, tungsten wires with a sufficiently strong gradient can switch the magnetization without an applied magnetic field. Therefore, exploiting the unique microstructure of mixed-phase W allows precise control of the local electronic current density and direction, as well as local spin-orbit torque efficiency, providing a new avenue for the design of efficient spintronic devices.

Introduction

To meet the growing demand for energy-efficient data retention, there has been extensive research into optimizing magnetic materials for use as memory elements. Spin-orbit torque-based magnetic random access memory (SOT-MRAM) is an attractive option for reading and writing of magnetic elements at high speed and with long endurance [1–4]. However, these devices are limited by the requirement of an applied magnetic field for current-induced switching, which impedes implementation into industrial devices [3, 5]. Thus, there have been many studies on engineering magnetic devices to switch without the need for an applied magnetic field [6–17]. For deterministic field-free SOT switching, the mirror symmetry of the device along the current direction must be broken. In sputter-deposited, polycrystalline multilayers, this has been achieved through lateral gradients made by oblique deposition [18, 19], vertical composition gradients [20, 21], tilted magnetic anisotropy [22, 23], non-uniform currents [24, 25], applied electric fields [26, 27], and selective damage through ion irradiation or laser ablation [28, 29]. However, these techniques are difficult to implement on the industrial scale because they either provide limited scalability or are not suitable.

One route forward to provide an industry-compatible approach is to locally tune the spin-orbit torque efficiency in the film plane, although relatively little attention has been paid to exploit this opportunity up to now [30–32]. This can be achieved by leveraging metastable materials, in which the metastable crystalline phase has a different spin-orbit coupling than the stable crystalline phase, thus providing a natural platform to precisely and locally tune spin-orbit torque efficiency. Thin films of tungsten, with both a metastable and a stable phase, offer this tunability. In addition, this material is widely applicable and has been investigated over decades for applications in superconductivity [33, 34], solar energy [35], microelectronics [36], and spintronics [37–41]. α\alpha-W is the stable, body-centered cubic crystal phase of the material, characterized by low resistivity (∼30 μΩ\sim30\,\mu\Omega-cm) and desirable electronic properties to replace Cu in ultra-narrow wires [42]. Meanwhile, the metastable β\beta-W phase with A15 crystal structure has been employed to exert spin-orbit torques on magnetic thin films due to its high spin-orbit coupling, which yields a high charge-to-spin conversion efficiency of ∼0.3−0.5\sim0.3 - 0.5 [37]. Despite the high resistivity of the β\beta-W phase (∼240 μΩ\sim240\,\mu\Omega-cm), this is a material of choice for state-of-the-art SOT-MRAM devices [2, 4, 43]. Due to the distinct properties of each of these two crystalline phases, there has been extensive work carried out to determine how to stabilize each phase and transform between the two [44–46]. However, efforts into understanding phase transformations in W has focused on global, rather than local, transformations, and local control of the W crystallographic phase has not been shown. With recent work demonstrating the use of direct-write laser annealing (DWLA) to locally enact heat-induced transformation of physical properties [47–49], it is now possible to tune the phase of W films on the nanoscale.

In this work, the local patterning of the W crystalline phase is implemented by DWLA to obtain precise control over the density of α\alpha-W grains in β\beta-W thin films, as revealed by transmission electron microscopy (TEM). With electrical measurements, a decrease of the film resistivity from 230 μΩ\mu\Omega cm to 30 μΩ\mu\Omega cm is detected, along with a decrease in the spin-orbit torque efficiency of W from 0.45 to <0.1< 0.1 as expected with this crystalline phase transformation. Field-free spin-orbit torque switching in W/CoFeB bilayers is then demonstrated by introducing a gradient in the W phase. These patterned lateral phase gradients in W provide a tunable platform to understand the physical mechanisms of field-free switching in polycrystalline systems with broken structural symmetry.

Tracking structural transformation of W with laser annealing

Metastable β\beta-W films are grown by sputtering. The β\beta-W phase of as-grown films is confirmed with x-ray diffraction, where only diffraction peaks corresponding to the β\beta-W phase are observed (see Supporting Information Figure S1). The W films are then locally modified using direct-write laser annealing [49] with laser fluences ranging from 0 J cm−2^{-2} to 31 J cm−2^{-2}. The maximum laser fluence used depends on the thermal conductivity of the substrate; for silicon nitride membrane substrates, the W films were annealed at a maximum laser fluence of 5.1 J cm−2^{-2}, while for W films on thermally oxidized Si substrates, the maximum laser fluence was 31 J cm−2^{-2}. In order to examine the microstructural changes in W that result from the local laser treatment, several regions of a 15 nm-thick W film grown on a silicon nitride membrane are annealed (see schematic in Figure 1a). Then, TEM is employed to image the annealed region in the film plane. To track the evolution of the W microstructure, a 50 μ\mum×\times50 μ\mum region in a W film is annealed with a linear gradient in laser fluence. In Figure 1b, TEM images taken along the entire 50 μ\mum fluence gradient at 8000×\times magnification are stitched together [50] and shown as a single image. Larger images of four regions along the gradient are shown in Figure 1c. In the as-grown region (uppermost panel), there are only a few sparse large grains visible at this magnification, which appear in black and white due to to their diffraction of the electron beam away from or towards the detector, while the gray uniform region is made up of very small grains. With laser annealing, the grain size and density increases as the laser fluence is increased (middle panels). Finally, at the maximum laser fluence, only large W grains are observed, and there are no gray regions remaining (lower panel). This trend is similar to behavior observed in W for in-situ TEM annealing studies [51]. From TEM images taken at 120,000× magnification, the size of the small grains are estimated to be 5-20 nm, while the large grains are 80-120 nm in size (seen in Supporting Information Figure S2a,b).

Figure 1. Structural transformation of W thin films. a) Schematic of DWLA of W. As the laser rasters over the W surface, the deposited thermal energy leads to a change in the crystalline phase. The laser fluence is increased linearly from left to right, indicated by the increasing intensity of pink in the bar below the schematic. The green spheres indicate W atoms in the β\beta-W phase, while the blue spheres indicate W atoms in the α\alpha-W phase. b) A series of stitched TEM images taken at 8000× magnification along the linear gradient in laser fluence for a 15 nm-thick W film on a silicon nitride membrane (scale bar=0.5 µm). c) Selected TEM images at 8000× magnification along the gradient. The uppermost panel is at the lowest fluence end of the gradient and appears visually uniform. For the images taken in regions with increasing laser fluence, there are an increasing number of large black and white grains. d) Selected-area electron diffraction performed in 2 µm steps along the fluence gradient, with the intensity shown as a function of scattering vector qq. As the fluence is increased from 0 J cm−2^{-2} to 5.1 J cm−2^{-2}, diffraction peaks corresponding to β\beta-W smoothly transition to diffraction peaks that correspond to α\alpha-W.

