LETTER

Jean-Christophe Le Breton1, Sandeep Sharma1,2,3, Hidekazu Saito2, Shinji Yuasa2 & Ron Jansen2

doi:10.1038/nature10224

Thermal spin current from a ferromagnet to silicon by Seebeck spin tunnelling

Heat generation by electric current, which is ubiquitous in electronic devices and circuits, raises energy consumption and will become increasingly problematic in future generations of highdensity electronics. The control and re-use of heat are therefore important topics for existing and emerging technologies, including spintronics. Recently it was reported that heat flow within a ferromagnet can produce a flow of spin angular momentuma spin currentand an associated voltage1. This spin Seebeck effect has been observed in metallic1,2, insulating3 and semiconductor ferromagnets4 with temperature gradients across them. Here we describe and report the demonstration of Seebeck spin tunnellinga distinctly different thermal spin flow, of purely interfacial naturegenerated in a tunnel contact between electrodes of different temperatures when at least one of the electrodes is a ferromagnet. The Seebeck spin current is governed by the energy derivative of the tunnel spin polarization. By exploiting this in ferromagnetoxidesilicon tunnel junctions, we observe thermal transfer of spins from the ferromagnet to the silicon without a net tunnel charge current. The induced spin accumulation scales linearly with heating power and changes sign when the temperature differential is reversed. This thermal spin current can be used by itself, or in combination with electrical spin injection, to increase device efficiency. The results highlight the engineering of heat transport in spintronic devices and facilitate the functional use of heat. Spin current is a central aspect of spintronics5,6 and is typically generated by spin-polarized charge current, through spinorbit interaction or by magnetization dynamics. Recently, the potential of thermoelectric effects in magnetic nanostructures711 has been recognized and emphasized12. Notably, the spin Seebeck effect has been observed in ferromagnetic materials with temperature differences between opposite ends14. A microscopic description has been given in terms of a spindependent Seebeck coefficient1,2, although spin current carried by magnons13 and mechanisms without a global spin current or spin accumulation in the ferromagnet4,14,15 have been proposed for some of the experiments1,3,4. The Seebeck spin tunnelling (SST) we describe here is distinctly different in that it involves a thermal spin current that is created in a magnetic tunnel contact. It is associated purely with the tunnel interface. We observe it in a tunnel junction that has one ferromagnetic electrode and one non-magnetic electrode (in our case silicon) held at different temperatures, TFM and TSi, respectively (Fig. 1). The temperature difference, DT 5 TSi 2 TFM, drives preferential tunnel transfer of electrons of either majority or minority spin from the ferromagnet to the silicon. A spin accumulation, Dm, is induced in the silicon, characterized by a difference in the densities and electrochemical potentials of electrons with their magnetic moments respectively parallel and antiparallel to the magnetization of the ferromagnet. To observe Seebeck spin tunnelling, we apply a heating current with density Jheating to the silicon electrode (Fig. 2), causing Joule dissipation in that electrode and raising its temperature with respect to that of the ferromagnetic electrode. Measuring under the condition of zero net
1

tunnel charge current, IT, between the ferromagnet and the silicon (Fig. 2a), this produces a voltage V(IT 5 0) 5 V0 1 DVTH. The first term includes the ohmic voltage drop over part of the heater resistance (the Cr/Au contact and the silicon) and the ordinary charge-related thermovoltage that develops across the contact to maintain IT 5 0. The second term is the SST voltage, that is, the additional voltage across the ferromagnetoxidesilicon tunnel contact due to the thermally induced spin accumulation, Dm, in the silicon. It is detected by measuring the change in voltage in response to an applied magnetic field, Bz, transverse to the spins in the silicon (Hanle geometry, with the magnetic field applied along the z axis, perpendicular to both the inplane magnetization of the ferromagnet and the injected spins). This causes spin precession and a reduction of Dm to zero with a characteristic Lorentzian line shape16, also allowing it to be distinguished from any effect Bz might have on V0. The detected voltage change, DVTH, is given by DVTH 5 V(Dm) 2 V(Dm 5 0) 5 TSP 3 Dm/2, where TSP is the tunnel spin polarization associated with the ferromagnet oxide interface. The voltage V is defined as VSi 2 VFM, where VSi and VFM are the respective potentials of the silicon and ferromagnetic electrodes. A significant thermal spin accumulation is observed in the silicon on heating it so that TSi . TFM, as evidenced by a Hanle curve (Fig. 2b) characteristic of spin polarization induced in the silicon16. The Hanle curve is identical for both directions of the Joule heating current, implying that the sign and magnitude of the spin polarization induced in the silicon are the same for both current directions. The peak amplitude scales quadratically with the heating current density (Fig. 2c) and, thus, linearly with the applied heating power (Fig. 2d, where the power, 2 given per unit volume of silicon, equals rJheating , with r the silicons resistivity). The observed scaling is consistent with a thermally induced 2 spin accumulation, which is expected to be linear in DT and Jheating for
TFM > TSi
Ferromagnet

