Vol 455 | 9 October 2008 | doi:10.
1038/nature07321
LETTERS
Observation of the spin Seebeck effect
K. Uchida1, S. Takahashi2,3, K. Harii1, J. Ieda2,3, W. Koshibae4, K. Ando1, S. Maekawa2,3 & E. Saitoh1,5
The generation of electric voltage by placing a conductor in a
temperature gradient is called the Seebeck effect1,2. Its efficiency
is represented by the Seebeck coefficient, S, which is defined as the
ratio of the generated electric voltage to the temperature differ-
ence, and is determined by the scattering rate and the density of
the conduction electrons. The effect can be exploited, for example,
in thermal electric-power generators and for temperature sensing,
by connecting two conductors with different Seebeck coefficients,
a device called a thermocouple1,2. Here we report the observation
of the thermal generation of driving power, or voltage, for electron
spin: the spin Seebeck effect. Using a recently developed spin-
detection technique that involves the spin Hall effect3–14, we mea-
sure the spin voltage generated from a temperature gradient in a
metallic magnet. This thermally induced spin voltage persists even
at distances far from the sample ends, and spins can be extracted
from every position on the magnet simply by attaching a metal.
The spin Seebeck effect observed here is directly applicable to the
production of spin-voltage generators, which are crucial for driv-
ing spintronic15–18 devices. The spin Seebeck effect allows us to
pass a pure spin current19, a flow of electron spins without electric
currents, over a long distance. These innovative capabilities will
invigorate spintronics research.
The spin Seebeck effect refers to the generation of spin ‘voltage’ as a
result of a temperature gradient. We define spin voltage as the spin-
current potential, which is represented by m" 2 m#, where m" and m#
respectively denote the electrochemical potentials for spin-up and
spin-down electrons14,20; a gradient in the spin voltage drives a spin
current. In a metallic magnet, spin-up and spin-down conduction
electrons notably have different scattering rates and densities16, and
thus have different Seebeck coefficients, as if two conductors with
different S values were inherently present in one magnet (Fig. 1b).
When a metallic magnet is subjected to a temperature gradient,
therefore, it should generate different driving powers of electrons
in different spin channels along the temperature gradient21–24. This
is the proposed scenario for the spin Seebeck effect: in the spin sector,
a magnet works in the same way as a thermocouple (Fig. 1a, b). This
driving power of electrons generates differing amounts of flow in the
two spin channels, that is, a spin current. To be more specific, we
consider a magnet with uniform magnetization subject to a uniform
temperature gradient. In this case, and when the length of the magnet
along the temperature gradient is much greater than the spin-
diffusion length25 of the magnet, the above scenario and thermody-
namic arguments26 predict the spatial distribution of m" and m# along
the temperature gradient shown in Fig. 2a.
However, this spin Seebeck effect remains to be observed. The
recently discovered inverse-spin-Hall effect10–14 (ISHE) is a powerful
tool for detecting spin voltage, and we use it to investigate the spin
Seebeck effect in the present study. The ISHE converts a spin current
into an electromotive force ESHE by means of spin–orbit scattering. A
spin current carries a spin-polarization vector s along a spatial
1
Department of Applied Physics and Physico-Informatics, Keio University, Yokohama 223-8522, Japan. 2Institute for Materials Research, Tohoku University, Sendai 980-8577, Japan.
3
CREST, Japan Science and Technology Agency, Sanbancho, Tokyo 102-0075, Japan. 4Cross-Correlated Materials Research Group, RIKEN, Wako, Saitama 351-0198, Japan. 5PRESTO,
Japan Science and Technology Agency, Sanbancho, Tokyo 102-0075, Japan.
778
©2008 Macmillan Publishers Limited. All rights reserved
direction JS. The relation between ESHE, JS and s is given by the
following vector product:
ESHE ~DISHE JS |s ð1Þ
The ISHE efficiency DISHE is enhanced in noble metals, such as Pt. By
measuring ESHE, the ISHE can be used to detect a spin current10–14.
