Influence of MgO tunnel barrier thickness on spin-transfer ferromagnetic resonance and torque in magnetic tunnel junctions

0.26 MB·English

by  Witold Skowroński

Influence of MgO tunnel barrier thickness on spin-transfer ferromagnetic resonance
and torque in magnetic tunnel junctions
Witold Skowron´ski,∗ Maciej Czapkiewicz, Marek Frankowski, Jerzy Wrona, and Tomasz Stobiecki
Department of Electronics, AGH University of Science and Technology, Al. Mickiewicza 30, 30-059 Krak´ow, Poland
Gu¨nter Reiss
Thin Films and Physics of Nanostructures, Bielefeld University, 33615 Bielefeld, Germany
3
Khattiya Chalapat and Gheorghe S. Paraoanu
1
0 Low temperature laboratory, Aalto University, P.O.Box 15100, FI-02015 Aalto, Finland
2
Sebastiaan van Dijken
n
NanoSpin, Department of Applied Physics, Aalto University School of Science, P.O.Box 15100, FI-00076 Aalto, Finland
a
J (Dated: January 31, 2013)

0 Spin-transferferromagneticresonance(ST-FMR)insymmetricmagnetictunneljunctions(MTJs)
3
withavariedthicknessoftheMgOtunnelbarrier(0.75nm<tMgO <1.05nm)isstudiedusingthe

spin-torque diode effect. The application of an RF current into nanosized MTJs generates a DC
]
l mixingvoltageacross thedevicewhen thefrequencyisin resonancewith theresistance oscillations
al arisingfromthespintransfertorque. Magnetizationprecessioninthefreeandreferencelayersofthe
h MTJs is analyzed by comparing ST-FMR signals with macrospin and micromagnetic simulations.
- From ST-FMR spectra at different DC bias voltage, the in-plane and perpendicular torkances are
s derived. Theexperimentsandfree-electronmodelcalculationsshowthattheabsolutetorquevalues
e
areindependentoftunnelbarrierthickness. Theinfluenceofcouplingbetweenthefreeandreference
m
layerof theMTJs on theST-FMR signals and the derived torkancesare discussed.
.
t
a
m I. INTRODUCTION agreement with free electron models [10].
-
d
n
II. EXPERIMENTAL
o High density magnetic random access memories can
c be implemented using current-induced magnetization
[ switching (CIMS) [1] which is caused by interactions be- The MTJ stack with a MgO wedge tunnel barrier

tween spin-polarized current and the magnetization of was deposited in a Singulus Timaris cluster tool sys-
1
v the free layer (FL) in magnetic tunnel junction (MTJ) tem. The multilayer structure consisted of the follow-
6 cells. Thisphenomenoniscalledthespin-transfer-torque ing materials (thickness in nm): Ta(5) / CuN(50) /
8 (STT) effect [2, 3]. Moreover, STT is utilized in MTJ Ta(3) / CuN(50) / Ta(3) / PtMn (16) / Co70Fe30(2)
1 nano-oscillators that generate signals in the GHz fre- / Ru(0.9) / Co40Fe40B20(2.3) / wedge MgO(0.7 - 1.1)
7 quency range [4–6]. In order to optimize MTJ param- / Co40Fe40B20(2.3) / Ta(10) / CuN(30) / Ru(7). The
.
1 eters, so that they can compete with existing memory slopeoftheMgOwedgebarrierwasapproximately0.017
0 and microwave technologies, it is necessary to fully un- nm/cm. The deposition process was similar to the one
3 derstandSTT.Thespin-torquediodeeffectenablesquan- used in our previous studies [11, 12]. After thin-film de-
1
titative measurements of STT parameters [7–9]. In this position,threedifferentpartsofthesamplewereselected
:
v work, we use the spin-torque diode effect to investi- forpatterningintonanometersizepillars(laterinthepa-
i gate the dependence of in-plane and perpendicular spin per referredto as S1,S2andS3,see Table I fordetails).
