week ending

PRL 109, 127202 (2012) PHYSICAL REVIEW LETTERS 21 SEPTEMBER 2012

Penetration Depth of Transverse Spin Current in Ultrathin Ferromagnets

A. Ghosh,1 S. Auffret,1 U. Ebels,1 and W. E. Bailey1,2,*
1
SPINTEC, UMR(8191) CEA/CNRS/UJF/Grenoble INP, INAC, 17 rue des Martyrs, 38054 Grenoble Cedex, France
2
Department of Applied Physics and Applied Mathematics, Columbia University, New York, New York 10027, USA
(Received 2 December 2011; published 18 September 2012)
We report a novel depth dependence for the penetration of spin current into ultrathin ferromagnets.
Ferromagnetic resonance measurements show that transverse spin current pumped into three structurally
distinct ferromagnets is attenuated, on reflection, by an amount proportional to the ferromagnetic layer
thickness, saturating abruptly at 1:2
0:1 nm. The observed power-law decay, differing significantly
from the (exponential) characteristic-length dependence for longitudinal spin current, confirms models of
spin momentum transfer which have been inaccessible to experiment.

DOI: 10.1103/PhysRevLett.109.127202 PACS numbers: 75.47.m, 76.50.+g, 75.76.+j, 75.78.n

In spin momentum transfer (SMT) [1,2], a spin- near-N interface Fj , spin current absorption in layers only
polarized current, injected into a ferromagnetic layer (F), a few angstroms thick can be measured as a perturbation on
transfers its angular momentum to the F magnetization M an otherwise robust signal.
as it is absorbed. The depth dependence of spin current Prior magnetotransport measurements have indicated
absorption in the F is of fundamental interest for SMT. the existence of a characteristic length for spin current
Since injected spins polarized transverse to M are absorbed absorption near the Fermi energy in 3d ferromagnets.
and exert torque [3,4], the penetration depth has been Giant magnetoresistance measurements [20–22] reveal an
framed in terms of the transverse spin coherence length exponential decrease of spin polarized current density with

c in the F [5–7]. This SMT length scale is thought to be increasing depth z in the ferromagnet as expðz=
SD Þ
very short,
c  2 nm in 3d metallic ferromagnets. where
SD is the spin diffusion length. These measure-
Experiments have not probed the dependence of SMT on ments refer to the longitudinal component of spin, parallel
the several-angstrom scale in the relevant thickness range and antiparallel to M. The exponential depth dependence
tF < 2 nm; coarser-resolution experiments performed on reflects a Poisson process for spin relaxation: spin-flip
thicker Co layers [8,9] include the possibility of some scattering events are uncorrelated over distance with uni-
variation in SMT efficiency. form probability distribution over depth.
The spin pumping effect [10–12] provides an alternative A different depth dependence of absorption has been
prospect to study the length scale of SMT. In a spin-valve predicted for transverse spin currents in F layers. The
structure (F1 =N=F2 ), precession of F1 sources (‘‘pumps’’) length scale is set by the transverse spin coherence length,
a spin current across the N and into F2 , where it is absorbed given to first order by
J  =jk"f  k#f j where k"ð#Þ
f are the
identically to spin-polarized electrical current injected majority (minority) Fermi wave vectors [7], or equivalently
through voltage, verified as torque on the magnetization hvg =2
ex , with vg as the spin-averaged group velocity
[13]. Spin pumping can be regarded as an inverse process and
ex the exchange splitting [23]. This quantity is esti-
to current-pumped precession [14,15]: Onsager relations mated at 1–2 nm near the Fermi energy in 3d ferromagnets
link interfacial spin torque and spin pumping coefficients [6]. The functional form predicted for total transverse spin
within a constant of proportionality [16,17] current absorption approximates an algebraically decaying
The real part of the spin mixing conductance g"#r sinusoid about a step function [24], with differences de-
contributes an interfacial Gilbert damping in F1 =N=F2 pending upon the Fermi surface integration [6,7,24–26].
trilayers [12,18]. This quantity, which is proportional to Experimental results in the regime t 
J exist only for hot
the Slonczewski spin torque coefficient for a N=F inter- electrons E  EF  5 eV, injected and detected from vac-
face, can be accessed conveniently through ferromagnetic uum using Mott polarimetry [23,27].
resonance (FMR) measurements of extended thin-film In this Letter, we have measured the depth-dependence
stacks. Because these measurements do not require device of transverse spin current absorption in three ultrathin
nanofabrication, they are rapid compared with SMT device ferromagnets and one antiferromagnet at EF using the
measurement for a given film configuration, allowing a spin pumping effect. Spin mixing conductances in thicker
larger number of layers to be characterized in finite time. polycrystalline structures are shown, for the first time, to be
Finite-size magnetostatic [19] and activation volume [9] in quantitative agreement with theory. In ultrathin films, we
effects do not enter in the measurement, facilitating observe transverse spin current absorption proportional to
interpretation. Moreover, because the Gilbert damping of thickness with abrupt saturation at a critical depth
c ¼
(thicker) Fi depends on the properties of the (ultrathin) 1:2
0:1 nm in the structurally diverse ferromagnets

