Observation of current-induced switching in non-collinear antiferromagnetic IrMn3 by differential voltage measurements

There is accelerating interest in developing memory devices using antiferromagnetic (AFM) materials, motivated by the possibility for electrically controlling AFM order via spin-orbit torques, and its read-out via magnetoresistive effects. Recent studies have shown, however, that high current densities create non-magnetic contributions to resistive switching signals in AFM/heavy metal (AFM/HM) bilayers, complicating their interpretation. Here we introduce an experimental protocol to unambiguously distinguish current-induced magnetic and nonmagnetic switching signals in AFM/HM structures, and demonstrate it in IrMn3/Pt devices. A six-terminal double-cross device is constructed, with an IrMn3 pillar placed on one cross. The differential voltage is measured between the two crosses with and without IrMn3 after each switching attempt. For a wide range of current densities, reversible switching is observed only when write currents pass through the cross with the IrMn3 pillar, eliminating any possibility of non-magnetic switching artifacts. Micromagnetic simulations support our findings, indicating a complex domain-mediated switching process.


Introduction
Antiferromagnetic materials (AFMs) provide a pathway to overcome the limitations of ferromagnet (FM)-based spintronic devices, thereby enabling new applications of spintronics in memory, computing, and terahertz electronics [1][2][3][4][5][6]. Their robustness against external magnetic fields, potential for high-density data storage, absence of stray fields, ultrafast dynamics, and high energy efficiency make them an excellent candidate for ultrafast magnetic random access memory [7,8], particularly at deeply scaled technology nodes that are important to emerging applications in artificial intelligence and unconventional (non-von Neumann) computing [9]. At the same time, their large spin coherence length [1,5,10,11] and excellent magnon propagation [12][13][14][15] make antiferromagnets promising candidates for information transmission in emerging computing concepts based on spintronics. Beyond computing applications, the ultrafast exchange-dominated dynamics of antiferromagnets also make them potentially of interest for the realization of roomtemperature, electrically tunable, and narrowband terahertz sources and detectors [3,[16][17][18].
Making memory units with AFMs requires experimental techniques to: (i) Electrically manipulate the AFM order (Néel vector) in a silicon-compatible material (i.e., write information), as well as to (ii) Electrically read out the orientation of the Néel vector (i.e., read information) once it has been modified. The first requirement was initially demonstrated in antiferromagnetic films with broken inversion symmetry (CuMnAs [8,[19][20][21], Mn2Au [22][23][24][25][26]) where a dampinglike Néel spin-orbit torque (SOT) was generated due to the current flowing in the bulk of the material, resulting in the motion of AFM domains. This was followed by studies in AFM insulators (NiO [27][28][29][30][31], Fe2O3 [32], CoO [33]), as well as thin films [34,35] and pillars [36] of metallic AFM materials interfaced with a heavy metal (HM) with large spin-orbit coupling (typically, Pt), where the mechanism of AFM order manipulation was the interfacial spin-orbit torque from the HM. In both types of experiments, the manipulation of AFM domains was directly observed using x-ray magnetic linear dichroism (XMLD) [20,37] and spin Seebeck microscopy [38] measurements. However, while helping to reveal the physics of AFM switching by electric current, these techniques are not directly applicable to practical AFM memory devices, where an electrical readout mechanism is needed. This second requirement -reliable electrical readout of the AFM order using magnetoresistive signals (i.e. the dependence of electrical resistance on the Néel vector orientation) -has proven to be challenging, and is the focus of the present work.
