Spin  pumping  and  spin  transport  in  magne0c  metal  and  insulator  

heterostructures    Eric  Montoya  

Surface  Science  Laboratory  Simon  Fraser  University  

Fe  

Au  

Fe  Au  

YIG  

Au  

Fe  Au  

Magne0c  metal   Magne0c  insulator  

Why  use  spin  currents?    We  can  eliminate  circumvent  these  problems:  •  Joule  Hea0ng  •  Circuit  Capacitance  •  Electron  migra0on  

Outline:  •  Introduce  spin  pumping  •  Spin  transport  in  Au  •  Spin  pumping  from  magne0c  insulator  (if  0me)  

Spin Dynamics (LLG) Landau Lifshitz Gilbert

γω

α=ΔH

Ferromagnetic Resonance (FMR)

FMR linewidth:

sMGγ

α =

Anritsu signal generator: 1-70 GHz electromagnet: 2.8 T

Bulk damping is noise in spin orbit (s-o) interaction Interface damping is spin pumping

Heff

M !

36GHz  

∂

�M

∂t= −γ

��M ×

�Heff

�+ α

��M × ∂

�n

∂t

��n =

�M

Ms

NMFM1 FM2

0 Lx

�I �m

�mNM

�I �m

=−gµB4πMs

Re [g↑↓]

��M × ∂n̂

∂t

�⊗ x̂ ∂

�M

∂t=

1

dFM

∂�m

∂t

�m =

−gµB�

�S

Re [g↑↓] =1

2

�

n

[|r↑,n − r↓,n|2 + |t↑,n − t↓,n|2]

|r↑(↓),n|2 + |t↑(↓),n|2 = 1 r↑,n∗

r↓,n + t↑,n∗

t↓,n � 0

g↑↓ =�

n

[1− Re[r↑,n∗

r↓,n + t↑,n∗

t↓,n��

≈ kF2

4π∼ N2/3

Spin  Pumping  

Spin  mixing  conductance  for  metals  

Density  of  electrons  per  spin  direc0on  in  NM  

Spin  current  arises  from  the  0me  retarded  response  to  interlayer  exchange  coupling  The  spin  current  can  be  expressed  as  an  accumulated  magne0c  moment  (/area)  

NMFM1 FM2

0 Lx

�I �m

�mNM

Spin  Pumping  

�I �m

=−gµB4πMs

Re [g↑↓]

��M × ∂n̂

∂t

�⊗ x̂ ∂

�mNM∂t

= D∂2

�mNM∂x2

− 1τsf

�mNM

D =vF 2τm

3

Boundary FM1/NM FM1/NM/FM2

x = 0�I �m− 12vF

�mNM = −D ∂

�mNM∂x

x = L ∂�mNM∂x = 0 −D

∂�mNM∂x =

12vF

�mNM

The  accumulated  magne0c  moment  then  diffuses  away  from  then  interface  

δsd = vF

�τsfτm3

0 100 200 300 400 500 6000

3

6

9

dAu !AL"

Α10"3s"

!sf  is  only  unknown  parameter  sf  is  only  unknown  parameter  

Spin  Pumping  Theory  

Fsd=F(!sf,  !m,  dNM,  vF)×Fb  sf,  !m,  dNM,  vF)×Fb  Fb=F(dFM,  4"MS,  g⇅,  g,  #B)  

•  GaAs(001)  Template  –  Atomic  H  etching  –  650eV  Ar+  spu^ering  (con0nuous  rota0on)  

–  4x6  reconstruc0on  RHEED  monitored  

•  16Fe  and  12Fe  have  different  FMR  fields  

•  At  16Fe  FMR,  16Fe  acts  as  spin  pump  and  12Fe  acts  as  spin  sink   GaAs(001)!

16Fe (2.3nm)!

12Fe (1.7nm)!20Au (4.1nm)!

300Au (61.2nm)!or!

20Au (4.08nm)!

