Spin polarization in half-metals probed by femtosecond spin ... - Nature

13 downloads 0 Views 2MB Size Report
Dec 14, 2008 - Because the direct energy transfer by Elliot–Yafet scattering is blocked in a half-metal, the demagnetization time is a measure for the degree of ...
ARTICLES PUBLISHED ONLINE: 14 DECEMBER 2008 DOI: 10.1038/NMAT2341

Spin polarization in half-metals probed by femtosecond spin excitation Georg M. Müller1 , Jakob Walowski1 , Marija Djordjevic1 , Gou-Xing Miao2 , Arunava Gupta2 , Ana V. Ramos3 , Kai Gehrke4 , Vasily Moshnyaga4 , Konrad Samwer4 , Jan Schmalhorst5 , Andy Thomas5 , Andreas Hütten5 , Günter Reiss5 , Jagadeesh S. Moodera6 and Markus Münzenberg1 * Knowledge of the spin polarization is of fundamental importance for the use of a material in spintronics applications. Here, we used femtosecond optical excitation of half-metals to distinguish between half-metallic and metallic properties. Because the direct energy transfer by Elliot–Yafet scattering is blocked in a half-metal, the demagnetization time is a measure for the degree of half-metallicity. We propose that this characteristic enables us vice versa to establish a novel and fast characterization tool for this highly important material class used in spin-electronic devices. The technique has been applied to a variety of materials where the spin polarization at the Fermi level ranges from 45 to 98%: Ni, Co2 MnSi, Fe3 O4 , La0.66 Sr0.33 MnO3 and CrO2 .

F

or the performance of a magneto-electronic device, the spin polarization of the flowing electrons is a major ingredient for a high signal strength and robust readout1 . The polarization also enables the manipulation of the magnetization itself 2 , and thus the realization of spin logic circuits3 . But what is the bottleneck to achieve a high spin polarization? Important for the spin transport through a device is the current polarization Pi , defined by Pi,α =

i↑,α − i↓,α , i↑,α + i↓,α

α = d,sp.

a

b

E –hω

Sc EF

I

(1)

The itinerant ferromagnets Fe, Co and Ni belong to the transition metals. Therefore, the band structure consists of hybridized bands with the different character of the underlying wavefunctions (s,p,d). Because of their highly delocalized nature, s,p-like bands predominantly contribute to Pi (ref. 4). From this simple picture, the apparent contradiction that a high spin polarization in transition-metal ferromagnets is destroyed by the less-polarized itinerant states is obtained. What are the workarounds to obtain a high contrast for both spin channels in spin-transport devices? One solution is manipulation of the ‘composition’. Recently, this was very successfully achieved by coherent tunnelling5,6 . By selecting a specific direction within the Brillouin zone, a 100% polarized band can be selected7 . Another successful approach is to manipulate the tunnelling probability by using a ferromagnetic insulator as a spin filter8 . A suppression of weakly polarized states at the Fermi level can also be realized in the material itself by chemical bonding. These materials are called half-metals9 . The classification of different half-metals due to their transport characteristics was given by Coey et al.10 . For transition metals with exchange split, fully polarized d-like bands, by hybridization with oxygen 2p states, the 4s states are shifted above the Fermi level and a 100% polarized material can be formed. A prototype for this half-metal of the first class is CrO2 (type IA ). On the other hand, by p-like

B

c

d Sc I

τ

–hω –hω

B

Figure 1 | Schematic representation of the ‘classical’ techniques to determine the spin polarization P of a material. a, Spin-polarized photoemission. b, Meservey–Tedrow technique. c, Andreev reflection using superconducting contacts10,12 (Sc). d, The approach using all-optical pump–probe experiments to determine the demagnetization time τm , which is related to the half-metallicity of the material.

bonding, the d states can be suppressed below the 4s band to achieve 100% polarization. Well-known examples are the half or full Heusler alloys, such as Co2 MnSi (also type IA ). For magnetite Fe3 O4 , a spin-polarized conduction channel is formed by hopping in the minority band. It belongs to a second class (type IIB ). A third class is the so-called transport half-metals such as the

1 IV.

Phys. Institut, Universität Göttingen, D-37077 Göttingen, Germany, 2 MINT Center, University of Alabama, Tuscaloosa, Alabama 35487, USA, 3 CEA Saclay, 91191 Gif sur Yvette Cedex, France, 4 I. Phys. Institut, Universität Göttingen, D-37077 Göttingen, Germany, 5 Department of Physics, Universität Bielefeld, D-33501 Bielefeld, Germany, 6 Francis Bitter Magnet Laboratory, Massachusetts Institute of Technology, Cambridge, Massachusetts 02139, USA. *e-mail: [email protected]. 56

NATURE MATERIALS | VOL 8 | JANUARY 2009 | www.nature.com/naturematerials © 2009 Macmillan Publishers Limited. All rights reserved.