Previous electrical switching experiments in AFM/HM systems have been performed on a single-cross geometry with the HM as the bottom layer. However, several recent reports have indicated that some of the electrically measured switching signals in such test structures may have non-magnetic origins due to thermal effects and atomic motion (e.g. electromigration) in the HM layer, particularly in cases where the applied current densities were high [39][40][41]. Chiang et al. showed that these non-magnetic resistive switching signals increased as the heat conductivity of the substrate reduced, pointing to the role of heat gradients in facilitating electromigration in Pt, which in turn gives rise to a switching signature that is not reflective of AFM switching [39]. A recent work by Cheng et al. [32] focusing on the Pt/Fe2O3 system showed that while switching at intermediate current densities may indeed be of magnetic origin, switching at higher current densities was dominated by signals from the Pt layer. Notably, in this material system, the antiferromagnetic switching was accompanied by a square-shaped behavior where the readout voltage was independent of the number of pulses applied in a particular direction, whereas the nonmagnetic switching of the Pt exhibited a sawtooth behavior, where increasing the number of write pulses in a particular direction gradually increased or decreased the magnitude of the readout voltage. It is noteworthy that a similar square-shaped switching signature was also observed due to AFM order manipulation in the PtMn/Pt and PtMn/Ta systems [36]. Nonetheless, both for device applications and fundamental studies of AFM switching, it remains a challenge to reliably and directly separate the nonmagnetic and magnetic contributions to resistive switching signals in AFM/HM structures. This is particularly difficult for AFM/HM devices because, unlike FM/HM structures, magnetoresistance effects are typically smaller in AFMs, switching current densities are higher, and there is typically no possibility to manipulate (or saturate) the magnetic order with magnetic fields, something which can be routinely done with modest magnetic fields for FM/HM samples.
Here we present a new experimental protocol based on differential measurements, which allows one to systematically and robustly separate the switching signals originating from the AFM and the HM layers in AFM/HM material systems, thereby demonstrating the magnetic origin of the switching. The antiferromagnetic material that we choose for this demonstration is IrMn3, grown on Pt as the HM, with both layers having a predominant (111) texture. This choice is in part motivated by the fact that this metallic AFM is compatible with conventional semiconductor manufacturing techniques, and can be deposited on arbitrary substrates (in our case, thermally oxidized silicon) in poly-crystalline form using ultra-high vacuum sputter deposition. At the same time, IrMn3 is a non-collinear antiferromagnet with a range of intriguing transport characteristics, including a large spin Hall conductivity [42,43], a large anomalous Hall effect in the absence of macroscopic magnetization [44-47], a theoretically predicted (ferromagnetic-like) spin-polarized charge current [48], and has been used in devices exhibiting large tunneling anisotropic magnetoresistance (TAMR) [49], making it of interest for both memory (switching) and highfrequency devices based on antiferromagnets.
The test structure used for the differential measurements is shown in Fig. 1. It consists of a double-cross Pt structure with six terminals, where an IrMn3 pillar is placed on one of the two crosses. This configuration allows for revealing the difference between signals coming from the bilayer IrMn3/Pt structure and Pt. Importantly, we compare two crosses in close proximity within the same device, to minimize the effect of wafer-to-wafer and across-wafer variations that may arise if we compared Pt-only structures with IrMn3/Pt devices from different wafers, or even when comparing devices from different parts of the same wafer. By using this test structure, we demonstrate switching in IrMn 3 pillars with diameters of 6 µm and 4 µm, using current pulses in opposite directions (180˚ switching) as well as using current pulses that are perpendicular to each other (90˚ switching). The results indicate that for a range of writing current densities (from 36 to 50 MA/cm 2 ) switching only occurs in the IrMn3 pillar, with no appreciable signal coming from the adjacent Pt cross, while for higher current densities (~ 60 MA/cm 2 ), both crosses show switching, indicating a signal that originates in the Pt layer. Interestingly, we also observe that the switching in the antiferromagnetic IrMn3 pillars [42,43] in fact has a sawtooth behavior, meaning that the shape of the readout signal by itself is not a reliable indicator of whether or not the switching is magnetic in origin. We further study the dependence of the switching behavior on the writing pulse number, which indicates a thermally activated domain wall motion mechanism underlying the observed switching. Micromagnetic simulations taking into account the cubic IrMn3 anisotropy, and pinning potentials from the spatial distribution of the anisotropy, support our findings and qualitatively describe the switching processes arising from the manipulation of complex multidomain structures.