0031-9007=12=109(12)=127202(5) 127202-1 Ó 2012 American Physical Society

 week ending
PRL 109, 127202 (2012) PHYSICAL REVIEW LETTERS 21 SEPTEMBER 2012

FIG. 2 (color online). Isolation of contribution to Gilbert

FIG. 1 (color online). Contributed Gilbert damping
ðtF1 Þ ¼ damping in F1 =Cu=F2 ðtF2 Þ, F1 ¼ Coð8 nmÞ, F2 ¼ CoFeBtCoFeB .
F2 ðtF1 Þ  noF2 ðtF1 Þ from the introduction of a second F2 Main panel: Derivative FMR spectra, 16 GHz, tCoFeB ¼ 0:5, 3.0,
interface in F1 ðtF1 Þ=Cuð5 nmÞ½=F2 =Alð3 nmÞ structures. and 10.0 nm. Note the increase in linewidth for the Co resonance
Inset: noF2 ðtF1 Þ and F2 ðtF1 Þ for CoFeB layers. (low field) as a function of tCoFeB . Inset: Separation of intrinsic
and extrinsic damping for different tCoFeB .
Ni81 Fe19 (‘‘Py’’), Co60 Fe20 B20 (‘‘CoFeB’’), and pure Co.
Identical behavior is seen in the antiferromagnet Ir80 Mn20 , absorption line, which appear as maxima in the experi-
with slightly higher
C ¼ 1:5
0:1 nm. The observed mental derivative (lock-in) signal @00 =@H (peak-to-peak
behavior is highly reminiscent of the spin-polarized hot- linewidth). The offset term
H0 is attributed to inhomo-
electron reflection in Ref [27]. Our measurements are geneities in the local resonance field. The difference
inconsistent with earlier experimental reports of an expo-
ðtF1 Þ ¼ F2 ðtF1 Þ  noF2 ðtF1 Þ isolates the effect on
nential onset of spin current absorption in F layers [28], Gilbert-type damping from the addition of the F2 interface.
and highlight the correlated, path-dependent nature of This additional Gilbert damping is taken as the effect of the
transverse spin-current absorption predicted by theory. spin pumping.
Layers have been prepared by sputtering on ion-cleaned In Fig. 1, we show a log-log plot of the contributed
Si=SiO2 substrates, seeded in every case with Tað5 nmÞ= damping
 as a function of the precessing bottom F1
Cuð5 nmÞ bilayers, capped in every case with 3 nm Al thickness tF1 . The plot shows
ðtF Þ ¼ Ktn , with n ¼
layers, oxidized in air. All samples have been characterized 1:05
0:05. The power law is in excellent agreement
by variable-frequency (2–24 GHz), swept-field FMR at with the inverse thickness dependence of contributed
room temperature. Care has been taken in the deposited damping predicted from spin pumping,
ðtF Þ ¼
sample series to isolate the effect of these covering layers jj@g"#eff =ð4Ms Þt1 "#
F . The parameter geff , in units of
alone, and the frequency range considered has facilitated nm2 , is the effective spin mixing conductance in channels
our isolation of the Gilbert damping constant  in the per unit area. This relationship may be expressed equiv-
measurements, as in prior work [29]. alently as the product of additional Gilbert relaxation
First, we show data which support the existence of the rate G and ferromagnetic layer thickness,
GtF ¼
spin pumping effect in our F1 =N=F2 heterostructures with
Ms tF ¼ 2 @g"#eff =4. Here the contribution also be-
the three different F layers considered. Three series of comes independent of Ms . We make use of the
GtF
F1 ðtF1 Þ=Cuð5 nmÞ=F2 =Alð3 nmÞ structures were prepared product in subsequent discussions.
for tF1 ¼ 3, 4.5, 6.0, 10.0, 17.5, 30.0 nm with and without The effective spin mixing conductances per interfacial
the F2 overlayer, with F1 =F2 combinations Py/Co, Py/ area g"#eff are listed in the second column of Table I. Spin
CoFeB, CoFeB/Co, for a total of 36 samples. The choices mixing conductances for specific interfaces Fi =Cu, as-
of tF2 ¼ 5 nm (Co) and 17.5 nm (CoFeB) avoid overlap of sumed here to be independent of growth order, can be
the F1;2 resonances. extracted from the measurements of g"#eff . We use
For each sample, the Gilbert damping  of F1 has been
extracted from the frequency-dependent linewidths
Hð!Þ 1=g"#eff ¼ 1=~
g"#F1=Cu þ 1=~
g"#F2=Cu ; (1)
in swept-field FMR measurement. The (intrinsic) Gilbert
1 1
damping  is isolated from the inhomogeneous broadening where g~1 1
F=N ¼ gF=N  2 gS;N is the Sharvin-corrected spin
pffiffiffi