NATURE MATERIALS DOI: 10.1038/NMAT2341

ARTICLES

a

b

Electrons Lattice Spin

3.0

τel¬lat

ΔT/ΔT∞

e¬ Electrons (Te)

τ el¬sp

τ lat¬sp

2.0

1.0

Lattice (Tl) 0

Spins (Ts)

0

c

d

5.0

7.5

2.0 1.5

τel¬lat

Electrons (Te)

τ el¬sp

τ lat¬sp

Lattice (Tl)

ΔT/ΔT∞



2.5 τ (ps)

1.0 0.5 0

Spins (Ts)

¬5

0

5

10

15

20

τ (ps)

Figure 2 | Three-temperature model as determined by rate equations. a–d, Schematic representation and numerical calculations for the case of a ferromagnetic metal (a,b) and for the case of the half-metal (c,d). In the case of the half-metal, the direct pathway for electron–spin interaction is blocked (c), resulting in a slow increase of the spin temperature (d).

manganite La0.66 Sr0.33 MnO3 . One spin sub-band has a much larger effective mass that barely contributes to the transport and Pi reaches a value of more than 90% (type IIIA ). A general extensive overview of the different families of half-metals in addition to the materials mentioned above—zinc-blende compounds, ruthenates and double-perovskites, chalcospinels, pyrites and organic halfmetals—is given by Katsnelson et al.11 . To characterize the value of the spin polarization, different techniques are available12 . They can be classified into two main subgroups: (1) those that determine the density of states n↓ and n↑ at the Fermi level using a photoemission experiment or (2) experiments determining the spin currents i↓ and i↑ in a spin-transport device (Fig. 1a–c). In a photoemission experiment (Fig. 1a), the density of states is mapped. The photoelectrons are attributed to their energy and spin. In Fig. 1b, the Meservey– Tedrow technique is shown. By applying a magnetic field to the superconducting electrode, the quasiparticle density shows Zeeman split resonances and the current through the barrier can be separated for both spin directions. In Fig. 1c, the superconductor is a probe for the diffusive transport (Andreev reflection). The electron probes the superconductor for a certain time before a Cooper pair is formed and a hole of opposite spin polarization is reflected. On the basis of these experimental techniques, Mazin and co-workers defined a more general spin polarization13,14 including a weighting of the density of states n↑,↓ by the Fermi velocity vF , where Pn accounts for the case in Fig. 1a, Pnv for the case of ballistic and Pnv 2 for diffusive transport in the case in Fig. 1c. This results in the definitions

i i nv ↑ − nv ↓ , i = 0,1,2 Pnv i = hnv i i↑ + hnv i i↓ (2)

and

PT =

nT ↑ − nT 2 ↓ 2



hnT 2 i↑ + hnT 2 i↓

,

where hi↑(↓) is defined as the sum over the corresponding quantity for majority (minority) spin n↑(↓) (EF ) weighted by their velocities v↑(↓) (EF ) raised to the power of i. As the transmission T 2 (T tunnelling matrix element) for an amorphous barrier such as Al2 O3 decreases with the Fermi velocity, predominantly s, p-like states contribute to the tunnelling current for the Meservey–Tedrow technique15 (Fig. 1b). Thus, values obtained by the Meservey– Tedrow technique PT and by Andreev reflection in the diffusive limit Pnv 2 are almost equal12 . A less common but direct method to probe the bulk density-of-states momentum resolved is spinresolved positron annihilation16 . However, using the ‘classical’ methods, the interpretation is not always clear, and sample preparation can be tedious, expensive or demanding. The finding by Zhang et al.17 of a very slow demagnetization of the spin system for the half-metal CrO2 in all-optical pump–probe experiments gave rise to the idea to implement a novel contact-free technique to characterize half-metallic materials for spin-electronic devices: if the demagnetization time τm of the spin system probed by the time-resolved Kerr rotation after femtosecond excitation is characteristically different (Fig. 1d), then it should be possible to use it as a measure of the degree of half-metallicity of a material. This opens up the way towards an alternative technique that overcomes the drawbacks of the ‘classical’ methods; it is fast, does not involve the preparation of a transport device and is suited for all different types of half-metal. The mechanism responsible for the slow demagnetization times observed is shown schematically in Fig. 2a,c, and on the basis of numerical calculations of the demagnetization time τm in Fig. 2b,d, for the ferromagnetic metal and a half-metal, respectively. The numerical calculation is based on the three-temperature model18 that artificially separates the electrons, spins and the lattice by defining three independent temperatures (Tel ,Tsp and Tlat ) that are interconnected by relaxation rates between spins and electrons τel–sp , electrons and lattice τel –lat and lattice and spins τlat–sp . The three temperatures have been modelled for parameters of Ni (Fig. 2b) and CrO2 (Fig. 2d). For the ferromagnetic metal, the understanding is that the Elliot–Yafet

NATURE MATERIALS | VOL 8 | JANUARY 2009 | www.nature.com/naturematerials © 2009 Macmillan Publishers Limited. All rights reserved.