Single  Layers  

Double  Layers  

Sample  Growth  by  MBE  

•  Van  der  Pauw  measurements  –  10K-‐300K  

•  Contribu0on  due  to  bulk  phonon  and  interface  sca^ering  

•  Using  Mathiassen’s  Rule:  

   •  Interface  sca^ering  

contribu0on  independent  of  temperature  

1

τm=

1

τp+

1

τi

τm =σmene2

!

R88295 = 2.5

!

R10295 = 6.1

Charge  Transport  

Ferromagne0c  Resonance  

•  FMR  followed  Gilbert  damping  phenomenology:  

•  Enhanced  Gilbert  damping  due  to  spin  pumping  is  an  interface  effect  

•  Spin  momentum  accumulates  at  the  Fe/Au  interface  

 

∆H(ω) = αω

γ+∆H(0)

��

��

��

��

�� �

�

��

��

�� �

�� �

��

�

�

��

�

�

0 10 20 30 400

50

100

150

f �GHz�

�H�Oe�

Gilbert  Damping  

•  $sp  greatest  in  ballis0c  limit  for  sp  greatest  in  ballis0c  limit  for  double  layer  

•  $sp  increases  with  decreasing  sp  increases  with  decreasing  temperature  for  double  layers  

•  $sp  decreases  with  decreasing  sp  decreases  with  decreasing  temperature  for  single  layers  

100 150 200 250 3000

3

6

9

Temperature !K"

Α!10"3 "

20Au  DL  

20Au  SL  

300Au  SL  

300Au  DL  

Relaxa0on  parameters  

0 50 100 150 200 2500.0

0.5

1.0

1.5

2.0

2.5

Temperature �K�

Τ�10�13

s�

τsf(290K)

τp(290K)= 8.4

τsf(90K)

τp(90K)= 22.1

•  !sf  increases  faster  than  !p  sf  increases  faster  than  !p  as  temperature  decreases  

•  !i  very  weakly  dependent  on  temperature  

Spin  flip  sca^ering  dominated  by  phonon  processes  

Combined  influence  of  temperature  dependent  spin  flip  sca^ering  at  interfaces  and  bulk  phonon  sca^ering?  

or  Mul0-‐phonon  sca^ering  that  does  not  contribute  strongly  to  

resis0vity?  

Previous  Studies  

three fitting parameters remained: S↑ ,S↓, and Pvac/Au. Unfor-tunately a unique fit to the data was not achieved with thedata in Fig. 6 alone.

The magnetoresistance data from Au/Fe/Au/Fe/GaAs!001" was used to help determine the specularity pa-rameters. A necessary simplifying assumption that all Fe/Auinterfaces had the same degree of specularity !S↑ and S↓"reduced the number of fitting parameters. The reflectivity atthe outer Au interface Pvac/Au was first set to zero, and thenall possible combinations of S↑ and S↓ that gave the correctconductivity for a parallel and antiparallel configuration inthe Au/Fe/Au/Fe/GaAs sample were determined, as shownby the curves joining the two sets of triangles in Fig. 7. Thepoints where these curves intersected gave the correct mag-netoresistance. The calculation was repeated for Pvac/Au=0.25 and 0.50. The points of intersection for the variousPvac/Au lay roughly on straight lines, shown by the twodashed lines in Fig. 7. The specularity parameters decreasedapproximately linearly with increasing Pvac/Au. The GMRdata alone indicated that the reflection from the outer Auinterface was mostly diffuse, Pvac/Au!0.5, and one spinchannel had a specularity in the range 0.6"S"0.8.

Fits to Au conductivity data in Fig. 6 were subsequentlyconstrained by the requirement that the parameters P ,S↑ andS↓ produce the correct magnetoresistance for Au/Fe/Au/Fe!lines A and B in Fig. 7". P=0 gave the best fit to the Auconductivity. The solid line in Fig. 6 represents the fit to thedata using either !P=0.0, S↑=0.55, and S↓=0.77" or !P=0.0, S↑=0.83, and S↓=0.53". When P was increased, the #2also increased. The dashed line in Fig. 6 represents !P=0.41, S↑=0.03, and S↓=0.65", the fit having the highest #2!satisfying the constraint of line A in Fig. 7". The results ofthe fits are displayed in Table II. A lower $Au would haveresulted in larger specularity parameters.