TSi > TFM Hot

Ferromagnet

Cold

Tunnel oxide Silicon

Cold
Silicon

Tunnel oxide

Hot

Spin polarization induced by Seebeck spin tunnelling

Figure 1 | Basic concept of Seebeck spin tunnelling. In a tunnel contact between a ferromagnetic electrode and a non-magnetic electrode, for instance silicon, a temperature difference between the electrodes causes a transfer of spin angular momentum from the ferromagnet to the non-magnetic electrode. The thermal spin current requires no tunnel charge current. The induced spin polarization is reversed when the temperature difference is reversed and is parallel or antiparallel to the magnetization of the ferromagnet.

Netherlands Foundation for Fundamental Research on Matter (FOM), 3502 GA Utrecht, The Netherlands. 2National Institute of Advanced Industrial Science and Technology (AIST), Spintronics Research Center, Tsukuba, Ibaraki 305-8568, Japan. 3Zernike Institute for Advanced Materials, Physics of Nanodevices, University of Groningen, 9747 AG, Groningen, The Netherlands.

8 2 | N AT U R E | VO L 4 7 5 | 7 J U LY 2 0 1 1

2011 Macmillan Publishers Limited. All rights reserved

LETTER RESEARCH
a
Thermal spin signal, V V0 (mV)

b
0.15 Si heating J = 830 A Si heating J = +830 A cm2 cm2

c
Thermal spin signal, V (mV) Fit V J 2 heating Hanle data 0.10 0.15

d
0.15 Fit V J 2 heating Hanle data 0.10

Cr/Au contact Joule heating current z y Cr/Au contact V x Bz

0.15

Hanle

To heat sink

0.05

0.05

0.05

Ferromagnet cold at TFM Tunnel oxide Silicon heated to TSi

0.00 1,000 500 0 Bz (Oe) 500 1,000 8 6 4 2 0 2 4 6 8

0.00

0.00 0 2 4 6 8 Heating power (nW m3 Si)

Jheating (102 A cm2)

Figure 2 | Observation of thermal spin current from ferromagnet to silicon by Seebeck spin tunnelling. a, Device layout, showing the 3-mm-thick, 800mm-wide, 4-mm-long silicon strip (purple), the 1.5-nm-thick SiO2/Al2O3 tunnel barrier (light grey) and the ferromagnetic Ni80Fe20 electrode (green). The silicon is heated by Joule dissipation from a constant current applied using two Cr/Au contacts at opposite ends, resulting in TSi . TFM. The voltage across the tunnel contact is measured in the absence of a tunnel charge current. The voltage lead simultaneously acts as a thermal sink that keeps the ferromagnet cold. b, Thermally induced spin accumulation in silicon, detected as a voltage

change, DVTH, in an external magnetic field, Bz, applied perpendicular to the tunnel interface (Hanle geometry). The plot shows results for heating current in two opposite directions. The magnetization (M) of the ferromagnet remains in the plane of the thin film owing to shape anisotropy. c, Measured DVTH (symbols) versus heating current density, Jheating, together with a quadratic fit (solid line). d, Measured DVTH (symbols) as a function of Joule heating power, and a linear fit (solid line). All data were measured for p-type silicon with a resistivity of 11 mV cm at 300 K, for a tunnel contact area of 180 mm 3 400 mm and for a base temperature of 300 K.