57

NATURE MATERIALS DOI: 10.1038/NMAT2341

ARTICLES a

b

Fe3O4

Δθ Kerr (arb. units)

Δθ Kerr (arb. units)

Ni

Co2MnSi

LSMO

CrO2

0

5 Δτ (ps)

10

15

0

250

500 Δτ (ps)

750

1,000

Figure 3 | Study of the spin polarization using all-optical pump–probe experiments to determine the demagnetization time τm . τm is related to the half-metallic nature of the material. a,b, Although Ni and the Heusler alloy Co2 MnSi show a fast demagnetization (a), for the half-metals Fe3 O4 (12 nm), CrO2 and La0.66 Sr0.33 MnO3 (LSMO), τm is two orders of magnitude larger (b). The experiments are conducted at room temperature. The data is normalized to the value at a delay of 15 ps in a and 1 ns in b.

spin-flip scattering is responsible for the ultrafast demagnetization19 (Fig. 2a) and results in a rapid temperature increase of the spin system. The Elliot–Yafet scattering process originates from the band mixing for spin-up and spin-down states at symmetry points close in energy owing to the spin–orbit coupling. The increase of the temperature of the spin system Tsp follows almost instantaneously the elevated temperature of the electron system Tel elevated by the laser pulse excitation (Fig. 2b). The timescale for the increase of the spin temperature is a few hundred femtoseconds. In a half-metal, this direct channel is blocked17 . There are no spin channels available for spin-flip scattering and the energy has to be transferred through lattice excitations (Fig. 2c). The lattice and the spins are only weakly coupled by anisotropy fluctuations of the crystal field arising from the spin–orbit interaction20,21 . Therefore, this is a very slow channel. The timescale for the increase of the spin temperature is a few hundred picoseconds (Fig. 2d) as found in a number of experiments17,20,22 . Comparing this technique to transport methods, one sees that the latter case averages over the device area in space (>104 nm2 ) and the data readout time (>1 µs). At elevated temperatures a depolarization and a loss of the half-metallic nature is found, arising from temperature-induced magnetic excitations23–25 . In contrast, in the time-resolved experiment, the excited electron itself probes the relaxation channels within a distance given by vF τe . This compares to a probing depth smaller than about six unit cells and a ‘readout’ time of a few femtoseconds (details of the approach are given in Supplementary Information, section (c) Temperature-induced magnetic excitations). In comparison, what are the expected length scales and timescales of the magnetic fluctuations, determined by the spin-wave stiffness and quadratic dispersion26,27 ? With increasing temperature, predominantly low-k (long-wavelength) magnons at the Brillouin-zone centre are populated following the Bose–Einstein statistics. As their occupation number increases, they form fluctuating domains and the magnon picture is replaced with that of ordered magnetic clusters, preserved even at TC (ref. 28). In contrast, high-energy spin waves are sparsely occupied. Electrons with lifetimes on the femtosecond scale probe a ‘frozen’ image of the much slower varying spin waves. The material has a locally 58

half-metallic character within this picture (details of the approach are given in Supplementary Information, section (c) Temperatureinduced magnetic excitations). Following the work of Soulen et al.12 , we compare different classes of material important for use in spin-electronic devices that exhibit different values of spin polarization (Fig. 3). The relaxation rates were analysed by studying the reflectivity R(τ ) and Kerr rotation θK (τ ) as a function of delay time τ in the femtosecond pump–probe experiment at room temperature. For the manganite La0.66 Sr0.33 MnO3 , room temperature corresponds to 0.82TC . The spin-wave energy near the zone boundary is a factor of three higher than kB TC . This allows the excitation of spin waves that have a short wavelength (∼0.2kBZ ). However, these are strongly damped with a short lifetime. As a representative of the transition-metal ferromagnets, we have chosen Ni. The demagnetization dynamics is well established18,19,29,30 . For a 15-nmthick Ni film on a Si(100) substrate, the electron–spin relaxation time is determined to be τel–sp = 0.1–0.14 ps (Fig. 3a). Because of the much higher heat capacity, the lattice temperature response is slower. The value for the electron–lattice relaxation time is τel –lat = 0.9 ps (not shown). The most rigid half-metal CrO2 was predicted to be half-metallic by Schwarz31 . Indeed, it is proven to be half-metallic by all three ‘classical’ methods (Fig. 1a–c) with values of P = 95–98% (refs 10,32–34). Taking this wellestablished half-metal (type IA ), we find for an epitaxial CrO2 film grown on TiO2 (100) substrate, an electron–spin relaxation time of τel–sp ∼ 125 ps, with a minimum of the magnetization at around τm = 200–300 ps (Fig. 3b). This is in sharp contrast to the ultrafast demagnetization of Ni. However, the electron– lattice dynamics is quite unchanged (τel –lat = 0.2 ps). Also for the half-metal of the second class Fe3 O4 (type IIB ), where the conduction at EF is dominated by localized minority spin transport, we find for a 12 nm thickness a very slow demagnetization time even larger than τm > 1,000 ps (Fig. 3b). For the half-metal La0.66 Sr0.33 MnO3 (type IIIA ), we find a demagnetization time τm of 400–600 ps (Fig. 3b). The reflectivity data reveal a very different slope as expected from the three-temperature model. Whereas the electron–lattice relaxation determines the short timescale (Fig. 2d), NATURE MATERIALS | VOL 8 | JANUARY 2009 | www.nature.com/naturematerials

