Spin-torque switching of a nano-magnet using giant spin hall effect

Since the discovery of spin transfer torque^1,2 as a means to alter magnetization, there has been a push to utilize nano-magnets to perform logic operations.^3,4 These proposals utilize the inherent non-volatility of magnets to build devices that are well-suited to performing non-Boolean computation.^5,6 The spin-switch proposed by Datta et al⁴ is one such device. A crucial component underlying the spin-switch as well as other spin-logic devices is an efficient means of electrically generating spin currents. The earliest approach to do this involves passing a charge current from a ferromagnet into a non-magnetic material.^7–10 The ferromagnet acts as a polarizer and preferentially injects spins of only one kind, creating a spin accumulation region in the non-magnetic material. For logic operations, once the spin current is generated it needs to propagate through a non-magnetic channel to communicate information between neighboring units which leads to further reduction in the spin current before it reaches the detector. To avoid this signal degradation and to generate spin currents more efficiently, the Giant Spin Hall Effect (GSHE) in some metals with high spin-orbit coupling^11–16 and Topological insulators^17–19 have been proposed as prime candidates to generate spin currents at room temperature. In the metals showing GSHE, longitudinal charge current flowing through a metal strip generates a transverse spin current at the surfaces of the metal strip due to high spin-orbit coupling. This spin current can be used to torque a nano-magnet in contact with the surface without the need for the spins to propagate over long distances. The efficiency of the charge-to-spin conversion in this method has been measured to be around 15-30%^11,12 depending on the GSHE material used. Recently, higher values have been reported for metals doped with impurities.¹³ In this letter, we report the experimental realization of the write-unit of the spin-switch proposed by Datta et al⁴ by demonstrating the current-induced switching of a CoFeB nano-magnet using GSHE of a β-Ta layer.

A Metal stack comprising of Ta(17) ∣ CoFeB(2) ∣ MgO(0.5) ∣ CoFeB(4) ∣ Ru+Ta cap(14) was sputtered as a blanket film onto a thermally oxidized Silicon wafer (all thicknesses in nm). A High-Resolution Transmission Electron Micrograph in fig. 1(a) shows the stack used for this study. The films were first patterned into strips using optical lithography with an AZ1518 mask and dry etching using Argon plasma. In this step, the sample was etched down to the SiO₂, leaving strips of the sputtered metal stack in the device region. Using electron-beam lithography, a bi-layer of HSQ ∣ PMMA was patterned into a hard-mask to etch elliptical Magnetic Tunnel Junction (MTJ) pillars of dimension 400 x 200 nm² using Argon plasma. The etching was timed so that only the layers (Capping layer ∣ CoFeB ∣ MgO ∣ CoFeB) down to the bottom Ta layer were etched. With the hard mask in place, a 90nm thick SiO_x film was deposited and lifted-off in the region directly above the MTJ using PG Remover at 85^∘C. The deposited SiO_x electrically isolates the GSHE Tantalum and the top electrical contact to the MTJ. With an additional optical lithography step, an electrical contact to the top of the MTJ was made using lift-off of electron-beam evaporated Ti ∣ Au. The final structure and a 3-dimensional cartoon of the device are shown in fig. 1(c), 1(d). The final device was annealed at 300^∘C in an inert Argon ambient for one hour to improve the switching characteristics of the MTJ.

In previous studies,^11,12,14 it was reported that in Tantalum and Tungsten, only the ultra-thin (typically 6nm-8nm), high-resistive β-phase shows the GSHE needed for charge-to-spin conversion through spin-orbit interaction. In order to confirm that the sputter deposited tantalum used in this study has the desired phase, we studied the crystal structure of sputter-deposited Tantalum films by using a Bruker D800 focus X-ray powder diffractometer with Cu K_α1 radiation. All the peaks indexed were assigned according to the tetragonal crystal structure of β-Ta and are verified from the ISDD standard data for β-Ta.The XRD spectra (fig. 1(b)) show that the films are poly-crystalline with the position of the peaks suggesting that for a reasonable range of thicknesses, the β-phase of Tantalum can be obtained. β-Ta crystallizes in the tetragonal phase with lattice parameters a = 10.194 Å   c = 5.313 Å. Using the measured XRD spectra we performed lattice parameter calculations along the primary diffracted direction. Comparing the lattice constants of the standard β-Ta unit cell with the extracted unit cell parameters for 10nm and 20nm film, we observed that there is an in-plane lattice compression in the Ta film for thicker films. The calculated lattice constants for 20nm and 10nm were a = 9.713 Å, c = 5.368 Å and a = 9.759 Å, c = 5.340 Å respectively. The crystal structure of the sputtered thin films was also confirmed through resistivity measurements of films over a thickness range 5nm - 20nm. Nearly independent of thickness, we found that the resistivity of the films was 190 - 200 μΩ-cm, which is characteristic of β-Ta.²⁰ So, we can conclude that the Tantalum used in our structure is indeed β Tantalum.The peaks for the 5nm thick film were too broad to show any preferred crystalline orientation. Different from earlier reports^11,12 that demonstrate spin switching using the Giant Spin Hall Effect, we observe that the β-phase of Tantalum persists for thicker films. Increasing the thickness of the Tantalum beyond the spin diffusion length (≈ 2nm)²¹ is not expected to contribute any additional spin current; however, thicker Tantalum may allow integration of this technology on different substrates where roughness might not allow the use of an ultra-thin Tantalum layer.

