Introduction

Precisely manipulating local magnetization with electrical stimuli has great potential for applications in data storage, in-memory computing, brain-like computing, and microwaves1,2,3,4,5,6,7,8. The spin-orbit interaction gives rise to a range of electromagnetic phenomena9,10,11,12, including the spin Hall effect (SHE), which enables the conversion of charge current into spin current. The spin-orbit torque (SOT) exerted by this spin current is regarded as a highly promising candidate for manipulating magnetization, given its independent read/write paths and rapid response rates13,14. A large amount of effort has been made to improve the charge-to-spin conversion efficiency to reduce the energy consumption of magnetization switching and enhance the device’s reliability15,16,17,18. However, achieving high spin Hall conductivity (SHC) and SOT efficiency simultaneously in conventional heavy metals or topological insulators remains challenging19,20,21. Despite the bulk SHE, the interfacial Rashba-Edelstein effect (REE) is another important source of SOT22. Moreover, the spin current tunneling is also highly dependent on the interface between non-magnetic and magnetic layers23,24, which makes interface engineering a powerful tool to enhance SOT efficiency.

The topological semimetallic SrIrO3 (SIO) is distinguished by a substantial intrinsic SHE originating from the Berry curvature of the topological band structure. Its sizable SHC ( > 104 ħ/2e S m−1) and charge-to-spin conversion efficiency make it a promising material for spintronic devices23,25,26. On the other hand, SIO shows a strong correlation between lattice, charge, spin, and orbit degrees of freedom, which facilitates the development and manipulation of exotic functionalities, like SOT27,28. It has been confirmed that a sharp interface between SIO and ferromagnetic oxide, as well as the SIO’s orthorhombic crystal lattice, can improve the SOT efficiency25,29. The introduction of strain can improve the SOT efficiency in heterostructures with SIO and an amorphous ferromagnetic layer, but it necessitates a meticulous control of Ir-O octahedra rotation26,30. In addition, the intercalation of antiferromagnetic NiO has been demonstrated to enhance the SOT efficiency in heterostructures comprising SIO and a magnetic layer23,31,32,33. This is attributed to the propagation of spin angular momentum (SAM) mediated by the thermal magnon of the NiO layer34,35. However, the spin current generated by the as-prepared SIO is limited, and the propagation of magnons only relatively increases the transmittance of the SAM without generating a new spin current source.

Here, we demonstrate an all-oxide epitaxial SIO/NiO (t)/La2/3Sr1/3MnO3 heterostructure to achieve a large SOT efficiency enhancement (51%) at room temperature by optimizing NiO thickness (optimal 2 nm). NiO can simultaneously transfer SAM from bulk SIO via spin waves efficiently, and further enhance the SOT efficiency due to the substantial interfacial REE. The interfacial orbital hybridization of Ir-5d/O-2p/Ni-3d shows essential roles in the interfacial REE, whereas the NiO/La2/3Sr1/3MnO3 interface contributes little to the spin torques. This engineered heterostructure enables stable room-temperature magnetization switching at low current density, benefiting from both the enhanced SOT efficiency and superior spin Hall conductivity inherent to the multilayer architecture.