© 2009 Macmillan Publishers Limited. All rights reserved.

NATURE MATERIALS DOI: 10.1038/NMAT2341

ARTICLES

1,000 0.01

100 Half-metal

E (meV)

τm (ps)

0.10

logτm > 6.0

10

1.00

1.0 10.0

|P| > 8.0 0.1 0.2

0.4

0.6

0.8

1.0

|P|

Figure 4 | Classification of the materials by the value of the spin polarization P and their demagnetization time τm . Left side: τm is obtained using the all-optical pump–probe experiment from Fig. 3. In addition, values for Gd (ref. 44), Fe, Co (ref. 45) and permalloy (Ni80 Fe20 ) are included (black squares). The lines are calculated using the two-relaxation-times model for different magnetic anisotropy values K1 (1 meV (black line) and 4 µeV per atom (blue line)). The error bars for τm are determined by the laser pulse width and given by a 5% interval θK, min . Right side: The materials on the right side, mostly from ref. 20, belong to the chalcospinels (yellow circles), manganites (light-blue squares), ruthenates and double-perovskites (grey circles). They are predicted to be either half-metallic or close to being half-metallic on the basis of band structure calculations (see Supplementary Information, Table SII). The energy scale on the right side given as a guideline relates to the timescale on the left through the Planck constant by τm = h/E. They are given for both graphs. Note the logarithmic scale for τm and E.

for a detailed analysis on the longer timescale, thermal diffusion has to be included. In contrast, a simple analysis in the case of La0.66 Sr0.33 MnO3 is not possible, because the reflectivity is strongly influenced by the intrinsic interdependence of conductivity and magnetism22 . An exception is observed for the Co2 MnSi Heusler alloy (Fig. 3a), also a half-metal of type IA . In contrast to CrO2 of the same type, the demagnetization time τm = 0.25 ps at room temperature is very fast and similar to that of Ni. In the following, we evaluate some possible reasons. Even though many Heusler compounds have been predicted to be half-metallic, the experimental verification tends to be more difficult. As opposed to CrO2 , where half-metallicity can be destroyed only by chemical decomposition into Cr2 O3 , the half-metallic character in the Heusler alloy depends on the ‘chemical’ order of the alloy, that is, all four cubic sublattices have to be ordered (L21 structure). In Co2 MnSi, anti-site Co defects can result in a defect band at the Fermi energy, which would destroy the spin polarization35 . Second, it has to be considered that the energy gap in the minority states channel is only 0.64 eV, which is much smaller than for the other half-metals CrO2 (1.8 eV) or Fe3 O4 (1.3 eV) showing a slow demagnetization time τm . (Detailed values for 1E and P (theory and experiment) classified by the different definitions and references are given in Supplementary Information. For Fe3 O4 , because of the lack of other values, Pn determined by photoemission experiments is used. For the other materials PT from tunnelling and Pnv 2 determined from Andreev reflection, in all cases comparable, is given.) Moreover, the position of the Fermi energy strongly depends on the composition of the Heusler alloy and thus the excitation gap 1sf for spin-flip excitations can be reduced. For tunnel junctions based on Heusler alloy electrodes, very high values of the tunnelling magnetoresistance are found36 ; approaching room temperature, the values decrease rapidly. The low-temperature tunnel-magnetoresistance experiments here result in a value of PT (20 K) = 66% for the Co2 MnSi film grown

