Letter Corrected: Publisher Correction

https://doi.org/10.1038/s41586-018-0853-0

Magnetic and magnetic inverse spin Hall effects in a

non-collinear antiferromagnet
Motoi Kimata1,2,5, Hua Chen3,6, Kouta Kondou2,4, Satoshi Sugimoto1,7, Prasanta K. Muduli1, Muhammad Ikhlas1,
Yasutomo Omori1, Takahiro Tomita1,2, Allan. H. MacDonald3, Satoru Nakatsuji1,2 & Yoshichika Otani1,2,4*

The spin Hall effect (SHE)1–5 achieves coupling between charge
currents and collective spin dynamics in magnetically ordered
systems and is a key element of modern spintronics6–9. However,
previous research has focused mainly on non-magnetic materials,
so the magnetic contribution to the SHE is not well understood.
Here we show that antiferromagnets have richer spin Hall properties
than do non-magnetic materials. We find that in the non-collinear
antiferromagnet10 Mn3Sn, the SHE has an anomalous sign change
when its triangularly ordered moments switch orientation. We
observe contributions to the SHE (which we call the magnetic SHE)
and the inverse SHE (the magnetic inverse SHE) that are absent in
non-magnetic materials and that can be dominant in some magnetic
materials, including antiferromagnets. We attribute the dominance
of this magnetic mechanism in Mn3Sn to the momentum-dependent
spin splitting that is produced by non-collinear magnetic order. This
discovery expands the horizons of antiferromagnet spintronics and
spin–charge coupling mechanisms.
The conventional SHE refers to transverse spin currents that flow
in non-magnetic conductors in response to an electric field and are
responsible for spin accumulation near sample surfaces1–5. The classical
dipolar coupling between magnetic moments and a magnetic field has
conventionally been used for electrical control of the magnetization of
a ferromagnet through, for example, a solenoid. Recent developments
in spintronics have led to further advancement, enabling us to use the
SHE, which couples charge currents directly to spin polarization via
the spin–orbit interaction1–7. This current-induced spin polarization
has drawn much attention because of its potential application for effi-
cient magnetization switching via the spin-transfer torque8. Material-
dependent spin Hall properties have therefore been studied mostly in
non-magnetic materials. Another possible reason for the scarcity of
research of the SHE in magnetic materials is the conventional wisdom
that the spin current, defined in the most commonly used way, is
time-reversal-invariant.
Because of spin–orbit coupling, spin is never conserved in solid-state
systems. For this reason, spin Hall currents, unlike charge Hall currents,
can be measured only indirectly, by examining observables, such as spin
accumulations or spin torques applied to adjacent ferromagnets7–9, that
also depend on spin relaxation at surfaces and interfaces. Indeed, all
previously studied current-induced spin Hall phenomena5 have been
successfully interpreted by combining the notions of bulk spin currents,
spin-mixing and relaxation phenomena near interfaces. An alterna-
tive approach11–13, which is useful to understand the magnetic SHE
(MSHE), is to view spin Hall effect phenomena directly, by targeting a
response that is ultimately measured; for example, the local spin-density
response to electric fields at the sample boundary. The dependence
of the SHE signals on magnetic-order parameter reversals can then
be simply interpreted in terms of the symmetries of well defined lin-
ear response functions. In this way, we demonstrate that the MSHE
originates from a reactive counterpart of the dissipative spin response
responsible for the ordinary SHE.
The material focused on here is Mn3Sn, which shows non-collinear
antiferromagnetic order of Mn magnetic moments that have an
inverse triangular spin configuration on stacked kagome lattices. The
almost perfect 120° non-collinear antiferromagnetic order leads to an
extremely weak ferromagnetic moment (about 0.002μB per Mn atom;
μB is the Bohr magneton) owing to canting. Despite the tiny total spin
polarization, the symmetry of Mn3Sn enables14,15 the appearance of an
anomalous Hall effect16–18 (AHE), which was recently observed exper-
imentally10. The AHE in Mn3Sn is surprisingly strong, comparable to
that of conventional ferromagnetic metals, and due to Bloch bands with
large momentum-space Berry curvatures10. A benefit of the inverse
triangular order is small magnetic anisotropy within the kagome plane,

a I b Fig. 1 | Device structure and experimental

Cu (normal metal) setup for spin-accumulation detection.
V a, Scanning electron microscope image of the
NiFe Surface spin spin-accumulation device. The microfabricated
[0001] (spin detector) accumulation, δ〈s〉
-
[0110] Mn3Sn single crystal is indicated by the red
dashed line. The blue square is a NiFe
T
B ferromagnetic electrode. The edge of the Mn3Sn
2 μm crystal is covered with a 30-nm-thick Al2O3
--
[2110] insulating layer (yellow areas) to avoid electrical
Mn3Sn single contact at the edge. The brown areas are
c R crystal non-magnetic Cu leads. b, Schematic of the
Current, I measurement setup. The external magnetic field
was rotated within the basal plane of the device
and the field angle θ was measured from the
B B
[0110] axis. c, Expected rectangular hysteresis
loop caused by spin accumulation. The jump in
the resistance corresponds to the magnetization
reversal of the NiFe layer.
1
Institute for Solid State Physics, University of Tokyo, Kashiwa, Japan. 2CREST, Japan Science and Technology Agency (JST), Saitama, Japan. 3Department of Physics, University of Texas at Austin,
Austin, TX, USA. 4Center for Emergent Matter Science, RIKEN, Wako, Japan. 5Present address: Institute for Materials Research, Tohoku University, Sendai, Japan. 6Present address: Department of
Physics, Colorado State University, Fort Collins, CO, USA. 7Present address: National Institute for Materials Science, Tsukuba, Japan. *e-mail: yotani@issp.u-tokyo.ac.jp