Results and discussion

SOT efficiency of SIO/NiO/LSMO with NiO intercalation

The epitaxial SIO (6 nm)/NiO (t)/La2/3Sr1/3MnO3 (LSMO) (30 nm) heterostructures are grown on SrTiO3 (001) substrate by pulsed laser deposition (see Methods). LSMO is selected as the ferromagnetic layer due to its high Curie temperature and similar crystal lattice and constants to NiO and SIO. Figure 1a, b illustrates the cross-sectional high-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) images of SIO/NiO/LSMO heterostructure grown on SrTiO3 (STO) substrate. The cubic rock salt NiO forms sharp interfaces with the perovskite SIO and LSMO layers. Additional STEM of the LSMO/STO interface is shown in Supplementary Information (Fig. S1). According to the HAADF image, different from the typical corner-sharing octahedra at the perovskite heterostructure interface, the Mn-O or Ir-O octahedra share the edge with Ni-O octahedra, which significantly reduces the distance between Mn-Ni or Ir-Ni, as shown schematically in Fig. 1c. The high-resolution X-ray diffraction (XRD) pattern demonstrates a single-phase cubic crystal structure of the epitaxial films (Fig. 1d). Diffraction peaks of LSMO (001) and (002), as well as their Laue oscillatory features, are observed at 23.11° and 47.25° near the substrate peak. Similarly, the epitaxial peaks of SIO (001) and (002) are observed at 22.32° and 45.69°, respectively. The rocking curves with small full-width at half maximum values show good crystal quality of the epitaxial films (Fig. S2). Figure 1e illustrates the variation of magnetization and resistivity versus temperature for the SIO/NiO (2)/LSMO heterostructure. Accompanied by the typical metal-insulator transition in LSMO, the magnetization of heterostructure drops at 352 K (Curie temperature of LSMO36), indicating that the device can be operated above room temperature. In addition, the LSMO film shows an in-plane easy magnetization axis, and the NiO layer thicker than 2 nm shows typical antiferromagnetic character (Fig. S3).

Fig. 1: Crystal structure and magnetic property.
figure 1

a, b Cross-sectional high-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) images of the SrIrO3/NiO/La2/3Sr1/3MnO3 (SIO/NiO/LSMO) heterostructure grown on SrTiO3 (STO) substrate. c Schematic representation of the stacked SIO/NiO/LSMO heterostructure (not drawn to scale). d High-resolution X-ray diffraction (XRD) ω-2θ scan of the SIO/NiO/LSMO heterostructure. The indices of crystallographic planes are given in parentheses, and the inset shows a cartoon of the stacked heterostructure. e The in-plane magnetization and longitudinal resistivity curves of SIO/NiO (2)/LSMO as a function of temperature.

Full size image

The second harmonic Hall resistance test method37,38, which exhibits low-frequency adiabatic and high-sensitivity properties, is employed to quantitatively resolve the SOT effective field. As illustrated in Fig. 2a, an alternating current Iac will induce a transverse spin current through SHE of SIO, which drives the magnetic moment oscillations of LSMO after SAM is transferred. By scanning the external magnetic field in the xy-plane (at an angle φ to the x-axis), the transverse (y-axis) Hall harmonic resistance will reflect the characteristics of the SOT effective fields. The \({R}_{{xy}}^{1\omega }\) of SIO/NiO (2)/LSMO exhibits a planar Hall resistance (RPHE) feature, which shows a sinφcosφ dependence with RPHE = 0.11Ω (Fig. 2b). The \({R}_{{xy}}^{2\omega }\) can be fitted as a superposition of damping-like (DL) term: \({R}_{{DL}+\nabla T}\cos \varphi\) and field-like (FL) term: \({R}_{{FL}+{Oe}}\left(2{\cos }^{3}\varphi -\cos \varphi \right)\), as shown in Fig. 2c. The \({R}_{{DL}+\nabla T}\) is linearly dependent on 1/(HK + Hext) and the \({R}_{{FL}+{Oe}}\) is linearly dependent on 1/Hext, as shown in Fig. 2d, e, respectively. After consideration of the current shunt in the SIO layer (Fig. S5), the efficiencies of the DL-and FL-effective fields (HDL, HFL) can be resolved, both of which initially increase with the intercalation of NiO, and reach their maximum at tNiO of 2 nm, and then gradually decrease as the NiO become thicker (Fig. S6).