Experimental protocol
The double-cross design allows for a differential DC resistance measurement between the two vertical arms (i.e. electrodes 2 and 3, hereafter referred to as R2-3), when a small reading current (100 µA in our experiments) is applied along the long arm of the device (between electrodes 1 and 4, hereafter referred to as I 1-4 read). Since any effect associated with the Pt layer is expected to be the same in both arms, any variations of the voltage difference between the two vertical arms (V 2-3 read = R2-3 I 1-4 read) are expected to mainly depend on the resistance changes associated with the IrMn3 pillar, in response to anisotropic magnetoresistance (AMR) and spin-Hall magnetoresistance (SMR) effects. Note that, apart from the new possibility for differential readout, this design can still be used in both of the usual configurations used for electrical switching (i.e. writing) of the AFM bit as will be shown in the paragraph below. We refer to these as 90˚ switching when the current pulses are applied perpendicular to each other, e.g. between contacts 1-4 and 2-6 or 3-5, and 180˚ switching when opposite write currents are sent through one of the vertical arms (contacts 2-6, as illustrated in Fig. 1b, or 3-5), or through the common horizontal arm (electrodes [1][2][3][4]. Readout in all cases takes place through the same procedure, by reading V 2-3 read while applying the reading current I 1-4 read. In addition to differential measurements of the AFM switching in the IrMn3 cross, this structure allows us to make in-situ reference measurements on the Pt cross within the same device. This is important since fabrication and patterning methods, even if nominally similar for two sets of samples, inevitably suffer from wafer-to-wafer and across-wafer variations in parameters such as the film thickness and lateral dimensions of metal lines, making it difficult to quantitatively compare control samples made on a different substrate to the AFM samples being studied. In the structure of Fig. 1, the control structure is already present in the final device and can be measured independently of the AFM arm to detect any signals originating from the Pt layer. As discussed in the next sub-section, this possibility is of particular interest for the 180˚ switching scheme when the writing pulse is only applied in the arm with the IrMn3 pillar (electrodes 2 and 6 in Fig. 1) or without it (electrodes 3 and 5).

Current-induced switching measurements
We first show results for the 180˚ switching configuration in two cases: (i) When the write pulse is applied in opposite directions along the arm with the IrMn3 pillar (between electrodes 2 and 6, The amplitude of the pulses is 20 mA (corresponding to a current density of 40 MA/cm 2 in the Pt layer), and the pulse width is 1 ms. Each write attempt consists of applying two pulses in the same direction (e.g. from electrode 2 to 6, or 3 to 5), and then measuring the differential voltage between the two vertical arms (V 2-3 read = R2-3 I 1-4 read) using a reading current of 100 µA (for further details of the electrical measurement, see the Methods section). Next, this process is performed for write attempts with the reverse write current direction (e.g. from electrode 6 to 2, or 5 to 3 in the case of the Pt reference measurement).
The results of this 180˚ switching experiment are shown in Fig. 2a. The results clearly show a change of the readout voltage (V 2-3 read) when the write pulse is applied in the cross with the IrMn3 pillar (data points indicated by squares), whereas no significant change in the readout voltage is observed when the write currents are applied in the arm containing only Pt (data points indicated by circles). This result unambiguously shows that the switching detected in the double-cross structure is only due to the IrMn3 pillar, with an average readout voltage difference of ~ 8 µV. In addition, when the number of applied pulses in the same direction increases, the output voltage continuously changes, increasing when they are applied from electrode 2 to 6 (red points), and decreasing when the direction of the pulses is reversed (blue points). This gradual change indicates that the switching takes place by the reorientation of the domain structure in the IrMn3 pillar. It is worth pointing out that this gradual switching behavior is similar to the previously detected sawtooth switching in various AFM material systems [19], while also bearing similarity to previously reported sawtooth-like non-magnetic signals in Pt-only devices [32,39]. In the present experiment, however, the differential voltage measurement, combined with the in-situ repetition of the experiment in the Pt reference arm, allows us to conclude that the switching (for a write current of 20 mA, equal to 40 MA/cm 2 ) is exclusively due to the IrMn3 pillar. It is worth noting that, therefore, the presence of a sawtooth signal shape by itself is not necessarily a reliable sign of a non-magnetic artefact (see also the micromagnetic calculations below for further discussion), and may indeed be associated with AFM switching, as is the case in our experiment.  The writing current is applied along the short arms of the double-cross structure. Square points refer to the case of a writing current applied in the arm with the IrMn3 pillar (square red points from electrode 2 to 6, I [2][3][4][5][6] Write, and square blue points from electrode 6 to 2, I  Write), and the circle data points correspond to the arm only with Pt (circle brown points from electrode 3 to 5, I [3][4][5] write, and circle dark blue points from electrode 5 to 3, I 5-3 Write). The differential voltage of the device only shows a measurable switching signature when the current pulse is applied through the IrMn3 pillar cross, i.e. in cases I [2][3][4][5][6] Write and I 6-2 Write. b 180˚ switching when the writing current is applied from electrode 1 to 4 (I 1-4 write, square red points) or vice versa (square blue points). In this case, the write current pulse always passes through the section with the IrMn3 pillar, showing a readout voltage with a similar switching behavior as the case of the write current I 2-6 write, but with larger voltage variation. Raw data for this figure (prior to subtraction of the background slope) can be found in Supplementary Note 3.
Next, we repeated the 180˚ switching experiment while applying both the writing and reading currents between the same horizontal electrodes (1 and 4 in Fig. 1). In this case, the writing pulse is applied through both of the crosses in the device, with and without the IrMn3 pillar, hence allowing one to expect that switching would also be observed in this case. The results are shown in Fig. 2b, confirming this expectation. The results are qualitatively similar to the previous 180˚ switching experiment in the IrMn3 arm, suggesting that the output voltage again changes because of the IrMn3 pillar. The fact that we observe switching of the IrMn3 in both cases, i.e., while applying the writing pulse perpendicular or parallel to the reading current, is a further indication that the IrMn3 pillar has a multidomain structure that can be manipulated by current pulses applied in either direction. Note that in this case (Fig. 2b), the peak-to-peak variation of the readout voltage (V 2-3 read) is approximately 20 µV, which is bigger than in the previous case (~ 8 µV in Fig. 2a).
This suggests that the current-induced motion of domain walls is more efficient in the case of horizontal currents (I 1-4 write) compared to vertical currents (I 2-6 write). Given that the current density flowing in the Pt layer is the same in both cases, we attribute this to an increase in the thermally activated depinning and motion of domain walls due to Joule heating, given the larger resistance of the longer horizontal device arm (~ 970 Ω) compared to the shorter vertical one (~ 480 Ω). The difference of the resistance in the two cases is in good correspondence with the respective distance between the measurement electrodes, which is 70 µm for the longer arm and 30 µm for the shorter one, respectively. Thermal activation effects and the depinning of domain walls are discussed in more detail in the next section.
Finally, we studied the switching behavior under the same conditions (20 mA write current and 1 ms pulse width), using a 90˚ switching configuration. In this case, the write currents are applied perpendicular to each other either in the left cross (I 1-4 write and I 6-2 write, labeled IrMn3 in Fig.   3 with data points indicated in squares) or in the right cross of the device (I 1-4 write and I 5-3 write, labeled Pt in Fig. 3 with data points indicated in circles). In each case, the voltages that are read out after horizontal write currents (I 1-4 write, which always passes through the IrMn3 cross) are indicated in red, while the voltages that are read out after vertical write currents (I 6-2 write,, which passes through the IrMn3 arm, and I 5-3 write, which passes through the Pt-only arm) are indicated in blue and dark blue, respectively. As expected, all cases indicate a clear switching signal, except the case of dark blue circles, which is the only case where the write current (I 5-3 write) does not pass through the IrMn3 pillar.