Figure 2b shows an illustration of the sample system used in the
present study. The sample consists of a 20-nm-thick, soft ferromag-
netic Ni81Fe19 film with a Pt wire attached to one end. The Ni81Fe19
layer was deposited on a sapphire substrate by electron-beam eva-
poration in a high vacuum, and the Pt layer was then sputtered in
an Ar atmosphere. Immediately before the sputtering, the surface of
the Ni81Fe19 layer was cleaned by Ar ion etching. The length, the width
and the thickness of the Pt wire are respectively LPt 5 4 mm, 100 mm,
and dPt 5 10 nm. We apply an in-plane external magnetic field, H,
along the x direction (Fig. 2d), except when collecting a set of angle-
dependent data (discussed below). The coercive force, HC, of the
Ni81Fe19 layer is around 15 Oe at 300 K, and the magnetization is
aligned along the external magnetic field direction when jHj . HC.
A temperature gradient =T is applied along the x direction by generat-
ing a temperature difference DT between the ends of the layer. Owing
to the direction of the temperature gradient (parallel or antiparallel to
a Thermocouple Metal A V
E
E T2
Metal B
∇T
T1
b Spin Seebeck effect
Metallic magnet m↑ – m↓
T2
∇T
T1
Figure 1 | The spin Seebeck effect. a, Illustration of a thermocouple. A
thermocouple consists of two conductors (metals A and B) connected to
each other. They have different Seebeck coefficients and, thus, the voltage V
between the output terminals is proportional to the temperature difference
T1 2 T2 between the ends of the couple. b, Illustration of the spin Seebeck
effect. In a metallic magnet, spin-up (") and spin-down (#) conduction
electrons have different Seebeck coefficients. When a temperature gradient is
applied, a spin voltage m" 2 m# proportional to the temperature difference
appears; a magnet functions just like a thermocouple, but in the spin sector.
NATURE | Vol 455 | 9 October 2008 LETTERS

H), Nernst effects26 are suppressed. The distance between the ends of
the Ni81Fe19 layer along the temperature gradient is L 5 6 mm, which
is much greater than the spin-diffusion length of Ni81Fe1925.
The mechanism of the measurement is as follows. In this set-up,
the spin voltage, m" 2 m#, which is thermally generated in the Ni81Fe19
layer, induces a spin current in the Pt layer through the Ni81Fe19–Pt
interface (Figs 2d and 3a). In the Pt layer, jm"2m#j decreases along the
z direction. Because the spin polarization, s, of this spin current lies
along the magnetization direction, when jHj . HC the spin current
High
a Ni81Fe19 b
∇T V Low
T + ∆T
H
induced in the Pt layer generates an electromotive force along the y
direction owing to the ISHE in the Pt layer (Fig. 2d and equation (1)).
Here we note that the sign of m" 2 m# reverses between the ends of the
Ni81Fe19 layer, as illustrated in Fig. 2a. Therefore, the direction of s in
the Pt layer and the sign of ESHE, the electromotive force induced by
the ISHE, are expected to reverse between the higher- and the lower-
temperature ends (Fig. 2d and equation (1)). This distinctive beha-
viour of the electromotive force is characteristic of the ISHE induced
a
Pt z
m↑ m↓ x
Ni81Fe19 y
T T + ∆T

m↑ m↓ ∇T
Pt T b c High d
High V Low
c V Low High High
V Low V Low
Ni81Fe19
H H H
6 6
T + ∆T ∇T ∇T ∇T
H 4 4
–L/2 0 L/2
∇T
x 2 2
T

V (µV)
d 0 0
+
Magnetization –2 –2
ESHE
–
–4 –4

–6 –6
– ESHE 0 6 12 18 0 6 12 18 0 6 12 18
∆T (K) ∆T (K)
+
σ T + ∆T e f g
z
Js
x H
y 21 K 21 K 21 K
∇T
T 18 K 18 K 18 K

15 K 15 K 15 K
V (µV)

Figure 2 | Experimental set-up. a, Illustration of the calculated distributions 12 K 12 K 12 K

of the electrochemical potentials m" and m# for spin-up and, respectively, spin-
down electrons in a metallic magnet. A temperature gradient =T is applied 9K 9K 9K
along the x direction. We write m" and m# respectively as m"5 mc" 2 ew and
6K 6K 6K
m# 5 mc# 2 ew, where e is the elementary charge, w is the electrostatic potential,