Fig. 2: Second harmonic voltage measurements and spin-orbit torque (SOT) efficiency analysis.
figure 2

a Schematic diagram of spin torques (τDL and τFL) in the second harmonic voltage measurement. b First-order harmonic resistance \({R}_{{xy}}^{1\omega }\) of SIO/NiO (2)/LSMO as a function of magnetic field angle φ (the angle between Hext and Iac). c Second-order resistance \({R}_{{xy}}^{2\omega }\) of SIO/NiO (2)/LSMO as a function of φ. The red and purple fitting line is the damping-like term (\({R}_{{DL}+\nabla T}\cos \varphi\)) and the field-like term (\({R}_{{FL}+{Oe}}(2{\cos }^{3}\varphi -\cos \varphi )\)), respectively. d The linear dependence of \({R}_{{DL}+\nabla T}/{R}_{{AHE}}\) versus \(1/{H}_{K}+{H}_{{ext}}\) for heterostructures with different tNiO. e The linear dependence of \({R}_{{FL}+{Oe}}/{R}_{{PHE}}\) versus \(1/{H}_{{ext}}\). f Damping-like SOT efficiencies (ξDL) as a function of tNiO for SIO/NiO (t)/LSMO, Pt/NiO (t)/LSMO, and NiO (t)/LSMO heterostructures. The dashed blue line is just guide to the eyes. Error bars represent standard deviation.

Full size image

The HDL/J is approximately one order of magnitude higher than the HFL/J and is the primary contributor to the SOT. The \({\xi }_{{DL}}=(2e/\hslash ){M}_{s}{t}_{{FM}}\cdot {H}_{{DL}}/J\) is employed to assess the DL-SOT efficiency, which plays a dominant role in the current-induced magnetization switching, as shown in Fig. 2f. It can be seen that the evolution of ξDL with tNiO can be divided into four stages. The source of DL-SOT can be divided into four parts: spin current originated from (i) SHE of SIO, and (ii) interfacial REE of SIO/NiO which is transferred by tunneling effect, and the magnon current which carries the SAM induced by (iii) SHE, and (iv) REE. The variation of these four parts with tNiO is summarized in Table 1. In the first stage, the insertion of NiO leads to a significant increase in ξDL, from 0.45 (tNiO = 0 nm) to the maximum of 0.68 (tNiO = 2 nm). A more than 51% increment in DL-SOT efficiency is obtained. The significant increase of ξDL at tNiO less than 2 nm originates mainly from the REE39 at the inversion asymmetric interface of SIO/NiO, which could generate an additional spin current. On the other hand, the NiO intercalation increases the transmittance of the SAM through the magnon current transportation. The antiferromagnetic order of NiO is gradually stabilized as tNiO exceeds 2 nm, which can be used as an effective carrier for SAM transport35, but it is difficult to amplify SAM according to momentum conservation. In the second state, as tNiO increases from 2 nm to 6 nm, the ξDL decreases rapidly from 0.68 to 0.28. This rapid decrease in ξDL after tNiO exceeds 2 nm is attributed to the sharp reduction of the tunneling spin current with the thickness of the insulating NiO layer. In the third state, as tNiO increases from 6 nm to 20 nm, the ξDL is almost maintained at a plateau of 0.22. This stability arises because the magnon current becomes the only medium for SAM propagation, which has a long propagating distance (above 20 nm) in the normal direction of the film. In the fourth stage of tNiO at 20–30 nm, the ξDL slowly decreases to 0, due to the magnons decay in NiO.

Table 1 Evolution trends of the SOT in SIO/NiO (t)/LSMO at four stages with different tNiO
Full size table

Pt/NiO (t)/LSMO contrast experiments are performed as shown in Fig. 2f. Although the REE at the Pt/NiO interface cannot be ruled out, the ξDL shows no increment as NiO rises to 2 nm, but rather a decrease from 0.085 to 0.083. This suggests that relying on magnon current propagation alone cannot amplify the SAM in the heterostructure. As tNiO increases from 6 to 10 nm, the ξDL of Pt/NiO (t)/LSMO is stabilized at ~0.06, confirming the efficient transport of magnons in NiO. Another contrast experiment is NiO (t)/LSMO without a heavy metal layer, where no SOT is observed with varied tNiO, implying that it is the SIO/NiO interface rather than the NiO/LSMO interface that contributes to the REE. In addition, our investigation also reveals that the surface terminations of the STO substrate exhibit negligible impact on the SOT efficiency (Fig. S7), underscoring the dominance of the SIO/NiO interface in governing SOT generation. And substrate-induced compressive strain can enhance the SOT efficiency ξDL (Fig. S8).