The measurements that are discussed below were performed on a 400 x 200nm² MTJ pillar patterned on a 17nm thick Ta layer. The width of the Tantalum strip is 6μm. The differential resistance of the MTJ pillar was measured using a standard low frequency a.c. lock-in technique with a sense current of 1μA. The measurement schematic is shown in figure 2.

Fig. 2(a) shows the measurement set-up for characterizing the resistance state of the MTJ using a lock-in amplifier and fig. 2(b) shows the additional connections in the set-up to toggle the magnet m using spins generated by passing a current I_SH through the Tantalum strip. An additional current source was connected across the Tantalum strip to pass a current pulse through it. Two separate measurements were performed on the device. In the first measurement, the set-up in fig. 2(a) was used to measure the differential resistance of the MTJ as a function of magnetic field swept in the x-direction(see Fig. 2). In the second measurement, a combination of a current pulse of the correct polarity and an external magnetic field was used to toggle the state of the magnet m. The schematic for the measurement set-up for this write action is shown in fig. 2(b). We denote current flowing in the positive (negative) y-direction through the Tantalum strip as positive (negative). As shown in the schematic, applying a negative current pulse creates an accumulation of spins on the surface of the tantalum polarized along the +x direction. Since the magnet, m, is magnetized in the negative x-direction it experiences a torque due to the accumulated spin. When the current I_SH exceeds a threshold value, the resulting torque aligns the magnetization of m in the direction of the spins accumulated at the m ∣ Tantalum interface.The resistance of the MTJ (R_MTJ) can be used as an indicator of the direction of magnetization of the magnet m with respect to M. Since we only apply a finite duration pulse, if we bias the MTJ at a magnetic field where m has a single, stable equilibrium state, the magnet relaxes to this state after the current pulse is removed. To ensure that only the current pulse determines the final state of m, the system should be biased in a B-field range where it is bi-stable i.e. it can exist in either of two stable states depending on the history of the sample. To summarize, the second measurement consisted of two steps - First, a current pulse was applied to impact the magnetization of m and second, the set-up shown in Fig. 2(a) was used to measure R_MTJ and sense if m has switched due to the pulse. Both of these steps were performed at a magnetic field where m is in a bi-stable state.

Fig. 3(a) shows the differential resistance of the MTJ as a function of magnetic field. As the field is swept from large positive to negative fields the two magnets switch at different external fields due to the difference in their thicknesses. The different magnetic configurations for the forward sweep and the reverse sweep are shown in fig.3(a). Please note that the magnetization directions in the insets of all the figures are drawn following the x,y,z-axes illustrated in fig.2.The Anti-parallel state (AP-state) has a higher resistance (R_AP) than the Parallel-state (P-state, R_P). As can be seen from the different configurations, both magnets can be switched at a sufficiently large external magnetic field. Also from the measured value of R_MTJ, one can see that the Resistance-Area(RA) product is lower than the value for typical MTJs. This low resistance is due to re-deposition of metal particles during the dry-etch patterning of the pillar. This reduces the amount of current flowing through the MgO barrier, which in turn reduces the Magneto-Resistance(MR) of the MTJ.