in the (110) orientation37 , which is found to be less than for CrO2 , Fe3 O4 and La0.66 Sr0.33 MnO3 . Generally, a lower polarization value is obtained for this direction as compared with the (001) direction, even for best-controlled growth38 . For a quantitative comparison between spin polarization at low temperatures and the demagnetization time τm measured by the all-optical approach, all values are compared in Fig. 4. In addition, the data for Fe, Co, permalloy and Gd are given. As τm changes by orders of magnitude owing to the thermal insulation between spins and electrons if the half-metallic nature appears, the value is plotted on a logarithmic scale. In the case where the polarization is as low as for Ni (PT ,nv 2 = 44–46%; refs 12,39) and for Co2 MnSi (PT = 66%; ref. 37), the demagnetization times are fast and in the picosecond range. For materials exhibiting a high polarization above 80% (Fe3 O4 Pn = −80% (ref. 40), CrO2 PT ,nv 2 = 95–98% (refs 10,32–34), La0.66 Sr0.33 MnO3 Pnv 2 = 96% (ref. 14)), the values found for the demagnetization time are two orders of magnitude higher. The last two are confirmed to be half-metals in spin-transport devices. It should be noted that in real spin-transport devices the full theoretical value of 100% can almost never be achieved and the values are mostly lower for various experimental reasons. The slow value of τm found for the half-metal scales with the strength of the spin–orbit interaction. This can be verified analogous to the case in ref. 20 where the values of τm and of the magnetic anisotropy constant K1 , originating from the spin–orbit interaction, reveal an inverse scaling relationship. On the basis of the above observations, we suggest an alternate approach to classify a half-metal from the determination of the logarithm of the demagnetization time log τm (τm [ps]). If log τm > 0.6, the fast electron–spin relaxation processes by Elliot– Yafet scattering are blocked and we expect the material to show half-metallic properties in spin-transport devices (Fig. 4, left side). Because of the logarithmic scaling, a strong sensitivity is inherent to the method. On the right side of Fig. 4, extra materials are included.

NATURE MATERIALS | VOL 8 | JANUARY 2009 | www.nature.com/naturematerials © 2009 Macmillan Publishers Limited. All rights reserved.

59

NATURE MATERIALS DOI: 10.1038/NMAT2341

ARTICLES

The factor AθD (T ) reflects the temperature dependence. In the intermediate range, both energy transfer channels contribute in parallel. Thus, the rates add reciprocally and we determine

Ni

τm =

Δθ Kerr (arb. units)

Co2MnSi

Fe3O4

LSMO

CrO2

0

5 Δτ (ps)

10

15

Figure 5 | All-optical pump–probe experiments shown on the short timescale to reveal the demagnetization by spin flips of the non-thermalized electrons. When the electron and spin systems become thermally isolated, the step-like feature arising from spin flips of non-thermal electrons gradually evolves from Ni towards the half-metals Fe3 O4 , La0.66 Sr0.33 MnO3 (LSMO) and CrO2 . For Fe3 O4 , a thicker film of 50 nm is shown as compared with that in Fig. 3. The experiments are conducted at room temperature. The data is normalized to the value at a delay of 15 ps to reveal the step-like feature for all samples.

They belong to the family of the chalcospinels, manganites, ruthenates and double-perovskites, which are predicted to be either half-metallic or close to being half-metallic on the basis of band structure calculations. However, the value of the spin polarization PT ,nv i is known experimentally for only a few of them. Interestingly, all of them show a slow τm . Provided that they are not insulating (no spin polarization Pn can be defined in this case), we can classify the materials as being half-metallic on the basis of our method. To develop a theoretical basis for our approach, by using Fermi’s golden rule, we calculate the electron scattering probability with a spin flip involved and relate the electron–spin relaxation time τel–sp with the spin polarization P n . We obtain, as derived in more detail in the Supplementary Information (section (b)), the following equation: τel–sp =

1 τel,0 . c 2 (1 − P n )

(3)

The parameter c describes the degree of admixture in the spin variables at the Fermi level as defined by Elliot and is related to the spin–orbit interaction ςSO by c ∼ ςSO /1Eexch (ref. 41). It scales inversely with the band splitting Eexch at a band crossing. τel,0 is given by the electron momentum scattering rate. On the other hand, the relaxation time τlat–sp is related to the anisotropy energy Eaniso as was proposed by Hübner21 , τlat–sp = 60

1 . AθD (T )|Eaniso |2

(4)

τel–sp τlat–sp . τel–sp + τlat–sp

(5)