A similar SOT enhancement phenomenon was observed at 60 K in a previous study of SIO/NiO/SrRuO3, which is attributed to the SAM carried by the magnon current31. However, the magnons are capable of transmitting SAM rather than enhancing the SOT efficiency35. In addition, the rapid decay of SOT below tNiO = 6 nm cannot be well explained by magnon transport, whose range is thought to be tens of nanometers35. Thus, the interfacial REE is proposed here to supply an additional spin current source, while the tunneling of interfacial spin current is introduced to give an integrated understanding of the SOT decay phenomenon with the increase of tNiO interlayer. The SIO/NiO interface exhibits partial discontinuity for ultrathin NiO films (tNiO < 2 nm, Fig. 1b), which is similar to the previous study31. Below this critical thickness, the REE increases significantly with tNiO. Upon achieving continuous NiO film stability (tNiO > 2 nm), the interfacial REE saturates, the magnon current stabilizes, and the spin current tunneling weakens with respect to tNiO. On the other hand, long-range magnon transport was demonstrated in Bi2Se3/NiO heterostructures35, but the inherent lattice mismatch between Bi2Se3 and NiO is worse than the case without NiO. Conversely, the high-quality NiO interlayer can enable the robust antiferromagnetism at extremely low thicknesses and significantly improve the interfacial REE in our study.

The electronic structures in the SIO/NiO heterostructure are calculated by density functional theory (DFT) to analyze the interfacial REE. Figure 3 presents the SHC (σSH) of the SIO/NiO heterostructure induced by spin splitting as a function of the Fermi level. When the spin-orbit coupling (SOC) is neglected and the NiO is set as nonmagnetic with a small contribution of magnetic moments, the σSH remains negligible at approximately 1.7 × 103 S m−1 ħ/2e. When considering SOC and NiO is also nonmagnetic, a finite σSH of 2.2 × 106 S m−1 emerges near the charge neutrality point (E – Ef = 0). Crucially, when considering SOC and the Zeeman field induced by antiferromagnetic ordering in NiO, σSH is significantly enhanced to 1.4 × 108 S m−1 ħ/2e at the neutral point, corresponding to a two-order-of-magnitude enhancement. These results suggest that the Zeeman field of antiferromagnetic NiO plays an important role in interfacial REE-induced spin splitting.

Fig. 3: Calculated spin Hall conductivity (σSH).
figure 3

σSH arises from the interface spin splitting at the SIO/NiO interface.

Full size image

Figure 4a shows the calculated Fermi surface of the SIO/NiO without considering the SOC, and NiO is set as nonmagnetic. It can be seen that the wave vectors with opposite spin directions are perfectly coincident at the Fermi level, therefore, net spin currents are hardly generated by applying electric field. When SOC is taken into account (Fig. 4b), a significant Rashba spin splitting of wave vectors is observed at the Fermi surfaces. At this point, a charge current will induce a considerable spin current. Furthermore, when NiO is configured as antiferromagnetic and SOC is considered, a more pronounced spin splitting is observed at the Fermi surface (Fig. 4c). In contrast, when only the antiferromagnetic exchange field is involved (excluding SOC), the spin splitting magnitude is much reduced (Fig. S10). These results support the hypothesis that the interfacial REE dominates the generation of spin currents, wherein the antiferromagnetic exchange field plays a pivotal role. Quantitatively resolving the spin splitting magnitude from the spin-resolved Fermi loops is challenging due to the fact that the splitting occurs across the entire two-dimensional Brillouin zone. However, the trends of spin splittings are in line with the calculated SHC (Fig. 3). Recently, Ariando et al. developed a novel experimental method to determine the Rashba parameter (α) through magnetoresistance measurements40. Based on this, we analyzed the magnetoresistance by considering weak antilocalization. The insertion of NiO can increase the α from 0.38 eV pm (tNiO = 0 nm) to 0.74 eV pm (tNiO = 2 nm), while reducing the spin scattering length from 15.3 nm (tNiO = 0 nm) to 8.0 nm (tNiO = 2 nm). These results are consistent with the DFT calculation (Figs. S11, S12).