Since the goal is to switch magnet m using GSHE, the B-field sweep was restricted to values between -25mT to 0mT and R_MTJ was measured during the forward and reverse sweep (fig. 3(b)). In this range, the top magnet(M) always points in the negative x-direction. Depending on the direction of the B-field sweep, the bottom magnet(m) switches at a different field value. In the region between -10.5mT and -5.5mT (shaded pink in fig. 3(b)), m is bi-stable, i.e. the magnet has two stable minimum configurations separated by an energy barrier.As the external B-field is swept from -25mT towards zero, the magnet m switches abruptly to the +x-direction. However, when the B-field is swept back towards -25mT from zero, the MTJ enters a meta-stable state characterized by a resistance smaller than the AP-state before switching abruptly to the –x-direction at approximately 11mT. This meta-stable state does not occur in the major loop for forward or reverse sweeps. For the current induced switching due to a finite duration pulse to dictate the final state of magnet m, one has to bias the MTJ at a magnetic field in this range; so that once the pulse is removed, m retains its state. Ideally, the center of the hysteresis loop should be at zero-external field. However, as is generally the case for structures without a well-optimized synthetic anti-ferromagnet,^11,12 the magnetic stray field/dipolar field of M offsets the hysteresis loop along the B-field axis. When the MTJ is in the P-state (see fig. 3(b)) at a large, negative B-field, the stray field from magnet M points along the +x direction, aiding the external field in the switching of magnet m. So, m switches at a field μ₀(H_c − H_dip), where H_c and H_dip are the intrinsic coercive field of the m and the dipolar field on m, originating from magnet M. Similarly, starting from the AP-state at 0 mT, m switches at a magnetic field μ₀(H_c + H_dip), because, the dipolar field from M counters the effect of the external field.

Fig. 4 shows the results of the current-induced switching. A positive (negative) current results in accumulation of spins polarized along the negative (positive) x-direction. The MTJ was initialized in its P-state in the bi-stable region by sweeping the magnetic field from -25mT to -9mT /-8mT/ -7mT. After a 5 ms current pulse was passed through the tantalum strip, the resistance of the MTJ was measured using the read-circuitry. To make sure the value of the resistance was stable, R_MTJ was measured ten times after the current pulse was applied. Note that all ten measurements of the resistance in fig.4 yielded nearly the same value (the data points lie on top of each other at every current density value). Current pulses starting from -2MA/cm² to -25MA/cm² were passed through the tantalum at each magnetic field. A successful switching event due to spin torque from the GSHE is expected to result in an increase in resistance of the MTJ to the value corresponding to the AP-state at this field (see fig.3(b)). P → AP switching was not observed at a bias field of -9mT or -8mT. When the device was initialized in the P-state at -7mT, a clear increase in R_MTJ is apparent when a current pulse of J_C = -15MA/cm² is applied. The final resistance after switching matched the resistance of the MTJ’s AP-state at this field in fig.3(b). This measured resistance corresponds to the meta-stable AP-state in the minor loop. Note that the direction of the Oersted field due to the current pulse is expected to torque m in a direction opposite to the torque due to spins created by the GSHE in β-Tantalum.^11,14 Also, self-heating can be ruled out as a probable cause of the switching since the magnet did not switch at the same current density range when the direction of the current flow was reversed.

In order to demonstrate switching from AP-state to P-state, the MTJ was initialized to its AP-state by sweeping the B-field from -25mT to 0mT and back to -7mT. In order to switch the magnetization of m to the -x-direction and put the system in the P-state, positive 5 ms current pulses were applied starting from 2MA/cm² to 25MA/cm² at each of the three magnetic fields (Figure 5). After each pulse, the resistance of the MTJ was measured ten times. For low current densities, all ten data points of R_MTJ for any given pulse amplitude were nearly identical. At J_SH ≈ 10MA/cm² however, the first data point (hollow square figure 5) of R_MTJ was observed to be lower than the resistance of the initial AP - state. The subsequent 9 data points (filled black circles) all showed a resistance equal to R_AP. For higher pulse amplitudes, a similar trend is evident from figure 5 but with the first data point of the resistance approaching the resistance of the desired P state. A similar effect was observed for the B = -9mT, but no successful switching to the P state was observed. At B= -9.5mT, all ten data points for any given pulse amplitude were identical. At a value of J_SH = J_C ≈16MA/cm², switching to the P state was observed for B = -9.5mT. But as higher amplitude pulses were passed through the tantalum, the MTJ switched to the AP state and back again to the P state for the next current pulse. We speculate that the switching from P to AP occurred due to Oersted fields, because, the direction of the Oersted field created by a positive pulse is expected to torque m (which is initially in the -x-direction) to point in the +x-direction. The switching back to the P-state is again consistent with the direction expected from the GSHE in tantalum. The back and forth switching could also be explained by the so-called back hopping observed in MTJs.²² In MTJs, after successful switching from an AP to P (P to AP) state, further increasing the current beyond threshold can result in the magnetization state hopping back to its initial AP (P) state followed by telegraph switching between P and AP states.