Two curves are calculated using equations (3)–(5) to elucidate two limiting cases: one for a weak-anisotropy material with K1 = 4 µeV and the other for a strong-anisotropy material with K1 = 1 meV (with τel,0 /c 2 = 0.18 ps). As shown in Fig. 4, for Pn → 1 the curves approach the values of τm = 1 ns and 4 ps respectively, which results in the criterion log τm > 0.6. A step-like feature observed in Fig. 3b for CrO2 and La0.66 Sr0.33 MnO3 is also characteristic for half-metals. The signal magnitude is typically of a few per cent of the total demagnetization20 , and reveals that the exclusion of spin-flip processes is active in generality only after the excited electrons are thermalized and below the level of the energy gap of the other spin channel as analysed by Zhang et al.17 . The data on the short timescale are given in Fig. 5. For Fe3 O4 , a different film thickness is shown (50 nm) because for the thinner film (12 nm) the step-like feature could not be resolved because of the small signal strength. From metallic Ni to half-metallic CrO2 , the step-like contribution becomes more and more pronounced (the data is normalized to the value at 15 ps to reveal the feature). For the thick 50 nm Fe3 O4 film, a less pronounced half-metallic behaviour between that of the Heusler and the La0.66 Sr0.33 MnO3 film is found. Further understanding may also give a measure of the half-metallicity and circumvent the special care that had to be taken to suppress magnetic precessional motion42 . The question of which definition of spin polarization P or type of half-metal is probed remains open. Because of the nature of the photoexcited electron probing its electronic surrounding, the optical technique determines the transport spin polarization Pnv 2 , which is closely related to PT (chosen as the abscissa in Fig. 4). This is most evident for the maganites, where the values for P vary strongly according to the different definitions and a slow demagnetization time τm is found in the all-optical experiment. In the beginning, we relate the breakdown of the half-metallic nature as the temperature is increased, indicated by the spin polarization P(T ) in transport experiments, to spin-wave excitations. Although P(T ) behaves proportional to the temperature dependence of the magnetization M (T ) as for CrO2 , where the room-temperature polarization is still 96% (ref. 32), a much stronger decay is observed for others: at present, different models explaining the breakdown of the spin polarization with temperature for the Heusler alloys are controversially discussed. As mechanisms, low-energy electron non-quasiparticle excitations (‘spin polarons’)23 , thermally activated spin mixing due to coupling to optical phonons24 or a change in hybridizations due to moment fluctuations25 are proposed. Within the demagnetization experiments, it is observed that the demagnetization time even increases in a critical region near TC . As a reason, the increase of the specific heat of the spin system cS ∼ (∂M 2 /∂T ) (in mean field approximation) was given22 . A comparison between the spin-current depolarization and the ultrafast demagnetization experiment as a local probe might provide further insight into the mechanism of spin-current depolarization in future experiments. Indeed, rigid transport spin polarization is the important quantity that determines the efficiency of a spin-electronic device. Putting this method into practice, by the choice of the wavelength for the excitation pulse, the energy can be adjusted to the value of the energy gap of the half-metal to further quantify the model. In summary, our experiments clearly demonstrate that the demagnetization time, which can be readily determined in a contact-free measurement using the optical pump–probe NATURE MATERIALS | VOL 8 | JANUARY 2009 | www.nature.com/naturematerials

© 2009 Macmillan Publishers Limited. All rights reserved.

NATURE MATERIALS DOI: 10.1038/NMAT2341

ARTICLES

technique, is a good measure for the transport spin polarization. Vice versa, the polarization dependence of the demagnetization time can be described in a model based on an Elliot–Yafet-type electron–spin interaction, which is an important further step towards verification of recent microscopic theories in femtosecond magnetization dynamics43 .

Methods The relaxation rates were analysed by studying the reflectivity R(τ ) and Kerr rotation θK (τ ) as a function of delay time τ in the femtosecond pump–probe experiment (laser pulse characteristics 1.5 eV, 60–80 fs, 20–40 mJ cm−2 ) at room temperature. A self-built Ti:sapphire oscillator and a regenerative amplification stage with a repetition rate of 250 kHz and ∼1 µJ per pulse were used (RegA 9050 short pulse version, Coherent). The Kerr rotation θK (τ ) was determined by a double-modulation technique29,30 . The probing depth of the Kerr rotation θK (τ ) is given by the optical penetration length within the magnetic film. The experiment is therefore not as surface-sensitive as electron transport, probing only a few monolayers near the interface. The Kerr rotation θK (τ ) is defined in the following as the asymmetric part, changing with the field direction θK ,− (τ ) = (1/2)(θK (τ ,M ) − θK (τ ,−M )), whereas the symmetric part θK , + (τ ) = (1/2)(θK (τ ,M ) + θK (τ ,−M )) mirrors the reflectivity change R(τ ).