X
torques on MgO tunnel barrier thickness. The tunnel Using a three-steps electron beam lithography process,
r
a barrierdeterminesthetransportpropertiesofthedevice, which included ion beam milling, lift-off and oxide and
as it affects the tunneling magnetoresistance (TMR) ra- conducting layers deposition steps, nanopillars with an
tio, the resistance area (RA) product and the coupling elliptical cross-section of 250 × 150 nm were fabricated.
between the FL and the reference layer (RL). We show The pillars were etched to the PtMn layer. The elec-
that the spin-torque ferromagneticresonance(ST-FMR) tric leads to each MTJ nanopillar consisted of coplanar
spectra contain a double resonance mode for very thin waveguideswhich were designed to match an impedance
MgObarriersduetostrongferromagneticinterlayercou- of 50 Ohms. To ensure good RF performance, the over-
pling. Moreover, the in-plane and perpendicular spin- lap between the top and bottom leads was about 4 µm2,
torques do not depend on MgO barrier thickness, in which resulted in a capacitance of less than 1 × 10−14
F.EachsetofMTJswithaconstantMgOtunnelbarrier
consisted of 10 - 15 nanopillars.
ST-FMR measurement were conducted in a frequency
∗Electronicaddress: [email protected] rangefrom2to12GHz. Intheseexperiments,theappli-
2
TABLEI:SummaryofstaticparametersofthepreparedMTJ
nanopillars.
Sample No. MgO thickness TMR RA product Hs 150
(nm) (%) (Ωµm2) (Oe) %) S1
( S2
S1 1.01 170 9.6 -21.7 R 100 S3
M
S2 0.95 165 6.24 -3.7
T
S3 0.76 110 2.86 47
50
cationofanRFcurrenttoanMTJgeneratedaDCvolt-
age (also called mixing voltage V ) across the device, 0
mix
when the current frequency was brought into resonance
-200 0 200
withtheresistanceoscillationsarisingfromtheSTT.The
Field (Oe)
MTJswereplacedinanin-planemagneticfieldatanan-
gle of β = 70◦ with respect to the easy magnetization FIG. 1: TMR vs. magnetic field loops of samples S1-S3.
axis (except for the case presented in Fig. 3(b)), so
that a large variety of angles θ between the junction’s
FL and RL could be obtained. We estimated θ from the
assumption,thattheresistanceRoftheMTJchangesas
follows:
12
a) S1 External magnetic field (Oe)
cos(θ)=(cid:18)RAP2+RP −R(cid:19)(cid:18)RAP2−RP(cid:19) (1) 180 655050000
450
where RAP and RP are the resistance of the MTJ for 6 400

an antiparallel and parallel alignment of the FL and RL 350
magnetization,respectively. Inordertoobtainthe clear- 4 300
250
estSTT results [13], the strengthand angleof the exter- 2
200
nal magnetic field was adjusted so that magnetizationof 150
0
the FL is perpendicular to the magnetization of the RL
(θ = 90◦). The magnitude of the RF input signal, con- 50 b) S2

nected to the MTJ through the capacitive lead of a bias ) 700
V 40 650
tee, wasfixed to -15dBm. This resultedin a RFcurrent
( 600
(IRF)between5µAand25µA,dependingonthesample mix 30 550
resistance. IRF was calculated on the basis of the non- V 500
resonant background signal, using a model proposed in 20 450
400
Ref. [8]. The bias voltage was fed through the inductive
10 350
lead of the bias tee. Vmix was measuredusing a AC cou- 300
pled lock-in amplifier, which was synchronized with the 0 250
amplitude modulated signal from the RF generator. In
this paper, positive bias voltageindicates electrontrans- 20 c) S3 550
port from the bottom RL to the top FL. 500
15 450
400
III. RESULTS AND DISCUSSION 10 350
300
250
5
Table IsummarizestheTMR,theRAproductandthe 200
static offset magnetic field (HS) for three sets of MTJs 150
0 100
with different MgO tunnel barrier thickness. The rep-
resentative TMR vs. magnetic field loops are presented 3 4 5 6 7
in Fig. 1. The high TMR ratio of 170% for a 1.01 nm Frequecy (GHz)
thickbarrierandtheexponentialdecreaseinRAproduct
FIG.2: ST-FMRspectraofsamplesS1(a),S2(b)andS3(c)
withdecreasingMgOthicknessconfirmgoodtunnelbar- measured with various magnetic field applied at an angle of
rier quality [11]. Similar TMR ratios and RA products β = 70◦ with respect to the easy magnetization axis. Only
were measured on full wafers using a current in-plane theRFsignal (without DC bias voltage) was supplied tothe
tunnelling (CIPT) technique before patterning [12]. The MTJ. Forsample S3(c) two closely spaced peaksare visible.
