Magnetization switching by magnon-mediated spin torque through an antiferromagnetic insulator

See allHide authors and affiliations

Science  29 Nov 2019:
Vol. 366, Issue 6469, pp. 1125-1128
DOI: 10.1126/science.aav8076

Toward magnonic devices

The field of magnonics aims to use spin waves (SWs) and their associated quasiparticles—magnons—as carriers of information. Compared with the movement of charge in conventional electronics, a major advantage of SWs is reduced Joule heating. However, SWs are trickier to direct and control. Two groups now go a step further toward magnon-based devices. Han et al. show that in multilayer films, domain walls can be used to change the phase and magnitude of a spin wave. Wang et al. demonstrate how magnon currents can be used to switch the magnetization of an adjacent layer.

Science, this issue p. 1121, p. 1125


Widespread applications of magnetic devices require an efficient means to manipulate the local magnetization. One mechanism is the electrical spin-transfer torque associated with electron-mediated spin currents; however, this suffers from substantial energy dissipation caused by Joule heating. We experimentally demonstrated an alternative approach based on magnon currents and achieved magnon-torque–induced magnetization switching in Bi2Se3/antiferromagnetic insulator NiO/ferromagnet devices at room temperature. The magnon currents carry spin angular momentum efficiently without involving moving electrons through a 25-nanometer-thick NiO layer. The magnon torque is sufficient to control the magnetization, which is comparable with previously observed electrical spin torque ratios. This research, which is relevant to the energy-efficient control of spintronic devices, will invigorate magnon-based memory and logic devices.

Spin current, a flow of spin angular momentum, is the essential ingredient for spin-transfer torque. One class of spin currents is the electrical spin current JS, which is associated with electron spins (13). When JS is absorbed by a magnet, the magnetization experiences an electrical spin torque, and its direction is reoriented (Fig. 1A). The electrical spin torque has opened the era for electrically controlled magnetic device applications—for example, magnetic random access memories (3). However, JS is commonly associated with charge flow so that Joule heat and corresponding power dissipation are unavoidable. Moreover, the spin propagation length (diffusion length) of JS is relatively short, typically on the order of nanometers (4), which prevents the delivery of spin information over long distances.

Fig. 1 Two types of spin-angular-momentum-transfer torque.

(A) Illustration of the magnetization (M) reorientation driven by the electrical spin torque (ST) by means of the electrical spin current JS. (B) Illustration of the magnetization (M) reorientation driven by the magnon torque (MT) by means of the magnon current JM. AFM, antiferromagnet; FM, ferromagnet.

These limitations could be overcome by another class of spin currents, the magnon current JM, for which the spin angular momentum is carried by precessing spin moments rather than moving electrons (5). The magnon torque associated with JM shows several advantages in comparison with the electrical spin torque. There is no electron movement in JM; therefore, much less Joule heat dissipation is expected. Moreover, JM can flow even in insulators for distances up to several micrometers (69), and thus, material systems for the magnon torque are not limited to electrical conductors. As JM approaches a magnet, it interacts with the magnetization through the exchange coupling, and consequently, the magnetization can be reoriented by the magnon torque (Fig. 1B). However, most studies on magnon currents have focused on long-distance transport (69), whereas experimental works on magnon-mediated spin torques (1013) have been limited to magnetic excitations or thermally driven domain wall motion (1416), and the magnetization switching has been realized only by the electrical spin torque. In this work, we observed a giant magnon torque with a ratio for the JM generation of ~0.3, which is comparable with the electrical spin torque ratio of topological insulators (1719), and we experimentally demonstrated that the magnetization can be efficiently switched by the magnon torque without any external magnetic field at room temperature.

We fabricated Bi2Se3 (8 nm)/NiO (tNiO = 0 to 100 nm)/NiFe (Py, 6 nm) structures, where tNiO is the thickness of NiO, with the Bi2Se3 (fig. S1 and materials and methods) (20) acting as a highly efficient spin current source (1719). In-plane current injection to the Bi2Se3 produces electrical spin currents at the Bi2Se3/NiO interface and excites JM in the NiO layer. We chose NiO, an antiferromagnetic insulator, as a magnon-current medium because magnons are the sole spin angular momentum carriers (69, 2124) in the NiO layer.