To gain crystallographic information about the two distinct phases in W, selected-area electron diffraction (SAED) using a 1.2 μm1.2\,\mu\mathrm{m} aperture was carried out along the gradient in 2 μm2\,\mu\mathrm{m} steps. Rings of intensity in SAED are observed (see Supporting Information Figure S2-Figure S2), typical of polycrystalline films. Therefore, the radial average of the SAED signal is taken and plotted as a function of the scattering vector qq in Figure 1d for each SAED image taken along the gradient. By comparing electron diffraction peaks to the positions of diffraction peaks calculated using Bragg’s law for the β\beta-W and α\alpha-W crystal structures, it can be seen that the broad diffraction peaks for the as-grown region of the W film correspond only to β\beta-W peaks, indicative of a uniform β\beta-W film with very small grains, while the region annealed at the highest laser fluence of 5.1 J cm−25.1\,\mathrm{J\,cm}^{-2} has only α\alpha-W peaks, with the narrow peak width typical of larger crystal grains. Between these two fluences, a smooth transition from diffraction data with peaks corresponding only to β\beta-W to diffraction data with peaks from both phases can be seen, thus indicating a mixed phase in this fluence window.

TEM image of an as-grown beta-W film

(a)

TEM image of an annealed alpha-W region in W film

(b)

SAED and associated TEM images for the as-grown film

(c)

SAED and associated TEM images for a mixed-phase film

(d)

SAED and associated TEM images for a mixed-phase film

(e)

SAED and associated TEM images for a W film at maximum laser fluence

(f)

Figure S2. a) TEM image of an as-grown β\beta-W film. b) TEM image of an annealed, α\alpha-W region in W film. (c-f) SAED and associated TEM images for four locations along the gradient which correspond to the four images shown in Figure 1d: c) the as-grown film, d-e) mixed-phase films with increasing laser fluence, and f) W film at the maximum laser fluence of 5.15.1 J cm−2^{-2}, which has entirely transformed to α\alpha-W. The film used for these TEM images is grown on a silicon nitride membrane.

From this, it can be concluded that the gray, small-grain region seen in the TEM images of Figure 1b,Figure 1c corresponds to β\beta-W, while the large grains are composed of α\alpha-W. This thermally-driven phase transformation is notable because, instead of a deformation of the β\beta-W lattice into an α\alpha-W lattice, an α\alpha-W lattice nucleates and expands [44], leading to a bimodal distribution of grains in the mixed-phase films [52]. Thus, a μm\mu\mathrm{m}-scale gradient will create many structurally asymmetric interfaces on the nanoscale, with 8080-120 nm120\,\mathrm{nm}-sized α\alpha-W grains interfaced with 55-20 nm20\,\mathrm{nm}-sized β\beta-W grains.

Electronic and spintronic properties

The microstructural transformation from β\beta-W to α\alpha-W observed in TEM is accompanied by dramatic changes in the electronic and spintronic properties of the films. For as-grown films, the average film resistivity ρW\rho_{\mathrm{W}} is 233 μΩ233\,\mu\Omega-cm, comparable to literature values for β\beta-W [52]. After annealing above a threshold fluence of 3 J cm−23\,\mathrm{J\,cm}^{-2}, ρW\rho_{\mathrm{W}} drops significantly (as seen in black in Figure 2a for a 14 nm14\,\mathrm{nm}-thick W film grown on a thermally oxidized Si substrate). At fluences higher than 20 J cm−220\,\mathrm{J\,cm}^{-2}, ρW\rho_{\mathrm{W}} stabilizes at 30 μΩ30\,\mu\Omega-cm, comparable to values for α\alpha-W films [52]. The fraction of α\alpha-W as a function of laser fluence, seen in blue in Figure 2a, is calculated using the measured resistivity values in black and a parallel resistor model with α\alpha-W and β\beta-W. This decrease in ρW\rho_{\mathrm{W}} as the laser fluence is increased eliminates the possibility that there is significant oxidation of the tungsten due to DWLA [49], since tungsten oxides have a higher resistivity than β\beta-W [53].

Transformation of electronic and spintronic properties of W on changing the phase with DWLA

Figure 2. Transformation of electronic and spintronic properties of W on changing the phase with DWLA. a) Resistivity ρW\rho_{\mathrm{W}} of 14 nm-thick W films as a function of laser fluence. Using a parallel resistor model, the fraction of α\alpha-W as a function of laser fluence is calculated from these resistivity values. b) Measurement geometry for second harmonic Hall measurements. An ac charge current jcj_{\mathrm{c}} flows along the xx-direction and the sample is rotated in an external magnetic field HH. At each angle φ\varphi between HH and xx, the second harmonic of the resistance along yy is recorded. c) The magnitude of the damping-like torque efficiency per current ξDLj\xi_{\mathrm{DL}}^{j} as a function of the fraction of α\alpha-W for 14 nm-thick W. Error bars are given by the standard deviation calculated from the fitting procedure used to determine ξDLj\xi_{\mathrm{DL}}^{j} (see Methods). d) The damping-like torque efficiency per electric field ξDLE\xi_{\mathrm{DL}}^{E} as a function of the fraction of α\alpha-W. Error bars are given by the propagated standard deviation from the fitting procedure used to determine ξDLE\xi_{\mathrm{DL}}^{E}. e-f) Thermoreflectance (R2ω/RR^{2\omega}/R) and reflectivity (RR) as a function of position across a 10 μ\mum-wide 7 nm W/2 nm CoFeB device. The thermoreflectance (shown in blue), is proportional to the current-induced heating across the device. e) Thermoreflectance of a β\beta-W device (depicted by the schematic below) is relatively constant across the device. f) For a sample with a gradient from 0% to 100% α\alpha-W (depicted by the color gradient from green to blue in the schematic below), the thermoreflectance is highly non-uniform.

The stability of mixed-phase W created with DWLA was investigated by measuring the resistivity of a W Hall bar with a gradient of phase over the span of several months. In contrast to unstable mixed-phase films grown by sputtering, which transformed to α\alpha-W within days [54], the resistivity of the annealed regions did not exhibit large changes in resistivity over time. For example, a 4%4\% increase in the resistivity of a W device with a gradient from 0%0\% to 50%50\% α\alpha-W over 10 μm10\,\mu\mathrm{m} was measured over five months (see Supporting Information Figure S3). Since a transformation from β\beta-W to α\alpha-W would result in lower resistivity, this increase indicates there is minor surface oxidation rather than transformation to α\alpha-W. This means that devices with phase-controlled W regions created with DWLA are stable for long-term use.

Figure S3. The resistance of a W wire with an annealed gradient is tracked over 150 days. Only a small (∼\sim 4.5%) increase in resistance is measured in this time frame.