H0 through
Hð!Þ ¼
H0 þ ð2= 3Þ!=; see, e.g., mixing conductance [30] and the Sharvin value for Cu is
Refs [10,29], and Fig. 2, inset, for an example.
H mea- gS;N ¼ 15:0 nm2 . The effective spin mixing conductance
sures the width between inflection points in a Lorenzian applies to F1 and F2 alike. Three linear equations for g1
eff

127202-2
 week ending
PRL 109, 127202 (2012) PHYSICAL REVIEW LETTERS 21 SEPTEMBER 2012

TABLE I. First two columns: effective spin mixing conduc-

tances g"#eff for F1 =N=F2 combinations, extracted from the data in
Fig. 1; second two columns: interfacial spin mixing conductan-
ces g"#F;N from g"#eff . See text for details.

F1 =N=F2 g"#eff ðnm2 Þ F=Cu g"#F=N ðnm2 Þ

Py/Cu/Co 15:0
1:5 Py 14:4
1:4
Py/Cu/CoFeB 15:3
1:5 CoFeB 16:0
1:6
CoFeB/Cu/Co 16:8
1:6 Co 15:7
1:6

can be written in terms of two values of g1 F=Cu each; the

system is solved for the three unknown interface values
gF=Cu . The ‘‘bare’’ conductances gF=Cu are tabulated in
Column 4 for comparison with calculated values.
We highlight the close agreement of the three polycrys- FIG. 3 (color online). Gilbert relaxation rate—F1 layer thick-
talline interfacial spin mixing conductances with each ness product contributed by ultrathin ferromagnets CoFeB, Py,
other and with theory. The three g~F=Cu values found from and Co and antiferromagnet IrMn to the F1 layer resonance in
measurements of
 all agree with the theoretical g"# [30] F1 =Cuð5 nmÞ=F2 ðtF2 Þ, AF2 ðtAF Þ. Note: Saturation level converts
for alloyed Co/Cu(111), 14:6 nm2 , within 10%. The to
 ¼ 1:9  103 for Co(8 nm) (
H ¼ 0:73 Oe=GHz),
measurements presented so far strongly support the idea
 ¼ 2:7  103 for Py(10 nm) (
H ¼ 1:1 Oe=GHz). Right
that the interfacial damping in Fi =Cu=Fj arises from the axis: Conversion to effective spin mixing conductance assuming
gL ¼ 2:1.
spin pumping effect.
Because of cancellation of terms in Eq (1), spin mixing
conductance at a top interface g"#N=F2 can be extracted
directly from the effective spin mixing conductance of conductance for the stack, and thus the damping of F1 .
the stack g"#eff . As measured, g"#N=F1 agrees closely with the In the data shown, it is possible to detect the effect of
angstrom-scale coverages of CoFeB (F2 ) on the low-field
Cu Sharvin conductance of 15:0 nm2 for F1 ¼ Co, Py;
"#;1 Co resonance (F1 ). Even as the CoFeB resonance itself is
thus the two g1 S;N =2 terms and single gN=F1 term cancel in at the threshold of visibility, not observed at 0.5 nm and
Eq. (1), and we can identify g"#eff ’ g"#N=F2 . In the following eventually observed at 3.0 nm, the spin current absorption