Joule heating. Similar results are obtained with a four-terminal geometry (Supplementary Information). For the maximum heating power (7.6 nW mm23), we find that DVTH 5 0.13 mV. If we assume that the TSP (30%) previously determined17 for Ni80Fe20/Al2O3 interfaces in metal tunnel junctions is the same in our Ni80Fe20/Al2O3/Si junctions, the thermally induced Dm is 0.87 meV. The SST thus produces a significant spin polarization in the silicon. The sign of the thermal spin accumulation is determined by direct comparison with electrically induced spin accumulation measured in the same tunnel contact, driving a non-zero tunnel current across the tunnel barrier using the three-terminal Hanle method16. The electrically induced spin accumulation gives rise to a Hanle signal (Supplementary Fig. 4) with a Lorentzian line shape and width similar to that found in previous work16 and to the thermally induced signal (Fig. 2), further supporting spin accumulation as the origin of the latter. The electrically induced spin accumulation is negative for negative tunnel current and positive for positive tunnel current (VSi . VFM, extraction16 of holes from p-type silicon), and the latter, positive, spin accumulation is the same sign as the thermally induced spin accumulation (Fig. 2). Thus, thermal transfer of spin with TSi . TFM produces a spin polarization with the same sign as that produced by electrical hole extraction. Such extraction induces in the silicon a magnetic moment parallel to the magnetization of the ferromagnet if the TSP for Ni80Fe20/Al2O3 interfaces is positive (dominated by majority spin electrons), as it is in metalbased tunnel junctions. Hence, the spin accumulation induced by the SST (Fig. 2; silicon heating with positive DV) corresponds to majority spin accumulation in the silicon (that is, a larger number of electrons with spin parallel to that of the majority spin in the ferromagnet, which is equivalent to a smaller number of holes with that spin direction). The observed thermal spin accumulation is quadratic in the drive current, allowing us to exclude as possible origins phenomena that are not even but odd functions of current, such as the spin Hall effect. Other known magnetic or thermomagnetic effects, such as the Hall, Nernst, Ettingshausen and RighiLeduc effects18, can be ruled out because they have a different symmetry and/or would not produce the Lorentzian magnetic field dependence characteristic of the Hanle effect (Supplementary Information). More conclusively, by measuring the Hanle signal in larger magnetic fields, for which the magnetization of the ferromagnet rotates out of the plane of the device, we performed an experimental test (Supplementary Information) that rules out all the known thermomagnetic effects. Because the field and injected spins are then no longer orthogonal, the spins no longer precess and a characteristic recovery of the spin accumulation is observed at high magnetic field values. This recovery occurs only if the spin accumulation

is produced by transfer of spins from the ferromagnet; hence, this observation excludes any source of spin accumulation that does not involve transfer from the ferromagnet. Unintentional electrical injection through shunting of part of the heating current by the tunnel contact can be ruled out for several reasons, including the signal having the wrong sign (Supplementary Information). The data in Fig. 2 thus unambiguously demonstrate thermal transfer of spin from the ferromagnet to the silicon. For the microscopic origin of Seebeck spin tunnelling, we consider electrons as the carrier of spin across the tunnel barrier. Direct (elastic) tunnelling results in a spin current without charge current if there is a different thermal electron distribution in the two electrodes and the TSP varies with electron energy. We illustrate the case with TSi . TFM in Fig. 3a. The excess density of electrons with energy above the Fermi level, EF, in the hot silicon creates a flow of electrons tunnelling to the cold ferromagnet. Simultaneously, the excess number of empty states below the silicon Fermi level creates an electron flow in the opposite direction. Even if the total charge current is zero, there is a net spin current because the two oppositely directed tunnel currents have different TSPs. This is a consequence of the energy variation of the spinpolarized electronic structure of ferromagnetic materials around EF. Indeed, for cobalt, CoFe and Ni80Fe20, it has been found that the spin polarization of electrons tunnelling through an Al2O3 barrier into a metal counter electrode is different for electron energies below and above EF, decaying most significantly in the second case19,20. The salient features of SST are illustrated using a free-electron elastic tunnelling model, including the energy variation of the TSP as a phenomenological parameter. It also explicitly includes the spin accumulation and the feedback it has on the tunnel spin and charge currents. The thermally induced spin accumulation is thus calculated self-consistently (Supplementary Information). First, a non-zero DT indeed produces a spin current and, thereby, a spin accumulation in the silicon (Fig. 3b). The spin accumulation, Dm, scales linearly with DT and changes sign when the hot and cold sides of the junction are interchanged. The Seebeck spin tunnelling coefficient can be defined uniquely as Sst 5 Dm/DT in units of volts per kelvin. It depends on properties of the non-magnetic electrode, in particular the spin relaxation time (Supplementary Information). The sign and magnitude of Dm depend crucially on the form of TSP as a function of energy, as shown in Fig. 3b for three functional forms selected for illustrative purposes. When the TSP is constant below EF but decays above EF (blue profile and symbols), the spin accumulation is positive for TSi . TFM, corresponding to majority spin accumulation with an induced magnetic moment in the silicon parallel to the
7 J U LY 2 0 1 1 | V O L 4 7 5 | N AT U R E | 8 3