overall offset field (HS) is shifted approximately 30 - 40
3
inates from magnetization precession in the FL [14]. A
similar behavior is observed for sample S1, wherein the
effectivecouplingbetweenFLandRLisweaklyantiferro-
M FL1 exp
20 a) = 70(cid:176) REF FL2 exp magnetic. However,forsampleS3,whichischaracterized
bystrongferromagneticcouplingbetweenFLandRL,an
FL sim
HEXT RL sim additional peak is measured. The origin of this double
15
resonancemodeisnotentirelyclear. Inpreviouspublica-
tions, it has been attributed to domain formation in the
) 10 FL[ 15],higher-orderspinwaveexcitations[16]andmag-
z
H netization precession in other layers of spin-valve MTJs
G 5 [17]. ToanalyzethedoubleresonancemodeinsampleS3
(
y in more detail, we performed macrospin simulations us-
nc 0 ing the model presented in Ref. [18]. This model, based
e
u on the Stoner-Wolfarth approach, assumes coherent ro-
Freq 20 b) = 30(cid:176) MREF FRLL eexxpp ttahteiosnysotfemtheenFeLrgaynwdeRfiLndmathgeneatnizgalteioonf.thBeyFmLinaimndizRinLg

H FL sim magnetizations with respect to the easy axis and on this
EXT
RL sim basis,wecalculatethedispersionrelation. Thesimulated
15
dispersionrelationsthatareobtainedforβ =70◦ andfor
β = 30◦ are presented in Fig. 3 together with the mea-
10 sure d ST-FMR spectra. For β = 30◦, the experimental
andsimulatedFMRmodesoftheFLandRLareingood
5
quantitativeagreement. We notethattheFMRsignalof
the RL is only measured when a large positive magnetic
0 field is applied to the nanopillar junctions. The reso-
nancefrequencyoftheRLdecreaseswithincreasingfield
-1.0 -0.5 0.0 0.5 1.0
strength in this field range. The frequency of the double
Field (kOe) resonance peak in the spectra for β = 70◦ (Fig. 3(a)),
on the other hand, increase with applied field strength.
FIG. 3: The dispersion relation of sample S3 measured with
the magnetic field applied at an angle of β = 70◦ (a) and β The experimentaldispersionrelationsnowcloselymatch
= 30◦ (b) with respect to the easy magnetization axis. The simulatedcurve. Basedonthisanalysis,weattributethe
solid anddashedlinesrepresentmacrospinsimulationsofthe doubleresonancemodetoinhomogeneousmagnetization
FL and RL, respectively. (a) At an angle of β = 70◦, the precession in the FL rather than FMR in the RL or any
resonance frequency of two slightly separated FL modes in- other magnetic layer of the MTJ stack.
crease with increasing magnetic field. (b) At an angle of β
◦ Tofurtherelucidatetheoriginofdouble-modeFLspec-
= 30 ,magnetization precessions in both theFL and RL are
tra, we simulated the resonance characteristics of MTJ
measured.