The diffuse scattering modeled by S↑ and S↓ describe theinfluence of interface imperfections. Unfortunately, no calcu-

lations or measurements were available for the spin asymme-try of Au impurities in Fe for comparison to the fits. It was,however, interesting to compare the spin asymmetry of diluteCu and Ag impurities in Fe since these elements were iso-electronic with Au and therefore were expected to scatter in asimilar fashion. Mertig recently calculated a spin asymmetry%↓ /%↑=8.20 for Cu and 12.22 for Ag.25 This would suggestthat Au defects at the Fe/Au interface would also diffuselyscatter minority electrons more strongly. Based on this argu-ment, one tends to favor the parameters from line A.

The fit of the Au thickness with first-principles calculationdescribed the data well for large thicknesses, but failed todescribe the conductivity for small thicknesses, in particularthe 5 ML Au film. One had expected a drop in the conduc-

FIG. 6. In situ measurement of the conductivity as a function ofAu thickness deposited on 28 ML Fe/GaAs!001". The two curvesare calculated using first-principles density functional calculations.The solid line !P=0, S↑=0.55, S↓=0.77" is the best fit given the setof parameters described by line A in Fig. 7 below. The dashed line!P=0.41, S↑=0.03, S↓=0.65" has the highest #2 among the param-eters on line A, which demonstrates the sensitivity of the fit to thefitting parameters.

FIG. 7. Fitting of the conductivity of 20 ML Au/10 MLFe/7 ML Au/28 ML Fe/GaAs!001" using first-principles calcula-tions. The filled points correspond to the parameters which give thecorrect sheet resistance for a parallel configuration of magnetic mo-ments !resistance at saturation in Fig. 2" and the open points givethe correct sheet resistance for an antiparallel !zero applied field"configuration. The triangles, diamonds, and squares are calculatedfor Pvac/Au=0, 0.25, and 0.5, respectively. The dashed lines labeledA and B are interpolations of the points of intersection which givethe correct GMR.

TABLE II. The GMR fitting parameters determined from Fig. 7that gave the best fit to the thickness dependence of the Au conduc-tivity shown in Fig. 6. The confidence interval column !CI" shownfor each set of solutions A and B was the range of the specularityparameters within the 90% confidence interval of the fit. The 90%confidence interval is a region in the parameter space of #2 where90% of experiments will be fitted by a set of parameters fallingwithin that region.

Best fit !A" CI !A" Best fit !B" CI !B"

P 0.0 0.0–0.16 0.0 0.0–0.20S↑ 0.55 0.55–0.34 0.83 0.83–0.74S↓ 0.77 0.77–0.72 0.53 0.53–0.32

MONCHESKY et al. PHYSICAL REVIEW B 71, 214440 !2005"

214440-8

single magnetic layer structure F/NM the boundary condi-tions at the F/NM interface are5

jm − 0.5vFmNM = − D!mNM

!x. !5"

For the outer interface we used a free magnetic moment con-dition

!mNM!x

= 0. !6"

For a magnetic double layer structure F1/NM/F2 the bound-ary conditions at the F1/NM interface are equivalent to Eq.!5". The boundary conditions at the NM/F2 interface are5

−!mNM

!x= 0.5vFmNM. !7"

The boundary conditions in Eq. !7" are valid for the casewhen the layer F2 is off resonance and therefore contributesnegligibly to spin pumping. The coefficient 0.5 correspondsto the effective transmission coefficient from the NM to Flayers and is given by Eq. !13" in Ref. 5. The right hand sideof Eq. !7" represents the magnetic current from NM into F2and acts as a driving field for the magnetic moment in F2.