2011 Macmillan Publishers Limited. All rights reserved

RESEARCH LETTER
a
Silicon (hot) E Empty states Extraction TSP above EF Ferromagnet (cold) E

b
1.0
Thermal spin accumulation (a.u.)

TSP
Majority spin accumulation

EF E

0.5 TSP 0.0 EF

EF Filled states Injection TSP below EF Majority spin

Density of states

E 0.5
Minority spin accumulation

TSP

EF E

Minority spin

1.0 1.0

0.5

0.0

0.5

1.0

T = TSi TFM (a.u.)

Figure 3 | Origin of Seebeck spin tunnelling and model calculation of salient characteristics. a, Spin-dependent density of states and its occupation for a tunnel contact with a hot silicon electrode and a cold ferromagnetic electrode, and the electron flows (four red arrows) that are induced by the temperature difference. For the cold ferromagnet, taken to be at TFM 5 0, all states below the Fermi energy, EF, are filled (dark blue) and those above EF are empty (light blue). For the hot silicon, a finite number of electrons with excess energy above EF exist, together with a fraction of empty states below EF. The sum of the four tunnel currents, two for each spin, results in a net flow of spin

because the tunnel spin polarizations above and below EF are unequal. b, Calculated value of the spin accumulation, Dm, induced in the silicon electrode of a tunnel contact between the silicon and the ferromagnet, shown as a function of the temperature difference, DT 5 TSi 2 TFM, under the condition of zero tunnel charge current. Results are shown for three different profiles of TSP versus energy (E), as depicted on the right. For DT . 0, TFM 5 300 K and TSi is increased, whereas for DT , 0, TSi 5 300 K and TFM is increased. See Supplementary Information for details and parameters. a.u., arbitrary units.

magnetization of the ferromagnet. If the TSP is instead taken to decay with energy below EF, the sign of Dm is reversed (pink profile and symbols). Most strikingly, Dm vanishes for any value of DT when the TSP is taken to be independent of energy but still non-zero (green profile and symbols). Hence, the induced spin accumulation does not depend on the sign of the TSP itself, but on the energy derivative of the TSP. The sign of the experimental result (Fig. 2; accumulation of majority spins for silicon heating) is consistent with the prediction of the model calculation (Fig. 3b, blue curve; Dm . 0 for DT . 0), considering that the TSPs for interfaces of transition-metal ferromagnets and Al2O3 in metal-based tunnel junctions are reported19,20 to decay significantly for energies above EF. For the magnitude, the model calculation predicts a saturation of Dm at about 10 mV K21, and TSi 2 TFM is estimated to be in the range of 0.1 K to a few kelvin at the most (Supplementary Information). The predicted Dm is below about 10 mV, which is much smaller than the measured signal. Thus, although the model provides a qualitative description of the salient observations, it does not provide quantitative agreement with the data. Part of this may be because it is a free-electron model, because the precise profile of the TSP as a function of energy is not known and because spin (and heat) transfer by inelastic, magnon-assisted tunnelling is not included. Such tunnelling has been shown to be important for the charge thermopower in magnetic tunnel junctions2123. These aspects may cause the model to underestimate the spin current across the tunnel barrier. However, the model also underestimates the electrically induced spin accumulation for which the spin current is rather well known; and control experiments16 and optical detection in spin-based light-emitting diodes24 have clearly established that the large spin accumulation exists in the bulk bands. This suggests that the quantitative disagreement is at least in part related to the description of the conversion of injected spin current into a spin accumulation and, subsequently, into a detected voltage. Next we demonstrate that the sign of the thermal spin current is reversed when DT is reversed. We compare devices in which the heating current is in the silicon (Fig. 4a) with devices in which it is in the ferromagnetic metal electrode (Fig. 4e). In the second of these cases, to improve the heating the metal electrode in the active part of the device consists of Ni80Fe20 (10 nm) and a gold layer only 10 nm thick, whereas in the first case the gold layer is 120 nm thick. A complication is that the current in the ferromagnet is accompanied by a voltage drop across it, such that any conventional magnetoresistance of the ferromagnet affects the signal via the term V0. For magnetic fields along the z axis, this results in a distortion of the Hanle curve that prevents accurate
8 4 | N AT U R E | V O L 4 7 5 | 7 J U LY 2 0 1 1