Fig. 3. Measured differential voltage (V 2-3 Read) for the 90˚ switching configuration with 20 mA write pulses.
Square data points indicate the case where the write current pulses are applied in the cross containing the IrMn3 pillar (I 1-4 write , red squares, and I 6-2 write , blue squares), and circle data points correspond to the case when the pulse is applied to the other cross (I 1-4 write, red circles, and I 5-3 write, dark blue circles). Note that one of the writing currents in the latter case (I 1-4 write) also passes through the cross with the IrMn3 pillar. The differential voltage, V 2-3 Read, only shows switching in the cases where the current pulse passes through the cross containing the IrMn3 pillar, i.e. all cases except write. Raw data for this figure (prior to subtraction of the background slope) can be found in Supplementary Note 3.
Specifically, when the 90˚ writing scheme is used in the arm containing the IrMn3 pillar, we observe a clear modification of the output differential voltage, as in the previous cases of 180˚ switching (square points in Fig. 3). However, when performing the same experiment on the right arm, only the horizontal write pulse between electrodes 1 and 4 (which also passes through the  configuration when the writing current is applied from electrode 1 to 4 (I 1-4 write, red squares) and from electrode 6 to 2 (I 6-2 write, blue squares), b 180˚ switching when the current pulse is applied from electrode 2 to 6 (I 2-6 write, red squares) and its reverse (I 6-2 write, blue squares), and c 180˚ switching between electrodes 1 and 4 (where I 1-4 write corresponds to red squares and I 4-1 write corresponds to blue squares). In all three cases, we observe a sawtooth-like switching of nonmagnetic origin at this high writing current, which corresponds to a current density of 60 MA/cm 2 in the Pt layer.
Hence, to obtain a measurable switching signal in Pt-only double-cross devices, we need to apply a current amplitude of 30 mA (60 MA/cm 2 ), whereas the IrMn3 switching in all the measured devices occurs for lower current densities. This is in agreement with the results of Figs. 2 and 3 (all at 20 mA), where no Pt switching was observed, and further indicates that the observed switching in Figs. 2 and 3 is due to AFM order manipulation, and hence has a different physical origin than that in Fig. 4, which is not magnetic.