The spin-orbit torque efficiency of W also changes significantly depending on the crystalline phase. To verify this, second harmonic Hall resistance measurements of spin-orbit torques were performed on W devices annealed homogeneously with a single laser fluence using CoFeB/MgO/Ta deposited on 14 nm14\,\mathrm{nm}-thick W as the magnetic layer with in-plane magnetization. In this technique, with the sample geometry illustrated in Figure 2b, an ac current in the bilayer produces current-induced torques leading to oscillations of the CoFeB magnetization. These in turn lead to an oscillating

Hall resistance that can be detected by measuring the second harmonic Hall resistance (Rxy2ωR_{xy}^{2\omega}). By fitting the angular, magnetic field, and current dependence of Rxy2ωR_{xy}^{2\omega}, the damping-like torque efficiency per current ξDLj\xi_{\mathrm{DL}}^{j} and per electric field ξDLE\xi_{\mathrm{DL}}^{E} of W can be extracted [55, 56] (see Supporting Information Figure S4 for example data). ξDLj\xi_{\mathrm{DL}}^{j} is a figure of merit describing how efficiently charge currents are converted to spin currents in a material and is related to the commonly cited spin Hall angle θSH\theta_{\mathrm{SH}} by ξDLj=TintθSH\xi_{\mathrm{DL}}^{j}=T_{\mathrm{int}}\theta_{\mathrm{SH}}, where TintT_{\mathrm{int}} is the spin transparency at the heavy metal/magnet interface [57].

Figure S4. a) Rxy2ω(φ)R^{2\omega}_{xy}(\varphi) for β\beta-W/CoFeB for an applied magnetic field of 128 mT. b) RHR_{\mathrm{H}} of β\beta-W/CoFeB as a function of μ0Hz\mu_{0}H_{z}. RAHER_{\mathrm{AHE}} and μ0HKeff\mu_{0}H_{\mathrm{K}}^{\mathrm{eff}} are extracted from this data. c) The scaled cos⁡φ\cos\varphi component of Rxy2ω(φ)R^{2\omega}_{xy}(\varphi) as a function of (μ0H+μ0HKeff)−1(\mu_{0}H+\mu_{0}H_{\mathrm{K}}^{\mathrm{eff}})^{-1}. Data is plotted for five different currents from 5 mA to 9 mA. d) The change in BDLB_{\mathrm{DL}} with IWI_{\mathrm{W}}. The slope of the linear relationship between the two (dashed black line) is proportional to ξDLj\xi_{\mathrm{DL}}^{j}. e) An estimate of the magnitude of ξFLj\xi_{\mathrm{FL}}^{j} as a function of α\alpha-W fraction.

ξDLj\xi_{\mathrm{DL}}^{j} as a function of α\alpha-W fraction is shown in Figure 2c. β\beta-W has ξDLj=−0.45±0.01\xi_{\mathrm{DL}}^{j}=-0.45\pm0.01, which is comparable to measured values of θSH\theta_{\mathrm{SH}} [37, 40, 58, 59]. However, small fractions of α\alpha-W lead to a steep decrease of ξDLj\xi_{\mathrm{DL}}^{j}, which may be exacerbated by a decreased interface quality between W and CoFeB after annealing at low laser fluences. ξDLj\xi_{\mathrm{DL}}^{j} continues to decrease with increasing fraction of α\alpha-W, albeit more gradually, and at 95% α\alpha-W, ξDLj=−0.07±0.01\xi_{\mathrm{DL}}^{j}=-0.07\pm0.01. ξDLE\xi_{\mathrm{DL}}^{E} (given by ξDLj/ρW\xi_{\mathrm{DL}}^{j}/\rho_{\mathrm{W}}) as a function of α\alpha-W fraction (seen in Figure 2d) provides a scaled spin-orbit torque efficiency accounting for non-uniform current flow through a W wire with a gradient in phase. Here, devices with larger fractions of α\alpha-W have a relatively high ξDLE\xi_{\mathrm{DL}}^{E} due to the significantly lower resistivity, although the large measurement error obscures a clear trend in ξDLE\xi_{\mathrm{DL}}^{E} as a function of α\alpha-W fraction.

To directly visualize the impact of adding a gradient in the W phase on the electronic properties of a device, the thermoreflectance of 10 μ\mum-wide 7 nm W/2 nm CoFeB devices was detected with current-modulated, spatially-resolved optical reflectivity measurements (see Methods). The reflectivity of the devices was measured as a laser scanned along the yy-direction transverse to an ac current applied to the devices along xx. The magnitude of the time-averaged reflectivity RR is determined by the microstructure of the film. The thermoreflectance (R2ω/RR^{2\omega}/R) is defined as the second harmonic reflectivity change R2ωR^{2\omega} normalized by RR. Applying an ac current results in harmonic Joule heating, which in turn leads to a harmonic variation of the refractive index. Hence, the thermoreflectance is a measure of the Joule heating-induced change in reflectivity, giving a qualitative insight into the current density [60]. A device with uniform, as-grown β\beta-W, shown in Figure 2e, is compared to a device with a gradient in phase going from 0% to 100% α\alpha-W over 10 μ\mum, shown in Figure 2f. The reflectivity (black line in Figure 2e-Figure 2f) gives an indication of the edges of the device, and regions with more α\alpha-W have a slightly lower reflectivity than β\beta-W, seen by the black reflectivity curve for the gradient device in Figure 2f. The thermoreflectance of the β\beta-W device is relatively constant (Figure 2e), reflecting the uniform current flow across the as-grown W. In contrast, the effect of strongly spatially-varying resistivity is exemplified by the thermoreflectance of the gradient device, depicted in blue in Figure 2f. Due to the low resistivity of α\alpha-W, the vast majority of the current passes through the far right region of the gradient. These results indicate that landscapes with smooth gradients in the spin-orbit torques and current density of W can be created.

Spin-orbit torque switching

To perform SOT switching, elliptical dots with minor (major) axis of 5.7 μ\mum (10 μ\mum) of 1 Ta/0.9 CoFeB/1 MgO/3 Ta (listed from the lowest layer on the substrate to the top layer, with all thicknesses in nm) are deposited on laser-annealed W devices. The 1 nm-thick Ta seed layer is included to stabilize perpendicular magnetic anisotropy in the CoFeB layer without the need for any post-growth annealing. 50 μ\mus current pulses are applied to the devices for spin-orbit torque switching experiments, and the Hall resistance RHR_{\mathrm{H}} is recorded after each current pulse to detect the magnetization orientation (see Methods). RHR_{\mathrm{H}} is also measured as a function of applied out-of-plane magnetic field μ0Hz\mu_{0}H_{z} in order to detect the RHR_{\mathrm{H}} values corresponding to uniform up/down magnetization. For devices uniformly annealed at a constant laser fluence, the critical switching current density jcj_{\mathrm{c}}, defined as the average of the two current densities where RHR_{\mathrm{H}} reverses sign from negative to positive and from positive to negative, at a given in-plane magnetic field follows the trend observed in second harmonic Hall measurements of lower spin-orbit torque efficiency for devices with a higher fraction of α\alpha-W (see Figure 2c), where the switching current increases significantlyelly for devices annealed at higher laser fluence (see Supporting Information Figure S5). For these uniformly annealed devices and devices with an unannealed β\beta-W film, complete magnetization reversal is only achieved for magnetic fields greater than 1 mT applied along the current direction due to the symmetry of the heterostructure (see Figure 3a), consistent with previous reports [5]. Without an applied in-plane field, no switching of the β\beta-W device is observed, as seen in Figure 3b with RHR_{\mathrm{H}} as a function of μ0Hz\mu_{0}H_{z} shown as a black dashed line, and RHR_{\mathrm{H}} as a function of the applied current density jj shown in blue.