Because antiferromagnetic ordering has a crucial role in magnon transport through antiferromagnets (25), we first characterized the thickness dependence of the antiferromagnetic ordering in the NiO layer by measuring hysteresis loops. The antiferromagnetic ordering can be estimated by the exchange bias field or coercivity; both increase as the antiferromagnetic ordering improves (26). The temperature-dependent exchange bias field Hex of Bi2Se3/NiO/Py is shown in Fig. 2A for various tNiO. Hex increases with the NiO thickness at tNiO ≤ 25 nm at 2 K. The blocking temperature Tb, at which the Hex becomes zero, increases as tNiO increases (Fig. 2B). A similar trend is observed for the coercivity Hc at room temperature (Fig. 2C); Hc increases with tNiO for tNiO ≤ 30 nm. Hc then slightly decreases with increasing tNiO, which is a typical behavior of antiferromagnetic films. The results suggest that the antiferromagnetic ordering in the NiO layer gradually improves as tNiO increases. The smaller Tb than that in NiO bulk (Fig. 2B) is related to the polycrystalline structure of NiO grown on Bi2Se3 (Fig. 2D). When the NiO layer is directly deposited on the c-plane sapphire substrate, it shows a good crystallinity with the dominant (111) plane. However, when the NiO layer is deposited on Bi2Se3, it is polycrystalline, possibly because of the lattice mismatch between the NiO and Bi2Se3 layers.

Fig. 2 Characterization of Bi2Se3/NiO/Py structures.

(A) The exchange bias as a function of temperature for various NiO thicknesses (tNiO). (B) The blocking temperature deduced from (A). (C) The coercivity as a function of tNiO at room temperature. (D) X-ray diffraction patterns of sapphire substrate/Bi2Se3 (8 nm)/NiO (100 nm) and sapphire substrate/NiO (100 nm).

We next characterized the magnon torque using the spin torque ferromagnetic resonance (ST-FMR) measurement (Fig. 3A) (17, 20, 27, 28). A radio frequency current (IRF; current density JC in the Bi2Se3 layer) was applied to the device and generated electrical spin currents with a spin polarization denoted in Fig. 3A with a red arrow at the Bi2Se3/NiO interface. JM is then induced in the NiO layer through the exchange interaction between the spins and nearby NiO moments (29). Passing through the NiO layer, JM exerts a magnon torque on the Py layer. Consequently, the Py magnetization is excited into the precession mode, generating a ST-FMR signal Vmix. Vmix (Fig. 3B, open symbols) from a representative device with tNiO = 25 nm is fitted by Vmix = VSFS + VAFA, where VSFS and VAFA are the symmetric and antisymmetric components, respectively. By adopting an established analysis method (17, 20, 27), we evaluated the spin torque ratio (θi = Ji/JC) for the symmetric component, which is analogous to the spin Hall angle in the electrical spin-orbit torque scheme (17, 27). θi is caused either by the electrical spin torque (stemming from JS, i = S) or by the magnon torque (stemming from JM, i = M), depending on the NiO thickness, which is discussed later.

Fig. 3 ST-FMR measurement of magnon torque.

(A) A diagram of the ST-FMR measurements, illustrating the magnetization precession driven by the spin torque, including the damping-like torque τDL and/or field-like torque τFL. The black arrow denotes the direction of IRF with a current density JC. The red and blue arrows indicate spin polarizations and magnon current JM, respectively. (B) A typical ST-FMR signal (open symbols) from a Bi2Se3/NiO (25 nm)/Py (6 nm) device at 10 GHz and 300 K with fits (solid lines), where the blue and green lines indicate the symmetric (VSFS) and antisymmetric Lorentzian (VAFA) component, respectively. (C) The spin torque ratio θi deduced from the ST-FMR data (solid circles) and the terahertz emission amplitude (open circles) as a function of tNiO at 300 K. The red curve is a fit using Eq. 1. For the fitting, we used ηθ = 0.8 and GA/F/σm=3×108 m–1. (Inset) The assumed magnon diffusion length (lm) as a function of tNiO. The star symbol corresponds to θi obtained from the ST-FMR measurement of the control device with 6-nm MgO insertion between Bi2Se3 and NiO layers. (D) Temperature dependence of θi for Bi2Se3/NiO (tNiO = 2, 5, and 25 nm)/Py (6 nm) devices.

Shown in Fig. 3C is θi versus tNiO at room temperature, at which the NiO/Py interface contribution is subtracted [fig. S2 and (20), section 2]. For the device without the NiO layer—the Bi2Se3/Py bilayer—θi is 0.67, which is consistent with a previous report for Bi2Se3 (18). We observed that θi abruptly decreases from 0.67 to ~0 by inserting only a 2-nm NiO layer between the Bi2Se3 and Py layers. In this NiO-thickness range, no noticeable exchange bias nor enhanced coercivity was observed, even at low temperature (Fig. 2). It indicates that the antiferromagnetic ordering is weak, and corresponding magnon torque has a negligible role in this thickness range. Therefore, the small θi would be of purely electrical origin—electron spin tunneling through a normal insulator such as MgO or SiO2 (21, 22).