determination of the induced spin accumulation. It was recently shown25, however, that the spin accumulation can also be detected with an in-plane field parallel to the tunnel interface (in the x direction; see Fig. 2a), giving rise to a so-called inverted Hanle effect. First we use the inverted Hanle effect to detect the spin accumulation induced by heating of the silicon, for the junction shown in Fig. 2. A clear, thermally induced inverted Hanle signal is observed (Fig. 4b), with sign inverted relative to the thermally induced Hanle effect (Fig. 2). Notably, the quadratic and linear scaling with, respectively, heating current and power (Fig. 4c, d) establishes the inverted Hanle effect as a good probe of thermally induced spin accumulation. We then perform a similar set of experiments with the heating current through the ferromagnet (Fig. 4fh). A significant inverted Hanle signal is observed for heating current through the ferromagnet in either direction (Fig. 4f). The signal also scales quadratically with heating current and linearly with power (Fig. 4g, h), which is characteristic of a thermally induced spin accumulation. Most importantly, the sign of DVTH is reversed; DVTH is positive for TFM . TSi and is negative for TSi . TFM. The reversal is not due to an unintentional difference between the tunnel contacts of the two devices, as electrical spin injection was similar for the two junctions (Supplementary information). Thus, the results establish another key feature of SST, namely that the thermal spin accumulation is reversed when DT is reversed. We note that the sharp features around zero field, which change in sign when the direction of the heating current is changed, arise from the aforementioned magnetoresistance of the ferromagnet. For a magnetic field applied along the x direction, these features can be clearly distinguished from the inverted Hanle curve and do not interfere with accurate determination of DVTH (Supplementary Information). We also note that the true magnitude of the spin accumulation is the sum of the Hanle and inverted Hanle amplitudes25. We expect that the efficiency of Seebeck spin tunnelling can be greatly improved by better thermal design of our devices, which is far from optimal. The temperature difference across the thin tunnel barrier can also be optimized by using dielectrics with small interface thermal conductivity and by making use of phonon mismatch at interfaces. The mechanism of SST suggests the development of a new family of spintunnelling materials that, for instance, are designed to have strong variations in TSP around the Fermi energy. Conversely, SST can be used as a new probe of the energy dependence of the TSP. Thermal transfer of spin offers ways to design spintronic devices with fundamentally different characteristics. The fact that SST operates without tunnel current or voltage alleviates issues related to stability and breakdown of the tunnel

2011 Macmillan Publishers Limited. All rights reserved

LETTER RESEARCH
a
Si heated
Bx
Inverted Hanle
SiO2 Au 300 nm Si device layer 3 m SiO2 1 m Si handle wafer 120 nm Amorphous SiO2/Al2O3 (1.5 nm) Tunnel area 180 400 m2 Ni80Fe20 (10 nm)

b
Thermal spin signal, V VSAT (mV)
0.00 0.10 0.20 0.30 0.40 0.50 0.60

c
0.00 0.10 0.20

d
Thermal spin signal, V (mV)
0.00 0.10 0.20 0.30 0.40 0.50 0.60 0

Joule heating of Si Current density Jheating

Inverted Hanle

0.30 0.40

Si heating J = 830 A cm2 Si heating J = +830 A cm2

2,000 0 2,000

Fit V

J2 heating

0.50 0.60 8

Fit V J 2 heating

Inverted Hanle data

8 6 4 2 0 2 4 6

Inverted Hanle data

2 4 6 8

Bx (Oe)