3 1 J A N U A R Y 2 0 1 9 | V O L 5 6 5 | N A T U RE | 6 2 7
© 2019 Springer Nature Limited. All rights reserved.
 RESEARCH Letter
B
a b
0.2 0.2
T = 30°
0.1 0.1
ΔR (mΩ)
ΔR (mΩ)
0 0
ΔR(–B)
–0.1 –0.1
–0.2 –0.2
–0.15 –0.10 0.05 0 0.05 0.10 0.15 –0.15 –0.10 0.05
B (T)
Fig. 2 | Spin-accumulation signal. a, b, Magnetic-field dependence
of resistance measured between the NiFe and Cu electrodes at room
temperature. The results in a and b were obtained after saturation of
Mn3Sn with a sufficiently large magnetic field B of −0.75 T and +0.75 T,
respectively. The corresponding spin structures of Mn3Sn are shown
in the insets; blue arrows represent Mn sublattice moments. The spin-
accumulation signal for each magnetic structure (ΔR(−B) and ΔR(+B) in
a and b, respectively) is defined as the difference in resistance at +0.15 T
which makes it easy to manipulate moment orientations using external
magnetic fields. Mn3Sn thus provides a remarkable material platform
that combines ferromagnet-like properties, such as a large AHE and
easily switchable magnetization, with desirable properties of antifer-
romagnets, namely, very small total magnetization, which generates
weak dipolar fields, and rapid magnetization dynamics. These factors
combine with strongly momentum-dependent exchange potentials to
make Mn3Sn ideally suited for observation of the MSHE.
To study the SHE in a non-collinear antiferromagnet, we built a
device consisting of ferromagnetic NiFe (blue square pad in Fig. 1a;
20 nm in thickness) and non-magnetic Cu electrodes formed on the top
surface of a microfabricated 100-nm-thick Mn3Sn single crystal. The
single crystal used for the microfabrication was grown by the Bridgman
technique and its composition was estimated to be Mn3.06Sn0.94
(see Methods). A thin 0.5-nm-thick MgO layer was inserted between
the Mn3Sn surface and the NiFe layer to minimize electrical current
flow through NiFe. An electric current was applied along the [2110] axis
of the crystal and the voltage was measured between electrodes aligned
along the [0110] axis, perpendicular to the current. The external mag-
netic field was rotated within the plane between the charge current and
the voltage probes (kagome plane) to minimize spurious voltage signals
due to the AHE. The magnetic-field angle θ was measured from the
[0110] direction. If current inside Mn3Sn generates spin accumulation
with a component parallel to the NiFe magnetization (Fig. 1b) it shifts
the electrochemical potential across the Mn3Sn/NiFe interface and can
therefore be detected as a voltage between NiFe and the Cu electrodes.
Because the switching field of the microfabricated Mn3Sn crystal (about
2,000 Oe; see Methods) is larger than that of NiFe (about 500 Oe)19, only
the NiFe magnetization is reversed when sweeping the external
magnetic field within ±1,500 Oe. As depicted in Fig. 1c, a rectangular
resistance–magnetic field (R–B) hysteresis loop corresponding to the
switching of NiFe magnetization is expected for a fixed polarization
direction of the accumulated spins.
Such a rectangular R–B hysteresis loop has been observed in the
magnetic-field dependence of the transverse voltage between NiFe and
the Cu electrodes (Fig. 2a, b). The plots in Fig. 2a, b provide evidence
that the rectangular loops originate from spin accumulation at
the Mn3Sn surface. The key observation in this work is that the spin-
accumulation signal changes indicate when the Mn3Sn sublattice
moments are reversed, as illustrated in the insets of Fig. 2a, b. This
unexpected sign change demonstrates that the dominant contribution
to the SHE in Mn3Sn is odd under time reversal (T ).
6 2 8 | N A T U RE | V O L 5 6 5 | 3 1 J A N U A R Y 2 0 1 9
© 2019 Springer Nature Limited. All rights reserved.
B
c
0.3
0.2
0.1
ΔR (mΩ)
ΔR(+B) 0
–0.1
–0.2
–0.3
0 0.05 0.10 0.15 –90 –60 –30 0 30 60 90
B (T) T (°)
and −0.15 T. The additional plateaus observed at intermediate resistance
probably arise from asymmetric domain-wall propagation in the NiFe
pad due to the asymmetric shape of the pinning potential, which is caused
by irregular edges or burr. c, Magnetic-field-angle dependence of the
spin-accumulation signal. The solid line is a fit by sin(2θ) and sin(4θ)
predicted by the model calculations. The error bars are estimated from the
uncertainty of the signal due to the relatively small signal-to-noise ratio.
We also measured the angular dependence of the spin-accumulation
signal ΔR(±B) by rotating the magnetic field by θ from the [0110]direc-
tion in the kagome plane, where ΔR(±B) = R(+0.15 T) − R(−0.15 T)
for the case of the magnetization of Mn3Sn pointing to the ±B direc-
tion. Figure 2c shows the angular variation of ΔR. Remarkably, ΔR
changes sign around 0° and shows oscillatory behaviour at higher θ.
This is in stark contrast to the behaviour observed for the conventional
SHE, where a simple cosθ dependence of ΔR is expected7. In this exper-
iment, we used the standard lock-in technique with an a.c. current. We
show below that this angular variation of ΔR is qualitatively consistent
with our model calculations for the MSHE in Mn3Sn.
We next show that this T -odd SHE is also manifested in a comple-
mentary inverse SHE (ISHE) experiment that measures the electric
current generated by precessing magnetization, that is, spin pumping6,20.
The measurement geometry (Fig. 3a) is designed to detect charge cur-
rents converted from spin currents pumped into Mn3Sn (100 nm thick-
ness) by an adjacent ferromagnet NiFe (10 nm thickness).
Figure 3b, c shows the d.c. voltage versus the d.c. magnetic field B,
measured at an input radio-frequency power of 26 dB m and frequency
f = 10 GHz, with the positive and negative magnetic field, respectively,
at θ = 40°. A resonance peak is observed at ±0.125 T and the voltage
spectrum is decomposed into symmetric and asymmetric Lorentzian
voltage components, indicated by the green and blue lines, respectively,
in Fig. 3d, e. The frequency dependence of the resonance field Br can
be fitted by the Kittel formula 2πf = γ Br (Br + μ0Ms ) (Fig. 3f), where
μ0 is the magnetic constant and γ is the gyromagnetic constant
(the ratio of the magnetic dipole moment to the angular momentum)
of the atom. In addition, the saturation magnetization Ms estimated
from the fit (Ms = 0.94 T) is typical of NiFe thin films, indicating that
the observed resonance corresponds to the ferromagnetic resonance
(FMR) mode of the NiFe layer.
An important feature of the symmetric Lorentzian voltage signal is
that its sign does not change upon reversal of B (Fig. 3d, e). This con-
trasts the conventional case for the combination of spin pumping and
the ISHE6. Because the symmetric Lorentzian voltage signal is greatly
enhanced in the NiFe/Mn3Sn layer compared with that in the NiFe
single layer, the observed symmetric voltage contribution is not arising
from the rectified anisotropic magnetoresistance voltage in the NiFe
layer21. The even dependence on the field is consistently observed over
a field range from 0.12 to 0.19 T, which exceeds the switching fields
of both the microfabricated Mn3Sn and the NiFe films in the present
device. We note that the symmetric Lorentzian voltage amplitude is