Received 30 January 2008; accepted 21 October 2008; published online 14 December 2008 References 1. Wolf, S. A. et al. Spintronics: A spin-based electronics vision for the future. Science 294, 1488–1495 (2001). 2. Kiselev, S. I. et al. Microwave oscillations of a nanomagnet driven by a spin-polarized current. Nature 425, 380–383 (2003). 3. Reiss, G. & Meyners, D. Logic based on magnetic tunnel junctions. J. Phys. Condens. Matter 19, 165220–165242 (2007). 4. Stearns, M. B. Simple explanation of tunneling spin-polarization of Fe, Co, Ni and its alloys. J. Magn. Magn. Mater. 5, 167–171 (1977). 5. Yuasa, S., Nagahama, T., Fukushima, A., Suzuki, Y. & Ando, K. Giant room-temperature magnetoresistance in single-crystal Fe/MgO/Fe magnetic tunnel junctions. Nature Mater. 3, 868–871 (2004). 6. Parkin, S. S. P. et al. Giant tunnelling magnetoresistance at room temperature with MgO(100) tunnel barriers. Nature Mater. 3, 862–867 (2004). 7. Butler, W. H., Zhang, X. G., Schulthess, T. C. & MacLaren, J. M. Spin-dependent tunneling conductance of Fe/MgO/Fe sandwiches. Phys. Rev. B 63, 054416 (2001). 8. Nagahama, T., Santos, T. S. & Moodera, J. S. Enhanced magnetotransport at high bias in quasimagnetic tunnel junctions with EuS spin-filter barriers. Phys. Rev. Lett. 99, 016602 (2007). 9. de Groot, R. A., Möller, F. M., van Engen, P.G. & Buschow, K.H.J. New class of materials: Half-metallic ferromagnets. Phys. Rev. Lett. 50, 2024–2027 (1983). 10. Venkatesan, M. in Handbook of Magnetism and Advanced Magnetic Materials Vol. 4, Novel Materials (eds Kronmüller, H. & Parkin, S. S. P.) (Wiley, 2007). 11. Katsnelson, M. I., Irkhin, V. Yu., Chioncel, L., Lichtenstein, A. I. & de Groot, R. A. Half metallic ferromagnets: From band structure to many-body effects. Rev. Mod. Phys. 80, 315–378 (2008). 12. Soulen, R. J. Jr. et al. Measuring the spin polarization of a metal with a superconducting point contact. Science 282, 85–88 (1998). 13. Mazin, I. I. How to define and calculate the degree of spin polarization in ferromagnets. Phys. Rev. Lett. 83, 1427–1430 (1999). 14. Nadgorny, B. et al. Origin of high transport spin polarization in La0.7 Sr0.3 MnO3 : Direct evidence for minority spin states. Phys. Rev. B 63, 184433 (2001). 15. Münzenberg, M. & Moodera, J. S. Superconductor-ferromagnet tunneling measurements indicate sp-spin and d-spin currents. Phys. Rev. B 70, 060402(R) (2004). 16. Hanssen, K. E. H. M., Mijnarends, P. E., Rabou, L. P. L. M. & Buschow, K. H. J. Positron-annihilation study of the half-metallic ferromagnet NiMnSb: Experiment. Phys. Rev. B 42, 1533–1540 (1990). 17. Zhang, Q., Nurmikko, A. V., Miao, G. X., Xiao, G. & Gupta, A. Ultrafast spin-dynamics in half-metallic CrO2 thin films. Phys. Rev. B 74, 064414 (2006). 18. Beaurepaire, E., Merle, J.-C., Daunois, A. & Bigot, J.-Y. Ultrafast spin dynamics in ferromagnetic nickel. Phys. Rev. Lett. 76, 4250–4253 (1996). 19. Koopmans, B., Ruigrok, J. J. M., Dalla Longa, F. & de Jonge, W. J. M. Unifying ultrafast magnetization dynamics. Phys. Rev. Lett. 95, 267207 (2005). 20. Ogasawara, T. et al. General features of photoinduced spin dynamics in ferromagnetic and ferrimagnetic compounds. Phys. Rev. Lett. 94, 087202 (2005). 21. Hübner, W. & Bennemann, K. H. Simple theory for spin-lattice relaxation in metallic rare-earth ferromagnets. Phys. Rev. B 53, 3422–3427 (1996). 22. Lobad, A. I., Averitt, R. D., Kwon, C. & Taylor, J. Spin–lattice interaction in colossal magnetoresistance manganites. Appl. Phys. Lett. 77, 4025–4027 (2000).