Figure 3. (a) Schematic of spin-orbit torque switching in β\beta-W/Ta/CoFeB. A current jcj_{\mathrm{c}} is applied along the xx-direction. The spin Hall effect in W generates a spin current jsj_{\mathrm{s}} with yy-polarized spins due to symmetry about the xzxz plane. (b) RHR_{\mathrm{H}} of β\beta-W/Ta/CoFeB as a function of μ0Hz\mu_{0}H_{z} (black dashed line) and the change in RHR_{\mathrm{H}} with jj with no applied magnetic field (in blue). (c) Schematic of spin-orbit torque switching with a gradient in W phase from β\beta-W (in green) to mixed β\beta-W/α\alpha-W phase (in blue). (d) RHR_{\mathrm{H}} as a function of μ0Hz\mu_{0}H_{z} (black dashed line) and change in RHR_{\mathrm{H}} as a function of jj (in blue) without any applied in-plane field for a device with a gradient in W phase. (e) Reliability of field-free current-induced switching of CoFeB on W with a phase gradient. In the upper panel, RHR_{\mathrm{H}} of the device is recorded after a series of current pulses with the magnitude shown in the bottom panel. (f) μ0Hx\mu_{0}H_{x} dependence of jcj_{\mathrm{c}} for devices with (in turquoise) and without (in black) W phase gradients.

Figure S5. a) The critical switching current density jcj_c for devices uniformly annealed at different laser fluences with an applied magnetic field of 20 mT is plotted in black. The effective anisotropy strength μ0HKeff\mu_0 H_{\mathrm{K}}^{\mathrm{eff}} of the CoFeB layer for each device is shown in blue. b) The change in anomalous Hall resistance as a function of in-plane applied magnetic field μ0Hx\mu_0 H_x for W/CoFeB devices annealed at different laser fluences. The fit to Equation (3) for the β\beta-W/CoFeB device is indicated by the black dashed line.

Devices with a gradient in the W phase created transverse to the current direction (depicted by the schematic in Figure 3c) exhibit different behavior than the uniform devices. For a device with a gradient in phase going from 0% to 20% α\alpha-W across the 10 μ\mum width, the magnetization fully switches at j=1.4×1011j = 1.4 \times10^{11} A/m2^{2} without an applied magnetic field. Here, the current density in Figure 3d-Figure 3f is defined as the total applied current divided by the cross-sectional area. In Figure 3d, RHR_{\mathrm{H}} as a function of jj with no applied magnetic field (in blue) is compared to RHR_{\mathrm{H}} as a function of μ0Hz\mu_{0}H_{z} (black dashed line), and it is seen that RHR_{\mathrm{H}} switches between uniform up (down) magnetization states with positive (negative) current pulses. The reproducibility of this field-free switching is tested by applying a pattern of current pulses, seen in the lower panel of Figure 3e, and measuring RHR_{\mathrm{H}}, seen in the upper panel of Figure 3e. It is observed that RHR_{\mathrm{H}} reliably switches to the maximum (minimum) values measured in Figure 3d with positive (negative) current pulses, indicating reproducible and complete magnetization reversal. jcj_{\mathrm{c}} as a function of applied in-plane magnetic field μ0Hx\mu_{0}H_{x} is compared for a gradient device and as-grown device with no gradient in Figure 3f. jcj_{\mathrm{c}} at μ0Hx=0\mu_{0}H_{x}=0 mT (indicated by the vertical black dashed line in Figure 3f) for the gradient device is positive, matching the polarity of jcj_{\mathrm{c}} for the uniform device with −μ0Hx-\mu_{0}H_{x} applied. This suggests that the gradient creates an effective −μ0Hx-\mu_{0}H_{x} that aids field-free switching.

To verify the presence of unconventional torques, which would produce this field-free switching, measurements of hysteresis loop shifts are performed for three W/Ta/CoFeB devices [61]: one with a gradient in the β\beta-W phase with α\alpha-W going from 0% to 20% over the 10 μ\mum device width (the steep gradient), one with a gradient from 0% to 10% α\alpha-W (the shallow gradient), and one with no gradient. Here, a dc current is passed through the device and the polar MOKE response as a function of μ0Hz\mu_{0}H_{z} is measured in a Kerr microscope, with hysteresis loops at applied dc currents of ±1×1011\pm1 \times10^{11} A/m2^{2} shown in Figure 4a. The shift in magnetic field of the hysteresis loop between positive and negative dc current, termed μ0Hzeff\mu_{0}H_{z}^{\mathrm{eff}}, without an applied magnetic field along xx is characteristic of unconventional torques in the system. For the device with no gradient, there is no loop shift outside the error of the measurement (the green data in Figure 4b). Meanwhile, both gradient devices (the turquoise and blue data in Figure 4b) have a measurable μ0Hzeff\mu_{0}H_{z}^{\mathrm{eff}} that increases with increasing current density. It is also observed that, for both gradient devices, there is a threshold current below which there is no measurable hysteresis loop shift, similar to a film with a gradient in Ta phase [32] and films where low crystal symmetry generates zz-polarized spins [62, 63]. In addition, the steep gradient has both a lower current threshold for the onset of a nonzero μ0Hzeff\mu_{0}H_{z}^{\mathrm{eff}} and a stronger dependence of μ0Hzeff\mu_{0}H_{z}^{\mathrm{eff}} on the current density. This indicates that the steep gradient has stronger unconventional torques that overcome the damping of the CoFeB at a lower current density than the shallow gradient.

Figure 4. a) Example of hysteresis loop shift measurements performed at a current density of 1×10111 \times10^{11} A/m2^{2} for a 0%-20% α\alpha-W gradient across 10 μ\mum. The purple (black) hysteresis loop is for positive (negative) current direction. The difference between the center of the loops for positive and negative current is defined as μ0Hzeff\mu_{0}H_{z}^{\mathrm{eff}}. b) The current dependence of μ0Hzeff\mu_{0}H_{z}^{\mathrm{eff}} for a steep gradient (0%-20% α\alpha-W) and a shallow gradient (0%-10% α\alpha-W) in β\beta-W over 10 μ\mum. A device without a gradient exhibits no loop shift. Error bars are given by the magnetic field step size used to collect hysteresis loops.

To determine the mechanism of unconventional torque generation, two possibilities are considered.

First, zz-polarized spins can be generated in materials with broken spatial symmetry, leading to unconventional torques that can deterministically switch a neighboring magnet. Second, in the case of W films with a phase gradient, there are several asymmetries in properties over the device which may contribute:

(i) There is inhomogeneous current flow through the W due to the differing resistivity of the β\beta-W and mixed-phase W. The patterned gradients designed for field-free switching have (at most) 0% to 20% α\alpha-W fractions, so the maximum resistivity change would be 230 μΩ\mu\Omega-cm to 105 μΩ\mu\Omega-cm from one edge to the other (see Figure 2a, for a fluence of ∼10\sim10 J cm−2^{-2}).