a b
2 T = 40°
NiFe 1 setup and sample structure. Mn3Sn (grey)
V (μV)
Brf
[0001]
- 0
[0110]
–1
T B
-- d 2
[2110]
1 symmetric (green line) and asymmetric (blue
V (μV)
Mn3Sn V line) Lorentzian voltage contributions (d, e).
0
–1
–0.20
f g
15 1.0
0.5
f (GHz)
V (μV)
10
0
5 –0.5
–1.0
0
0 0.05 0.10 0.15 0.20 –90
Br (T)
almost independent of the resonance field above 0.12 T, which is the
switching field of Mn3Sn in the present device (see Methods). We also
note that the symmetric voltage contribution shows slight asymme-
try in the polarity of the applied field. This indicates mixed odd and
even dependences on the field direction. Assuming the coexistence of
ordinary odd ISHE and even magnetic inverse SHE (MISHE) contri-
butions, the MISHE contribution is about 80%. Figure 3g shows the
angular dependence of the MISHE contribution on the magnetic field
for f = 10 GHz. Here the definition of the magnetic-field angle θ is the
same as in the spin-accumulation device. The angular variation of the
MISHE voltage can be fitted by sin(2θ) and sin(4θ) components, and
the voltage sign changes around θ = 0, similar to that in the MSHE
detected in the spin-accumulation experiment. This angular depend-
ence is totally different from the expected angular dependence in con-
ventional ISHE voltage, but qualitatively consistent with our model
calculations below (see also Methods). We note here that all the exper-
imental results discussed above were obtained at room temperature.
Here we explain the origin of the MSHE and MISHE, and their dom-
inance over their conventional counterparts in Mn3Sn, by using the
language of quantum kinetic theory22,23. As described above, we view
the SHE directly as a linear response of spin density to an electric field
at the boundaries of a bulk material. The spin density can be expressed
in terms of the non-equilibrium single-particle density matrix, which
must satisfy the quantum Liouville equation:
∂ρ i
+ [H , ρ] = 0
∂t ħ
where t is time, ħ is the Dirac constant and H is the total Hamiltonian,
including the Bloch Hamiltonian of the perfect crystal HB and contri-
butions from both disorder and external driving fields. For qualitative
purposes, it is safe to use a relaxation-time approximation for the disorder
contribution to [H, ρ]. (In the quantum kinetic approach, we have ignored
possible contributions from skew scattering, which can dominate the SHE
and ISHE in materials with weakly spin–orbit-coupled bands. Our discus-
sion is thus more suitable for systems that exhibit strong SHE and ISHE
effects.) Adding the well known driving-field contribution, we obtain a
simple algebraic equation for the steady-state linear response ρ(1):
i ∂ρ (0) ρ (1)
[HB, ρ (1)] + · eE + =0
ħ ħ∂k τ
where τ is the relaxation time, k is the wavevector, E is the electric field, e
is the electron charge and ρ(0) is the equilibrium density matrix. In good
Letter RESEARCH
c Fig. 3 | Spin pumping and MISHE.
T = 40° a, Schematic of the spin-pumping measurement
and NiFe (blue) layers are attached to each
other. The magnetic field is tilted by an angle
θ from the [0110] direction. Brf is the applied
radio-frequency field. b–e, Detected voltage
e for positive (c) and negative (b) magnetic
fields. These spectra can be decomposed into
f, Frequency dependence of resonance fields.
The solid line shows the Kittel formula with a
–0.15 –0.10 –0.05 0.05 0.10 0.15 0.20 saturation magnetization of 0.94 T. g, Magnetic-
B (T) B (T) field-angle dependence of the symmetric
voltage contribution, which shows an even
dependence on magnetic-field polarity. The
solid line is a fit by sin(2θ) and sin(4θ), as
expected by the model calculation. The
smaller sin(4θ) component in the pumping
measurement compared to that obtained in
the spin-accumulation measurement is
–60 –30 0 30 60 90 consistent with our model calculation
T (°) (see Methods).
metals τ is large and we can ignore the last term when solving for the
elements of ρ(1) that are off-diagonal in band index. The non-equilibrium
spin density δ⟨s⟩ = tr[ρ (1)(ħ/2)σ ], where σ is the spin Pauli matrix vector
and angle brackets denote the expectation value of spin in the non-
equilibrium steady state, can then be separated into contributions from
diagonal (d; intra-band) and off-diagonal (od; inter-band) elements of ρ(1):
eħ β ∂f β α
δ⟨s α⟩d = − τE ∑ ∬ n vnnσnn δ(r ′ − r )dr ′ dk
2 n ∂ε n
eħ β
δ⟨s α⟩od = − E ∑ ∬ (fm −fn ) (3)
2 m≠n
 vmnβ α 
σnm 
× Im  2
δ (r ′−r )dr ′dk
 (ε m−ε n) 
where τ is the relaxation time, v is the velocity operator, ε is the band
energy, f is the Fermi–Dirac distribution function, r and r′ represent
the position and δ is the Dirac delta function. Because the spin-density
response functions relate the electric field, which is even under T , to
the spin density, which is odd under T , they must change sign under
T . Because the relaxation process, accounted for by the relaxation time
τ, is odd under T , only the diagonal contribution can be non-zero in a
T -invariant non-magnetic system. Moreover, because the electric field
is a polar vector that is odd under spatial inversion I , whereas the
(1) spin density is an axial vector even under I , spatial-inversion symmetry
must be broken for such a response to occur24–26. It is precisely for this
reason that spin accumulation occurs at the boundaries of a bulk mate-
rial, where spatial inversion symmetry is broken locally. These symme-
try considerations allow us to associate the ordinary SHE with the
intra-band spin-density/electric-field response. By contrast, the MSHE
observed in Mn3Sn is due entirely to the inter-band spin-density/
electric-field response and can be non-zero only when HB is not T -
invariant (that is, only in magnetic systems), and because it is not multi-
plied by τ, it must be odd under magnetization reversal. The microscopic
origins of the ordinary SHE and the MSHE are entirely distinct.
To qualitatively explain the main experimental features of the MSHE
in Mn3Sn, we constructed an s–d type toy model that has the same sym-
metries as the NiFe/Mn3Sn bilayer system with the interface normal
(2) to the [0001] axis of Mn3Sn, and calculated the MSHE response using
the quantum kinetic approach explained above. The model includes s
electrons that are itinerant and are coupled to the localized magnetic
moments due to d electrons through exchange coupling. The model
3 1 J A N U A R Y 2 0 1 9 | V O L 5 6 5 | N A T U RE | 6 2 9
© 2019 Springer Nature Limited. All rights reserved.
 RESEARCH Letter
a b
B B
Fig. 4 | Model calculation of angular variation for the MSHE.
a, b, Bilayer kagome toy model mimicking the (0001) surface of Mn3Sn
with different magnetic-field directions of θ = 0° and 30°. The sublattice
moments rotate in opposite directions to that of the external magnetic
field. c, Magnetic-field-angle dependence of the interband contribution to
consists of two kagome layers, AB-stacked (Fig. 4a, b), with on-site
exchange fields forming the pattern of the Mn moments in Mn3Sn
(a Rashba-type spin–orbit coupling that breaks the in-plane mirror
symmetry) and an on-site potential difference between the two layers
(see also Methods).
The results for the inter-band contribution δ⟨s⟩od (equation (3)),
projected to the magnetic-field direction B^ as in the experimental con-
figuration, are plotted in Fig. 4c. In the plot we take into account the