The presence of magnon torque is evident for larger values of NiO thickness in which magnons are the only spin-angular-momentum carriers. We found that θi starts to increase as tNiO increases above 10 nm (Fig. 3C). θi shows a peak value of ~0.3 at tNiO = 25 nm and then gradually decreases with further increasing tNiO up to 100 nm. The peak θi value is of similar magnitude to the electrical spin Hall angle of topological insulators (1719) and is higher than that of heavy metals Pt and Ta (30).

We used independent terahertz emission measurements to double check the NiO thickness–dependent behavior [fig. S3 and (20), section 3]. The terahertz emission amplitude characterizes the spin-to-charge conversion, which is the reciprocal process of the ST-FMR measurements (charge-to-spin) (31, 32). A similar trend in Fig. 3C by using both ST-FMR and terahertz techniques validates our observations of NiO thickness–dependent magnon torques. As a control experiment, we inserted a 6-nm MgO layer between Bi2Se3 and NiO layers to block the spins through the NiO layer (33), and the obtained θi is negligible (Fig. 3C, star symbol). This rules out the possibility that observed torque is generated at the NiO/Py interface [table S1 and (20), section 4]. The temperature dependence of θi measured for Bi2Se3/NiO (tNiO = 2, 5 and 25 nm)/NiFe devices is shown in Fig. 3D. The θi shows a peak at a certain temperature close to the antiferromagnetic transition in the NiO layer for each device, at which the enhanced spin fluctuations and magnons facilitate the spin transport, which is in line with the previous reports (23, 24, 34, 35). These results confirm the magnon-originated nature of the spin torque.

To gain insight into the thickness dependence of θi, we estimated a transverse magnon current at the NiO/Py interface using a simple drift-diffusion model. Because the NiO layer is polycrystalline (Fig. 2D), we assumed that magnon propagation in the NiO layer is diffusive (25, 36). In this model, the resultant transverse magnon current Jm at the NiO/Py interface isJm=θJiC=ηθJC2GA/FlmκGA/Flm(1+κ2)+2πσm(1κ2)(1)where θ is the spin Hall angle of Bi2Se3 layer; η is the angular momentum loss at the Bi2Se3/NiO interface; GA/F is the interfacial magnon conductance for the transverse component between NiO and Py (36), which is analogous to the mixing conductance of the spin transport theory (37); κ = exp(–tNiO/lm); and lm and σm are the magnon diffusion length and magnon conductivity of the NiO layer, respectively [(20), section 5]. The red curve in Fig. 3C is the fit of experimental data to Eq. 1. For the fitting, we assumed that lm increases with tNiO and saturates at ~30 nm (Fig. 3C, inset); the value of 30 nm was estimated from the exponential decrease of θi for tNiO ≥ 25 nm in Fig. 3C; the saturation lm of ~30 nm is consistent with that recently reported in polycrystalline NiO (38). This assumption is motivated by the experimental results shown in Fig. 2; the behaviors of Hex (or Tb) and Hc with tNiO suggest an improvement of antiferromagnetic ordering with tNiO. To support this assumption, we performed atomistic lattice spin model calculations for the relation between the antiferromagnetic ordering and lm and found that they are correlated (fig. S5). The reasonable fit shown in Fig. 3C implies that the tNiO-dependent change of lm is a possible explanation for the experimental observation, even though a more detailed study is required for a quantitative understanding.

Last, we demonstrated magnetization switching induced by the magnon torque in the Bi2Se3/NiO/Py heterostructure at room temperature (Fig. 4). In switching devices, the Py layer was patterned into a rectangular shaped island to prohibit current shunting through the Py layer (Fig. 4, A and B) (20). The switching results for tNiO = 25 nm are shown in Fig. 4, C to F, measured with a magneto-optic Kerr effect (MOKE) microscope by injecting a pulsed current I. We first saturated the magnetization in the Py island (Fig. 4C, yellow box) along the +y axis with an in-plane magnetic field. Then, we removed the field and applied I (JC ~ 1.27 × 107 A cm–2) along the +x axis. We found that the Py magnetization was switched to the –y axis, indicated by the contrast change from dark to light in the yellow box (Fig. 4D). We then initialized the Py magnetization along the −y axis (Fig. 4E) and applied I (JC ~ −1.27 × 107 A cm–2) along the −x axis. The Py magnetization was switched to the +y axis (Fig. 4F). This bidirectional switching depending on the current polarity excludes the possibility that the switching is governed by Joule heating–induced effects. This magnon torque–driven magnetization switching is reproducible in other devices (fig. S6). We also demonstrated magnetization switching induced by magnon torques in Bi2Se3/NiO (25 nm)/Co40Fe40B20 (CoFeB) trilayers at room temperature (fig. S7).

Fig. 4 Magnetization switching induced by magnon torque in the Bi2Se3/NiO/Py devices at room temperature.