The purpose of the studies presented in this paper was toidentify the spin diffusion coefficient and spin flip relaxationtime in Au. We carried out two experiments. D and !sf weredetermined by FMR employing a single magnetic structureF/NM. Similar experiments were done by Mizukami et al.6

on the Cu /permalloy /Cu /Pt films. In addition the propaga-tion of spin current in NM was investigated by time andspatial resolved Kerr effect technique. For this case we useda double magnetic layer F1/NM/F2 where F1 was used forspin pumping and the layer F2 was used as a detector of thespin current.

The Fe films were deposited at room temperature on acommonly used 4"6-GaAs!001" reconstructed template.The 4"6 surface reconstruction was obtained by annealingthe GaAs wafer at #600 °C following hydrogen cleaningand Ar+ sputtering at 650 eV. The following structures weregrown: !a" nAu /16Fe /GaAs!001", where n=20, 80, 150,200, 250, 300 and the integers represent the number ofatomic layers; and !b" 20Au /12Fe /300Au /16Fe /GaAs!001"and 20Au /12Fe /300Ag /16Fe /GaAs!001".

The FMR studies were carried out using standard micro-wave spectrometers using 10, 24, 36, and 73 GHz, see de-tails in Ref. 7. For both bulk and interface Gilbert damping#H is strictly linearly dependent on the microwave angularfrequency $, #H=%!$ /&".

The NM layer increases magnetic damping when itsthickness becomes comparable to the spin diffusion length'sd=vF!!sf!el /3"0.5. For dNM('sd#H is given only by theintrinsic Gilbert damping of the Fe layer. For dNM)'sd the#H increases by the loss of spin momentum in NM. Theequations of motion !1" and !4" with the boundary conditions!5" and !6" were solved self-consistently, and were employedfor fitting the measured spin pumping coefficient %, see Fig.1. The spin pumping Gilbert damping parameter, %sp, as afunction of the Au thickness was fitted with the following

parameters: %intr=3.5"10−3, g̃↑↓=2.4"1015 cm−2, !el=1.2"10−14 s, !sf=15"10−14 s, and the Fermi velocity was as-sumed to be *F=1.4"108 cm /s. The fitted parameters resultin the spin diffusion length 'sd of 34 nm. It is interesting tonote that Kurt et al.8 studied the spin diffusion length byusing current–perpendicular-to-plane !CPP" giant magnetore-sistance !GMR" measurements using polycrystalline Au /Cuspacers. They obtained 'sd=35 nm, which is very close toour result. The ratio r=12.5 indicates that in our samples !sfis one order of magnitude larger than !el.

20Au /12Fe /nAu /16Fe /GaAs!001" structures were em-ployed in the study of propagation of spin currents across theNM film. Time-resolved magneto-optical Kerr effect!TRMOKE" measurements are an ideal tool for investigatingthe propagation of spin currents in these structures. Strobo-scopic measurements of magnetization precession in the10 GHz frequency range were carried out with picosecondtime resolution and submicrometer spatial resolution, using acoplanar transmission line carrying repetitive picosecondmagnetic excitation pulses. After excitation the 100 fs dura-tion laser pulses probed the top 12Fe layer via the perpen-dicular component of precessing magnetization !polarMOKE" at the delay time tD, see detailed description in Ref.9. Spin currents generated by the bottom 16Fe layer propa-gated across the normal metal spacer and resulted in rf exci-tations of the top 12Fe film. The resonant frequencies of theFe layers are strongly affected by the interface anisotropies,see Ref. 10. Therefore the 12Fe and 16Fe films have theirresonant frequencies 4.5 GHz apart and therefore the spincurrent induced magnetization precession in the 12Fe filmcan be in principle easily distinguished. However, the iden-tification of absorbed spin current is complicated by the pres-ence of a direct TRMOKE signal from the bottom 16Fe layerwhich becomes observable when the spacer thickness is lessthan 250 atomic layers. A Au spacer with the thickness of300 atomic layers was sufficient to suppress the signal fromthe bottom 16Fe film. No measurable MOKE signal was ob-served on the 300Au /16Fe /GaAs!001" sample. Therefore,further studies with the Au spacer were carried out using the20Au /12Fe /300Au /16Fe /GaAs!001" structure. The timedependence of the picosecond resolved Kerr signal and itsfast Fourier transform !FFT" are shown in Figs. 2!a" and