Fig. 4: Density functional theory (DFT) calculated interfacial Rashba spin splitting.
figure 4

(a), (b) The DFT calculated Fermi surfaces of SIO/NiO (a) without and (b) with considering the spin-orbit coupling (SOC). NiO is set as nonmagnetic. (c) Fermi surface of SIO/NiO with considering SOC, and NiO is set as antiferromagnetic. Arrows represent spin polarization vectors. Colors are used to distinguish energy bands and do not represent any physical quantities. Schematic of the (d) Ir-5d/Ni-3d orbital hybridization at the SIO/NiO interface and the (e) Ir-5d/Mn-3d orbital hybridization at the SIO/LSMO interface. (f) Schematic of the SOT action in SIO/NiO/LSMO heterostructure. The interfacial Rashba-Edelstein effect (REE) of SIO/NiO and the SHE of SIO provide the spin current source, and the spin angular momentum carried by the magnons is injected into the LSMO layer, thus driving the magnetic moment oscillation.

Full size image

The Rashba splitting arises as a cooperative phenomenon requiring both SOC and broken inversion symmetry. The SOC Hamiltonian is expressed as41: \({H}_{{SO}}=\frac{e{\hslash }^{2}}{4{m}_{e}c}{{\boldsymbol{\sigma }}}\cdot \left({{\boldsymbol{k}}}\times \nabla \left({V}_{{nuc}}+{V}_{{\mathrm{int}}}\right)\right)\), where \({V}_{{nuc}}\) represents the symmetric nuclear potential of the atomic lattice, and \({V}_{{\mathrm{int}}}\) donates the asymmetric interfacial potential induced by the heterostructure. Atomic structural asymmetry at the interface generates a gradient in \({V}_{{\mathrm{int}}}\), while the strength of SOC governs the efficiency with which these gradient splits degenerate spin states. Consequently, the activation of SOC is essential for observing Rashba splitting, as demonstrated in Fig. 4b. In contrast, when both SOC and the effective Zeeman field (originating from antiferromagnetic NiO) are absent, the spin-polarized tunneling channels retain their intrinsic symmetry, resulting in no measurable spin splitting, as shown in Fig. 4a.