and mc" and mc# are the spin-dependent chemical potentials, which depend on
3K 3K 3K
the temperature T and on the densities n" and n#. Thus, the gradient of m" can
be expanded as =m" 5 (hmc"/hn")=n" 1 (hmc"/hT)=T 2 e=w (and that of m#
∆T = 0 K ∆T = 0 K ∆T = 0 K
similarly). The first term on the right-hand side reflects the particle-density
(n") modulation (accumulation term) and the second term the entropy
–100 0 100 –100 0 100 –100 0 100
contribution; hmc"/hT reflects the spin-subband entropy. Because, in H (Oe) H (Oe)
comparison with our sample size (L 5 6 mm), =n" and =n# decay within a
spin-diffusion length25 (,5 nm in NiFe) of the ends, the accumulation terms Figure 3 | Measurements of electromotive force. a, Illustration of the spin-
have a minor role and the difference between =m" and =m# is given by dependent electrochemical potentials, m" and m#, induced by a temperature
=(m" 2 m#) 5 eSS=T, where SS 5 (1/e)[hmc"/hT 2 hmc#/hT]. For a metallic gradient in the Ni81Fe19–Pt sample system. DT denotes the temperature
magnet, in general, hmc"/hT ? hmc#/hT. This implies that a temperature difference between the ends of the Ni81Fe19 layer. b, c, DT dependence of the
gradient induces a finite gradient of spin voltage, m" 2 m#. Because of the electric voltage difference V between the ends of the Pt wire in the
symmetry, m" 2 m# crosses zero at the centre of the sample. b, c, Illustrations of Ni81Fe19–Pt sample for H 5 100 Oe when the Pt wire is attached to the lower-
the measurement set-up in the present study. The sample consists of a temperature (300 K, b) and higher-temperature (300 K 1 DT, c) ends of the
Ni81Fe19 film with a Pt wire attached to one end. The surface of the Ni81Fe19 Ni81Fe19 layer. For H 5 100 Oe, the magnetization of the Ni81Fe19 layer is
film is a 24-mm2 rectangle. The chemical composition was confirmed using aligned along the magnetic field direction. The error bars represent the 95%
fluorescent X-ray microscopy. H denotes the external magnetic field vector confidence level (62s.d.). d, DT dependence of the electric voltage difference
(with magnitude H). To generate a temperature gradient in the film, the V between the sides of a plain Ni81Fe19 film (no Pt layer) for H 5 100 Oe, in
temperature difference between two Cu blocks connected to either end of the the direction perpendicular to that of the temperature-gradient. The
sample was controlled in a vacuum, using heaters and thermocouples attached measurement was performed at the lower-temperature end of the film. The
to the blocks. d, Illustration of the spin Seebeck effect induced by a errors are negligibly small. e, f, H dependence of V between the ends of the Pt
temperature gradient =T in a Ni81Fe19 film and the ISHE induced in Pt wires wire in the Ni81Fe19–Pt sample for various values of DT, measured when the
attached to the ends of film. Here Js and ESHE respectively denote the spatial Pt wire is attached to the lower-temperature (e) and higher-temperature
direction of the thermally induced spin current in the Ni81Fe19 film and the (f) ends of the Ni81Fe19 layer. g, H dependence of V between the sides of a
electromotive force generated by the ISHE in the Pt wire. The spin plain Ni81Fe19 film, in the direction perpendicular to that of temperature
polarization vector, s, of the spin current lies along the magnetization gradient. The measurement was performed at the lower-temperature end of
direction. the film. Scale bars in e, f and g, 10 mV.
779
©2008 Macmillan Publishers Limited. All rights reserved
LETTERS
by the thermally generated spin voltage. Hence, measuring this elec-
tromotive force allows us to detect the spin Seebeck effect with high
sensitivity. To do this, we measure the electric voltage difference, V,
between the ends of the Pt wire of the sample system at 300 K, as
illustrated in Fig. 2b, c.