(ii) The anisotropy strength of the CoFeB depends on the W phase and thus will differ across a gradient device. In order to quantify this asymmetry, planar Hall effect measurements of CoFeB with the W annealed at constant fluences are used to extract the effective magnetic anisotropy as a function of laser fluence. As the fraction of α\alpha-W increases, the effective magnetic anisotropy strength decreases by at most 25% (see Supporting Information Figure S5a,b).

(iii) There are thermal gradients produced by the non-uniform current flow and differing thermal conductivities between α\alpha-W and β\beta-W.

These non-uniform physical properties could combine with the spin-orbit torque gradient to switch the magnetization without an applied field. To examine the relative likelihood of each of these possibilities, a combination of COMSOL Multiphysics® software [64] and MuMax3 [65] are used to simulate W/CoFeB properties. For these simulations, a current density of 3×10113 \times10^{11} A/m2^2 through the W layer and a gradient from 0% to 100% α\alpha-W over 10 μ\mum were employed to test if the combined current density gradient, anisotropy gradient, and inhomogeneous Oersted field are sufficient for field-free switching. This simulated gradient is steeper than the gradients used in the experiments for field-free switching in order to observe the largest possible changes in current-induced asymmetries across the device. In these simulations, no deterministic magnetization reversal is observed (see Supporting Information § 12 and Supporting Information Figures S6 and S7 for modeling of current density, non-uniform Oersted field, and micromagnetic simulations). While there is a contribution from thermal gradients, it is not expected to be sufficient for field-free switching based on published works with current density gradients, where an estimated 40 K temperature gradient did not play a dominant role in magnetization reversal [24].

While anisotropy and current density gradients are not adequate to produce field-free switching in micromagnetic simulations, deterministic field-free switching is demonstrated in simulations by including zz-polarized spins (see Supporting Figure S7). This raises the question of what the underlying physical origin is for zz-polarized spin generation in polycrystalline films where each of the two crystalline phases is high-symmetry. Broken electronic symmetry has recently been proposed as a way to generate zz-polarized spins [25], and the resistivity difference in the W phases may create this electronic asymmetry. Another possibility is an asymmetry resulting from the change in microstructure on annealing. As seen in the TEM images, while the α\alpha-W and β\beta-W unit cell are each high-symmetry, there are many grain boundaries with an α∣β\alpha|\beta interface (see Figure 1), creating lateral interfaces between different crystal structures and spin Hall conductivities [66], which may provide the necessary asymmetry for zz-polarized spin generation.

Conclusions

Local patterning of the crystalline phase of W with direct-write laser annealing provides a means to tune spin-orbit torque efficiency and current-induced magnetization switching. The density and size of α\alpha-W grains in a β\beta-W film can be adjusted on the nanoscale and, as a result, the film resistivity can be modified by almost an order of magnitude, and the strength of spin-orbit torques can be tuned from 0.45 to ∼0.07\sim0.07. Gradients in these properties are leveraged to switch the magnetization of a neighboring CoFeB layer by current without an applied magnetic field. The large resistivity difference between phases can be employed to precisely control the current magnitude and direction at all points in a wire. By combining the field-free switching shown here with non-uniform current directions, ultrahigh efficiency of spin-orbit torque switching is likely [67], providing a new route for energy-efficient spintronic devices. Furthermore, given the broad applications of W films across various industries, the ability to locally tune the crystalline phase of a film with accompanying changes in resistivity, thermal conductivity, superconductivity, and optical properties holds great promise for applications beyond spintronics.

Experimental Section

Film growth

5-15 nm-thick W films are grown by magnetron sputtering on Si substrates with 100 nm-thick SiOx coating. For TEM measurements, the films are instead grown on a 200 nm-thick silicon nitride membrane. To ensure β\beta-phase growth in films thicker than 6 nm, the chamber was exposed to 1 sccm N2\mathrm{N}_{2} flow for 5 minutes before growth to increase the chamber base pressure, as high sputtering base pressure is known to assist stabilization of the β\beta-W phase [68] [41]. Because the laser fluence required to change the W phase is quite large and would likely damage the magnetic properties of any magnet grown underneath W, the W is always grown on the substrate and annealed first, then placed back into the chamber for growth of any additional layers.

Transmission electron microscopy

15 nm-thick W films were sputtered on a 200 nm-thick silicon nitride membrane for TEM measurements. After growth, the films were laser annealed. Because the thermal conductivity of the membrane is significantly lower than that of the 500 μ\mum-thick thermally-oxidized Si substrate, the laser fluences used on the membrane were also much lower (5.1 J cm−2^{-2} compared to the 31 J cm−2^{-2} used on the thermally-oxidized Si substrate) to reproduce the same transformations from β\beta-W to α\alpha-W. The equivalence in terms of the transformations obtained by these laser fluences was verified by comparing x-ray diffraction on annealed W films grown on the thermally-oxidized Si substrates to electron diffraction of annealed regions of W films grown on silicon nitride membranes.

Electrical measurements

For resistivity measurements, 10 μ\mum-wide Hall bars with a 15 μ\mum distance between the longitudinal voltage contacts are fabricated from W films on thermally oxidized Si substrates using UV photolithography in combination with liftoff. The resistivity of each Hall bar device is measured by passing through a 0.1 mA current and measuring the voltage across two longitudinal voltage leads. Then, different Hall bars are annealed uniformly with DWLA at different laser fluences. The resistivity of each device is then re-measured such that the resistivity can be directly compared before and after annealing.

For second harmonic Hall measurements, 10 μ\mum-wide W Hall bars with a 10 μ\mum distance between the longitudinal voltage contacts are uniformly annealed at different laser fluences. Then, a window in photoresist above the Hall bar is defined with photolithography. The W surface within the window is cleaned with low-power Ar plasma etching. Subsequently, an in-plane magnetized 2 mm CoFeB/1 nm MgO/3 nm Ta film is sputter deposited, and then all unwanted material outside the window is removed with liftoff. This sequence of fabrication steps, with the CoFeB film deposited last, ensures that the magnetic properties are not altered by the laser annealing. For spin-orbit torque switching measurements, the 10 μm×10\mu\mathrm{m} \times10 μm\mu\mathrm{m} W Hall bars are annealed with a laser fluence gradient transverse to the current direction, producing a phase gradient from pure β\beta-W to mixed phase W. Subsequently, the W surface is cleaned with Ar plasma etching and an out-of-plane magnetized 1 Ta/0.9 CoFeB/1 MgO/2 Ta film is deposited on the W film. The Ta seed layer is used to ensure that the CoFeB has perpendicular magnetic anisotropy as-grown. The CoFeB elliptical dots with minor (major) axis of 5.7 μm\mu\mathrm{m} (10 μm\mu\mathrm{m}) are then defined with photolithography and Ar plasma etching. After applying a current pulse, the anomalous Hall resistance is measured with a small test current of 0.1 mA.