The interfacial Rashba interaction in oxide heterostructures is strongly correlated with the charge transfer at the interface42. In this context, we attempt to analyze the contribution of interfacial orbital hybridization, accompanied by interfacial charge transfer, to Rashba splitting. At the SIO/NiO interface (Fig. 4d), the unpaired d-electrons in Ir+4 (5 d5) occupy the dxz and dyz orbits, while the unpaired d-electrons in Ni+2 (3 d8) occupy the two eg orbits, \({d}_{{z}^{2}}\) and \({d}_{{x}^{2}-{y}^{2}}\). The Ir-5dxz,yz can hybridize directly with the out-of-plane Ni-3\({d}_{{z}^{2}}\) without bridging through O-2p, which facilitates larger Rashba SOC at the SIO/NiO interface. The decomposed Fermi surface on each d-orbit further confirms that the orbital hybridization between Ir-5dxz,yz and Ni-3\({d}_{{z}^{2}}\) indeed dominate the Rashba spin splitting (Fig. S13). In contrast, at the SIO/LSMO interface (Fig. 4e), the Ir-5dxz,yz is hybridized with the out-of-plane Mn-3\({d}_{{z}^{2}}\) orbits, but needs to be bridged by O-2p orbits. Moreover, there are ~2/3 Mn3+ (3d4) and ~1/3 Mn4+ (3d3) in LSMO, and the 2/3 Mn3+ has a possibility of 50–60% to provide the occupied 3\({d}_{{z}^{2}}\) electron for orbital hybridization43. The indirect orbital hybridization of Ir and Mn, together with the lower Mn-3\({d}_{{z}^{2}}\) electrons occupancy are not favorable for interfacial Rashba SOC. Therefore, the interface of SIO/NiO/LSMO has a much stronger REE compared with SIO/LSMO and the introduction of SIO/NiO results in enhanced spin current and SOT efficiency.

Furthermore, when the edge-sharing Ir-O/Ni-O octahedral interface is artificially set as a corner-sharing Ni-O/Ir-O octahedral interface, the Rashba splitting of its Fermi surface is heavily reduced (Fig. S14), which is related to the indirect orbital hybridization of Ir-5d/Ni-3d via O-2p. Therefore, this approach of analyzing the REE at the SIO/NiO interface from the perspective of orbital hybridization can be extended to a wider range of metal-oxygen framework structures. By now, the large ξDL in the SIO/NiO/LSMO heterostructure with optimal tNiO = 2 nm is elucidated, originating from the combined action of the intrinsic SHE of SIO and the interfacial REE. The SHC of SIO, deduced via ξDLof 0.45 (Fig. 2f), is estimated to be about 1.53 × 105 ħ/2e S m−1 at tNiO = 0 nm, while increasing to approximately 2.3 × 105 ħ/2e S m−1 at tNiO = 2 nm after considering the contribution of interfacial REE. Such a high SHC is comparable to that of conventional heavy metals yet surpasses most oxides and topological insulators, as shown in Fig. 5. Moreover, the heterostructure exhibits a significantly enhanced SOT efficiency (ξDL up to 0.68) compared to heavy metals, collectively demonstrating the potential of the SIO/NiO/LSMO heterostructure for developing high-performance spintronic devices with ultralow power consumption and robust operational stability.

Fig. 5: Comparison of ξDL and spin Hall conductance (σSH) for some representative spin current source materials13,14,26,51,52,53,54,55,56,57,58,59.
figure 5

Error bars denote experimental uncertainties. Variations may stem from measurement techniques (e.g., harmonic Hall or spin torque ferromagnetic resonance), interfacial conditions, or theoretical approximations.

Full size image

SOT-driven magnetization switching

Subsequently, we focus on the SOT-driven magnetization (M) switching. The SIO/NiO (t)/LSMO heterostructure was patterned into eight terminal devices as shown in Fig. 6a43. The write pulse currents IPulse-I and IPulse-II are applied along [110] and [\(\bar{1}\)10] direction, respectively. The Hall resistances Rxy are read out. When the M of the LSMO rotates in the (001) plane, the Hall resistance Rxy exhibits a sinφcosφ dependence (Fig. S4). Figure 6b shows the Rxy induced by pulse currents at different Hext applied along the [100] direction, and the pulse current density is 2.0 × 107 A cm−2 (in SIO layer) with a pulse width of 100 μs. It can be seen that when Hext = 0 or 3.5 mT (close to the coercivity), IPulse-I stabilizes Rxy at about −0.048 Ω, corresponding to M along −45° ([\(1\bar{1}0\)]) direction, while IPulse-II stabilizes Rxy at 0.048 Ω, corresponding to M along 45° ([110]) direction. The current-induced M switching coincides with the direction of DL torque τDL (Fig. 2a). Increasing the Hext and exceeding the saturation magnetization field (18 mT), the ∆Rxy decreases until it vanishes, indicating that the M has been fixed by Hext and cannot be driven by the current-induced SOTs. Notably, no significant ∆Rxy is observed in monolayer SIO under IPulse-I/IPulse-II driving (Fig. S15), confirming that the ∆Rxy in SIO/NiO/LSMO is due to SOT-driven M switching rather than thermal effects.