opposite rotation of the local moments and the total weak magnetiza-
tion (Fig. 4a, b). The angular variation of δ⟨s⟩od ⋅ B^ is qualitatively con-
sistent with the experimental results in Fig. 2c, and has the expected
even-order Fourier components due to the T -even nature of δ⟨s⟩od ⋅ B^.
For example, the small peaks around ±70°, similar to that in Fig. 4c,
are due to a higher-order Fourier component of δ⟨s⟩od ⋅ B^ and are a
reflection of the crystallinity of the Mn3Sn surface.
The ISHE (and MISHE) can be addressed by the same quantum
kinetic approach by calculating the non-equilibrium electric current
generated by a time-dependent exchange field; see Methods for a more
complete account of the formalism, along with general symmetry
considerations.
In conclusion, we have demonstrated experimentally and theoreti-
cally that the chiral non-collinear antiferromagnet Mn3Sn has non-trivial
T -symmetry-breaking counterparts of the conventional SHE and
ISHE. Although the unusual properties of Mn3Sn make it ideal for
observing these effects, they should be present in any single-crystal
magnetic system and should be detectable using measurement geom-
etries determined by the relevant symmetry considerations. This new
type of SHE that is controllable by a magnetic field adds to the arsenal
of spintronics and suggests new application possibilities27.
Online content
Any methods, additional references, Nature Research reporting summaries, source
data, statements of data availability and associated accession codes are available at
https://doi.org/10.1038/s41586-018-0853-0.
Received: 25 November 2017; Accepted: 25 October 2018;
Published online 16 January 2019.
1. Dyakonov, M. I. & Perel, V. I. Possibility of orientating electron spins with current.
Sov. J. Exp. Theor. Phys. Lett. 13, 467–469 (1971).
2. Hirsch, J. E. Spin Hall effect. Phys. Rev. Lett. 83, 1834–1837 (1999).
3. Kato, Y., Myers, R. C., Gossard, A. C. & Awschalom, D. D. Observation of the spin
Hall effect in semiconductors. Science 306, 1910–1913 (2004).
4. Wunderlich, J., Kaestner, B., Sinova, J. & Jungwirth, T. Experimental observation
of the spin-Hall effect in a two-dimensional spin–orbit coupled semiconductor
system. Phys. Rev. Lett. 94, 047204 (2005).
5. Sinova, J. et al. Spin Hall effects. Rev. Mod. Phys. 87, 1213–1260 (2015).
6. Saitoh, E. et al. Conversion of spin current into charge current at room
temperature: inverse spin-Hall effect. Appl. Phys. Lett. 88, 182509
(2006).
7. Kimura, T., Otani, Y., Sato, T., Takahashi, S. & Maekawa, S. Room-temperature
reversible spin Hall effect. Phys. Rev. Lett. 98, 156601 (2007).
8. Liu, L. et al. Spin-torque switching with the giant spin Hall effect of tantalum.
Science 336, 555–558 (2012).
9. Jungwirth, T. et al. Spin Hall effect devices. Nat. Mater. 11, 382–390 (2012).
6 3 0 | N A T U RE | V O L 5 6 5 | 3 1 J A N U A R Y 2 0 1 9
© 2019 Springer Nature Limited. All rights reserved.
δsod along field direction (×10–3)
3
2
1
0
–1
–2
–3
–90 –60 –30 0 30 60 90
T (°)
the spin density induced by the current (along x) for the toy model. The
component of δ⟨s⟩od parallel to the external magnetic field is plotted
because the spin detector NiFe in Fig. 1 is magnetized along the external
field and thus detects spin accumulation parallel to the field direction.
10. Nakatsuji, S. et al. Large anomalous Hall effect in a non-collinear
antiferromagnet at room temperature. Nature 527, 212–215 (2015).
11. Freimuth, F. et al. Spin-orbit torques in Co/Pt(111) and Mn/W(001) magnetic
bilayers from first principles. Phys. Rev. B 90, 174423 (2014).
12. Hals, K. M. D. & Brataas, A. Spin-motive forces and current-induced torques in
ferromagnets. Phys. Rev. B 91, 214401 (2015).
13. Freimuth, F., Blügel, S. & Mokrousov, Y. Direct and inverse spin-orbit torques.
Phys. Rev. B 92, 064415 (2015).
14. Chen, H., Niu, Q. & MacDonald, A. H. Anomalous Hall effect arising from
noncollinear antiferromagnetism. Phys. Rev. Lett. 112, 017205 (2014).
15. Suzuki, M.-T., Koretsune, T., Ochi, M. & Arita, R. Cluster multipole theory for
anomalous Hall effect in antiferromagnets. Phys. Rev. B 95, 094406 (2017).
16. Chien, C. L. & Westgate, C. R. The Hall Effect and its Applications (Plenum,
1980).
17. Nagaosa, N., Sinova, J., Onoda, S., MacDonald, A. H. & Ong, N. P. Anomalous Hall
effect. Rev. Mod. Phys. 82, 1539–1592 (2010).
18. Tian, Y., Ye, L. & Jin, X. Proper scaling of the anomalous Hall effect. Phys. Rev.
Lett. 103, 087206 (2009).
19. Morota, M. et al. Indication of intrinsic spin Hall effect in 4d and 5d transition
metals. Phys. Rev. B 83, 174405 (2011).
20. Tserkovnyak, Y., Brataas, A. & Bauer, G. E. W. Enhanced Gilbert damping in thin
ferromagnetic films. Phys. Rev. Lett. 88, 117601 (2002).
21. Harder, M., Cao, Z. X., Gui, Y. S., Fan, X. L. & Hu, C.-M. Analysis of the line shape
of electrically detected ferromagnetic resonance. Phys. Rev. B 84, 054423
(2011).
22. Vasko, F. T. & Raichev, O. E. Quantum Kinetic Theory and Applications (Springer,
New York, 2005).
23. Culcer, D., Sekine, A. & MacDonald, A. H. Inter-band coherence response to
electric fields in crystals. Phys. Rev. B 96, 035106 (2017).
24. Aronov, A. G. & Lyanda-Geller, Y. B. Nuclear electric resonance and orientation of
carrier spins by an electric field. JETP Lett. 50, 431–434 (1989).
25. Edelstein, V. M. Spin polarization of conduction electrons induced by electric
current in two-dimensional asymmetric electron systems. Solid State Commun.
73, 233–235 (1990).
26. Garate, I. & MacDonald, A. H. Influence of a transport current on magnetic
anisotropy in gyrotropic ferromagnets. Phys. Rev. B 80, 134403 (2009).
27. Železný, J., Zhang, Y., Felser, C. & Yan, B. Spin-polarized current in noncollinear
antiferromagnets. Phys. Rev. Lett. 119, 187204 (2017).
Acknowledgements This work is partially supported by CREST (grant numbers
JPMJCR15Q5 and JPMJCR18T3), the Japan Science and Technology Agency,
Grants-in-Aid for Scientific Research (16H02209, 25707030), Grants-in-Aids
for Scientific Research on Innovative Areas (15H05882, 15H05883, 26103001,
26103002) and the Program for Advancing Strategic International Networks to
Accelerate the Circulation of Talented Researchers (R2604) from the Japanese
Society for the Promotion of Science. H.C. and A.H.M. were supported by
SHINES, an Energy Frontier Research Center funded by the US Department of
Energy, Office of Science, Basic Energy Sciences, under award SC0012670.
Author contributions Y. Otani and S.N. planned the experimental project, and
M.K., Y. Omori, M.I., T.T., S.S., P.K.M. and K.K. fabricated the devices. M.K., S.S. and
K.K. performed the experiments and collected data. H.C. and A.H.M. performed
the theoretical analyses. M.K., H.C., K.K., S.N., A.H.M. and Y. Otani wrote the
manuscript, and M.K., K.K. and H.C. prepared the Methods section and the
figures. All authors discussed the results and commented on the manuscript.
Competing interests The authors declare no competing interests.
Additional information
Extended data is available for this paper at https://doi.org/10.1038/s41586-
018-0853-0.
Reprints and permissions information is available at http://www.nature.com/
reprints.
Correspondence and requests for materials should be addressed to Y. Otani.
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.