(A) Illustration of the structure of the magnon torque switching device with an isolated Py rectangle defined on top of the NiO layer. (B) Optical microscope image of a device with electrodes, where the sample functional region is indicated with a red dotted box and an isolated Py rectangle is denoted with a yellow box. (C to F) MOKE images for magnon-torque-driven magnetization switching in the Bi2Se3/NiO (25 nm)/Py device by injecting a pulsed current I along the [(C) and (D)] +x axis or [(E) and (F)] –x axis at room temperature. (G to J) MOKE images for a Bi2Se3/NiO (5 nm)/Py device by injecting I along the [(G) and (H)] +x axis or [(I) and (J)] –x axis at room temperature. (K to N) MOKE images for the Bi2Se3/NiO (25 nm)/Cu (6 nm)/Py device by injecting I along the [(K) and (L)] +x axis or [(M) and (N)] –x axis at room temperature. In (C) to (N), the dark contrast represents the magnetization along the +y axis, and the light contrast represents the magnetization along the –y axis. The direction of magnetization is indicated with white arrows. The current density JC in the Bi2Se3 layer is denoted underneath each image.

The switching experiments on another device of Bi2Se3/NiO (5 nm)/Py are shown in Fig. 4, G to J, in which θi estimated from ST-FMR is negligible (Fig. 3C). Following the same measurement procedures as in Fig. 4, C to F, we could not observe the magnetization switching, even with a larger JC. This behavior was reproducible in other devices with tNiO = 5 nm. It excludes a possibility that the current-induced Oersted field, which is also present in the devices with tNiO = 5 nm, is the origin of the magnetization switching observed in the devices with tNiO = 25 nm. We also performed switching experiments on other devices with isolated Py islands at various tNiO and found that the switching efficiency is qualitatively consistent with θi estimated from ST-FMR (fig. S9). Therefore, our results provide unambiguous evidence that the switching is governed by the magnon torque.

From the above experiments, one question is whether direct exchange coupling between NiO and Py magnetic moments is essential for large magnon torques or not. To answer this, we performed switching experiments in Bi2Se3/NiO (25 nm)/Cu (6 nm)/Py devices, in which the Cu layer breaks direct exchange coupling but allows spin transmission through the transfer of spin angular momentum from magnons to electrons (Fig. 4, K to N). Both the Py and Cu insertion layer were patterned into a rectangular shaped island to eliminate current shunting through the Cu/Py bilayer. Following the same measurement procedures as in Fig. 4, C to F, we observed robust magnetization switching with JC ~ 1.55 × 107 A cm–2, which was reproducible in other devices with a Cu insertion (fig. S8). The observations confirm that direct exchange coupling between the NiO and Py magnetic moments is not essential for the magnon torque–induced magnetization switching.

Our demonstration reveals the ability of magnon torque to switch magnetization, a process that is as energy efficient as electrical spin torques. This research will broaden the scope of not only magnon-related studies that are mainly focused on magnon transport (69, 2124, 39) but also spintronics studies, in which advances have relied largely on the electrical spin torque. In this study, we induced the magnon torque in an antiferromagnetic insulator by injecting the electric current to a Bi2Se3 layer as a proof of principle. However, our work is just a starting point to explore the magnon torque–driven magnetization switching. We expect that all-magnon-driven magnetization switching, without involving electrical parts, could be achieved in the near future.

Supplementary Materials

Materials and Methods

Supplementary Text

Figs. S1 to S9

Table S1

References (4153)

References and Notes

  1. Materials and methods are available as supplementary materials.
Acknowledgments: We thank J. W. Yu for illustration-making assistance. Funding: This work was supported by SpOT-LITE program (A*STAR grant, A18A6b0057) through RIE2020 funds and the National Research Foundation (NRF), Prime Minister’s Office, Singapore, under its Competitive Research Programme (CRP award NRFCRP12-2013-01). K.-J.L. was supported by the National Research Foundation of Korea (NRF-2015M3D1A1070465 and 2017R1A2B2006119) and the KIST Institutional Program (project 2V05750). Y.W. was supported by the Fundamental Research Funds for the Central Universities (project 82232016). Author contributions: Y.Wa. and H.Y. conceived of the research; Y.Wa., Y.Y., R.M., and K.C. fabricated the devices; D.Z., Y.Wa., and E.L. grew films; Y.Wa., S.D.P., J.L., K.L., and S.S. performed measurements; G.G. and K-J.L. contributed to the theoretical model; S.-H.O., D.-H.K., and K-J.L. performed the atomistic model calculation; Y.Wa., K-J.L. and H.Y. wrote the manuscript; all authors commented the manuscript; and H.Y. led the project. Competing interests: The authors declare no competing interests. Data and materials availability: Data reported in this paper are archived online at Harvard Dataverse (40).
View Abstract

Stay Connected to Science

Navigate This Article