Micromagnetic simulations
To better understand the experimental results, and in particular the role of the domain structure of the IrMn 3 pillar, we performed micromagnetic simulations. The micromagnetic framework that was used is based on two strongly coupled Landau-Lifshitz-Gilbert (LLG) equations (see Supplementary Note 5 for a detailed description of the model). In the simulations, the IrMn3 thin film has a cubic anisotropy [50,51] as shown in Fig. 5a. This figure plots 1 anis  − rescaled to the range from 0 to 1, where anis  is the anisotropy energy density. In this anisotropy configuration, there are 8 possible domain types oriented along the diagonals of the cubic cell, which are labeled in Fig. 5a as 1, 2, 3, 4, and their opposite directions. Note that the red color in Fig. 5a indicates a more stable state, favored by the anisotropy. From the x-ray diffraction measurements (see Supplementary Note 1), it is evident that one of the ( ) 111 directions is primarily oriented along the out-of-plane direction, which we define as the z-direction. If we further assign the horizontal arm of the device to be the x-axis and the vertical one to be the y-axis (see Fig. 1   To study the dynamical properties of the DWs, we first performed a set of micromagnetic simulations to calculate the behavior of each of the six DWs (Figs. 5b-g) by itself, in the presence of an electric current [52]. The DW velocity increases linearly as a function of the current, with mobility that depends on the DW type because of the different relative orientation between the DW magnetization and the spin polarization p (perpendicular to the applied current). Table 1 ( )     where one sees larger variations of the differential output voltage when the longer, more resistive horizontal arm is used to apply the write current.
To further investigate this hypothesis and measure the minimum current required for switching in this IrMn3/Pt device, we repeated the same experiments at different current amplitudes in the range from 2 to 20 mA. The procedure consisted of applying three write attempts (as defined in the previous section), and measuring the differential voltage after each attempt, using a reading current of 100 µA. After this, we reversed the direction of the writing pulse and repeated the same procedure at the same amplitude, but in the opposite direction. These steps were repeated while increasing the write current amplitude in 1 mA steps and keeping the pulse width constant at 1 ms.
The results following this protocol for the 180˚ switching scheme in the vertical (terminals [2][3][4][5][6] and horizontal (terminals 1-4) write current directions are shown in Fig. 6. corresponds to the amplitude and direction of the writing current applied before the differential voltage measurement.
The graphs correspond to cases a when the current pulse is applied between electrodes 2 and 6, I 2-6 write, and b when the writing current is applied between electrodes 1 and 4, I [1][2][3][4] write. For both cases, the device resistance increases due to Joule heating as we apply more write attempts, and for current pulses above 18 mA the output voltage shows a switching behavior, indicating a threshold current density of ~ 36 MA/cm 2 for switching in this structure.
For both cases, the total resistance of the device shows a non-linear increase as a function of the amplitude of the writing pulse, followed by a resistance peak. A similar trend in the resistance has been previously measured in MnTe thin films as a function of temperature [53,54]. This suggests that the measured resistance background in Fig. 6 is associated with the Joule heating of the device by the applied current, and its effect on the AFM state, namely the spin fluctuations of the Mn atoms which are bigger when the device temperature is closer to the Néel temperature. It is worth noting that similar behavior has also been measured in the tunnel anisotropic magnetoresistance of Ir0.2Mn0.8/MgO/Ta magnetic tunnel junctions [55], as well as in the Seebeck coefficient of thin IrMn3 films [56]. In the latter work, the Néel temperature of IrMn3 is measured to be ~ 350 K for a 4 nm thin film, while the Néel temperature for bulk IrMn3 is around 700 K.
Based on this, while we did not perform systematic measurements of the Néel temperature in our samples, we can expect our films to have a Néel temperature within the above range. Considering that the measured temperature increase due to a single 20 mA writing current pulse in our samples is ~96 K as shown in Supplementary Note 6, it is reasonable to ascribe this resistance trend to the temperature increase of the IrMn3 pillar.
Note that the increase of the differential voltage is also larger when the write pulse is applied parallel to the reading current direction, i.e. terminals 1-4, which has a larger electrical resistance compared to the vertical arms, thus also generating more Joule heating. However, for both cases (I 1-4 write and I 2-6 write), the threshold current required for switching in the IrMn3 pillar is approximately the same, i.e. ~ 18 mA, corresponding to a threshold current density of ~ 36 MA/cm 2 .
In summary, the experiments reported here provide clear evidence of the electrical switching of Néel order in an IrMn3 pillar by spin-orbit torques from an adjacent Pt layer. The proposed differential measurement approach can be applied to a variety of AFM/HM material systems in a straightforward manner, including those with insulating AFM films. We expect that this will allow for reliable electrical measurement of the resistive signatures of AFM switching induced by SOT, which will be important not only for the fundamental understanding of current-induced AFM domain dynamics, but also as an electrical readout method for practical applications in memory and computing devices.
Finally, it is worth mentioning that the range of current densities where AFM switching can be observed without artifacts from the HM layer depends on the threshold switching current density of the AFM. Reducing the latter will allow for the operation of AFM/HM memory devices at even lower currents than those presented here, where the possibility of resistive signals due to the HM layer will be further reduced. One possible pathway to do so could be the further reduction of the IrMn3 thickness compared to the 10 nm in this work. In particular, it is worth noting that Néel temperature reduction as a function of the thickness of IrMn3 has been previously observed. Bulk IrMn3 has a Néel temperature above 700 K [57], which can be reduced to room temperature for thickness values below 3 nm [55,56]. The reduction of the Néel temperature (while still keeping it above room temperature by selecting an appropriate IrMn3 thickness) will be accompanied by a reduction of the exchange constant, which in turn affects the threshold current. Thus, reducing the thickness of IrMn3 may result in a reduction of the threshold current density to switch the AFM state, providing an even wider range of current densities where artifact-free AFM switching can be electrically observed.