discussion, a proportionality of Gilbert damping enhance- in CoFeB can be measured through an increase of Co
ment
GtF1 with tF2 implies a proportionality of spin linewidth of 10 Oe (19%).
current absorption near the N=F2 interface with tF2 . The frequency dependence of the Co(8 nm) linewidth as
We next present, in Figs. 2 and 3, measurements of the a function of CoFeB thickness (inset), plotted as
Hð!Þ, is
onset of spin current absorption in ultrathin layers with used to separate intrinsic and extrinsic linewidth. The
magnetic order. We use the thickness-dependent onset of linear fits to
Hð!Þ for each value of tCoFeB indicate that
the enhanced damping as a measure of spin current ab- the line broadening due to CoFeB coverage arises from
sorption. The effective spin mixing conductance g"#eff of both intrinsic (!-dependent) and extrinsic (constant) com-
F1 ðtF1 Þ=Cuð5 nmÞ=F2 ðtF2 Þ, tF2 ¼ 0–15 nm is character- ponents, denoted as
H0 . The contribution to intrinsic
ized through the damping enhancement
 of F1 . In these (Gilbert) relaxation is attributed to spin pumping.
three series, we have used the F1 ðtF1 Þ : F2 combinations Figure 3 presents the central result of our Letter. We plot
Co(8 nm):Py, Co(8 nm):CoFeB, and Py(10 nm):Co. the product
GtF1 for the three ferromagnets CoFeB, Py,
Sample data are presented in Fig. 2. We show field-swept Co, and for the antiferromagnet IrMn. The spin current
FMR spectra at 16 GHz for Coð8 nmÞ=Cuð5 nmÞ= absorption of these four layers, measured through the
CoFeBðtCoFeB Þ, tCoFeB ¼ 0:5, 3.0, and 10.0 nm. damping as a function of F2 coverage, is strikingly similar.
Resonances are well separated (through choice of the F1 For the ferromagnets CoFeB, Py, and Co, there is a linear
thickness) and the low-field Co(8 nm) resonance is moni- increase of the effective spin mixing conductance g"#eff , as a
tored as a function of CoFeB coverage. The results are not function of coverage, rising to a maximum value
sensitive to the thickness of F1 in the range used. The and cutting off at a critical thickness t ¼
C ,
C ¼ 1:2

higher-field resonance for the ultrathin CoFeB, which 0:1 nm. The observed saturation values of
GtF1 ’
decreases in resonance field HB as a function of tCoFeB 410
20 Mhz  nm for the F layers are equal within
due to surface anisotropy, is not material in the data experimental error, and consistent with the values deter-
reduction. mined for the thicker trilayers in Table I. For the antiferro-
A consequence of Eq. (1) is that the spin mixing magnet layer IrMn,
C is significantly larger (1.5 nm)
conductance of N=F2 affects the effective spin mixing and the contributed relaxation rate significantly smaller,