nanopillarsusingoommfsoftware[19]withanadditional
extension enabling calculations of TMR and STT effects
Oe with respect to the wafer-level measurements due to [20]. In these micromagnetic simulations, elliptical mul-
dipolarmagnetostaticstray-fieldcouplinginthenanopil- tilayersystemswith a 2 nm thick FL,a 1 nm thick MgO
lar junctions. For the MTJ with a 1.01 nm thick tunnel tunnel barrier,a2 nm thick high-anisotropyRL,antifer-
barrier,antiferromagneticstray-fieldcoupling dominates romagnetically coupled to a 2 nm thick exchange-biased
the interaction between FL and RL (HS = -21.7 Oe). pinned layer (PL), were used. The area of the junction
A reduction of the barrier thickness to 0.76 nm reverses was identical to the experimental structures. The inter-
the sign of the offset field (HS = 47 Oe). In this case, layerexchangecouplingandanisotropyenergieswereex-
the FL and RL couple ferromagnetically due to direct perimentallydeterminedbymagneticandmagnetotrans-
interactions across the thin MgO tunnel barrier. portmeasurements. Variationoftheferromagneticinter-
layer exchange coupling from 0 to 19 µJ/m2 in the sim-
ulations yielded results comparable to the experimental
A. ST-FMR data. WenotethatdipolarcouplingbetweentheFLand
RL is intrinsically calculated and taken into account in
oommf. Thusdependingonthestrengthoftheinterlayer
TypicalST-FMRsignals(withoutDCbiasvoltage)for
exchange coupling (input parameter), the effective cou-
samples S1 - S3 are presented in Fig. 2. We note that
pling between FL and RL varies from antiferromagnetic
a single symmetric peak is measured for sample S2 in
toferromagneticinaccordancewiththeexperimentalre-
a wide magnetic field range. For this sample, the cou-
sults on samples S1 - S3.
pling between FL and RL is negligible. Moreover, the
monotonic increase of the resonance frequency with ap- The dynamic simulations were conducted in the fol-
plied magnetic field indicates that the FMR signal orig- lowing way: first, an externalmagnetic field was applied
4
mental behavior of sample S3 with a 0.76 nm thin MgO
tunnel barrier. The simulations thus confirm that the
the double resonance mode originates from inhomoge-
2
a) S2, J = 6 J/m Anisotropy const. neous magnetization precession in the FL of the MTJ
2 H = 450 Oe 3 nanopillar stack due to strong interlayer exchange cou-
5 kJ/m
3 pling between FL and RL.
18 kJ/m
1
B. Torques and torkances
)
u.
a. 0 In order to obtain the STT components, i.e., in-plane
(
AC torque τk and perpendicular torque τ⊥, from the ST-
V FMRmeasurements,weusedthemodelpresentedinRef.
3 b) S3, J = 19 J/m2 Anisotropy const. [13]. Here, we assume a simplified formula for V :
3 mix
H = 450 Oe 10 kJ/m
3 1∂2V
2 18 kJ/m Vmix = 4 ∂I2 IR2F (2a)
1 ∂2V ¯hγsinθ
+ I2 [ξ S(ω)−ξ ΩA(ω)],
1 2∂I∂θ4eMSVolσ RF k ⊥
(2b)
0 where h¯ is the reduced Planck’s constant, γ is the gyro-
3 4 5 6 7
magneticratio,e istheelectroncharge,Vol isthevolume
Frequency (GHz)
ofthe FL,MS is thesaturationmagnetizationofthe FL,
σ is the linewidth, ξ = 2(e/¯hsinθ)(dV/dI)dτ /dV and
k k
FIG. 4: Simulated ST-FMR curves for weak (a) and strong ξ⊥ = 2(e/¯hsinθ)(dV/dI)dτ⊥/dV are the magnitudes of
(b) ferromagnetic interlayer exchange coupling. In oommf the symmetric S(ω)=[1+(ω-ωm)2/σ2]−1 and asymmet-
simulations, a voltage step was used to excite magnetization ric A(ω)=[(ω-ωm)/σ]S(ω) lorentzians components, and
precession in the FL of a MTJ structure. The dimensions of Ω⊥=γNxMeff/ωm, Nx=4π+(Hz-Hasin2β)/Meff, where
the simulated and experimental junctions are identical. The ωmistheresonantfrequency,Hz isthesumoftheapplied
existence of closely-spaced double-peak ST-FMR signal for
external magnetic field and the offset field acting on the
strong coupling is independentof the anisotropy constant.