FIG. 1. Dependence of the additional damping by spin pumping, %sp, on theAu cap layer thickness dAu in the Au /16Fe /GaAs!001" samples. The !•"symbols represent the measured data from the #H dependence on micro-wave frequency f , #H!f". #H!f" followed well a linear dependence on f .The error bars were determined from small slope variations in the #H!f"measurements. The solid line shows fitting using the spin pumping theorywith the following parameters.: g̃↑↓=2.4"1015 cm−2, !el=1.2"10−14 s, and!sf=15"10−14 s.

07C509-2 Kardasz et al. J. Appl. Phys. 103, 07C509 "2008#

Downloaded 19 Apr 2011 to 142.58.93.159. Redistribution subject to AIP license or copyright; see http://jap.aip.org/about/rights_and_permissions

T.  Monchesky  et.  al.  Phys.Rev.B.  71  (2005)   B.  Kardasz  et.  al.  J.Appl.Phys.  103  (2008)  

From  80-‐5nm  thickness  of  Au  !i  increases  by  a  factor  of  12      !sf  only  increases  by  a  factor  of  1.5  sf  only  increases  by  a  factor  of  1.5  

!sf  can  only  weakly  be  dependent  on  interface  sca^ering  sf  can  only  weakly  be  dependent  on  interface  sca^ering  

Temperature  dependence  of  !sf  governed  by  mul?-‐phonon  scaAering    sf  governed  by  mul?-‐phonon  scaAering    

Spin pumping at YIG/Au interface

recently new ideas and systems being developed for generation of pure spin currents for driving Spin Transfer Torque (STT) devices

John Slonczewski has shown higher spin efficiency can be achieved by thermal gradients

using Magnetic Insulator (MI)/NM heterostructures

new emerging field spincoloritronics

Arne Brataas and Gerrit Bauer have shown that the spin pumping generation is determined at MI/NM interfaces by spin mixing conductance

?????what is at the YIG/Au interface ????

!

g"#

J. Slonczewski, PRB 82, 054403 (2010)

!

g"# =

1

2| r

n

" $ rn

#|2 + | t

n

" $ tn

#|2( )

n

%

!

tn

"#= 0

!

rn

"#=1$ e

i%n"#

g!" = 1# cos(!n! #!

n

")( )

n

$

B. Heinrich et al. PRL, 107, 066604 (2011) C. Burrowes et al. APL , 100, 092403 (2012)

Spin  mixing  conductance  in  magne0c  insulators  

XPS

YC

O Fe

YIG  surface  chemistry  

YIG:  Y3Fe2(FeO4)3    

•  Grown   on   (111)   Gd3Ga5O12  substrate   by   PLD   at   700C   and  0.1Torr  O2  

•  Thickness  d=9nm  (low  angle  XRD)  •  4"Ms=1.31kG(SQUID),   g=2.027  Ms=1.31kG(SQUID),   g=2.027  

(FMR)  •  Surface  roughness  0.5nm  (AFM)    

As  prepared  YIG  has    surface  deficiency  of  Fe  

 Common  for  even  thick  

PLD  prepared  YIG  

!!

!

!

0 10 20 30 400

10

20

30

40

50

60

frequency !GHz"

!H!Oe"

7.3 7.6 7.9

H !kOe"

Amplitude!a.u."

#a$

7.3 7.6 7.9

H !kOe"Amplitude!a.u."