127202-3
 week ending
PRL 109, 127202 (2012) PHYSICAL REVIEW LETTERS 21 SEPTEMBER 2012

FIG. 4 (color online). Normalized plot of spin current absorption. Open symbols: Ferromagnets Co, Py, CoFeB, and antiferromagnet
IrMn. Closed symbols: Paramagnetic layers Ru, Pt, and Pd. Thicknesses are normalized to the critical thickness
c (F, AF) or
characteristic spin flip length
sf (PM); enhanced damping is normalized to its saturation value. Inset: Non-normalized data for
paramagnetic layers,
ðtPM Þ; lines are exponential fits. See text for details.

GtF1 ’ 270
20 Mhz  nm. There is a weaker depen- of precession during transit, with
J as the thickness for a
dence of inhomogeneous broadening
H0 on overlayer rotation of .
thickness tF2 , consistent with its classification as extrinsic For (near-Fermi surface) transport experiments, wave
(defect-related, magnetostatic) in origin; see Supplemental vectors of electrons propagating forward through the fer-
Material [31]. romagnet span 2 of the solid angle. The distribution of
The ‘‘universal’’ plot in Fig. 4 compares spin current wave vectors with respect to the surface normal implies a
absorption in the three ultrathin ferromagnets and one distribution of transit times through F, and therefore a
antiferromagnet with that previously known in paramag- distribution of electron spin rotations about M while cross-
nets. The data in Fig. 3 are normalized in thickness to
C ing F. Fermi-surface averaging thus leads to algebraic
and damping size effect to
Gmax tF1 , and plotted alongside decay of the injected spin current for a given depth z:
similarly prepared structures which substitute paramagnets calculations predict a nonexponential convergence, as a
Ru, Pd, and Pt for F2 . The paramagnetic layers show function of tF , of spin current absorption [24] or spin
an exponential depth dependence of spin current absorp- mixing conductance [7] at a N=FðtF Þ interface, which
tion, similar to that observed by others [32,33] as oscillates about the saturation value with a period of 2
J .
1  expð2t=
SF Þ;
SF is a characteristic length for spin We observe linear convergence, as a function of tF ,
relaxation in the paramagnet. Variations in the saturation towards the saturation value of the spin mixing conduc-
value of
 for the three PM layer coverages can be tance g"# . The result highlights the path-dependent nature
interpreted through different interface conductances of transverse spin current absorption. The net transverse
gCu=PM for the three Cu/PM interfaces [29]. The linear spin current absorption does not result from a Poisson
thickness dependence of spin current absorption in the scattering process of uncorrelated spin-flip scatterers, but
magnetically ordered materials can be distinguished easily rather an angular average of continuous spin rotations for
from the exponential thickness dependence in the each electron wave vector in F. The absence of overshoot
paramagnets. or oscillations would nevertheless seem to imply a mecha-
The three F layers are structurally diverse, with fcc order nism which suppresses interference. Tight-binding calcu-
for Py, mixed fcc/HCP for Co, and disorder likely for lations [25,26] have predicted that point defects (Fe in
CoFeB. Nevertheless, the onset of spin current absorption Ni80 Fe20 ) are fully effective in suppressing oscillations
is identically proportional to thickness in these layers [34]. predicted for Cu/Ni; a very similar tF =
C to cutoff depen-
The absence of discontinuity after interface formation dence is predicted with
C  0:7 nm for (100) and
suggests bulk effects in the FM. A thickness proportion- 1:1 nm for (111) structures [26]; the latter is close to
ality of transverse spin current rotation about M in the hot- our result for the FM layers. The longer value of
C found
electron polarimetry measurements, both in transmission for the bulk antiferromagnet is consistent with weakened
[23] and reflection [27], has been interpreted as the effect exchange; IrMn is nearer its Curie point of 400 C.