23. Chioncel, L. et al. Nonquasiparticle states in Co2 MnSi evidenced through magnetic tunnel junction spectroscopy measurements. Phys. Rev. Lett. 100, 086402 (2008). 24. Dowben, P. A. & Skomski, R. Finite-temperature spin polarization in half-metallic ferromagnets. J. Appl. Phys. 93, 7948–7950 (2003). 25. Lezaic, M., Mavropoulos, Ph., Enkovaara, J., Bihlmayer, G. & Blügel, S. Thermal collapse of spin polarization in half-metallic ferromagnets. Phys. Rev. Lett. 97, 026404 (2006). 26. Pickart, S. J. et al. Spin-wave dispersion in ferromagnetic Ni and fcc Co. Phys. Rev. 156, 623–626 (1967). 27. Ye, F. et al. Evolution of spin-wave excitations in ferromagnetic metallic manganites. Phys. Rev. Lett. 96, 047204 (2006). 28. Halilov, S. V., Eschrig, H., Perlov, A. Y. & Oppeneer, P. M. Adiabatic spin dynamics from spin-density-functional theory: Application to Fe, Co, and Ni. Phys. Rev. B 58, 293–302 (1998). 29. Koopmans, B., van Kampen, M., Kohlhepp, J.T. & de Jonge, W. J. M. Ultrafast magneto-optics in nickel: Magnetism or optics? Phys. Rev. Lett. 85, 844–848 (2000). 30. Djordjevic, M. & Münzenberg, M. Connecting the timescales in picosecond remagnetization experiments. Phys. Rev. B 75, 012404 (2007). 31. Schwarz, K.-H. CrO2 predicted as a half-metallic ferromagnet. J. Phys. F: Met. Phys. 16, L211–L215 (1986). 32. Dedkov, Y. S. et al. Room-temperature observation of high-spin polarization of epitaxial CrO2 (100) island films at the Fermi energy. Appl. Phys. Lett. 80, 4181–4183 (2002). 33. Anguelouch, A. et al. Near-complete spin polarization in atomically-smooth chromium-dioxide epitaxial films prepared using a CVD liquid precursor. Phys. Rev. B 64, 180408 (2001). 34. Parker, J. S., Watts, S. M., Ivanov, P. G. & Xiong, P. Spin polarization of CrO2 at and across an artificial barrier. Phys. Rev. Lett. 88, 196601 (2001). 35. Picozzi, S., Continenza, A. & Freeman, A. J. Role of structural defects on the half-metallic character of Co2 MnGe and Co2 MnSi Heusler alloys. Phys. Rev. B 69, 094423 (2004). 36. Sakuraba, Y. et al. Giant tunneling magnetoresistance in Co2 MnSi/Al–O/Co2 MnSi magnetic tunnel junctions. Appl. Phys. Lett. 88, 192508 (2006). 37. Schmalhorst, J. et al. Transport properties of magnetic tunnel junctions with Co2 MnSi electrodes: The influence of temperature-dependent interface magnetization and electronic band structure. Phys. Rev. B 75, 014403 (2007). 38. Hattori, M., Sakuraba, Y., Oogane, M., Ando, Y. & Miyazaki, T. Tunnel magnetoresistance effect in magnetic tunnel junctions using a Co2 MnSi(110) electrode. Appl. Phys. Express 1, 021301 (2008). 39. Kim, T. H. & Moodera, J. S. Large spin polarization in epitaxial and polycrystalline Ni films. Phys. Rev. B 69, 020403 (2004). 40. Dedkov, Y. S., Rüdiger, U. & Güntherodt, G. Evidence for the half-metallic ferromagnetic state of Fe3 O4 by spin-resolved photoelectron spectroscopy. Phys. Rev. B 65, 064417 (2002). 41. Elliot, R. J. Theory of the effect of spin–orbit coupling on magnetic resonance in some semiconductors. Phys. Rev. 96, 266–279 (1954). 42. Müller, G.M., Münzenberg, M., Miao, G.-X. & Gupta, A. Activation of additional energy dissipation processes in the magnetization dynamics of epitaxial chromium dioxide films. Phys. Rev. B 77, 020412 (2008). 43. Stamm, C. et al. Femtosecond modification of electron localization and transfer of angular momentum in nickel. Nature Mater. 6, 740–743 (2007). 44. Melnikov, A. et al. Nonequilibrium magnetization dynamics of gadolinium studied by magnetic linear dichroism in time-resolved 4f core-level photoemission. Phys. Rev. Lett. 100, 107202 (2008). 45. Cinchetti, M. et al. Spin-flip processes and ultrafast magnetization dynamics in Co: Unifying the microscopic and macroscopic view of femtosecond magnetism. Phys. Rev. Lett. 97, 177201 (2006).

Acknowledgements Support by the Deutsche Forschungsgemeinschaft within the priority program SPP 1133 and SFB 602 is gratefully acknowledged. In MIT, the work is supported by NSF and ONR grants; at the University of Alabama, support by a NSF MRSEC grant is gratefully acknowledged. M.M. thanks J. Stöhr and B. Koopmans for fruitful discussions.

Author contributions M.M. designed research and model; G.M.M., J.W., M.D. and M.M. carried out research (femtosecond dynamics), G.M.M. main contributor to experiment execution and data analysis; G.X.M., A.G., A.V.R., K.G., V.M., K.S., J.S., A.T., A.H. and G.R. carried out research (spin transport, film growth and characterization); M.M. wrote the manuscript; J.S.M. and M.M. coordinated research.

Additional information Supplementary Information accompanies this paper on www.nature.com/naturematerials. Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions. Correspondence and requests for materials should be addressed to M.M.

NATURE MATERIALS | VOL 8 | JANUARY 2009 | www.nature.com/naturematerials © 2009 Macmillan Publishers Limited. All rights reserved.

61