Fig. 6: SOT-driven magnetization switching.
figure 6

a Schematic of an eight-terminal M-switching device. A small read current is applied along [100], and the Hall resistance is read along [010]. The write pulse currents IPulse-I and IPulse-II are applied along the [110] and [\(\bar{1}\)10], respectively. b Rxy as a function of write current pulses in SIO/NiO (2)/LSMO with different Hext along [100]. 3.5 mT is close to the coercivity, and 50 mT exceeds the saturated field. c Rxy of SIO/NiO (2)/LSMO as a function of IPulse-I and IPulse-II. The IPulse-I stabilizes Rxy to −0.048 Ω while IPulse-II stabilizes Rxy to 0.048 Ω. d The stability of SOT-driven M switching of SIO/NiO (2)/LSMO in 20 consecutive sets of operations. e The switching ratio (the change in Rxy to the change in RPHE: ∆Rxy/∆RPHE) as a function of current density (SIO layer) in SIO/LSMO and SIO/NiO (2)/LSMO heterostructures. f tNiO-dependent critical switching current in SIO/NiO (t)/LSMO.

Full size image

The current density-dependent M switching in SIO/NiO (2)/LSMO without Hext is shown in Fig. 6c. It can be seen that IPulse-I and IPulse-II can switch M to −45° and 45°, respectively, and the critical switching current density (Jth) is 11.6 × 106 A cm−2 (9.6 mA). Figure 6d illustrates the M switching by successive IPulse-I/IPulse-II with JSIO = 20.0 × 106 A cm−2, and the stable ∆Rxy indicates the good stability of this oxide-based SOT device. Figure 6e shows that the switching ratio of M can reach more than 90% in both SIO/LSMO and SIO/NiO (2)/LSMO devices at suitable current densities. Moreover, the tNiO-dependent Jth for M switching is depicted in Fig. 6f. One can see that the Jth of heterostructure at tNiO = 0 nm is 14.1 × 106 A cm−2, which is reduced by about 18% to 11.6 × 106 A cm−2 at tNiO = 2 nm. Continuously increasing tNiO will slowly increase Jth, which coincides with the variation of ξDL with tNiO in Fig. 2f. The Jths are comparable to conventional heavy metal (Supplementary Information, Table S1), and this current-driven magnetization switching enables diverse Boolean logic operations in a single SIO/NiO (2)/LSMO device via programmable input pulse current magnitude and direction (Fig. S16–17).

Conclusions

A SOT-driven magnetization switching device based on an all-oxide SIO/NiO/LSMO heterostructure has been optimized. By modifying the thickness of the NiO intercalation, the damping-like SOT efficiency is increased by more than 51%, which is attributed to the Rashba-Edelstein effect at the SIO/NiO inversion asymmetric interface. Antiferromagnetic NiO functions as an effective medium for the magnon, which carries spin angular momentum for a long distance of more than 20 nm in the film’s normal direction. The device exhibits an SHC of greater than 105 ħ/2e S m−1 and a ξDL of up to 0.68. This configuration demonstrates robust room-temperature current-driven magnetization switching at low current density. These findings facilitate the development of oxide heterostructures for low-power, high-stability in-memory computing integrated spintronic devices.