Materials and sample fabrication
The materials used in this study were deposited on thermally oxidized silicon substrates, using sputter deposition under an Ar pressure of 2.5 mTorr at room temperature, in a physical vapor deposition system with base pressure less than 1 × 10 -8 Torr. The IrMn3 films were deposited by co-puttering from Ir and Mn targets at 10 W and 40 W, respectively. All samples were capped with a 2.5 nm MgO protecting layer. The Pt(5)/IrMn3(10) structure (thicknesses in nanometers), was patterned into the six-terminal devices shown in Fig. 1, using photolithography and dry etching techniques. The IrMn3 was patterned on top of the Pt double cross as a micro-pillar with nominal diameters of 4 and 6 µm. Finally, after etching of the heavy-metal layer and the pillar, Cr(5)/Au(80) electrodes were grown by electron beam evaporation to form the electrical contacts of the device.

Magnetic characterization
To confirm the antiferromagnetic character of the sputtered IrMn3 films, a test structure of Ta(5)/Pt(8)/Co(0.8)/Pt(1)/IrMn3(7)/MgO/Ru (thickness in nanometers) was magnetically characterized using a vibrating sample magnetometer (VSM) measurement system from Lake Shore. The samples were characterized in an out-of-plane field range between -500 and 500 Oe.

Electrical measurements
Electrical current pulses were applied in the six-terminal devices using two 6221 Keithley current sources for the longitudinal (I 1-4 write) and vertical pulse directions (I 2-6 write and I 3-5 write). The reading current was always set to 100 µA between electrodes 1 and 4, as shown in Fig. 1. The differential voltage was measured between electrodes 2 and 3 using a Keithley 2182A nano-voltmeter while applying the reading current, using the delta mode option of the nano-voltmeter.

Micromagnetic simulations
In our micromagnetic model, we consider IrMn3 to have a face-centered cubic unit cell with lattice constant = 0.4 nm [51], where only the Mn atoms are magnetic [58]. Therefore, the magnetic unit cell is a bcc cell with a lattice constant = √2/2 ≈ 0.28 nm [51]. Even though IrMn3 is a noncollinear antiferromagnet with three sublattices, at a mesoscopic scale it can be described by means of two strongly coupled LLG-Slonczewski equations. In this framework, the inhomogeneous intra-lattice exchange constant is 11 = 4.55 × 10 −7 erg/cm, and both the homogeneous and inhomogeneous inter-lattice exchange constants are 0 = 12 = −2.286 × 10 −6 erg/cm [50]. We considered the damping parameter to be 0.1  = [50], while the spin Hall angle is 0.1 SH  = [59]. The saturation magnetization of Mn was considered to be = 153.75 emu/cm 3 . Finally, the anisotropy is known to be cubic, with anisotropy constant = −6.2 × 10 6 erg/cm 3 [50].

Data Availability
The data that support the plots within this paper and other findings of this study are available from the corresponding authors upon reasonable request.