Methods
Device fabrication. The single crystal used in the present study was prepared by
using the Bridgman technique. The detailed sample preparation method can be
found in ref. 28. A thin flake was fabricated from a single crystal of Mn3Sn using a
focused ion-beam system. The final processing of Ga+ ion milling was performed
by a low acceleration voltage of 5 kV to reduce the damaged layer at the surface. The
polished thin single crystal flake was then transferred to a thermally oxidized Si
substrate by using a W needle manipulator. Then, the surface of Mn3Sn was milled
by Ar dry etching to remove the residual damaged layer. The MgO/NiFe and Cu
electrodes were formed by electron beam lithography and lift-off processes. The
edge of the Mn3Sn flake was covered by a 30-nm-thick Al2O3 layer to avoid direct
contact between the sample edge and the electrodes. The resistivity of the processed
Mn3Sn crystal was 360 μΩ cm at room temperature, which is comparable to the
bulk value. In addition, the metallic behaviour of the temperature dependence of
the resistivity was confirmed. The MgO insertion layer was prepared by a method
similar to that used in a previous study29. Thus, the interfacial resistance is expected
to be about 0.1 Ω μm2.
Measurement method. The spin accumulation measurement was performed by
the standard lock-in technique with an a.c. current of 137 Hz. For the spin-pumping
experiment, the NiFe/Mn3Sn bilayer was put between the signal and the ground
lines of the coplanar waveguide. As a result, a perpendicular a.c. magnetic field was
applied to the sample. The operation frequency was 7–20 GHz, and the maximum
input power was 26 dB m. The d.c. voltage signal was measured by a nanovoltmeter.
All measurements were performed at room temperature.
Switching field of microfabricated Mn3Sn crystal. The switching field of the
microfabricated Mn3Sn crystal was determined from the AHE resistance measured
in the same configuration as that of the spin-accumulation experiment. In this
configuration, no AHE voltage is expected if the crystallographic kagome plane
is completely parallel to the plane of the voltage probes, because the Mn magnetic
moments lie in the kagome plane. However, a small residual AHE voltage is still
observed owing to a small misalignment between the kagome plane and the voltage
probe plane. Extended Data Fig. 1a shows the AHE voltage, measured in the same
configuration as that of the spin-accumulation measurement. The sharp resistance
jumps observed around ±2,000 Oe correspond to the reversal of the sublattice
magnetization of Mn3Sn. This switching field is higher than that observed in bulk
samples. We confirmed the increase of the switching field in several microfabri-
cated crystals of Mn3Sn30, which might originate from the residual strain produced
during microfabrication. Note that this sublattice switching field of Mn3Sn is higher
than that of NiFe long rectangular pad (about 500 Oe)19.
In the device used in the spin-pumping experiment, we measured the rectified
d.c. voltage under a.c. excitation. As shown in the Extended Data Fig. 1b, a voltage
jump of the d.c. voltage is observed around ±1,200 Oe, which corresponds to the
magnetization reversal of Mn3Sn. Although the mechanism of this d.c. voltage
change is not clear at present, this signal might originate from the rectified AHE
voltage. The mechanism of this change is now under detailed investigation.
Enhancement of symmetric voltage signal by spin pumping in NiFe/Mn3Sn
bilayers. Extended Data Fig. 2 shows a comparison of the d.c. voltage signal in
NiFe/Mn3Sn and a NiFe single layer under FMR excitation. Here we normalize the
voltage signals with the asymmetric voltage amplitude, which arises purely from
the rectified anisotropic magnetoresistance (AMR) of both samples. The important
feature of these two figures is that the symmetric voltage contribution in the NiFe/
Mn3Sn bilayer is greatly enhanced compared with that in the NiFe single layer.
Thus, a symmetric voltage contribution (in addition to the rectified AMR and/or
the Seebeck effect due to microwave heating31) exists in the NiFe/Mn3Sn sample.
Also, the dominant contribution of the symmetric voltage does not change its sign
with magnetization reversal. This behaviour is totally different from that observed
in the conventional ISHE, anomalous Nernst effect and spin Seebeck effect due
to the out-of-plane heat flow during FMR excitation, but is consistent with the
expected behaviour of the MISHE.
Frequency and angle dependence of spin-pumping experiment. Extended Data
Fig. 3a shows the frequency dependence of the ratio between symmetric and asym-
metric voltage contributions (Vsym/Vanti) for f = 10–13 GHz. As can be seen in the
figure, the value of Vsym/Vanti is almost constant with frequency. Extended Data
Fig. 3b shows the angular dependence of the symmetric voltage contribution at
several microwave frequencies. The observed angle dependences in the whole
frequency range are similar, with almost the same amplitude, and are well explained
by a dependence on sin(2θ) and sin(4θ), as expected from the model calculation.
These results also support that the observed symmetric voltage contribution does
not arise from the rectified AMR but from the MISHE, because the amplitude of
the symmetric voltage contribution largely changes with microwave frequencies
in the case of the AMR rectification mechanism21. It should be also noted that the
anomalous Nernst effect voltage and the spin Seebeck effect voltage are expected
to have a cosθ dependence, which is completely different from the experimental
observations.
© 2019 Springer Nature Limited. All rights reserved.
Letter RESEARCH
Symmetry analysis of current-induced spin accumulation in the SHE and
MSHE. By considering the current-induced spin accumulation at a material’s
boundary to be a local spin-density response to an external electric field, we can
use a symmetry-based approach to understand this behaviour qualitatively. Below
we briefly review this method. More details can be found in ref. 32.
We write a general tensor quantity in three dimension as aij…r, where each
subscript runs through x, y, z. The tensor transformation under a proper rotation
in SO(3) represented by the 3 × 3 rotation matrix tij is
amn ... z = tmitnj . . . tzraij... r (4)
where summation over repeated indices is assumed. For an improper rotation,
which has determinant |t| = −1, we have to distinguish between polar and axial
tensors. The former always transform like equation (4), whereas the latter have to
be multiplied by an additional factor, the determinant of t:
amn ... z = ∣t∣ tmitnj . . . tzraij... r (5)
Any linear response function L connecting a thermodynamic force f and a flux
j of certain thermodynamic variables q can be written as a tensor (here we assume
f and j to be vectors for simplicity):
jα (r , t) = Lαβ (r)fβ (r , t) (6)
where Lαβ is determined by the equilibrium properties of the system in the
absence of perturbations, and α, β, … and m, n, … are Cartesian coordinates.
The symmetry constraints on L are summarized by Neumann’s principle, which
states that any type of symmetry exhibited by the point group of the crystal is
possessed by every physical property of the crystal. Therefore, for a given tensor
quantity that corresponds to a response function of the system, its components
must satisfy equation (4) or (5) for any symmetry operation in the point group
of the unperturbed system. In systems with high symmetry this requirement can
greatly reduce the number of components of a response function tensor that is
allowed to be non-zero.
The components of the response function L must also satisfy the Onsager
relation:
Lαβ (M , B) = εαεβLβα(−M , −B) (7)
where M and B are the magnetization and the magnetic field, respectively, and εα
and εβ are +1 (−1) if the thermodynamic variables qα and qβ are even (odd) under
time reversal. This general form of the Onsager relation is needed when the ther-
modynamic force and the flux do not belong to the same variable, for example, in
the case of magnetoelectric effects.
To study the behaviour of the response functions under rotation of the magneti-
zation alone, it is usually helpful to expand them in a power series of the directional
cosines of the magnetization vector θα = Mα/|M|:
Lαβ = aαβ + aαβγ θγ + aαβγδ θγθδ +  (8)
Neumann’s principle can then be used in combination with Onsager’s relation
to analyse the tensor coefficients at each order of this expansion.
We now consider in general the phenomenon of electric-field-induced spin
density, which includes the special case of spin accumulation due to the SHE,
described by the linear response relation:
δsi = χijEj (9)
Because spin is an axial vector whereas the electric field is a polar vector, χij is
an axial tensor and, according to Neumann’s principle, it must vanish if the system
has inversion symmetry. This is the reason why spin accumulation due to the SHE
appears on the surface only when the bulk of the system has inversion symmetry.
Moreover, because the spin changes sign under time reversal whereas the electric
field is even (that is, the spin does not change sign under time reversal), the exist-
ence of χij requires time-reversal symmetry breaking due to either magnetism or
dissipative processes.
In the ordinary SHE in non-magnetic systems, the induced spin accumu-
lation requires dissipative processes. We consider the surface of an isotropic,
inversion-symmetric bulk material, which is a good approximation of the usual
polycrystalline non-magnetic conductors used for demonstrating the SHE. The
surface has continuous rotation symmetry with respect to its normal, taken to
be z, as well as vertical mirror planes. Assuming that the current is applied along
the x direction, the vertical mirror perpendicular to y will leave it unchanged
because the current or the electric field is a polar vector. For equation (9) to hold,
δs must be along y, because a mirror reflection reverses spin components parallel
to the mirror but leaves the component perpendicular to it unchanged. The orthog-
onality between the current direction, the surface normal and the polarization of
 RESEARCH Letter