127202-4
 week ending
PRL 109, 127202 (2012) PHYSICAL REVIEW LETTERS 21 SEPTEMBER 2012

Summary.—We have measured the thickness-dependent [18] B. Heinrich, Y. Tserkovnyak, G. Woltersdorf, A. Brataas, R.
transverse spin current absorption in ultrathin ( 1 nm) Urban, and G. Bauer, Phys. Rev. Lett. 90, 187601 (2003).
polycrystalline ferromagnets using the spin pumping ef- [19] R. D. McMichael and M. D. Stiles, J. Appl. Phys. 97,
fect. Spin mixing conductances for thicker films >1:1 nm 10J901 (2005).
[20] Q. Yang, P. Holody, S.-F. Lee, L. L. Henry, R. Loloee, P. A.
agree closely with theoretical predictions. Below this limit,
Schroeder, W. P. Pratt, and J. Bass, Phys. Rev. Lett. 72,
we observe a strict proportionality in thickness which 3274 (1994).
differs from both the longitudinal spin current absorption [21] S. Steenwyk, S. Hsu, R. Loloee, J. Bass, and J. Pratt,
in ferromagnets and the spin current absorption in W. P., J. Magn. Magn. Mater. 170, L1 (1997).
paramagnets. [22] L. Piraux, S. Dubois, A. Fert, and L. Belliard, Eur. Phys.
We thank M. Chshiev for discussions, acknowledge J. B 4, 413 (1998).
U.S. NSF-ECCS-0925829, the Bourse Accueil Pro [23] W. Weber, S. Riesen, and H. C. Siegmann, Science 291,
No. 2715 of the Rhône-Alpes Region, and the French 1015 (2001).
National Research Agency (ANR) Grant No. ANR-09- [24] M. D. Stiles and A. Zangwill, Phys. Rev. B 66, 014407
NANO-037. (2002).
[25] K. Carva and I. Turek, Phys. Rev. B 76, 104409 (2007).
[26] S. Wang, Y. Xu, and K. Xia, Phys. Rev. B 77, 184430
(2008).
[27] W. Weber, S. Riesen, C. H. Back, A. Shorikov, V. Anisimov,
*[email protected] and H. C. Siegmann, Phys. Rev. B 66, 100405 (2002).
[1] M. Tsoi, A. G. M. Jansen, J. Bass, W.-C. Chiang, [28] T. Taniguchi, S. Yakata, H. Imamura, and Y. Ando, Appl.
M. Seck, V. Tsoi, and P. Wyder, Phys. Rev. Lett. 80, Phys. Express 1, 031302 (2008).
4281 (1998). [29] A. Ghosh, J. F. Sierra, S. Auffret, U. Ebels, and W. E.
[2] J. Katine, F. Albert, R. Buhrman, E. Meyers, and D. Ralph, Bailey, Appl. Phys. Lett. 98, 052508 (2011).
Phys. Rev. Lett. 84, 3149 (2000). [30] Y. Tserkovnyak, A. Brataas, G. Bauer, and B. Halperin,
[3] J. Slonczewski, J. Magn. Magn. Mater. 159, L1 (1996). Rev. Mod. Phys. 77, 1375 (2005).
[4] Y. B. Bazaliy, B. Jones, and S.-C. Zhang, Phys. Rev. B 57, [31] See Supplemental Material at http://link.aps.org/
R3213 (1998). supplemental/10.1103/PhysRevLett.109.127202 for the
[5] S. Zhang, P. M. Levy, and A. Fert, Phys. Rev. Lett. 88, dependence of inhomogeneous broadening
H0 on tF2 .