Jheating (102 A cm2)

Heating power (nW m3 Si)

e
Thermal spin signal, V VSAT (mV)

f
0.16
Amorphous SiO2/Al2O3 (1.7 nm) Tunnel area 200 200 m2 Au Au (10 nm) 120 nm

g
FM heating J = 910 kA cm2 FM heating J = +910 kA cm2

h
0.16

Thermal spin signal, V (mV)

FM heated
Bx
Inverted Hanle
Ni80Fe20 (10 nm)

Inverted Hanle data Fit V J 2 heating

0.16

Inverted Hanle data

Fit V J 2 heating

0.12

SiO2 300 nm Si device layer 3 m SiO2 1 m Si handle wafer

0.08

Inverted Hanle

0.12

0.12

0.08

0.08

0.04

0.04

0.04

0.00

0.00

0.00

800

800

2,000

4,000 6,000

8,000

Bx (Oe)

Jheating (105 A cm2)

Heating power (nW m3 metal)

Figure 4 | Sign reversal of thermal spin current by heating the silicon or the ferromagnet. a, Layout of the device for silicon heating. b, Thermally induced spin accumulation for silicon heating in the same junction as in Fig. 2, but now detected with an external magnetic field, Bx, applied parallel to the tunnel interface (inverted Hanle geometry). c, d, Measured DVTH (symbols) versus heater current density (c) and heating power (d), with corresponding fits. e, Layout of the device for heating of the ferromagnetic electrode (FM). In the

tunnel area, the metal electrode consists of 10-nm-thick Ni80Fe20 and 10-nmthick gold. fh, Similar sets of inverted Hanle data as in bd, but now with the ferromagnetic electrode heated. For all data, the inverted Hanle signal is defined as the voltage at a given Bx minus the saturation voltage, VSAT, at large Bx, with voltage polarity consistently defined as VSi 2 VFM. All data were measured for p-type silicon with a resistivity of 11 mV cm at 300 K, for a tunnel contact area of 200 mm 3 200 mm and for a base temperature of 300 K.
15. Adachi, H. et al. Gigantic enhancement of spin Seebeck effect by phonon drag. Appl. Phys. Lett. 97, 252506 (2010). 16. Dash, S. P., Sharma, S., Patel, R. S., de Jong, M. P. & Jansen, R. Electrical creation of spin polarization in silicon at room temperature. Nature 462, 491494 (2009). 17. Min, B. C., Motohashi, K., Lodder, J. C. & Jansen, R. Tunable spin-tunnel contacts to silicon using low-work-function ferromagnets. Nature Mater. 5, 817822 (2006). 18. Nolas, G. S., Sharp, J. & Goldsmid, H. J. Thermoelectrics: Basic Principles and New Materials Developments Ch. 1 (Springer, 2001). 19. Valenzuela, S. O., Monsma, D. J., Marcus, C. M., Narayanamurti, V. & Tinkham, M. Spin polarized tunneling at finite bias. Phys. Rev. Lett. 94, 196601 (2005). 20. Park, B. G., Banerjee, T., Lodder, J. C. & Jansen, R. Tunnel spin polarization versus energy for clean and doped Al2O3 barriers. Phys. Rev. Lett. 99, 217206 (2007). 21. Wang, Z.-C., Su, G. & Gao, S. Spin-dependent thermal and electrical transport in a spin-valve system. Phys. Rev. B 63, 224419 (2001). 22. McCann, E. & Falko, V. I. Giant magnetothermopower of magnon-assisted transport in ferromagnetic tunnel junctions. Phys. Rev. B 66, 134424 (2002). 23. McCann, E. & Falko, V. I. A tunnel junction between a ferromagnet and a normal metal: magnon-assisted contribution to thermopower and conductance. J. Magn. Magn. Mater. 268, 123131 (2004). 24. Jansen, R. et al. Electrical spin injection into moderately doped silicon enabled by tailored interfaces. Phys. Rev. B 82, 241305 (2010). 25. Dash, S. P. et al. Spin precession and decoherence near an interface with a ferromagnet. Preprint at http://arxiv.org/abs/1101.1691 (2011). Supplementary Information is linked to the online version of the paper at www.nature.com/nature. Acknowledgements We are grateful to S. P. Dash for help with the device fabrication and discussions, T. Yorozu for the finite-element calculations and A. Yamamoto for making the finite-element program available to us. This work was financially supported by the program Controlling spin dynamics in magnetic nanostructures of the Netherlands Foundation for Fundamental Research on Matter (FOM). Author Contributions J.-C.L.B. and R.J. designed the experiments. J.-C.L.B. and S.S. fabricated the devices. J.-C.L.B., S.S., H.S. and R.J. contributed to the measurements. R.J. developed the model calculation. All authors contributed to the planning, discussion and analysis of the research, and to the writing of the manuscript. Author Information Reprints and permissions information is available at www.nature.com/reprints. The authors declare no competing financial interests. Readers are welcome to comment on the online version of this article at www.nature.com/nature. Correspondence and requests for materials should be addressed to R.J. (ron.jansen@aist.go.jp).