accumulated spins known from the microscopic Dyakonov–Perel theory of SHE- em is the sublattice-dependent
between two layers; J is the on-site exchange splitting; ^
induced spin accumulation is thus a symmetry requirement. direction of the on-site exchange field depicted in Extended Data Fig. 4; λR is the
We now argue that in magnetic systems there can be new contributions to δs strength of the Rashba-type spin–orbit coupling due to inversion-symmetry break-
that are odd under magnetization reversal. Consider, for example, the same system ing; and d^im, jn is a unit vector pointing from site im to site jn. The Hamiltonian is
as above but with the bulk material having ferromagnetic magnetization along x. a 12 × 12 matrix after Fourier transformation to momentum space.
This in-plane magnetization removes a lot of symmetries present in the non- We can then calculate δ⟨s⟩d and δ⟨s⟩od using equation (3) or equation (11). The
magnetic system, thus allowing more components of χij to be non-zero. Among results for the inter-band contribution, which is odd under magnetization reversal,
them, we are especially interested in possible components that are odd under mag- are plotted in Extended Data Fig. 5. For this current direction, the in-plane spin-
netization reversal. First, the system is invariant under a π rotation around z (C 2z) accumulation vector is perpendicular to the weak magnetization direction when
plus a time reversal operation (T ). Thus the part of χij that is odd under magnet- the rotation angle is 90°. This can be understood by analysing the symmetry of the
ization reversal should be invariant under this combined C 2z T operation. However, model, which we detail below.
C 2z T changes the sign of the electric field along x. Therefore, the left-hand side of The toy model has the same symmetry as the (0001) surface of Mn3Sn described
equation (9) must also change sign under this operation. This is not possible for by the point group C3v, which includes a three-fold rotation axis C3 and three
any spin components in the x–y plane, but a finite δsz is allowed. Thus for a poly- vertical mirror planes parallel to it. The group elements are:
crystalline ferromagnet with magnetization parallel to an in-plane electric field or
current, symmetry allows a spin-accumulation vector perpendicular to the surface 1, 3(2¯⊥), ± 3z (13)
that is odd under magnetization reversal, in addition to the conventional SHE- which can be generated by two rotation matrices:
induced in-plane spin accumulation, which is even under magnetization reversal.
Next, we introduce a quantum kinetic formalism that can be used to calculate  − 1/2 3 /2 0
− 1 0 0   
the response function χij for a given system. We start from equation (2). In the HB 
2¯ x =  0 1 0, 3z = − 3 /2 − 1/2 0
 (14)
eigenstate representation, the diagonal (intra-band) and off-diagonal (inter-band)  0 0 1   
elements of ρ(1) can be determined from the equations:  0 0 1 
By expanding χ to the first order of the directional cosines of the weak mag-
∂ρn(0) (1)
ρnn
· eE + =0 netization direction and applying equation (4) in combination with equation (14)
ħ∂k τ to the linear-order coefficient aαβγ we found that only the following components
(1)
(10)
i (1) (∂ H ) ρ are non-zero
(εm − εn)ρmn + (ρm(0) − ρn(0)) k B mn · eE + mn = 0
ħ ħ(εm − εn) τ ayyy = − axxy = − ayxx = − axyx ayyz = azyy = ayzy = axxz = azxx = axzx azzz (15)
(In the good metal limit, the τ dependence in the second equation in equation If we focus on the places in Extended Data Fig. 5 where δ⟨s⟩od vanishes, the three
(10)  can be ignored.) Solving equation (10) and substituting ρ (1) into panels suggest axxx = ayxy = azxy = 0, in agreement with equation (15). Specifically,
δ⟨s⟩ = tr[ρ(1)(ħ/2)σ] yields equation (3) or, in terms of χ = δ⟨s⟩/E , which is the vanishing of δ⟨sy⟩od when the weak magnetization is also along y can be under-
defined in equation (9), stood as follows: the magnetic order in Extended Data Fig. 4 when all the moments
are rotated by π/2 has a mirror-reflection symmetry M with respect to a plane
eħ ∂fn parallel to y and going through site 3, plus time reversal T . Thus χodαβ must be even
χdαβ = − τ∑ β
∫ ∂εn vnnσnndk
α
2 n under this operation. In addition, this operation will change the sign of the electric
(11) field along x. Therefore the symmetry-allowed components of δ⟨s⟩od must change
eħ  v β σα 
χodαβ = −
2
∑ ∫ (fm −fn )Im  (ε mn−εnm) 2  dk sign under this operation, which means that only δ⟨sx⟩od can be non-zero, in agree-
m≠n m 
n  ment with Extended Data Fig. 5.
Finally, we note that non-zero χod from equation (11) relies on spin-dependent
where v is the velocity operator. Because of the factor τ in the expression for inter-band coherence that generally comes from spin–orbit coupling. However, it
χdαβ, which accounts for the required change of sign under time reversal, χdαβ has is known that non-collinear magnetic ordering, which can be viewed as a position-
to be even under magnetization reversal. By contrast, χodαβ must be odd under dependent SU(2) field applied on itinerant electrons, can lead to similar effects as
magnetization reversal. This quantity corresponds to the MSHE defined in the spin–orbit coupling. Indeed, we found that even if we let λR = 0 in the model of
main text. We note that in the relaxation-time approximation, we have ignored equation (12), χod is still non-zero if J ≠ 0. We thus expect that the large MSHE
possible contributions from skew scattering, which lead to additional Fermi surface observed in Mn3Sn is due to the effective spin–orbit coupling in its Bloch bands
contributions to χ that are of higher order in the impurity potential and can be odd provided by the non-collinear magnetic ordering, rather than the very small atomic
under magnetization reversal. As mentioned in the main text, the present approach spin–orbit coupling of Mn.
is justified for clean systems with strong spin–orbit coupling in the Bloch bands. Quantum kinetic theory for the ISHE and the MISHE. We turn next to the
We may consider that spin accumulation in the kinetic theory is similar to the ISHE and the MISHE. Inspired by the treatment of the current-induced spin
Edelstein effect in two-dimensional inversion-symmetry-breaking systems. Indeed, accumulation above, we describe the ISHE in terms of the non-equilibrium elec-
χdαβ corresponds to the original Edelstein effect, as can be verified through explicit tric current driven by a time-dependent exchange field. For this case, the kinetic
calculations using a Rashba Hamiltonian. Similarly, we can use a Rashba model equation is:
with an in-plane magnetization along x to calculate χod, which requires time-
reversal-symmetry breaking in the Bloch bands, and verify that for an electric field i ∂ρ(0) ˙ ρ(1)
along x only the z component of the induced spin density is non-zero33. [H , ρ(1)] + Δ+ =0 (16)
ħ ∂Δ τ
Toy model calculation of current-induced spin accumulation in Mn3Sn. In this
section we use the quantum kinetic approach explained above to calculate the spin where Δ is a time-dependent exchange field representing the effect of the precess-
accumulation in a toy model that has the same symmetries as the NiFe/Mn3Sn ing magnetization. The non-equilibrium electric current calculated using
bilayer system with the interface normal to the (0001) axis of Mn3Sn. The model ⟨j⟩ = tr[ρ(1)ev ] can be decomposed into diagonal and off-diagonal contributions:
consists of two kagome layers, AB-stacked, with on-site exchange fields forming
the pattern of the Mn moments in Mn3Sn, a Rashba-type spin–orbit coupling that β ∂fn
⟨j α ⟩d = eτΔ˙ ∑ β α
∫ ∂εn σnnvnndk
breaks the in-plane mirror symmetry, and an on-site potential difference between n
the two layers. The Hamiltonian is written as: (17)
β  σβ vα 
† †
⟨j α ⟩od = eΔ˙ ∑∫ (fm − fn )Im mn nm 2 dk
H =t ∑ cim αcjnα + Λ ∑ cim αcimα m≠n  (εm − εn) 
⟨im, jn⟩α iα, m < 4