#b$9nmYIG 9nm YIG/6.1nm Au/4.3nm Fe/6.1nm Au

g!" #1.3$1014cm

%2

12%  efficiency  compared  to  Fe  

Spin  pumping  from  YIG  

∆H = 15.9Oe ∆H = 21.2Oe Transfer  of  angular  momentum  to  Fe  layer  is  seen  as    

loss  of  angular  momentum  in  YIG  layer  

Damping!  

Evalua0ng                in  Ar+  etched  YIG  

�� ��

�

�

�

�

0 10 20 30 400

20

40

60

80

100

120

f �GHz�

�H�Oe�

9YIG(etched)/6.1Au

9YIG(etched)/6.1Au/4.3Fe/6.1Au  

! = 0.0069

! = 0.0014

g!" = 5.1#1014cm

$2

g!"

Low angle Ar+ etched 10minutes

0.8kV, 400o C

70% of predicted by first principles calculations

X. Jia et al. Europhysics Letters 96, 17005 (2011)

50% of Fe/Au

!"#$"#%&'#()%*&+(,-&

./01.&2(&

34&5&& 34&5&

./01.&2(&678&

698&

6:8&

6$8&

metallic state of Fe decreases rapidly g!" # 3.7$10

14cm

%2

required change for increased g!"

XPS on YIG

740 735 730 725 720 715 710 705 700

XPS

signa

l

Eb [eV]

Prolonged  etching  lead  to  metallic  Fe  -‐>  Suppresses  pumping  

Conclusions:

spin pump/sink effect can be used to investigate the spin transport parameters in magnetic nanostructures

spin pumping at YIG/Au is efficient 70% of theory calc. 50% of Fe/Au

evidence that a time retarded interlayer exchange coupling creates spin pumping

Efficiency of spin pumping comparison Microwave driven: for f=10GHz and Θ=90o 2x1010 Thermal excitation: for ΔT=10 K , Vcoh=2.7x103 nm3 ωeff=2x102 MHz 1.0x108 STT (60% polarization): for 2x106 Acm-2 : 2x1010

!

!

4!"g!"sin

2# !

4!g!"!

2kBTYIG

m #TAu( )

VcohM

s

$

%

&&

'

(

))

!eff = "2kB TYIG

m !TAu( )VcohM

s

"

#

$$

%

&

''

J. Xiao et al. PRB 81, 214418 (2010)

nm-2

Acknowledgements: NSERC, CIfAR , NSF, NIST

Simon Fraser University Physics, Magnetism

Prof. Mingzhong Wu, Dr.Young-Yeal Song, and Dr. Yiyan Sun

Physics Department Colorado State University

Fort Collins, USA

Ken Myrtle Research Assistant

Dr. Bartek Kardasz research associate

Dr. Erol Girt Professor

Dr. Capucine Burrowes PDF

Dr.  Bret  Heinrich  Prof.  Emeritus  

artist: Mr. Clarence Mills Charles Eyrich MSc Candidate

Dr. Monika Aurora PhD Candidate

Eric Montoya PhD Candidate

Dr. Wendell Huttema PDF

Conclusions  Temperature  dependence  of    !sf  sf  

 from  290K  to  90K  

!sf  increases  by  factor  of  10  sf  increases  by  factor  of  10  !p  increases  by  factor  of  4  !i  negligible  dependence  

   

Temperature  dependence  of  !sf  governed  by  mul0-‐phonon  sca^ering    sf  governed  by  mul0-‐phonon  sca^ering    

Thickness  dependence  of    !sf  sf    

from  80nm  to  5nm  !sf  increases  by  factor  of  1.5  sf  increases  by  factor  of  1.5  

!p  constant  !i  increases  by  factor  of  12  i  increases  by  factor  of  12  

 

deposition of Fe on bare YIG results in metallic state of Fe

no spin pumping

YIG surface H atom etching showed in XPS a strong presenceof metallic state of Fe

no spin pumping

spin current blockade by metallic Fe

704 708 712 716

Binding Energy [eV]

704 708 712 716 720

Binding Energy [eV]

metallic state of Fe

1 MLFe deposited at 500o C

YIG state of Fe