precessing FL, Ha is the in-plane anisotropy field of the
FL and4πM is the effective out-of-planeanisotropyof
eff
the FL. We neglected the terms (2c) - (2g) of Ref. [13]
at an angle with respect to the magnetic easy axis. Af- because in our case θ ≈ 90◦.
ter relaxation,a voltagestepwasappliedto exertaSTT
Figure 5a presents a comparison of the in-plane
on the FL. The voltage step amplitude was adjusted, so
torkance in samples S1, S2, and S3. The absolute value
thatthe FLmagnetizationoscillationschangedthe MTJ
ofthein-planetorkanceincreaseswithdecreasingbarrier
resistance by a few Ohms. The used values correspond
thicknessanditonlyweaklydependsonDCbiasvoltage.
to an AC current of a few µA, which closely mimic the
According to Slonczewski’s free electron model for elas-
experimental conditions and ensures that the magneti-
tic tunneling in symmetric MTJs, the in-plane torkance
zation oscillations are within the linear regime. Finally,
is proportional to the differential conductance measured
the resonancespectra wereobtainedby Fouriertransfor-
for parallel alignment of FL and RL [21]:
mation of the time-derivative damped oscillation of the
simulated tunneling magnetoresistance.
Figure 4 presents the simulated ST-FMR spectra for dτk ¯h 2p dI
= (3)
two MTJ nanopillars that closely resemble experimental dV 2e1+p2 (cid:18)dV (cid:19)
k
samples S2 and S3. The simulations confirm that the
magnetization of the RL does not precess under these By using Jullieres model to derive the spin polariza-
conditions(β =70◦)intheinvestigatedfrequencyrange. tion of the tunneling current p at V = 0 V, we found a
For a weak interlayer exchange coupling energy of J = goodmatchbetween our experimentaldata and theoret-
6 µJ/m2 (sample S2), a single resonance peak is sim- icalcalculationsbasedonEq.3(Fig. 5(a)). Theabsolute
ulated for different FL anisotropy energies - Fig. 4a - torque values in Fig. 5(b) were obtained by numerical
and different magnetic field strength (not shown), which integration of the data in Fig. 5(a). Obviously, the in-
fulfills the Kittel dispersion relation. For a larger ferro- plane torque varies linearly with DC bias current and it
magnetic coupling energy of J = 19 µJ/m2 (sample S3), is independent of MgO tunnel barrier thickness. These
an additional broad resonance peak was resolved in the results are in good agreement with previously published
simulations (Fig. 4(b)), regardless of the FL magnetic experimental data in Refs [8, 9, 13, 22] and calculations
anisotropy. This behavior is reminiscent to the experi- based on an ab initio approach[23, 24].