+J ∑ †
em · σαβcim
^ αcimβ
The appearance of the same integrands in the above equations as in equation
(12) (11) is related to Onsager reciprocity. However, it is crucial to realize that whereas
imαβ
† equation (11) describes a d.c. response to a static electric field, the response func-
+ λR ∑ i(z^ × d^im, jn) · σαβcim αcjnβ
tions in equation (17) are implicitly time-dependent because of the precessing
⟨im, jn⟩αβ
exchange field Δ and can only give rise to a d.c. current or voltage after averaging
where i, j label unit cells; m, n label sublattices; α, β label spin-up or spin-down, over the precession period. This can be made more explicit by expanding jα in a
respectively; t is the spin-independent hopping; Λ is the on-site potential difference power series in the small dynamic component of the exchange field δΔ:

© 2019 Springer Nature Limited. All rights reserved.


γ to be non-zero. To give a concrete example, we use the same bilayer kagome model
⟨j α ⟩ = (ρ0αγ + ρ1αβγδΔβ + )Δ˙ (18)
By considering that the integral of any periodic function’s derivative over a
period is zero, we can conclude that in the small cone-angle limit the d.c. part of
⟨j α ⟩ comes from the components of ρ1αβγ that are antisymmetric in β and γ. To be
specific, taking the dynamic components δΔ to be δΔz and δΔx (for precession
around y), we can have only the following quadratic terms of δΔ and Δ̇:
x z z x z x
δΔx Δ˙ , δΔz Δ˙ , δΔx Δ˙ + δΔz Δ˙ , δΔx Δ˙ − δΔz Δ˙ (19)
However, only the last term cannot be written as a total derivative. Thus only
the part of ρ1αβγ that is antisymmetric in β and γ is relevant for the d.c. current.
We note that after performing this expansion, ρ1αβγ is time-independent andγ
determined only by the ground state of the system. Moreover, because both δΔβ Δ˙
and the current are odd under time reversal, the interband contribution to ρ1αβγ
does not require time-reversal-symmetry breaking in the ground state and should
not change sign under magnetization reversal. However, the antisymmetric com-
γ mentioned earlier. Because δΔβ Δ˙ is even, τ is odd and τ−1ρ1,d
bination εαβγ δΔβ Δ˙ is odd under magnetization reversal, which can be seen from
the Landau–Lifshitz equation. Therefore, the d.c. current given by the interband
contribution in equation (17) changes sign under magnetization reversal and cor-
responds to the ordinary ISHE. To see this more clearly, we consider the usual
bilayer system formed by a layer of a polycrystalline ferromagnet and a layer of
polycrystalline non-magnetic metal, with the interface normal to z. As argued
above, the interband contribution to ρ1αβγ does not require time-reversal-symmetry
breaking in the ground state and should be determined in the lowest order by the
αβγ
symmetry of the non-magnetic system. We thus study the constraint of ρ1,od
enforced by the continuous rotation symmetry around z, described by the rotation
matrix:
 cosθ − sinθ 0
 