Thermoreflectance measurements

The reflectivity and thermoreflectance of films was probed with a 520 nm-wavelength laser beam that was focused through an objective lens (100×\times magnification, NA = 0.9), resulting in a spot size of ∼0.5\sim0.5 μm\mu\mathrm{m} [69]. The incident beam was set to be linearly polarized and at normal incidence to the film’s surface. The reflected light intensity RR was detected using a Thor labs DET36A photodetector. The laser beam was scanned along a line transverse to the current direction.

Static (time-averaged) and dynamic responses of the system to a current-induced stimulus are resolved by applying alternating (sine wave) currents and demodulating the signal from the photodetector using a lock-in amplifier (LIA). The harmonic changes of the reflectivity are linked to the temperature dependence of the film’s refractive index nn. The temperature is modified due to Joule heating which scales with the square of applied current density [60]. Therefore, the harmonic change in the reflectivity is completely captured by the second harmonic signal from the photodiode R2ωR^{2\omega}. Thermoreflectance relates the normalized change in reflectivity R2ω/RR^{2\omega}/R to the change in temperature ΔT\Delta T through: R2ω/R=cTRΔTR^{2\omega}/R = c_{\mathrm{TR}}\Delta T, where cTRc_{\mathrm{TR}} is the material and wavelength dependent thermoreflectance coefficient [70].

Acknowledgements

Each author contributed to this work as follows: L.J.R. and A.H. conceived the project with input from L.J.H. L.J.R. and A.F. synthesized, laser annealed, and characterized W films and devices. A.F. performed x-ray diffraction and resistivity measurements. E.M. performed TEM measurements, and the TEM data was analyzed by L.J.R. S.D. performed second harmonic Hall resistance measurements, and the data was analyzed by S.D. and L.J.R. E.K., N.K., and T.G. performed reflectivity and thermoreflectance measurements and analyzed the data, with help from P.G. L.J.R. performed spin-orbit torque switching and loop shift measurements and analyzed data with input from A.H. and P.G. L.J.R. wrote the manuscript with help from A.H. and L.J.H. All authors contributed to finalizing the manuscript. L.J.R acknowledges funding from the ETH Zurich Postdoctoral Fellowship Program 22-2 FEL-006. The Laboratory for Magnetism and Interface Physics acknowledges support from SNF Grant No. 200021-236524. We thank the cleanroom staff at the Laboratory for Nano and Quantum Technologies at the Paul Scherrer Institute for technical support. We acknowledge the usage of the instrumentation provided by the Electron Microscopy Facility (EMF) at PSI and we thank the EMF team for their help and support. We also thank Min-gu Kang, Paul Noël, and Aurélien Manchon for helpful discussions. The raw data that supports this study will be available at the Zenodo repository https://doi.org/10.5281/zenodo.17937763.

Further structural characterization

X-ray diffraction scans of 14 nm-thick W films grown on thermally oxidized Si substrates confirm that the films, as-grown, are composed of only β\beta-W, as seen in Figure S1 (green data). After laser annealing at high fluence of ∼25 J cm−2\sim25\ \mathrm{J}\ \mathrm{cm}^{-2}, the film can be transformed completely to α\alpha-W, confirmed by the disappearance of β\beta-W peaks and the appearance of a strong α\alpha-W peak (blue data). The stability of the as-grown β\beta-W films was confirmed by measuring a series of x-ray diffraction scans over several months. It is observed that, even 5.5 months after growth, only β\beta-W diffraction peaks are detected in XRD data, seen in Figure S1.

Figure S1. a) X-ray diffraction scans for 14 nm-thick W films grown on thermally oxidized Si substrates before (in green) and after (in blue) laser annealing at ∼25 J cm−2\sim25\ \mathrm{J}\ \mathrm{cm}^{-2}. In the as-grown film, only film peaks corresponding to β\beta-W are detected. After laser annealing, only the α\alpha-W diffraction peak is visible. b) X-ray diffraction scans of an as-grown 14 nm-thick W film recorded at 0, 38, and 169 days after growth. For all scans, only the β\beta-W film peaks are visible. This confirms that the metastable β\beta-W films are stable over several months.

The grain sizes of β\beta-W and α\alpha-W are estimated by measuring individual grains in transmission electron microscopy (TEM) images taken at 120,000×120{,}000 \times magnification. β\beta-W grains are quite small, with sizes measuring from ∼5−20\sim5 - 20 nm (Figure S2), while α\alpha-W grains are are ∼80−120\sim80 - 120 nm in size (Figure S2). The diffraction intensity as a function of scattering vector shown in Figure 1e was generated by taking the radial average of selected-area electron diffraction (SAED) images taken at different locations along the gradient. Four representative SAED images, with each next to the TEM images of the area where they were recorded, are shown in Figure S2-Figure S2. The aperture for the SAED images was 1.21.2 μ\mum.

Electrical characterization of mixed phase stability

The longitudinal resistance RR of a 10 µm-wide and 10 µm-long W wire with a lateral gradient in β\beta-W, going from 0% to 50% α\alpha-W, was tracked over 150 days. After 150 days, the resistance had increased by ∼\sim 4.5%, as shown in Figure S3. Since a transformation of β\beta-W to α\alpha-W would cause a decrease in resistance, this confirms that annealed, mixed-phase W remains stable, and the nucleated grains of the stable α\alpha-W crystalline phase do not cause the remainder of β\beta-W to transform to α\alpha-W. The small increase in resistance is attributed to surface oxidation of W.

Details of second harmonic Hall measurements

The spin-orbit torque efficiency as a function of α\alpha-W fraction, seen in Figure 2c, was extracted using second harmonic Hall measurements on 14 nm-thick W devices that have been uniformly annealed. 2 nm of CoFeB, which has in-plane anisotropy, with a 1 nm MgO/3 nm Ta cap was deposited on top of the W device as the magnetic layer. We employ the standard analysis technique to extract the spin-orbit torques from second harmonic Hall data as described in Ref. [71] [56]. The current and magnetic field dependence of the second harmonic Hall signal is fit to extract the damping-like field and associated damping-like torque efficiency. Representative data for the as-grown W/2 nm CoFeB device is shown in Figure S4.