5
were only measured in asymmetric MTJs with different
C) ) FLandRL electrodes. Inourjunctions,the composition
-19 0 15 a) b) 2 Nm and thickness of the CoFeB electrodes are the same
/dV(1 10 01 -19 (10 ||
d 5 -1
IV. SUMMARY
-2
0
)
C
-19 V (10 12 c) d) -00.0.1 -19 0 Nm) wtoirtIqnhuvesaudrmiieomddeaMreygff,Oewctteu.nWhnaeevlembaeinrarvsieuesrrteitdghaiactkesdnyemMssmTueJstirnincgaStnhToep-FsilpMlainrRs-
/d 0 SS12 (1 signal for samples with tMgO > 0.9 nm. In this case,
d -1 S3 comp. -0.2 the coupling between FL and RL is weakly antiferro-
-2 S3 raw magnetic. Contrary,doubleandclosely-spacedresonance
-0.4 -0.2 0.0 0.2 0.4 -1 0 1 modes were obtained for MTJs with a 0.76 nm thick
Voltage (V) Current (mA) tunnel barrier. Macrospin and micromagnetic simula-
tions indicate that the asymmetric double-peaks orig-
FIG. 5: Bias dependence of the in-plane torkance (a), in-
inate from inhomogeneous magnetization precession in
plane torque (b), perpendicular torkance (c) and perpendic-
the FL causedby ferromagneticcouplingto the RL.The
ular torque (d) for MTJs with different MgO barrier thick-
ness. The solid lines in (a) represent calculations based on in-plane and perpendicular torques scale with DC bias
Eq. 3. The torque values are numerically integrated from currentand they are independent of MgO tunnel barrier
experimentally determined torkances. τ⊥ for sample S3 was thickness.
compensated for an error originating from asymmetric ST-
FMR resonances.
Acknowledgement
Experimental data on the perpendicular torkance are
summarized in Fig. 5(c). For samples S1 and S2, the We would like to thank Singulus Technologies AG
torkance decreases with DC bias voltage and dτ /dV = for consultation and technical help with MgO wedge
⊥
0forzeroDCbiasvoltageaspredictedbytheoreticalcal- MTJs preparation. We also acknowledge help of Michal
culations. However,adiscrepancyisobservedforsample Wilczyn´ski, Piotr Ogrodnik and Renata S´wirkowicz for
S3. In this sample, strong ferromagnetic coupling be- a fruitful discussion regarding STT models. Project
tween the FL and RL of the MTJs results in asymmet- supported by the Polish National Science Center grant
rical double resonance modes in the ST-FMR spectra. 515544538, Polish Ministry of Science and Higher Ed-
The fitting procedure based on Eq. 2 therefore intro- ucation Diamond Grant DI2011001541 and Swiss Con-
duces an error in the experimental torkance values for tribution by NANOSPIN PSPB-045/2010 grant. W.S
this sample. A good match with theoretical calculations and T.S. acknowledges the Foundation for Polish Sci-
isobtainedwhenthisartifactiscompensatedbysubtrac- ence MPD Programme co-financed by the EU European
tion of a constant torkance value. Figure 5(d) illustrates Regional Development Fund. G.R. acknowledges sup-
that the absolute perpendicular torque varies quadrati- port from the DFG (contract RE 1052/21-1). G.S.P
callywithDCbiascurrent. Moreover,τ issimilarforall and K.C. acknowledge support from the Commission of
⊥
samples. Wenotethatdifferenttorqueversusbiasdepen- Higher Education of Thailand and Academy of Finland
dencies have been measured recently. Especially, it has (nos. 129896,118122,and 135135). S.v.D. acknowledges
been shown that the shape of τ (V) curves can change financial support from the Academy of Finland (grant
⊥
from quadratic to linear [25, 26]. However, such effects no. 127731).
[1] Huai, Y., Albert, F., Nguyen,P., Pakala, M., and Valet, 803 (2008).
T. Appl. Phys. Lett. 84, 3118 (2004). [6] Skowron´ski, W., Stobiecki, T., Wrona, J., Reiss, G., van
[2] Berger, L. Phys. Rev. B 54, 9353 (1996). Dijken, S. Appl. Phys. Express 5, 063005 (2012).
[3] Slonczewski, J. C. J. Magn. Magn. Mater. 159, L1 [7] Tulapurkar, A. A., Suzuki, Y., Fukushima, A., Kubota,
(1996). H., Maehara, H., Tsunekawa, K., Djayaprawira, D. D.,
[4] Petit,S.,Baraduc,C.,Thirion,C.,Ebels,U.,Liu,Y.,Li, Watanabe, N.,and Yuasa, S. Nature 438, 339 (2005).