t =  sinθ cosθ 0 (20)
  to the magnetic-field direction. However, it is crucial to realize that the external
0 0 1 
αβγ
For equation (4) to hold for the rank-3 polar tensor ρ1,od , the summation on
the right-hand side of equation (4) must only have terms proportional to
cos2θ + sin2θ or 1. We can eliminate more components of ρ1,odαβγ
by considering in
addition vertical mirrors parallel to z. We found that only the following compo-
nents are allowed to be non-zero and can be antisymmetric in βγ:
ρxxz = ρyyz , ρxzx = ρyzy (21)
which means that a d.c. current along x has to be from magnetization precessing
in the z–x plane, or that a d.c. current along y has to come from magnetization
precessing in the y–z plane. These conclusions are consistent with the description
of conventional spin pumping plus ISHE.
We now argue that the MISHE, namely, the FMR voltage signal in the Mn3Sn–
NiFe system that is even under magnetization reversal, comes from the intra-band
(equation (17)) contribution to ρ1αβγ. This is not contradictory to our former con-
clusion that the integrand in the first equation in equation (11) (which is identical
to that in the first equation in equation (17)) is even under time reversal. In fact, it
is precisely for this reason that the MISHE requires time-reversal-symmetry break-
αβγ
ing in the ground state of the system. More explicitly, ρ1,d /τ must be odd under
magnetization reversal. We can thus expand this rank-3 polar tensor in odd pow-
ers of the directional cosines of the equilibrium magnetization Δλ:
τ−1ρ1,d
αβγ
= aαβγλΔλ + aαβγλμν ΔλΔμ Δν +  (22)
The leading-order contribution is described by a rank-4 axial tensor aαβγλ. For
the geometry used in ordinary spin-pumping plus ISHE experiments, where the
static magnetic field is along y and magnetization precesses in the z–x plane, a d.c.
current along x requires non-zero axxzy − axzxy. However, if the system has contin-
uous rotation symmetry around z and vertical mirrors, components such as aαzγλ
and aαβzλ, where the unspecified indices are either x or y, must be identically zero.
Therefore, in conventional spin-pumping plus ISHE experiments using polycrys-
talline films, the d.c. signal that is even under magnetization reversal is absent.
However, the NiFe–Mn3Sn system has a lower symmetry determined by the sin-
gle crystalline Mn3Sn surface, which may allow the desired component axxzy − axzxy
© 2019 Springer Nature Limited. All rights reserved.
Letter RESEARCH
as in equation (12) but with an extra term accounting for the precessing exchange
field from the NiFe layer:
†
HNiFe = − ∑ Δ(t) · σαβcim αcimβ
imαβ (23)
Δ(t) = |Δ|[sinθsinφ(t), cosθ , sinθ cosφ(t)]
Physically, we expect that the electronic states of Mn3Sn are coupled to the
precessing exchange field from NiFe at the interface, whereas the non-collinear
moments of Mn3Sn are static because the spin-wave gap of the latter is much larger
than the driving frequency.
We can carry out the complete symmetry analysis based on the C3v group for
the leading-order contribution in equation (22) described by the rank-4 axial ten-
sor. But here we just point out that if the static magnetic field is along y, the d.c.
current can only be along y. This isγ because of the same MT  symmetry that we
αβγ
is even under
MT , the current must be odd under MT , which is only possible if the in-plane
current is along y. This is indeed verified by explicitly calculating ⟨j α ⟩d using equa-
tion (17) for the toy model described above, as shown in Extended Data Fig. 6. We
note that the d.c. current is proportional to the integrals of the curves in the figure
over a precession period. We also checked that upon magnetization reversal, the
time average of ⟨j α ⟩d does not change sign.
Finally we mention that similarly to the case of MSHE in Mn3Sn, the spin–orbit
coupling needed by MISHE can also be provided by the non-collinear magnetic
order of Mn3Sn only. An explicit example using the toy model with λR = 0 is shown
in Extended Data Fig. 7.
Angular dependences of MSHE/MISHE and interface symmetry breaking. To
directly compare with the angular dependence of the MSHE signal measured
experimentally (Fig. 2c), we need to project the δ⟨s⟩ from our model calculation
magnetic field couples to the total (weak) magnetization. As all sublattice
moments are rotated in the anticlockwise direction, this weak magnetization
rotates in the opposite, clockwise, direction and vice versa. This unique property
of non-collinear antiferromagnets should be properly taken into account when
calculating the angular dependence of MSHE and MISHE data using our model.
Figure 4 was obtained by including this effect. However, from the symmetry point
of view, the projection of δ⟨s⟩od onto the magnetic-field direction should always
have even-order Fourier components only, because δ⟨s⟩od ⋅ B^ is even under the
reversal of B^. Moreover, the fact that δ⟨s⟩od ⋅ B^ vanishes when B^ is along [0110] in
our model is due to the MT symmetry on the (0001) surface of ideal Mn3Sn. The
nontrivial angular dependence of δ⟨s⟩od ⋅ B^ reflects the crystallinity of the interface
structure. Similarly we have calculated the angular dependence of the MISHE
signal, that is, τ−1ρ1xβz (antisymmetric in βz) according to the experimental geom-
etry, using our model. The result is shown in Extended Data Fig. 8, which also has
the expected even-order Fourier components because the MISHE is even under
field reversal and vanishes at the angles where the MISHE is forbidden by symme-
try. It should be noted that because the MISHE and MSHE are not reciprocals of
each other, they do not necessarily have the same angular dependence.
Data availability
The data that support the findings of this study are available from the correspond-
ing author upon reasonable request.
28. Ikhlas, M. et al. Large anomalous Nernst effect at room temperature in a chiral
antiferromagnet. Nat. Phys. 13, 1085–1090 (2017).
29. Fukuma, Y. et al. Giant enhancement of spin accumulation and long-distance
spin precession in metallic lateral spin valves. Nat. Mater. 10, 527–531 (2011).
30. Narita, H. et al. Anomalous Nernst effect in a microfabricated thermoelectric
element made of chiral antiferromagnet Mn3Sn. Appl. Phys. Lett. 111, 202404
(2017).
31. Shiomi, Y. et al. Spin–electricity conversion induced by spin injection into
topological insulators. Phys. Rev. Lett. 113, 196601 (2014).
32. Birss, R. R. Symmetry and Magnetism (North-Holland, Amsterdam, 1966).
33. Taniguchi, T., Grollier, J. & Stiles, M. D. Spin-transfer torques generated by the
anomalous Hall effect and anisotropic magnetoresistance. Phys. Rev. Appl. 3,
044001 (2015).
RESEARCH Letter

Extended Data Fig. 1 | Switching field of microfabricated Mn3Sn signle
crystals. a, External-magnetic-field dependence of AHE resistance in
the spin-accumulation device. The resistance jump around ±2,000 Oe
corresponds to the magnetization reversal of the microfabricated Mn3Sn
crystal. The observed AHE resistance is much smaller than that expected
in the bulk samples because the AHE signal in this device arises from the
slight tilting of the sample’s basal plane from the crystallographic kagome
plane. b, External-magnetic-field dependence of d.c. voltage under the
application of an a.c. current in the spin-pumping device. The sharp
voltage jumps around ±1,200 Oe correspond to the magnetization reversal
field Hc of Mn3Sn.

© 2019 Springer Nature Limited. All rights reserved.

 Letter RESEARCH

Extended Data Fig. 2 | Comparison of FMR-induced d.c. voltage signals respectively. The voltage signals are normalized by the asymmetric voltage
in NiFe/Mn3Sn bilayer and NiFe single layers. External-magnetic- amplitude. The symmetric voltage contribution in the NiFe/Mn3Sn bilayer
field dependence of d.c. voltage signal under FMR excitation. The green (left) is considerably enhanced compared with that of the Ni80Fe20 single
and blue lines are symmetric and asymmetric voltage contributions, layer (right); Py, permalloy.

© 2019 Springer Nature Limited. All rights reserved.

RESEARCH Letter

Extended Data Fig. 3 | Frequency and angular dependence of spin- f = 10–13 GHz. The value of Vsym/Vanti is independent of the microwave
pumping experiment. a, Frequency dependence of the ratio between frequency. b, Magnetic-field-angle dependence of symmetric voltage
symmetric (Vsym) and asymmetric (Vanti) voltage contributions for signals at several microwave frequencies.

© 2019 Springer Nature Limited. All rights reserved.

 Letter RESEARCH

Extended Data Fig. 4 | Toy model for MSHE and MISHE calculations. λR = 0.2t. The horizontal axis shows the wavevector k nomalized by the
Left, bilayer kagome toy model mimicking the (0001) surface of Mn3Sn. lattice constant a.
Right, band structure of the toy model with Λ = 0.1t, J = 1.5t and

© 2019 Springer Nature Limited. All rights reserved.

RESEARCH Letter

Extended Data Fig. 5 | Angular dependence of interband contribution Λ = 0.1t, J = 1.5t and λR = 0.2t. Only the first half of the rotation
of current-induced spin density. Interband contribution to the current (anticlockwise, from 0 to π) is shown. The second half is related to the first
(along x) induced spin density for the toy model in equation (19) with through time reversal.

© 2019 Springer Nature Limited. All rights reserved.

 Letter RESEARCH

Extended Data Fig. 6 | Calculated intraband contribution of time- magnetization for the model described by equations (12) and (23) with
dependent current induced by magnetization precession. Intraband Λ = 0.2t, J = 0, λR = 0.2t, θ = 0.1, |Δ| = 0.5t, EF = 2t and kBT = 0.1t
contribution to the time-dependent current induced by precessing (EF, Fermi energy; kB, Boltzmann constant).

© 2019 Springer Nature Limited. All rights reserved.

RESEARCH Letter

Extended Data Fig. 7 | Calculated intraband contribution of time- induced by precessing magnetization for the model described by equations
dependent current induced by magnetization precession without spin– (12) and (23) without spin–orbit coupling. The other parameters are
orbit interaction. Intraband contribution to the time-dependent current Λ = 0.2t, J = 0.6t, θ = 0.1, |Δ| = 0.5t, EF = 2t and kBT = 0.1t.

© 2019 Springer Nature Limited. All rights reserved.

 Letter RESEARCH

Extended Data Fig. 8 | Model calculation of angular dependence for J = 1.5t, λR = 0.2t, EF = 0 and kBT = 0.1t (for faster convergence of
MISHE. MISHE versus in-plane magnetic-field direction, calculated numerical integration over the Brillouin zone).
using the same geometry as the experiment. The parameters are Λ = 0.1t,

© 2019 Springer Nature Limited. All rights reserved.