236601 (2002). [32] J. Foros, G. Woltersdorf, B. Heinrich, and A. Brataas,
[6] J. Zhang, P. Levy, S. Zhang, and V. Antropov, Phys. Rev. J. Appl. Phys. 97, 10A714 (2005).
Lett. 93, 256602 (2004). [33] S. Yakata, Y. Ando, T. Miyazaki, and S. Mizukami, Jpn.
[7] M. Zwierzycki, Y. Tserkovnyak, P. J. Kelly, A. Brataas, J. Appl. Phys. 45, 3892 (2006).
and G. E. W. Bauer, Phys. Rev. B 71, 064420 (2005). [34] Structural issues of the ultrathin FM do not appear to be
[8] F. J. Albert, N. C. Emley, E. B. Myers, D. C. Ralph, and important. Our FM2 layers are deposited on identical
R. A. Buhrman, Phys. Rev. Lett. 89, 226802 (2002). underlayers, (111)-textured columnar Cu, verified by
[9] W. Chen, M. J. Rooks, N. Ruiz, J. Z. Sun, and A. D. Kent, TEM consistent with prior work for Ta seeding [35]. All
Phys. Rev. B 74, 144408 (2006). measurements are above the threshold for ferromagnetism
[10] R. Urban, G. Woltersdorf, and B. Heinrich, Phys. Rev. at 4 Å [36] and quasi-layer-by-layer grain-epitaxial growth
Lett. 87, 217204 (2001). has been seen in similarly prepared (sputtered) films in the
[11] B. Heinrich, Y. Tserkovnyak, G. Woltersdorf, A. Brataas, past [37]; for Py and Co, two (111)-monolayers at 4 Å are
R. Urban, and G. E. W. Bauer, Phys. Rev. Lett. 90, 187601 likely to be continuous if perhaps intermixed. For the PM
(2003). layers Pt and Pd, separate TEM investigations of similarly
[12] Y. Tserkovnyak, A. Brataas, and G. E. W. Bauer, Phys. prepared ultrathin Co/Pt [38] and Pd/Cu/Pd/Co multi-
Rev. Lett. 88, 117601 (2002). layers [39] demonstrate similarly abrupt interfaces, with
[13] G. Woltersdorf, O. Mosendz, B. Heinrich, and C. H. Back, intermixing confined to 1–2 monolayers.
Phys. Rev. Lett. 99, 246603 (2007). [35] S. N. R. Nakatani, K. Hoshino, and Y. Sugita, Jpn. J. Appl.
[14] S. Kiselev, J. Sankey, I. Krlvorotov, N. Emley, R. Phys. 33, 133 (1994).
Schoelkopf, R. Buhnnan, and D. Ralph, Nature [36] W. E. Bailey, S. E. Russek, X.-G. Zhang, and W. H. Butler,
(London) 425, 380 (2003). Phys. Rev. B 72, 012409 (2005).
[15] W. Rippard, M. Pufall, S. Kaka, S. Russek, and T. Silva, [37] W. E. Bailey, S. X. Wang, and E. Y. Tsymbal, J. Appl.
Phys. Rev. Lett. 92, 027201 (2004). Phys. 87, 5185 (2000).
[16] A. Brataas, Y. V. Nazarov, and G. E. W. Bauer, Phys. Rev. [38] G. Bertero and R. Sinclair, J. Magn. Magn. Mater. 134,
Lett. 84, 2481 (2000). 173 (1994).
[17] A. Brataas, Y. Tserkovnyak, G. E. W. Bauer, and P. J. [39] M. Sakurai and T. Shinjo, J. Magn. Magn. Mater. 128, 237
Kelly, arXiv:1108.0385. (1993).

127202-5