In this section, we analyze the apparent current-induced switching from P → AP and AP → P quantitatively. The width of the pulses used in this study lies within the thermal activation regime of spin transfer torque (STT) switching,^23,24 where the critical current density J_C is a function of the height of the thermal barrier(E_b) separating the P and AP state and the current-pulse width (τ) used for switching. Using the standard model of thermally activated switching,^23,24 one can calculate the intrinsic critical current (J_C0) using the equation : I

where τ₀ = 1ns represents the reciprocal of attempt frequency,²⁵ J_C is the current density(labelled J_SH in fig.4 and fig.5) needed to observe switching of m. J_C0 can be expressed for an in-plane magnet as ²⁵ 

where t is the thickness of the magnet, α = 0.01 is the damping constant, M_s and H_c are the saturation magnetization and the coercive field of the magnet and Θ_SH is the spin hall angle. Both in the P → AP switching and AP → P switching cases, the GSHE switching exhibited a B-field dependence. At B = -8mT, the center of the bi-stable region, the external field cancels the dipolar field on m, resulting in a zero net-external field. For a single-domain magnet with a uniaxial anisotropy, the barrier at zero external field can be calculated using the formula E₀ = M_sH_cV/2, where M_s = 1.1 x 10⁶A/m is the saturation magnetization, H_c is the coercive field and V is the volume of the CoFeB magnet. For m initialized in the -x-direction, as the field is swept closer to the right edge of the minor loop, the barrier for P → AP switching is lowered as compared to E₀. The lowering of the barrier can be estimated using

where H_ext is the external field and H_c is the coercive field of the magnet after canceling out the dipolar field. The above expression assumes that m is single-domain and its energy is dictated only by the uniaxial anisotropy and an external field along the easy-axis of the magnet. For the B=-7mT, when a P → AP switching was observed, the barrier height is E_b ≈ 0.36E₀. Using this calculated value of E_b and a J_C = 15MA/cm², determined from the P → AP switching measurement, in equations (1) and (2) we estimate a Θ_SH of 5.9%. No GSHE switching was observed for a J_SH upto 20MA/cm² at a B-field of -8mT where the net field on magnet m is zero and E_b ≈ E₀. Using equation (1), we predict that a current density of ≈55MA/cm² is needed to generate the necessary J_C0 to switch the magnet m, for a barrier height of E₀. This is well outside the range of pulse amplitudes used in our measurements and thus consistent with our experimental finding. Unlike the P → AP switching, the AP → P switching is not abrupt, leading us to conclude that the switching may be driven by nucleation of domains. This is why equation (3) cannot be used in this context for a quantitative analysis. However we note that a very similar value of J_C = 16MA/cm² for AP → P switching is consistent with our previous discussion.

The calculated value of the spin hall angle is lower than the one estimated in Liu et al.¹¹ for a 6nm thick tantalum layer switching a 1.6nm thick CoFeB nano-magnet of area 350 x 100nm². The low RA product of the MTJ in our work negatively impacts the estimated spin hall angle. The spin current needed to torque the magnet m is generated only from the charge current flowing laterally in the Tantalum strip under it. Due to the low RA product of the MTJ, some of this current is shunted away from the tantalum into the MTJ and does not contribute to the GSHE. Due to the sidewall-shorting problem described above, the charge current as it reaches the section of the tantalum strip close to the MTJ is flowing not just in the tantalum and the magnet m, but also through the magnet M and the capping layers. All the current densities reported in this work have not taken this fact into account. The true current density that is contributing to the spin current generation through GSHE is lower. By taking this shunting effect into account the true Spin hall angle of the tantalum is likely to be ≈ 2 times larger than the value estimated above.

In conclusion, we have demonstrated current-induced magnetization switching of a patterned CoFeB magnet using Giant spin hall effect from a tantalum thin-film - an important step towards building a spin-switch for logic applications. We have used the thermal activation model of spin-torque to estimate a spin hall angle.We have also shown that thicker tantalum films can be deposited in the β-phase and still show GSHE.

The authors thank Chia-Ching Lin, Vinh Quang Diep and Prof. Supriyo Datta for their input and helpful discussions and Dr. Cem Akatay for help with the TEM imaging. The authors also thank Dr. Vo Tuan and the rest of the CNSE staff for providing the thin film stacks used in this study and the Birck Staff for supporting the fabrication effort. This work was supported by the Nanoelectronics Research Initiative (NRI) through the Institute for Nanoelectronics Discovery and Exploration (INDEX) center.