Methods

Sample preparation

SrIrO3/NiO (t)/La2/3Sr1/3MnO3 heterostructure were grown on SrTiO3 substrate by pulsed laser (KrF, 248 nm) deposition method. The La2/3Sr1/3MnO3 (30 nm) film was grown at 600 °C with a pulsed laser at an energy density of 1.0 J cm−2 and an oxygen atmosphere of 150 mTorr. NiO (t nm) and SrIrO3 (6 nm) films were grown at 650 °C and 75 mTorr oxygen atmosphere with laser energy densities of 1.5 and 1.3 J cm−2, respectively. After deposition, the oxygen pressure was increased to 200 Torr, and the temperature was reduced to 20 °C at 10 °C min−1. Subsequently, the heterostructures were patterned into 10 μm (width) and 60 μm (length) Hall bars and eight-terminal devices by the standard photolithography and ion milling technology.

Sample measurements

The crystalline quality of the film was characterized by a high-resolution X-ray diffractometer (Smartlab, Rigaku) and a spherical aberration-corrected scanning transmission electron microscopy (Spectra 300, Thermofisher). The magnetic property was measured by a vibrating sample magnetometer. In second harmonic measurements, a sinusoidal current of 1 mA at 133 Hz was applied to the Hall channel by a Keithley 6221 current source, while the transverse first- and second-order harmonic Hall resistance were read off with two lock-in amplifiers (SR 830). The pulse current driving the magnetization switching is applied through a Keithley 6221, and the Hall resistance is read out by a voltameter Keithley 2182.

Density functional theory calculation

The calculations in this study are conducted with codes: Vienna ab-initio simulation package (VASP)44,45,46 and OPENMX47,48. The slab model includes two layers of rock-salt NiO and two layers of Pbnm-SrIrO3 (Fig. S9). The geometry structures of the relevant models are optimized with VASP until the residual forces are less than 0.02 eV Å−1. The projector augmented wave (PAW) potential is used, and the plane-wave cutoff is 550 eV. The GGA-PBE approximation is utilized to describe the exchange-correlation effect of electrons. The Brillouin zone (BZ) is sampled with a resolution better than 0.02 Å−1. The Fermi energy is determined by letting the occupation of electrons equal valence charges. For the static calculations on spin and electronic structures, the linear-combination-of-atomic-orbital (LCAO) formalism is utilized as implemented in OPENMX. The energy cutoff for real-space integration is 550 Ry. The noncollinear algorithm and the j-dependent pseudopotential are employed to capture the SOC and relativistic behavior of electrons. The GGA-PBE approximation is also used in calculations with OPENMX. The Fermi surface interpolations and orbital decompositions are done with the kSpin method49.

The calculations on spin Hall conductance are based on the Boltzmann equation with constant relaxation time approximation (CRTA)50:

$${\sigma }_{S,{xz}}^{y}=\frac{\tau e}{{V}_{{cell}}}\sum \limits_{m}{\int }_{{BZ}}\frac{{dk}}{{\left(2\pi \right)}^{2}}\frac{\partial {f}_{{{\boldsymbol{k}}}m}}{\partial {E}_{{{\boldsymbol{k}}}m}}{s}_{{{\boldsymbol{k}}}m}^{y}{v}_{{{\boldsymbol{k}}}m}^{z}{v}_{{{\boldsymbol{k}}}m}^{x}$$

\({f}_{{{\boldsymbol{k}}}m}={\left(\exp \left(-\frac{{E}_{{{\boldsymbol{k}}}m}}{{k}_{B}T}\right)+1\right)}^{-1}\) is the Fermi-Dirac function describing the equilibrate distribution of electrons at 300 K. k labels the BZ point. m is band index. τ is the mean relaxation time of distribution function. \({s}_{{{\boldsymbol{k}}}m}^{y}{v}_{{{\boldsymbol{k}}}m}^{z}\) is the spin-current propagating in z-axis with spin-polarization along y-axis, contributed by the km state. \({v}_{{{\boldsymbol{k}}}m}^{x}\) is the x-component of the velocity of km state. The BZ integration is conducted with a dense grid of 48×48×1.