The angular dependence of the second harmonic Hall resistance, Rxy2ω(φ)R_{xy}^{2\omega}(\varphi), is shown in Figure S4, and the data is fit to the following model [56]:

Rxy2ω(φ)=(12RAHEBDLμ0H+μ0HKeff+R∇T)cos⁡φ+RPHEBFL+BOeμ0H(2cos⁡3φ−cos⁡φ)(1)R_{xy}^{2\omega}(\varphi) = \left(\frac{1}{2}R_{\mathrm{AHE}}\frac{B_{\mathrm{DL}}}{\mu_0 H+\mu_0 H_{\mathrm{K}}^{\mathrm{eff}}}+R_{\nabla T}\right)\cos\varphi+R_{\mathrm{PHE}}\frac{B_{\mathrm{FL}}+B_{\mathrm{Oe}}}{\mu_0 H}\left(2\cos^3\varphi-\cos\varphi\right) \tag*{(1)}

where φ\varphi is the angle between the current direction and the applied magnetic field, RAHER_{\mathrm{AHE}} is the anomalous Hall resistance, μ0H\mu_0 H is the applied magnetic field, μ0HKeff\mu_0 H_{\mathrm{K}}^{\mathrm{eff}} is the effective anisotropy field, R∇TR_{\nabla T} is the anomalous Nernst resistance, BDLB_{\mathrm{DL}} and BFLB_{\mathrm{FL}} are the damping-like and field-like fields, and BOeB_{\mathrm{Oe}} is the Oersted field contribution.

For each device, RAHER_{\mathrm{AHE}} is extracted with a separate measurement of the Hall resistance RHR_{\mathrm{H}} as a function of applied out-of-plane magnetic field μ0Hz\mu_0 H_z, seen in Figure S4. The difference between the two dashed lines in Figure S4, which are linear fits to the high-field region of RHR_{\mathrm{H}}, gives 2RAHE2R_{\mathrm{AHE}}, and the field where RHR_{\mathrm{H}} reaches this saturated value is defined as μ0HKeff\mu_0 H_{\mathrm{K}}^{\mathrm{eff}}. The magnitude of the cos⁡φ\cos\varphi term scaled by RAHER_{\mathrm{AHE}} is plotted as a function of (μ0H+μ0HKeff)−1(\mu_0 H+\mu_0 H_{\mathrm{K}}^{\mathrm{eff}})^{-1} in Figure S4 for five different currents. The slope of this linear fit is proportional to BDLB_{\mathrm{DL}}, which is plotted as a function of the current flowing through W (IWI_{\mathrm{W}}) calculated using a parallel resistor model, in Figure S4. BFL+BOeB_{\mathrm{FL}}+B_{\mathrm{Oe}} is obtained with the same procedure (plotting the magnitude of the 2cos⁡3φ−cos⁡φ2\cos^3\varphi-\cos\varphi as a function of H−1H^{-1}). We estimate the magnitude of the Oersted field from BOe≈μ0IW/2wHBB_{\mathrm{Oe}}\approx\mu_0 I_{\mathrm{W}}/2w_{\mathrm{HB}} where wHBw_{\mathrm{HB}} is the 10 μ\mum Hall bar width. Finally, the slope of the linear relationship between BDL(FL)B_{\mathrm{DL(FL)}} and IWI_{\mathrm{W}} is directly proportional to the damping-like (field-like) torque efficiency ξDL(FL)j\xi_{\mathrm{DL(FL)}}^j through the following relation:

ξDL(FL)j=2eℏMstCFBBDL(FL)AIW(2)\xi_{\mathrm{DL(FL)}}^j=\frac{2e}{\hbar}\frac{M_{\mathrm{s}}t_{\mathrm{CFB}}B_{\mathrm{DL(FL)}}A}{I_{\mathrm{W}}} \tag*{(2)}

where MsM_{\mathrm{s}} is the saturation magnetization of CoFeB, tCFBt_{\mathrm{CFB}} is the thickness of CoFeB, and AA is the cross-sectional area of the device. ξFLj\xi_{\mathrm{FL}}^j is shown in Figure S4. For β\beta-W, ξFLj≈0\xi_{\mathrm{FL}}^j\approx0. Devices with a small fraction of α\alpha-W appear to have a significant field-like torque, but this may be due to a current distribution in the device where the Oersted field approximation fails. Thus, for micromagnetic simulations of W/CoFeB, ξFLj\xi_{\mathrm{FL}}^j is set to 0.

Spin-orbit torque switching

The critical switching current jcj_c of out-of-plane magnetized W/0.9 nm CoFeB devices with the W layer uniformly annealed in each device at a different laser fluence was determined by tracking the changes in RHR_{\mathrm{H}} with current. As the laser fluence, and thus the fraction of α\alpha-W, was increased, jcj_c also increases significantly (black data in Figure S5). A decrease in μ0HKeff\mu_0 H_{\mathrm{K}}^{\mathrm{eff}} of the CoFeB layer is also observed due to a decrease of perpendicular magnetic anisotropy (shown in blue).

The effective anisotropy strength of CoFeB on different W devices shown in Figure S5 is extracted with planar Hall effect measurements, seen in Figure S5. The change in RHR_{\mathrm{H}} as a function of μ0Hx\mu_0 H_x is fit to the following equation:

RH=R01−(μ0Hxμ0HKeff)2(3)R_{\mathrm{H}} = R_0 \sqrt{1 - \left(\frac{\mu_0 H_x}{\mu_0 H_{\mathrm{K}}^{\mathrm{eff}}}\right)^2} \tag*{(3)}

where R0R_0 is a scaling factor, and μ0HKeff\mu_0 H_{\mathrm{K}}^{\mathrm{eff}} is given by μ0HKeff=μ0(Ms+HK⊥)\mu_0 H_{\mathrm{K}}^{\mathrm{eff}} = \mu_0(M_s + H_{\mathrm{K}}^{\perp}) where HK⊥H_{\mathrm{K}}^{\perp} is the perpendicular anisotropy field. μ0HKeff\mu_0 H_{\mathrm{K}}^{\mathrm{eff}} is defined as positive (negative) when the film has perpendicular (in-plane) anisotropy. An example of the fit is indicated by the black dashed line in Figure S5.

Finite element and micromagnetic simulations

COMSOL Multiphysics® software [64] is used to simulate the current density and Oersted field of W devices with phase gradients. A 6 nm-thick W film with a gradient from 0% to 100% α\alpha-W over 10 μ\mum was simulated, which is much steeper than the 0% to 10% and 0% to 20% α\alpha-W gradients used in experimental devices. This steep gradient was chosen in order to observe the largest possible changes in properties across the device (see schematic in Figure S6). The nominal current density jj of 3×10113 \times10^{11} A/m2^{2} (see Figure S6), which is slightly higher than the current used for field-free switching experiments, is applied for 50 μ\mus. This was used to simulate the highest possible Oersted field (Figure S6) to which the device could be exposed in experiments. The current density and Oersted field are shown at two xx positions along the Hall cross: at x1x_1, where the current density flows through a straight region, and at x2x_2, where the wires for Hall voltage detection are placed. This accounts for changes in the current density due to the Hall voltage leads.

Figure S6. a) Schematic of device geometry for COMSOL simulations. A linear gradient in β\beta-W going from 0% to 100% α\alpha-W across a 10 μ\mum-wide, 6 nm-thick W wire is simulated, and the current density and Oersted field are evaluated at two positions on the Hall cross: x1x_1, where the device is 10 μ\mum wide (black solid line), and x2x_2, at the location of the wires for Hall voltage detection (black dashed line). b) jj across the device for x1x_1 (black solid line) and x2x_2 (black dashed line). c) The zz-component of the Oersted field BzB_z for the current density shown in (b).