M.,Wang,P.,andDieny,B. Phys. Rev. Lett.98,184420 [8] Sankey, J. C., Cui, Y., Sun, J. Z., Slonczewski, J. C.,
(2007). Buhrman, R. A., and Ralph, D. C. Nature Phys. 4, 67
[5] Deac, A. M., Fukushima, A., Kubota, H., Maehara, H., (2007).
Suzuki, Y., Yuasa, S., Nagamine, Y., Tsunekawa, K., [9] Kubota, H., Fukushima, A., Yakushiji, K., Nagahama,
Djayaprawira,D.D.,andWatanabe,N. Nature Phys.4, T., Yuasa, S., Ando, K., Maehara, H., Nagamine, Y.,
6
Tsunekawa,K.,Djayaprawira,D.D.,Watanabe,N.,and macher, H.W. J. Appl. Phys. 110, 023906 (2011).
Suzuki,Y. Nature Phys. 4, 37(2007). [19] Donahue, M. J. and Porter, D. G. Interagency Report
[10] Wilczyn´ski, M., Barna´s, J., and S´wirkowicz, R. Phys. NISTIR 6376 NIST,USA(1999).
Rev. B 77, 054434 (2008). [20] Frankowski, M., Czapkiewicz, M.
[11] Skowron´ski, W., Stobiecki, T., Wrona, J., Rott, K., http://www.layer.uci.agh.edu.pl/M.Frankowski
Thomas,A.,Reiss,G.,andvanDijken,S. J.Appl.Phys. [21] Slonczewski,J.C.andSun,J.Z. J.Magn.Magn.Mater.
107, 093917 (2010). 310, 169 (2007).
[12] Wrona, J., Langer, J., Ocker, B., Maass, W., Kanak, J., [22] Wang, Chen and Cui, Yong-Tao and Katine, Jordan A.
Stobiecki,T., and Powronik, W. InJ. Phys.: Conf. Ser. and Buhrman, Robert A. and Ralph, Daniel C. Nature
200, 052032, (2010). Phys. 7, 496 (2011).
[13] Wang,C.,Cui,Y.,Sun,J.,Katine,J.,Buhrman,R.,and [23] Heiliger, C. and Stiles, M. Phys. Rev. Lett. 100, 186805
Ralph,D. Phys. Rev. B 79, 224416 (2009). (2008).
[14] Cornelissen,S.,Bianchini,L.,Devolder,T.,Kim,J.,Roy, [24] Jia, X., Xia, K., Ke, Y., and Guo, H. Phys. Rev. B 84,
W. V., Lagae, L., and Chappert, C. Phys. Rev. B 81, 014401 (2011).
144408 (2010). [25] Chanthbouala,A.,Matsumoto,R.,Grollier,J.,Cros,V.,
[15] Lee,K.,Deac,A.,Redon,O.,Nozi´eres,J.,andDieny,B. Anane,A.,Fert,A.,Khvalkovskiy,A.V.,Zvezdin,K.A.,
Nature Mater. 3, 877 (2004). Nishimura, K., Nagamine, Y., Maehara, H., Tsunekawa,
[16] Helmer, A., Cornelissen, S., Devolder, T., Kim, J. V., K., Fukushima, A., and Yuasa, S. Nature Phys. 7, 626
van Roy,W., Lagae, L., and Chappert, C. Phys. Rev. B (2011).
81, 94416 (2010). [26] Matsumoto,R.,Chanthbouala,A.,Grollier,J.,Cros,V.,
[17] Sankey, J., Braganca, P., Garcia, A., Krivorotov, I., Fert, A., Nishimura, K., Nagamine, Y., Maehara, H.,
Buhrman,R.,andRalph,D. Phys.Rev.Lett.96,227601 Tsunekawa, K., Fukushima, A., and Yuasa, S. Appl.
(2006). Phys. Express 4, 063001 (2011).
[18] Serrano-Guisan, S.,Skowron´ski, W., Wrona,J., Liebing,
N., Czapkiewicz, M., Stobiecki, T., Reiss, G., and Schu-