barrier and decay of TSP at high bias that degrades electrical transfer of spin. It can also be linked to optics using lasers as the heating source, allowing optical creation and control of spin polarization in any semiconductor, including silicon, without the need for circularly polarized light and optical orientation. But above all, Seebeck spin tunnelling offers prospects to create spin currents in a more power-efficient way, either by itself or acting jointly with electrical spin injection, and allows the functional use of heat in electronic devices and circuits.
Received 23 February; accepted 24 May 2011. Published online 29 June 2011. 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. Uchida, K. et al. Observation of the spin Seebeck effect. Nature 455, 778781 (2008). Slachter, A., Bakker, F. L., Adam, J.-P. & van Wees, B. J. Thermally driven spin injection from a ferromagnet into a non-magnetic metal. Nature Phys. 6, 879882 (2010). Uchida, K. et al. Spin Seebeck insulator. Nature Mater. 9, 894897 (2010). Jaworski, C. M. et al. Observation of the spin-Seebeck effect in a ferromagnetic semiconductor. Nature Mater. 9, 898903 (2010). Zutic, I., Fabian, J. & Das Sarma, S. Spintronics: fundamentals and applications. Rev. Mod. Phys. 76, 323410 (2004). Chappert, C., Fert, A. & Nguyen van Dau, F. The emergence of spin electronics in data storage. Nature Mater. 6, 813823 (2007). Johnson, M. & Silsbee, R. H. Thermodynamic analysis of interfacial transport and of the thermomagnetoelectric system. Phys. Rev. B 35, 49594972 (1987). Fukushima, A. et al. Peltier effect in sub-micron-size metallic junctions. Jpn. J. Appl. Phys. 44, L12L14 (2005). Gravier, L., Serrano-Guisan, S., Reuse, F. & Ansermet, J.-Ph. Spin-dependent Peltier effect of perpendicular currents in multilayered nanowires. Phys. Rev. B 73, 052410 (2006). Hatami, M., Bauer, G. E. W., Zhang, Q. & Kelly, P. J. Thermoelectric effects in magnetic nanostructures. Phys. Rev. B 79, 174426 (2009). Slonckzewski, J. C. Initiation of spin-transfer torque by thermal transport from magnons. Phys. Rev. B 82, 054403 (2010). Bauer, G. E. W., MacDonald, A. H. & Maekawa, S. Spin caloritronics. Solid State Commun. 150, 459460 (2010). Xiao, J., Bauer, G. E. W., Uchida, K., Saitoh, E. & Maekawa, S. Theory of magnondriven spin Seebeck effect. Phys. Rev. B 81, 214418 (2010). Sinova, J. Spin Seebeck effect: thinks globally but acts locally. Nature Mater. 9, 880881 (2010).

7 J U LY 2 0 1 1 | V O L 4 7 5 | N AT U R E | 8 5

2011 Macmillan Publishers Limited. All rights reserved