Figure 3b and Fig. 3c show V at H 5 100 Oe (.HC) as a function of
DT when the Pt wire is on the lower- and, respectively, higher-
temperature ends of the Ni81Fe19 layer. Measurements were made
using identical samples. The magnitude of V is proportional to DT at
both ends of the layer. Figure 3d shows that the electric voltage signal
disappears in a plain Ni81Fe19 film to which no Pt wire is attached (see
also Fig. 3g). This signal also disappears in a plain Pt film. These
observations clearly demonstrate that the observed V signal shown
in Fig. 3b, c is not associated with the conventional Seebeck effect or
the Nernst effects26 in either an unconnected Ni81Fe19 film or a Pt
layer, but is only generated when the two layers are electrically
connected.
Significantly, Fig. 3b, c also shows that the sign of V for finite values
of DT clearly reverses between the lower- and higher-temperature
ends of the layer. This unconventional behaviour of V is the feature
predicted for the ISHE induced by the spin Seebeck effect described
above (Fig. 2d). This is therefore direct evidence for the operation of
the spin Seebeck effect in the Ni81Fe19 film.
To further support this interpretation, we measure the magnetic
field dependence of V in the same system. In Fig. 3e and Fig. 3f, we
show V as a function of H for various values of DT, measured at the
lower- and, respectively, higher-temperature ends of the layer. The
external magnetic field is applied along the x direction. Notably, in
the low-field range, 215 Oe , H , 15 Oe, hysteresis loops appear in
all the data. These hysteresis loops correspond to the magnetization
hysteresis that arises in the Ni81Fe19 layer in response to the variation
of H. This demonstrates that the V signal is affected by the magne-
tization, that is, the spin-polarization direction in the Ni81Fe19 layer.
When jHj . HC, the sign of V at each end of the layer can be reversed
by reversing H. This sign reversal is consistent with the prediction of
the ISHE induced by the spin Seebeck effect, through equation (1)
(we note that s is reversed by a magnetization reversal).
Figure 4a shows the voltage difference, V, measured with respect to
h, the angle the in-plane magnetic field makes with the x direction.
a Lower T Higher T
H x H x
q q
Ni81Fe19
Pt ∇T
1
V(q)/V(q = 0)
0 0
–1
0° 90° 180° 270°
q xP (mm)
Figure 4 | Dependence on magnetic field angle and position of Pt wire.
a, Dependence on magnetic field angle of the electric voltage difference V
between the ends of the Pt wire in the Ni81Fe19–Pt sample, for H 5 100 Oe
and DT 5 10 K, when the Pt wire is attached to the lower-temperature (blue
circles) and higher-temperature (red squares) ends of the Ni81Fe19 layer. An
780
©2008 Macmillan Publishers Limited. All rights reserved
NATURE | Vol 455 | 9 October 2008
We fix the temperature difference, DT, and the magnetic field
strength, H, at 10 K and 100 Oe. Because H . HC, the magnetization
in the Ni81Fe19 layer and s are aligned along the external magnetic
field direction. At both the lower- and the higher-temperature ends
of the layer, V varies with h in a sinusoidal pattern and vanishes when
h 5 90u, a situation consistent with the prediction of the ISHE
described in equation (1). The sign of V reverses between the lower-
and higher-temperature ends for each h value, a result which also
supports the aforementioned prediction of the ISHE induced by the
spin Seebeck effect (Fig. 2d).
The spin voltage induced at the higher-temperature end of the
Ni81Fe19 layer with temperature difference DT is (m"2m#) 5 eSSDT/
2, where SS is the spin Seebeck coefficient. This spin voltage injects
spins into the Pt layer and generates the electric voltage
V < hPtgNiFe–Pt(LPt/dPt)SSDT/2 owing to the ISHE, where hPt is the
spin Hall angle11–14 of Pt and gNiFe–Pt is the spin-injection efficiency14.
It is important to note that V is greatly amplified by the very large
aspect ratio, LPt/dPt < 4 3 105, of the Pt layer, despite the small tem-
perature gradient in the Ni81Fe19 layer. For V/DT 5 0.25 mV K21,
gNiFe–Pt < 0.2 and hPt 5 0.0037 (ref. 12), we estimate the spin
Seebeck coefficient for Ni81Fe19 to be SS 5 22 nV K21 at 300 K,
which is different from the conventional Seebeck coefficient of
220 mV K21 measured for the present Ni81Fe19 film. This implies
that the spin relaxation decreases spin accumulation in this large-
area film.
Up to now, we have discussed the spin voltage at the ends of the
Ni81Fe19 layer. It is also interesting to enquire whether the spin volt-
age obeys the calculated spatial variation illustrated in Fig. 2a far from
the ends. To investigate the spatial distribution of the spin voltage, we
measure V using Pt wires placed at different positions on the Ni81Fe19
films. In each case, the Pt wires are placed perpendicular to the x
direction (Fig. 4b). During the measurement, an external magnetic
field of 100 Oe (.HC) is applied along the x direction. In Fig. 4b, we
show V as a function of xP, the position of the Pt wire along
the temperature gradient from the centre of the Ni81Fe19 layer, for
various values of DT. With DT applied, V clearly increases for xP . 0
and decreases for xP , 0. For each value of DT, V varies almost
linearly with xP. The calculation shown in Fig. 2a reproduces these
experimental results very well, on the basis that V is proportional to
b
x
H H H
Pt
3
0
∇T ∇T
–3
∆T = 0 K
V (µV)
∆T = 10 K
∆T = 20 K
–5
360° –3 –2 –1 0 1 2 3
in-plane magnetic field is applied at an angle h to the x direction.
b, Dependence of V on xP, the position of the Pt wire, measured using four
separate samples and seven different values of xP on Ni81Fe19 films. In all the
samples, the Pt wires are placed perpendicular to the x direction and an
external magnetic field of 100 Oe (.HC) is applied along the x direction.
NATURE | Vol 455 | 9 October 2008
m"2m# at each position on the Ni81Fe19 layer. This experiment yields
the result that, when using the spin Seebeck effect, spins can be
extracted from every position on the Ni81Fe19 surface simply by
attaching a metal contact.
The spin Seebeck effect can be applied directly to the construction of
thermal spin generators for driving spintronic devices, thereby opening
the door to thermo-spintronics. The present study has also shown that,
in combination with the ISHE, the spin Seebeck effect allows us to
make effective thermoelectric generators and temperature-gradient
sensors, as the electric voltage generated can be amplified simply by
increasing the length, LPt, of the Pt wire. The spin Seebeck effect gen-
erates a pure spin current over a long distance in a magnet, without
electric currents. This is in striking contrast to any of the other known
methods for the injection of pure spin current10–14,27,28, where spin
current and spin voltage disappear over distances greater than the
spin-diffusion length25 (for example ,500 nm in Cu and ,5 nm in
Pt). This length scale severely limits experiments on pure spin currents.
The long-distance nature of the spin Seebeck effect observed here
immediately suggests a way of applying a variety of experimental tools
to spin currents. We anticipate that the spin Seebeck effect will trans-
form basic research on spin currents and lead to dramatic advances in
spintronic devices.
Received 6 May; accepted 4 August 2008.
1. Ashcroft, N. W. & Mermin, N. D. Solid State Physics 253–258 (Saunders College,
1976).
2. Maekawa, S. et al. Physics of Transition Metal Oxides 323–331 (Springer, 2004).
3. Dyakonov, M. I. & Perel, V. I. Current-induced spin orientation of electrons in
semiconductors. Phys. Lett. A 35, 459–460 (1971).
4. Hirsch, J. E. Spin Hall effect. Phys. Rev. Lett. 83, 1834–1837 (1999).
5. Takahashi, S. & Maekawa, S. Hall effect induced by a spin-polarized current in
superconductors. Phys. Rev. Lett. 88, 116601 (2002).
6. Murakami, S., Nagaosa, N. & Zhang, S.-C. Dissipationless quantum spin current at
room temperature. Science 301, 1348–1351 (2003).
7. Sinova, J. et al. Universal intrinsic spin Hall effect. Phys. Rev. Lett. 92, 126603
(2004).
8. Kato, Y. K., Myers, R. C., Gossard, A. C. & Awschalom, D. D. Observation of the
spin Hall effect in semiconductors. Science 306, 1910–1913 (2004).
9. 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).
10. Saitoh, E., Ueda, M., Miyajima, H. & Tatara, G. Conversion of spin current into
charge current at room temperature: Inverse spin-Hall effect. Appl. Phys. Lett. 88,
182509 (2006).
11. Valenzuela, S. O. & Tinkham, M. Direct electronic measurement of the spin Hall
effect. Nature 442, 176–179 (2006).
©2008 Macmillan Publishers Limited. All rights reserved
LETTERS
12. Kimura, T., Otani, Y., Sato, T., Takahashi, S. & Maekawa, S. Room-temperature
reversible spin Hall effect. Phys. Rev. Lett. 98, 156601 (2007).
13. Seki, T. et al. Giant spin Hall effect in perpendicularly spin-polarized FePt/Au
devices. Nature Mater. 7, 125–129 (2008).
14. Takahashi, S. & Maekawa, S. Spin current in metals and superconductors. J. Phys.
Soc. Jpn 77, 031009 (2008).
15. Wolf, S. A. et al. Spintronics: a spin-based electronics vision for the future. Science
294, 1488–1495 (2001).
16. Žutić, I., Fabian, J. & Das Sarma, S. Spintronics: fundamentals and applications.
Rev. Mod. Phys. 76, 323–410 (2004).
17. Maekawa, S. (ed.) Concepts in Spin Electronics (Oxford Univ. Press, 2006).
18. Chappert, C., Fert, A. & Van Dau, F. N. The emergence of spin electronics in data
storage. Nature Mater. 6, 813–823 (2007).
19. Slonczewski, J. C. Conductance and exchange coupling of two ferromagnets
separated by a tunneling barrier. Phys. Rev. B 39, 6995–7002 (1989).
20. Valet, T. & Fert, A. Theory of the perpendicular magnetoresistance in magnetic
multilayers. Phys. Rev. B 48, 7099–7113 (1993).
21. Cadeville, M. C. & Roussel, J. Thermoelectric power and electronic structure of
dilute alloys of nickel and cobalt with d transition elements. J. Phys. F 1, 686–710
(1971).
22. Gravier, L., Serrano-Guisan, S., Reuse, F. & Ansermet, J.-P. Thermodynamic
description of heat and spin transport in magnetic nanostructures. Phys. Rev. B 73,
024419 (2006).
23. Tsyplyatyev, O., Kashuba, O. & Fal’ko, V. I. Thermally excited spin current and
giant magnetothermopower in metals with embedded ferromagnetic
nanoclusters. Phys. Rev. B 74, 132403 (2006).
24. Hatami, M., Bauer, G. E. W., Zhang, Q.-F. & Kelly, P. J. Thermal spin-transfer
torque in magnetoelectronic devices. Phys. Rev. Lett. 99, 066603 (2007).
25. Bass, J. & Pratt, W. P. Jr. Spin-diffusion lengths in metals and alloys, and spin-
flipping at metal/metal interfaces: an experimentalist’s critical review. J. Phys.
Condens. Matter 19, 183201 (2007).
26. Callen, H. B. Thermodynamics Ch. 17 (Wiley, 1960).
27. Hirohata, A., Xu, Y. B., Guertler, C. M., Bland, J. A. C. & Holmes, S. N. Spin-
polarized electron transport in ferromagnet/semiconductor hybrid structures
induced by photon excitation. Phys. Rev. B 63, 104425 (2001).
28. Jedema, F. J., Filip, A. T. & van Wees, B. J. Electrical spin injection and
accumulation at room temperature in an all-metal mesoscopic spin valve. Nature
410, 345–348 (2001).
Acknowledgements The authors thank Y. Suzuki, S. E. Barnes, Y. Fujitani, G. Tatara,
K. M. Itoh, H. Kuwahara and M. Matoba for discussions. This work was supported
by a Grant-in-Aid for Scientific Research in Priority Area ‘Creation and control of
spin current’ (19048028) from MEXT, Japan, a Grant-in-Aid for Encouragement of
Young Scientists (A) from MEXT, Japan, the global COE for the ‘High-level global
cooperation for leading-edge platform on access spaces’ from MEXT, Japan, a
Strategic Information and Communications R&D Promotion Programme from MIC,
Japan, and the Next Generation Supercomputing Project of Nanoscience Program
from IMS, Japan.
Author Information Reprints and permissions information is available at
www.nature.com/reprints. Correspondence and requests for materials should be
addressed to E.S. (saitoheiji@appi.keio.ac.jp).
781