BOeB_{\mathrm{Oe}} and jj as a function of yy (geometry given in Figure S6) are input into a MuMax3 [65] simulation to determine if these conditions are sufficient for field-free magnetization reversal. Changes in CoFeB anisotropy across the laser fluence gradient, seen in Figure S5, are also included. A magnetic simulation volume of 250 nm ×\times 400 nm ×\times 1 nm (x×y×zx \times y \times z) divided into 2 nm cells is implemented with Ms=700M_s = 700 kA/m and the exchange energy Aex=1×10−11A_{\mathrm{ex}} = 1 \times10^{-11} J/m. Unless otherwise stated, the Gilbert damping α\alpha is set to 0.20. 50 regions with length along xx of 250 nm and width along yy of 8 nm are defined, and each region has the associated current density, Oersted field, anisotropy energy, and spin-orbit torque efficiency corresponding to the phase gradient going from β\beta-W to α\alpha-W. Because a module for spin-orbit torques is not available, they are modeled using the spin transfer torque module. This is possible because a current along the zz-direction for spin transfer torques is equivalent to a current along the xx-direction for spin-orbit torques. The orientation of the polarized spins is set by the magnetization orientation of the magnetic “fixed layer” in the spin transfer torque module, and only the damping-like torque is considered (the field-like torque is set to zero with the “epsilon prime” parameter). After the system is relaxed for 0.2 ns, positive jj, with densities scaled to account for the current distributions shown in Figure S6, are applied for 1 ns, and the non-uniform BOeB_{\mathrm{Oe}} associated with the current shown in Figure S6 is also included. Then, jj and BOeB_{\mathrm{Oe}} are removed, and the system is allowed to relax for 5 ns. The out-of-plane magnetization mzm_z is recorded as a function of time during this process.

When the spin polarization was set so that there are only conventional yy-polarized spins, the comcombined changes in current density, Oersted field, magnetic anisotropy, and spin-orbit torques across the device were not sufficient to switch the magnetization at both location x1x_1 (solid line in Figure S7) and x2x_2 (dashed line in Figure S7). Complete magnetization reversal from mz=+1m_z = +1 to mz=−1m_z = -1 was only achieved when zz-polarized spins were added to the simulation, seen in Figure S7 (at x1x_1 with the solid line and at x2x_2 with the dashed line). When the magnetization is initialized to the mz=−1m_z = -1 state, the current does not switch the magnetization, as expected for deterministic reversal.

Figure S7. Time traces of mzm_z before, during, and after application of a 1 ns square current pulse (applied at t=0.2t=0.2 ns) obtained from micromagnetic simulations including both jj and BOeB_{\mathrm{Oe}}. The shaded pink region indicates the time period where the current pulse is applied. The solid (dashed) lines are simulated at position x1x_1 (x2x_2). jj and BOeB_{\mathrm{Oe}} for each position are taken from Figure S6 and S6c, respectively. a) Response of mzm_z to spin-orbit torques starting from mz=+1m_z=+1 and mz=−1m_z=-1, assuming only yy-polarized spins, where no magnetization reversal is observed. b) Response of mzm_z to spin-orbit torques starting from mz=+1m_z=+1 and mz=−1m_z=-1, assuming yy- and zz-polarized spins, where deterministic magnetization reversal is only observed when the initial state of mz=+1m_z=+1. c) The magnetization reversal for yy-polarized spins (in grey) and yy- and zz-polarized spins (in blue) for different values of α\alpha (ΔKu=1.5×105\Delta K_u=1.5 \times10^5 J/m3^3) at x1x_1 starting at mz=+1m_z=+1. For all α\alpha values, switching is only observed when zz-polarized spins are included. d) The magnetization reversal for two scenarios: yy-polarized spins (in grey) and yy- and zz-polarized spins (in blue) for different ΔKu\Delta K_u across the device (α=0.20\alpha=0.20) at x1x_1 starting at mz=+1m_z=+1. Full magnetization reversal is only seen when zz-polarized spins are included (in blue).

To check the robustness of the micromagnetic simulations, the same simulation is carried out for three different Gilbert damping (α\alpha) values. The magnetization dynamics at x1x_1 and x2x_2 are nearly identical (Figure S7,Figure S7), so additional simulations were only carried out at x1x_1. Regardless of the damping value used, deterministic magnetization reversal is only observed when zz-polarized spins are included (blue lines in Figure S7). Finally, the impact of a thermal gradient is estimated by doubling the anisotropy gradient across the device from ΔKu=(8.5 to 7.5)×105 J/m3\Delta K_u = (8.5\ \text{to}\ 7.5) \times10^5\ \mathrm{J}/\mathrm{m}^3 to ΔKu=(8.5 to 6.5)×105 J/m3\Delta K_u = (8.5\ \text{to}\ 6.5) \times10^5\ \mathrm{J}/\mathrm{m}^3. When only yy-polarized spins are used, the magnetization relaxes to a multidomain state, seen by the fact that mzm_z plateaus at ∼0.6\sim0.6 (grey line in Figure S7). When zz-polarized spins are included, deterministic switching is still achieved (blue lines in Figure S7).

References

  1. [65]Arne Vansteenkiste, Jonathan Leliaert, Mykola Dvornik, Mathias Helsen, Felipe Garcia-Sanchez, and Bartel Van Waeyenberge. The design and verification of MuMax3. AIP Advances, 4(10), 2014.DOI
  2. [66]Fei Xue and Paul M Haney. Staggered spin Hall conductivity. Physical Review B, 102(19):195146, 2020.DOI
  3. [67]Sergei M Vlasov, Grzegorz J Kwiatkowski, Igor S Lobanov, Valery M Uzdin, and Pavel F Bessarab. Optimal protocol for spin-orbit torque switching of a perpendicular nanomagnet. Physical Review B, 105(13):134404, 2022.DOI
  4. [68]Jiaxing Liu and Katayun Barmak. Topologically close-packed phases: deposition and formation mechanism of metastable β-W in thin films. Acta Materialia, 104:223–227, 2016.DOI
  5. [69]Christian Stamm, Christoph Murer, Marco Berritta, Junxiao Feng, Mihai Gabureac, Peter M Oppeneer, and Pietro Gambardella. Magneto-optical detection of the spin Hall effect in Pt and W thin films. Physical Review Letters, 119(8):087203, 2017.DOI
  6. [70]T Favaloro, J-H Bahk, and Ali Shakouri. Characterization of the temperature dependence of the thermoreflectance coefficient for conductive thin films. Review of Scientific Instruments, 86(2), 2015.DOI
  7. [71]Shilei Ding, Min-Gu Kang, William Legrand, and Pietro Gambardella. Orbital torque in rare-earth transition-metal ferrimagnets. Physical Review Letters, 132(23):236702, 2024.DOI

Paper details

vlnc:W_x36fb9zt
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents