 

2.:
:m f. k

all-Lu ”I

i. . .55... . s. a.
“if. hufﬁng?!
.1.

.,,.........9w%
If 11 1
if 5.3% -

A . .
....rd......i13.u

yum"...

 

(Mk)

A
K‘)

LA)

This is to certify that the

dissertation entitled

Studies of Current-Perpendicular—to Plane
(CPP) Magnetoresistanoe

presented by

Khalid Fatthi Eid

has been accepted towards fulﬁllment
of the requirements for

Ph ‘D ‘ degree in Bhysics

JbWW

J . Bass, Major professor

S.D. Mahanti, Associate

Chairperson for Graduate
Date 08/15 /02 Studies

M5 U i: an Afﬁrmative Action/Equal Opportunity Institution 0-12771

 

 

 

LIBRARY

Michigan State

University

 

 

PLACE IN RETURN Box to remove this checkout from your record.
To AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE

DATE DUE

DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

6/01 cJCIRC/DaieDuepGSp. 15

 

 

 

STUDIES OF CURRENT-PERPENDICULAR—TO-

PLANE (CPP) MAGNETORESISTAN CE
By

Khalid F atthi Eid

A DISSERTATION
Submitted to
Michigan State University
in partial fulﬁllment of the requirements
for the degree of
DOCTOR OF PHILOSOPHY
Department of Physics and Astronomy

2002

ABSTRACT
STUDIES OF CURRENT-PERPENDICULAR-TO-PLANE (CPP)
MMAGNETORESISTANCE
By

Khalid F atthi Eid

Scientiﬁcally, measurements of Giant Magnetoresistance (GMR) in ferromagnetic/non-
magnetic (FIN) metal multilayers with Current ﬂow Perpendicular to the layer Planes
(CPP-MR) are of interest because they can provide unique insights into the physical
phenomena underlying GMR. Technologically, the CPP-MR is of growing interest
because it may be able to provide the larger MRS needed for next generation devices.
Most CPP—MR data have been reasonably well described by two simple models, the two-
current series-resistor (2CSR) model, in which the only lengths are the thicknesses of the

F- and N—metals, t1: and tn, and the Valet-Fen extension to when spin-ﬂipping occurs,
where the characteristic lengths are the spin-ﬂipping (or spin-diffusion) lengths lip and
1?; . This thesis consists of several experiments that test and/or use these models.

(1) Recently, new data were claimed to be evidence that there are additional length
scales in the CPP-MR, the mean-free-paths for elastic scattering, AN and M: (actually two
lengths in the F-metal). After reproducing the reported results, we extended them in three
different ways to test the claimed interpretation. We conclude that the mean-free-paths

are not responsible for the behavior of the data; rather we propose spin-memory-loss at

the F/N interfaces as the primary explanation.

(2) The CPP-MR would be more competitive for devices if the total speciﬁc
resistance, area times resistance, AR, the change in AR with magnetic ﬁeld, AAR, or both
can be increased. We investigated three potential ways to increase them: (a) by alloying
the F metal layer with an impurity expected to produce a large change in resistivity
without decreasing the anisotropic scattering that is fundamental to GMR, (b) by creating
additional interfaces within individual ferromagnetic layers, and (c) by inserting a source
of strong spin-ﬂipping, in hopes of reducing the contribution of contact or lead
resistances to AAR.

(3) Ruthenium (Ru) is now of great interest in GMR because Ru layer inserts as thin
as 0.6 nm can ‘pin’ the magnetizations of thin Co layers anti-parallel to each other,
forming an ‘artiﬁcial anti-ferromagnet’. For CPP-MR use, it is necessary to know
various properties of Ru and of Ru-based interfaces. We present a study of the CPP-MR
properties of Ru and its interfaces with Co and Cu. The most important results for
devices are that the resistivity of sputtered Ru is small enough and the spin-ﬂipping
distance (spin-diffusion length) is large enough so that they don’t constitute problems for
device use.

(4) We also use Ru to provide a sensitive new way to study spin-asymmetric
scattering at some interfaces of interest.

(5) Lastly, there is now great interest, both for basic physics and for devices, in
studying the CPP-MR and current-driven excitations in submicron-size multilayer pillars.
We describe a method for producing such pillars by combining optical and electron beam

lithographies, and present a few measurements on samples we have prepared this way.

COPYRIGHT

KHALID FATTHI EID

2002

iv

c5335! Jg‘i 345“ cr‘.‘

 

ACKNOWLEDGEMENTS

I would like to thank my thesis advisor Professor Jack Bass for his
encouragement, criticism, and guidance. He always had the time to patiently
discuss my thesis research with me and give advice. I also thank Professor
William P. Pratt Jr. for useful discussions and insightful evaluation of research
ideas, and Professor Norman Birge for encouragement and valuable discussions.

I would like to thank Professor Kamal Rashid from En-Najah University for
helping me apply to the graduate school and encouraging me during my studies.

I also wish to thank Dr. Reza Loloee for teaching and helping me in the lab.
Reza’s friendship and support made the lab a pleasant place to do research.

I would also like to thank Dr. B. Bi and Dr. J. Caballero for teaching me many
of the lithography skills and Dr. R. Slater and N. Verhanovitz for helping me with
the equipment and the programming. I also thank the people I worked with in the
labs: M. DaMish, A. Zambano, H. Kurt, M. Crosser and S. Urazhdin.

Finally, I would like to thank my friends in Lansing for friendship and for
making my life in Michigan State a memorable experience. They include M.
Hozain, M. AI Haj Darvvish, A. Shinnaq, Dr. A. Ahmad, J. Siryani, Dr. K.
Ghosheh, S. Arrnagan, Dr. N. Saglam, G. Mustafa, G. Al zoubi, S. Ahmad, ‘M &
M’ AI khasawneh, A. Masadeh, A. Abdo, M. Shatnawi, S. Berber, M. Chatal, and

M. Sayyaheen.

vi

TABLE OF CONTENTS

List of Tables

List of Figures

1

Introduction

1.1 Deﬁnitions and background

1.2 Basic ideas in GMR

1.3 GMR Device Applications

1.4 CPP-MR Theory & History Crucial to the Present Thesis
1.4.1 High Current-Density Effects in F/N Multilayers

1.5 ' The Present Thesis

Sample Preparation, Characterization, and Measurement
2.1 Introduction

2.2 Common Preparation Procedures
2.3 Characterizing Samples
2.4 Microscopic Samples
2.4.1 Preparation
2.4.2 Characterization
2.4.3 Measurement
Theoretical Background
3.1 Free-Electron-Based Semi-Classical Models
3.2 Real Fermi Surface Effects
3.3 High Current Density Effects on Magnetization

3.3.1 Spin-Transfer Switching

Search for Mean-Free-Path Effects in the CPP-MR

4.1 Introduction

4.2 AR (H) from the initial state to ﬁrst saturation

4.3 Magnetic Order

4.4 Magnetoresistance Tests of Mean-Free-Path Effects

4.5 Effect of Spin-Memory-Loss
4.6 Studies of CIP-MR in Co/Cu based multilayers and CPP_MR in
Co/Ag Based Multilayers

4.7 Summary and Conclusion

Methods for Enhancing the CPP-MR
5.1 Introduction
5.2 Enhancing the CPP-MR by Alloying Co with Zr

vii

ix

100

110
113

116
116
117

5.2.1 Giant Magnetoresitance of CoZr/Cu systems

5.2.2 Dusting CoZr/Cu Interface with Co

5.2.3 GMR in CoZr/Ag

5.2.4 Effect of F eMn growth on MR

5.2.5 Search for Magnetically dead layers

5.2.6 (3095er and C030Z1‘20
5.3 Internal Interfaces

5.3.1 Data and analysis

5.3.2 Incomplete Interfaces versus Interfacial Spin Memory Loss
5.4 Spin-Memory-Loss to enhance GMR

5.4.1 Data and Analysis

5.4.2 Fitting Data to the 2CSR Model

6 CPP-MR studies with Ruthenium (Ru)

6.1 Ru for CPP-MR

6.1.1 CPP-MR of Co/Ru
6.2 Studies of Spin-Asymmetries at Interfaces using Ru.

7 GMR in Multilayers with Submicron Area
8 Summary and Conclusions
Appendix

Bibliography

viii

120
130
133
136
140
143
147
151
156
166
167
172

177
178
181
192

204

209

211

217

 

 

2.1

2.2

2.3

2.4

2.5

2.6

5.1

5.2

LIST OF TABLES

VdP resistivity and superconducting transition temperature of sputtered
Nb. Tc(Nb) bulk z 9.2 K.

VdP Resistivity of CuGe Films.

Van der Pauw Resistivities of Co, Py, Cu and Ag ﬁlms. Last Column
shows if each sample was sputtered at the beginning or end of each
sputtering run.

Van der Pauw Resistivity of CoZr ﬁlms.

Average resistivities (in Man) both at 4.2 K and room temperature.
Ion-Milling rates of materials used in the lab.

CoZr parameters found from the ﬁts to our data.

The Composition of the samples used to study the effect of incomplete
interfaces and separated-like effect (due to the internal interfaces).

ix

34

35

35

36

37

50

126

159

1.1.

1.2.

1.3.

1.4

1.5.

1.6.

1.7.

2.1.

2.2.

LIST OF FIGURES

Cartoon of the s- and d- subbands of a ferromagnetic metal and a
normal metal. There is a net magnetic moment in the ferromagnet but not
the normal metal.

Schematic diagram of scattering in the ‘up’ and ‘down’ channels when the
magnetizations of the ferromagnetic layers are (A) antiparallel or (B)
parallel to each other. The star indicates stronger scattering. Scattering is
only shown in bulk F.

AR(H) for (A) an EBSV, and (B) a hybrid spin valve. AR(H) for the hybrid
SV is symmetric around H = 0, while that for the EBSV is not. Numbers
indicate layer thicknesses in nm.

Square root test of the 2CSR model. Adding Ge impurities that greatly
reduce the value of ,1 in Cu does not affect the applicability of the model.

(From [68]).

Testing the 2CSR and VF models. Adding Sn to Ag reduces the mean-
free-path without greatly reducing the spin-diffusion-length. The Co/Ag
and Co/AgSn data fall on the 2CSR model prediction of a single straight
line going through the origin. Adding Pt or Mn, in contrast, reduces the
spin-diffusion length, and causes the data to fall below the 2CSR model
line (see text). (From [68]).

AAR vs. (1», for Py(tpy)/Cu(20nm)/Py(tpy) EBSVs. Filled circles are the data,
along with the 2CSR (dashed line) and the VF (solid line) model with 1;;
= 5.5 nm. From [74].

AR(H) for an interleaved sample (A) and a separated sample (B).
Samples were already taken to saturation in the —H direction. There are
differences both in shape and size of the change in AR with H.

(A) A CPP mask. A screw through the middle circular hole holds the mask
in place, but allows it to be rotated. The three openings are used to sputter
the bottom Nb (1), the multilayer (2), and the top Nb (3). The blocking
position protects the sample before and after sputtering. (B) A ﬁnished
sample with Indium blobs on the Nb strips.

EDX spectrum of a CuGe ﬁlm. The table on the bottom shows the detailed
analysis of the spectrum.

10

11

19

20

21

24

30

31

2.3.

2.4.

2.5.

2.6.

2.7.

2.8.

2.9.

2.10.

2.11.

EDX spectrum of a CoZr ﬁlm, with a small concentration of Zr. The
analysis on the bottom of the plot shows 2.5% Zr.

EDX spectrum of a CoZr ﬁlm. The percentage of Zr given by the
analysis is 19%, which is expected to be well in the amorphous regime.

AR of two different thicknesses of CoZr in Nb/CoZr/Nb sandwiches. Any
dependence of AR on magnetic ﬁeld H is small.

Dependence of AR on tech. Samples both with and without Cu are
used in the straight line ﬁt. The ﬁt gives ARcw/Nb: 5.8 i 1 from2 and

peoz,=(19 i 1 Mom).

AR versus too for Nb/Co(tc°)/Nb sandwiches. Data are shown for two
different sputtering runs. The two straight lines are best ﬁts, giving an
average value of ZARNb/CO= 6.5 i 1 f sz, and pa, = 3.9 :l: 0.7 uﬂcm.

Au electrodes deposited on a 0.5”x0.5” Si substrate using
photolithography. The outside pads are large so that measuring electrodes
can be connected to them. The small samples will be deposited in the
middle of those electrodes. All the Au pads are connected to avoid
shocking the sample; these connections are removed when the sample is
ready to be measured. The crosses are alignment marks for subsequent
steps.

(A) Schematic diagram of the double exposure to the electron beam. After
baking the ﬁrst layer, all the area enclosed by the dotted line is exposed to
electrons. The area enclosed by the solid line is then exposed to electrons
after spinning and baking the second electron-beam-resist layer. After
developing, the part of the second layer between the dotted line and the
solid line will be ‘free hanging’, creating an undercut. (B) shows what the
substrate should look like after developing. The shaded area is the
undercut.

Photomicroscope picture showing the undercut (discoloration of the top
layer) of Fig. 2.93. The electron-beam resist is washed out completely
from the center area where the multilayer will be sputtered.

SEM images of the Al/Ti/Cu masks. Masks were dipped in PR
developer for 5 seconds. Ideally the mask should look like the mushroom
shape shown in Fig. 2.12. In A and B the substrate was tilted by 30°. In C
there is no tilt.

xi

32

33

38

39

39

42

44

45

47

2.12.

2.13.

2.14.

2.15.

3.1.

3.2.

4.1.

4.2.

Cartoon of the ideal shape of the AW i mask after the short wet
etchinginPRdeveloper. Ideally the mask would look a mushroom to protect
the Al part from SiO to be able to dissolve the mask after the ion milling
and SiO evaporation.

Ion milling thickness as a function of ion milling time. The slope gives the
ion-milling rate in urn/min. The intercept with the horizontal axis indicates
the existence of an oxide layer that ion mills slowly on the Cu surface. The
ion milling rate of Cu is (26 i 1) urn/min.

Picture of a sample after the ion milling of the multilayer (the line in
the middle) and depositing the SiO insulating layer. SiO is the dark
rectangular window in the middle. There are four crosses on the comers of
the picture (inside the dark squares) that serve as alignment marks during
the electron beam exposure. These alignment marks are covered by SiO,
which makes them less visible.

A ﬁnished device ready to be measured. Each of the top contacts (5 Ag
slabs) connects one of the top Au ﬁngers with one from the bottom and,
hopefully, has electrical contact with the top layer of the pillar in the
middle. The pillars appear as dark spots in the middle of the overlap
between the top contact and the multilayer.

Cartoon of an [F1/N/F2/N] N multilayer sandwiched between two
superconducting layers, the arrows indicate the magnetization directions
of the magnetic layers, showing an AP alignment.

2CSR model representation of the multilayers when 1;, << tp. The two
spin channels mix completely in the middle of each ferromagnetic layer
then split at around 1er ﬁom the PIN interface. So the two spin channels
are no longer independent.

AR(H) vs H at 4.2K for interleaved (top of each pair) and separated
(bottom of each pair) multilayers of Co/Cu with N= 4, 6, and 8. The unit
scales for each coupled pair are identical. The open squares and dashed
curves show how AR(H) varied when ﬁrst taken from the as-prepared state
to the saturated state. The ﬁlled circles and solid curves Show how it
varied after having been taken to saturation.

AR(H) vs. H curves for interleaved (top) and separated (bottom)
samples. Both samples on the left have the thinner Co layer with ICO= lnm
sputtered before the layer with (Co: 6nm. The two samples on the right
have the opposite order.

xii

48

50

51

53

57

64

78

79

4.3.

4.4.

4.5.

4.6.

4.7.

4.8.

4.9.

4.10.

Schematic, approximate pictures of inferred magnetic ordering for
interleaved and separated samples in their ‘as-prepared’ state and after
they have been taken to saturation and then the ﬁeld reversed only enough
to ﬂip the thicker Co layers.

M vs H at 12K for interleaved (x) and separated (+) samples with Cu and
CuGe with N = 8 for ten = tong, = 20 nm. P and AP states are indicated.

M vs H at 12K for center-cut pieces of Cu interleaved (top) and
separated (lower) samples including both data from the as-prepared state
to saturation and then from positive saturation to negative saturation and
back again.

Specular neutron reﬂectivity data and ﬁts plotted as a function of the
wavevector Q, of an interleaved sample [Co(6)/Cu(20)/Co(1)/Cu(20)]g at
18 K in a 250 G ﬁeld. Only the R' ' (dark squares) and R+ + (open circles)
reﬂectivities are shown. (The SF reﬂectivities are effectively at the
background level). The magnetization proﬁle for the Co layers that is
shown in the bottom graph was obtained from the ﬁt. The apparent
differences in moments in the thin and thick layers are artifacts (see text).

Specular neutron reﬂectivity data and ﬁts plotted as a function of the
wavevector Q2 of a separated sample [Co(6)/Cu(20)]g[Co(l)/Cu(20)]g at 18
K in a 250 G ﬁeld. Only the R" ' (dark squares) and R’r + (open circles)
reﬂectivities are shown. (The SF reﬂectivities are effectively at the
background level). The magnetization proﬁle for the Co layers that is
shown in the bottom graph was obtained from the ﬁt. The apparent
differences in moments in the thin and thick layers are artifacts (see text).

Specular and diffuse reﬂectivity data for the separated sample
[C0(6)/CU(20)]8[C0(1)/CU(20)]3 at 18 K in the as-prepared state. The data
were taken in a 3 G guide ﬁeld. The R' ' (dark squares), R+ + (open
squares), R+ ' (dark circles), and R' + (open circles) reﬂectivities are all
shown. The diffuse data have not been corrected for polarization
efficiencies. The dotted lines mark the superlattice peaks in the diffuse
data.

AR of interleaved and separated samples identical to samples in Fig. 4.1
except that CuGe replaces Cu. Symbols and curves have the same
meaning as in Fig. 4.1.

Effect of further increasing CuGe thickness on AR of interleaved and
separated samples. The convention for symbols is the same as in the
previous ﬁgures.

xiii

81

84

85

89

90

92

94

95

4.11.

4.12.

4.13.

4.14

4.15.

4.16.

4.17.

AR vs H for Cu, CuGe with tcuGc = 10,20,30,and 40 nm. The number of
repeats is 3 for all samples. The speciﬁc resistance ARp increases with
CuGe thickness, and AAR decreases, but the difference between separated
and interleaved samples persists. Data shown are after samples were taken
to saturation.

AR vs H for multilayers with short zl/t in all layers. The difference
between separated and interleaved samples has grown, disagreeing with
mean-free-path effects.

AR vs H curves. All samples are of the separated type. The bottom two
have a 1 nm FeMn layer in the middle of the Cu layer that separates the
Co(6) layers from the Co(l). Strong spin-ﬂipping by F eMn decouples the
two sides of the multilayers giving two separate peaks on each side of
H = 0.

. (AAR)s,p/(AAR)1mvs N. The dotted and long-dashed curves are predictions

using VF theory with the parameters for Co/Cu of [25] plus 1%? = 60 nm,

l§£= 500 nm, and ($36” = 130 mm. The vertical bars indicate the ranges

of uncertainties due to uncertainties in these values of 151:. The short-
dashed curve for Co/Cu and the solid curve for Cu/CuGe add 61 = 0.25 to
the prior calculations. Now the error bars indicate the ranges of
predictions for 61 = 0.25 :l: 0.1. For these curves, the error bars due to
uncertainties in 13;: are smaller than the symbols for the data.
(AAR)s¢p/(AAR)1m should be equal to l for all N, according to the 2CSR
model. The uncertainties in the data are comparable to the symbol sizes.

Calculated values of AAR vs 6, for interleaved and separated
multilayers with N = 6. The solid curves include the Co/Nb contacts; the
dashed curves are the same calculations but without those contacts. The
ﬁlled circles correspond to Eq. 4.1 with n = 5 for the interleaved case and
n = l for the separated one.

Calculated values of ARM) (left scale) and 1/ AR AP . AAR (right scale) for

multilayers with total thickness = 360 nm and either ﬁxed too = 6 nm or too
= to... Filled circles are for the 2CSR model. Solid curves are VF

calculations with IS“: = 500 nm, 1,33 = 60 nm, and 5, = 0.25.

AAR vs tea for symmetric (ﬁlled triangles) and asymmetric (open
circles) Co/Cu exchange-biased spin-valves. The dashed curves are prior
ﬁts [25] assuming no spin-memory-loss at the Co/Cu interfaces. The solid

curves just add an interfacial spin-memory-loss of 6, = 0.25.

xiv

96

98

99

101

105

107

110

4.18.

4.19.

5.1.

5.2

5.3.

5.4.

5.5.

5.6.

5.7.

5.8.

Comparison of CPP-MR and CIP-MR for separated and interleaved Co/Cu
multilayers. Symbols and curves have the same meanings as in Fig. 4.1.
The open squares for the separated CPP-MR were noisy, as shown by the
ﬂuctuations in the points for H = 200-500 0e. To reduce uncertainty at H
z 0, that square is the average of six very low ﬁeld measurements. All
other points are single measurements.

AR(H) vs H at 4.2K for interleaved (top) and separated (bottom)
multilayers of Co/Ag with N = 8. The scale units for each coupled pair are
identical. The open squares and dashed curves are for the initial state and
the ﬁlled circles and solid lines are after samples were taken to saturation.

Predicted AAR of CoZr/Cu symmetric EBSVs using parameters given in
the text and assuming inﬁnite spin-diffusion—length, [Eli-’2' , (solid line) or
IE3” = 8 nm (dotted line). The predictions are compared to data for
Co/Cu and CoFe/Cu EBSVs.

ARM» for CoFe and Co EBSVs as functions of layer thickness along
with the predicted ARp for CoZr using the parameters given in the text.

ARA}: vs t for CoZr/Cu EBSV data, compared with ‘best ﬁt lines’ for
CoFe/Cu EBSVs (broken curve), Co/Cu EBSVS (dotted curve), and the
2CSR model for CoZr/Cu described in the text above.

AAR vs I for Co, CoFe, and CoZr symmetric EBSVs. This ﬁgUre is a
replot of Fig. 5.1 with the addition of our CoZr symmetric EBSV data.

AAR vs tCoZr for symmetric EBSVs. The dotted curve is a best to just the
three EBSV data sets for AAR. The solid curve is the best ‘common ﬁt’ to
all of the samples, exchange biased and hybrid.

Using the parameters ﬁ'om the ﬁts to AAR ﬁts the data for ARAp quite well.
The dashed curve is the 2CSR model ﬁt described in the text above. The
best ﬁts are closer to the data than the 2CSR model prediction.

AR vs H for symmetric EBSVs with toozr = 8 nm (A) and 25 nm (B). In
both cases, the hysteresis curves are nicely square, indicating well-formed
AP states.

AAR vs tCoZr for EBSVs. (A) Symmetric EBSVs; (B) Asymmetric
EBSVs with tpy = 24 nm; (C) Asymmetric EBSVs with tpy = 6 nm. The
dashed curves represent a ﬁt to just these three sets of data. The solid
curves represent a ‘common ﬁt’ to all of the exchange biased and hybrid
samples with CoZr/Cu. In C, the two ﬁts fall on top of each other.

XV

111

112

119

119

124

125

127

127

128

129

5.9.

5.10.

5.11.

5.12.

5.13.

5.14.

5.15.

5.16.

5.17.

5.18.

AAR vs tCoZr for hybrid SVs with tpy = 24 nm (A) or 6 nm (B). The dashed
curves represent a ﬁt to just these two sets of data. The solid curves
represent a ‘common ﬁt’ to all of the CoZr/Cu data in Figs. 5.8 and 5.9.

Co dusting of CoZr/Cu symmetric EBSVs. Sputtering a 1 nm thick
layer of Co at each CoZr/Cu interface does not increase AAR.

Comparing AAR for dusting CoZr/Cu interfaces by Co, dusting Co/Cu
interfaces by CoZr, and no dusting. There is a systematic reduction in AAR
with the order of preparing the sample (indicated by numbers) in the
sputtering run. Dusting a Co/Cu interface with CoZr reduces AAR, and
dusting a CoZr/Cu interface with Co increases it.

AR(H) curves for EBSVs with Cu spacer layers (A), and Ag
spacerlayers (B and C). The pinning is strongest for the Cu sample A and
the total speciﬁc resistance, ARp, is lowest. Pinning the Ag sample at a
higher temperature (C) gave a slightly better result than at the usual
pinning temperature (B).

AAR of CoZr/Ag based symmetric EBSVs and a comparison with Cu. Ag

gives larger AAR values, especially at small thicknesses, indicating the
CoZr/Ag interface gives a larger contribution to MR than the CoZr/Cu
interface.

AAR vs 1cozt in CoZr/Ag/Py hybrid samples. The ﬁt uses the same
parameters as in Fig.5.l3 plus 7= 0.8 for the CoZr/Py interface. As in the
Cu-based samples the ﬁt is below data, requiring a slightly larger value
of 7.

AAR vs teal, in symmetric EBSVs. Changing the metal on which FeMn is
grown affects AAR. Growing FeMn on Ag gives the highest AAR.

Effect of FeMn growth on ARAp. Growing FeMn on Ag gives a
signiﬁcantly higher ARAp than when growing it on Cu or CoZr. The V-F
model predicts only a slight increase in ARM) for the Ag samples over the
Cu samples. The ﬁts for Ag and Cu are on top of each other.

ARM: vs tow. As predicted by the 2CSR model, ARAp grows linearly with
tCoZr- The slope is the same for Ag and Cu samples, showing that CoZr
has the same resistivity in both. But the intercepts are different, pointing
to the F eMn as the source of the larger ARAp for Ag.

Hysteresis loops of CoZr(t)/Cu]~ multilayers with ﬁxed total CoZr
thickness, t1 = 480 A . The saturation magnetization is almost the same
for all samples, regardless of the individual CoZr layer thicknesses. Any
magnetically dead CoZr layer at the CoZr/Cu interface must be thin.

xvi

130

132

132

134

135

136

138

139

139

141

5.19.

5.20.

5.21.

5.22.

5.23.

5.24.

5.25

5.26.

5.27.

5.28.

5.29.

5.30.

u/N vs tCoZr thickness for CoZr/Cu multilayers. The ordinate intercept
gives the thickness of any dead layer.

u/N versus CoZr layer thickness for CoZr/Ag multilayers. The ordinate
intercept indicates little or no magnetically dead layer at the CoZr/Ag
interface

AR(H) versus H, for [Conrioo.x(5)/Cu(20)/Py(6)/Cu(20)] m/Py(6). x is 97.5
in A and 95 in B. Vertical and horizontal scales are the same for direct
visual comparison.

AAR vs leozr. Different ﬁts are shown: 15; = inﬁnity or 2 nm, ,6 = 0.5 or
0.03, y= 0.77, and AREozt/cf 0.51.

AAR vs rm. for CoZr(t)/Cu(20)/Py(6) EBSVs.

AR vs H for a CoZr/Cu symmetric EBSV. The amorphous CoZr does not
give a noticeable AAR within the ﬂuctuations of the data.

Predicted change in AAR (A) and ARp (B) and AR/Rp (C) with number of
inserted internal Co/Cu interfaces.

AR(H) curves for samples (A) with 4 internal interfaces and (B)
without. To allow direct comparison, both the vertical and horizontal
scales are the same for both samples. Introducing the interfaces increases

AR, AAR, and the Hg for the Co.

Increase of ARp with M in [Py(6)/Cu(5)/(Co(t/n)/Cu(0.5))n/Cu(4.5)]3
multilayers versus prediction (solid curve).

AAR vs M. compared with 2CSR prediction. Data shows a signiﬁcant
increase in AAR at M = 4, then AAR levels off faster than predicted.

Effect of the thin Cu spacer layer thickness on AAR. When Cu layer
thickness is 0.4 nm (ﬁlled boxes) AAR is smaller than when Cu thickness
is 5 nm (ﬁlled circles). The one sample with to“: 0.6 nm (ﬁlled triangle) is
slightly higher than the 0.5 nm.

AR(H) curves for (A) Interleaved, (B) Separated and (C) Regular
samples. Regular means samples without any internal interfaces. The
reduction in AAR due to the difference in ordering of the layers between A
and B was already seen in Ch. 4. Note that samples A and B also have
larger ARs than sample C.

xvii

142

142

144

145

145

146

150

152

153

154

155

160

5.31.

5.32.

5.33.

5.34.

5.35.

5.36 .

5.37.

5.38.

5.39.

5.40.

AR vs H for samples identical to the sample in Fig. 5.30 except for the Cu
thicknesses. In A the Cu layer between neighboring Co layers is 0.5nm
and the rest are 20mm. In B the Cu layer between neighboring Py layers is
0.5 nm and the rest are 20 nm. In C the Cu layer between neighboring Co
or neighboring Py layers is 0.5 nm and the Cu layer between Co and Py
is 5 nm.

R(H) for an Ag based multilayer. Using a 4 nm thick Ag spacer layer is not
enough to decouple Py and Co layers and an AP state is not achieved in
this sample, giving sharp instead of rounded or ﬂat peaks.

Effect of internal interfaces on AR(H) curves for samples with Ag spacer
layers. Introducing eight internal interfaces( part A) increases AAR and
ARp.

Dependence of ARp on the total number of internal interfaces M. solid
boxes are for data and solid line is the 2CSR model prediction. The data
agree with prediction within the ﬂuctuations.

AAR dependence on the number of internal interfaces in Co/Ag based
multilayer.

Cartoon of the multilayer (with N = 1) with FeMn layers to enhance GMR.
A single FeMn layer is inserted in the middle of the Cu layer next to Nb on
each side of the multilayer. The Cu layers in direct contact with Nb are
assumed to become superconducting by the proximity effect.

AAR(N) in [Co(6)/Cu(20)/Co(1)/Cu(20)]1v multilayers. The ﬁlled circles
are data points and the solid line is a no-adjustable-parameters prediction
of the 2CSR model. There is a good agreement between the two,
especially at small N.

Comparison of 2CSR model prediction of ARp(N) with data for
[Co(6)/Cu(20)/Co(l)/Cu(20)]~ multilayers. The data falls below the
predictions, which may be because the Co/Cu interfaces are not fully
established in the 1 nm Co layers.

AAR vs N for multilayers with (ﬁlled boxes) or without (ﬁlled circles)
F eMn. Structure is
[Cu(10)/FeMn(1)/Cu(10)/[Co(6)/Cu(20)/Co(1)/Cu(20)]N/FeMn(1)/Cu(5).
The ‘regular’ samples are identical, just without the FeMn layers. AAR
increases by ~ 90% forN= 2 to ~ 9% forN= 4.

ARp vs N. Effect of the FeMn layers on the total speciﬁc resistance, ARp.
Boxes and circles have the same meaning as in the previous ﬁgure.

xviii

161

163

164

165

165

167

168

168

169

170

5.41.

5.42.

5.43.

5.44.

5.45.

6.1.

6.2.

6.3.

6.4.

6.5.

6.6.

AR/R vs N with (circles) and without (squares) FeMn inserts. Since both
AAR and ARp increase with introducing the F eMn layers, the
magnetoresistance, AR/R, changes only slightly.

AAR versus N. Data (ﬁlled boxes) for samples with FeMn thin layers,
2CSR model predictions for the case without FeMn (solid curve),
assuming that F eMn just removes ZARs/p (dashed curve), and assuming
that FeMn removes ZARs/p and turns on asymmetry at the outer two Co/Cu
interfaces (dotted curve).

Effect of rpm" on ARp (top) and AAR (bottom) for
Nb/Cu(10)/FeMn(t)/Cu(1 0)/[Co(6)/Cu(20)/Co(1)/Cu(20)]2/FeMn(t)/ Cu
(10)/Nb samples. Increasing tpeMn beyond 1 nm essentially does not
change ARp and reduces AAR slightly.

ARp (top) and AAR (bottom) vs N for FeMn inserts with (triangles) or
without (squares) additional thin Co layers outside of the FeMn layers.
Adding the Co layer gives no gain in AAR, but increases ARp.

Effect of the position of FeMn on (A) ARp and (B) AAR. Growing the
FeMn on the top of the multilayer gives a larger AAR and a slightly
smaller ARp than when FeMn is grown before the multilayer.

Spin-memory-loss detector. In the experiments to be described, X is either
a single layer of Ru or a Cu/Ru multilayer.

AAR on a log scale vs tau. The steep drop at small rm, is due to the
formation of two Cu/Ru interfaces. After the two interfaces are
established, AAR drops exponentially with IR“, and the slope of the
straight line on this semi-log graph is the inverse of the spin—diffusion-

length in Ru 1:: = (14 21:1)nm.

ARA]! vs N for a Py-based with EBSV with an [Cu/Rub, insert. The slope
of the best ﬁt line gives 2.412%, = 2.0 i 0.2 mm?

AAR on a log scale vs the number of [Ru(3)/Cu(5)] bilayer inserts. This
straight line ﬁt (on the log scale) of the data gives a spin-ﬂip probability of
electrons at the Cu/Ru interface ~ 35% per interface.

AR vs H curve showing an inverse CPP-MR, i.e., ARAp < ARp. The small
AAR compared to noise, and the background change with H, make it hard
to quantify the data.

AR(H) for the hybrid sample [Py(6)/Cu(20)/Ru(2)/Co(1.5)/Ru(2)/Cu(20)]6
showing inverse CPP-MR.

xix

171

173

174

175

176

178

180

180

181

181

184

6.7.

6.8.

6.9.

6.10.

6.11.

6.12.

6.13.

6.14.

6.15.

6.16.

6.17.

6.18.

6.19.

CIP R(1-I) loop for the sample of Fig. 6.6. The CIP-MR is also inverse.

Progression of AR(H) curve with increasing Co layer thickness. MR is
inverse for small thicknesses of Co and switches to normal for large Co
thickness.

AAR vs too . The data follow a nearly straight line, allowing an estimate
of room" = (-0.15 i 0.05). Arrows in the plot indicate that those data points
involve only a lower bound on the AP state.

AR(H) of a [Co/RU]2o multilayer. The antiferromagnetic coupling is too
strong to be overcome even at our highest applied ﬁeld of about 8 k6.

Average AR(H) for a [Co(3)/Ru(1.6)]io/Co(3) multilayer vs H. Each point
is repeated 100 times and averaged to get reliable results.

ARAP VS N. Analysis of the slope of the straight-line ﬁt gives ZARCOmu =
(1-2 i 0.23) f sz, and analysis of the intercept gives ZARCo/Nb= (7.3 i
2.3) mm2 (see text).

2CSR Model predicted AAR as a function of too for a Co(t)/Ru(10)/Co(t)
EBSV. AAR(t) is non-monotonic due to the competition between the
contributions from the Co/Ru interfaces and bulk Co.

AR(H) for Co(15)/Ru(10)/Co(15). Fields were chosen based on the AR(H)
carves of Co(15)/Cu(20)/Co(15) EBSVs. AAR is much less than predicted
in Fig. 6.13.

Drawing of the “spin-asymmetry-scattering detector” used to study Co/X
interfacial 7bo/x- If yea/x = 0, then AAR is also zero. Even small values of
mom will give nonzero MR, making the detector quite sensitive.

Reference sample: AAR a: 0.4 chm, is non-zero when there is no Ru in
contact with the Co.

AR(H) curve. AAR is very small in the curve. Solid boxes are taken for the
AP state and solid circle are for the P state.

AR(H) for samples with X = FeMn(2, 5). The values of AAR are very
close to zero in this case, just as in Fig. 6.17.

AR for samples with X = Cu(10)/FeMn(l)/Cu(10). AAR is now a non-
zero positive value, which indicates that the Co/Cu interface is now
contributing to GMR.

185

186

187

188

189

190

191

192

195

196

197

198

199

6.20.

6.21.

6.22.

7.1.

7.2.

7.3.

7.4.

A.l.

A2.

A3.

A.4.

AR for samples with X = Cu(l0)/FeMn(2)/Cu(2)/Co(2)/Cu(3). The AAR
values are almost identical to the ones without the Co layer.

AR for samples with X = Ru(2)/Cu(20). Putting the Ru next to Co gives a
negative AAR indicating that the outer Co/Ru interface is active and
contributes to GMR.

AR for samples with X = Ru(l or 2)/Cu(5 or 10)/FeMn(2)/Cu(10 or 15).
Adding the FeMn layer changes AAR very slightly from its value without
FeMn.

R vs H at room temperature. The sharp drop on the outer side of the two
peaks indicates a single-domain magnetization in the lithographed Co
layer.

R vs H of the same sample in Fig. 7.1 taken at 4.2 K. The R(H) curve
is asymmetric now probably due to the formation of an antiferromagnetic
CoO layer on one or both Co layers. Arrows indicate the direction of
changing H.

R vs I curve at 4.2 K of the same sample of Figs. 7.1 and 7.2. There is an
overall drop of resistance with current. The two hysteretic loops between i
10 and 21:20 mA are probably due to current-driven switching of the
magnetization. Open boxes and ﬁlled circles are for two sweeps of current.

R vs II at room temperature. Relatively ﬂat peaks indicate a good AP
alignment. The change in resistance is higher than in Fig. 7.4, and the drop
in the resistance after the AP peak is sharp indicating single domain
switching of the magnetizatio.

AR vs H for CoZr/Cu/Py asymmetric EBSVs. A good AP state is produced in
samples with thin CoZr, while the samples with thicker CoZr show a poorer AP
alignment. This behavior is common for both 19, = 6 or 24 nm.Horizontal scale is
the same for all samples to allow direct comparison of the switching ﬁelds.

AR vs H for CoZr/Cu/Py hybrid spin valves. Curves are sharper for samples with
thicker CoZr layers indicating the lack of a good AP alignment.

AR vs H for CoZr/Ag/Py hybrid spin valves. The AP alignment gets quite
uncertain as the CoZr layers become thicker, as seen from the bottom part.

AR vs H for CoZr/Ag symmetric EBSVs. Unlike in the asymmetric and the
hybrid spin valves that contain Py, the AP alignment seems good for both thin
and thick CoZr in the symmetric EBSVs. Pinning is poorer than the Cu-based
EBSVs (see Figs. 5.7 and 5.12).

200

202

203

205

206

207

208

212

213

214

215

A.5.

ARA, vs tool, for [Py(tpy)/Cu(20)/CoZr(t)/Cu(20)],o/Py(24) hybrid multilayers with
11», = 6nm (top) or 24 nm (bottom). The ﬁgure shows data, best common ﬁt to all
AARs of CoZr samples (solid line), and best ﬁt to AARs of the hybrid samples

alone (dotted line). The offset is most likely because the resistivity of Py is less
than the value used in the ﬁts.

xxii

216

Chapter 1

INTRODUCTION

1.1. DEFINITIONS AND BACKGROUND

Magnetoresistance, MR, is the fractional change in resistance of a sample upon the
application of a magnetic ﬁeld. For the samples studied in this thesis, it is now traditional
to deﬁne its magnitude as the difference between the sample resistance, R(H), at
magnetic ﬁeld H and at the saturation ﬁeld, HS, R(Hs), divided by R(Hs):

MR(H) = [R(H) - R(Hs)]/R(Hs)-
H5 is the ﬁeld above which R(H) becomes constant (independent of H).

When the sample is a ferromagnetic/normal metal, F/N, multilayer this MR is called
Giant magnetoresistance, GMR [1-6], because one of the original experiments on F/N
multilayers in 1988 [1] gave a room temperature MR ~ 50% (~100% at 4.2K). This MR
is much larger than the MRS of both non-magnetic and magnetic metals at room
temperature; for example, the 2% MR of the magnetic alloy Permalloy (NiMFex with x ~
0.8) [7] is large enough that Py thin ﬁlms were used as the read-head sensors in
computers until they were recently replaced by GMR multilayers [8,9]. The MR of a
simple ﬁlm made of a ferromagnet like Py is called anisotropic MR (AMR), because the
resistance is different for current ﬂow parallel or perpendicular to the direction of the
applied magnetic ﬁeld. AMR is also present in GMR multilayers, but is usually small

compared to the GMR. However, GMR is now used as a universal descriptor of the MR

of magnetic multilayers, even though in some cases the GMR can be less than the AMR
[31

In the original GMR experiments [1,2], the direction of current ﬂow was in the plane
of the layers of the multilayer (i.e. parallel to the F/N interfaces). This sense of current
ﬂow is called Current-in-Plane (CIP) and gives rise to CIP-MR. Current can also ﬂow
perpendicular to the plane of the layers giving Current Perpendicular to Plane (CPP)-MR.

The discovery of GMR stimulated great interest not only to understand the physics
underlying GMR, but also to exploit it technologically and to use it as a sensor for
probing other phenomena. GMR has been used as a sensor to study the coupling between
ferromagnetic layers when separated by nonmagnetic spacer layers [1 0-14] and, more
recently, generation of magnons and magnetization switching by high intensity polarized
direct electric current [15-21]. The discovery of GMR accelerated the emergence of a
new technology called spintronics [22-24], in which the spin-dependent properties of
electrons are crucial for devices, and that has potential to replace much of the traditional
silicon-based electronics technology based only upon the electron’s charge.

The earliest experiments on GMR, and all of the device applications so far, involved
the CIP geometry, where a typical thin ﬁhn multilayer has a resistance of several ohms,
making measuring GMR simple.

The ﬁrst CPP-MR measurements were reported in 1991 by our group, and showed
that the CPP-MR could be several times larger than the CIP-MR [25]. It also quickly
became clear that the CPP-MR can often be described by simpler equations than the CIP-
MR, allowing simpler analysis of experimental data [26,27]. These initial

measurements involved samples with macroscopic area through which the CPP-current

ﬂows (A ~ 1 m2), and thus very small resistance ~ 10 n9. To obtain uniform current
ﬂow through the sample, and thus to be able to measure the speciﬁc resistances of main
interest (ARAp, ARp (deﬁned below), and their difference, ARAp - ARp = AAR) all
quantitatively, the measuring technique used crossed superconducting strips and required
low temperatures and sophisticated equipment [25,28]. AR, the product of A and the
sample resistance R, is the intrinsic quantity in the CPP geometry [29] (i.e., it is the
inverse of the conductance per unit area). ARM» and ARp represent the values of AR
when the magnetizations of adjacent F -layers in the multilayer are aligned anti-parallel
(AP) or parallel (P) to each other. AAR represents the maximum change in speciﬁc
resistance with change in magnetic ﬁeld, and the CPP-MR = AAR/ARp.

Soon afterward, CPP resistances ranging ﬁom fractions of ohms to ohms were
achieved by using optical lithography to prepare multilayer samples with areas A ~ (umz)
[30-33], or by electrodepositing multilayers into nanotubes with diameters Z 30 nm in
plastics subjected to radiation and then etched [6, 34-39]. Both techniques allow CPP-
MR measurements to be extended to room temperature. For the lithographed samples,
however, because they were wider than long, the current density was not uniform,
making it difﬁcult or impossible to reliably determine AR. While the electrodeposited
samples were long and thin enough to give uniform current ﬂow, until recently
measurements involved an unknown number of wires in parallel and, so far,
electrodeposition has been limited to only a few F/N metal (or alloy) pairs [39].
Measurements have also been made of the MR with current-at-an-angle to the planes
(CAP-MR) [4,40-42], and a combination of CIP-MR and CAP-MR data extrapolated to

the CPP-MR [4,41]. The present thesis focuses upon measurements of the CPP-MR

using the low temperature, ‘large area’ technique just noted, and some preliminary
studies of the CPP-MR using a combination of optical and electron-beam lithography to
produce samples with areas A ~ 0.01 umz. Measurements of the CIP-MR will only be
presented occasionally for purposes of comparison.

In section 1.2 we review the simple phenomenological picture of the source of GMR
outlined in ref. [1]. Section 1.3 describes present uses of the CIP-MR for devices, the
potential for use of the CPP-MR, and the main problems to be solved if the CPP-MR is to
become competitive. Section 1.4 brieﬂy reviews those facets of the ﬁeld of CPP-MR that
are of direct relevance to the topics of this thesis. More complete reviews can be found in
[3-5, 43] and the theses of S. F. Lee [44] and Qing Yang [ 45]. The chapter concludes

with Section 1.5, an outline of the studies in this thesis.
1.2. BASIC IDEAS IN GMR

GMR arises from spin-dependent scattering in the F-metal and at the F/N interfaces
[1]. Such scattering depends upon the band structures of the F and N metals, the
impurities present in both the F and N metals, and the intermixing of the F and N metals
at the F/N interface. The band structure of a ferromagnetic metal is moment—dependent;
that is, the d-bands with one moment direction are shifted relative to the d-bands with the
Opposite moment direction. At the Fermi surface this shifting leads to a difference in the
density of states for d-band electrons depending on their moment direction. Fig. 1 .1
shows a cartoon of the s-bands and d-bands of an N-metal and an F-metal.

This unbalanced occupation of the d-bands gives rise to a net magnetization in the F -
metal, even without the presence of an external ﬁeld. The direction in which more

electron moments point is called the majority direction, and the direction in which fewer

 

point is called the minority direction. In the simplest picture, the s-electrons are assumed
to carry most of the current. If electrons don’t ﬂip their moments, then majority s-
electrons will scatter mostly into other majority s-electron states, because there are only
few majority d-electron states available at the F errni energy. Minority electrons, in
contrast, can scatter more often into the larger number of minority d-states at the Fermi
energy. These different scattering processes give rise to a different scattering probability
for electrons with moments opposite to the local magnetization of the ferromagnetic
metal, and electrons with moments along the local magnetization. This difference is
called spin-anisotropy [4]. If no moment ﬂipping occurs, then these two ‘moment
channels’ (or ‘spin channels’) will not mix, and can be treated as two separate channels of
conduction equivalent to resistances in parallel. Flipping of the electron moments mixes
these channels, making the picture more complex.

These different scattering rates can be described phenomenologically in terms of
different resistivities for majority and minority electrons—i.e. electrons with their
moments along or opposite to the local magnetization M [46]. The different scattering
rates make the resistivity of the majority electrons, p’, different ﬁ'om that of the minority

ones, p’. The F /N interfaces also have spin-dependent scattering, which we characterize

similarly by speciﬁc resistances ART and AR". Instead of using the four parameters p’,

l t
p’, RT, and R’, we use a dimensionless bulk spin-asymmetry parameter, ,8 = iI—lo—T— ,
.0 'l' P
i-R’
a dimensionless interfacial spin-asymmetry parameter, 7 = T—‘l- , the measured bulk
R + R

resistivity of the F-metal, pp, and an interfacial speciﬁc resistance, ARp/N. Chapter 3 will
describe in more detail the use of these parameters to describe the CPP-MR.

To explain how GMR results from such spin-dependent scattering, consider the F/N/F
trilayer in Fig.1.2 and assume that minority electrons are more strongly scattered than
majority ones, both in the F -metal and at the F/N interfaces (i.e., assume 13F > 0 and 'YF/N >
0). Further, consider a ﬁxed direction in space (e.g. the direction of the initially applied
magnetic ﬁeld H), and that the magnetizations of the F-layers are allowed only to be AP
(Case A) or P (case B) to each other. Lastly, divide the electrons entering the trilayer into
ones with moments up (in the direction of the initially applied H) and moments down,
and do not allow their moments to ﬂip as they cross the multilayer.

When the F-layer moments are (AP) to each other (case A), both up and down
electrons will be scattered strongly in one F -layer and weakly in the other, and their
resulting resistances will be identical. The net resistance will be half of each of these two
equal resistances. When, in contrast, the F-layer moments are parallel (P) to each other
(Case B), one of the electron channels (the ‘up’ electrons in Fig. 1.2) will be scattered
only weakly in both F-layers, the resistance of this channel with be small, and will ‘short’
out the total resistance. The net resistance will now be even smaller than this shorting
resistance, and thus smaller (perhaps even much smaller) than the resistance in the AP
case, Rp << RAP. If the relative alignment of magnetizations is between parallel and
antiparallel, electron moments will sometimes ﬂip as the electrons transit the multilayer
(since the electrons will no longer be in quantum eigenstates of the system). While the
total resistance will usually be intermediate between those of the parallel and antiparallel

states, the analysis becomes much more complex. Because of the resulting simplicity of

analysis, we will limit our study to only the P and AP states, focusing on ARM, ARp, and
their difference AAR = ARAp — ARp.

We assumed above that the electrons don’t ﬂip their moments as they transit the
multilayer. If the electrons ﬂip their moments as they pass through the N-layer, the GMR
will be reduced. In the extreme case of many ﬂips, the electrons will no longer care
whether the F-layer moments are aligned P or AP, and the GMR will vanish. We call
such effects of spin-ﬂipping ‘losing spin-memory’.

In a bulk metal, spin ﬂipping leads to spin-memory-loss, and is measured by a length
called the ‘spin-difﬁrsion length’ lsp [47-50]. Because spin-ﬂip scatterings are usually
only a small fraction of the total scattering events (at low temperatures, the ﬁaction
coming from spin-spin exchange coupling and spin-orbit coupling), [SP is usually longer
than the electron mean-free-path, A, in the same layer. Thus we normally expect the
inequality 13;: >> A to be valid in a given F- or N-metal.

We assumed in the above analysis that we could produce both AP and P states. The
P-state can be produced by applying a magnetic ﬁeld strong enough to align the moments
of all F-layers along the ﬁeld. The AP state, however, is more elusive and three different
ways have been used to produce it.

1. Using the antiferromagnetic exchange coupling that occurs between two F-layers
separated by a certain small thickness of a given N-layer [1, 2, 10-14]. The
discovery of this coupling originally lead to the discovery of GMR. For even
smaller N-layer thicknesses the exchange coupling is ferromagnetic, and for larger

N-layer thicknesses it oscillates with N-layer thickness, getting weaker as that

thickness increases until its effect becomes negligible at thicknesses of a few nm
[11,13].

2. Using an exchange biased spin valve, EBSV, in which one of two F-layers is
heated and cooled in a ﬁeld while in contact with an antiferromagnetic (AF) layer.
The exchange coupling between these layers increases the switching ﬁeld st of
the F-layer [51,52], in many cases to a value much larger than that of a second
‘free’ F-layer separated from the ﬁrst by an N-layer thick enough to minimize
exchange coupling between the two F-layers. The moment of the ‘free’ F-layer
can then be switched back and forth while that of the ‘pinned’ F-layer is held
ﬁxed.

3. Using a hybrid multilayer (or hybrid spin-valve (SV), composed of two F-layers
(again separated by a sufficiently thick N-layer) with very different switching
ﬁelds. These different ﬁelds can be achieved by using either: (a) two different F-
metals with very different switching ﬁelds, or (b) only one F-metal, but with two
very different layer thicknesses, since the switching ﬁeld depends upon the layer
thickness.

The ﬁrst method, using antiferromagnetic coupling, was widely used in early GMR
studies, but will not be used in this thesis because it does not allow variation of the N-
layer thickness continuously to make systematic studies. We will use the other two
methods. Fig. 1.3 shows examples of AR(H) curves for an EBSV (A) and a hybrid spin
valve (B) studied in the present thesis. The main difference between the two is the shift
in the AR(H) curve to one ﬁeld direction from H = 0 in the EBSV, which is characteristic

of exchange biased samples [51]. Such EBSVs are often the structure of choice for CIP

devices because they allow reversible CIP-MR measurements at low ﬁelds and because

the magnetization of the pinned layer remains stable over many free layer switches [53,

54].

Normal
Metal

 

 

 

4s

 

 

 

Ferromagnet

 

 

 

 

3d

Fig. 1.1. Cartoon of the s- and d- subbands of a ferromagnetic metal and a normal metal.
There is a net magnetic moment in the ferromagnet but not the normal metal.

 

Upi

upi

I
CPP CP
downi downi

 

 

 

 

 

 

 

(B)

 

' ‘ {ll/{’1 ' ' ’

N I
lllllll

 

’-’r/////. ‘
~ ”xx/x. -

 

 

-‘-‘//////.

 

 

 

 

 

 

 

Fig. 1.2. Schematic diagram of scattering in the ‘up’ and ‘down’ channels when the
magnetizations of the ferromagnetic layers are (A) antiparallel or (B) parallel to each
other. The star indicates stronger scattering. Scattering is only shown in bulk F.

10

22. We,
.

 

 

 

 

 

 

 

 

 

 

Fan/Py(24)l0u(20)le(24) .
21.5 E .j
"E 21 :— i i i. \\ ( ) .2
i I i \. 3
C 20.5 E ii \ it i
2: 20 : ‘ i \ _
m : \\
< 19.5 \¥ \ _
19 ' N L . if
18.5LA424111-4414911.i.1.i.ihii.’
-50 0 50 100 150 200 250
H (0e)

22 If- ,
[Co(6)iCu(20)iCo(1wearing 3
21 .j
“2 20E. )1
C :
z: 19 1
2t : . E s
18 _- I . _‘
17 ‘i— '4’

" ‘ 4100035100 ‘ '0‘ ‘ ‘500‘ 1000‘ ‘
H (De)

Fig. 1.3. AR(H) for (A) an EBSV, and (B) a hybrid spin valve. AR(H) for the hybrid SV
is symmetric around H = 0, while that for the EBSV is not. Numbers indicate layer

thicknesses in nm.

1.3. GMR DEVICE APPLICATIONS

CIP-GMR found its way into devices in a relatively short time [24,55,56], and is an
elegant example of how advances in solid state physics can lead to technology.

The ﬁrst GMR-based general-purpose motion detector became commercially
available in 1994, just six years after the discovery of GMR [57]. In a motion detector,
the GMR multilayer sensor is placed in the vicinity of the moving part of interest, which

is shaped so as to give a time varying magnetic ﬁeld at the sensor. The moving part can

11

either be magnetic (e. g., steel), or coated with a magnetic material. The sensor changes
its resistance in response to the changing ﬁeld. This technique can be used to detect
rotational motion (such as for motors), or linear displacement (such as for automatic
breaking systems) [56,58].

In 1998, IBM offered the ﬁrst large-scale commercial application of GMR as the
magnetic ﬁeld sensor in the read heads of magnetic hard discs in computers [24]. This
rapid implementation of GMR into read heads occurred because a GMR thin ﬁlm with R
~ 20 Q and CIP-MR ~ 6% (by now somewhat larger) simply replaced a very similar Py
thin ﬁlm with R ~ 20 Q and AMR ~ 2%. Read heads require the ability to switch
resistance from high to low values in a relatively low magnetic ﬁeld and stability over a
range of temperatures. These requirements exclude some samples with very high MR,
such as single crystal Fe whiskers [59] and the colossal magnetoresistance, CMR,
observed in oxides [60,61].

Demonstrations of GMR random access memory were made as early as 1997.
Several companies now have established research programs for developing nonvolatile
GMR memory devices [24].

All the devices mentioned above use the traditional CIP geometry, because of its
easy-to—measure resistance. However, the CIP-MR looks as if it has leveled off.
Companies are now looking for systems with larger MR5 to read the smaller bits under
development for higher density magnetic memory, or to serve as switchable, non-volatile
magnetic memory elements. Some companies [62] are investigating F/N metal CPP-
based read heads and memories, due to the large potential CPP-MR. However, an

obstacle to implementing CPP-MR devices with all-metal systems is the small speciﬁc

12

resistance of present day multilayers, AR ~ 10 film2 in samples with total layer
thicknesses limited to ~ 50 nm. Even at A = 0.01 micron2 the total CPP resistance is only
~ 1 Q. A competitor for both read heads and magnetic memories, is magnetic tunnel
junctions, giving tunneling MR (TMR) [56,63]. Present junctions with large room
temperature TMRs (3040%) have A123 (2 60,000 mm?) too large for read heads [56].
Research is underway to ﬁnd ways to reduce AR without reducing the TMR. Large AR
is not a fundamental problem in magnetic memory, for which TMR is under intense
investigation [64].

Extensive spintronics studies of semiconductors are now also underway. Some
semiconductors have been shown to conserve the current polarization over fairly long
distances [65,66]. But how to inject a large spin-polarized current into them from
physical contacts is still a topic of intense discussion and research [67].

We will explore in this thesis some of the potential ways to increase the total speciﬁc
resistance, AR, and/or AAR of metallic multilayers, in hopes of making the CPP-MR more

viable for devices.

1.4. CPP-NIR THEORY & HISTORY CRUCIAL TO THIS THESIS

Fert initially explained GMR phenomenologically in terms of the two-current model
discussed above [1 ]. This picture was quickly converted into transport equations to
describe the CIP-MR in multilayers. Because of the low symmetry of the CIP-MR, these
transport equations for a simple F/N multilayer contain many parameters [3], among

which are the mean-free-path in the N-layer, AN, and the mean-free-paths in the F-layer,

A; and/1:. Qualitatively, for an electron crossing the multilayer to recognize the relative

l3

orientation of the moments of two adjacent F-layers, the N-layer thickness must be less
than 1N. More quantitatively, AN appears in the transport equations as a ratio with the
layer thickness m in the argument of exponential ftmctions—functions involving exp (-
twig—making AN (and similarly it; and it) ‘characteristic lengths’ in the CIP-MR. In
addition, current ﬂow through the layers of multilayer in the CIP geometry is not
uniform; the current tends to shunt through the lower resistivity layers (usually the N-
layers). Lastly, the CIP-MR seems to be very sensitive to roughness of the F/N
interfaces, since the electrons, on average, move parallel to these ‘thin ﬁlm’ interfaces.

Zhang and Levy [26], based upon a free-electron analysis assuming diffuse scattering
of electrons, ﬁrst pointed out that, if no spin ﬂipping occurs, the CPP-MR transport
equations should be very simple, just series resistor equations for two parallel current
channels. Unlike the CIP-MR equations, the mean-free—path lengths in the N and F
metals are not characteristic lengths, but enter only via the layer resistivity parameters 9.1
and pp. Qualitatively, in the CPP geometry, the current-carrying electrons must travel
through every layer and interface, and thus must ‘experience’ the actual magnetic order,
regardless of the mean-free-path lengths. The only lengths in the CPP-MR equations are
then the F- and N-layer thicknesses, IF and m.

Lee et a1. [27] (and later Valet and Fert [47]) ﬁrst wrote down the two-current series-

resistor, 2CSR, model equations using the parameters ,BF, 7pm, pp, and ARp/N deﬁned

1‘ l
(PF+P1=)= PF and

4 1—p2

 

above. Deﬁning an alternative F -layer resistivity, p;

14

T Jr
. . . AR +AR .
Similarly, ARp/N =( ”N 4 FIN) = 14R”: , leads to the followmg 2CSR model
‘7

 

equations for AR Ap and AAR for a simple [F/N]N multilayer with N repeats.

 

2 0 t 2
AAR = N [ﬂthF +27ARF/N] (1.2)
ARA,

An alternative form of Eq. 2 that we will use in this thesis is

 

WARM/1R») = N(.BP;4F + 27’ Ath/N) (1-3)
If the leﬁ hand side of Eq. 1.3 is plotted against N for samples with ﬁxed IF, the data
should fall on a straight line passing through the origin, and the slope of this line should
be independent of the non-magnetic metal resistivity, m, and also independent of the
total sample thickness, IT.

The samples in this thesis are all made with superconducting (S) leads. All
experimental evidence so far indicates that the contribution of these leads can be
incorporated into the above equations simply by adding to Eq. 1.1a a constant (and
independently measured) term ZARs/F, to give Eq. 1.1b.

AR AP = ZARSW + NpNtN + NpgtF + 2NAR,§,N (1 . 1b)
However, the validity of this procedure is yet to be proved theoretically. Eqs. 1.2 and 1.3
remain unchanged. These 2CSR model equations (1.1b-1.3) have been found to work
surprisingly well for Co/Cu and Co/Ag multilayers, which consist of pure metals with
long spin-diffusion lengths. That is, a wide variety of data in these systems have been
found to be consistent with the forms of Eqs. 1.1b-1.3 with parameters that are

independent of the layer thicknesses tp and m [27,28]. Eq. 1.3 predicts that a plot of the

15

leﬁ-hand-side ‘square root’ versus N for multilayers with ﬁxed to, = 6 nm should fall on
a straight line passing through the origin, independent of the total multilayer thickness or
the resistivity of any N-metal or N—alloy in which spin-ﬂipping is weak. Fig. 1.4 for
Co/Cu and Cu/CuGe, and Fig. 1.5 for two different sets of Co/Ag plus Co/AgSn, both
show exactly this predicted behavior for non-magnetic metals Cu or Ag. Adding a little
Ge to Cu or Sn to Ag increased the resistivity by about a factor of 20 [27,68], but
produced only weak spin-ﬂipping [50], thereby still leaving the 2CSR model applicable.
These data were taken as strong evidence that Eqs. 1.1-1.3 contain the essential physics
of the CPP-MR for systems where spin-ﬂipping is negligible, and that the mean-free-
paths in the F- and N-layers do not play a signiﬁcant separate role from their appearance
simply within the parameters pp and Au.

In 1993, Valet and Fert [47] argued that the characteristic lengths in the CPP-MR are
the spin-diffusion lengths, lg]: in the N-metal and [E]: in the F-metal. Using a free-
electron Boltzmann Equation model, they showed that the 2CSR model only describes

the CPP-MR in the limit 1;} >> m and 1's} >> tp. They wrote down the more general

equations (which we call the Valet and Fert (V F) model) that applies when the spin-
diffusion—lengths are not that long.

For later use, we note that the VF model assumes no band structure and is also strictly
valid only when 15p >> A, a condition expected to be valid for most metals [47,69,70].
Said another way, the VF equations are the leading terms in a Boltzmann equation
analysis and the corrections are ~ il/lsp. While it is not clear that this condition has
always been met in experimental studies [43], the results of those studies have seemed to

be consistent with the VF equations [47]. As one example, derivations of values of [SF in

16

N—metal alloys applying the VF equations to data such as those in Fig. 1.5 were found
[50,68] to be consistent with independent estimates from measurements of spin-orbit
scattering [71] or, where appropriate, with calculations of spin-spin exchange [72]. The
heavy impurity Pt is expected to produce strong spin-orbit scattering, and the ‘magnetic’
impurity Mn is expected to produce spin-spin exchange scattering. In both cases, the
presence of spin-ﬂipping causes the data to fall below the straight line of Eq. 1.3 (Fig.

1.5), and the values of 15; were derived by ﬁtting the data to the VP model.
Measurements of 1g}. were extended to additional alloys in [73].

The VF model was also found to be necessary to understand measurements of how
AAR for symmetric Py-based EBSVs varied with the Py thickness, tpy. Instead of AAR
continuing to grow with increasing tpy as expected from the 2CSR model, AAR saturated
in value, becoming constant beyond a certain value of tpy (Fig. 1.6) . This behavior was

attributed to a ﬁnite 1:; ~ 5.5 nm [74], a value subsequently conﬁrmed independently

from studies of electrodeposited nanowires [39]. Also, with this ﬁnite 1;} , VF theory was

found to describe surprisingly well the values of ARM: and AAR for Co/Cu/Py/Cu hybrid
spin-valves, using only the previously determined parameters for Co/Cu and Py/Cu

without adjustment [75]. Later studies of spin-memory-loss in Co suggest that it has a

much longer value, [E]? = 60 i 20 nm [76,77]. All of these results constitute evidence that

the characteristic lengths in the CPP-MR are spin-diffusion lengths, not mean-free-paths.
We conclude this discussion of prior work with brief mention of three additional

studies of importance to the present thesis.

17

From studies of Deviations from Matthiessen’s rule due to addition of two different
impurities at once to a host F-metal, Campbell and Fert [46] list estimated values of ,6 for
alloys of Co, Fe, and Ni with a variety of impurities. For certain impurities (especially
Cr, V, and Ru), these values of ,6 are negative, i.e. the scattering anisotropy is opposite to
the usual one. Our group performed experiments both to look for negative ,Bs in
appropriate alloys and to test the absolute values of the Campbell-Fert estimates in a few
cases. As predicted, evidence of negative ﬂs were found for FeV, NiV, FeCr, and NiCr
[78-80] Reasonable agreement in magnitudes with the Campbell-Fert values were found
for NiFe and CoFe (both with positive ,6) and for NiCr (negative ﬂ) [81,82]. The method
and equations used for determining the sign of ,B will be described in Chapter 3.

In an attempt to test the applicability of the 2CSR model, our group examined AAR
for Co and Py based hybrid spin-valves in two different forms: ‘interleaved’
[Co/N/Py/N]N , and ‘separated’ [Co/N]N/[Py/N]N , with N = Cu or Ag [83]. When these
measurements were ﬁrst made, the spin-diffusion lengths in both Co and Py were
expected to be long. In such a case, AR(H) should be the same for interleaved and
separated samples, because the 2CSR model doesn’t change as the order of the layers
varies. To initial surprise, a clear difference was found between the values and shapes of
AR(H) for the two kinds of samples (see, e.g. Fig. 1.7 below). This difference led to the

measurements of AR(H) for Py-based EBSVs described above, and the discovery of a
short 13. Subsequently, Bozec et al. found similar differences in [Co/Cu/Fe/Cu]~ hybrid

spin-valves [84]. They tentatively attributed the differences either to a short Is]: in Fe or

to spin-memory-loss at the PIN interfaces.

18

Lastly, as will be discussed in Ch. 3, VP theory predicts that inserting a source of spin-
memory-loss judiciously into a multilayer with large lead resistance can reduce the effect
of the lead resistance on AAR, and thereby increase AAR. Indirectly, such behavior is
already evident in the ﬁtting of the Py-based EBSV data described above. Direct
evidence for such behavior was presented by Gu et al., who inserted a strong spin-

ﬂipping alloy, FeMn, into one of the Py layers in a Py-based EBSV [85].

 

' 1
80 - ta.=0mn '-
. o 00/01! ,.
* II (SO/WM“) 3 r
60 -- 9’ -
.. 9” a

[ARt(Ho)lAR.(Ho)-A&(Ha)l]"' (fﬂm')
8 8
| l
O 0‘
I

 

 

 

O 20 4O 60 80
Bilayer No. N

Fig.1.4 Square root test of the 2CSR model. Adding Ge impurities that greatly reduce the
value of A in Cu does not affect the applicability of the model. (From [68]).

19

 

a:
O
I
‘a.
l

     

° CO/MB/tu) '
o Co/Ag(6/B) ""
'3 c0/ A831194") ‘
v eta/Wax) -
’ (kl/WW") -

l l J l l L l l I l I l

0 20 40 60 80
Bilayer N o. N

N
O
I

6‘

5 '0

V : a,’ 11m
E: 60 F- 3:01:11: "'4
3.3 I ’ Z
5’ L ’ 4
J; 40 r- "
a j; ,7 I
5

5

g

 

 

 

Fig.1.5 Testing the 2CSR and VF models. Adding Sn to Ag reduces the mean-ﬁee-path
without greatly reducing the spin-diffusion—length. The Co/Ag and Co/AgSn data fall on
the 2CSR model prediction of a single straight line going through the origin. Adding Pt

or Mn, in contrast, reduces the spin-diffusion length, and causes the data to fall below the
2CSR model line (see text). (From [68]).

20

 

 

 

 

 

. - I r I 1 l
0.0 0 . l 0 20 30
try (nm)

Fig. 1.6. AAR vs. 1;», for Py(tpy)/Cu(20nm)/Py(tpy) EBSVs. Filled circles are the data,
along with the 2CSR (dashed line) and the VF (solid line) model with 13 = 5.5 nm. From

[74].
1.4.1. High Current-Density Effects in F/N Multilayers

A new avenue for GMR related studies in sub-micron sized samples emerged in 1996,
when Slonczewski [15] and Berger [16] predicted that the magnetization direction of
magnetic layers could be switched in a controlled way by a large enough electric current
density. In a sense, this phenomenon is the inverse of GMR, where controlled changing
of the relative magnetic order of a multilayer leads to a change in the current ﬂowing
through it (via a change in resistance). They predicted also [15,16] that a high current
density could excite magnons (quantized spin waves) in the magnetic materials when an
applied ﬁeld limited the ability of the magnetization to completely switch. Such current-
induced switching makes CPP-MR potentially attractive for devices like MRAM,
because current-induced switching would be more localized, and probably also allow

simpler circuit design, than magnetic-ﬁeld—induced switching. The magnon generation

21

opens the possibility for high frequency generation. The large current densities needed
for such switching cannot be applied to macroscopic samples with superconducting
contacts, making it impossible to study current induced switching in them. These
phenomena need small areas of current ﬂow, which have been achieved using point
contacts [17,18], e-beam lithographically produced submicron sized samples [192]], or
electrodeposited nanowires [6,39].

Tsoi et al [17], using point contacts, ﬁrst observed current-induced spin waves. The
Cornell group [19] soon observed the current-induced sWitching of magnetization in
nano-pattemed, pillar shaped trilayers of Co/Cu/Co, and such switching has been
conﬁrmed by others [20,21]. There are two different switching mechanisms, the self
magnetic ﬁeld of the current [21], and spin transfer from the polarized current to the
layers [19,20]. From simple arguments, one expects the self-ﬁeld switching to scale as ~
1/d, and the spin—transfer switching to scale as ~ 1/d2, where d is the sample diameter.
Thus, self-ﬁeld switching should dominate for large samples, and spin-transfer switching
for small ones (e. g. d s 200 nm). More precisely, the crossover diameter depends upon
the magnetic and GMR parameters of the materials, including those determining the
torque, magnetization relaxation , and magnetic anisotropy. The two sources can be
distinguished experimentally by their different hysteresis curves for R vs 1. Self ﬁeld
switching should give an R(I) symmetric about I = O, with peaks in resistance on each
side of I = 0 (like the MR(H) curve for a simple F/N multilayer— Fig. 1.3B). The peaks
are at the AP states, with the high ﬁeld limits in both directions always being P-states.
Spin-transfer torque, in contrast, is present only when the two F -layer magnetizations are

not collinear (or anti-collinear). This torque depends upon the direction of current ﬂow

22

and is characterized by an asymmetric R(I) curve. Current ﬂow in one direction
stabilizes antiparallel alignment of two magnetic layers and current ﬂow in the opposite
direction stabilizes parallel alignment. The switching from one alignment to the other
causes a jump in sample resistance, up in one case, down in the other. The dominant
physics underlying these phenomena is still a matter of debate among theorists
[15,21,22].

1.5. THE PRESENT THESIS

Aﬁer this introduction, we brieﬂy describe sample preparation and characterization in
Ch. 2 and provide background on theory in Ch. 3. The rest of the thesis then consists of a
variety of experiments as follows.

Ch. 4 involves a search for mean-free-path effects. As noted above, several searches
for evidence that mean-free-paths might be characteristic lengths in the CPP-MR did not
ﬁnd any clear evidence of such effects. However, in 2000, Bozec et al [88] claimed to
have seen ‘mean-free-path effects’. They used multilayers composed of alternating layers
of 6 nm and 1 nm thick Co layers (separated by thick Cu layers) having the two different
orderings described above: [Co(6)/Cu(20)/Co(1)/Cu(20)]~ = interleaved, and
[Co(6)/Cu(20)]N/[Co(1)/Cu(20)]1v = separated. Similarly to what was described above
for Co/Ag/Py/Ag and Co/Cu/Fe/Cu, they found qualitative and quantitative differences
between the MR curves of the two types of samples. Fig. 1.7 shows an example of our
conﬁrmation of the presence of such differences. Because of the relatively long spin-
diffusion lengths in Co (~ 60 nm) and Cu (~ 500 nm) [5,76], they argued that the
differences could not be accounted for by either the 2CSR model (which predicts that the

two curves should be the same), or by the VF model (since the spin-diffusion lengths in

23

Co and Cu are too long). They attributed the diﬁ‘erences to mean-free-path effects,
claiming that the difference is due to ‘real-metal’ effects that appear when the layer thick»

nesses are shorter than the As in the Co and Cu layers. In Ch. 4 we test these claims.

23 Y'Tjj'v" IIIIIIIII I 111111111
22»
p
p

 

     

.1

—

 

 

AR (f n m’)

 

 

 

 

17o 4W§6641oroo ...... 1 .500
”(09)

Fig. 1.7 ARC-I) for an interleaved sample (A) and a separated sample (B). Samples were
already taken to saturation in the —H direction. There are differences both in shape and

size of the change in AR with H.
Ch. 5 describes studies designed to enhance the CPP-MR. By enhancing CPP-MR,

we mean increasing the total speciﬁc resistance AR and/or the speciﬁc resistance change,
AAR. As described earlier, such enhancement would help to make the CPP-MR in
metallic multilayers more viable for devices. In this thesis, we tried to enhance CPP-MR
in three different ways. A) By alloying ferromagnetic Co with an impurity Zr that greatly
increases its resistivity, yet is not expected to reduce its spin-asymmetric scattering

parameter ,6. B) By introducing ‘internal’ interfaces within an F-layer, which we

24

expected to increase both AR and AAR without substantially increasing the total sample
thickness. The interface contribution is particularly significant when the resistivity of the
bulk metal is small, as in Co/Cu multilayers. C) By introducing a source of strong spin-
memory-loss into the superconducting layers of an EBSV. The arguments for why this
introduction might increase AAR are given in Chs. 3 and 5.

In Ch. 6 we determine the GMR properties of Ru and Co/Ru interfaces. Co/Ru
multilayers have acquired importance lately because of strong antiferromagnetic coupling
between Co layers separated by a very thin layer of Ru. This coupling is now used in
devices with structure: Co(3)/Ru(0.6)/Co(3)/Cu(tc“)/F, where F is any ferromagnetic
metal of suitable GMR properties, and thicknesses are in nm. This structure is called a
synthetic antiferromagnet and is characterized by a very stable anti-parallel magnetization
structure for the two 3 nm thick Co layers. The net moment in the two Co layers is zero,
which makes them insensitive to external magnetic ﬁelds, while a third F layer is free to
rotate in small ﬁelds. To provide the basic understanding needed to evaluate the potential
of Ru in CPP-MR devices, we study the spin-memory loss in bulk Ru and at the Cu/Ru
interface and the spin-dependant scattering at the Co/Ru interface. We will also discuss
other aspects of Ru in CPP devices.

In Ch. 6 we also use Ru and the CPP-MR to probe spin-dependent scattering at
several different interfaces.

In Ch. 7 we present some CPP-MR measurements on such samples at both room
temperature and 4.2 K. This work was intended as the start of a program of systematic
studies of current-driven excitations.

Ch. 8 contains a summary and our conclusions.

25

Chapter 2
SAMPLE PREPARATION, CHARACTERIZATION

AND MEASUREMENT

2.1 INTRODUCTION

The process of sample preparation, as well as characterization and CPP-MR
measurement in our F/N multilayered samples, varies substantially depending on sample
area. All samples in this study are prepared in our labs and fall into one of two categories:

1) Macroscopic samples, with area of current ﬂow A z 1.25 m2.

2) Microscopic samples, with area ~ 1x10'8 mmz.
Multilayers of both types are deposited by sputtering onto single-crystal Si substrates
with dimensions 0.5”x0.5” and a native SiOz thin layer on the substrate surface.

In addition to its importance to the preparation process, sample area is of key
importance because it determines the resistance of the multilayer. The two types of
samples also have very different contact materials. Macroscopic samples are sandwiched
between two crossed superconducting leads to provide a uniform current through the
sample. The large resistances of microscopic samples make it unnecessary to have
superconducting contacts, since noble-metal contacts have low enough resistivities
compared with the multilayer that the current in the multilayer should be near uniform.

We use the same sputtering system to deposit the multilayers of both the macroscopic

and microscopic samples. Except for some details, the same steps are used in preparing

26

both kinds of samples. The macroscopic samples are sputtered through a special mask,
rotatable in situ, that has three openings of different shapes and a blank ‘blocking
position’ that protects the substrate before and after sputtering. The microscopic samples
are sputtered through a mask with a single square opening plus a blocking position.

In the next section we discuss the sputtering process that is common to both kinds of
samples. We then describe characterization studies, also relevant mostly to both kinds of
of samples. Except for some data speciﬁc to the present samples, most of what we
describe here has been presented in prior theses [44,45] and publications [28,29]. We
conclude with a more detailed discussion of the preparation and characterization of
microscopic samples, a topic newer to this thesis.

The small (R~ 10'8 Q) resistances of the macroscopic samples are measured at 4.2K
with a Superconducting Quantum Interference Device (SQUID) based bridge circuit
described elsewhere [28,44]. The sample holder, including a cylindrical hand wound
superconducting magnet (capable of ﬁelds up to 1.5T [44]) around the sample, is lowered
into a standard 100 liter liquid helium dewar so that both the sample and the SQUID are
below the helium level. It takes about 20 minutes to cool the sample slowly in the
helium gas above the liquid helium and about 20 minutes to warm the sample up again to
room temperature in the neck of the helium dewar. This ability to cool and warm a
sample in well under an hour has led to the name ‘quick dipper’ for the sample holder.

The large resistances (R ~ 1 Q) of the microscopic samples allow them to be

measured at room temperature and 4.2K with a standard digital nanovoltmeter.

27

2.2 COMMON PREPARATION PROCEDURES

We start the fabrication process by cleaning the substrates with Alconox, rinsing them
thoroughly with water, and then cleaning them with acetone and alcohol, using ultrasonic
agitation in each step. The substrates are dried using nitrogen gas to remove any traces of
alcohol from the surface. Cleaned substrates are loaded into sample holders and placed
on a rotating sample positioning and movement assembly (SPAMA) plate in an ultra-
high-vacuum—compatible sputtering chamber. To achieve the desired low pressure, the
chamber is baked for a few hours to reduce the amount of water vapor, while a high-
speed cryopump pumps upon the chamber. A liquid nitrogen cold (Meissner) trap inside
the sputtering system freezes out water molecules both before and during sputtering, thus
further improving the effective background vacuum.

A chimney is placed over each target to collimate the particle ﬂow to the substrate,
eliminating cross contamination. A shutter is placed between the targets and the sample
holder, to open or block particles from reaching the substrates.

Four dc magnetron triode sputtering guns, plus two diode guns, give us the ability to
make sophisticated multilayers. Target materials are high-purity metals or alloys, mostly
purchased from commercial companies. A CuGe target was prepared in our laboratory
using RF melting. A CoZr target was originally a pure-Co target, into which we drilled
small holes and then inserted Zr pieces in the holes. We tuned the concentration of Zr by
adjusting the number of holes with Zr inserts.

Two ﬁlm-thickness monitors, placed on the same plate as the substrates, and at the

same distance from the sputtering guns, monitor the deposition rates from the different

28

targets. These rates are measured for all guns being used, just before each multilayer is
sputtered.

The motion of the sample plate, and the shutter assembly, are both controlled by a
computer, which also calculates the time needed for depositing different layers to get the
right thicknesses. The background pressure before admitting Ar gas is S 2x10’8 Torr.
The Ar sputtering pressure is 2.5 mTorr. During sputtering, the substrates are held at
temperatures (measured by a thermocouple on the SPAMA plate) between —3 0°C and
20°C. Each sample has its own separate mask, which protects the sample when other
samples are being sputtering. A mask for the macroscopic samples is shown in Fig.
2.1A. Opening 1 is used to sputter the bottom Nb lead, opening 2 to sputter the
multilayer, and opening 3 to sputter the top Nb lead. After the sample is ﬁnished, the
mask is rotated back to the initial blocking position. A vacuum-compatible wobble stick
allows in situ manual rotation of the mask into these four positions. Fig. 218 shows a
ﬁnished macroscopic sample, onto which Indium contacts have already been soldered.
Contacts to the sample from the SQUID-based measuring system (described elsewhere
[29]) are made by pressing superconducting leads into the Indium contacts. Simple
square CIP masks are used to prepare ﬁlms for Van der Pauw [90] measurements and for
sputtering microscopic samples.

Since the intrinsic quantity in the CPP geometry is the speciﬁc resistance, AR, we
must also measure the area of overlap A of the two crossed superconducting strips. We
do this using a Dektak surface proﬁler [44,45] to measure the widths of the two Nb leads.
The area A is then just the product of their widths. Uncertainties in determining these

widths lead to an uncertainty in the area of about :t 5%.

29

Blocking

Position

 

Fig. 2.1. (A) A CPP mask. A screw through the middle circular hole holds the mask in
place, but allows it to be rotated. The three openings are used to sputter the bottom Nb
(1), the multilayer (2), and the top Nb (3). The blocking position protects the sample
before and aﬁer sputtering. (B) A ﬁnished sample with Indium blobs on the Nb strips.

2.3 CHARACTERIZIN G SAMPLES

Samples are characterized by Electron Energy Dispersive X-Ray Scattering (EDX),
resistivity measurements, and magnetization measurements. The periods of repeated
multilayers are also checked with x-rays, examples of which are given elsewhere [44].

A) EDX Studies

EDX allows us to ﬁnd the concentrations of impurities in alloys. High-energy
electrons in an electron microscope hit the target ﬁlm of interest, and some excite core
electrons from target atoms. An electron ﬁ'om a higher energy level jumps down to the
vacancy, emitting an x-ray photon with energy characteristic of the particular initial and

ﬁnal orbits and, thus, of the atom producing it. The x-rays are detected by a

30

crystal kept at liquid nitrogen temperature, and analyzed to determine the target
composition. Sample x-ray spectra are shown in Figs. 2.2-2.4. Such spectra gave

concentrations of CuGe = 2.5 i- 0.2 at. % and CoZr = 2.5 i 0.2 at.% and 20 :t l %.

CIIK : CuK

1.00 2.00 3.00 {.00 5.00 6.00 7.00 0.00 9.0010.0011.0012.0013.0014.00

 

 

IDA! Phinhoz Quantification (standardlels)
Element Reunited

31mm: It 96 E 3 K— Ratio 2 A F

Cu! 97.22 97.56 0.9762 1.00.13 0.9998 1.0030
M 2.78 2.“ 0.0247 0.9522 0.9337 1.0000
Totll 100. 00 100. 00

 

Element Net Into . Bkgd Into . Into. Error P/B
M 196.27 4.16 0.71 £7.18
60: 2.57 3.10 9.13 0.83

 

 

 

Fig. 2.2. EDX spectrum of a CuGe ﬁlm. The table on the bottom shows the detailed
analysis of the spectrum.

31

'Zr

 

2.00 8.00 6.00 8.00 10.00 12.00 14.00 16.00 18.00

 

IDA! EhiRhoz Quantification (Standardloaa)
Ila-ant Ker-alizad

non-at It ’6 At X K-Ratio Z A I'

2th 3.84 2.52 0.0210 0.9299 0.5875 1.0000
COR 96.16 97.48 0.9610 1.0031 0.9962 1.0000
Totll 100. 00 100. 00

 

llanont Not Into. Bkgd Into. Into. Error P/B
Zrh 15.16 16.60 6.73 0.91
Cox 590.43 11.00 0.75 53.66

 

 

 

Fig. 2.3. EDX spectrum of a CoZr ﬁlm, with a small concentration of Zr. The analysis on
the bottom of the plot shows 2.5% Zr.

32

 

 

ZJL

 

 

 

 

 

 

Cox

 

 

 

 

 

 

 

 

 

 

 

 

 

 

I. m 7i
\ZrL % £1 1 ]\ RR

0.80 1.60 2.40 3.20 4.00 4.80 5.60 6.40 7.20 8.00

 

 

 

 

 

 

 

Inn: PhiRhoz Quantification (Standardlass)
llamant Normalized

Element It ’6 At X K- Ratio 2 A F

21L 26.70 19.05 0.1665 0.9450 0.6600 1.0000
COX 73.30 80.95 0.7302 1.0222 0.9745 1.0000
Total 100.00 100.00

 

llannnt Nat Into. 8ng Into. Into. Error PIE
Zrn 53.55 6.41 1.44 8.36
Cox 199.10 6.63 0.72 30.05

 

 

 

Fig. 2.4. EDX spectrum of a CoZr ﬁlm. The percentage of Zr given by the analysis is
19%, which is expected to be well in the amorphous regime.

33

B) Resistivity

We separately measure the resistivities of all components of the multilayers to
provide parameters for our ﬁts and to check for reproducibility of the sputtered metals.
We measure these resistivities by two methods: the Van der Pauw (V dP) method [44, 45]
for all metals, and by sandwiching F-metals between superconducting strips.

VdP measurements are made at both 4.2 K and room temperature (RT), the later used
to check for reasonable values of the ‘phonon-resistivity’. For Nb, we also measure its
superconducting transition temperature, Tc. Tables 2.1-2.4 contain detailed measured
VdP resistivities for most of the metals and alloys used in this thesis. Table 2.5 contains

a summary of the results with uncertainties (2 standard deviations).

 

 

 

 

 

 

 

 

 

 

 

Table 2.1. VdP resistivity and superconducting
transition temperature of sputtered Nb. Tc(Nb) bulk
z 9.2 K.

Sample# T Thickness (A°) Apacm) Tc (K)
1074-18 room 3000 23.1

1074-1a 12K 6.9

1180-8!) 12K 2000 5.5 9.1

1 1 90-7b 1 2K 3000 5.1 9.1 5
1201 J!) room 3000 19.7

1201-7b 12K 5.0 9.13

 

 

 

 

 

 

34

 

Table 2.2. VdP Resistivity of CuGe Films

169-81!
169-83
169-8b
169-8b

171-8b
171-8b

182-83
l82-8a
l82-8b
182-8b

190-81!
190-83
l90-8b
190-8b

Temp.

300
4
300
4

300
4

 

Table 2.3. Van der Pauw Resistivities of Co, Py, Cu and Ag
ﬁlms. Last Column shows if each sample was sputtered at the
beginnin or end of each sputtering run.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

leampleiilmateriai t(nm) cm em) Order of sputterinL
1150-83 Co 225 13 6.1 beginm
1150-8b Co 225 12.5 3.8 end
1151-8A Co 270 12.8 6 beginninL
1151-83 Co 180 13.2 6.7 end
1152-8A Co 300 11.3 4.1 end
1201-7a Co 300 11.6 4.8 belinm
1201-8a Co 300 11.4 4.4 end
1179-8a Co 200 11.5 4.9
1191-8a Co 300 11.1 4.6
1094-88 Py 300 11.1
1153-8A Py 300 21.6 10.6
1197-8a Cu 500 2.71 0.7 beginnirl
1197-8b Cu 500 2.68 0.8 end
1198-8a Cu 500 2.94 1.1 beginning
1199-8a Cu 500 3.15 1.2 middle
1198-8!) AL 500 2.49 0.8 end
1199-8b A 500 2.73 1.0 end

35

Table 2.4. Van der Pauw Resistivity of CoZr ﬁlms

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

[Sampieil hiekness (Aﬁ pﬂcm) mummy
(309752725
1007-81: 3000 20.6
1008-88 3000 23.5
1042-8a 3000 20.3
1043-8b 3000 20.25
1040-83. 3000 20.9
1040-8b 3000 20
1074-10 3000 28.9 21.1
1074-8b 3000 29.2 21.7
1089-80 3000 30.8 22.8
1106-5: 3000 18.6
1108-50 3000 17.7
Cosongg
1173-8a 3000 132.4 136.4
1173-80 3000 132.6 136.1

 

 

 

 

 

 

For CoZr, Bass [91] and Campbell and Fert [46] list a resistivity increase of 4 uﬂcm
per percent of Zr impurity, but this value is uncertain since it comes from only a single
source. Assuming additivity of residual resistivities (Matthiessen’s rule for alloying), and
our measured value of ~ 5 110cm for poo, our estimated values of pea;r for our nominally

2.5% Zr suggest a higher value of about 5 uQcm per atomic percent Zr.

36

Table 2.5. Average resistivities (in uQcm) both at 4.2 K and room temperature.

 

 

 

 

 

 

 

 

 

Material p(4.2K) p(RT2 Ap =p(RD - p(4.2K) pphomn [91]
Nb 5.1 i 1.6 20.6 i 1.8 15.5 14.4
CuGe(2.5%) 8.0 i 0.9 10.0 i- 1.0 2.0 1.65

Cu 0.96 i 0.46 2.86 :i: 0.50 1.9 1.65

Ag 0.9 i 0.1 2.55 i 0.15 1.65 1.6

Py 10.9 10.3 21.9-10.5 11

Co 5.1i1.9 12.1 i 2.2 7 5.5
CoZr(2.5%) 21 i 2 28.8 i 2 7.8 5.5

Ru z 8.5

 

 

 

 

 

 

 

The residual resistivity at 4.2 K comes from scattering by imperfections in the metals
and impurities in the alloys. p(RT) has an extra contribution ﬁom scattering by
phonons. The third column in Table 2.5 lists the measured increase in resistivity (Ap) for
each metal. The fourth column lists the room temperature phonon resistivity (pph) for the
high purity metal from [91]. In all cases, Ap and pm, are similar.

One exception from the observed increase of resistivity with temperature is CosoZrzo.
For the CosoZrzo alloy the room temperature resistivity is smaller than that of liquid
helium temperature. Combined with a resistivity, p > 100 pﬂcm, this behavior suggests
an amorphous structure of those samples.

For F-metals or F-alloys, we also measure AR vs I for a sandwich structure of the
form Nb(100)/F (t)/Nb( 100). The slope of the AR versus t gives the F-metal resistivity pp,
and the ordinate intercept is 2ARWF. The sandwich method is limited to 4.2 K by the use
of superconducting Nb leads. Fig. 2.5 shows that any magnetic ﬁeld dependence of AR
for sandwiched single F-layers is small. Fig. 2.6 shows AR vs tCoZr for CoZr inserts. The
CoZr plot shows that results with and without 10 nm inserts of Cu between the CoZr and

the Nb are similar, strongly suggesting that the Cu becomes superconducting by

37

proximity with the superconducting Nb. Since the differences between data with and
without Cu are modest, we ﬁt both kinds of samples together to a single straight line, the
slope of which is pcoz,= (19 i l) uﬂcm. Happily, this value is consistent with the value
of malt: 21 2t 2 uﬂcm found from VdP measurements (table 2.5). This consistency is
heartening, since the CPP and CIP resistivities of layers need not be identical, because of
the columnar grth of the layers perpendicular to their layers [5]. The intercept with
the ordinate axis is 2ARCoz,/Nb= 5.8 i 1 f sz, a value compatible both with the
independent value of 6.5 :i: l f sz for Co/Nb shown in Fig. 2.7 and with earlier values
[5]. Finally, Fig. 2.7 shows AR vs tea for Nb/Co/Nb sandwiches for two different
sputtering runs. The intercept with the AR axis is already discussed above, and the slopes

give poo: 3.9 i 0.7 uﬂcm, consistent with the VdP value, pCO= 5 .1 i 2 uQcm.

 

30 l ' I 7 77 i 17 f q

. i
25 0' e e e e e a e e e e e ‘0
20 r tcw= 100 nm (K1072-1) _

10L; I I I I I I I I I I I T

tcw= 20 nm «(1071-4)

 

 

4L1 1911141111:

-300 -200 -100 o 100 200 300
H (De)

 

Fig. 2.5. AR of two different thicknesses of CoZr in Nb/CoZr/Nb sandwiches.
Any dependence of AR on magnetic ﬁeld H is small.

38

00
O

 

 

 

 

 

 

 

: ----- AR= 5.8 + 0.19 t -";
25 _- ;. -
A : ’ooo' -
9' 20 1 x" j
E :
C. _ f
.h 15 ",4?
V '0'
m 10 — go"! 1
< 'x" o NbICoZrINb i 3
.1 I NbICuICoZrICuIN .
5 f ----- LineFit, Both j
o #4 . 1 1 r n J 1 . . r 1 A 1 . 1 r r A 1 . r .
0 20 40 60 80 100 120
t(nm)

Fig. 2.6. Dependence of AR on 1002:- Samples both with and without Cu are used in the
straight line ﬁt. The ﬁt gives Axum? 5.8 :1 1 mm2 and peozF(19 :i: 1 pﬂcm).

 

    

 

 

 

 

' 1 —AR= 6.5+0.036t
16 - co 0 -
. '-"-AR=6.4+ 0.042t
Co
"E 12
c: C
h >
m 8
< ’ e
» O .
4 _ w. Nb(100)ICo(t)INb(100)-
. +RunK1137§ .
"I”Run K1164§
o .421M..11-..n1...9141..
0 60 100 150 200 250
tc°(nm)

Fig. 2.7. AR versus too for Nb/Co(tc°)/Nb sandwiches. Data are shown for two different
sputtering runs. The two straight lines are best ﬁts, giving an average value of ZARNb/CO =
6.5 :i: 1 fﬂmz, and poo = 3.9 i 0.7 poem.

39

C) Magnetization

Finally, we also used magnetization measurements to determine the magnetic states
of multilayers, switching ﬁelds, and to search for magnetically dead layers. Some
magnetization studies are presented in Chapter 4, and the results of a search for

magnetically dead layers in CoZr/Cu and CoZr/Ag multilayers will be described in Ch. 5.

2.4 MICROSCOPIC SAMPLES

Microscopic samples need far more care in preparing and handling than do large
ones. After the sample is sputtered, most of the subsequent preparation is carried out in a
class 100 Cleanroom facility in our department. Cleanroom atmosphere is required for
such small samples, since a tiny piece of dust can ruin the whole device. Once samples
are taken into the Cleanroom, they are only taken outside for ion milling followed by
evaporating a SiO insulator and top contacts. Preparing microscopic-area samples
involves a series of lithographic steps as well as thermal evaporation, sputtering, and ion
milling.

Equal care has to be exercised while measuring the resistance of the finished pillar-
shaped microscopic samples. Their high resistance and small size make a small
instantaneous current ﬂow due to the discharge of static electricity from the operator
enough to destroy the sample. To avoid this possibility, all features are shorted until the
connections to the measuring equipment are established. Only then is shorting removed
and the measurement process started. Both before and during measurements, anyone

handling the samples must be grounded.

40

In the following, we discuss brieﬂy each step in the preparation process of the small

samples, and characterization and measurement procedures.

2.4.1 Preparation

Step 1. Photolithography (Patterning the large Pads)

Before the multilayer sample is sputtered, current and voltage electrodes must be
deposited onto the substrate. We use clean, single-crystal (100) Si wafers of 3” diameter
to deposit electrodes simultaneously on as many as 18 of the 0.5”x0.5” substrates. We
begin the process by spinning S1805 photoresist (PR) to get a thickness t ~ 500 nm. We
then bake in an oven in ambient atmosphere at 95C for 45 minutes to remove the solvent
material and harden the PR layer. The PR is sensitive to UV light, which breaks chains
between molecules, making the exposed material easy to dissolve in a liquid called the
developer (KOH). PR that has not been exposed to UV is less soluble in the developer.
The wafer is then placed under UV (~3 seconds) through 8 Cr mask that blocks light from
the substrate, except for the areas chosen to be the current and voltage pads. After
selective exposure of the features, the wafer is immersed in the developer # (452) for 45
seconds. The developing process washes out the exposed PR, opening the Si surface to
the outside but leaving the unexposed areas intact. After developing, the wafer is placed
in a Phillips Thermal Evaporater where we evaporate ~ lnm of Ti followed by ~ 70 nm
of Au under a vacuum ~ 2x10“7 Torr. The Ti layer is to improve the adhesion of the

deposited Au to the Si substrate, since Au does not adhere strongly to SiOz.

41

We then put the wafer in acetone for some time. This step is called lift-off. Acetone
dissolves the PR layer, washing out the metal layers on top of it and leaving only the
metal layers (Ti/Au) that have landed directly on the Si surface.

At this stage the sample looks like Fig. 2.8. The light parts are Au electrodes that will
carry the current to the sample and will be used to measure the voltage drop across the

sample. The dark parts are where the Si substrate is not covered by Au.

 

Fig. 2.8. Au electrodes deposited on a 0.5”x0.5” Si substrate using photolithography.
The outside pads are large so that measuring electrodes can be connected to them. The
small samples will be deposited in the middle of those electrodes. All the Au pads are
connected to avoid shocking the sample; these connections are removed when the sample
is ready to be measured. The crosses are alignment marks for subsequent steps. .
The crosses in Fig. 2.8 are alignment marks that will be crucial in subsequent electron-
bearn lithography steps, where features must be aligned to within ~ 0.1 pm.
Step 2. Depositing the Multilayer

Since all our multilayers are prepared by sputtering, which produces an uncollimated

ﬂux of atoms, we need to have a large undercut. To obtain this undercut we use a bilayer

of electron beam resist, Copolymer 9%l PMMA 4%, where we ﬁrst spin the Copolymer

42

9% for 60 seconds at 2500 RPM, bake it at 147° C and then expose it to an e-beam over a
chosen area #1 (enclosed by the dashed line in Fig. 2.9A). This exposure step is followed
directly by spinning PMMA 4% for 60 seconds at 3000 RPM, baking at 147° c for at
least 1 hr, and then exposing the bilayer to an e-beam for a second time, but now over an
area #2 (enclosed by the solid line in Fig. 2.9A), which is smaller than area #1 by 1 pm
on each side (see Fig. 2.9). Note that there was no developing after the ﬁrst exposure.
Proper alignment is essential for this and all other steps. The next step is developing,
where the sample is dipped in EBL developer for 65 seconds, followed by Isopropanol
for 30 seconds then water for 25 seconds. EBL attacks both PMMA and Copolymer,
while Isopropanol mainly attacks Copolymer, adding to the undercut size. Since the
bottom layer was originally exposed to the e-beam over a wider area than the top layer,
all area #1 in the bottom layer gets dissolved, while the top only develops to the limits of
the smaller area #2. This procedure gives a 1 pm wide undercut, large enough for
sputtering the multilayer and successful lift off, without any sharp edges that can make it
impossible to deposit top contacts properly. If desired, the 1 micron undercut could be
varied just by changing the borders of areas one and two in Fig. 2.9. Fig. 2.10 shows a
picture of the sample at this stage, with the light areas around the developed feature
indicating undercuts. We check for the presence and extent of the undercut by examining
the sample under a photo-microscope. As an alternative, 8 trilayer layer technique
involving Copolymer 9%/Al/PMMA 4%, was tried for this step. But the above technique
proved to be easier and better for subsequent liﬁ off.

The shape of the multilayer is chosen to allow for a large contact area with the bottom

leads to reduce the contact resistance, but narrower in the middle to minimize the overlap

43

area between the bottom portion of multilayer left aﬁer ion milling (bottom electrode)
and the top-contact leads. The smaller this overlap area, the less the chance of pinholes

through the SiO insulating layer.

3 (B)

\\\\\\\\\\\\\\\ ‘

|

 

Fig. 2.9. (A) Schematic diagram of the double exposure to the electron beam. Aﬁer
baking the ﬁrst layer, all the area enclosed by the dotted line is exposed to electrons. The
area enclosed by the solid line is then exposed to electrons after spinning and baking the
second electron-beam-resist layer. After developing, the part of the second layer between
the dotted line and the solid line will be ‘free hanging’, creating an undercut. (B) shows
what the substrate should look like after developing. The shaded area is the undercut.

 

Fig. 2.10. Photomicroscope picture showing the undercut (discoloration of the top layer)
of Fig. 2.9B. The electron-beam resist is washed out completely from the center area
where the multilayer will be sputtered.

After developing, the sample is loaded into the sputtering system, and the multilayer
is created. Starting from its bottom, the structure of the multilayer is: thick noble metal
electrode/ thick bottom Co layer/ thin noble metal layer/ thin Co layer/ thick noble metal
top cap. The bottom noble metal layer will not be ion milled and will provide the bottom
electrode for the transport measurement. To simplify processing, the thick bottom Co
layer will be only partially milled.

A shadow mask is used to sputter only on the middle portion of the sample, making
the lift off easier. For lift off, samples are inserted into Acetone for nearly 1 hr, during
which everything gets washed out except the multilayer in the area within the solid lines
in Figs. 2.9A and 2.93. Preparing the multilayer by sputtering is advantageous over
evaporation, since the poor collimation leads to very smooth edges, which helps make the

insulator cover the multilayer properly and make the top pads continuous.

45

Step 3. Ion Milling Mask (Al/Si bilayer or Al/Ti/Cu trilayer)

The next step is to create a sub-micron sized mask to protect the portion of the
multilayer underneath, while unmasked parts are ion milled down to within the bottom
Co layer. This mask has to be created using electron beam lithography and thermal
evaporation. The ﬁrst step is to spin a layer of Copolymer 9% at 3000 RPM for 60
seconds followed by baking at 147 0C for 1 hr. Then we spin a layer of PMMA 2% at
5900 RPM for 60 seconds and bake it at 147 0C for at least 1 hr. The use of PMMA 2%
rather than 4%, and high spinning speed, are to minimize the thickness, to improve
resolution and ultimately the size and shape of the mask feature. Five spots are then
exposed to electron beam (since we have ﬁve masks or pillars on each multilayer).
Careful focusing and correction for astigmatism are crucial in this step. The sample is
then developed exactly as in step 2, and the mask is evaporated. Historically, this mask
was either a bilayer of Al/Si or a trilayer of Al/Ti/Cu. Upon dipping the sample in
photoresist developer (KOH), the Al dissolves much faster than the other materials,
creating a mushroom-like undercut as shown for an Al/Ti/Cu trilayer in ﬁgure 2.11A, B,
and C. It takes only 3 to 6 seconds to create a decent undercut. The creation of the
undercut helps get a good top contact, as will be seen in the next section. Unlike in step 2,
here we expose the bilayer of electron beam resist only once, since the mask material is
evaporated and not sputtered.

Figs. 2.11 and 2.12 show an Al/Ti/Cu mask at this stage and a cartoon showing the

ideal shape of the mask. The elongated shape of the mask is to impose shape anisotropy

46

 

Fig. 2.11. SEM images of the Al/Fi/Cu masks. Masks were dipped in PR developer for 5
seconds. Ideally the mask should look like the mushroom shape shown in Fig. 2.12. In A
and B the substrate was tilted by 30°. In C there is no tilt.

47

Al/Ti/Cu mask

'/ M “nil-Yer

Substrate

   

 

 

WM . .

, 2.-.... .mﬁaaxgkunannanauﬁ-nmu 1.5.132“... .-,. ...- .3... .1... ,- ._ .1 .l. .-.

Fig. 2.12. Cartoon of the ideal shape of the Al/T i mask after the short wet
etchinginPRdeveloper. Ideally the mask would look a mushroom to protect the Al part
from SiO to be able to dissolve the mask after the ion milling and SiO evaporation.
on the magnetic layers, which will take the shape of the mask, so that we have a single
easy axis along the longer side, to control the magnetic states of the multilayer during
measurements.
Step 4. Ion Milling and Insulator Evaporation

At this point, we could in principle, start ion milling the multilayer to shape the pillar-
like samples. However, gold ion mills much faster than the multilayer, and our 75 nm
thick gold electrodes would disappear before the sample was ion milled. To solve this
problem, we protect the pads with a bilayer of electron beam resist, opening a small
window through the resist around the multilayer by exposure to electrons and developing.
All the area within the window gets ion milled, including the multilayer, but the covered
gold pads remain protected by the electron beam resist. The electron beam resist bilayer
is made of Copolymer 9% / PMMA 4% , baked, exposed, and developed as in step 3,
except that the exposed area is the window that deﬁnes the ion milling area. During ion
milling, the sample has to be anchored to a good heat sink, so that the sample temperature
doesn’t get too high. Overheating the sample can burn the electron beam resist, making it

difficult to dissolve in acetone during lift off. Ion milling is performed after reaching a

48

background pressure ~ 10'7 Torr, and samples are loaded using a magnetic arm to avoid
breaking the vacuum. The Ar pressure during the ion milling is 3x10‘4 Torr, and the
voltage and current readings are 500 Volts and 20 mA, respectively. To insure
reproducibility in our ion milling rates, we delay the milling until after the ion mill
stabilizes. Unmasked parts of the multilayer are then ion milled all the way down to the
bottom Co layer, which is only ion milled partly through. Leaving the bottom Co layer
unpattemed helps controlling the magnetic state of both Co layers since it reduces the
magnetostatic dipolar interaction that favors antiparallel coupling between patterned Co
layers. The depth to which we ion mill through the multilayer is controlled by
independently measuring the ion milling rates of all constituents of the multilayer. We
measure the ion-milling rate for each metal by preparing ﬁlms, spinning, and baking
photoresist on it, exposing to UV radiation through a photo mask, then developing in
KOH. Each ﬁlm is then ion milled for some time, dipped in acetone to remove the
photoresist, and the depth is measured using AF M or a Dektak. The ion-milling rate is the
slope of the line of thickness versus time. Fig. 2.13 shows the ion milling results of Cu
with a straight-line ﬁt. The straight line does not go through zero but intercepts the tirne-
axis, which seems to indicate a Cu oxide layer that ion mills slower than Cu. Table 2.6
shows the ion milling rates of other materials that we have measured. The error is taken
from the scatter in the data (twice the standard deviation), or by trying two extreme ﬁts,

whichever is larger.

49

 

 

160 'Hirirﬂi'ir'li'ﬂlii"Iii'iéi'w‘,

. ,v :

.— ----- Depth = -13 + 25.5t / 1

En”? —:

1:: g E." 5

v - I :

i: 80: {a :
H

e * x! "

Q 40~ 3' _, *1

' ion Milling Rate ofCu :

L {-3 j

.1' .1

 

 

0 1 2 311415L‘6H 7
time(min)

Fig. 2.13. Ion milling thickness as a function of ion milling time. The slope gives the
ion-milling rate in nm/min. The intercept with the horizontal axis indicates the
existence of an oxide layer that ion mills slowly on the Cu surface. The ion milling

rate of Cu is (26 i l) nm/min.

Table 2.6. Ion-Milling rates of
materials used in the lab

 

 

 

 

 

 

 

 

 

 

 

material Rate (nmlMin)
Cu 26 :i: 1

AI 11 :t 2

Au 38 i 2

Co 8.8 i 0.4
FeMn 7 :t 2

Py 5.5 :1: 2.5
Nb 3.0 :1: 0.2

 

After the multilayer is ion milled and the pillar shapes created, we evaporate SiO onto
the sample in the same chamber with the ion mill. The SiO is evaporated while the

sample is rotating at 200 - 300 RPM, to minimize the chance of pinholes through the ~ 35

50

nm thick SiO layer. After evaporating the insulating SiO layer, the sample is dipped in
acetone to wash out both the electron beam resist bilayer and the SiO over it, leaving SiO
only in the window that was deﬁned earlier by electron beam lithography. Finally, the
sample is dipped in KOH for a few minutes, in hopes of completely dissolving the Al
layer and exposing the top capping metal on the now pillar-shaped multilayer.

The sample at this stage is shown in ﬁgure 2.14. The dark rectangle surrounding the
multilayer is the window opened using electron-beam lithography, which has been ion

milled and then covered by a SiO insulating layer.

Multilayer

Alignment
marks

SiO

 

Fig. 2.14. Picture of a sample after the ion milling of the multilayer (the line in the
middle) and depositing the SiO insulating layer. SiO is the dark rectangular window in
the middle. There are four crosses on the comers of the picture (inside the dark squares)
that serve as alignment marks during the electron beam exposure. These alignment marks
are covered by SiO, which makes them less visible.

5) Top Contact

The top pads are also patterned using electron beam lithography. A bilayer of
Copolymer 9% / PMMA 4 % electron beam resist is spun, baked, and exposed as in steps

3 and 4. We try to make as large as possible overlap or contact area between those top

51

pads and the original electrodes laid down using photolithography. The top pads are
chosen narrow (3 microns), to reduce the chance of a short with the bottom electrode
through possible pinholes in the insulator. After developing the sample, it is taken to the
ion milling chamber, where a quick ion mill is performed to expose a fresh metal layer on
top of the pillar capping layer. This milling is followed directly by evaporation of the top-
contact metal, either Ag or Au, without breaking vacuum between the ion milling and
evaporation. Evaporating immediately is important to minimize any oxide layers that
would give a high contact resistance. As seen in ﬁgure 2.15 the only places where the top
pads touch another metal are at the point of contact with the pillar multilayer, in the
middle, and at both ends where they touch the extension pads. Only the middle portion of
the sample is left open for the quick ion milling and for evaporation, since the whole
sample is covered with an Alurninurn-foil mask with 8 ~ 2mm diameter hole in the
middle. This masking makes lift off easier and faster. After lift off and removal of the

electron beam resist bilayer and the metal covering it, the sample is ready to be measured.

52

139-13;.

.: -.
r :1 x
are ”imiic‘iﬁ ‘ -
A ‘ . . {.1 -_ ~ .-
' ,;...

 

Fig. 2.15. A ﬁnished device ready to be measured. Each of the top contacts (5 Ag
slabs) connects one of the top Au ﬁngers with one ﬁ'om the bottom and, hopefully,
has electrical contact with the top layer of the pillar in the middle. The pillars appear
as dark spots in the middle of the overlap between the top contact and the multilayer.

2.4.2 Characterization

Characterization is most important during the lithography steps, as a lot of time can
be wasted on undiscovered bad samples. Most characterization is done using the optical
microscope in the clean room to inspect samples and check for proper developing time,
adequate cleanliness, proper lift off, and basically any step in the lithography process.
When trying to create the mushroom-shaped mask, sometimes the whole mask gets
dissolved and the step needs to be repeated. Failure to do so would waste much time on a
sample that can never work. Figs. 2.8, 2.10, 2.11, 2.14, and 2.15 are examples of pictures

taking during preparation.

53

2.4.3 Measurement

As seen from Fig. 2.8, the Au pads are initially all shorted together. After the sample
is prepared and ready to be measured, it is mounted in a quick dipper that has all of its
terminals shorted and grounded as a precaution to not burn out the samples. The quick
dipper has six current wires and six voltage wires, each of which can be shorted to the
ground by a switch. The operator is also grounded by a bracelet to discharge any static
electricity. Silver paint is used to connect the current and voltage wires to the Au pads on
the sample. After the silver paint dries, the shorting connections between the different Au
pads ( see Fig. 2.8) are opened, after which they are electrically connected only through
the pillar-shaped multilayer. After the Quick Dipper is attached to the current source, and
the nanovoltrneter switched on, the particular current and voltage terminals to be used are
opened to let current ﬂow through the multilayer. The magnetic ﬁeld, H, used to measure
the R(H) curves at room temperature is provided by an electromagnet with a computer
controlled power supply. For R(H), the voltage across the sample is measured six times;
once at zero current, then two repeats at pre-selected positive current and the same-
magnitude negative current, and ﬁnally at zero current. This procedure eliminates the
effects of any offset in the nanovoltrneter. Switching of current direction for each
measurement cannot be used for current—driven magnetization switching, because the
current direction must be ﬁxed. For measurements at 4.2 K the quick dipper is inserted
into the liquid-heliurn dewar and a superconducting magnet is used to provide the

magnetic ﬁeld.

54

Chapter 3

THEORETICAL BACKGROUND

As described in Chapter 1, theoretical analysis of transport in PM multilayers began
with a phenomenological two-current model by Baibich et a1. [1] in which electron spins
never ﬂip. Subsequently, more rigorous theories, both semi—classical and fully quantum
mechanical, were developed [3,4], some including spin-ﬂipping. Initial models
addressed the CIP-MR, but since we study the CIP-MR only peripherally in this thesis,
we do not go into those models, referring the interested reader to appropriate reviews [3].
Rather, we focus on analyses of the CPP geometry, where higher symmetry usually
simpliﬁes the analysis [3,4,26,47]. Since all of our experimental data are taken on
samples with collinear magnetizations, we limit our discussion to calculations for which
the layer magnetizations are mono-domain and collinear. It is convenient to divide the
discussion in this chapter of models of the CPP-MR into two categories: free-electron
based models involving diffuse scattering—essentially semi-classical models neglecting
quantum size effects; and real Fermi surface based models, which are inherently
fundamentally quantum mechanical. We describe the main features of, and issues
underlying, these models in sections 3.1 and 3.2, respectively. In section 3.3, we provide
a brief description of the concepts underlying a new phenomenon that can be viewed as

the inverse of the CPP-MR, current-driven magnetization reversal.
3.1 FREE-ELECTRON-BASED SEMI-CLASSICAL MODELS

As ﬁrst pointed out by Zhang and Levy [26], free electron analysis (i.e., assuming

single parabolic conduction bands for both the F - and N-metals, but different scattering

55

times along with spin—dependent scattering in the F -metal), for the case of diffuse
electronic transport and neglecting spin-ﬂipping, together lead to a 2CSR model for the
CPP-MR [26]. In this model, the only lengths are the layer thicknesses t1: and m, and all
of the parameters (both ‘bulk’ and ‘interfacial’) that we deﬁned in Chapter 1 are
independent of the layer thicknesses. Valet and F ert (VF) used a Boltzmann equation
formalism to extend this model to include spin-ﬂipping within the N- and F -metals [47],
adding as parameters the spin-diffusion lengths in the F - and N-layers, 1;, and 1;. . Later
studies based upon the Kubo formalism or the Landauer-Buttiker formalism conﬁrmed
these results under the conditions speciﬁed. Another way to describe these models is to
say that the 2CSR and VF models are expected to apply if the transport is diffuse and if
scattering from adjacent interfaces is completely incoherent. For details see [3 ,4].

In Chapter I, we wrote down the 2CSR model equations (1.1-1.3) for a simple [F/N]N
multilayer. For ARA? we gave ﬁrst the equation (1.1a) neglecting contacts, and then the
equation (1.1b) including superconducting contacts. Eqs. 1.2 and 1.3 remain the same
with or without the superconducting contacts, because such contacts just add a constant
term to AR Ap.

In this section we analyze in detail the 2CSR model equations for a more complicated
hybrid multilayer of the form [F1/N/F2/N]N (see cartoon in Fig. 3.1). We choose this
more complicated multilayer because the simple [F/N]N multilayers were worked out in
detail in other dissertations [44,45], because hybrid multilayers are used repeatedly in this
thesis, and because the hybrid case has important differences with the simple case. The
essential equations for a hybrid multilayer were ﬁrst written down in [92]. We assume

that the two F-metals have switching ﬁelds different enough to produce AP alignment, as

56

discussed in Ch. 1. Fig. 3.1 shows the AP state, with arrows indicating the direction of

magnetization of each metal.

 

ccccccccccccccccc
aaaaaaaaa
aaaaaaaaaaaaaaaaa
IIIIIIII
eeeeeeeeeeeeeeeee

 

 

 

eeeeeeeee
eeeeeeeeeeeeeeeee

 

aaaaaaaa
aaaaaaaaaaaaaaaaa
---------
ooooooooooooooooo
eeeeeeee

 

 

 

eeeeeeeee

 

aaaaaaaa

 

IIIIIIIII

 

 

 

 

 

 

 

 

 

eeeeee --------
.................
.........
.................
........

 

 

-----------------
uuuuuuuuu
aaaaaaaaaaaaaaaaa

eeeeeeee
eeeeeeeeeeeeeee
a I a a

 

 

 

 

aaaaaaaaaaaaaaaaa
000000000
eeeeeeeeeeeeeeeee
eeeeeeee
nnnnnnnnnnnnnnnnn
aaaaaaaaa

 

 

 

eeeeeeeeeeeeeeeee
--------
IIIIIIIIIIIIIIIII

 

 

 

 

 

 

Fig. 3.1. Cartoon of an [F1/N/F2/N]N multilayer sandwiched between two
superconducting layers, the arrows indicate the magnetization directions of the magnetic
layers, showing an AP alignment.

The total CPP speciﬁc resistance, AR, in each of the AP or P states equals the parallel
sum of speciﬁc resistances of the two spin channels; the speciﬁc resistance of each
channel is the series sum of resistivities of individual layers times their thicknesses, plus
the interfacial speciﬁc resistances.

AP State

In the AP state the total speciﬁc resistance for spin up electrons is:
_ T 1‘ T T 1 1
AR“, (up) — ZARSA, + 2NpNtN + Np“!Fl + 2NARH,N + antpz + 2NARF2,N .
Using the conventions for pg'¢ and ARE: in Chapter 1, and assuming that

ARSTW =AR§,F =2ARS,Fand pg=p1t =2pN ( i.e. no spin asymmetry in electron

scattering in the normal metal or at the interface with the superconductor), gives :

57

AR..- (up) = 4ARs1r + We. + 2Nparaa - 20+ 4NARan11— r.) +
ZNPiztrz <1+ﬁz)+4NAR;2a (1+72)

= 2(2ARS,F + 211w:N + Npl'ntn + 2NAR;1,N + 1\/p;,tF2 + 2NAR;,,N)
" 210501301161 ‘1’ 2NARF1/N7i ‘ (Npi‘zt 11.52 + 2NARF2/N7 2 )1

Similarly:

AR AP (down) = 2(2ARS,F + 2NpNtN + Npi‘rtrr + 2NAR;1 ,N + Nplf-ztn + 2NAR;2,N)
'1' ZinFlt F1161 ‘1‘ ZNARir/Nl’ 1 " (Npizt 17.52 '1' 2NARi‘2/N7 2 )1

The net speciﬁc resistance in the AP state is given by:

1 = 1T + 1¢
ARA. ARAP ARM,

 

 

OI'

ARM. = 271125,, + 2111er1,, + Npgnn + 211/ARI.”N + Nat-21F, + 217/11,22,N

[pFltFlﬂi '1' ZARile’r ‘(Piz’rzﬂz + 2ARFZ/N72 ”2 (3'1)

- N2 0 e
2.119,,F + 2NpNtN + antn + 211/ARFl ,N + Npgztp, + 2NAR;,,N

P State

The same analysis for the AP state also applies for the P state giving:

AR, (up, down) = 2(2ARS,F + 21w)”:N + NpgltFl + 211/.4102“N + Np,‘.,tF2 + 2NAR;,,N)
:1: 2N [pFltFiﬂi ‘1' 2Akin/1471 '1' P92117162 1‘ 2Akin/0172]

Thus:

ARp = 2.423,. + 211mm:N + Np;,t,., + 2NAREW + Npﬁztn + 211141;,N —

N2 [pFitFlﬂl +2ARFl/N71+p;‘2tl’2ﬂ2 +2AR;-2,N72]2 (3-2)
2.419,,F + 2NpNtN + Np;,t,., + 2NAR£W + Np;.,tF2 + 211/.4152,N

AAR = ARAp - ARp is then given by [92].

(mt-11F, p, + 2AREi/N71Xpiztp2ﬂ2 + ZAREaer) (3 3)

AAR = 4N2 ‘1 4- e I!
ZARS/F ‘1‘ 2NpNtN '1' NpFltFl + ZNARFUN + Nppztn + ZNARn/N

58

Much of this thesis is built around this equation, which allows the applications potential

of a given multilayer to be evaluated by estimating the sizes of its ARA]! and AAR.

Examining the equation reveals a few important features:

1)

2)

3)

The only length scales are tn, (1:; and (N.

i) The last term in equation 3.1 is very small (a 0.1 m2 for
[Co(6)/Cu(20)/Co(l)/Cu(20)]6, layer thicknesses in nm) and can be ignored,
making ARAp equal to the denominator of equation 3.3 for AAR.

ii) Doping the N layer with a dopant that increases the resistivity and leaves the
spin-diffusion-length 1g: unchanged increases ARM: by the difference in
resistivity multiplied by the thickness of the normal metal, and decreases AAR
accordingly, but does not make any other difference.

If one of the ferromagnetic metals (or interfaces) has negative asymmetry
parameters (both ,6 and t5), and the other has positive ones, AAR becomes negative
and the CPP-MR becomes ‘inverse’—i.e., ARAp < ARp. If one metal has positive
parameters and the other has one positive and one negative, then whether AR is

normal or inverse depends upon the thickness of that metal—i.e., upon the

relative magnitudes of ﬂFpEtF and 27m, AR;N for that F-metal.

4) AAR can be enhanced by any of the following:

i) Doping F metals with impurities that increase ,8 and/or pi.

ii) Creating more interfaces in the multilayer. (This is most efﬁcient in
Co/Cu systems where the bulk Co contribution is small).

iii) Eliminating terms that contribute to the denominator but not the

numerator (e.g. ZARs/r).

59

The 2CSR model assumes no spin ﬂipping anywhere in the multilayer. We now
consider a simple [F/N]N multilayer as in Ch. 1, but when the spin-diffusion lengths in the
F-metal, 1;, , or the N-metal, 1;} , or both, become comparable to the layer thicknesses I}:
or m. The equations must then be generalized as described by VF [47]. The VF analysis
should apply at low temperatures, where electron-magnon and electron-phonon scattering
are both ﬁ'ozen out, so that there is still no mixing of the two-currents. However, spin-

ﬂipping due to spin-orbit or spin-spin scattering is now allowed.
VF showed that, in the limit where 1st >> 4r and 13. >> 2N , the Boltzmann equation

for CPP transport through a multilayer reduces to a pair of macroscopic transport

equations (with corrections of order 11131:, which are usually expected to be small in

 

metals).

id’s =;’;—;:S_ 34

0'8 62 182 ( ' )
-3325

Js— e 62 (3.5)

Here as is the conductance, ,US the electrochemical potential, 15 the spin-diffusion length,
and J3 the current density, all in channel 5 (s = up or down). Each of the above equations
applies locally to every individual layer in the multilayer.

The ﬁrst equation is a statement of the balance between spin accumulation and spin
ﬂipping over the length scale 15, and the second equation is Ohm’s law. The above
equations include only macroscopic parameters (i.e. as and Is) and are, accordingly,
called macroscopic equations. Eqs. 3.4 and 3.5 are to be solved in each layer and the

solutions matched at the layer interfaces to ﬁx the integration constants involved.

60

The interfacial boundary conditions are:

Js(z=zi+)=JS(z=zi‘) (3.6)

ys(z=zi*)-,us(z=zi')=ARsJS(z=zi) (3.7)
where 2, is the position of the ith F/N interface.

Equation 3.6 assumes no spin ﬂipping at interfaces (i.e. the components of current in
each spin channel are conserved at F/N interfaces) and equation 3.7 expresses the
discontinuity of the electrochemical potential due to interface resistance (i.e. the change
in electrochemical potential equals current, J times speciﬁc resistance, AR, which is just
Ohm’s law). Solving the coupled equations 3.4 and 3.5 subject to boundary conditions
3.6 and 3.7 in each layer yields complicated expressions for CPP resistance, even for the

special cases of P or AP. For inﬁnite multilayers VF derived the following equations:

 

 

 

 

 

 

AR(P AP) = N (AR0 + 2AR(S,,l AP)) (3.8)
Where 4R0 = (1-ﬂ2)pi +pntn +2(1 -72)AR£/N- (3.9)
2
(ﬂ;— 7)2 -——CO o-———th[ [N N]+— 0th[ IF 1+ ﬂ.
AR:l = p NISF ZISF P:Isrc 2131: ARp/N
1 1 t 1 1 1
COthI—IN— —CO thl F ]+ . { 0th[ tN N+] .COthltF1}

Pnlsrco 2131: Pr’sr ler ’4er Pnlsrc ler Prlsr 21sFr

(3.10)

61

2 2
M tanh[—t§—] + J;— coth[—t—F—

 

 

 

 

 

 

ARE], = PNIsli 2135i lilil: Zlir
1 IN 1 ’1: l 1 1,, 1 1,.
I" tanh[§N— 'lF coth[21F ]+ AR' { IN tanh[21N ]+ .1], coth[—F—]}
PN SF 317 PF SF SF rm PN SF SF Pr sr ler
(3.11)

When 15;: >> t in both ferromagnetic and normal metal layers, these equations reduce to
the simple 2CSR model formulae for resistance.

For ﬁnite numbers of layers, the VF equations must be solved numerically on a
computer. Each layer is characterized by its resistivity (or resistivities for the F -layer)
and two lengths, the layer thickness and the layer spin diffusion length 13;. At the layer
boundaries, the chemical potentials and currents are matched, taking account of
interfacial speciﬁc resistances where needed but neglecting any interfacial spin memory
loss. For later use, and to try to clarify some signiﬁcant behaviors, we consider a few
cases in which the VF equations simplify.

Limiting Cases

(1) As already noted, when 1;, >> tF and 13. >> 1N , the VF equations reduce to the

2CSR model equations 1.1 - 1.3.

(2) When 1:]: << ’r in a multilayer, the VF equations reduce to a set of simple

resistors in a series-parallel combination, as shown in Fig. 3.2. Each resistor consists of a

parallel combination of two entities, each N-layer, the two bounding F/N interfaces, and

those portions of the adjacent F-layers within 1;} of the interface. These parallel

combinations are coupled in series by the central (tF —21§F) of each F-layer with an

‘unenhanced’ F-layer resistivity, pp.

62

(3) In an F/N/F trilayer with given 13‘. , and only tN allowed to vary, AAR decays as

’N

-(—)
IN C I O I I O I

AAR~e SF . This observatlon Will be used in Ch. 6 to measure 15°]: in Ru, and a srmllar

argument for interfaces will be used to measure spin-memory-loss at Cu/Ru interfaces.

(4) In an F/N/F trilayer, or an EBSV, if 1;, << 1,, the VF analysis modiﬁes Eq. 1.2

 

0 F e 2
(ﬁr Prlsr +7r/N ARF/N) (3.12)

to: AAR = 4 . F .
ZPFISF '1' PN’ N ‘1' 2’1er

Comparing Eq. 3.12 with Eq. 1.2 shows that the ‘MR-active’ region has been reduced to

just the central N-layer, its two interfaces with the F-layers, and the portions of the two F-

layers within lip of the central N-layer. In the numerator, this reduced MR-active region

means that t]: is replaced by 1%}. In the denominator, everything outside of this MR

active region disappears; i.e. ARpeMn/F , ARs/F , and the contributions from the two F-

layers beyond 1gp , are all gone. An initially surprising consequence of these changes is
that the denominator can decrease more than the numerator increases, thus increasing
AAR (i.e., AAR for ﬁnite IE1: is larger than AAR for 1's} = oo). We caution, however, that
ARM» does not similarly decrease, but retains exactly the same form as in Eq. 1.1.

(5) Finally, by analogy to item (4), AAR can be increased by removing the
contributions to the denominator of AAR of large ‘lead resistances’ by judicious insertion
of a source of strong spin-memory-loss (SSML). An example of such an SSML would be
a 1 nm thick layer of F eMn, which has been shown to produce > 99% ‘spin-memory-
loss’ [73]. If the leads are normal-metals, thin FeMn layers can be inserted near to the

outer edges of the multilayer, constraining the MR active region to between themselves,

63

and removing all of the lead resistance ‘outside of themselves’. If the leads are supercon-

ducting, then their contribution 2ARs/p might be replaced by a smaller 2 ARE/FeMn An

example of such a result when FeMn was inserted within a Py-layer was given in [85]. In
this thesis we examine what happens when FeMn is inserted into the Cu layers (assumed

initially superconducting) separating Nb leads from the outer F-layers of a sample.

1 F 1 1
1 SF ARF/N Pain ’4er p F SF

1
SF ARF/N PN’N ARF/N PF SF P1:

pl-‘(tF “21:10

 

prop-21130
“\NV‘

1F 1 Tr tr
PF’sr Akin, erN 4er p514} Pr’sr ARgm erN 412:," PF’SF

Fig. 3.2. 2CSR model representation of the multilayers when 1;}: << tp. The two spin
channels mix completely in the middle of each ferromagnetic layer then split at around
1;? from the F/N interface. So the two spin channels are no longer independent.

 

 

 

 

Spin-Flip Scattering at Interfaces

As noted above, the original V-F model neglected spin-ﬂipping at F/N interfaces.
Including spin ﬂip scattering at interfaces requires modiﬁcation of either the equations or
the analysis procedure. Fert and Lee [93] generalized the VF model to allow for

interfacial spin-memory loss by replacing equations 3.6 and 3.7 by:

use: )-Js(z.' )1 -/J-a(z: ) — 1-512: )1 = Helga--5 w. 2., (3.13)
b

ﬂs(zi)-.Us(zi-)=rs~ls(zi) (3.14)

64

where

 

 

Js(z.*)+Js(zf) (3 15)

JS(zi)= 2

and (,u_S - #5 )2.- is also the mean value of 215 - p_s at z,

ARI?F is a boundary speciﬁc resistance associated with spin-ﬂip scattering at the interface

(no spin ﬂipping => —> 0 ).

 

ARE”

In Eq. (3.13), the current in each spin channel is no longer conserved at interfaces. The
discontinuity in current is due to spin ﬂipping, which causes relaxation of spin
accumulation at interfaces. Of course the total current, J, is still conserved and uniform
over the multilayer (i.e. J. + J. = J).

We at MSU developed a different approach to treat spin ﬂipping at interfaces [73]. in
studies of spin-memory-loss at N1/N2 interfaces (i.e. the interfaces between two different
normal metals) the interface was treated as a homogeneous slab of N1N2 alloy of ﬁnite
thickness, (1, resistivity, p1, and spin-diffusion-length, 1,3,. This extra interfacial ‘layer’ is
subject to the constraint:

101-’1 = ARM/NZ (3-16)

The strength of spin ﬂipping at N1/N2 interface is measured by a quantity 6 deﬁned as

o:-‘,’-. (3.17)
[SF

The spin-ﬂip probability at N1/N 2 = e'a. This additional ‘layer’ is then treated as an extra

layer in the multilayer, except that the ‘interfaces’ N1/N1N2 and N1N2/N2 are assumed

‘transparent’ (i.e. have zero AR). Because the thickness t; can be factored out of the

65

constraint equations 3.15 and 3.16, it has no effect on the outcome of this analysis and
can be chosen arbitrarily.

In the present thesis, we have adapted this model to include F/N interfaces. Since the
F/N interface has a spin-direction dependent speciﬁc resistance, we simply assume that
the interfacial FN alloy has a bulk anisotropy parameter ,6], chosen equal to 7, and apply
numerical analysis similar to that used for the N1/N2 interfaces. For simplicity of
presentation, we describe all of the effects of our inclusion of interface spin-ﬂipping in
VF theory in Ch. 4. There we not only use it to analyze data on ‘interleaved’ and
‘separated’ hybrid multilayers, but also show that adding a modest spin-memory-loss at
F/N interfaces (6, ~ 0.25) does not strongly perturb our prior analyses of [Co/Cu]~ and
similar ‘simple’ multilayers in which we neglected any such loss, but does more

signiﬁcantly perturb the analysis of Co/Cu EBSVs.
3.2 REAL F ERMI SURFACE EFFECTS

Bauer and coworkers [4,94,95] have argued that the ﬁrst issue to be considered
concerning real Fermi surface effects is whether the scattering from adjacent interfaces is
coherent or incoherent. If it is incoherent, then the speciﬁc resistances of neighboring
interfaces are independent of each other and simply additive. In the absence of spin
ﬂipping, one then ends up with a 2CSR model, with only the parameters changed from
their free electron values. In support of such a picture for real multilayers, Stiles [96] and
Xia et a1. [95] showed that no-free-parameters calculations of AR for ‘single interfaces’ in
Ag/Au and Co/Cu agree with the reported experimental values [5] to within mutual

uncertainties.

66

If, however, scattering from adjacent interfaces is at least partly coherent, then the
contributions to AR from these interfaces will not be simply additive. The total speciﬁc
resistance, ART, will then, in general, depend upon the separations between interfaces,
more speciﬁcally upon the ratios of those separations to the mean-free-paths in the N-
and F-layers. If, to simplify, one assumes that the thickness tr: of the F-layer is longer
than its mean-free-paths, but that the thickness IN of the N-layer can be varied from
shorter than to longer than its mean-ﬁee-path, then ART should vary with IN. Because the
speciﬁc resistance of two interfaces, ZARp/N, is usually much larger than the speciﬁc
resistance (pNtN) of the N-layer, this change in ART can be viewed as a change in the
interface resistance. Various aspects of such phenomenon have been discussed by
several authors [4, 97-99]. For convenience, we follow [89] in calling such effects
‘mean-free-path’ effects. In Ch. 4 we will test a claim in [89] of having observed such
effects.

3.3 HIGH CURRENT DENSITY EFFECTS ON MAGNETIZATION

In a pioneering work, Slonczewski [15] and Berger [16] predicted that a spin
polarized current can generate magnons or completely reverse (switch) the magnetization
direction. Tsoi et. al. [17,18] soon observed magnon generation, and then the Cornell
group [19] observed the magnetization reversal. The large current densities needed for
both phenomena prohibit observing them in samples with superconducting contacts.

As discussed in Ch.1, two different processes can cause magnetization switching in
pillar-shaped samples with submicron areas (nanopillars). The ﬁrst is the helical magnetic
ﬁeld, or self ﬁeld, induced by the current, and the second is associated with spin transfer.

Both the self ﬁeld and the spin transfer effects compete and can be observed in the same

67

current range or different ranges depending on the magnetic parameters of the materials
and the diameter of the nanopillar. The dominating process in a given sample can be
determined experimentally (see Ch. 1) from the shape of the R(I) curve, which is

symmetric around zero for self-ﬁeld switching and asymmetric for spin-transfer.

3.3.1 Spin-Transfer Switching

When Slonczewski predicted the spin transfer switching of magnetization, he
attributed it to spin transfer from the polarized current to the local magnetization, which
exerts a torque on both of the ferromagnetic layers. He assumed ballistic transport, that
the minority electrons will get totally reﬂected from the Co layer, and that the majority
electrons will go through it without scattering (perfect spin ﬁlters). The component of
electron spin perpendicular to the magnetization gets reﬂected at the Co/Cu interface and
is responsible for the spin transfer to the local magnetization; thus if the magnetizations
of both layers are collinear there will be no spin-transfer switching. Slonczewski used a
trilayer of thick Co/spacer Cu layer/thin Co layer. The advantage of having the thick
layer is that it needs a very large current density to switch and can thus be considered
ﬁxed. When electrons are ﬂowing from the thick Co layer to the thin, the majority
electrons emerging out of the thick Co will exert a torque on the thin layer to switch it to
the parallel alignment with the thick layer. When the current direction is reversed, the
minority electrons that get reﬂected at the Co/Cu interface between the thick Co layer and
the Cu spacer layer will go back to the thin Co layer and switch it in the direction

opposite to the thick layer’s magnetization. This way, one magnetization state is

68

stabilized for each polarized current ﬂow direction. The equation of motion of the

magnetization of the thin Co layer in the Slonczewski model is:

A

(IS

A A A A A dS Ig(6) A A
—=S H C-4IIM S.bb- —-
dt X 7[ cf ( ) ] a dt

_, st (3.18)
elSl

 

In this equation, 8 is the spin direction of the F layer, M is the magnetization, C is the
easy axis of magnetization, I is the current, a is damping coefﬁcient, y is the
gyromagnetic ratio, b is the direction perpendicular to thin ﬁhn, and gm) is a coefﬁcient
that depends on the current polarization. In the absence of damping (a = 0) and if Ig(@
= 0 the solution of the above equation is steady elliptic precision of spin around C axis,
with time-independent amplitude, A. Magnetic damping, the second term in the above
equation, damps A as time passes. On the other hand, the spin-transfer term, the last term,

can amplify or damp A, depending on the current direction. The critical currents for

switching are:
13” = 1;; = aﬂ|§|[Heﬂ(0) + 27rM]/g(0) (3.19)
1:; = 15 = aw|§|1Ha<zr> - 27er/g(7r) (3.20)

An applied magnetic ﬁeld will favor the parallel alignment, making it harder to
switch the magnetizations by spin transfer. This causes both 15 and IE to shift in the

same direction (i.e. towards more positive values, according to our convention of current
direction). When the external ﬁeld is large enough, it is no longer possible to switch the
magnetization completely. Rather, the current will excite steady spin waves in the

magnetic layer. The excitation of spin waves shows up as a spike in the dV/dI curve

69

because the spin waves reduce the parallel component of magnetization, making AR
intermediate between that of parallel and antiparallel.

Weintal and coworkers [88] generalized the Slonczewski approach to include both the
ballistic and diffuse scattering regimes.

Heide [86] argued that spin induced switching occurs due to the exchange interaction
between the spin accumulation and the F (Co) metal. This interaction, according to
Heide, creates an effective ﬁeld that favors either the ferromagnetic or antiferromagnetic
coupling between the two Co layers, depending on the current ﬂow direction, and is
called nonequilibriurn exchange interaction (N EXI). This effective ﬁeld interaction is
very similar to the RKKY interaction, but is not oscillatory—i.e. only one magnetization
state is energetically favored for each current direction. Heide argued that the length
scale for the spin accumulation-induced effective ﬁeld is the spin-diffusion-length and
that the second Co layer was not needed for the switching of the ﬁrst, provided that the
current is spin polarized.

Recently, Zhang, Levi, and F ert [87] argued that both the spin transfer (spin torque)
and the effective ﬁeld mechanisms occur on equal footing in the exchange interaction.
They included the exchange interaction between local moments and spin accumulation of
conduction electrons to study the current induced switching in the diffuse transport
regime. Another signiﬁcant ﬁnding of Zhang et al. was that the component of spin
accumulation perpendicular to the magnetization was the important one and that the
length scale over which it relaxes is ~ 1-2 nm. The Zhang et a1. model requires two
ferromagnetic layers, which have to be noncollinear to produce spin induced switching.

Those ﬁndings are at odds with the Heide argument that the longitudinal component of

70

spin accumulation matters and that the appropriate length scale is the spin-difﬁrsion-
length, 15: ~ 60 nm.

Thus, it remains for experiments to sort out what actual processes dominate the
polarized current switching of the magnetization, and what length scales are relevant to
those processes. Since we do not yet have a set of systematic samples to do so, we will

not go into details of those theories any further.

71

Chapter 4
SEARCH FOR MEAN-FREE-PATH EFFECTS IN

THE CPP-MR

4.1 INTRODUCTION

In the CIP-MR, the mean-free-paths in the N-metal, AN, and those for electrons with

moments along, 111:, or opposite to, 1%, the F-layer magnetization are characteristic
lengths [3]. In contrast, in the CPP-MR, according to the theory of Valet and F ert (VF)
[47, Ch. 3] the only scaling lengths are the spin-diffusion lengths (spin-ﬂip lengths) in the
. N- and F-metals, 1g], and lg}. The mean-free-paths appear only indirectly through the
layer resistivities, which are parameters in the CPP analysis. Since scattering events with
spin-ﬂips are usually only a fraction of all scattering events, 13,.- is normally expected to
be longer than 7L. When these lengths are much longer than their respective layer
thicknesses, 115in >> no, and [SF >> tp, they drop out of the CPP-MR, yielding a simple
two-current series resistor (2CSR) model. In the 2CSR model, the only lengths are the
layer thicknesses (N and tr, which appear only algebraically. As described in Ch. 1, this

dependence only on IN and tr: was tested early on by comparing data in the square root

form of Eq. 1.3 for samples with the same F-metal, Co, but different N-metals, Cu vs

Cu(4%Ge) or Ag vs Ag(4%Sn), that changed AN by up to a factor of 20, but left 1;“, and

1'5} long enough so that the 2CSR model should still apply. Agreement of those data

with the 2CSR model predictions implied that the model represented a good

72

approximation to the physics underlying the CPP-MR of these metals, and seemed to
show that AN, ,1; , and 1%- were not CPP-MR scaling lengths.

Recently, however, the applicability of the 2CSR and VF models was challenged
[89], based upon a combination of theoretical analysis with observation of qualitatively
different CPP-MRS for two different geometries of samples composed of Co layers with
thicknesses too = 6 nm or 1 nm alternated with Cu layers thick enough (tcu = 20 nm) to
eliminate exchange coupling between the Co layers. The geometries are called
‘interleaved’ [Co(6)/Cu(20)/Co(1)/Cu(20)]~ or ‘separated’ [Co(6)/Cu(20)]~[Co(l)/Cu(20)]~,
where italic N indicates the number of repeats. Because the saturation magnetic ﬁeld Hs
of Co layers of macroscopic area increases with decreasing too, the use of two very
different values of tea should allow for anti-parallel (AP) alignment of the magnetic
moments of thin and thick Co layers over a limited ﬁeld range. Analysis of experimental
data should thus be straightforward. Because the only lengths in the 2CSR CPP-MR
equations are too and ten, those equations predict that the MRs of ‘separated’ and
‘interleaved’ samples should be identical. In contrast to this expectation, [89] found the
maximum CPP-MRS of separated samples to be only about half as large as those for
interleaved ones and AAR(H) = AR(H) — ARp for the separated samples to have a more
complex shape.

We began our study of mean-free-path effects by trying to reproduce the data of [89].
The ﬁlled circles and solid curves in Fig. 4.1 for interleaved and separated samples with
N = 4, 6, and 8, conﬁrm the qualitative differences in AAR(H) that they found. To allow

direct visual comparison of the forms and magnitudes of data for interleaved and

73

separated samples, in all ﬁgures showing AR(H), the scale divisions for interleaved and
separated samples are the same.

Because the data for separated and interleaved samples differ, we must be careful in
deﬁning the experimental quantities to be discussed in this chapter. AR(H) and AAR(H) =
AR(H) — ARp represent the measured quantities as in Fig. 4.1. Their values will usually
differ for interleaved and separated samples with the same numbers of layers N. The
symbols AAR and CPP-MR = AAR/ARp will be reserved for cases where AR(H) equals
ARAp for the state with magnetizations of adjacent layers aligned AP. We’ll see in
section 4.2 that, deﬁned this way, AAR and AR Ap are essentially independent of whether
the sample is separated or interleaved.

The data in [89] were taken on samples with layers sputtered in the order
[Co(l)/Cu(20)/Co(6)/Cu(20)]1v, whereas the data in Fig. 4.1 and most of our other
samples below are on samples sputtered in the order [Co(6)/Cu(20)/Co(l)/Cu(20)]~. Fig.
4.2 shows that we get the same differences if we use instead the order of [89].

We focus ﬁrst on the ﬁlled circles and solid curves, representing AR(H) after the
samples were taken to above their saturation ﬁelds, H5. The open squares and dashed
curves, showing how AR(H) varied when the samples were ﬁrst taken to saturation, will
be discussed in section 4.2. The solid curves of AR(H) for the ‘interleaved’ samples
consist of a single broad peak, symmetric about H = 0. This is the qualitative behavior
expected from the series resistor model as the magnetic order of the layer magnetizations
in the sample reorients from P at high H to AP for values of H between the saturation
ﬁelds of the too = 6 nm and 1 nm layers. The solid curves of AR(H) for the ‘separated’

samples, in contrast, have more complex shapes, with maximum values of AAR(H) only

74

about half those for the ‘interleaved’ samples. The smallest N shown (N = 4), gives a
sharp peak followed by a broad plateau. The largest (N =8), gives two separate peaks,
almost equal to the sum of separate contributions from the 6 nm and 1 nm Co layers.

As noted in Ch. 1, differences in AAR(H) for interleaved and separated samples
similar to those in the solid curves in Fig. 4.1 were reported previously in samples of Co
and Py with Ag, or Co and Fe with Cu. In the ﬁrst case, the difference was attributed to
short 1;}. In the second, short 13.} was taken as part of the explanation, but possible spin-
memory-loss at F/N interfaces was also proposed. In contrast, the authors of [89] argued
that IE}? ~ 60 nm [76] and IE}? ~ 500 nm [73] were too long for spin-memory-loss within
the Co and Cu to be important. They argued that the series resistor and VF models break
down when the mean-free-paths become larger than the layer thicknesses—that ‘mean-
free-path effects’ then appear. Their argument is not simply one of changing magnitude
of AR(H) as the resistivities pp and W of the F and N layers change. (We’ll see in section
4.4 that the series resistor and VF models can approximately account for the observed
changes in magnitude of AR(H) due simply to changes in no and/or pp.) Rather they
argued that ratios t/l appear explicitly in equations that differ from those of both the
series resistor and VF models. If they are correct, then the differences in behavior in Fig.
4.1 should gradually disappear as the ratio W. is increased.

In this Chapter, we test this claim in three ways. First, we replace Cu by a dilute

Cu(2%Ge) alloy (hereafter just CuGe) with a short mean-ﬁee-path, 2cm. ~ 8 nm, but a
long spin-diffusion length, [EEG‘~ 130 nm [68]. Such alloying should mainly reduce Ac...
For sammes identical to those in Fig. 4.1, except with Cu replaced by CuGe, increasing

the ratio too/Zoo by up to a factor of 25 (from ~ 1/5 for our sputtered pure Cu to ~ 5 for

75

the thickest CuGe layers) produces no signiﬁcant change from Fig. 4.1 in either the forms
or relative magnitudes of interleaved and separated samples. Second, in addition to
replacing Cu by CuGe, we also replace the 6 nm Co layers by 15 or 30 nm thick layers of
Co(2%Zr) (hereafter just CoZr) and the 1 nm Co layers by 30 nm thick layers of Py, both
of which have mean-free-paths much shorter than these layer thicknesses. If anything,
the differences increase. Such behavior is opposite to that expected from mean-free-path
effects, but is expected if the CPP-MR is dominated by short spin-diffusion lengths in
CoZr and Py. Lastly, we insert a 1 nm thick layer of FeMn into the middle of the central
layer of a separated Co/Cu sample. The total AR contributed by the FeMn layer and its
interfaces with Cu is comparable to that contributed by alloying the 20 nm thick Cu layer
with Ge, From ‘mean-free-path’ effects, we would thus expect the separated data to
move slightly toward the interleaved data. In contradiction, the separated data move
away from the interleaved data, in just the way expected for a strong increase in ‘spin-
memory-loss’ due to the FeMn layer.

The rest of this chapter is organized as follows. In section 4.2 we discuss the
signiﬁcance of the open circles and dashed lines in Fig. 4.1, representing the behavior of
AR(H) while the samples are ﬁrst taken from their as-prepared state to saturation. We
infer that the magnetic structures of the samples in their as-prepared states must in some
way be closely similar to the structures in their AP state. In section 4.3 we use
magnetization (M), and polarized neutron scattering measurements both to conﬁrm this
inference and to show that the magnetic structures in interleaved and separated samples
do not differ in spurious ways. The absence of any signiﬁcant spurious difference in

magnetic structures is essential to any simple analysis of the data. With conﬁdence in

76

such absence, we turn in Section 4.4 to the three tests of mean-ﬁee-path effects listed
above. We shall see that all three tests strongly suggest that the differences between the
data for interleaved and separated samples are due not to mean-free-path effects, but
rather to spin-ﬂipping in the multilayers. In section 4.5 we ﬁrst examine whether the data
can be explained by the best available spin-diffusion lengths in the bulk Co and Cu. We
conclude that these lengths can only partially explain the data. We propose an
explanation based upon spin-ﬂipping at the Co/Cu interfaces. In section 4.6 we

summarize and conclude.

4.2 AR(H) FROM THE INITIAL STATE TO FIRST SATURATION

Fig. 4.1 shows that, once the samples have been taken to saturation, the differences
between AAR(H) for interleaved and separated samples are large. in contrast, in the as-
prepared states of the samples, the differences in AAR(H) are much smaller. If we deﬁne
AAR(0) = AR(as-prepared) — ARp for both interleaved and separated samples, and
AAR(peak) = AR(peak) — ARp for interleaved samples, then we ﬁnd that AAR(0) for
separated samples is essentially the same as AAR(peak) for interleaved ones, and that
AAR(0) for interleaved samples is about 85% of AAR(peak) for interleaved ones.
Because of the differences in Hg for 1 nm and 6 nm thick layers of Co, we expect the
magnetizations of adjacent layers in interleaved samples to be oriented nearly AP at the
‘peak’ of AR(H) after the samples have been taken to saturation. Separated samples also
have AAR(0) z AAR, simultaneously arguing that adjacent layers in the as-prepared states

of separated samples are also ordered AP, and that AAR is experimentally well-deﬁned,

77

independent of whether the thicknesses of adjacent layers are the same or different. In
contrast, for interleaved samples AAR(0) z 0.85A AR, indicating that, in the as-prepared

state, adjacent layers are not fully AP ordered.

 

 

      

 

 

 

 

1‘ V‘VTVf‘rvrvpvv—VVIVIvvtwvv‘virﬁ'vf‘ 23_""r'*"r"r'rﬁrvr"—"r'rj"
1 I
I Cu (114) Interleaved ; : Cu(xs) Interleaved 3
15,5L . 12. -
: A : 1179-1:
N I 1
E 21- -1
a i . i
.. 20E -j
v F 4
K 19. ' J
< , 1
> d
1.; .1
I O
I> “ I
.nn.l..n.ln...IJmelnnnnl...n
*7 Ir 7117 I *I'VV'I
2, errors) . Separated
A : 0 1171-1
"E 21:- -
,
c:
.-
v

 

 

 

 

 

 

 

 

.
10'
b
.

 

 

 

AR(rnm’)

 

 

 

 

Fig. 4.1. AR(H) vs H at 4.2K for interleaved (top of each pair) and separated (bottom of
each pair) multilayers of Co/Cu with N= 4, 6, and 8. The unit scales for each coupled
pair are identical. The open squares and dashed curves show how AR(H) varied when
ﬁrst taken ﬁ'om the as-prepared state to the saturated state. The ﬁlled circles and solid
curves show how it varied after having been taken to saturation.

78

 

 

‘ V'TWTW f‘TYVY‘ vv‘ﬁvj vvvvvvvvvvvvvvvvvvvvvvv

20 ’- [Co1lCu20ICoBICuZO]. g 19 {- [Co6lCu20ICo1lCuZO].
1300-2 3 '

 

 

 

 

 

 

 

19 [CO1IC020LIIC06IC0201'
z 9. 3 F
N : Separated '-, 1308-3

 

AR(tnm’)
a: :1

cl
VTY'VYY

 

‘

.l
A.

I V VY '

 

 

 

 

   

Fig. 4.2. AR(H) vs. H curves for interleaved (top) and separated (bottom) samples. Both
samples on the left have the thinner Co layer with tCO= lnm sputtered before the layer
with too= 6nm. The two samples on the right have the opposite order.

We interpret these AR results as follows. As just noted, the close agreement between
AAR(Peak) z AAR for interleaved samples and AR(O) for separated ones shows that both
the peak states of interleaved samples, and the as-prepared states of separated samples,
are close to the ideal AP state with magnetizations of adjacent Co layers reversed. The
slightly smaller values of AR(O) for the interleaved samples indicate that their magnetic
orders deviate slightly ﬁ‘om such AP states, a discrepancy we discuss below. For the
interleaved samples, oppositely directed magnetizations for thick and thin layers in as-
prepared and peak states would lead to M at 0 if the layers were single domain. In
Section 4.3 we will show that M(Hinitia1 = 0) z 0, and evidence that the domains in as-
prepared samples are micron-sized is given in [100]. Based upon these two results, we

argue that the as-prepared states of both interleaved and separated samples consist mostly

79

of nricron-sized AP ordered domains that extend from the top to the bottom of the
sample, but with the magnetizations of either the thicker or thinner Co layers oriented
randomly ‘positive or negative’ in the layer planes. As discussed in [100], such ordering
is likely due to the fringe ﬁelds from the ends of magnetic domains. To simplify, if the
bottom Co-layer consists of small magnetic domains pointing only in opposite directions,
such as those shown on the left sides of both drawings in Fig. 4.3, then the ﬁinge ﬁelds
from these domains will act on the newly growing Co layer above to orient its domains
anti-parallel to those in the bottom layer. If the layer thicknesses are all the same, this
process could yield the ‘ideal’ as-prepared anti-parallel order in vertical domain
structures shown in Fig. 4.3. In separated samples, the thicknesses of adjacent layers are
also all the same, except at the one boundary between thick and thin layers. In the as-
prepared state, we, thus, expect almost as ideal AP domain structures as in simple [F/N]N
multilayers. In interleaved samples, in contrast, adjacent layers always have different
thicknesses. If the ﬁrst layer is thick, its fringing ﬁelds will not be fully ‘taken up’ by the
thinner second layer, letting some ﬁeld lines extend up to the third layer and compete
against the smaller fringing ﬁelds from the second layer. This difference is compatible
with a less ideal AP order of the as-prepared states of interleaved samples and the smaller
values of AR(O) in Fig. 4.1. Fig. 4.3 shows the ideal pictures of the differences between
as-prepared samples of interleaved and separated samples, and the same samples after
being taken to H5 and then having the ﬁeld reversed only enough to ﬂip the thicker
layers. In the as-prepared states, half of both thicker and thinner Co layers would already
be oriented along any given ﬁeld direction. Thus, as each different layer thickness ﬂips,

the changes in magnetizations should be only about half as large as when the same layers

80

ﬂip after the sample has been taken to saturation. These are the behaviors seen in
magnetization measurements that we postpone to Fig. 4.5 below. In section 4.3, we will
see that magnetization and neutron scattering measurements both support this picture. In
section 4.4 we will see that similar behaviors are seen also in Co/CuGe multilayers. Thus

the picture seems to apply generally to interleaved and separated multilayers based upon

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Co and Cu or CuGe.
interleaved Separated

As-Prepared After H, As-Prepared After H,
Co —’ Co ‘— Co *— Co "’ Co ‘— Co ‘—
Cu Cu Cu Cu Cu Cu
C“ ‘— Co —> C0 —’ Co ‘— Co -> C0 *—"

C

Cu Cu 11 Cu Cu Cu
Co —> C0 <- Co 4—— Co _> C0 ‘_ Co
CU CU CU Cu CU CU
Co ‘— Co —’ Co —’ Co ‘— Co —' Co —'

 

 

 

 

Fig. 4.3. Schematic, approximate pictures of inferred magnetic ordering for interleaved
and separated samples in their ‘as—prepared’ state and after they have been taken to
saturation and then the ﬁeld reversed only enough to ﬂip the thicker Co layers.

4.3 MAGNETIC ORDER

In this section we look for evidence of any unwanted differences in magnetic order
between interleaved and separated samples. We do this in two ways. (1) By directly
comparing total magnetizations M for interleaved and separated samples. (2) By

scattering polarized neutrons from interleaved and separated samples.

81

(1) Comparison of M for interleaved and separated samples

To avoid the strong diamagnetism of a superconductor, magnetizations M were
measured at 12K, above the superconducting transition temperature (~ 9K) of our
sputtered Nb. M was measured both on entire samples—allowing comparison of absolute
magnitudes for interleaved and separated samples, and also on the A ~ 1.2 mm2 ‘central
pieces’ through which the CPP current ﬂows. Similar results were obtained for both.

Fig. 4.4 compares the values of M for whole interleaved and separated samples with
both Cu and CuGe non-magnetic metals. In both cases, the two Ms are practically
identical. Irnportantly, the ﬁelds at which the AP states occur are the same to within our
measuring uncertainties. Fig 4.5 shows similar behaviors for pieces of samples involving
just the areas through which CPP current ﬂowed. In both the top and bottom ﬁgures, the
ratios of changes in magnetization as thick and thin layers ﬂip are about the expected 6 to
1. The peaks in AR(H) for the separated samples in Fig. 4.1 after saturation occur about
where the average layer magnetizations pass through zero for the thick layers (~ 100 Oe)
and the thin layers (~ 725 Oe), respectively. Fig. 4.5 contains the variations when the
samples were ﬁrst taken from their as-prepared states to above H5, in addition to those
when they were cycled to below —Hs and back to above HS. In the as-prepared states,
both magnetizations in Fig. 4.5 are close to zero. As H is increased ﬁom zero, M initially
increases rapidly up to H ~ 200 Oe, then more slowly to above Hs. Both changes are
only about half as large as those over the same ﬁeld range after the samples have been
taken to above HS. These differences in magnetization changes before and after
saturation mean that the magnetic structures of the samples differ before and after the

samples are taken above H3.

82

(2) Polarized neutron reﬂectivity (These studies were done by J .A. Borchers and K.V.
O’Donovan of NIST)

The magnetic structures in these samples were also characterized using polarized
neutron reﬂectivity (PNR), which is sensitive to the interaction between the sample’s
magnetization and the magnetic moment of the neutron. PNR measurements were
performed at the NIST Center for Neutron Research on multilayers prepared in the same
way as those for the CPP-MR studies, except that the samples were squares, 1.27 cm on a
side, and sputtered directly onto the Si substrates instead of onto Nb. All PNR
measurements were made at 17 — 18 K after cooling in zero ﬁeld in a closed-cycle
reﬁigerator. An electromagnet provided a maximum ﬁeld of 0.2 T. Using neutrons of
wavelength it = 0.475 nm on the NG-l reﬂectometer, Borchers and O’Donovan
measured the non- spin-ﬂip (NSF) reﬂectivities, R+ + and R' ', as well as the spin-ﬂip (SF)
reﬂectivities, R+ ’ and R' +. (The + and - signs in the superscript describe the polarization
state of the incident and scattered neutron spins. These spins are aligned either parallel,
+, or antiparallel, -, to the applied ﬁeld.) The selection and detection of the neutron spin
state is described elsewhere [101]. The NSF reﬂectivities sense the chemical structure of
the multilayer and the component of the ill-plane magnetization parallel to the applied
ﬁeld. The SF reﬂectivities are primarily sensitive to the projection of the in-plane
magnetization that is perpendicular to the applied ﬁeld. They measured both the specular
and diffuse (i.e., off-specular) reﬂectivity as a function of the wavevector Qz=47zsin61//i,
where 6 is the angle of the incident and scattered neutrons relative to the sample surface.
We obtained the diffuse data by offsetting the sample normal by 0.1° relative to the

specular scattering

83

 

'H'i" "l—'" 'l""i""i"' 'q
[Cu (x8), T =12K j
ﬁ-x interleaved ”unaware”;

:+ Separated 1’ :
. g (M i3
§ .

‘Nu

_

O

h

g AP§

2:" :

-1: P [xii .:

I l :1: 1

: it :

'ZExxlﬁ artisans!!!" _

-3.n.rl....l....l....1....1....‘
I

IITIIFIIII

A
m
V

.1.

M (10"3 emu)

 

 

 

 

 

 

gene. (in, r '= 12k 1 j

A zgx Interleaved ”1.... "”3.
E 1 -_+ Separated f ray 1 _i
o E iii P E
9 0:— Apg’l“ SAP -3
E -13. P 1% (bl .3
E E i X‘s?
_2Triltxaut Irritant! i

-3 ‘

"-‘1’000500' ‘0‘ ’ 5001000 ’
H (De)

Fig. 4.4. M vs H at 12K for interleaved (x) and separated (+) samples with Cu and CuGe
with N = 8 for to“ = tong. = 20 nm. P and AP states are indicated.

84

.°
N

 

 

 

10.1007 "Qty-"1W
A 01 [Interleaved 5 ( Q
{:5 ET =12 K ; 05 i
01 ECenter :j: g
4“) O 7 e : j
‘- E 8 I i
E -0.1 :— mj .0 g
5' i “a
1’ 1 L. 4»

 

 

O
O
O
. 41-
. —-1
. -.
-11-
71-
4’
C -‘
Cl -‘
CI -1
C
11W

._ Separated .’
§T=12K 5

9
u-‘L

llll

  

1_L1

 

_ Center

11 rrrr

 

M ( 10'3emu)

I
P
A
I WTTI
~. . . O
l 141

. .0000... "
—
d

021.-.--.-
4000-500 0 5001000

H(Oe)
Fig. 4.5. M vs H at 12K for center-cut pieces of Cu interleaved (top) and separated

(lower) samples including both data from the as-prepared state to saturation and then
from positive saturation to negative saturation and back again.

 

 

condition and then scanning Q2. These difﬁlse measurements provide general
information about growth-axis correlations among in-plane ferromagnetic domains with
dimensions less than 100 um [100] and were also subtracted ﬁom the total reﬂectivity to

give the specular reﬂectivity. After correcting the specular data for the efﬁciencies of the

85

polarizing elements (ranging from 92% - 100%), the ﬁts give a proﬁle of the vector
magnetization as a function of depth for regions of the multilayer with ferromagnetic
domains larger than about 100 um within the sample plane [102-104].

PNR data generally support this picture of randomly oriented, micron-sized domains
within the sample plane that become single domains in a saturating ﬁeld. We examined
an interleaved sample [Co(6)/Cu(20)/Co(1)/Cu(20)]8 and a separated sample
[CO(6)/CU(20)]3[CO(1)/Cu(20)]8 with magnetoresistance and magnetization behaviors
similar to those shown in Fig. 4.1 and 4.4, respectively. After cooling in zero ﬁeld and
saturating in a ﬁeld of 2000 G, we obtained the reﬂectivity of both samples in a ﬁeld of
250 G (which coincides with the peak in the magnetoresistance of the interleaved
sample), and in ﬁelds of 74 G, 114 G, and 725 G (the latter two of which are near the two
peaks in the magnetoresistance of the separated sample). These results were compared to
the reﬂectivity data measured in a saturating ﬁeld > 1600 G. The data obtained at all of
these ﬁelds were characterized by the absence of specular spin-ﬂip (SF) scattering. This
result indicates that after saturation the Co moments are aligned either parallel or
antiparallel to the applied ﬁeld direction at all ﬁelds. Upon ﬁeld cycling, the reversal of
the Co moment directions for the thick and thin layers thus occurs via domain formation,
rather than moment rotation within the sample plane.

Figures 4.6 and 4.7 show the NSF specular reﬂectivity data and corresponding ﬁts for
the interleaved and separated samples, respectively, in a ﬁeld of 250 G. In Fig. 4.6 the
data for the interleaved sample are dominated by superlattice peaks above the critical
angle at Q = 0.21, 0.32, 0.44, 0.58, 0.71, 0.85 nrn'l z 2 7rn/ D, where D = 45.4 nm is the

multilayer repeat length and n is an integer. (Note that this approximation does not

86

explicitly hold for the peaks at low Q due to dynamical effects.) In contrast, the R' ' data
for the separated sample in Fig. 4.7 show superlattice peaks at Q = 0.29, 0.52, 0.76 nm'1
z 2 Irn/ D) where D) = 24.6 nm, and the R+ + data have superlattice peaks at Q = 0.35,
0.64, 0.95 nm'1 z 2 7rn/ D2 where D; = 19.6 nm. D1 and D2 correspond to the repeat
lengths of the tlrick- and thin-layer parts of the multilayers, respectively, within the
separated sample. The superlattice peaks for each multilayer appear in different
reﬂectivity cross sections because the thin Co layers are antialigned relative to the 250 G
ﬁeld while the moments in the thick Co layers are aligned parallel to the ﬁeld. (In a
saturating ﬁeld, both sets of superlattice peaks are evident only in the R' ' data.) In
general, the relative intensities of the superlattice peaks in the R+ + and R' " data are
sensitive to the magnitude and orientation of the moments in the thin and thick Co layers
averaged across the sample plane. The magnetization proﬁles at the bottom of Figs. 4.6
and 4.7 were generated from the ﬁts to the corresponding PNR data. (Note that the
moments in the thin Co layers appear to be substantially smaller than those in the thick
Co layers, even though the moments in both the thick and thin layers saturate in H = 200
O to almost 100% of the bulk Co value.) This illusion occurs because the ﬁts are
sensitive to the structural interfacial roughness averaged across the sample plane, which
exceeds the 1 nm thickness of the thin layers along the grth direction.) Consistent
with the magnetization results (Fig. 4.4 & 4.5), our ﬁts reveal that the Co moments in the
thin and thick layers are aligned AP to each other for both samples at 250 G, as shown
schematically in Fig. 4.3 (“After H5”). The magnetizations of the thin Co layers are,
however, slightly reduced from their saturation values in both cases, suggesting that small

in-plane domains form within these layers as the ﬁeld approaches the coercive ﬁeld.

87

Specular reﬂectivity data at 73, 114 and 725 G conﬁrm that the thin and thick Co layer
moments reverse directions, as expected, at different coercive ﬁelds. After saturation in
-2000 G, the moments in the thick Co layers in both the interleaved and separated
samples ﬂip when the ﬁeld is increased to approximately 90 G, while the moments in the
thin Co layers remain parallel to the original ﬁeld direction. This behavior is evidenced,
in part, by a large reduction of the ﬁtted moments for the thick Co layers compared to
their saturation value, as well as by an increase in the diffuse scattering [103]. As stated
previously, this moment reversal occurs via the formation of small, in-plane domains.
Upon increasing the ﬁeld to 725 G, the moments in the thin Co layers then reorient
parallel to the applied ﬁeld. Fits to the 725 G reﬂectivity data for both samples indicate
that the thin Co moments, averaged across the sample plane, are reduced from their
saturation value. The data suggest that the reversal of the thin Co layer moments also
proceeds via formation of small, in-plane domains. The reﬂectivity data thus conﬁrm
that the peak/plateau in the magnetoresistance of the interleaved/separated sample near
250 G (Fig. 4.1) is associated with an anti-parallel alignment of the moments in the thin
Co layers relative to those in the thick Co layers. The peaks in the magnetoresistance of
the separated sample near 100 G and 725 G (Fig. 4.1) are associated with the reversal of
the magnetization direction of the Co moments in the thick and thin layers respectively.
The orientation and magnitude of the thin and thick Co layer moments thus have a similar
ﬁeld dependence for both the interleaved and separated samples. Any differences in
magnetic structures between interleaved or separated samples in their nominal AP or P

states appear too small to explain the different magnetoresistances in Fig. 4.1.

88

 

 

 

 

 

 

 

 

' I I ' I I I
0 .— -_ ..
. ' R
-1 _ 0°15, ‘ ++ —1
Q .0" 99° A R '
.2 339153.
*6 '2 ' i . . '
g ‘e 9:. -..',~',é>‘=._
w 10': ~°.- Q
n: -3 F W 'a a 21 .o'. ~
v aw i 1:11:11:
o {if i '18 -’ '- ‘
or i ' . . =
3 4 - Interleaved . Wt 1,3,1», . .-
.1- 3' .
H = 250 G ~ 1 in 1
-5 - ..
a I L I m I a I a a
0.0 0.2 0.4 0.6 0.8 1.0 1.2
-1
Qz(nm )
2 I I I ' I I
”a. 1 . '
3 1 - 1 .J
5 1
8 o 1
g . 1 .
0 _1 l a i r 11 a l
100 200 300 400
Depth (nm)

Fig. 4.6. Specular neutron reﬂectivity data and ﬁts plotted as a function of the wavevector
Q, of an interleaved sample [Co(6)/Cu(20)/Co(1)/Cu(20)]3 at 18 K in a 250 G ﬁeld. Only
the R' ' (dark squares) and R+ + (open circles) reﬂectivities are shown. (The SF
reﬂectivities are effectively at the background level). The magnetization proﬁle for the
Co layers that is shown in the bottom graph was obtained from the ﬁt. The apparent
differences in moments in the thin and thick layers are artifacts (see text).

89

 

  

Log10 (Reflectivity)

 
 

Separated
-5 - H = 250 G

 

 

 

 

_l

O

 

 

 

Co Moment (“3)

 

o . 100 200 I 300 . 400
Depth (nm)

Fig. 4.7. Specular neutron reﬂectivity data and ﬁts plotted as a function of the
wavevector Q, of a separated sample [CO(6)/CU(20)]3[CO(1)/CU(20)]8 at 18 K in a 250 G
ﬁeld. Only the R' ' (dark squares) and R+ + (open circles) reﬂectivities are shown. (The
SF reﬂectivities are effectively at the background level). The magnetization proﬁle for
the Co layers that is shown in the bottom graph was obtained from the ﬁt. The apparent
differences in moments in the thin and thick layers are artifacts (see text).

90

The PNR data also provide information about the magnetic structures of the samples in
their as-prepared states. Fig. 4.8 shows the specular and diffuse reﬂectivity data for the
separated sample. Magnetic SF scattering is evident, and the NSF data show broad
features near the structural superlattice reﬂections. Preliminary ﬁts to the specular data
indicate that some fraction of the sample with in—plane domain sizes larger that 100 pm is
ordered with the moments in the thick Co layers aligned antiparallel to those in the thin
Co layers, and some fraction is ordered with alternating moments of adjacent layers. The
situation is not as simple as the ideal case of Fig. 4.3. However, the as-prepared data also
show substantial magnetic diffuse scattering (Fig. 4.8b) indicative of magnetic ordering
with in-plane ferromagnetic domain sizes smaller than 100 um [102,103]. Distinct peaks
are apparent at Q = 0.22, 0.42, 0.65 nm". These magnetic peaks lie halfway between the
structural superlattice peaks in the NSF data at Q z 2 Irn/ D1. (D1 = 24.6 nm is the repeat
distance of the part of the multilayer with thick Co layers in the separated sample.) The
presence of these peaks indicates that the repeat length of the magnetic structure within
the bottom multilayer is twice that of the multilayer itself. Thus the as-prepared state of
the thick layer part of the multilayer is ordered (partially) in small, ferromagnetic in-
plane domains that are oriented antiparallel relative to each other across the 20 nm Cu
layers, as depicted in Fig. 4.3 (“As-Prepared”). The in-plane direction of these domains
is apparently random since the diffuse magnetic intensity is evenly distributed in the NSF
and SF reﬂectivities. Unfortunately the diffuse data do not provide any direct
information about the magnetic ordering of the thin-layer part of the separated

multilayers.

91

 

Log,o(Reﬂectivity)

 

 

 

 

°° Diffuse - R" o R“+

m; W“? 3 '

N
I

O
I

 

 

Reﬂectivity1o(countsl30 s)

 

o
o
o
N
.0
o
a:
o
on
A
o

-1
Qz(n m )

Fig. 4.8. Specular and diffuse reﬂectivity data for the separated sample
[Co(6)/Cu(20)]g[Co(l)/Cu(20)]3 at 18 K in the as-prepared state. The data were taken in
a 3 G guide ﬁeld. The R' ' (dark squares), R” + (open squares), R+ ' (dark circles), and R' +
(open circles) reﬂectivities are all shown. The diffuse data have not been corrected for
polarization efficiencies. The dotted lines mark the superlattice peaks in the diffuse data.

92

4.4 MAGNETORESISTANCE TESTS OF MEAN-FREE-PATH

EFFECTS

In this section, we examine differences in AR(H) between interleaved and separated
multilayers with various constituents. If t/zi is the controlling parameter in the differences
between interleaved and separated samples in Fig. 4.1, then greatly increasing it in either
the F or N metal should cause the data for separated samples to approach those for
interleaved ones, provided that Is}: remains long enough.

As a ﬁrst test of this expectation, we replaced Cu by Cu(2%Ge)—hereafter just

CuGe—which from p = 8.0 :l: 0.5 pﬂcm (see Ch. 2) we estimate a short 2.0.0., ~ 8 nm

[91], but long 135‘“ ~ 130 nm [71]. For (CuGe = 20 nm and 40 nm (Figs. 4.9 - 4.11), the

ratios taco/deuce are 2.5 and 5. For tong, = 20 nm and N = 2 (Fig. 4.10), the ratio tT/IEI‘!Ge
< 1, where t1 = 60 nm is the total CuGe thickness between outer Co layers. For fence = 10
nm and N = 3 (Fig. 4.11), ml [330° < 0.4 is even smaller. Our data thus cover the range of

mean-free-paths in the N-layer from ~ 5 times the layer thickness to ~ 1/5 of the layer
thickness, while in some of our data the spin-diffusion length in the N-layer exceeds the
sum of the thicknesses of those layers. Comparing Figs. 4.] and 4.9 shows that replacing
Cu by CuGe did not reduce the differences between AR(H) for interleaved and separated
samples, independent of the number of layers, N, or the thicknesses of the Cu or CuGe
layers. We conclude this part of the discussion by noting that the differences in absolute
magnitudes of the Cu and CuGe data in Figs. 4.1 and 4.9 are not an issue in the analysis
of mean-ﬁee-path effects, as they can be explained straightforwardly within the 2CSR
model. According to qu.1.l-l.3 of the 2CSR model in Ch. 1, replacing Cu by CuGe

should increase ARAp and ARp by simply the increase in speciﬁc resistance,

93

2N(pcl,g¢tc‘,gc - pollen), due to this replacement, and should decrease AAR by the ratio of

the total anti-parallel speciﬁc resistances, AREP "/ARC"G°. Comparison of the data of

Figs. 4.1 and 4.9 shows that this simple picture properly accounts for most of the
differences observed. Small residual deviations can reasonably be attributed to a

combination of data ﬂuctuations plus corrections due to the actual ﬁnite (but relatively

long) spin-diffusion lengths, which require the theory of Valet and Fert.

 

 

 

 

 

 

 

40 ........ ,--ﬁ,-...,...f
: CuGe (x6) tCuGe I 20 nm:
see :- lnterleeved 1170-11
a“ u“ .
s E 3.3 'I 3
c: G :
.- Q- .
v v 3...; '. d
g g L "I. 4
38 L a. 4
l’ '* .
"'2 ECuGe (x4) tCuGe I 20 nm‘; : l
21 . Sepereted 11804!: 3.; Cue-(x6) _ tCuGe-ZOnm 1
“A . ‘1. : «f‘ : Sepereted '., 1169-6 3
C no} . G
h- . e-
m a:
< 1:

 

 

 

 

 

 

 

 

 

 

 

42 E 30001131 tCuGe-I zonmi
"A
E
G
~
I
<
{cw-(x!) -: tCuGe-zonm-f
A = Separated ". 1182-1 3
N ‘7 r '. 1
E s '- a
G 3 ‘:
.- l
K
<

 

 

 

 

0 500
H (0e)
Fig.4.9. AR of interleaved and separated samples identical to samples in Fig. 4.1 except
that CuGe replaces Cu. Symbols and curves have the same meaning as in Fig. 4.1.

94

 

 

'60 '(x'z'I' ' ' ' ' ' ' Interleaved :

 

 

 

 

   

 

 

 

 

 

 

 

 

 

 

  
 

    
 

 

 

 

 

 

 

 

 

 

 

 

 

E'Cu'(x2')q I I Separated: t CuGe (x2) ,- Separated

10.4 - - 15 2 r
:1“; 20m 1182-7 3 :- - ; 20nm 1182-8

E ' F 1, ‘: E 16.1 f ‘-
(2 '~. : C: .
2; 2'. 16‘
n: a: E
< < 15.91.
‘ 15.8‘
-1000-500 0 500 1000 .1000-500 0 500 1000
H(Oe) H(Oe)

20,5. 5.188158) """ 1181185083: 26 8 E ' 005013“ 'Ingrieaved 3.
A - 30m“ I . l A . ‘l" 40nm 1188-2 '
c: . ' é ‘ \ 3
- 20.3 c, 26 6: 13‘ 3
E v 26 5 _ bamo'. 3
< 20.2 < 26 4’ W i

. 1
61.8503)“ ' ' ' ' 866311213173 133$? (£21 ' 'T ' 805515130?
.. 18.4 f. 30nm 1189-5 ; A 23.5 . 400m 1188- 1
N > 1
’ ‘ E 23.4; 1.
E 23.3? 17.: ’7:
E” 23.25 “M WW?
< > 1.. :
23.1 .;
3 23¢ ............................. 1

4000-500 0 500 1000 4000-500 0 500 1000

H(Oe) H(Oe)

Fig.4.10. Effect of further increasing CuGe thickness on AR of interleaved and separated
samples. The convention for symbols is the same as in the previous ﬁgures.

95

12

 

 

    
    

 

 

 

 

 

 

 

 

 

 

 

 

 

    
 
  

    

   

 

 

 

 

 

 

 

 

 

 

 

 

 

  
 
 

 

 

 

"c'd‘IiSIT'T'""" 't'c"-'20'n'ni; 10.4 ‘m3l*"”" Vida -10Fm1
»lnterleeved " - :lnterleeved " .
11.0 L “70-0; 152; 1177-2
"A b -t “a :
E ’ ‘ E 1. L
‘3 11 . J G E
r. . :17 0 _.
10.0; L ‘7" f
: ‘ 17.4‘ ’
1 P r- ‘3
’..3'frfmfu-.fluul.-. h
'Tqu'f)“ ‘ - - . IT” h t“ 'IIZ'OVIIIII; 10.4 L'Cfuvéer (iifj ' ' '7 ' ' ' $.11. {117313114
. Z 1
12 .3'9"““' 1170-0: ".2 :3"’""" 1177.3;
N‘. E 4 ”A : 1
E ‘ E 18:- a
c j c _ 1
=- . 2.17.8 F -
‘ I- 1
5 1 5 17.0 g 4
‘ 17.41. 4"
17.2 .A A+A1AAA414ALALA LJ‘1A_AA_A‘AAAA
-‘|000 -500 0 500 1000
H (De)
20.2 ' I I 'V—fﬁ'Vﬁ ' V 2". ."f ‘ 'VV'r'77'v~vv v ﬁv‘
ECuGble) tum-101m:- 2":CUGEIX3) tam-3011111;
b ' EIN‘.‘ V - ' ‘1
2° blnterleeved : '0! Cd 1100-43
; «as- 20.7:- -'
In - c 2u E
a: : E
: 20.4 ; .-;.
10.41 , ".3 E _3
" """T' ' """fT '- Ecu 5'23 r ' ﬂ ' I
EgrpcetdfeSd) tcw' - 20 "m3 23-9 ESepGere‘ted) tcm' - 30 "m -I
21.0 - 1100-0- E 1191-7:
. 3 we 22.. ’- “E
21.4: ~ - 5
. ‘ c 22e7 : :
. - b :
:13 5 22.8E
22.0 _
3‘ 22.4 E
20.. F4..-2A--.1..-.12...1-.--1-4.._. 22.3:
4000 .500 0 500 1000
H (0e)
34-3 C'G'v(3)' t'
u e x
. O
3‘7 glnterleeved c' ' 1
u... 04.03- 3
E ’ ’
c: 04.0‘_
:044’ -
040 f
:42 I
2. 1 {6113;};in ' Twig“. - 40 um;
' E8.P.'.E.d ‘1'0..‘—:‘
«A 2.5 1
5 : :
4 4
E20.0_ . .3
20.0 E -:
20.0 R i

 

 

 

‘ 21‘0‘0‘0‘ ‘-000‘};‘(1g‘.‘)‘066 ‘ ‘1‘0‘0‘0‘ ‘

Fig.4.11. AR vs H for Cu, CuGe with tCuGe = 10,20,30,and 40 nm. The number of
repeats is 3 for all samples. The speciﬁc resistance ARp increases with CuGe thickness,
and AAR decreases, but the difference between separated and interleaved samples

persists. Data shown are after samples were taken to saturation.

96

In addition to the claim of mean-free-path effects in Co/Cu multilayers in [89], one of
the authors of that paper subsequently argued [99] that the differences between
interleaved and separated multilayers seen earlier in [83] and [84] might also arise from
mean-free-path effects instead of ﬁnite spin-diffusion lengths. To test this possibility, we
replaced the 6 nm and 1 nm thick layers of Co in interleaved multilayers with much
thicker layers of alloys that have much higher resistivities and thus much shorter mean-
free-paths, but also short spin-diffusion lengths. Based on prior studies, we chose 30 nm
thick layers of Py (a well studied ‘low Hs’ alloy) to replace the 6 nm thick Co layers, and
15 and 30 nm thick layers of CoZr (an alloy with a large resistivity per atomic percent
impurity [91]) to replace the 1 nm thick Co layers. Independent measurements of the
resistivities of our sputtered Py and CoZr give pp, = 110 nan (Ch. 1) and pool: = 195
nﬂm (Ch. 1 and 5). Such values reduce the ratios A/t by factors of ~10-20 between IQ, =
6nmandtcOz,=15and30nmandby~60betweentc°=1nmandtpy=30nm.
Contradicting expectation if the ratios t/It are important, Fig. 4.12 shows that these large
increases of the ratio in every layer did not decrease the difference between interleaved
and separated data. Rather, the differences are larger than those in Fig. 4.1. Increased
differences are expected if ﬁnite spin-diffusion lengths in the Py [74] and CoZr (Ch. 5)

layers partly ‘decouple’ the Py and CoZr layers in the separated multilayers.

97

Annual”)
8 2 8

AR (In 111’)

 

v v V v v e

I r I I

. mmvcwuzo

V0011

  

30)ICuGe(20)).

 

 

 

 

 

 

 

Separated

 

[Pyt30IICuGe(20)L/[Colr(30)lCuGe(20)].

v

.1

1211-1

q

 

 

AR (In 111’)

 

AR( mm“)

AR (19 m2)

 

122
120 C
110
110 C
114
112

110 C

 

. [py(3oyCuGe(20yCoZr(30)ICuGe(20)].

 

 

AJALAA‘.1.A.A1A.

V r‘ ' V V V V

L4;21.2Lxlx

YYﬁ'v‘

[PymmucomII,IICozr130IICuG-IzoII,

v V 1+

 

120 g
110 L

C SOP-rated '? 1211-4
110 E 9

I

: 9

I b
112 r
110 :-
1“ A 2 L A n A A A l ‘4‘ L A A 2 2 A

-150 -100 -00 o 00 100 150
H(Oe)

 

VV‘V—Y—IV

AlAAAAleAAL

vv’v—v—jrwvv

LA

L4

Ivvvv

[Py(30)ICuGe(30)ICoZr(16)ICuGe(30)].

 

AELAA

Iv—v—WV

 

AAIAAAA

 

 

.,..-.r.
00:
"Aur-
b
E [ Sepereted
Cu-
e- .
v
a: t
(82-
I
so.

 

450 400 -50

ALILLA

V—felvvvv

“l

0
H(Oe)

ALAL_AA

I Y I

[Py(30)ICuGe(30)] .IICoZr(15)ICuGe(30)] .

1211-8

 

LALL

YV

1'17"

 

A‘L

100

150

Fig. 4.12. AR vs H for multilayers with short 2V! in all layers. The difference between
separated and interleaved samples has grown, disagreeing with mean-free-path effects.

98

Lastly, we placed a 1 nm thick layer of FeMn in the middle of the central Cu layer of
a separated Co/Cu sample to see if it would bring the separated data closer to the
interleaved data, as expected for ‘mean-free-path effects’. In [73] we showed that 1 nm
of FeMn causes > 99% spin-memory loss. If this loss is dominant, inserting the F eMn
layer should completely isolate the contributions from thick and thin Co layers,
producing exactly the opposite effect expected from ‘mean-free-path effects’. Fig. 4.13

shows that the data support the dominance of spin-memory-loss. Inserting the F eMn

gives two separate peaks, with AAR(H) almost zero in between.

[Co(6)/Cu(20)]NI[Co(1 )l(3u(20)]N

 

 

2
l
.3
00

am
cub
N
01

 

AR (1
F3

-
vvw vva'Y

 

 

 

 

 

.e
.3
0|

 

 

-1 -0.5 0 0.5 1
H(kOe)

[Co(6)/Cu(20)]N_1ICo(6)ICu(10)IFeMn(1)lCu(10)I[Co(1)ICu(20)]N
m ﬁr ....................... : 18.5 f ............................ ‘

 

 

.
1
° I
I
L

    

 

 

 

 

 

””-‘1‘”f075“”0”010”?” ' W‘-‘1‘”-0L.§“0‘“6:““‘
H(kOe) 110100)

 

 

Fig. 4.13. AR vs H curves. All samples are of the separated type. The bottom two have a
1 nm FeMn layer in the middle of the Cu layer that separates the Co(6) layers from the
Co(l). Strong spin-ﬂipping by FeMn decouples the two sides of the multilayers giving
two separate peaks on each side of H = 0.

99

4.5 EFFECTS OF SPIN-MEMORY-LOSS

From the data in section 4.4, we conclude that the ratio [Vt is not driving the
differences between interleaved and separated data—there are no ‘mean-free-path
effects’ of the kind described in [89]. In this section we ﬁrst ask what contribution to the
data of Figs. 4.1 and 4.9 can come from VF theory taking the best available values for the
bulk spin-diffusion lengths, 1511‘, in Co, Cu, and CuGe. We will see that these ﬁnite spin-
diffusion lengths can explain between 20% and 50% of the observed behaviors for
samples with different values of N. We then show that the remainder can be consistently
explained by spin-memory-loss at the Co/Cu interfaces.

A. Spin-Memory-Loss in bulk Co, Cu, and CuGe.

Published estimates of 1%,? lie in the range 60 i 20 nm [76,77]. For Cu, we estimated
1E}: ~ 500 nm [73], with uncertainty from 400 mu to 1 pm. For CuGe, we estimate [Elf-"3°

= 130 i 30 nm from the data in [68]. For quantitative comparison of separated (Sep) and
interleaved (Int) samples, we examine the ratio (AAR)s¢p/(AAR)Im. Experimentally, we
take (AAR)1m at the peak of the interleaved data, and (AAR)sep at the plateau just beyond

the ﬁrst peak of the separated data. The results for IN = 20 nm are shown as a function of

N in Fig. 4.14 for samples with both Cu and CuGe. The predictions of VF theory for this
ratio, assuming 1%: = 60 nm, ($19: 500 nm, and 15“}?Ge = 130 mm are the dotted and long-

dashed curves, with error bars showing the ranges of uncertainties due to the uncertainties
in 15;: listed above. These two curves account for 20% to 50% of the observed differences

between interleaved and separated samples. We, thus, disagree with [89] that ﬁnite

100

values of 13}: are irrelevant to the differences in Fig. 4.1, but we agree that a further

mechanism is needed to fully explain those differences. To account for them by [$8

alone would require 1%,? ~ 10 nm, a value that seems too short.

 

1 ﬁt I V l V Y I I I j T Y V l

 

 
 
 
  

 

 

 

 

 

 

...-.....,..
t O Co/Cu :
1 ...... m... D COICUGC ‘
+1 0.8 -‘ \ --..., -1
5‘- - ‘ ""---..., Cu No 25% —,— .
A s ‘ ---...__~
% § ‘ ‘ Mun...“ ..... ~.. J
< 0.6 ‘ ‘ § ‘ “m... e "'
v CuGe No 25% ‘ s ‘
\ ‘ s ‘
8
Am 0.4 .‘u... , 0 .-
m "~-91’-’.V1‘P25/° 1
<0 . “‘---
g 0.2 — I! -
L CuGe with 25%
0 4 4 l j l 4 e l 1 J l l n e l e n 1 1 l 2 1 PL 1 L 4‘
3 4 5 6 7 8

Fig. 4.14. (AAR)s,p/(AAR)1m vs N. The dotted and long-dashed curves are predictions

using VF theory with the parameters for Co/Cu of [25] plus 1?? = 60 nm, [E]? = 500 nm,

and [Elm = 130 mm. The vertical bars indicate the ranges of uncertainties due to
uncertainties in these values of 13p. The short-dashed curve for Co/Cu and the solid curve
for Cu/CuGe add 51 = 0.25 to the prior calculations. Now the error bars indicate the
ranges of predictions for 6, = 0.25 :t 0.1. For these curves, the error bars due to
uncertainties in 13;: are smaller than the symbols for the data. (AAR)s¢p/(AAR)IM should be
equal to 1 for all N, according to the 2CSR model. The uncertainties in the data are
comparable to the symbol sizes.

101

B. Possible Spin-Memory-Loss at Co/Cu interfaces.

An intriguing possibility is spin-memory-loss (spin-ﬂipping) at F/N interfaces
[74,84]. With no direct evidence for such loss, our analysis must be indirect. We
recently showed that spin-memory—loss can be large at Nl/N2 interfaces; values for
V/Cu, Nb/Cu, and W/Cu ranged from ~ 6%/interface for V/Cu up to ~ 60%/interface for
W/Cu [73]. We attributed these losses to spin-orbit-induced spin-ﬂipping in high-
resistivity interfacial alloys. Adding additional ﬂipping due to magnetic disorder in
alloyed regions at F/N interfaces could well enhance interfacial spin-memory-loss.

In this section we ﬁrst show that modest spin-memory-loss at F/N interfaces wouldn’t
destroy the agreement of published data with 2CSR and VF models. We then show that
modest spin-memory-loss at the Co/Cu interface can ‘explain’ our data for Co/Cu and
Co/CuGe.

The model we use to treat spin-memory-loss at interfaces was outlined in Ch.3. VF
theory treats spin-polarized transport through a multilayer in terms of spin-dependent,
spatially varying, electrochemical potentials and currents. Each layer is characterized by
two lengths, the layer thickness and the layer spin diffusion length 15;. At the layer
boundaries, the chemical potentials and currents are matched, taking account of
interfacial speciﬁc resistances where needed but neglecting any interfacial spin memory
loss. F ert and Lee [93] included interfacial spin memory loss by deﬁning a ‘spin-ﬂip’
interface resistance. We use an alternative procedure for PM interfacial spin ﬂips that
builds upon our treatment of spin-memory-loss at interfaces between nonmagnetic metals

[73]. We represent the F/N interface by an additional ‘layer’ (1), taken to be a

102

homogeneous slab of thickness t1, resistivity p}, ‘bulk’ scattering anisotropy ,6, (chosen
equal to the interfacial 71cm) and spin diffusion length 1le , subject to the constraint

AR},N = pl’t ,. VF theory then involves matching electrochemical potentials and currents
at the boundaries of this I-layer with the F and N-metals. Such matching depends only on

two parameters, the interfacial spin-memory-loss parameter 5, -=- t, / IslF , and the
product ple’F. However, the constraint AR;,N = pjt, lets us write P113]? = ARﬁm / 6 ,,

leaving only a single independent parameter, 6,, to control the process, independent of

any particular choice of t, —i.e. t, turns out to be arbitrary.

We have written programs to solve the VF equations for interleaved and separated
samples up to N = 8 (equivalent to a standard [F/N]N multilayer up to N = 16). As an
example of a solution, we show in Fig. 4.15 the calculated dependences of AAR upon 6,
for interleaved and separated Co/Cu samples with N=6, taking the Co and Cu parameters

in [5] and assuming, for simplicity, 152°: 13%": 00. The only unknown in the problem is

then 6,. The solid curves include the Nb/Co interface resistance, ARC-0”,",= 3 fnmz, at
each end of the sample. For the interleaved sample, AAR passes through a weak
maximum with increasing 6,. Over the range from 61 = 0 to 61 = 1, AAR varies by little
more than 10%. In contrast, for the separated sample, AAR decreases monotonically and
rapidly; by 6} = 1 it has dropped by ~75%. To clarify the source of the weak maximum in

Fig. 4.15 we also display as dashed curves the same calculation, but now taking

ARNWC°= 0; i.e., neglecting any contact resistance. For 6, =0, the larger AAR for the

interleaved and separated samples follows from Eq. 3.3 in Ch.3 simply by setting

103

ARNb/Co= 0 in the denominator ARAp. Both dashed curves now decrease monotonically
with increasing 6; (increasing interfacial spin memory loss), and join nicely onto the solid
curves for large 61. By large 61, the two Co/Nb interface resistances have dropped out of
the problem—in this limit, ARNb,Co can be set to zero. The initial rise in AAR in the solid
curve for the interleaved sample is a consequence of ARWCo gradually dropping out of

the denominator of Eq. 3.3 as 6, increases. This rise is an example of a counterintuitive

behavior of CPP transport in the presence of a large contact resistance——increasing spin

memory loss can initially cause the CPP-MR to increase [26].

 

 

 

 

 

7 F 7 I T IT f f
‘-~. =6
6 ‘ ‘~,‘ Interleaved -
A 5 “ ‘ s "
N I ‘. l
a .1
M 3 _ '7Separated “Klo . _
< : “ ~“~:.e .
< 2 : 1 . .1)
1 j- _
O: l l--:IA‘A-TA:A-
0 05 1 1.5 2 25
8

Fig. 4.15. Calculated values of AAR vs 5, for interleaved and separated multilayers
with N = 6. The solid curves include the Co/Nb contacts; the dashed curves are the same
calculations but without those contacts. The ﬁlled circles correspond to Eq. 4.1 with n = 5
for the interleaved case and n = 1 for the separated one.

104

In the limit 6, >> 1, the samples reduce to a linear combination of units consisting of
the Cu interlayers bounded by two thin slices of Co/Cu interfaces (of thickness 1le ). AAR

can then be described by Eq. 4.1, which is a simple extension of Eq. 3.3 to this limit.

 

- I 2
2Pllsr + pCutCu

where for separated samples n = 1 and for interleaved samples n = (ZN-1). As expected,
the solutions of Eq. 4.1 (solid circles in Fig. 4.15) agree with the numerical solutions
(curves) in the large 6] limit. These two limits of Eq. 3.3 are equivalent to the results of
the ‘non-local’ model of [89]. AAR for the interleaved samples is the sum of
contributions from 2N-1 sandwiches composed of two Co/Cu thin slices on opposite sides
of each Cu layer, because the magnetizations reverse from each Co layer to the next.
Irnportantly, as shown by the open boxes and dotted curves in Figs. 4.1 and 4.9, the same
value of AAR is seen in the as-prepared state of separated samples, and almost the same
value is seen in the as-prepared state of interleaved ones. We saw in section 4.2 that the
magnetizations of adjacent layers also reversed from Co layer to Co layer in the as
prepared state of the separated samples and approximately also in the as-prepared state of
the interleaved ones. In contrast, after the separated samples have been taken to
saturation (solid symbols in Figs. 4.] and 4.9), the magnetizations of adjacent Co layers
at the intermediate ﬁeld corresponding to the ‘AP’ state are all parallel to each other,
except at the boundary between the two different Co thicknesses. AAR then contains a
contribution from only the one pair of Co layers for which the adjacent layer

magnetizations are opposite to each other (i.e. now n = l in Eq. 4.1).

105

We now consider what effect ﬁnite interfacial spin memory loss with 6, = 0.25 would
have on the Co/Cu parameters that were previously derived assuming no spin-memory-
loss [5]. Those parameters were derived using [F/N]N multilayers in which all of the F-
and N-layers were identical, i.e. samples effectively ‘interleaved’. The solid or dashed
curves for interleaved samples in Fig. 4.15 show that AAR is not sensitive to the presence
of 6;, until 61 becomes larger than about 1. More generally, a non-zero 61 S 0.25 makes

little change in the linear variations with the thicknesses tp and IN of the quantities ARM)

 

and J AR AP . AAR used to determine 2CSR model parameters for multilayers of the form

[Co/Cu]N [6,7]. To show this behavior quantitatively, Fig. 4.16 uses the parameters for

 

Co and Cu given in ref. [5] to calculate ARM: and J AR AP . AAR for ﬁxed too = 6 nm and

for too = (Cu as functions of N. The ﬁlled circles are calculations with the 2CSR model

assuming no spin-memory-loss anywhere. The solid curves are the VF model including

spin-memory loss both in the bulk Co and Cu and at the Co/Cu interfaces—13%“ = 500

run, If; = 60 nm, and 51 = 0.25. The differences between the ﬁlled circles and solid lines
are small, and the forms of the two sets of data are very similar. Recalculating the solid
curves assuming, instead, If; = [SC]? = 00 makes almost no change. A similar analysis for

an inﬁnite multilayer also produces little change. We conclude that adding interfacial
spin-ﬂipping with 61 S 0.25 has little effect on the previous 2CSR model analysis of data
for [Co/Cu]~ [5] and [Co/Ag]N [28] multilayers with equal Co layer thicknesses, only
requiring renormalization of the model parameters by amounts small enough to lie within

the previously speciﬁed uncertainties.

106

40 I I I I I I 40

 

 

 

 

 

”a?
D
a; Q
>
E J”
”av
11 H:
< N
a:

< A
B
BIN)
5 - °° - 5 V

O J l l l I l 0

6 8 10 12 14 16
N

 

Fig. 4.16. Calculated values of ARM» (left scale) and JAR“, . AAR (right scale) for
multilayers with total thickness = 360 nm and either ﬁxed too = 6 nm or (Co = tCu. Filled
circles are for the 2CSR model. Solid curves are VF calculations with 15C}.u = 500 run, 193?

= 60 nm, and 61 = 0.25.

In contrast to having little effect on both interleaved samples and simple [Co/Cu]}v
multilayers, Fig. 4.15 shows that including interfacial spin-ﬂips with 51 = 0.25 has a large
effect on separated samples, and we’ll see in Fig. 4.17 that it can also cause larger
changes in Co-Cu based exchange-biased spin-valves.

In Fig. 4.14, the short-dashed curve for Co/Cu and the solid curve for Co/CuGe in Fig.
4.14 show our calculations of (AAR)s¢p/(AAR)Im adding 61 = 0. 25 to the spin-diffusion
lengths listed above. The uncertainties in these curves due to those in 13}: are no larger
than the sizes of the symbols for the data. The error bars shown for these curves are for

6} = 0.25 i 0.1. This value of 6, can “explain” our data.

107

In two initial papers [105,106], we referred to the value of 61 = 0.25 as corresponding to
25% spin-memory-loss at an F/N interface. For spin-memory-loss at N1/N2 interfaces,
the ratio 62 = t/Isp appears in AAR as a simple exponential, exp(-6), and exp(-0.25) z 0.78
corresponds to a 22% reduction in AAR. When 6 is not too large, it thus approximates the
percentage spin-memory-loss. In contrast, Fig. 4.15 illustrates that the situation for F/N
interfaces is more complex—cg, in interleaved or simple [F/N]N multilayers, 61 = 0.25
can produce little change in AAR. So in this thesis we simply describe the results
produced by 61 = 0.25, and never refer to a consequent percentage spin-memory-loss.

We conclude this discussion of spin-memory-loss at Co/Cu interfaces by: (a) arguing
that the loss observed is plausible if the interface is viewed as a random alloy of Co and
Cu, and (b) showing that some previously published data provide indirect support for

such loss.

(a) We estimate 6, = t1/ IslF for an assumed Co-Cu 50-50 alloy with thickness t1 assuming

that spin-ﬂips are due to spin-orbit scattering. Since the spin-ﬂip cross-section of Co
in Cu is rather uncertain [71], and we don’t have reliable information about the ‘size-

effect’ parameter p11; [91] for the interfacial alloy, we must make assumptions. We
assume that p12,, has the typical value p171; = 1 ﬁlm2 [91], and that p1!) =AREO,Cu ,
where ARENCU = 0.5 me2 is the measured Co/Cu interfacial resistance [5]. Dividing

the second relation by the ﬁrst, we ﬁnd t/l; z 0.5. To calculate 6,, we must relate ,1;

to ls’r- To do so, we assume that the minority and majority mean-free-paths in the

interface have the ratio set by our measured 7 for Co/Cu interfaces—i.e. that AI and

108

2*, are related by 711/1} = ARéomu/ARg-omu: (ma/(1-» z 7 [6]. Since the
transport mean-free—path is the sum of the spin-dependent mean-free-paths, the

effective mean-free-path related to the interface resistivity p1, is essentially just the

long one, III. In contrast, from Fert et a1. [72], IslF is related to the short mean-free-

path A} and the mean-free-path for spin-orbit scattering, 2.55 by 151,.- =1/(4I’1sr )/ 3 ,
because the x1, under the square root is determined by the equation l/zl, = (1/ 1}) +

(ll/1% ). If, from ref. [71] we take Asp ~ 20 A} (20 times the transport mean-ﬂee-

 

path), we obtain 1le z \/ ( dilli )[ 20 /( 7x3 ) ] ~71}. Our estimate then becomes 61 =

t1/ 1;; = t,/ 11; = 0.2, large enough to ‘justify’ our experimental estimate 61 ~ 0.25.

(b) As noted above, adding an interfacial 61 = 0.25 has minimal effect on AAR for simple
F/N multilayers with identical F layers, or for interleaved multilayers. Such an
addition can, however, affect AAR in exchange-biased spin-valves (SVs). In an
earlier paper [107] we reported that data for both symmetric (equal Co layer
thicknesses) and asymmetric (one Co layer thickness ﬁxed and one variable) Co/Cu
exchange-biased SVs fell below the values predicted using non-adjustable,
independently determined, CPP-MR parameters for Co, Cu, and Co/Cu interfaces.
Fig. 4.17 shows those data and those ﬁts (dashed curves). We also noted [107] that
most of the discrepancies could be removed by a combination of modest variations in
the assumed parameters plus corrections for failure to reach a fully AP state in the
symmetric SVs with thick Co layers. The solid curves in Fig. 4.17 show that an

interfacial spin-memory loss of 6; = 0.25 provides an alternative possible explanation.

109

These curves (and the uncertainty bars shown) were calculated using exactly the same
parameters as for the dashed curves, but simply adding interfacial spin-memory loss

of 51 = 0.25 i 0.1. This completely non-adjustable procedure ﬁts both sets of data

 

 

 

 

surprisingly well.
1..wT,....,.....ﬁj..,j..
. ,I' A ‘ .
0.8 :' o’ ‘ 4 1
l- 1” .4
NE 0.6 _- x, “ ‘ - 0.8 I:
G : ‘ I, A .. 7’
‘5 0.4-E "'A -O.6 A
m " ,..---""- l 5
<1 ' ,.-—-" 4 B
< 0.2 ‘ '."D O - O4 VN
' ‘ .a'. . . O ‘
. .-. - o O e——>;
O - 0' - 0.2
O .
L 1 1 1 l 14 1 1 l 1 1 1 1 I 1 J 1 1 l 1 1 1 1 d 0
O 10 20 30 40 50
t (nm)
Co

Fig. 4.17. AAR vs too for symmetric (ﬁlled triangles) and asymmetric (open circles)
Co/Cu exchange-biased spin-valves. The dashed curves are prior ﬁts [25] assuming no
spin-memory-loss at the Co/Cu interfaces. The solid curves just add an interfacial spin-

memory-loss of 61 = 0.25.

4.6 STUDIES OF CIP-MR IN Co/Cu BASED MULTILAYERS AND

CPP-MR IN Co/Ag BASED MULTILAYERS.

For completeness, we also measured the CIP-MRS of interleaved and separated Co/Cu

multilayers. If mean-free-paths are the dominant lengths scales in both the CIP- and

110

CPP-MRS, the two might be expected to show similar differences between interleaved
and separated samples. Indeed, we previously reported similar differences for
[Py/Ag/Co/Ag] multilayers [83], but attributed them to different underlying physics,
mean-free-path limitations for CIP, as noted in the introduction above, and spin-diffusion
length limitations, mostly in the Py, for CPP. Fig. 4.18 shows that we ﬁnd similar
differences in the CIP-MRS and CPP-MRS for Co/Cu multilayers. Again, we attribute
them to different physics—the CIP-MRS to expected mean-free-path effects there, and
now the CPP-MRS to a combination of spin-memory-loss in the bulk materials and at the

interfaces as argued above.

CPP CIP

 

 

   

 

 

 

 

 

 
   

 

   
   
 

 
 

 

 

 

      

£60.53? ' ' ' ' Watergate
.\° as
I! a:
E- 5.
IL 0.
(‘5 6
prﬁTervvavvvvl rivvv'vvvvfrfv‘rj vvvvvvvvv 'vvvv‘
30%- Cu (x8) - Cu (x8) Separated .
s 'l . 8’- -‘
°\. 25 - 11 J °\° L ‘5.
a: 20 :- ° .3 :1: EL 11.,“ -
§ e 3 s I "
a 15 :- 4. n'. 4: _
o 10E— —3 0 2E
05.... . -1.A--- .0 b o 3“an ,
4000-500 0 5001000 - 000-500 0 5001000
H(G) mm

Fig. 4.18. Comparison of CPP-MR and CIP-MR for separated and interleaved Co/Cu
multilayers. Symbols and curves have the same meanings as in Fig. 4.1. The open
squares for the separated CPP-MR were noisy, as shown by the ﬂuctuations in the points
for H = 200-500 0e. To reduce uncertainty at H z 0, that square is the average of six
very low ﬁeld measurements. All other points are single measurements.

111

Lastly, to see if the results of Fig. 4.1 are highly sensitive to details of the mutual
solubilities and band structures of the F - and N-metals, we checked the differences
between interleaved and separated samples with Cu replaced by Ag, which is insoluble in

Co. Fig. 4.19 shows data for N = 8 similar to those in Fig. 4.1.

 

30 E Ag(x8) Interleaved

..
h

    

 

 

h
-
20 llIllllllllllllllLlllllllll

 

 

]_l'

3 Ag(x8) g Separated -

 

 

 

20 llLLJlllLllllllllllllllllllll
-1000 -500 0 500 1000
H(Oe)

Fig. 4.19. AR(H) vs H at 4.2K for interleaved (top) and separated (bottom) multilayers of
Co/Ag with N = 8. The scale units for each coupled pair are identical. The open squares and
dashed curves are for the initial state and the ﬁlled circles and solid lines are after samples were
taken to saturation.

 

112

4.7 SUMMARY AND CONCLUSIONS

In this chapter we have done the following:

(1) Conﬁrmed the differences in CPP-MR reported in [89] for interleaved
[Co(6)/Cu(20)/Co(l)/Cu(20)]1v and separated [Co(6)/Cu(20)]N[Co(l)/Cu(20)]N
multilayers aﬂer they are taken to above their saturation ﬁelds, and shown that the
differences persist in samples with repeat numbers as small as N = 2.

(2) Showed that the CPP-MRS in the as-prepared states of separated samples are as
large as those at the peaks of interleaved samples. We attribute this result to
magnetic structures of the separated samples in the as-prepared state in which
magnetizations of adjacent layers are reversed, as distinct from the state where the
magnetizations of thick and thin layers are reversed. This result conﬁrms the
argument in [89] that the magnetic order of adjacent layers in separated samples is
crucial.

(3) Used both magnetization and polarized neutron reﬂectivity measurements to show
that: (a) our samples do reach closely anti-parallel (AP) states and at the same
ﬁelds for interleaved and separated samples, and (b) any unwanted differences
between the magnetic structures of interleaved and separated samples are small.

(4) Showed that the differences in CPP-MR remain essentially unchanged when Cu is
replaced by Cu(2%Ge) with thicknesses that increase the ratio ten/30. of mean-
free-path to layer thickness in the Cu from about 1/5 to 5, while still leaving the
spin-diffusion length long. We infer that mean-free-paths are not scaling lengths

in the CPP-MR of present day samples.

113

(5) Showed that the differences remain unchanged, or even increase, when the Co
layers are replaced by Py and CoZr layers for which t/l increases by more than
factors of 10. This result is evidence that the CPP-MR scaling lengths are spin-
diffusion lengths, not mean-free-paths.

(6) Showed that insertion of a thin (I = lnm) FeMn layer into the middle of the central
Cu layer of a separated Co/Cu multilayer produced changes in AAR(H) of just the
kind expected for a strong increase in ‘spin-memory-loss’ due to the FeMn layer.
Because the total AR contributed by the FeMn layer and its two interfaces with Cu
was only comparable to the total AR contributed by alloying the 20 nm thick Cu
layer with Ge, any mean-free-path effects due to the FeMn can be ruled out.

(7) Conﬁrmed the claim in [89] that the best available spin-diffusion lengths in Co and
Cu are too long to fully explain the observed differences in CPP-MR between
interleaved and separated samples, but showed that these lengths are short enough
to account for part of those differences.

(8) Argued that the rest of the differences can plausibly be attributed to spin-ﬂipping
at Co/Cu interfaces with spin-ﬂipping parameter 6, a t , / 1le z 0.25, and showed

that the same amount of ﬂipping: (a) can resolve discrepancies between predictions
and previously obtained data on Co-Cu exchange-biased spin-valves, and (b) is
consistent with a simple estimate of effects of spin-orbit scattering in an interfacial
layer composed of a Co-Cu alloy.

(9) Showed that the differences between interleaved and separated states are similar in

the CIP- and CPP-MR5, although we interpreted the two cases differently.

114

(10) Showed that similar differences between interleaved and separated states persist
when the Cu is replaced by Ag, which we take as evidence that the differences
between separated and interleaved samples aren’t sensitively dependent on details
of mutual solubility or band structures.

We conclude that the use of the 2CSR and VF models in most published work on the
CPP-MR still seems justiﬁed, although if interfacial spin-ﬂipping at F/N interfaces is
eventually conﬁrmed, some parameters may need to be slightly modiﬁed. The results of

this chapter are published in refs. [105,106,108].

115

Chapter 5

METHODS FOR ENHANCING THE CPP-NIR

5.1 INTRODUCTION
For applications of the CPP-MR in devices, one wants to maximize AAR and/or ARAp
for a sample of ﬁxed total thickness. According to Eq. 1.2, AAR can be increased by

1 1 1 1 1 1 O 1 1 .
Increasmg ,6, increasmg pp, Increasmg 7m: A Rm: , increasmg the number of interfaces, or

eliminating the contribution of the lead or contact resistance to the denominator of AAR,
provided that increasing one parameter does not cause offsetting decreases in others. In
this chapter we describe controlled studies involving one or more of these changes.

In 5.2, we describe an attempt to increase both AR» and AAR in Co/Cu multilayers
by alloying the Co with Zr. Zr was chosen based upon the tables of resistivity per
atomic percent impurity in Bass [91] and of ,6 in Campbell and Fert [46]. The former
shows that Zr produces a relatively large increase in resistivity in Co, and the latter that
Zr is one of only a few impurities expected to produce a ,6 equal to or larger than that in
nominally pure Co (60, = 0.46 [5]).

In 5.3, we describe a successful attempt to increase both AAR and AR» in Co/Cu and
Co/Ag multilayers by adding ‘internal interfaces’ within the Co layers. The procedure is
to divide each Co layer into a series of thinner layers by inserting Cu or Ag ‘inserts’ so
thin (I ~ 0.5 nm) that they add little to the total sample thickness and (probably via
pinholes) couple the thin Co layers ferromagnetically so that they continue to behave as a

a single ‘large layer’.

116

In 5.4, we describe a successful attempt to increase AAR of a Co/Cu multilayer by
inserting thin (1thn = 1 nm) layers of F eMn into the Cu portion of its superconducting
Cu/Nb contacts. As explained in Ch. 3, VF theory predicts that inserting a source of
strong spin-ﬂipping into either magnetic or non-magnetic layers of a CPP-MR structure
should eliminate from the denominator of Eq. 1 .2 anything outside of the source of strong
spin-ﬂipping. What will happen upon inserting the source into part of the supposedly

superconducting contact layer was unclear before the present measurements.

5.2 ENHANCE THE CPP-MR BY ALLOYING Co WITH Zr

Alloying F-layers is well-known as a way to enhance GMR. As examples, NiagFeog
(Permalloy = Py) and Coo.9Feo,1 are used in CIP-MR devices and both also enhance the
CPP-MR over the values for the nominally pure metals Ni or C0 [5]. Fig. 5.1 shows how
Coo_91Feo,09 enhances AAR in symmetric EBSVs of Co with Cu. The primary source of
this enhancement is an increase in ,6 from ~0.46 for Co [68] to ~ 0.65 for CoFe [82].
Because adding Fe increases the Co resistivity only weakly [82], ARM: for CoFe is only
slightly larger than for Co alone, as shown in Fig. 5.2.

We anticipated that CoZr would increase both AAR and AR» more than is done by Fe
in Figs. 5.1 and 5.2. Zr produces a relatively large increase in the resistivity of Co per
atomic percent impurity [46,91], but was not expected to reduce ,6 from the value for Co
[46]. We chose 2.5% Zr as large enough to dominate the scattering, and thus to produce

the ,B of CoZr, but small enough, hopeﬁilly, to leave the interfacial parameters those of
Co and the spin-diffusion-length 1%,?” long enough to not excessively truncate AAR for

the thickness of CoZr to be studied. We hoped, thereby, to determine the CPP-MR

117

parameters of dilute CoZr alloys. At the end of this section we will present some data for
C095er and CosoZrzo, the latter above the concentration (~ 5-7% Zr) at which CoZr is
expected to become amorphous [109,110]. As explained in Ch. 2, the Zr alloys were
sputtered from targets into which cylinders of Zr were embedded, the number
determining the Zr concentration.

To show what we hoped to achieve, the solid lines in Figs. 5.1 and 5.2 compare 2CSR
and VF predictions for AAR and ARA]: for a C097,5Zr2,5 alloy with the measured values for
CoFe and Co. As most of our data are for the 2.5 atomic % Zr alloy, hereafter we
designate it by CoZr, and use subscripts only for other Zr concentrations. Our CoZr

predictions use the V-F model, assuming a resistivity poozr = 19.5 “9cm (corresponding

to our measured resistivity for 2.5%Zr—see Ch. 2), ﬂcOz, = 0.5 from Campbell and Fert
[46], and the interface parameters of Co, i.e. 7cm = 7CD = 0.77 and A Raw/Cu = ARCo/Cu =

0.5 me2[5]. The other parameters for CoFe-based EBSVs are taken from ref. [82].

118

Predicted AAR of CoZr/Cu EBSV,
compared to that of Co and CoFe

 

 

 
  
 

 

 

  

 

 

 

 

A c6 1 T e. I . . I .4 . r
E] CoFe
CoZr-inﬁnite ............
N; 1.5 ~_ ----- CoZr-15:8 ---- C] _
c: : ‘3 D 1
11. 1 J
v . q
n:
< 0.5 A A A A 1
0 111.1 im1114111.1 l..1u trill lmhl
0 5 10 15 20 25 30
t(nm)

Fig.5.1. Predicted AAR of CoZr/Cu symmetric EBSVs using parameters given in the text
and assuming inﬁnite spin-diffusion-length, [gab , (solid line) or 1%,?” = 8 nm (dotted
line). The predictions are compared to data for Co/Cu and CoFe/Cu EBSVs.

 

  
 
 

 

35 ....,....,.eee,....,....,....,,...
[ Predicted AR”(t) for CoZr compared to l
: best fits for Co and CoFe q
30 ~—
CoZr

 

M
(II

 

 

 

 

 

NA

E Co

Ci

*

v2 20 b .'_'._,--_

m c ........... ##M

“ 15: ————— .
1O r.111111.111L1111..11111.1.11111114‘

0 5 10 15 20 25 30 35

Fig. 5.2 ARA]: for CoFe and Co EBSVs as functions of layer thickness along with the
predicted ARp for CoZr using the parameters given in the text.

119

To show the effect of a ﬁnite spin-diffusion length, the AAR predictions assume spin-
diffusion lengths of inﬁnity (solid curve) or 8 nm (dashed curve) (less than the value for
CoFe [82]). The predictions in both ﬁgures are much larger than those for Co/Cu and
larger than those for CoF e.

The large resistivity of Zr in Co, coupled with the expected large ,6, earlier spurred a
search for a large CIP-MR with CoZr alloys. However, Bouziane and coworkers [111-
113] found a lower CIP-MR for Cocgzrz/Cu multilayers than for Co/Cu. They concluded
[112,113] that the reduced CIP-MR for CoZr was due to poor CoZr/Cu interface quality.
Dusting the CoZr/Cu interfaces with thin layers of Co recovered the CIP-MR of Co/Cu
multilayers, and dusting Co/Cu interfaces with thin layers of CoZr reduced the CIP-MR.
We still decided to try CoZr in the CPP geometry, on the basis that the dominant
electrons impact the interfaces differently in the CPP-MR and CIP-MR. The dominant
electrons for the CIP-MR are those moving parallel to the interfaces, whereas the

dominant ones for the CPP-MR are those moving perpendicular to the interfaces.
5.2.1 Giant Magnetoresistance of CoZr/Cu Systems

Here, we present a systematic study of the CPP-MR in a variety of CoZr based
multilayers, along with detailed analyses of the scattering parameters of CoZr and CoZr
interfaces with the normal metals Cu and Ag. We began with symmetric CoZr/Cu
EBSVs. When the results did not agree with our expectations, we added asymmetric
CoZr/Cu/Py EBSVs and later CoZr/Cu! Py/Cu hybrid spin-valves to try to pin down the
parameters and clarify the source(s) of the deviations from expectations. To further
explore those sources we later examined similar samples with Ag replacing the Cu. To

simplify the discussion, we will describe all of these latter results in groups.

120

We begin with the 1 data for our symmetric EBSVs of the form
Nb(250)/Cu(20)/FeMn(8)/CoZr(t)/Cu(20)/CoZr(t)/Cu(20)/I\Ib(250), with all thicknesses
in run. Figs. 5.3 and 5.4 compare the measured values of ARAp and AAR for symmetric
CoZr spin-valves with the equivalent values for CoFe and Co. The 2CSR model
predictions were already described for Figs. 5.1 and 5.2 above. The zero on the ordinate
axis of Fig. 5.3 is suppressed to eliminate most of the contributions of the FeMn and
superconducting contacts, allowing more direct comparisons between the contributions of
CoZr, Co, and CoFe to ARM». We see that the ARM» data for CoZr fall near the simple
2CSR model predictions; i.e., adding CoZr increases the total speciﬁc resistance almost
as much as predicted. In contrast, the AAR data fall well below the predictions. This
latter behavior is similar to that found for the CIP-MR [108-110]. Obvious possibilities

for contributors to the small values of AAR are: (a) weaker than expected interfacial

scattering anisotropy 7 and/or interface resistance ZAR"; (b) weaker than expected bulk
anisotropy, ﬂ; or (c) a short spin-diffusion length, [Elf-’2'. The low ordinate intercept

suggests a low interfacial contribution to the CPP-MR. The relative ﬂatness of the data at
high tool: suggests a short 15;. The low values of AAR suggest a small ,6.
We began to try to understand these data by extending our analysis to the Valet-Fert

(VF) model using most of the same parameters as for the 2CSR model analysis above,

but now leaving three parameters adjustable: the Spin-diffusion length in CoZr, [Efe’z' , the
bulk spin-asymmetry parameter in CoZr, ﬂoczr, and the interfacial spin-asymmetry

parameter for CoZr/Cu, yew/Cu. Because the numerator of AAR is dominated by the

product row/Cu AREoZﬂCu (see Eq.l.2), ﬁxing AREoZﬂCu and leaving only yew/Cu

121

adjustable is sufﬁcient to let us characterize the data. However, this procedure leads to
uncertainty in the ‘true’ separate values for parameters 7Com“ and ARCoZr/Cu' Moreover,

the scattered data of Fig. 5.4 alone were not sufficient to pin down the values of all three
parameters very well. In particular, as EBSVs have only two active F/N interfaces,
EBSV data are only moderately sensitive to interfacial contributions to the CPP-MR.

For these reasons, we also measured several additional sets of samples.

First, to try to achieve better AP states, we made asymmetric EBSVs combining

pinned CoZr layers with free layers of Py, which has a lower switching ﬁeld than CoZr.

We started with Py thickness tpy = 6 nm, about equal to the Py spin—diffusion length (13:

~ 6 nm [74]), to keep the Py contribution to the total AAR down, but not make the Py
switching ﬁeld too large (in large area samples, the switching ﬁeld decreases with
increasing layer thickness). Finding that samples with large tooz, didn’t seem to reach
good AP states (see the appendix), we tried thicker tpy = 24 nm, much larger than 1;}.

Second, to enhance the importance of the CoZr/Cu interfaces, and to give larger
signals overall, we made multilayer hybrid SVs combining layers of CoZr and Py. Again
we started with tpy = 6 nm, and then extended measurements to tpy= 24 nm.

Detailed analyses showed that the best-ﬁt parameters to AAR for different sets of
samples were not identical. Rather than describe separate ﬁts to each different set of
data, for simplicity we show ﬁts to just two groups of data: (a) the EBSV AAR data and
(b) the hybrid SV AAR data, and then, for comparison, also a ‘common’ ﬁt to all of the
AAR data together. The separate ﬁts are shown as dashed curves and the common ﬁt as

solid curves. These three ﬁts illustrate the uncertainties in the derived parameters.

122

Fig. 5.5 compares these two best ﬁts with the AAR data for the symmetric CoZr

EBSVs already shown in Fig. 5.4. For the dashed curve, the best—ﬁt parameters are: ,BcOz,

= 0.4, yew/Cu = 0.35, and [SCFOZ' = 8nm. For the solid curve, the parameters are: ﬂcOz, =
0.34, roch/Cu = 0.52, and 1;?” = 8nm , These best ﬁt parameters were determined

using only data for AAR. It is therefore important to check how well they describe the
ARM: data for these same samples. Fig. 5.6 shows that they both predict the values of
ARAp quite well, certainly much better than the 2CSR model did with the original
‘guessed parameters’ for CoZr and CoZr/Cu listed above. We end the discussion of
symmetric EBSVs by showing in Fig. 5.7 that the magnetoresistance hysteresis curves
indicate that we achieved good AP states for both thin and thick CoZr layers.

To continue our analysis, Fig. 5.8 compares these same two ﬁts with our data for the
three different sets of EBSVs. For the samples containing Py, the parameters for Py and
the Py-based interfaces are taken from earlier papers and are non-adjustable. These

parameters are: ppy= 110 nﬂm, ,pr= 0.75, Wy/Cu = 0.7, Apr/c“= 0.51 mez,
1;}? = 5.5nm , and ZApr/Nb= 6.1 me2. Surprisingly, despite the lower reversing ﬁelds

for the Py layers, the hysteresis curves for the EBSVs with Py were not as good as for
those with just CoZr. Hysteresis curves given in the appendix imply good AP states for
thinner CoZr layers, but not so good for thicker ones. The thick CoZr data points in Fig.
5.8B and C should, thus, probably be viewed as only lower bounds. Indeed, the thick
CoZr data points for tpy = 6 nm actually decrease with increasing tam. As shown in the
appendix (Fig. A5), the agreement between the ‘predictions’ for ARM: and the data are

also not as good in the asymmetric EBSVs as they were in the symmetric ones, with the

123

‘predictions’ now lying a bit above the data and growing with increasing teal, a bit too
fast. At least part of the problem may lie with the Py parameters, which are not the focus

of the present work.

 

 

 

35 I I I I T HT fr 1 T T T I I I I I I I I I T 7 I I T f I I I I I I
O CoZr data :
CoZr Predicted -
30 Co Predicted —
---- CoFe Predicted ‘

 

25

20

AP

AR (me2)

15

 

 

 

 

P-
10 IlJlLlllLlLlJllilLllll4llll111411

0 51015 20 25 30 35
t(nm)

Fig. 5.3. ARAp vs I for CoZr/Cu EBSV data, compared with ‘best ﬁt lines’ for CoFe/Cu
EBSVs (broken curve), Co/Cu EBSVs (dotted curve), and the 2CSR model for CoZr/Cu
described in the text above.

124

Comparison of Prediction with Data, Co and CoFe

 

 

 

 

 

 

 

 

 

2 e .ITeIMIchrmm
CoZr-inﬁnite
L ----- CoZr-lsfa .
D CoFe --------- E J
O CoZr-Data """""
A1I5'- A C0 .0... -
NE a. D
c: D D
H- _
v
at , . ‘
4 ..' ’ ‘ A . :2: i
a. . A A T
< . B 2 ° -
m .
011111111.1%.1.L111.1#1111114
0 5 10 15 20 25 30

t(nm)

Fig. 5.4. AAR vs t for Co, CoFe, and CoZr symmetric EBSVs. This ﬁgure is a replot of
Fig. 5.1 with the addition of our CoZr symmetric EBSV data.

Finally, we turn to the hybrid data, for which the AAR data are collected together in
Fig. 5.9. Hybrid SVs have the advantage that both ARM» and AAR are larger than for the
EBSVs, and thus easier to measure. Fig. 5.9 contains the best ﬁt curves for both the

hybrid data alone (dashed curves with parameters ,BcOz, = 0.3, mob/Cu = 0.6, and

Cult
1 SF

= 6nm) and the ‘best common’ ﬁt’ (solid curves with the parameters already
given). The lower [Eli-’2' in the hybrid multilayers could be partly due to the lack of AP
state at moderate CoZr thicknesses (see Appendix for examples of magnetoresistance
hysteresis curves). As for the asymmetric EBSVs, the ARAp predictions are

systematically a bit too large, and the hysteresis curves suggest possible problems with

AP states for thick CoZr layers (see Appendix).

125

Table 5.1 summarizes the parameters from the three sets of ﬁts given above. Two of
these parameters are unexpected.

First, the small value of yum/Cu = (0.3 — 0.5) relative to 7tc/Cu = 0.77. Because 2.5%
Zr is such a small impurity content, we had expected the Zr atoms to be largely irrelevant
to the interfacial properties. However, the CPP-MR analyses of the data in Figs. 5.5 and
5.8 agree with the conclusions of the CIP-MR studies that the CoZr/Cu interface is less
effective in producing MR than the Co/Cu interface. We will see below, however, that
interfacial dusting of CoZr/Cu interfaces with Co, or Co/Cu interfaces with CoZr, gives
less clearcut results than those reported for the CIP-MR.

Second, ﬂew ~ (0.34) is smaller than the value 0.5 listed by Campbell and F ert [46].

Here, however, the uncertainty of the Campbell-Fert value is unknown.

Table 5.1. CoZr parameters found from the ﬁts to our data.

 

 

 

 

 

Sample Set ﬂCoZr 7003 [€ng (nm)
Expected 0.5 0.77 ?
EBSV ﬁts 0.4 0.35 8
Hybrid ﬁts 0.3 0.6 6
Common ﬁt 0.34 0.52 8

 

 

 

 

 

 

126

 

08 IIII IrII rIll 177 IiIi ITII WY!#
0 _ I I r I I _‘

  

CoZr/Cu symmetric EBSV

   

 

AAR(me2)
f:

 

 

 

 

 

 

0.2 i
E Common Fit 3
0.1 _ ----- EBSV-Flt ;
1; 0 Data 1
0 5 10 15 20 25 30
t nm
CoZr( )

Fig. 5.5. AAR vs (Caz: for symmetric EBSVs. The dotted curve is a best to just the three
EBSV data sets for AAR. The solid curve is the best ‘common ﬁt’ to all of the samples,
exchange biased and hybrid.

 

 

 

 

 

 

 

 

 

I T I I I T I I I I I I I I I I I I I I I I I I I I I I I 1 I’ I I I

J

; CoZr/Cu Symmetric EBSV ;

30 _— j

A E ‘
NE 25 _
c: 20 .
II- I
v '

a 15 _ ,

tr“ 3

0 Data _

< 10 F— —COmmon Flt

I ----- Best EBSV Fit .

5 L ----- ZCSRMprediction ;

0' u 111. inilrllml. 111111,]

011‘151111011151 20 25130 35
t(nm)

Fig. 5.6. Using the parameters from the ﬁts to AAR ﬁts the data for ARAp quite well. The
dashed curve is the 2CSR model ﬁt described in the text above. The best ﬁts are closer to
the data than the 2CSR model prediction.

127

21.1
A 21
N
E 20.9
C}
“_ 20.8
n: 20.7
4 20.6
20.5
204

 

'-200 -100 0 100 200 300 400

 

IIIIIIIIIIIIIIIIIIﬁIITIIrITII

FeMn/CoZr(8)ICu(20)/CoZr(8)

 
 

k1174-2

  

I I I I I I I r T

r H I I I I I I I I

 

11111111111111111111L

    

  

 

 
    

L4 11L 1 L 1 L1 1 I 1 L14 #1 l 1 l l l l l l l I

IITIIIIIIIIIIIIIIIIIIIITITIII

FeMnlCoZr(25)/Cu(20)ICoZr(25)
(B)

I.

 

k1 174-3

 

 

 

IIIIIIIIIIIIIIIIIIIIIIIIIITIIIIT Ir

   

 

 

 

 

*
1111111111141111111J11111111111|lrlrrrr

l J J l l L l I l J I 1 l A l l L l l L l 1

H (De)

Fig. 5.7. AR vs II for symmetric EBSVs with tCoZr = 8 nm (A) and 25 nm (B). In both
cases, the hysteresis curves are nicely square, indicating well-formed AP states.

128

0.8

 

    

 

 

0.7 :—
«A 0.5 E“
E 0.5;
Ci :
- 0.4:—
n: 0.3;
q : CoZr/Cu symmetric EBSV

< 0.2 , _:

Common Fit

0.1 . ----- EBSV-Fit 1

r 0 Data 1

 

 

 

 

 

1.05 l

 

CoZr/Cule24 EBSV

P
03
0|
T

 

AAR(me2)
O
N

L- -—Common Flt
----- AdR-EB
0 Data

 

 

I l I L l l I l l

 

 

 

 

 

 

 

 

 

 

o i
i
‘1
11
E] 0 3 _ CoZrICu/Py6 EBSV
< 0.2 ~ Common Fit ‘2
1. ----- 15st Fit «
0.1 :— . DC“ 1
0:‘1‘141+urhr.1..,,1,,,,1”I_1
0 5 10 15 20 25 30

t c ozt(nm)

Fig. 5.8. AAR vs tCOZr for EBSVs. (A) Symmetric EBSVs; (B) Asymmetric EBSVs with
tpy = 24 mm (C) Asymmetric EBSVs with tpy = 6 nm. The dashed curves represent a ﬁt
to just these three sets of data. The solid curves represent a ‘common ﬁt’ to all of the
exchange biased and hybrid samples with CoZr/Cu. In C, the two ﬁts fall on top of each

other.

129

 

M
(II
.4

1
—1

N
O

.3
0|

 

.3
0
fr

F [CoZr(t)lCu(20)le(24)/Cu(20)]")le(24) “f

AAR(me2)

(II
I I I

—Common Flt ;
t ----- rid Flt
_ I Data

 

 

 

L I
ﬁﬁﬁ V 1'

 

.3
0|

 

 

AAR(me2)
3

i [CoZr(t)ICu(20)le(6)ICu(20)]1oIPy(6) 1

 

 

 

 

 

 

 

5 ~ _
: CommonFlt
----- HybrldFlt
[ I 0111
o i 1 14 L1 . i 1 r . . i r T . r 1
0 5 10 15 20

t c oz'(nrn)

Fig. 5.9. AAR vs teczr for hybrid SVs with tpy = 24 nm (A) or 6 nm (B). The dashed
curves represent a ﬁt to just these two sets of data. The solid curves represent a
‘common ﬁt’ to all of the CoZr/Cu data in Figs. 5.8 and 5.9.

5.2.2 Dusting CoZr/Cu Interface With C0:

To see if some property of the CoZr/Cu interface is causing the apparent low yew/Cu
of Table II, we ‘dusted’ the CoZr/Cu interfaces with 1 nm of Co to create Co/Cu
interfaces. The structure of those samples is FeMn/CoZr(t)/Co(l)/Cu(20)/Co(l)/CoZr(t)].

If Zr impurities reduce 7, dusting with Co should increase AAR. Fig.5.10 shows that Co

dusting did not signiﬁcantly change AAR.

130

To look at the effect of dusting more carefully, we prepared a set of symmetric EBSV

samples with the following structures:

A) [FeMn/CoZr(l2)/Cu(20)/COZr(l2)]

B) [FeMn/CoZr(10)/Co(2)/Cu(20)/Co(2)/CoZr(1 0)

C) [FeMn/Co( l 2)/Cu(20)/Co(l 2)]

D) [F eMn/Co(l 0)/CoZr(2)/Cu(20)/CoZr(2)/Co(10)]
We chose the thickness to be 12 nm total for each ferromagnetic layer because our ﬁts are
most sensitive to the value of YCch/Cu at this thickness. Fig. 5.11 shows data for the
above samples, where we made a pair of each structure to check reproducibility. All eight
samples were prepared in a single sputtering run. The numbers in Fig. 5.11 show the
order in which each of the samples was sputtered. First, there is a systematic decrease in
AAR with the order in which each sample was sputtered. That is, comparing pairs of
samples with the same structure (samples 1 and 6 (having structure C), 2 and 5, 3 and 8,
and 4 and 7) shows that the earlier the sample was sputtered, the larger is AAR. The issue
of more relevance here is to compare samples with different structures. Comparing
samples with structure (C) Co/Cu/Co, with samples of structure (D),
Co/CoZr/Cu/CoZr/Co, shows that introducing CoZr slightly reduces the AAR compared
to a ‘bare’ Co/Cu interface. Comparing samples A and B shows the same effect, but less
pronounced. This experiment suggests that CoZr/Cu interface is at least partly
responsible for the reduced MR effect in CoZr-based spin valves, since dusting Co/Cu

interfaces with CoZr thin layers reduces AAR and dusting CoZr/Cu interfaces with Co

increases AAR.

131

Effect of Co dusting on CoZrlCu interface

1* = t + t
otal CoZr Co
0.7 I I Y fl’ﬁﬁ Ifr“ V V ITYﬁ I V I I I V I I T‘ITY T‘I I Y Vi
P <
4

 

 

 

 

 

 

 

t I
. I? I .
, — I s

()6: . I .

A : I -
— I 5

NE 0.5_ I .
I . I 1

G 0.4:— I f
“- I I. 1
CK . .4". I thoMy .
: 1| QLDnbdair ;

<1 0.2— 1| , _
0.1}Il’1 {

i i
o’lrrgiinlilr...1.1.1lrliil.1.11 .1

0 5 10 15 20 25 30 35

tTotal (nm)

Fig.5.10. Co dusting of CoZr/Cu symmetric EBSVs. Sputtering a 1 nm thick layer of Co
at each CoZr/Cu interface does not increase AAR. .

Co_CoZr_C u_Sym_EBSV
X(nm)12ICu, or X(10nm)IY(2nm)ICu

 

 

 

 

 

 

 

0.7_ ,...,.sﬁ,..-,-..,.s., 1
0.6 4 .3
A E 1 CoICu 1 1i 1
V 0.5 _- 1 CoICoZrICu 1 2 '2
' E i: ‘ COZTICU l X E '
0 4 »_ >< CoZrlColCu e _
G . C I I I
u— : :
I! E Co (No dusting): 1,6. 3
g 0-2 .- Co(CoZr dusting): 2,5. -.
E CoZr (N0 dusting): 3,8. j
0-1 .- CoZr(Co dusting): 4,7. -.
0i..h11411...r.1.1.111..14..1:

0 2 4 6 8 10 12 14

trotal(nm)

Fig.5.11. Comparing AAR for dusting CoZr/Cu interfaces by Co, dusting Co/Cu
interfaces by CoZr, and no dusting. There is a systematic reduction in AAR with the order
of preparing the sample (indicated by numbers) in the sputtering nm. Dusting a Co/Cu
interface with CoZr reduces AAR, and dusting a CoZr/Cu interface with Co increases it.

132

5.2.3 GMR in CoZr/Ag:

To see if the unexpectedly small CPP-MR seen in CoZr/Cu multilayers is unique to
Cu, we also tried CoZr/Ag multilayers. We found that AAR is systematically larger for
CoZr/Ag, and the increase seems to be associated mainly with the CoZr/Ag interface.
But there are complications, in that there is a signiﬁcant increase in ARAp in going from
Cu to Ag that is too large to be explained in terms of the difference in resistivities or
interface ARs for Cu and Ag. The difference may be due to problems with the FeMn
layers in the CoZr/Ag. The quality of F eMn seems to change when grown on Ag instead
of Cu. Fig.5.12 compares AR(H) for a Cu sample (A) and and a Ag sample (B), plus the
same Ag sample (C) after reannealing to a higher temperature. We see that FeMn grown
on Cu has better pinning properties than that grown on Ag (i.e. pinning persists to a
higher ﬁeld), and also gives smaller total resistivity (ARAp). The total speciﬁc resistance
in the sample of Fig. 5.123 is larger than that of Fig. 5.12A. This higher ARM: persists in
all Ag-based samples. We take this higher ARM: and different details of AR(H) curve of
Ag based samples as an indication that the FeMn grown on top of Ag has higher
resistivity and poorer pinning properties than FeMn grown on Cu. Fig.5.12C displays
the AR(H) curve for the same sample of Fig. 5.128 reannealed at a higher temperature.
After the second annealing the total speciﬁc resistance rose a bit, and a ‘ﬂatter top’ to the

AR(H) curve suggests a closer approach to antiparallel alignment.

133

16.8

 

1110433 I
15.7 :— CulFeMn/CoZﬂ7yCu(20)lCoZr(7) «j
15.5 ‘ i

3
3
16.5
3
3

AR(mm’)

15.4 _—

 

15.3 :

 

 

K1073-3 _.
Ag, 1st Pin 5

  

b

b

p
I

p-

y-

     
 

22.5 j

22.4 :

AR(mm’)

22.3 :

22.2 ’

 

 

 

vvvvv I V v V l’ v ‘1' T I f r

K1073-3

23-5 Ag, 2nd Pini

23.4 g

23.3 ;

AR(fom’)

23.2 :

23.1 :

 

 

 

23'

Fig.5.12. AR(H) curves for EBSVs with Cu spacer layers (A), and Ag spacer layers (B
and C). The pinning is strongest for the Cu sample A and the total speciﬁc resistance,
ARp, is lowest. Pinning the Ag sample at a higher temperature (C) gave a slightly better
result than at the usual pinning temperature (B). tCoZr = 7 nm in both samples.

Fig.5.13 compares AAR versus CoZr thickness for symmetric EBSVs with structure
[Ag/FeMn/CoZr(t)/Ag(20)/CoZr(t)] with AAR for Cu based EBSVs (from Fig. 5.5). The

Ag-based samples give larger AARs. The higher extrapolated ‘ordinate intercept’ for Ag,

134

combined with similar modest rises with increasing tool: for both Ag and Cu, suggest that
the larger AARs are mainly the result of mob/Ag being larger than CoZr/Cu. The ﬁts
shown use the ‘common ﬁt’ parameters given in table 5 .1, except that for Ag, yew/Ag =
0.8 [28] replaces yew/Cu = 0.52. With this larger yew/Ag, the ﬁt to the CoZr/Ag data is
reasonable, suggesting similar contributions from CoZr whether CoZr is grown on Ag or

on Cu.

The same behavior also occurs in hybrid samples, where slightly higher values of
yew/Ag are needed to better ﬁt the CoZr/Ag data. Fig.5. 14 shows AAR of hybrid samples
with structure [CoZr(t)/Ag(20)/Py(24)/Ag(20)]10/Py(24). Using the same ‘common ﬁt’
parameters but with m/Ag = 0.8 roughly ﬁts the data, within its rather large scatter.

FeMnlCoZr(t)IN(20)/CoZr(t) , N = Cu, Ag
Same parameters for ﬁts except y.

 

 

 

 

 

1 W. .5,

O Data-Cu I

O Data-Ag ‘

0 8 ’ Common Fit, Cu ----------- ‘

A - F ----- Common FILAg ””1." ’ “

 

 

 

11114LllliL

0 L 51‘11101‘141511120 25 30
(nm)

 

 

t
CoZr

Fig. 5.13 AAR of CoZr/Ag based symmetric EBSVs and a comparison with Cu. Ag gives
larger AAR values, especially at small thicknesses, indicating the CoZr/Ag interface gives
a larger contribution to MR than the CoZr/Cu interface.

135

 

[CoZr(t)IAg(20)/Py(24)IAg(20)]1ole(24)

3o ﬁujﬁTfr

: A A 'l
25 :— ‘ .......... ‘- 33‘loi —

. 11 ..... A

t A ..-'

1. o

- 1'. A _,
20 ﬂ. .

l} .1

A _

A

01

I I
1_l

 

AAR(me2)

 

 

 

 

 

10 _— —
5: A Data _
: ----- Common Fit :
o h i r i r L l L r 1 1 r r r l 1 r l r . 1 l i r l 4
0 5 10 15 20 25
t (nm)

CoZr

Fig. 5.14. AAR vs 1ch: in CoZr/Ag/Py hybrid samples. The ﬁt uses the same parameters
as in Fig.5.l3 plus r= 0.8 for the CoZr/Py interface. As in the Cu-based samples the ﬁt is
below data, requiring a slightly larger value of y.

5.2.4 Effect of FeMn Growth on MR

As mentioned above, Fig.5. 12 shows a difference in the CPP-MR “hysteresis” curves
depending on the type of nonmagnetic spacer layer. The Cu-based sample has a lower
total speciﬁc resistance, AR, and a stronger pinning of the CoZr layer in contact with
FeMn. The relatively large difference in ARAp between samples with Cu and Ag suggest
that the change is coming from the FeMn, since the resistivities of the rest of the layers
are too small to account for the difference. Since the pinning is also weaker in the Ag
samples, we suspected that FeMn grown on top of Ag behaves differently from F eMn
grown on Cu. Antiferromagnetism in FeMn is known to depend on the crystal structure

of the FeMn [116], which in turn depends on the growth. F eMn is antiferromagnetic in

136

the FCC (or 7 phase), and nonmagnetic in the BCC structure ([3 phase). The presence of
pinning in our samples indicates the F eMn is antiferromagnetic when grown on Cu and
may be partly so on Ag. We made a systematic set of samples to study the effect on the
CPP-MR of EBSVs of growing FeMn on Cu, Ag, or CoZr. This section is limited to the
symmetric EBSVs and we will show that F eMn is most likely the source of the
differences in Fig. 5.12. The samples used in this search for FeMn growth effects had the
following structures:

A) Cu(20)/FeMn(8)/CoZr(t)/Cu(20)/CoZr(t)

B) Ag(20)/FeMn(8)/CoZr(t)/Ag(20)/CoZr(t)

C) Cu(2)/FeMn(8)/CoZr(t)/Ag(20)/CoZr(t)

D) Cu(20)/CoZr(t)/Cu(20)/CoZr(t)/FeMn(8)

Fig. 5.15 shows AAR versus toozr for all four structures. Sample B has the highest AAR
on average, which is surprising since the GMR-active region in B is identical to that in C.
We do not yet have an explanation for this difference. Except for structure B, the other
three samples give basically the same AAR to within ﬂuctuations.

To test whether the differences in AR Ap between the different structures are due to
differences at the CoZr/N interface, we plot ARM: as a function of tCoZr for the four
samples along with two predictions using the VF model. The differences are even more
visible in Fig. 5.16 where samples of structure B have a higher ARM: than the other three.
The difference between ARA]: of B and C rules out the possibility that the CoZr/N
interface is causing the differences in ARA]: since the only difference in them is the
material on which the FeMn layer is grown. Since there is a small possibility that CoZr

itself may change in structures in A through D, we plot in Fig. 5.17 ARM: as a function of

137

CoZr thickness for both samples A and B, ﬁtting each to a straight line. The slopes of the
two lines are the same, which means that the resistivity of CoZr is the same for both types
of nonmagnetic spacers, Cu and Ag. The value of the slope also agrees with the CoZr
resistivity found earlier ( Ch. 2) using the Nb/CoZr(t)/Nb sandwiches and Van der Pauw
measurements. Dividing the slope of each line in Fig. 5.17, 360 :t 50 nﬂm, by two gives
peoz,= 180 i 25 nﬂm. However, the intercepts on the ARM» axis are different for the two
sets of samples; 13 i l me2 for Cu and 20 i 1 f szfor Ag. From Fig. 5.17, we ruled
out the CoZr/N interfaces as the source of this difference. We, thus, end with only the

FeMn layer as the source of the increased AR when grown on Ag.

Effect of FeMn growth on AAR

 

 

 

 

 

 

 

1 ﬂ T I I ‘ f1 1 1 T {j Y I T I T Y 1' I
A 008 _ ‘ ! J
"E _ + O X X j
i x 0 X
0 6 o y C o . 1
G . 1- . . . —1
- .9 . «
Io- L O i
v ~ 0 O Q X z 0 -
m 0.4 - x . O _
4 : + 0 +
_ .3. o CulFoMn/CoZr/Culc
< 0 2 _ to. o AgIFeMnICoZrIAgIC _
* d X CulFeMnlCoZr/AglCo *
_ + CulCoZrICu/CoZr/FoM
01144411111...111111111111111111111
O 5 10 15 20 25 30 35
t (nm)
CoZr

Fig. 5.15. AAR vs tooz, in symmetric EBSVs. Changing the metal on which FeMn is
grown affects AAR. Growing F eMn on Ag gives the highest AAR.

138

ARM, changes with the growth of FeMn.
Fits shown for Ag, Cu regular SVs. ‘

 

 

 

 

 

 

 

 

30
25 *
A
N
E 20
C: .
u- 15 ,
V
IL 1 Q
< 10 L o Cu-FeMn-CoZr-Cu-CoZr
a: : o AgoFeMn-CoZr-Ag-CoZr
< + Cu-FeMn-CoZr-Ag-CoZr
5 ’ X Cu-CoZr-Cu-CoZr-Fsﬂn
f ----- Flt ( Regular A9)
I ---Fit (Regular Cu)
0 b41114”.1L1h1144mm1..11111..11.4.1hi.1T
O 5 10 15 20 25 30 35 40
t (nm)
CoZr

Fig. 5.16. Effect of FeMn growth on ARM». Growing FeMn on Ag gives a signiﬁcantly higher
ARAp than when growing it on Cu or CoZr. The V-F model predicts only a slight increase in
ARA}. for the Ag samples over the Cu samples. The ﬁts for Ag and Cu are on top of each other.

 

Effect of FeMn growth on ARAP

30 ....,.-..,ﬁ.,.,..-.,,..., ,.,,,
‘ 1

 

AR (me2)

 

 

 

 

 

. :

<10; 1

5 _ —ARAP(t)Cu= 13+ 0.36t 3

; ----- mama)”: 20 + 0.36 t 3

oh....11.4111111111111.1.114...11411l

O 5 10 15 20 25 30 35
tCoZr(nm)

Fig.5.l7. ARM: vs tow. As predicted by the 2CSR model, ARA]: grows linearly
with teen. The slope is the same for Ag and Cu samples, showing that CoZr has
the same resistivity in both. But the intercepts are different, pointing to the FeMn
as the source of the larger ARM» for Ag.

139

5.2.5 Search for Magnetically Dead Layers

The discussion above shows that the CoZr/Cu and CoZr/Ag interfaces are potential
sources of complications in the physics of the multilayers. The existence of a
magnetically dead layer CoZr layer at these interfaces might inﬂuence the CPP-MR. This
possibility prompted us to search for dead layers at CoZr/Cu and CoZr/Ag interfaces.

To check for dead layers we prepared multilayered ﬁlms of structure:
[CoZr(t/N)/X(5)]N , where N is the usual number of repeats in the multilayers.

The total magnetic moment, 11,, of the multilayer is given by: Lt! = ugh“, x t, where [15mg
is the magnetic moment per unit thickness of CoZr, and t is the total thickness of the
magnetic CoZr material. ugh”, is a constant having no effect on the analysis. For ﬁxed
total thickness of CoZr, t1, number of layers, N, each with thickness t, and a magnetically
dead layer of thickness to at each interface, the total magnetic moment of the multilayer
should be:

Ht=PShec1Xtt=lJShectX(Nt-2N10) (5-1)

Plotting ut/N versus I should then be a straight line with ordinate intercept at t = 2 to.

Fig.5.18 shows the magnetization curves of 6 multilayers with ﬁxed tam = 48 nm and
variable N. The inset shows only a very slight reduction, if any, in the total saturation
magnetic moment of the multilayers as N is increased up to 48. This near constancy
indicates that any dead layer at the CoZr/Cu interface must be very thin. Fig.5.19 shows
the total magnetic moment divided by the number of repeats, ut/N, versus 10,2," Using
Eq. 5.1 gives 2to = l.O6/(O.6l) z (1.8 i 3.6) A0 :> to z (0.9 i2)A°. The best ﬁt is only
half a dead layer. Within our errors, we cannot rule out either no dead layer at all, or a

single dead monolayer. In the same way we determined from Fig. 5.20, that at CoZr/Ag

140

interfaces, any dead layer would be: 210 =1/5.5 a (0.2105) A° z» to z (0.1 i- 0.3)A°.

Here, we can rule out a dead layer at the CoZr/Ag interfaces.

 

   

 

    

 

 

 

 

 

 

CoZr/Cu deadlayer
0 0032 [CoZr(th)ICu(5)]j, N indicated in Caption
0.0024: "“"""‘
’5 0.0016 :—
E 0.0008 —
:7 o g .
5 r :' ‘ 11:31-31:31,
. 0 - a x .
E-0-0003 _ --o--k1131-2b(x3) 1
. . +k1131-3 48 .
£00016 “ .' -+—k1131-4:::2)) ;
' : : - €+~k1131-6a(x2) -
-0.0024 ‘ s _
4100321.. .. ....1.........1...a
-3000 -2000 -1000 0 1000 2000 3000
H (De)

Fig.5.18. Hysteresis loops of [CoZr(t)/Cu]~ multilayers with ﬁxed total CoZr thickness,
tT = 480 A0. The saturation magnetization is almost the same for all samples, regardless
of the individual CoZr layer thicknesses. Any magnetically dead CoZr layer at the
CoZr/Cu interface must be thin.

14]

CoZr/Cu dead layer

 

 

 

 

 

 

 

 

 

0.0015 e... ,--.

3 0.001:—

E t

3 .

Z

10.0005
oi-.....-..,.i. °° ”W" ‘° ”'1
0 50 100 150 200 250

((71)

Fig. 5.19. p/N vs teal, thickness for CoZr/Cu multilayers. The ordinate intercept gives
the thickness of any dead layer.

 

 

 

   

 

 

 

 

 

 

CoZr/Ag deadlayer
0.0014“.-. .,f.
; —plN' - -1.0e-06 + 5. 54e-06t
0.0012 5 _
0.001 1
E? I :
E 0.0000 . -
3 i 3
2 0.0006 i J
\ ’ -4
1 1
0.0004 5 .
C 1
0.0002 _~ m- _
0: °° " mil" "° 1
0 50 100 150 200 250

t (A)
Fig. 5.20. W versus CoZr layer thickness for CoZr/Ag multilayers. The ordinate
intercept indicates little or no magnetically dead layer at the CoZr/Ag interface.

142

5.2.6 C095er and C030Zl’20

In this section we study samples with higher Zr content. The two concentrations of Zr
we chose were 5% (still polycrystalline), to see how the CPP-MR behaves with
increasing Zr, and 20% (expected to be amorphous), which is a magnetically ultrasoﬁ
structure [111,114] with potential for applications if it has a decent CPP-MR. Fig. 5.21
shows AR(H) curves for two Coloaerx(5)/Cu(20)/Py(6) hybrid samples. F ig.5.21A is for
x = 2.5,and B is for x = 5 sample, There are three important features of Fig. 5.21. The ﬁrst
is that the switching ﬁelds for CoZr are the same and not noticeably small for both Zr
contents, indicating that Co(5% Zr) is still polycrystalline (since amorphous CoZr is
magnetically ultrasoft [109 ,114 ,115]). 5% Zr is probably just below the critical value at
which CoZr becomes amorphous. The second feature is that AAR is smaller for the
C095er sample than the C097,5Zr2_5 one. The third feature is that the total ARp of the
sample with x = 5 is larger, as expected for higher Zr content that increases the resistivity.
Combining the lower AAR and the higher ARp makes it clear that increasing the Zr
content from 2.5% to 5% reduces the MR signiﬁcantly, instead of enhancing it. To make
sure that the reduction in MR in Fig. 5.21 is not a coincidence we present more data and
analysis for samples of C095Zr5. To have two different sets of results we investigate two
different types of samples. Hybrid multilayers [CoZr(t)/Cu(2O)/Py(6)/Cu(20)]10’Py(6),
and asymmetric EBSVs [FeMn/CoZr(t)/Cu(20)/C0Zr(t)]. Results for the hybrid samples
are displayed in Fig. 5.22 along with different V-F model ﬁts. Fig. 5.213 shows that a
decent AP state can be established for the 5 nm thick CoZr, yet Fig. 5.22 shows that AAR
of the 5 nm sample has decreased from that of the 3 nm thick CoZr. Production of a good

AP state means that the less rapid growth of AAR is not due to the lack of AP alignment,

143

but to something else—probably a very small ,6. To ﬁt the data requires ,6 z 0.03 (short-

dashed curve).

52
50

AR(fQ m2)

62

AR(fQ m2)

48”

[Py(6)lCu(20)/CoZr(5)/C

u(20)]x10 I Py

(6)

 

42_
40“

'T'YT

Y I V V I I r

k1007-2

E Co Zr
48_

46:

97.5 2.5

1

   

 

r I

I T ﬂ
.1

(A)

     

 

 

so:
50;
56:

52'
50’

 

 

w969-2b

0095er

l #1

 

l J 1 l

 

 

400

-200

0
H (De)

200

 

Fig. 5.21. AR(H) versus H, for [Conrl00.x(5)/Cu(20)/Py(6)/Cu(20)]m/Py(6). x is 97.5 in
A and 95 in B. Vertical and horizontal scales are the same for direct visual comparison.

144

CoZr(5%)IPy Hybrid SV

 

 

 

 

 

 

35 l. 1 1 r l 1 l 1 r 1 r T ﬁr ﬁ' l r r I 1
0 om
: — — llF_b-0.46_g-0.77_1‘-0.51 _
30 f ---- llF_b=0.03_g=0.77_1‘=0.51 —_
h2_b-0.03_g-0.77_r'-0.51 .
A : ----- l=2_b=0.5_g=0.77_r‘=0.51 __ __ .. — - - 1
: ”<1: -------------------------
C} 20 r _ ,- ’ j
u- ; _ y
0: 1 5 L l
" o
"-
<1 ' «5.. _
10 :‘O m.._-..-‘.- .
< _ um...___...~--.--.- —
5 r ‘
o h 1 1 1 L l 4 1 1 1 l 1 1 4 1 1 1 1 1 1
0 5 10 15 20
t (nm)
CoZr

Fig. 5.22. AAR vs tCoZr- Different ﬁts are shown: 13]: = inﬁnity or 2 nm, ,6 = 0.5 or 0.03,
y= 0.77, and ARE°Z,,CH= 0.51.

FeMnlCoZr(t)ICu(20)IPy(6) EBSV

 

 

 

 

 

 

1r' l 1 1 11"YI**‘rI"*T
)- . 1
F o
A 008’ ‘1
N
G 006'— (3 _
h _ .
V L (JO Q l
m Ol4b _
. C .
< o 00“er 1
F O
< 0.2 - cosuzru J
1 , 1
p. 1
ov.1..111..11mm111111111D111111111D1‘

 

0 5 1O 15 20 25 30 35
tCoZr(nm)
Fig. 5.23. AAR vs tool, for CoZr(t)/Cu(20)/Py(6) EBSVs.

145

Having tested the CPP magnetoresistance of polycrystalline CoZr alloy with two
different concentrations, we look for a CPP-MR in CosoZrzo. A high resistivity (>100 1.1.0
cm), which decreases with increasing temperature, suggests an amorphous state. Fig. 5.24
shows that AR(H) curve displays little or no variation with H. This result is not

completely surprising, since an amorphous metal should have a very short spin-diffusion-

 

 

length.
32 ﬁf“l""TT‘l'l""l""l""lr"'
FeMnICoZr(6)ICu(20)ICoZr(6)

A 31.5 L k1175-5 L
N L /- Amorphous CoZr 4
E

Cl

l.—

0:

<

 

 

 

30.1111 1111 11.11.14...
-200 -150 -100 -50 0 50 100 150
H(Oe)

Fig. 5.24. AR vs H for a CoZr/Cu symmetric EBSV. The amorphous CoZr does not give
a noticeable AAR within the ﬂuctuations of the data.

146

5.3. INTERNAL INTERFACES.

F/N interfaces typically have speciﬁc resistances (2 A R'Fm ~ 1 me2 [5]) much larger

than those of a few monolayers of the F-metal. In such a case, it is, in principle, possible
to insert additional interfaces into the F -layers of an F/N multilayer and signiﬁcantly
increase both the total speciﬁc resistance AR and the change in speciﬁc resistance AAR,
without greatly increasing the total multilayer thickness [53,117]. Inserting
‘n -1’ thin N-layers into the F-layers divides the latter into ‘n’ thinner layers, producing
M = (Zn-2) ‘internal interfaces’. One would like the N-layers to be thick enough to
completely separate the interfaces on both sides, leaving some unmixed N-layer in
between; we’ll call these ‘complete interfaces’. For sputtered multilayers, unavoidable
intermixing then will require the N-layers to be at least 2-4 monolayers (ML) thick, equal
to 0.4-0.8 nm [118]. But, to maximize AAR, the ‘F-layers’ should behave as single
magnetic entities with switching ﬁelds similar to those of the unmodiﬁed F-layers. To
achieve this condition, the N-layers must be thin enough to give ferromagnetic coupling,
either by the RKKY interaction or by means of pinholes. They must, thus, be thinner
than the IN z 0.9-1.0 nm for the ﬁrst antiferromagnetic coupling peak [11,12]. It is not
clear in advance that both conditions can be satisﬁed at once for a given F/N multilayer.
For the CIP-MR, the preferred geometry for devices is the EBSV, because its
hysteresis curve allows local control of the magnetization [53] and its magnetic structure
is stable over many magnetization reversals [54]. For the CPP-MR, the situation is more
complex, because a standard EBSV has only two active F/N interfaces, and the AR of the
pinning AF layer is in series with the multilayer. A hybrid spin valve may often be

competitive, since it can contain more repeats, thereby enhancing both AR and AAR. In

147

the present study, we chose a hybrid spin valve with 3 repeats (N = 3) to enhance the

signal. We chose Co for the F-layer and Cu (later also Ag) for the inserted N-layer
because 1%,? ~ 60 nm [5] is long enough to be neglected and because our knowledge of all
of the Co/Cu parameters [5,43] lets us predict without adjustment the values of ARp,
AAR, and the CPP-MR with which to compare our data. We pair the Co layers with Py
layers, for which we also know all of the parameters in advance [74], and which have a
much smaller saturation ﬁeld, thus ensuring a good AP state for the data analysis.

Fig. 5.25 shows the 2CSR model predicted enhancements of ARp, AAR, and CPP-MR
= AR/R as a function of M for a multilayer with structure
[Py(6nm)/Cu(4.5)/[Co(t/n)/Cu(0.5)]n/Cu(4.5)]3. We see that all three quantities increase

with increasing M, and that adding only 4 internal interfaces should increase AAR by
100%, AR(P) by 20% and AR/Rp by more than 80%, with less than 5% increase in total

thickness. The predictions are for continuous Co, Cu layers with well-deﬁned interfaces.

They assume the parameters given in section 5.2-- ARE),Cu = 05ij2 , p230 = 7.5110cm ,

,BCO =0.46, and yea/Cu = 0.77, p;}, = 24,1190m nay/Cu = 0.7, etc. We chose the Py thickness

of 6nm to roughly equal 1;]! [74], since making it thicker would increase the thickness of

the device and ARp more than it would increase AAR (see Eq. 3.12). We introduce
interfaces only into the Co, both to simplify the analysis by eliminating effects of ﬁnite

[SF in Py and because the effect of internal interfaces is less pronounced in Py due to the

larger values of ﬁg», and [7%, compared to 2 A RPy/Cu [5]. Note that the hybrid spin valve

contains two different thicknesses of non-magnetic spacer (Cu or Ag). For Cu, 4 nm

thick layers are thick enough to separate Py magnetically from Co, and our initial studies

148

use 0.5 nm thick Cu to produce the ‘internal interfaces’. For Ag, we found that 4 nm was
not enough to decouple the Co and Py, so we use 10 nm of Ag instead.

We begin our discussion with the Cu-based samples, which we have studied in most
detail. We then present less complete data on Ag-based samples. In both cases, we
initially ﬁnd substantial enhancements of ARp and AAR with increasing M. But, then the
growth with increasing M becomes less than predicted. These results lead us to examine
two possible explanations: ‘incompletely formed interfaces’ and ‘spin-memory-loss at
interfaces’. We conclude that both probably contribute, with the case for spin-memory-

loss at interfaces the strongest.

149

10

 

  

 

 

 

 

 

 

 

 

 

 

I
)-
.3:
«E _
c: ‘—
1.— _
V i-
m 4‘
4 — .
< 2; S
h- -1
,- MI Al 11 14*
ﬁ jﬁj‘ITVﬁIWﬁi
20
A
NE I
15 .1
G -1
Io- : I
v L .
Q10— —
m - .
< s: (B) ;
~ 1
.- 4 I 41 L1 11 g
'- [ ITWV'T I q
40'- L
o I
.\ F .
Q30— q
K - 1
\ ; I
120 -1
<1 : (C) :
1o; _‘
E .
1.
L1 1 1 1|: 1
0 2 4 6 8 10

Fig.5.25 Predicted change in AAR (A) and ARp (B) and AR/Rp (C) with number of
inserted internal Co/Cu interfaces.

150

5.3.1 Data and Analysis
Resistances and Magnetoresistances of Samples with to. = 0.5 nm

Figs. 5.26A and B compare data on AR(H) vs H for a sample with 4 internal
interfaces and a similar sample with no internal interfaces. In both cases, the data round
off at their peaks, seeming to indicate a decent AP state. We see that adding the internal
interfaces increases the saturation speciﬁc resistance ARp by ~ 10%, AAR by ~ 100%,
and only modestly increases the switching ﬁeld of the Co layers.

Fig.5.27 compares our experimental values of ARp for Cu based multilayers as a
function of M with predictions from the 2CSR model. As predicted, ARp increases
almost linearly with M. But the data lie a bit below the prediction and, more importantly,
grow more slowly with M than predicted (see dashed line).

Figure 5.28 compares AAR as a function M with the 2CSR model predictions. There
is an initially large enhancement of AAR, as much as 100% for M = 4. Thereafter the data

grow more slowly with M.

151

k1126-8
[Py6lCu4/{Co(1)ICu(0.5)}3/Cu(3.5)]3
19..e1...,...,ﬁ...

 

‘I TIerﬁlTj'l

17;
163— A

15

AR(mez)

 

 

[YTrYIIIY

14

 

 

 

1 ”"1111‘25-0m
16; [Py(6)lCu(10)lCo(3)lCu(10)]3

I11

 

 

 

 

A . I
N 15— 2
Cl 14; L
V P f b
0: 13— f 1‘ _
< r ,l l '\ l
12— -. ,1 / \ _
1”11.1..1111211111.11111L11111114
-800 -400 0 400 800
H(Oe)

Fig. 5.26. AR(H) curves for samples (A) with 4 internal interfaces and (B) without. To
allow direct comparison, both the vertical and horizontal scales are the same for both
samples. Introducing the interfaces increases AR, AAR, and the Hg for the Co.

152

N
0|

 

I 1 j I ﬂ I I I I I r I I I I r I I j T I

 

 
 

0 Data 4
2CSR Model

  

 

N
O

 

 

 

 

 

NA
E 15 .
E g 9 :
Z 10 __ J
m _
< 5 _ [Py(6)ICu(4)I[Co(6In)ICu(0.5)]nlCu(3.5)]3
o 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 ﬂ
0 2 4 6 8 10 12

Fig.5.27 Increase of ARp with M in [Py(6)/Cu(5)/(Co(t/n)/Cu(0.5))../Cu(4.5)]3 multilayers
versus prediction (solid curve).

153

.3
N
1

ﬂ

 

 

 

 

 

 

 

 

 

E 0 Data 1
10 — ZCSR Model
A " 1
N _ :
E 8 _ i
C} _ o :
~1— 6 ' 4
m i i
< 4 . ‘i
< 2 . [Py(6)lCu(4)/[Co(6ln)ICu(0.5)]nlCu(3.5)]3

‘. 3
o r . 1 1 1m 1 1 L 1 1 . 1 1 1 1 1 1 1 1 1 1 1 1 l

0 2 4 6 8 10 12

Fig. 5.28. AAR vs M. compared with 2CSR prediction. Data shows a signiﬁcant increase
in AAR at M = 4, then AAR levels off faster than predicted.

154

 

 

 

 

 

 

 

A : I I :
N 5 L I .2
E o . ,
c: 4 :— . . 1
u- : t tsu= 0.6 nm ‘
"" 3 L I t =o.5 nm L
O: i O t:=0.4 nm 1
< 2 L " 1
< 1 E [Py(6)lCu(4)/[Co(3ln)ICu(tCu)]nICu(3.5)]3 S
: t = 0.6, 0.5, 0.4 nm 3
0’11111111Eu 1 1111;141111‘

 

5 6 7 8 9 1o 11
M

Fig.5.29. Effect of the thin Cu spacer layer thickness on AAR. When Cu layer thickness
is 0.4 nm (ﬁlled boxes) AAR is smaller than when Cu thickness is 5 nm (ﬁlled circles).
The one sample with tc..= 0.6 nm (ﬁlled triangle) is slightly higher than the 0.5 nm.

Lastly, Fig. 5.29 compares values of AAR for Cu thicknesses ten = 0.4, 0.5, and 0.6
nm (only one point) for M = 6, 8, and 10. In all three cases, AAR is smaller for to, = 0.4
nm than for 0.5 nm In the only case studied, AAR is slightly larger for (Cu = 0.6 run than
for ten = 0.5 nm, but probably within ﬂuctuations.

Taken together, these data show that re“ = 0.5 nm is certainly the minimum thickness
needed to produce two complete Co/Cu interfaces, and suggest that it might not even be
quite sufﬁcient to do so. The slower than expected growth of AAR with M in Fig. 5.28
might be due in part to such incomplete interfaces. However, another possibility is spin-
memory-loss at Co/Cu interfaces, as inferred from the data and analysis in Ch. 4.

Inserting internal interfaces leads to a sample structure similar to that of ‘separated’

155

samples in Ch. 4. To further examine these two possibilities we prepared additional

samples, which we describe in the next section.
5.3.2 Incomplete Interfaces versus Interfacial Spin Memory Loss

To see whether (and if so how) the data of Figs. 5.27-5.28 are affected by incomplete
interfaces or interfacial spin—memory-loss, we made a series of Py and Co based hybrid
spin valves, all of which have total thicknesses of Py and Co equal to 18 nm, all but one
of which have ﬁxed individual ferromagnetic layers with tea = tpy = 2 nm and a total of 17
Co/Cu and 17 Py/Cu interfaces, but each of which has a different order of the F-layers
and/or different Cu thicknesses. Two of the samples are ‘interleaved’ in the sense
deﬁned in Ch. 4, and the other four are composed of combinations of ‘interleaved’ and
‘separated’ portions that we designate ‘mixed’.

In Fig. 5.30 we begin with three samples with to“ large enough (20 nm) so that all of
the interfaces should be ‘complete’. Sample C is an interleaved sample of the form
[Py(6)/Cu(20)/Co(6)/Cu(20)]3 with 5 interfaces each of Co/Cu and Py/Cu. We use it as a
reference sample, equivalent to an M = 0 sample from Figs. 5.27 and 5.28, except that it
has (Cu = 20 nm instead of to“ = 4. Because of the low resistivity of Cu, this difference
should be minor. Interestingly, ARp and AAR for this sample fall nicely on the predicted
(solid) curves in Figs. 5.27 and 5.28. Sample A is also a simple interleaved sample, but
of the form [Py(2)/Cu(20)/Co(2)/Cu(20)]9, containing 17 interfaces each of Co/Cu and
Py/Cu. It shows the usual behaviors from Ch. 4 for an interleaved sample, a single
peaked CPP-MR, with an ‘as-prepared’ value of AR just a bit smaller than the ‘peak’
value. Sample B is a mixed sample of the form [{Py(2)/Cu(20)}3/{Co(2)/Cu(20)}3]3.

As usual for ‘separated samples’, its AR is larger in the ‘as prepared’ state than at the

156

subsequent ‘peak’ after the sample has been taken to saturation, and AAR between the ‘as
prepared’ and parallel (P) states is close to that for the ‘peak’ and P states of the
interleaved sample A. However, unlike the more complex structure seen in simple
‘separated samples’ in Ch. 4, this mixed sample has only a single MR-peak on each side
of H = O. This difference is a consequence of the ‘mixed’ nature of the sample.

With the data for these samples in hand, we now turn to samples having the same
order as sample B in Fig. 5.30, but some Cu layer thicknesses are reduced to (Cu = 0.5 nm
to test effects of possible incomplete interfaces. We are especially interested in how the
values of ARp, AAR, and stitch for these samples compare with those for sample B.

The structure of sample A in Fig. 5.31 is identical to that of sample B in Fig. 5.30,
except that the thickness of the Cu layers between the Co layers is reduced to ten = 0.5
nm. The structure of sample B is again the same as B in 5.30, except that now the
thickness of the Cu layers between the Py layers is reduced to to, = 0.5. In sample C, the
Cu layer thicknesses between neighboring Py layers and neighboring Co layers are all
reduced to (Cu = 0.5 nm, and the Cu layer thicknesses between the Py and Co layers are
ﬁxed at 5 nm to approximate those in the ‘internal interface’ samples in Figs. 5.27 and
5.28.

Comparing the data of the three samples in Fig. 5.31 with those of samples A and B
in Fig. 5.30 we see the following.

(1) The similarities in behaviors of the data for samples A and B in Fig. 5.31 with those
for interleaved and separated samples in Ch. 4, strongly suggest that the difference

between the AAR z 6 fﬂm2 between the peak and saturation states for sample B and the

157

AAR z 8 me2 for the same states in Sample A is due to the source ascribed in Ch. 4,
which we proposed to be interfacial spin-memory-loss.

(2) The values of AAR z 6 ﬁlm2 for the peak to saturation values of all three samples in
Fig. 5.31 and of sample B in Fig. 5.30 are all closely the same. This agreement means
that this AAR is insensitive to the thickness of the Cu layers, and strongly suggests that

the main limitation on AAR in Figs. 5.28 is due to interfacial spin-memory-loss.

(3) The variations in switching ﬁelds of Co between sample B in Fig. 5.30 and the three
samples in Fig. 5.31 correlate with the ‘effective thickness’ of the Co layer. That is, high
switching ﬁelds are required for thin, widely separated Co layers, but much lower
switching ﬁelds are required for thin, closely coupled Co layers.

(4) Considering only sample B from Fig. 5.30 and the three samples from Fig. 5.31, ARp
decreases as the separation between adjacent Co and adjacent Fy layers is reduced. That
is, ARp is largest when ten = 20 nm between both Py and Co layers and smallest when to,
= 0.5 nm between both. However, as illustrated by the unpredicted difference between
the values of ARp for samples A and B in Fig. 5.30, this interpretation can only be
preliminary, because ﬂuctuations can bias analysis of only a few samples.

(5) Comparing the initial states of sample B in Fig. 5.30 and those of the three samples in
Fig. 5.31, we see that the AR for the initial state of B in Fig. 5.30 is higher than that for its
peak state, the ARs for the initial states of the two samples in Fig. 5.31 with only one set
of thin Cu layers are about the same as for their peak states, and the AR for the sample
with both sets of Cu layers being thin is less than that of its peak state. We presume that
the magnetizations of widely separated (uncoupled) adjacent Co or Py layers tend to align

anti-parallel to each other in the as-prepared state (as for the ‘separated’ samples in Ch 4),

158

whereas the ferromagnetic coupling between adjacent Co or Py layers separated by only

ICU = 0.5 nm causes them to tend to align parallel, thereby decreasing the value of AR in

the as-prepared state.

Table 5.2 The Composition of the samples used to study the effect of incomplete
interfaces and separated-like effect (due to the internal interfaces).

Effect Of lntemal Interfaces: Thickness and

Notes

Rinmal <Rstable

Rinitial<Rstab|e

Rinitial >Rstable

Rinitiak-Rstable

Rinitial~Rstable

Rlnitial<Rstablo

 

159

[Py(2)lCu(20)ICo(2)/Cu(20)].

 

    
   

 

 

 

   

 

       

 

 

32» u
311 A _
A
N . .
.5
V
m 28.
<
26
11111111111...1.114411111L111l
v"rv'vt'rﬁrvvvrv'vr'avrvfrlﬁ
[{PylzllCu(20)},l{00(2)lCu(20)},l, -
26~ —
"A . B
E 24
g 1
E: .
22
20
‘- 1111111111111111111 1.1 111"
22 _ [was)Icumoycmsycum”3 _
1
A 201 .
N . .
E C
Q
a:
<
1111 11114411111111.11141111

 

 

 

 

.306 .3406 o 400 300
H (De)

Fig.5.30. AR(H) curves for (A) Interleaved, (B) Separated and (C) Regular samples.
Regular means samples without any internal interfaces. The reduction in AAR due to the
difference in ordering of the layers between A and B was already seen in Ch. 4. Note that
samples A and B also have larger ARs than sample C.

160

 

Y Y Iﬁ ﬂ I rfT r7 1 Y I V V W I V V 7 r? V I I 'j V

: k1242-5 q
, [{PY(ZVCUl20)},’{C°(2)’Cu(0-5)},’CU(1 95)], ,

241 _
.1" i 4
522~ _
r1 : (A) j
(201 -
1

18; «

IAAlAALlAAAlAAAlALAlAAAlAAA114+
Ivrlivvlvvvlv

 

 

L k1242-7 ‘
24 _ [{py(2)ICu(o.5)},ICu(19-51I{Co(2)ICu(20)).lal

 

 

A
N .
5 > +
v L (B) _‘
m 20 l
<

18— —
l .1
l
16 1 1L1L1111111111L111111141 1 I
11,11.17..,1r1r+11,1.1,14rr471

F
. “242-8 .
. [(PvlleIIlO-Ull,lCIll4-WICOIZVC0l0-5ll,’cul4-5ll, .
20~ -

AR(fom’)

 

 

 

 

lAlAlll‘AA'AAJJALIIA‘AIAAAILAL

406 400 0 400 800

 

H(Oe)

Fig. 5.31. AR vs H for samples identical to the sample in Fig. 5.30 except for the Cu
thicknesses. In A the Cu layer between neighboring Co layers is 0.5nm and the rest are
20nm. In B the Cu layer between neighboring Py layers is 0.5 nm and the rest are 20 nm.
In C the Cu layer between neighboring Co or neighboring Py layers is 0.5 nm and the Cu
layer between Co and Py is 5 nm.

161

Ag Samples:

We now turn to less extensive studies with Ag as the nonmagnetic spacer layer. Since
Ag has a larger lattice parameter mismatch with Co [28], the Co/Ag interfaces may be
rougher than Co/Cu. Since Ag and Co are completely immiscible [28], there may also be
less chemical mixing at their interfaces.

Fig.5.32 shows R(H) for a [Py(6)/Ag(4)/[Co(0.6)/Ag(0.5)]6/Ag(3.5)]3 multilayer. The
sharp peaks in the data mean that a good AP state was not achieved in this multilayer. 4
nm of Ag between Co and Py does not seem to be enough to decouple the ferromagnetic
layers from each other. Perhaps larger interfacial roughness enhances ferromagnetic
coupling between the Co and Py layers across the 4 nm Ag layer. Figs. 5.33 A and B
show that we achieved at least better approximations to AP states by using 10 nm of Ag
to separate the Py and Co layers and 0.8 nm of Ag to separate neighboring Co layers.
Comparing 5.33A and B shows that introducing 8 interfaces increased AAR by ~ 40%.
Note that the magnetic ﬁeld range is smaller for Ag than it was for Cu. Since the AP
states are somewhat uncertain for Ag, our curves should probably be viewed as providing
only a lower limit on those states. The use of a slightly thicker (4 nm) Co layer in the Ag
samples does not produce a signiﬁcant change in the MR or switching ﬁeld compared to
the 3 nm in the Cu samples.

F ig.5.34 shows ARp vs M for samples of the form
[Py(6)/Ag(1O)/[Co(4/n)/Ag(0.8)]n/Ag(9.2)]3. Here the 2CSR model ﬁts the data better
than with the Cu samples. This better agreement might be due to more nearly complete
interfaces, both due to less intermixing of Co and Ag and to the thicker ‘thin’ Ag layers

(0.8 nm vs 0.5 nm for Cu).

162

Fig.5.35 shows that introducing a pair of Co/Ag internal interfaces increases AAR by
~ 30%, but, as for Co, adding still further ones gives less increase than predicted.
Although we haven’t performed additional experiments to check for sure, we presume

that interfacial spin-ﬂipping (spin-memory-loss) is affecting the Ag data as we argued it

 

 

 

 

 

did for the Cu data.
R1 1 18-3
[Py(6)lAg(4)/{Co(0.6)IAg(0.5)}6IAg(3.5)]3
14.6 f—
1‘
A 14.2 :—
C} _
c 1_
V 1
m 13.8 :—
-400 -200 0 200 400

H (De)

Fig.5.32. R(H) for an Ag based multilayer. Using a 4 nm thick Ag spacer layer is not
enough to decouple Py and Co layers and an AP state is not achieved in this sample,
giving sharp instead of rounded or ﬂat peaks.

163

k1127-5
[Py(6)IIAg(10)I{Co(1)IAg(0.8)}4IAg(9.2)]3

IIYIfIITYTVTTII'I I I r

22E fl _-
: l 3

p
L.

 

"'1

N
O

—

AR(mm’)
a;

.3
a:

 

 

.3 N
0 A

AR(fom’)
:1‘

 

 

 

40‘0” ”L‘zo‘o‘” ””0““ ”3013‘” “too
H(Oe)

 

Fig.5.33. Effect of internal interfaces on AR(H) curves for samples with Ag spacer layers.
Introducing eight internal interfaces( part A) increases AAR and ARp.

164

 

 

 

10 — I Data -
. — ZCSR Model

 

 

[Py(6)IAg(10)I[Co(4ln)lAg(0.8)]nlAg(9.2)]3

 

 

 

0111111L111AL1111

o 2 4 6 81110

Fig.5.34. Dependence of ARp on the total number of internal interfaces M. solid boxes
are for data and solid line is the 2CSR model prediction. The data agree with prediction
within the ﬂuctuations.

 

   
  

 

 

 

 

 

 

12 1 1 1 1 1 1 1 1 1 1 1 1 1 v 1 1 1 1 f
C .
1 I Data
10 j —2CSR Model ,
A p 1
N - a
E a _ .
L 1
Cl 1 .
u- 6 — s
v r _
m 1.
<1 4' -
<
2 ~ [Py(6)IAg(10)I[Co(4ln)IAg(O.8)]nlAg(9.2)]3 1
o h 1 1 1 1 1 A 1 1 1 1 1 1 1 1 1 1 1 1 A l
0 2 4 6 8 10
M

Fig.5.35. AAR dependence on the number of internal interfaces in Co/Ag based
multilayer.

5.4 SPIN-MEMORY-LOSS TO ENHANCE GMR

In this section we try to introduce strong spin-memory-loss by inserting a 1 nm F eMn
layer in each of the two Cu layers separating Co from Nb in multilayers of structure
Nb(1OO)/Cu(20)/[Co(l)/Cu(2O)/Co(6)/CU(20)]N/Nb(l00). We wish to see if this strong
spin ﬂipping increases AAR by eliminating the contact resistance ZARs/p from the
denominator of Eq. 1.3.

Our group [85] showed earlier that, introducing a thin (mm = 1 nm) layer of FeMn
that produces strong spin ﬂipping [73] into the free Py layer of a Py-based EBSV
enhanced AAR monotonically as the spin-ﬂipping layer was moved towards the outside of
the Py layer. Our procedure here differs, in that the FeMn layer is placed outside of the
multilayer in a Cu layer that otherwise separates the outer F -layer from the

superconducting Nb leads.

If we simply assume that the FeMn eliminates the MRS”: term in the denominator of

Eq. 1.2 we obtain:

(1017-1111161 ‘l‘ 2ARFl/N71)(pl:2tF2ﬂ2 + ZAszmrz) (5.1)
21«112,1,,M,,,Cu + 2NpNtN + Npggn + 2NAR§W + prntn + 2NAR;2,N

AAR = 4N 2
Another possible result of introducing the FeMn between Nb and the multilayer might be
to make the outermost Co/Cu interface contribute to AAR. In the absence of the FeMn,
the Cu layer next to Nb becomes superconducting by proximity with the superconducting
Nb, perhaps making the Co/Cu(superconducting) interface not MR active. If the
resistance of the outer Co/Cu interface becomes active, it would further enhance AAR,

especially for a small number of layers, the case of most interest for devices. More

details of F/S interfaces will be examined in Chapter 6.

166

5.4.1 Data and Analysis

Fig. 5.36 shows the structure of the multilayers used in this study. We chose
[Co(6nm)/Cu(2O)/Co(l)/Cu(20)l1v multilayers for three reasons: First, we have the results
of studies of such multilayers without FeMn inserts in Ch. 4. Second, the Co/Cu
interface makes a stronger contribution to GMR than bulk Co layers of these thicknesses.

Third, Co has a fairly long spin-diffusion length [5], thereby leaving scope for the FeMn

to have a signiﬁcant effect.

Superconducting
By proximity

Fe

eeeeee
oooooooooooooooooooo
ooooooooooooooooooooooooooooooooooooooooo
eeeeeeeeeeeeeeeeeeee
oooooooooooooooooooooooooooooooooooooooooo
oooooooooooooooooooo
oooooooooooooooooooooooooooooooooooooooooo
oooooooooooooooooooo
llllllllllllllllllllllllllllllllllllllllll

    

 

IIIIIIIIIIIIIIIIIIII
eeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeee
oooooooooooooooooooo
oooooooooooooooooooooooooooooooooooooooooo
QQQQQQQQQQQQQQQQQQQQ
eeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeee
cccccccccccccccccccc
eeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeee
oooooooooooooooooooo
eeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeee
oooooooooooooooooooo

eeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeee
oooooooooooooooooooo
oooooooooooooooooooooooooooooooooooooooooo
oooooooooooooooooooo
IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII
oooooooooooooooooooo
ooooooooooooooooooooooooooooooooooooooooo
oooooooooooooooooo
eeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeee
oooooooooooooooo
ooooooooooooooooooooooooooooooooo
ooooooooooooooooo
00000000000000000000000000000000000

 

eeeeeeeeeeeeee
IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII

Fig. 5.36. Cartoon of the multilayer (with N = 1) with FeMn layers to enhance GMR. A
single FeMn layer is inserted in the middle of the Cu layer next to Nb on each side of the
multilayer. The Cu layers in direct contact with Nb are assumed to become
superconducting by the proximity effect.

Fig. 5.37 shows that the 2CSR model with no adjustable parameters [5] predicts rather
well our experimental values for AAR vs N in samples with no FeMn. Fig. 5.38 shows
that it also describes well the increase in ARp with increasing N, but does not get the
absolute magnitude correct. The source of this discrepancy might be that the ICC = lnm

Co layers are too thin to be fully continuous, and thus do not contribute fully to ARp.

167

 

 

6 l 0 Data r
chRmodeI

 

 

 

Vj I 7 Y’Yﬁ I I I I Y

YrIIY

AAR(me2)

 

 

 

 

F ig.5.37. AAR(N) in [Co(6)/Cu(20)/Co(l)/Cu(20)]N multilayers. The ﬁlled circles are
data points and the solid line is a no-adjustable-parameters prediction of the 2CSR model.
There is a good agreement between the two, especially at small N.

 

25ﬁ111ﬁ"1‘ﬁ1F71’1—'T

 

0 Data
— 268 R-model

 

 

 

20}

ARP(me2)
s 3

(II
I I T l I
L A l

 

 

04111111111L11L11111l
0 2 4 6 8 10

 

Fig.5.38 Comparison of 2CSR model prediction of ARp(N) with data for
[Co(6)/Cu(20)/Co(l)/Cu(20)]~ multilayers. The data falls below the predictions, which
may be because the Co/Cu interfaces are not fully established in the 1 nm Co layers.

168

Fig.5.39 compares data for multilayers of the form of Fig. 5.36 with data for
multilayers without FeMn. For convenience, we call the samples without F eMn ‘regular
samples’. Fig.5 .39 shows that introducing FeMn increases AAR signiﬁcantly, especially
for the smaller numbers of layers in the multilayer. For example, AAR increases by about
90% for N=2, and by about 46% for N=3. The effect gets weaker for larger number of
repeats, in part because the effects of the contact resistance, 2ARs/p, and of the additional

Co/Cu interfaces that might become GMR active, get smaller with increasing N.

 

 

 

 

 

 

 

3 L I Iﬂ ' ‘ T I I ' 1 T -' 1 1 1 1 T] r rﬁ I l v v r r l 1 fr 1 l r—r q
1 I 1
205 ‘T J
A : . 1
N " q
E 2 _ I .2
* 4
"" 1.5 ; o a
v :
n: 1 :L I
< » .
< t o 1
- I 1nm FeMn
0'5 if 0 O No FeMn l
o 1 A 1 1 1 +4 1 1 1 1 1444 1
1 2 3 4 5

Fig.5.39. AAR vs N for multilayers with (ﬁlled boxes) or without (ﬁlled circles) F eMn.
Structure is [Cu(1 0)/FeMn(1 )/Cu(l O)/[Co(6)/Cu(20)/Co(l)/Cu(20)]N/FeMn(1)/Cu(5). The
‘regular’ samples are identical, just without the FeMn layers. AAR increases by ~ 90% for
N=2to~9%forN=4.

169

 

20 TVYT—IITYIIIITTIfTﬁIfI l 1 r1 Tﬁ III I (Til 1171

 

 

 

 

 

 

1
A15— I I I —
N 1-
E _ I O -
v _ .
:1?L 1
_ . -
< 5* I 1aneMn “
t O NoFeMn ‘
1 J
0 11 11111111 11111111 111 11111k
1 2 3 4 5

Fig.5.40. ARp vs N. Effect of the FeMn layers on the total speciﬁc resistance, ARp.
Boxes and circles have the same meaning as in the previous ﬁgure.

170

 

 

 

 

 

 

20 I HI I ITTIWTI I I IT I I I I I l I I Ijj IfI I I I I I I I I I I
I
~ I
15 e . 2
i I
O
E
a 1° ‘ 1
<1 I 1
5 — I 1aneMn —1
O NoFeMn
o ’1 1 lJlJ 1L1 ll 1+Lgl 11411114111 1 141 1141141 1 l

 

11.5 2 2.5 3 3.5 4 4.5 5
N

Fig. 5.41. AR/R vs N with (circles) and without (squares) FeMn inserts. Since both AAR
and ARp increase with introducing the F eMn layers, the magnetoresistance, AR/R,
changes only slightly.

Fig. 5.40 shows that inserting the FeMn layers also increases ARp. This increase in

ARp is largest at small number of repeats N.

Fig. 5.41 is a combination of the data of Figs. 5.39 and 5.40. Fig. 5.41 shows that
inserting the FeMn layers slightly decreases AR/R. So inserting thin layers of FeMn to
introduce strong spin ﬂipping can signiﬁcantly increase both ARp and AAR, without
greatly reducing AR/R. Being able to increase both AAR and ARp is highly desired for
CPP-MR potential applications, especially if AR/R increases with them or stays

unchanged.

l7l

5.4.2: Fitting Data to the 2CSR Model:

Fitting the data to models helps us better understand how the FeMn enhances AAR.
As discussed at the beginning of the section, introducing strong spin ﬂipping at the ends
of the multilayer can make the VF model for AAR reduce to equation 5.1. The
introduction of FeMn might also have two other possible effects. It may destroy the
coherence between the current-carrying electrons and the holes created by Andreev
reﬂection, or stop the Cooper pairs completely and limit them to outside the area between
the two FeMn layers. Either of these two effects could turn the outside Co/Cu interfaces
MR active. For our ﬁts we use parameters found earlier [5,43], including
11121111111111, = 0.6 11121112, 131.111,: 875 nan , and Axum“: 1 1:21:12 .

Fig. 5.42 compares the data with three 2CSR model curves. The data do fall above
the dot-dash line expected without any effects of FeMn. But they fall below both the
dashed line 2CSR model prediction assuming equation 5.1, and the top dotted line, which
adds to the numerator of equation 5.1 two additional Co/Cu interfaces.

We conclude that introducing tpcMn = lnm into the Cu layer next to superconducting
Nb does increase AAR, but not by as much as would be expected if it were fully inside a
non-superconducting metal. To check whether 1 nm of FeMn was enough or not, we
tried samples with FeMn layer thickness 2 nm or 3 nm. Fig. 5.43A shows that increasing
the F eMn layer thickness does not change ARp from the samples with 1 nm of FeMn. Fig.
5.43B, however shows that increasing the thickness of the FeMn decreases AAR slightly,
which means the magnetoresistance would go down for the thicker FeMn layers. So, Fig.

5.43 shows that 1 nm of FeMn is better than any thicker layers to enhance the CPP-MR.

172

 

5..f-.4......1ﬁ+f. ,,,.,,ﬁ1,,,

I Data ‘
— 208R Prediction, No FeMnm 1
- - Predicted, Outer ColCu=ColS ,
----- Predicted-FeMn-Co -

 

I

I

I

I

 

 

 

AAR(me2)

 

 

 

 

Fig.5.42. AAR versus N. Data (ﬁlled boxes) for samples with FeMn thin layers, 2CSR
model predictions for the case without FeMn (solid curve), assuming that FeMn just
removes ZARs/p (dashed curve), and assuming that F eMn removes ZARs/F and turns on
asymmetry at the outer two Co/Cu interfaces (dotted curve).

173

 

20 I I I I I I I I I I I I I I T I T I I I I I I Ij I I I I I I I I T

 

 

 

 

 

L.
1.
A15” I I I .2
N t I
E :
G _
u- 10— —
V b l
n- l
a: .
< 5— (A) _
;
L 1 ﬂ1ﬁ11ﬁ111r11fﬁ11 1 I114
1
A I .
N I
E 1; 1
C} :
“- C
c: 0.5} —
3 - (B)
Ob““"*“l“"l““1““"+"1"L‘
0 0.5 1 1.5 2 2.5 3 3.5
nm
tFeMn( )
F ig.5.43. Effect of mm on ARp (top) and AAR (bottom) for

Nb/Cu(l 0)/FeMn(t)/Cu( l 0)/[Co(6)/Cu(20)/Co(1)lCu(20)]2/FeMn(t)/Cu(10)/Nb samples.
Increasing (FeMn beyond 1 nm essentially does not change ARp and reduces AAR slightly.

To further check whether the outer Co/Cu interface contributes to GMR, or still has
no asymmetric scattering, we replaced the Cu/FeMn/Cu layers on each side of the
multilayer by Cu/FeMn/Cu/Co/Cu. Now the Cooper pairs should stop outside the outer

Co layers, and all the volume between the two outer Co layers should be normal. Fig.5.44

174

shows that adding the Co layers had little effect on AAR, while increasing ARp. This

result shows again that 1 nm of F eMn is enough to cause the maximum enhancement

 

 

 

 

 

 

 

 

 

 

 

 

attainable in those samples.
25 I‘rIT I I II I frfr‘ I I I I I I I I ' I I I I I I I V I I I I If?
1
20 . —
1. A
A A
N
E 15 - I I I _.
Cl I
*
v
0.10 :’ "
m ~ I FeMn 1
< : A FeMn-Cu-Co
5 - 7 _
l.1 l #1141 l l 1111 111 1 11 l
.ﬁ l ‘ l ' I r l ‘Tﬁ Ti ' l ' I l
% A
Z I
2.5 '_ _l
A C
N
E 2 - l .1
u- 15 — i
v d
a: 1 1 1|
< 7 _
. I FeMn
< : A FoMnICuICo
0.5 L— ' T
0 b1 1 1 1 1 1 1 111 1 1 1 1 l
1 2 3 4 5

Fig.5.44. ARp (top) and AAR (bottom) vs N for FeMn inserts with (triangles) or without
(squares) additional thin Co layers outside of the FeMn layers. Adding the Co layer gives
no gain in AAR, but increases ARp.

Finally, to check for possible structural changes due to introducing FeMn into the

multilayers, we looked for the difference in ARp and AAR when the FeMn is sputtered on

175

the bottom or top of the multilayer. F ig.5.45 shows that any difference is small. The
reasons for the behaviors of F eMn in the superconducting layer are not yet clear. Some

further information about such interfaces will be presented in Ch. 6.

 

 

 

 

 

 

 

 

 

 

 

 

 

15..-.,j...j....,.....”WV...
o
i I l
A C I g .
NE 10 '_ . _
: o
u- .
v :
a_ _
n: 5 p 'i
< r I FeMn onBothSldes -
i O FeMn on Top Only I
: (A) O FeMn on Bottom Only 3
’ l
1 1 1 Li 1 J Lin 4 1
I IT IT I 'ﬁ—I ﬁrﬁ
1
I i
1 _ I 3 _
A . . ‘
N L- -1
E 0.8 — 8 i
V I I
0: 0.4"— l
; I FeMn on Both Slcles
_ O FeMn on Top Only _
0'2 r (B) O FeMn on Bottom Only
0 b 1 l 1 L 1 1 l 1 1 L4 1 l l l l l l l l J
:6 8 5
a " 8
5 8
t3 Pocltlon of FeMn Layer

Fig.5.45. Effect of the position of FeMn on (A) ARp and (B) AAR. Growing the FeMn on
the top of the multilayer gives a larger AAR and a slightly smaller ARp than when FeMn
is grown before the multilayer.

176

CHAPTER 6

CPP-MR STUDIES WITH RUTHENIUM (Ru).

There is now great interest in Ruthenium (Ru) for CIP-MR technology as the basis for
a Co/Ru/Co ‘synthetic antiferromagnet’ (SAF) [13,14,119-121], in which a very thin (tn.l
= 0.6 nm) Ru layer strongly couples two Co layers (tc0 ~ 3 nm) antiferromagnetically.
This coupling is used in read heads of structure [AF/Co(3)/Ru(0.6)/Co(3)lCu/F] [119],
where F is a ferromagnetic alloy with high contribution to the GMR and AF is an
antiferromagnetic material that pins the Co layer in direct contact with it. SAFs offer
better stability over many reading cycles than a normal antiferromagnet alone.

To estimate whether Ru can play a signiﬁcant role in CPP-MR technology, it is
necessary to know its CPP parameters: pg“, rec/Ru, ARCo/Ru, and 15,9. In addition, CIP-
MR measurements on Co/Ru multilayers have been interpreted as yielding a negative
interfacial scattering anisotropy parameter [122, 123]. In the CPP-MR this would
translate into a negative 7Com...

In the ﬁrst part of Ch. 6, we determine the CPP properties at 4.2K of sputtered Ru and
examine AAR for a variety of different sample structures involving Ru as the N-metal.
Among the results obtained is evidence that mom“ is indeed negative. Some of this work
has already been published [124]. In the second part of Ch. 6, we use this negative rec/Ru
to construct a multilayer structure that lets us investigate in a controlled way the

anisotropic scattering properties (speciﬁcally )1) of various interesting interfaces, 1. We

177

do so by constructing a multilayer that gives AAR s 0 for any I that has 71 = 0. AAR is
then very sensitive to the properties of interfaces I that have non-zero values of )1.
6.1 Ru FOR CPP-MR:

The resistivity of Ru was shown in Ch.2 to be pRu= 9.5 :l: 0.5 pﬂcm . We begin our
studies of CPP-MR properties of Ru by determining the spin-diffusion length in Ru, 1% ,
and spin-memory-loss at Ru/Cu interfaces. We measure these using a Py-based EBSV,
spin-memory-loss detector [73] with structure (Fig. 6. 1 ):
[Nb(l 50)/Cu(20)/FeMn(8)/Py(24)/Cu(1 0)/X/Cu(10)/Py(24)/Cu(20)/Nb(Nb(l 50). X is
the material creating the spin-memory-loss to be measured, in the present case either a
single layer of Ru of thickness tRu or a [Cu/Rub; multilayer with variable N. We measure
how AAR for this EBSV decreases with increasing tau or N, and ﬁt the data numerically

using the theory of VF and the parameters for Py-based EBSVs as described in Ref. [74].

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Pinned Py Test material Free Py

Fig. 6.1. Spin-memory-loss detector. In the experiments to be described, X is either a
single layer of Ru or a Cu/Ru multilayer.

178

Fig. 6.2 shows how log AAR decreases with increasing tn“. For no Ru, AAR has the
expected value for a Py-based EBSV, ~ 2.1 f Qm2[73]. As tau increases, AAR ﬁrst
decreases rather rapidly, and then more slowly. The slower decrease is consistent with a
straight line on the log-linear plot, indicating an exponential decay. The initial falloff in
the curve is due to the formation of two Cu/Ru interfaces. Those interfaces suppress AAR
both by increasing the speciﬁc resistance of the sample (see Eq. 3.3) and due to spin-

ﬂipping at the Cu/Ru interface. The slower exponential falloﬁ‘ is what is expected from

the V-F model (AAR ~ e‘m 13F) ), for ﬁnite 1;}; in the Ru. From the slope of this line we

estimate 13: = 14 :t 1 nm for our sputtered Ru. This spin-diffusion length is much longer
than the thickness of Ru layers used in synthetic antiferromagnets. To fully ﬁt all of the
data in Fig. 6.2 we must know also ARCu/Ru and the spin-memory—loss parameter for
Cu/Ru interfaces.

To measure ARCu/Ru and the spin-memory-loss parameter 6cm at Cu/Ru interfaces,
we replaced the single Ru layer by a [Cu(5)/Ru(3)]N multilayer. Fig. 6.3 shows ARAp vs

N. Since each bilayer contains two Ru/Cu interfaces, a linear ﬁt to the data gives
ZARCu/Ru = 2.0 i 0.2 me2. Fig. 6.4 shows AAR vs N. The rapid dropoff means that

Cu/Ru interfaces cause strong spin-memory loss. To estimate the spin-memory-loss per

interface, we use the V-F model and subtract the spin-memory-loss contribution of bulk

Ru. We obtain a spin-ﬂipping probability of 35 i 5% per Cu/Ru interface.

179

 

 

 

 

 

 

 

10_.........,.ﬁ..,..,.er,,
P C Data] 1
A ‘ Interface Formation """ ﬁt
N /
E 1 _. Exponential Decay ‘
G F .0. ------
. --..,_
u- .........
; 01 r . ..... .
< 1
FeMany(24)ICu(10)IRu(t)ICu(10)IPy(24) '
o_o1411I1.1..1..IJI.....,,,,
0 5 10 15 20 25
t(nm)

Fig.6.2. AAR on a log scale vs tau. The steep drop at small tan is due to the formation of
two Cu/Ru interfaces. After the two interfaces are established, AAR drops exponentially
with tn“, and the slope of the straight line on this semi-log graph is the inverse of the spin-

diffusion-length in Ru 1%: = (14 j: 1) nm.

 

 

 

351""""“""l"""""'“H
3 l
30 } ”-9 ------ 8 :
A 25 E ' -
NE :- ........ 5 ' '-
C} 20?". -
u— :
Va 15 _ ----- ARAP=20.6+2.1 N _‘
< _ 3
E 10 E- FeMn(8)IPy(24)ICu(10)I[ Ru(3)lCu(6) 1N ICu(10)IPy(24) {
5 L 1
I 1
t
o n 1 . . 1 . 14 . .4 1 1 1 L

 

0 1 2 3 4 5 6

Fig. 6.3. ARA]: vs N for a Py-based with EBSV with an [Cu/Rub, insert. The slope of the
best ﬁt line gives MR0“, = 2.0 i 0.2 mm?

180

 

10 I I I I U I I l—I‘ I l ITﬁ'j It I l I I I—' I V

 

 

 

 

 

 

FeMany(24)/Cu(10)I[Cu(5)IRu(3)]NICu(10)IPy(24)
NA '. e o ta
_. ' a _
E 1 E ----- Fit 5
C: .
H— ..
V o'- .
n: '.
4 0.1 r '. 1
E c. 5
< ., 1
5... '
0.01 ...1.4. J. 1.11...1.I.1.4,

o 2 4 6 810 1214
N

Fig. 6.4. AAR on a log scale vs the number of [Ru(3)/Cu(5)] bilayer inserts. This straight
line ﬁt (on the log scale) of the data gives a spin-flip probability of electrons at the Cu/Ru
interface ~ 35% per interface.

6.1.1. CPP-MR of Co/Ru.

Motivated by literature evidence [46] of negative anisotropy (i.e. negative ,6) in CoRu
alloys, we decided to examine the contribution of Co/Ru interfaces to the CPP-MR, given
that the interfaces of sputtered Co/Ru multilayers can be thought of as a CoRu alloy. In
the process, we studied simple [Co/Ru] 1v multilayers, symmetric Co-based EBSVs with
Ru as the N-layer, and [Py(6)/Cu(20)/Ru(2)/Co(t)/Ru(2)/Cu(20)]” hybrid multilayers
with Co/Ru interfaces. We hoped that a negative ﬂ for the CoRu alloy would produce a
negative 7 at the Co/Ru interface. Since bulk Co has a positive ,6, Co layers having

interfaces with Ru would have to be thin for the interface contribution to dominate. As

shown in Eq. 3.3, a negative 7 for Co/Ru can be inferred from a negative AAR for a

181

hybrid multilayer in which the other ferromagnetic metal (i.e. Py) has positive
parameters. We ﬁrst used the hybrid multilayers
[Co(6)/Cu(20)/Ru(l)/Co(l)/Ru(l)/Cu(20)]1v because we had studied such multilayers
thoroughly in Chapter 4 (without the Ru layers). If 7Com“ is negative, and two Co/Ru
interfaces contribute more to the CPP-MR than does 1nm of Co, then the above
multilayer should give an inverse GMR, characterized by AR(H) curves having ARAp <
ARp, Fig. 6.5 shows AR(H) for the above multilayer. Although the curve is noisy,
because of the small value of AAR and the need for a relatively large H to switch the
magnetization of the Co(6) layer, and despite a gradual rise in the background AR with H,

it seems clear that the CPP-MR is inverse.

k1277-44
[Co(6)/Cu(20)/Ru(1)lCo(1)lRu(1)/Cu(20)]6
33.8....,....,....,....,j..ﬁ,ﬁ..
* i
: 1
33.7 g. .3

 

33.6

33.5

AR(mez)

33.4 h

 

 

l I J l J

l l I l l

 

33W

-150 -1000 -500 O 500 1000 1 500
H (Oe)

Fig. 6.5. AR vs H curve showing an inverse CPP-MR, i.e., ARAp < ARp. The small AAR

compared to noise, and the background change with H, make it hard to quantify the data.

 

182

To get a larger AAR at smaller ﬁelds, we replaced the thick Co layer by a 6nm thick
Py layer, giving [Py(6)/Cu(20)/Ru(l)/Co(1.5)/Ru(l)/Cu(20)]5. Py gives a larger
contribution to AAR than Co (because both ,6 and p are larger for Py than for Co), and
also needs only a smaller switching ﬁeld. Fig. 6.6 shows a clear inverse MR. Since the
anisotropy parameters for bulk Py, the Py/Cu interface, and bulk Co are all positive, only
a negative rat the Co/Ru interface can create such a negative AAR.

Fig. 6.7 shows that the sample of Fig. 6.6 also has an inverse CIP-MR. We presume
that this inverse CIP-MR is also due, at least partly, to a negative rat the Co/Ru interface.
If the negative CIP-MR was only due to the AMR of Py, then the switching ﬁeld should
be only that of Py. Comparing Figs. 6.6 and 6.7 we see that the switching ﬁelds are the
same and have magnitude z 450 Oe, which is the switching ﬁeld of the Co layer, not the
Py one (i.e. the MR depends on both the Co and the Py, which is peculiar to GMR, not to
AMR). The sharp peak in the CIP-MR at small ﬁelds might be associated with the AMR
of Py. As in many other cases, the CIP-MR is much smaller than the CPP-MR.

If our analysis above is correct, then as the Co thickness, tea, increases, AAR should
decrease in magnitude and ﬁnally switch sign, becoming positive when Co is thick
enough to overcome the negative contribution of the Co/Ru interface to the CPP-MR.
Fig.6.8 shows the progression of AR(H) curves with increasing too for three different
samples. The Co/Ru interface dominates at low Co thickness giving an inverse MR,
whereas bulk Co dominates at higher thickness giving normal MR. The three curves in

Fig. 6.8 are plotted on the same scale to allow direct comparison between the signs and

magnitudes of AAR.

183

To estimate the value of m we made a series of samples with varying too like those
of Fig. 6.8. Fig. 6.9 shows AAR versus ICC. AAR follows an almost straight line with tea,
switching from inverse to normal at too z 5 nm . This straight line can be explained with
Eq. 3.3 in Chapter 3, where we see that the ‘bulk’ Co contributes to AAR in two places.
First, it adds to the total speciﬁc resistance in the denominator. But this addition is small,
due to the relatively small resistivity of Co. The more important contribution is through
the numerator, where AAR is proportional to too, thus explaining the straight line in Fig.

6.9. Using this ﬁgure, we estimated the value of rec/Ru to be

— (0.15 i 0.05).

 

 

 

 

 

43.644..,...,... ,
’ k1302-2
C [Py6lCu20IRu2/Co1.5IRu2ICu201x6
43.4 “‘2
"A
E 43.2 _
Cl :
h r-
V 43 —
m .
< I
42.3 _
42.6h“‘P1---nl-.L.n.4L.‘
4000 -500 o 500 1000

H(Oe)

Fig. 6.6. AR(H) for the hybrid sample [Py(6)/Cu(20)/Ru(2)/Co(1.S)/Ru(2)/Cu(20)]6,
showing inverse CPP-MR.

184

a: 1.743 f

1.742 :

K1302-2R

 

[Py6/Cu20/Ru2/Co1.5/Ru2/cu201x6

 

 

l J

 

 

 

i-‘sbd'”o""566"'
H(Oe)

Fig. 6.7. CIP R(H) loop for the sample of Fig. 6.6. The CIP-MR is also inverse.

185

“ Viﬁ‘ﬁﬁTVV'IWYUIVYI‘

k1302-2
[Py6ICu20IRu2ICo1.5IRu2ICu201x6

 

 

 

 

 

 

 

 

 

 

 

N 43.5 ; E
E - - ,. _ .
C:
h
I! 43
<
k1321-7 E
[Py6lCu20IRu2/Co3lRu2/Cu20]x6 3
N 41 ; 3
E : :
c: 5 3
'0- . 1
I! 40.5:— i
< t ;
g 3
40;::::;::..'T+..::::::l
k1321-3 3
[Py6lCu20IRu2ICo1OIRu2ICu201x6 ;
«7‘ 45.5i 3
G I j
“" t :
V l‘ q
n: 45 L _‘
44.5 b i L. l 1 L 1 . 1 1 . . 1 . 1 . . . ‘
-1ooo -500 o 500 1000

H(Oe)

Fig. 6.8. Progression of AR(H) curve with increasing Co layer thickness. MR is inverse
for small thicknesses of Co and switches to normal for large Co thickness.

186

0.3

[Py6ICu20

IRu2ICo(t)IRu2/Cu20].5

 

 

T I U T T I r I j 5

I l 1 Pl I l

 

L I l l l j I j

J

 

 

2 4 6
t(nm)
Fig. 6.9. AAR vs tCO . The data follow a nearly straight line, allowing an estimate of 7Com

= {-0.15 i 0.05). Arrows in the plot indicate that those data points involve only a lower
bound on the AP state.

8'10

12

To further study the CPP-MR properties of multilayers containing Ru we made

This choice should yield a good AP state.

187

samples with structure: Nb/Cu/lCo(3)/Ru(0.6)]N/Co(3)/Cu/Nb. We chose the Co and Ru
thicknesses to give coupling between the Co layers at the ﬁrst antiferromagnetic peak.

Fig. 6.10 shows AR(H) curve for the
multilayer: Nb/Cu/[Co(3)/Ru(0.6)]2o/Co(3)/Cu/Nb. Even at H = 8 kG, we do not get the

ﬂat AR(H) evidencing parallel alignment. This behavior agrees with literature results

showing very strong antiferromagnetic coupling in Co/Ru multilayer with Ru layers
thickness IR“ ~ 0.5-0.6 nm [14,119].

To avoid such high ﬁelds, we switched to IR" = 1.6 nm, at the second AP coupling
peak. This change allowed us to switch the magnetic state from AP to P within the
available ﬁeld range. Moreover, since the Co/Ru interface is quite broad [125], the

thicker Ru should create better interfaces then tau: 0.6 nm.

 

IUﬁ'IITIIIU'lrU

1VL2 ..r4..,. .r...,
. k1 271 -7 .
L Nb/CuI[Co(3)IRu(0.6)] mlCo(3)ICuINb j

"‘ 3013r

30J1;

AR(mm2

30'-

 

 

 

111

100.0,. ”.0... ”4.00.0.
H(Oe)

Fig. 6.10. AR(H) of an [Co/Rube multilayer. The antiferromagnetic coupling is too
strong to be overcome even at our highest applied ﬁeld of about 8 kG.

 

-8000 8000

188

k1306-7
[Co(3)lRu(1.6)]1oICo(3)
22.61gm....,...,...rﬁﬁr,...,...,.ﬁd

o . i

22.6 :— —‘

 

22.59 3 1

pp (m m2)

- ..

b- .1
o

h —1

- -<

h -1

F ‘
O

>- -1

h -

v -4

h d
I

.— _.

b ‘1

 

AVG AR-C

 

22.55 g .4}

l l I l l 141L111 l l l l l J 11141111411111

22.54
-800 -600 -400 -200 0 200 400 600 800

H (Oe)

Fig. 6.11. Average AR(H) for a [Co(3)/Ru(1.6)]m/Co(3) multilayer vs H. Each point is
repeated 100 times and averaged to get reliable results.

In Fig. 6.11, AAR of the multilayer [Co(3)/Ru(1.6)]lo/Co(3) was so small that we
needed to measure each point 100 times and take the average value of AR(H) at that
point. This extremely small AAR is due to the opposite contributions of bulk Co and the
Co/Ru interfaces, which almost cancel each other at this particular Co thickness ( See
Fig. 6.9). The small CPP-MR in this multilayer agrees with those of the hybrid samples
in Fig. 6.9. The larger value of AAR in Fig. 6.10 could be due to CoRu alloying playing a
larger role, giving a negative ,6, and thereby making AAR larger than when the Ru is 1.6

nm thick.

189

Fig. 6.12 shows ARM» versus N, the number of Co/Ru bilayers, for (Ru = 1.6 nm. In
principle, the slope of the line gives 2ARCo/Ru, and the ordinate intercept gives ZARNb/CO.
However, each additional bilayer also adds bulk contributions from the Co and Ru layers.
Taking the slope as 1.6 i 0.16 fﬂmz, and taking p‘cO = 7 i 2 119cm and pRu= 9.5 $1
uﬂcm, respectively, we estimate ZARCO/Ru = 1.20 i 0.23 f sz. The value of 7.5 i 2.3
f sz for the ordinate intercept equals ZARs/F + p*c°1c°, because we have an extra Co
layer on top of the multilayer. This leaves ZARNb/CO = 7.3 i 2.3 f m2 , which is a bit

higher than, but overlaps with, our ‘standard’ value of ZARNb/Co = 6 i- 1 mez,

 

      
 

I ﬁT T I I I I I I I I I I j I I I I I I I

; NblCu/[Co(3)IRu(1.6)]NICo(3)ICuINb

2
ARAP(me)
a :3 a

.L
O

1

—AR = 7.5 +1.56 N
AP

 

 

T

 

1—
o 1 1 l 1 l J 1 l 1 l l 1 l 4 l J; P l l l l 1 l L

o 5 1o 15 20 25
N

Fig. 6.12. ARAp vs N. Analysis of the slope of the straight-line ﬁt gives 2ARCo/Rn = (1.2 i
0.23) f m2, and analysis of the intercept gives ZARc°mb= (7.3 :1: 2.3) f sz (see text).

To better control the AP state, and achieve a wider variety of MR data for Co/Ru, we

also made EBSVs with structure: Nb/Cu(10)/F eMn(8)/Co(t)/Ru( 10)/Co(t)/Cu(10)/Nb.

The resulting values of AAR could be even smaller than those of multilayers, depending

190

on the Co layers thickness, limiting our ability to analyze those samples quantitatively. .
Fig. 6.13 shows the 2CSR model based prediction of AAR(tc°) for a
Co(t)/Ru(lOmn)/Co(t) EBSV. The prediction shows how difﬁcult it is experimentally to
measure such EBSVs. Fig. 6.14 shows AR(H) for an EBSV of composition:
Nb/Cu/FeMn/Co(15)/Ru(10)/Co(l5)/Cu/Nb. Each point is the average of 100 points.
We took the points at ﬁelds where we expected AP or P states based on Co/Cu/Co
EBSVs. The change in speciﬁc resistance is much smaller than predicted in Fig. 6.13.

Either our choise of AP ﬁeld is incorrect, or some of the parameters used in Fig. 6.13 are

 

abit off.
0.1 ......... .....ﬁ....4ﬁ .......
L .
: Co(t)IRu(10)ICo(t) EBSV
A 0.08 - ZCSR Model Prediction «
N .
E
C} 0.06
I.—
“ 0.04
<
< 0.02
o .

 

 

 

 

 

t(nm)

Fig. 6.13. 2CSR Model predicted AAR as a function of tea for a Co(t)IRu(10)/Co(t)
EBSV. AAR(t) is non-monotonic due to the competition between the contributions from
the Co/Ru interfaces and bulk Co.

191

 

[ﬁr I I I I I f

15.45 ...,.f.....,...
i k1324-44

 

 

 

Each point is averaged 100 times

15.44 - -

A h I
N >-

E 15.43 — . —

C:

u— ; .

V 15.42 — ’ .’ —

a: 4 . -

r 0 ~

< _ :

15.41 — 1

15.4 ~ 1 1 L l 1 1 1 l 1 1 1 l 1 1 1 l 1 1 1 l 1 1 4 1

-200 o 200 400 600 800 1000
H (Oe)

Fig. 6.14. AR(H) for Co(15)/Ru(10)/Co(15). Fields were chosen based on the AR(H)
carves ofCo(15)/Cu(20)/Co(15) EBSVs. AAR is much less than predicted in Fig. 6.13.

6.2 STUDIES OF SPIN-ASYMMETRIES AT INTERFACES USING

Ru.

In this section we describe a way to use Ru in a Py— and Co-based hybrid spin-valve
to allow sensitive studies of spin-asymmetric-scattering at certain interfaces. The
interfaces we study are Co/S (where S is Cu, initially presumably superconducting by
proximity effect with the superconducting Nb), Co/FeMn, Co/Cu, and Co/Ru.

The structures we use have the form Nb/Cu(1)/Py(24)/Cu(20)/Ru(0.6)/Co(2)/X/Nb
(Fig. 6.15) where X will be:

(1) Cu (20)

(2) FeMn(t)/Cu(20)

192

(3) Cu(10)/FeMn(2)/Cu(1 0)

(4) Cu( 1 O)/FeMn(2)/Cu(2)/Co(l )/Cu(3)

(5) Ru(2)/Cu(20)

(6) Ru(2)/Cul 0)/FeMn(2)/Cu(10).
As a reference, to establish the magnitudes of expected deviations of AAR from zero and
to choose the magnetic ﬁeld range needed to achieve AP and P states, we also made a
sample with no Ru: Nb/Cu(20)/Py(24)/Cu(20)/Co(2)/Cu(20)/Nb. The AR curve for this
sample is shown in Fig. 6.16. The value of AAR z 0.4 f sz agrees with the 2CSR
model predictions. Note that the very different switching ﬁelds of the Py layer (~ 50 0e)
and the Co layer (> 400 Oe) give good AP alignment over an extended ﬁeld range.

As indicated in the introduction to this chapter, the structure above is chosen to give
AAR 5 0 in case #1 when X generates no asymmetric scattering. We explain how this
occurs using the 2CSR model Eq. 3.3, which we generalize to distinguish between the

two different interfaces of the Co layer.

 

_4 (pryPRyth +7Py/Cu ARRy/Cu)(ﬂCo pCotCo +7Co/Ru ARCo/Ru +7Co/XAR2‘0/X) (61)
ARNb/Py +pCutCu +pRyth +ARRy/Cu +pCotCo +ARCo/Ru +ARCo/X +ARNb/X

AAR
For simplicity, we have assumed that the very thin Ru layer to the leﬁ of the Co layer
contributes only to the Co/Ru interface (i.e., we neglect any bulk Ru contribution). In
practice, this equation has to be modiﬁed due to the ﬁnite spin-diffusion length in Py and
also if X generates spin-memory-loss. But, as such changes do not affect the main point
of our argument, which focuses upon the right-hand term in the numerator of Eq. 6.1, we

omit them to simplify this discussion. Any quantitative ﬁts to the data can take proper

account of all such subtleties.

193

The numerator of AAR in Eq. 6.1 is the product of two terms. The leﬁ-hand term

depends only upon the properties of Py and Cu, and is strictly positive because both ,pr
and mm are positive. In the right-hand term, in contrast, ,BCO p20 tea is positive, but the

interfacial contribution from 7Com“ A R}, , Ru is negative. Using the Ru parameters earlier
in Ch. 6, and our standard Co and Co/Cu parameters [5,43], we have chosen the Co
thickness, (Co = 2 nm, to make ﬂea p20 x(2 nm) z 7Co/Ru ARE. , 11.. , thereby making AAR 5 0

for the ﬁrst two terms on the right. The sign and magnitude of AAR should, thus, be very
sensitive to whether 7A in third term on the right is zero or non-zero.

From prior work [68], we expect this third term to be zero for a simple Co/S interface,
where the superconductor S is either Nb alone or Cu turned superconducting by the
proximity effect with the neighboring Nb. Said another way, prior studies showed the
Co/S interface to be MR inactive (rec/s = 0) [28, 68]. We tested this expectation ﬁrst.

(1) X = Cu.

If rcws E 0, we expect AAR E 0 for X = Cu(20), since the Cu should be turned
superconducting by proximity effect with the adjacent Nb. Fig.6.17 shows AR for this
sample. Because AAR is so small, we no longer took a complete AR(H) hysteresis curve
as in Fig. 6.16. Rather we cycled between a few selected ﬁelds where we expected to get
AP and P states and measured AR at each ﬁeld 100 times and averaged. Happily, the
change in resistance is close to zero (5 0.02 n9). The measuring sequence was as
follows. H = + 700 Oe (to give a P state), H = + 200 Oe expecting the system to remain
approximately in the P state, then -60 0e to give an AP state, -70 0e to repeat the AP

state, -700 to again give the P state, and then a reversed cycle to check for symmetry and

194

double the data set. The sequence was intended to allow us to correct for the size of any
superﬁcial slow parabolic rise of the resistance with magnetic ﬁeld that occurs for single
Nb/F/Nb sandwiches (i.e., that is not GMR related). In the end, the changes in AR were

small enough that any such correction is not crucial.

 

 

 

OOOOOOOOOOOOOOOOOO
--------------------

oooooooooooooooooooo

..........
uuuuuuuuuu

..........
nnnnnnnnnn
..........

IIIIIIIIII

  

oooooooooooooooooooo
oooooooooooooooooo

oooooooooooooooooooo

iifiNbiiiififi Cu

   
  
 
    
     

"
I
J
I

I O
.;.
r': ;
, _
0..

cccccccccc

ttttttttttt

oooooooooo

IIlllIlllIlllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllll

aaaaaaaaaa

cccccccccc

 

 

 

 

 

 

 

 

 

   

 

 

Fig. 6.15. Drawing of the “spin-asymmetry-scattering detector” used to study Co/X
interfacial rec/x. If rec/x = 0, then AAR is also zero. Even small values of 7Com will give
nonzero MR, making the detector quite sensitive.

(2) X = F eMn (1)

We showed elsewhere that 1 nm of FeMn in a non-superconducting environment is
enough to almost completely (> 99.9%) ‘quench’ spin-memory. Here we examine what

happens if one side of the FeMn is in contact with a ferromagnet (Co) [123,124] and the

other is in contact with a superconductor. Fig. 6.18 shows that AAR s 0. The Co/FeMn

interface is also not magnetically active (i.e. mmm -:- 0).

195

k1366-8
Nb1 50IC u20le24ICu20/CoZICu20/N b1 50

 

 

 

 

 

 

 

ﬂ T E I Ti 1 frr I 1 fr
i
11.6
A b
N 1.
E 11.4
a _
F 1—
v
11.2
m _
< 1—
11
10.8PPL1111111111111111...11..1...
-600 400 -200 0 200 400 600
H(Oe)

Fig. 6.16. Reference sample: AAR a: 0.4 chm, is non-zero when there is no Ru in
contact with the Co.

196

k1333-6
NblCule6/Cu20lRu2/COZICu20le

 

 

 

 

1007 U I TI I I I I I I T I—I—V I 1 I I I 1 IﬁTl I I I T y y r

; I AP-State :

10.69 - -

NA . .

E 10.68 _— _

V " .1

m 10I68 '- —1
< .

P . q

1- . 1

10.68 _ 1—1

: l I

: . j

10.67 I I In 1 114 14 14 1 1 l 1! 1 l 1 1 1 l 1 I 1

 

 

 

-500 .400 -200 0 200 400 500
H (Oe)

Fig.6.17. AR(H) curve. AAR is very small in the curve. Solid boxes are taken for the
AP state and solid circle are for the P state.

197

k1333-3 k1366-3

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Py6/Cu20IRu2IC02IFeMn2ICu20 Py24lCu20/Ru2lCoZ/FeMn5ICu20
14.53 ............................. 18.47 . - . . - w. 5-
14.52i ' 9'5“" '3 i' N a 95m :
g o I AP-State ; 1345 g o I AP-State J
N; 14.51 Cb - NE [ 7 2
14.5 L 2 “-45 r . . ~
C : 1 G . .
E “"9 E" ' 0 E 13.44 L ,
< 14.43 E- -: < I .
5 3 18.43 - I o l
1447 : ~ ‘1 t I
I I 3 I I
1‘.“* ............ 1..J_1 .1..41+-4‘ 18.42 L..1...1-4_.1...14..1-..1...J_._..
500-400-200 11 200400500 500-400-200 0 200400500
H (Oe) H (Oe)
k1366-6
Py24lCu20IRu2ICo2IFeMn5/Cu20
17.24 ' ' 9‘5““ 3
; I AP-State ;
17.23 :- o —1
NA 2 I
E 17.22 :- . .j
g 17.22 . '
m 17.21 g a
< 17.21 :1 -f
I I 2
17.2 3 I ’ «j
I I 1
17.2‘...1...1...1...1...1...1...1...‘
500 -400 .200 0 200 400 500
H (Oe)

Fig. 6.18. AR(H) for samples with X = FeMn(2, 5). The values of AAR are very
close to zero in this case, just as in Fig. 6.17.
(3) X = Cu(10)/FeMn(2)/Cu(10):

As an antiferromagnet and a strong spin-ﬂipping alloy, FeMn might block Cooper
pairs ﬁom penetrating to the multilayer (thereby turning the Cu layer between F eMn and
Co normal), or destroy the coherence between the electrons and the Andreev reﬂected
holes from the superconductor [128]. In either case, the Co/X interface could become a
Co/Cu interface. Fig. 6.19. shows AR(H) for four samples with this structure. Separating

the FeMn layer from the Co by Cu changes AAR from being a 0 to a positive value.

Apparently the Co/Cu interface has became GMR active. However, the positive AAR is

198

much less than predicted by the 2CSR model. This disagreement may be attributed to two

different reasons. The spin-memory-loss at the Cu/Ru interface (spin-ﬂip probability

z3 5%), or (even more interesting) possible spin-memory-loss at the Co/Ru interface.

k1333-1
Py6lCu20IRu2/CoZICu10/FeMn2/Cu10

 

 

 

 

 

 

 

 

 

 

 

 

14.23 i . P531111 1
i " ”5““ I 13.15
.4" 14.25 ’- I 1 A .
CE: _ ' - g 13.14
,_ 14.24 _- . ~_
"' » "' 13.12
1: - . a:
< 14.22 ~ .. <
- 0 3 13.1
. . O I
14.2 — I «
’ HO
. 1.1..1.4_.1.. 1. . 13.08
-600 400 -200 0 200 400 600
H (De)
k1365-1
Py24lCu20/Ru2lCoZICu5/FeMn2/Cu15
I' 'I'T‘ IR'VWIi—‘Y I" I I ; 15.98
0 P-State I 1
18.42 I -
I AP-State .
A . : A “-95
°‘ .4 ~ . N
E 18 I ; E
a . - 15.94
,_ 15.33 7 t E .
m 13.35 L ' ; a: 15.31 ’
< ’ . j <
13.34 _- 1 15.39
C .
18.32H,.4111.11144444414 1.1.1 .
600 400 -200 0 200 400 600
H (De)

k1333-5

Py6lCu20lRu2ICoZICu10IFeMn2ICu10

 

*v—v*

O P-State
L I APsStatO

 

 

 

 

 

 

 

 

I’ I V Y I

 

11111411111111111111111411114

.1

I
_ I
I . I .
‘II.|III.J AI4_1AIA
£00 400 400 0 200 400 600
H (Oe)
k1365-6
PyZAIC UZOIRUZICOZICU5IFOMI'IZIICU1 5
:- T_T 1 I . I. ' 1 v ' '
L . P514111 '-
: I 41951315
I
* o

L LLJ 4A 1 A

l

 

 

500-400-200 0 200 400 500
H (Oe)

Fig. 6.19. AR for samples with X = Cu(10)/FeMn(1)/Cu(10). AAR is now a non-zero
positive value, which indicates that the Co/Cu interface is now contributing to GMR.

(4) x = Cu(l 0)/FeMn(2)/Cu(2)/Co(2)/Cu(3).

We inserted a ferromagnetic Co layer to see if it would enhance the effect of FeMn
seen in case (3). The idea was that the Co layer might block the passage of Cooper pairs
ﬁ'om the Nb, making everything to its leﬁ normal instead of superconducting. As
Fig.6.20 shows, introducing the Co layer added little or nothing to AAR from what had

been seen with the FeMn alone.

199

O P-State .
16.26 1 I AP-Statel 1
- ’ 1
NA ’ ‘
E 16.24 ~ I i _
G .
V 16.22 - ..
0: - .
16.2 - ° . _.
. O
- l
16.18 ...1...1...1...14-.1...1..444..
-600 -400 -200 0 00 400 600
H (De)
k1333-7
Py6lCu20/Ru2IC02ICu10IFeMn2/Cu2lCoZICu3
[ O P-State i I l
14 6 _ I AP-State .1 . ..
. . 7 I ‘
NA " 1
E 14.58 - J
C} 1 -
"' 1
V 14.56 _
n: .
< ' :
14.54 - .
. .
. O O .
. 0 l
14.52 . 1 1..-1..1..-1.. 1 .1 .

k1333-2
Py6lCu20IRu2ICoZ/Cu10IFeMn2ICu2ICoZ/Cu3

 

 

 

 

 

 

 

 

 

 

 

 

 

 

-600 -400 -200 O 200 400 600

H (Oe)

Fig. 6.20. AR for samples with X = Cu(10)/F eMn(2)/Cu(2)/Co(2)/Cu(3). The AAR values
are almost identical to the ones without the Co layer.

200

(5) X = Ru(2)/Cu(20):
If Ru is simply turned superconducting by the proximity effects with the
superconducting Nb through the Cu, then this insert should still give AAR 5 O. In
contrast, Fig. 6.21 shows a modestly strong negative AAR. This behavior seems to
contradict our earlier argument that a Co/X interface has no GMR contribution when
X is a normal metal in proximity to superconducting Nb. But we saw in Section 6.1
that a Cu/Ru interface causes a signiﬁcant spin-memory loss ~ 35%. Perhaps the
outer Ru/Cu interface could be responsible for making the Co/Ru interface GMR
active[129].
(6) X = Ru/Cu/FeMn/Cu

To check the plausibility of our explanation for case 5, we put FeMn into the
middle of the Cu outside the Ru layer. Comparing Fig. 6.22 with Fig. 6.21 shows that
inserting the F eMn produced almost no change in AAR. This suggests that the effect

of the Ru layer is similar to the FeMn layer, but with less strength.

201

14.25

AR(mm’)
.2
N

d
P
i
"I

14.1 F

13.54
13.52

13.5
‘0- 13.48

1 3.44 P
1 3.42

Fig. 6.21. AR for samples with X = Ru(2)/Cu(20). Putting the Ru next to Co gives a
negative AAR indicating that the outer Co/Ru interface is active and contributes to

GMR.

k1366-1
Py24lCu20/Ru2lCoZ/Ru2lcu20

 

TTf‘IIYTY

 

r '''''' 1
O1 Pstato
I AP-State

 

 

444141141111 14141111 1 I 11414—14141

.3

l
i

 

 

-300 .400 .200 0 200 400 300
H (De)

k1366-5
Py24/Cu20/Ru2/Co2/Ru2/cu20

 

13.43 :

l l l l 'l'

I

 

l l l

O P-State
I AP-State

IIﬁrfrI Tjij

 

 

 

L44L11l 1411111111

-600 .400 .200 0 200 400 300
H (Oe)

202

q

 

k1333-4
PyﬁlCuZOIRuZICoZ/RUZICM OIFeMnZICu10

 

 

 

 

 

 

 

 

 

 

 

 

15.03} -
"A1504: "
g :0 j

’ O
“15.02: 0 .-
05 ’ J
< 157 0 P4531133 .

I I AP-Stato 1

14.33, 1

. I, .f

1‘.”....1...1.-.1.441...1...1...1...
500-400-200 0 2 400
H(Oe)
k1355-2

Py24lCu20lRu2JCoZlRu1ICu5lFoMn2lCu15

t1; [VVYIVVITTYYI—vvvIvvv—Ivvvlvr—l

20.23? ' .:
A : O ‘
N 20.24- J
E . 1

I J
E 2012: o P-Stato :
g 20.2:— 1

20.13: . J

z 1

20.16 ...4...1...1...1...1...1...1...l

-300 400 .200 0 200 400 500

Fig.6.22. AR for samples with X = Ru(l or 2)/Cu(5 or lO)/FeMn(2)/Cu(10 or 15).

H (De)

 

 

 

 

 

 

 

 

 

 

 

 

 

k1365-4
Py24lCu20IRu2/CO2JRu2ICu5lFoMnZICu15
: .2

18.42- 0
I. o ’ 1
N . .
E 18.4: 1
c . 1
518'“: o P-suto :
V ; I AP-Stato :
II513.33. ~ .
< ; 1
18.34: . I 3
18.32h...1...1...14..1...1...1..._1...

400-400-200 0 200 400 600

H(Oe)

1113664
Py24ICu2OIRu2ICoZIR02ICu5IFoMn2lCu15
.-..,...,..4,-.-,...,.-.,...,.‘.‘
I0 3

17.4- . -
. . :
a""1731“ ‘
E ' E 2
c . 4
~17“? o P531110 1
V I I AP-Stat. I
0:17.34- .
< : 1
17.32; %
‘ 1
: I" ‘

17.3 ...1...1...1.4.4...1.-41-..1-.

600-400-200 0 200 400 600

H(Oe)

Adding the F eMn layer changes AAR very slightly from its value without F eMn.

203

Chapter 7
GMR IN MULTILAYERS WITH SUB-MICRON

AREAS

In this short chapter we present some preliminary data on trilayers of
Co(30)/Ag(lO)/Co(1.5) lithographed into nanopillars. As noted out in Ch. 1, the study of
F/N multilayers with diameters ~ 100 nm has attracted attention because of their potential
for devices and the new prospectives on physics that they open.

Here we present room temperature (RT) and liquid-helium-temperature
measurements of resistance in a few such samples. Only a portion (~ 3 nm) of the
‘thickness’ of bottom Co layer is lithographed to avoid the magnetostatic antiparallel
interaction created by free poles when both F -layers are lithographed to the same small
area. The pillar has a diamond shape to use shape anisotropy to try to achieve a single
magnetic domain in the thin Co layer (See Fig. 2.10). The pillar diameter is z 150 nm.
Fig. 7.1 shows the R versus H curve for a Co(30)/Ag(10)/Co(l.5)/Ag(60) sample. The
sample resistance of ~ 1 Q is measured without the need for the SQUID-based system,
enabling measurements at room temperature. Because the peaks are sharp, we can put
only a lower bound on the AP state. However, the peaks are reproducible and symmetric.
The sharp switching from the AP to the P states at the outer side of each peak indicates
that the magnetization of the patterned thin Co layer switches as a single domain (i.e.

bulk rotation rather than domain wall formation and motion).

204

 

Fig. 7.2 shows the R(H) curve for the same sample at 4.2 K. Note ﬁrst that the
changes in resistance AR are almost the same in Figs. 7.1 and 7.2. Second, as expected,
the resistance at 4.2 K is less than the room temperature resistance, due to freezing out of
the phonon scattering at 4.2K. Third, also as expected, the switching ﬁelds are higher at
low temperature. Fourth, the R(H) curve at 4.2K is not symmetric around H = 0, but is
shifted towards negative H values. This asymmetry could be due to exchange bias created
by a C00 layer on one or the two of the Co layers. C00 is antiferromagnetic with a
Curie temperature less than room temperature, which may be why there is not any

exchange bias at room temperature.

Co(20nm)lAg(10)ICo(1 .5)IAg(60)

 

 

 

 

 

 

 

 

 

 

0.933 , , .

C N ]

0.986 1 j

E 0.934 — j

z ..

Q

0: 0.932 — __

0.93 — M
0.973lwm.1........-....
-0.4 -0.2 0 0.2 0.4

H(~1030e)

Fig. 7.1. R vs H at room temperature. The sharp drop on the outer side of the two
peaks indicates a single-domain magnetization in the lithographed Co layer.

205

Co 20IAg 10ICo 1.5/Ag 60(nm)

0.739
0.733 :-
0.737 :— —E
A 0.786 :— 1 —f
v 0.735 E— » {
0.734 i —f
0.733 f— - —f
0.732 “V ‘{

0.781

 

IWrTITIII

11114114

 

 

 

-2”T1.£s"‘-1‘”-b..‘r.”‘o”‘0.é”‘1 1.5 2
H(kOe)

Fig. 7.2. R vs H of the same sample in Fig. 7.1 taken at 4.2 K. The R(H) curve is
asymmetric now probably due to the formation of an antiferromagnetic CoO layer on one
or both Co layers. Arrows indicate the direction of changing H.

 

Our current-driven switching data are not as clean as the ﬁeld-dependent switching
shown above. Fig. 7.3 shows the R0) curve of the same sample at 4.2 K. There is an
overall drop in sample resistance with current that we do not yet understand (it might be
due to some tunneling through the SiO insulating layer). The most important features of
Fig. 7.3 are the two steps on each side of the zero. The jump in resistance, AR, in each of
those two hysteretic features is equal to the jump in resistance in the R(H) curve. Those
equal jumps in R suggest that the source of the changes in R is the same in both cases (i.e.
magnetization switching of the thin Co layer). The two hysteretic features are symmetric

around zero current, suggesting that the switching is caused by the self-ﬁeld (induced

206

magnetic ﬁeld) of the current as discussed in Ch.1. The two features were reproducible

for two different ﬁeld sweeps.

0.33 _

0.62

0.61

0.6

VII (Ohm)

0.59

0.58

Co(20)lAg(10)/C

o(1.5)

 

I I T

IIIIIIII

 

l l

-20

Q

-15

IIIIITIIII'

l llJlllllLllll

-10 -5 0

|(mA)

II.

7 l

5 10

111 IJJJI

I [ff Iﬁ

—

Id

1
15 20

 

Fig. 7.3. R vs I curve at 4.2 K of the same sample of Figs. 7.1 and 7.2. There is an overall
drop of resistance with current. The two hysteretic loops between i 10 and 3:20 mA are
probably due to current-driven switching of the magnetization. Open boxes and ﬁlled
circles are for two sweeps of current.

Fig. 7.4 shows R(H) at room temperature for a sample with the same structure, but a

better AP alignment, which leads to a larger change in resistance. The MR of this sample

is about 3%, larger than the value in ref [20] and comparable to the results of [19].

207

 

I I I I 17 I ﬁr I

Co(20nm)lAg(10)lCo(1.5)/Ag(60)
1.37...,.......

I I I

1.33 _—

9”"
l

_ l
1 .35 L 1”

Q
l
0

k1.

l

_ 1' j
_ / 9 .
C 1 i
1.33 — ,1 U \ —
I New 2

6
I
1
I

1.34 :-

R (o)

- /
L ,I/"
1 I32 ’- M

1.31h'41111111111...1..41..
-0.6 -0.4 -0.2 0.2 0.4 0.6

 

  

 

 

Fig. 7.4 R vs H at room temperature. Relatively ﬂat peaks indicate a good AP alignment.
The change in resistance is higher than in Fig. 7.4, and the drop in the resistance after the
AP peak is sharp indicating single domain switching of the magnetization.

We conclude that we have successfully prepared a couple of nanopillars with

resistances ~ 1 Q, measured their MRs both at room temperature and 4.2K, and measured

R(I) on one sample at 4.2K.

208

Chapter 8

SUMMARY AND CONCLUSIONS

This dissertation studied several topics in CPP-MR aimed at understanding the
physics underlying GMR and trying to improve the CPP-MR to make it more viable for
devices. The study included testing of the main theories used in the ﬁeld, enhancing CPP-
MR by three different methods, studying Ru for the CPP-MR, studying spin-asymmetric-
scattering at Co interfaces with other metals, and preparing and studying multilayers with
submicron dimensions.

First, we made a systematic study searching for mean-free-path effects that were
claimed to appear as deviations from the two-current series-resistor (2CSR) or Valet-Fert
(VF) models. We have shown in several different ways that the data involved cannot be
explained by mean-free-path effects, but are most likely due to spin-memory-loss in the
multilayers, especially at ferromagnetic/non-magnetic (F/N) metal interfaces. These
results are important both for clarifying the relevant length scales of the CPP-MR and for
fundamental understanding of GMR.

Second, we successfully implemented three ideas to enhance the CPP-MR, although
none produced effects as large as we had hoped. (A) Alloying of Co with Zr impurities
increased the total speciﬁc resistance, AR, but not the more important quantity, the
change in speciﬁc resistance, AAR, with magnetic ﬁeld. (B) Introducing internal
interfaces into the Co layers of Co/Cu multilayers signiﬁcantly increased AR and AAR,

without substantially increasing the total multilayer thickness. (C) Inserting FeMn inside

209

the Cu layers separating Co layers from superconducting Nb layers increased AR and
AAR, especially for multilayers with few repeats.

Third, we determined the CPP-MR parameters of sputtered Ru, an element of present
interest for technology. We showed that the Ru spin-diffusion-length is long enough, and
its resistivity small enough, that they do not limit the potential of Ru for devices. We
also found a negative scattering-anisotropy parameter, 7, at the Co/Ru interface.

Fourth, we presented a new method for high precision measurements of spin-
asymmetric-scattering at C0 interfaces with other metals, alloys, and superconducting
Nb. This method helped us verify some experimentally assumptions. The probe also
provided some intriguing results about the GMR behavior of Ru in proximity with Nb.

F iﬁh, we prepared samples with submicron areas and studied the CPP-MRS at both
4.2K and room temperature. Those samples have room temperature resistances ~ IQ, and
we manipulated their resistances using either an external magnetic ﬁeld or a direct

current.

210

 

APPENDIX

MORE MAGNETORESISTAN CE DATA ON COZR-BASED

MULTILAYERS

This appendix shows additional AR(H) curves for CoZr-based multilayers and two
ﬁgures comparing ARAp with the best ﬁts to the AAR data. The data presented in the
appendix is a completion to data presented in Section 5.2. The ARO-I) curves shown here
are for CoZr/Ag symmetric EBSVs, FeMn/CoZr/N/Py(tpy) asymmetric EBSVs, and
{Py/N/CoZr/N] 10/Py hybrid spin valves. Here, N = Cu or Ag and tpy= 6 or 24 nm. Fig.
A.1 shows AR(H) curves for Cu/FeMn/CoZr/Cu/Py asymmetric EBSVs. A good AP state
is produced for thin CoZr for both Py thicknesses, while the samples with thicker CoZr
do not show such a good AP alignment. Based on Fig. A.1 only a lower bound can be put
on AAR of the asymmetric samples with thick CoZr.

Figs. A2 and A3 show that the AP state is even less controlled in the hybrid samples
than in the exchange-biased samples whether using Cu (Fig. A.2) or Ag (Fig. A.3) as the
nonmagnetic spacer.

Fig. A.4 shows that the AP state is best produced in the Ag-based symmetric EBSVs .
This was also observed in the Cu-based EBSVs ( Fig. 5.7). When using Ag however, the
pinning is weaker than when using Cu, probably due to changes in the quality of FeMn

when grown on Cu from when it is grown on Ag. Finally Fig. A.5 shows the parameters

211

giving the best ﬁts to the AAR data in Ch. 5 give superior estimates to ARM» in

[FeMn/CoZr/Cu/Py] asymmetric EBSVs.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

     

 

 

 

 

 

 

11.9 A 222 ,..Af...,...,.--,,-2
an’comsycwzws) : FONNCOZKZOVCIKZOWB)
17.3 q
A E (A) k1040-2 3
N r .
E 17.7 _ .1
c : :
._ 17.3 : 3
I: 17.5 i 3
< z s
17.4: g .
17.3.1. .1...1...1...1...1 21.2.1...1.L.1...1...1. .4
-200 0 200 400 300 300 .200 0 200 400 300 300
H (00) H (09)
200. I ‘ ' I ‘ j I I V ' ' I ' V Y 7 T V 22.. f v T r W Y Y I v v v I v v v I v f 1 ‘
_ FoMNCoZr(8)ICu(20)IPy(24) 22. 4 _ FMCOMZOVCMZWM _:
20.3 .
A I A 22.2 _ . k1041-2
N ’ N
52°" : E 22 ,
E202 : E 21.3 ;
v v I
m 20 m 21.3: :
< : ( 21.4;- 1
1°" _ “-2.
19.6b 21.4444441114.1...1...‘
.200 0 200 400 300 300
H(Oe)
23_-...-T...,.A...A...-
‘ FeMnlCoZr(30y6u(20)le(24) i
k1041-6

 

 

 

 

 

Fig. A.1. AR vs H for CoZr/Cu/Py asymmetric EBSVs. A good AP state is produced in
samples with thin CoZr, while the samples with thicker CoZr show a poorer AP
alignment. This behavior is common for both tpy = 6 or 24 nm.Horizontal. scale is the
same for all samples to allow direct comparison of the switching ﬁelds.

212

 

 

  

  

 

 

 

 

 

 

 

 

 

       
    

 

 

 

 

 

 

 

 

 

 

 

 

 

 

54 E [incwzoycaraycmzonml Pym -? l [Py(6)ICu(20)ICoZr(15)ICu(20))mI Pym I
. . ,
52 3 a . 1
A ; (A) k1 007.7 1 A k1 008-1 .
N 50 r 1 N .1
CE: ‘8 3 i g :
a. 43 } ~1- :
V : 1 V J
m 44 _' 1 m :
< 42 g g < 1
40:L -i ‘
38..1....1....1....L.4..1..I .1....1.L241....:..'4:If:
.200 -100 0 100 200 -200 .100 0 100 200
H (09) H (Oe)
70 ? [Py(24)lCu(20)/CoZr(7)/Cu(20)]x1olPy(24) 9° §1Py(24)ICu(20)ICozu15)ICu(20)]mI Py(24) j
A ' "A 35 L 1
N 65 C
E E ; (D)
c G 80 r
u- 60 "' .
E : E 75 i
< I < :
55 _ . 70 '
I
: 1
50.,1. .11....1....LL4.414. 65..1LLL4J4...1....14...1..
.200 .1 oo o 100 zoo -200 -100 0 100 200
H (De) H (09)
105_.....A.....,....T.-.....‘
. [wan/Cu(201/ooz:r(30)ICu(20)1,u uI Py(24) .
"“100 ' k1042-6 :
E ’ 1
E 95 i
0: l
< 90
: 1
85 . . 1 . . . . 1 . . . 4 1 4 . . . 14 . . . 14
-200 -100 0 100 200

H (Oe)
Fig. A.2. AR vs H for CoZrlCu/Py hybrid spin valves. Curves are sharper for samples
with thicker CoZr layers indicating the lack of a good AP alignment.

213

85"‘lﬁ"l'fﬁlljll"'l‘
h [PY(2WA9(20)ICOZV(4)IA9(20)1 m I M24)

 

‘ a

on
0
IT!!!
'J
l

.. K1091-1
6%

 

 

 

 

 

 

 

 

 

 

 

 

 

115mewlxrrlwwvﬁwn
C [Py(24)lAg(20)/CoZr(18)IAg(20)] ml Py(24) 3
110 _ k1091-22 .
A 1
NE 105 _— _:
_ 100 _— 1
n: 95 L ﬁ-
< : .
so} _:
' ”.1..- >

 

 

 

 

-600 1400 -200
H (Oe)

Fig. A.3. AR vs H for CoZr/Ag/Py hybrid spin valves. The AP alignment gets quite
uncertain as the CoZr layers become thicker, as seen from the bottom part.

214

 

AgIFeMnlCoZr(6)IAg(20)/CoZr(6)

N

9"

on
I

  

k1 074-2

  

 

 

   

  

.4001 I 4-200. M o I L 4200‘
H (Oe)

400 600

 

l

27-6 — Ag/FeMnICoZr'(2b)iA§(20))Co2ﬁ2'0)

 

27" . k1073-5

.41

27.2 ’

 

AR(mez)
u

 

 

26.8 ’

 

 

26.5WJMHII”51......1‘
--4oo -2oo o 200 400 600

H (Oe)

Fig. A.4. AR vs H for CoZr/Ag symmetric EBSVs. Unlike in the asymmetric and the
hybrid spin valves that contain Py, the AP alignment seems good for both thin and thick
CoZr in the symmetric EBSVs. Pinning is poorer than the Cu-based EBSVs (see F igs. 5.7
and 5.12).

 

215

100 "T Y T I T 1 Y Y I Y 1 I I I I 1 r fl ﬁt 1 I

 

 

 

 

 

 

 

 

 

NA

E .—
C}

I.-

v: _
1r 0 Data-P ‘

< — Coming: Flt 1 j

20 — ----- Best Hybrid Fit 1 _.
0 b1 11 1 11 11 1 11 1111 11 1 #4 11 1 11
0 5 10 15 20 25
t (nm)

OOZI'

 

 

 

 

 

 

 

 

 

NA
£5 _
Ci .
- - i.

a. I 0 Data H id 4

m“ 40 — Commzszlthz -
: ----- Best Hybrid Flt ;
< - 1
20 f -

E

0 _1 1 1 1 1 1 1 1+1 1 1 1 m 1 1 1 1 1 1 1 1 1 1
0 5 10 15 20 25
tCoZr(nm)

Fig. A.5. ARA]: vs (Caz: for [Py(tp,)/Cu(20)/CoZr(t)/Cu(20)]mle(24) hybrid multilayers
with tpy = 6nm (top) or 24 nm (bottom). The ﬁgure shows data, best common ﬁt to all
AARs of CoZr samples (solid line), and best ﬁt to AARs of the hybrid samples alone
(dotted line). The offset is most likely because the resistivity of Py is less than the value
used in the ﬁts.

216

References

[1] MN. Baibich, J.M Broto, A. Fert, F. Nguyen Van Dau, F. Petroff, P. Etienne, G.
Creuzet, A. Friederich, and J. Chzelas, Phys. Rev. Lett. 61, 2472 (1988).

[2] G. Binash, P. Grunberg, F. Saurenbach, and W. Zinn, Phys. Rev. B 39, 4828 (1989).

[3] RM. Levy, Solid State Physics, H. Ehrenreich and D. Tumbull, Eds., Academic
Press, NY, 47, 367 (1994).

[4] M.A.M. Gijs and G.E.W. Bauer, Adv. in Phys. 46, 285 (1997).

[5] J. Bass and WP. Pratt Jr., J. Magn. Magn. Mater. 200, 274 (1999).

[6] J .-Ph. Ansermet, J. Phys.: Condens. Matter 10 223 (1998).

[7] A. Barthelemy, A. Fert, R. Morel, and L. Steren, Physics World 7, 34 (1994).
[8] See IBM web page--www.ibm.com.

[9] See Seagate web page-~www.seagate.com.

[10] P. Grunberg, R. Schreiber, Y. Pang, M.B. Brodsky, and H. Sowers, Phys. Rev. Lett.
57, 2442 (1986).

[11] D. H. Mosca, F. Petroff, A. Fert, P. A. Schroeder, W. P. Pratt, Jr., R. Loloee, and S.
Lequien, J. Magn. Magn. Mater. 94 L1 (1991).

[12] SS. Parkin, N. More, and KP. Roche, Phys. Rev. Lett. 64, 2304 (1990).
[13] SS. Parkin, R. Bhadra, and KP. Roche, Phys. Rev. Lett. 66, 2152 (1991).
[14] MD. Stiles, J. Magn. Magn. Mater. 200, 322 (1999).

[15] J. Slonczewski, J. Magn. Magn. Mater. 159, L1 (1996).

[16] L. Berger, Phys. Rev. B 54, 9353 (1996).

[17] M. Tsoi, A.G.M. Jansen, J. Bass, W.-C. Chiang, M. Seck, V. Tsoi, and P. Wyder,
Phys. Rev. Lett. 80, 4281 (1998).

[18] M. Tsoi, A. Jansen, J. Bass, W.C. Chiang, V. Tsoi, and P. Wyder, Nature (London)
406, 6791 (2000).

[19] EB. Myers, D.C. Ralph, J .A. Katine, R.N. Louie, and RA. Buhrman, Science 285,
867 (1999).

217

[20] J. Grollier, V. Cros, A. Hamzic, J .M. George, H. Jaﬂ‘res, A. Fert, G. Faini, J. Ben
Youssef, and H. Le Gall, Appl. Phys. Lett. 78, 3663 (2001).

[21] K. Bussmann, G.A. Prinz, R. Bass, J.-G Zhu, Appl. Phys. Lett. 78, 2029 (2001).
[22] Gary A. Prinz, Phys. Today, 48, 58 (1995).

[23] Gary A. Prinz, Science, 282, 1660 (1998).

[24] R&D Magazine, page 14, July 1999.

[25] WP. Pratt Jr., S.F. Lee, J .M. Slaughter, R. Loloee, P.A. Schroeder, and J. Bass,
Phys. Rev. Lett. 66, 3060 (1991).

[26] S. Zhang, and RM. Levy, J. Appl. Phys. 69, 4786 (1991).

[27] SF. Lee, W.P. Pratt, Jr., Q. Yang, P. Holoday, R. Loloee, P.A. Schroeder, and J.
Bass, J. Magn. Magn. Mater. 118, L1 (1993).

[28] SF. Lee, Q. Yang, P. Holoday, R. Loloee, J .H. Hetherington, S. Mahmood, B.
Ikegami, K. Vigen, L.L. Henry, P.A. Schroeder, W.P. Pratt, Jr., and J. Bass, Phys.
Rev. B 52, 15426 (1995).

[29] C. Fierz, S.F. Lee, J. Bass, W.P. Pratt Jr., and RA. Schroeder, J. Phys. Cond. Matt.
2, 9701 (1990).

[30] M.A.M. Gijs, S.K. Lenczowski, and J .B. Giesbers, Phys. Rev. Lett. 70, 3343 (1993).
[31] SK. Lenczowski, et al., J. Appl. Phys. 75, 5154 (1994).

[32] J. Spallas, Y. Huai, S. Vernon, B. Fuchs, B. Law, lD.R. Kania, D. Kroes, M. Thomas,
D. Okane, Z.C.H. Tan, IEEE Trans. Magn., 32, 4710 (1996).

[33] W. Vavra, S.F. Cheng, Antia Fink, J .J . Krebs, and G.A. Prinz, Appl. Phys. Lett. 66,
2579 (1995).

[34] A. Blondel, J .P. Meier, B. Doudin, and J .-Ph Ansermet, Appl. Phys. Lett. 65, 3019
(1994)

[35] B. Doudin, A. Blondel, and J .-Ph. Ansermet, J. Appl. Phys. 79, 6090 (1996).

[36] L. Piraux, J .M. George, J .F. Despres, C.Leroy, E. Ferain, R. Legras, K. Ounadjela,
and A. Fert, Appl. Phys. Lett. 65, 2484 ( 1994).

218

[37] L. Piraux, S. Dubois, C. Marchal, J .M. Beuken, L. Filipozzi, J .F Despres, K.
Ounadjela, and A. Fert, J. Magn. Magn. Mater. 156, 317 (1996).

[38] K. Liu, K. Nagodawithana, P.C. Searson, and CL. Chhien, Phys. Rev. B 51, 7381
(1995).

[39] A. Fert and L. Piraux, J. Magn. Magn. Mater. 200, 338 (1999).

[40] T. Ono and T. Shinjo, J. Phys. Soc. Jpn. 64, 363 (1994).

[41] RM. Levy, S. Zhang, T.Ono, and T.Shinjo, Phys. Rev. B 52, 16, 049 (1995).

[42] M.A.M. Gijs, M.T. Johnson, A. Reinders, P.E. Huisman, R.J.M. van de Veedronk,
(81.19;?)Lenczowski, and R.M.J. van Gansewinkel , Appl. Phys. Lett. 66, 1839

[43] J. Bass, W.P. Pratt, and RA. Schroeder, Comments Cond. Matt. Phys. 18, 223
(1998)

[44] Shang—Fan Lee, Ph.D. Dissertation, Michigan State University, 1994.
[45] Qing Yang, Ph.D. Dissertation, Michigan State University, 1995.

[46] LA. Campbell and A. F ert, E.P. Wohlfarth (Ed), Ferromagnetic materials, Vol. 3,
North-Holland, Amsterdam, 1982, p. 751, Chapter 9.

[47] T. Valet and A. Fert, Phys. Rev. B 48, 7099 (1993).
[48] RC. van Son, H. van Kempen, and P. Wyder, Phys. Rev. Lett. 58, 2271 (1987).
[49] M. Johnson and RH. Silsbee, Phys. Rev. B 37, 5312 (1988).

[50] Q. Yang, P. Holoday, S.-F. Lee, L.L. Henry, R. Loloee, P.A. Schroeder, W.P. Pratt
Jr., and J. Bass, Phys. Rev. Let. 72, 3274 (1994).

[51] J. Nogués and Ivan K. Schuller, J. Magn. Magn. Mater. 192, 203 (1999).

[52] B. Dieny, V. Speriosu, S. Metin, S.S.P. Parkin, B.A. Gurney, P. Baumgart, and DR.
Wilhoit, J. Appl. Phys. 69, 4774 (1991).

[53] K. Nagasaka, Y. Seyama, L. Varga, Y. Shimizu, and A. Tanaka, J. Appl. Phys. 89,
6943 (2001).

[54] Luc Thomas, Mahesh G. Samant, and Stuart S.P. Parkin, Phys. Rev. Lett. 84, 1816
(2000).

219

[55] Gary A. Prinz, J. Magn. Magn. Mater. 200, 57 (1999).
[56] K]. Kirk, Contemporary Physics 41, 61 (2000).

[57] J .M. Daughton, J. Brown, B. Chen, R. Beech, A. Pohm, and W. Kude, IEEE Trans.
Magn. 30, 4608 (1994).

[58] Robert Slater, Ph.D. Dissertation, Michigan State University, 2001.
[59] WA. Reed and E. Fawcett, Phys. Rev. 136, A422 (1964).

[60] SJ. Jin T.H. Tiefel, M. McCormack, R.A. F astnacht, R. Ramesh, L.H. Chen,
Science, 264 413 (1994).

[61] AP. Ramirez, J. Phys: Cond. Matt. 9, 8171 (1997).

[62] Abstracts F D-Ol , FD-02, and FD-OS, 46th Annual Conf. On Magn. & Magn. Mat.,
Seattle, Wash,, Nov. 12-16 (2001).

[63] J .S. Moodera, L.R. Kinder, T.M. Wong, and R. Meservey, Phys. Rev. Lett. 74, 3273
(1995)

[64] See, e.g. www.motorola.com (Search MRAM); New York Times, Jan. 18, 2001, Pg.
D6.

[65] DD. Awschalom and N. Samarth, J. Magn. Magn Mater. 200, 130 (1999).

[66] J .A. Gupta, X. Peng, A.P. Alivisatos, D.D. Awschalom, Phys. Rev. B 59, R1042]
(1999)

[67] SA. Wolf, D.D. Awschalom, R.A. Buhrman, J .M. Daughton, S. von Molnar, M.L.
Roukes, A.Y. Chtchelkanova, and D.M. Treger, Science 294 1488 (2001).

[68] J. Bass, P.A. Schroeder, W. P. Pratt Jr., S.F. Lee, Q. Yang, P. Holody, L.L. Henry,
and R. Loloee, Mat. Sci. & Eng. B 31, 77 (1995).

[69] Richard Magno and Joe H. Pifer, Phys. Rev. B 10, 3727 (1974).

[70] RH. Silsbee, A. Janossy, and P. Monod, Phys. Rev. B 19, 4382 (1979).
[71] P. Monod and S. Schultz, J. Phys. (Paris) 43, 393 (1982).

[72] A. Fert, J .-L. Duvail, and T. Valet, Phys. Rev. B 52, 6513 (1995).

[73] W. Park, D.V. Baxter, S. Steenwyk, l. Moraru, W.P. Pratt Jr., and J. Bass, Phys.
Rev. B 62, 1178 (2000).

220

[74] SD. Steenwyk, S.Y. Hsu, R. Loloee, J. Bass, and W.P. Pratt, Jr., J. Magn. Magn.
Mater. 170, L1 (1997).

[75] W.P. Pratt, Jr., Q. Yang, L.L. Henry, P. Holody, W.-C.Chiang, P.A. Schroeder, and
J. Bass, J. Appl. Phys. 79, 5811 (1996).

[76] L. Piraux, S. Dubois, A. Fert, and L. Belliard, Euro. Phys. J. B 4, 413 (1998).

[77] A.C. Reilly, W.-C. Chiang, W. Park, S.Y. Hsu, R. Loloee, S. Steenwyk, W.P. Pratt
Jr., and J. Bass, IEEE Trans. Magn. 34, 933 (1998).

[78] C. Vouille, A. F ert, A. Barthelemy, S.Y. Hsu, R. Loloee, and RA. Schroeder, J.
Appl. Phys. 81, 4573 (1997).

[79] C. Vouille, A.Barthelemy, F. Elokan Mpondo, A. Fert, P.A. Schroeder, S.Y. Hsu, A.
Reilly, and R. Loloee, Phys. Rev. B 60, 6710 (1999).

[80] PA. Schroeder, P. Holody, R. Loloee, C. Vouille, A. Barthelemy, A. Fert, and S.Y.
Hsu, J. Magn. Magn. Mater. 177-181, 1464 (1998).

[81] W. Park, R. Loloee, J .A. Caballero, W.P. Pratt, Jr., P.A. Schroeder, and J. Bass, J.
Appl. Phys. 85, 4542 ( 1999).

[82] A.C. Reilly, W. Park, R. Slater, B. Ouaglal, R. Loloee, W.P. Pratt Jr., and J. Bass, J.
Magn. Magn. Mat. 195, L269 (1999).

[83] W.-C. Chiang, Q. Yang, W.P. Pratt, Jr., R. Loloee, and J. Bass, J. Appl. Phys. 81,
4570 (1997).

[84] D. Bozec, M.J. Walker, B.J. Hickey, M.A. Howson, and Nathan Wiser, Phys. Rev. B
60, 3037 (1999).

[85] J .Y. Gu, S.D. Steenwyk, A.C. Reilly, W. Park, R. Loloee, J. Bass, and W.P. Pratt Jr.,
J. Appl. Phys. 87, 4831 (2000).

[86] C. Heide, Phys. Rev. Lett. 87, 197201 (2001).
[87] S. Zhang, P.M. Levy, and A. Fert, Phys. Rev. Lett. 88, 236601 (2002)
[88] X. Waintal, E. Myers, P. Bouwer, and D. Ralph, Phys. Rev. B 62, 12317 (2000).

[89] D. Bozec, M.A. Howson, B.J. Hickey, S. Shatz, N. Wiser, E.Y. Tsymbal, and D.G.
Pettifor , Phys. Rev. Lett. 85, 1314 (2000).

[90] L.J. van der Pauw, Philips Research Reports, 13, l (195 8).

221

[91] J. Bass, in Metals, Electronic Transport Phenomena, edited by K.H. Hellwege and
J .L. Olson, Landolt-Bomstein, New Series, Group 3, Vol. 15, (Springer-Verlag,
Berlin, 1982).

[92] J .J . Krebs, W. Vavra, G.A. Prinz, S.F. Cheng, and A. Fink, J. Appl. Phys. 79, 6084
(1996).

[93] A. Fert and s.r=. Lee, Phys. Rev. B 53, 6554 (1996).

[94] KM. Schep, J .B.A.N. van Hoof, P.J. Kelly, G.E.W. Bauer, and J .E. Inglesﬁeld,
Phys. Rev. B 56, 10805 (1997).

[95] K. Xia, P.J. Kelly, G.E.W. Bauer, 1. Turek, J. Kudmovsky, and V. Drchal, Phys.
Rev. B 63, 64407 (2001).

[96] MD. Stiles and DR. Penn, Phys. Rev. B 61, 3200 (2000).

[97] W.H. Butler, X.-G. Zhang, and J. M. MacLaren, J. Appl. Phys. 87, 5173 (2000).
[98] A. Shpiro and PM. Levy, Phys. Rev. B 63, 14419 (2000).

[99] Y. Tsymbal and D.G. Pettifor, Solid State Physics 56, 113 (2001).

[100] J .A. Borchers, J .A. Dura, J. Unguris, D. Tulchinsky, M.H. Kelley, C.F. Majkrzak,
S.Y. Hsu, R. Loloee, W.P. Pratt Jr., and J. Bass, Phys. Rev. Lett. 82, 2796 (1999).

[101] or. Majkrzak, Physica B 221, 342 (1996).

[102] S.K. Sinha, in Neutron Scattering in Materials Science 11, Eds. D.A. Neumann,
T.P. Russell, and B.J. Wuensch, MRS Symp. Proc. 376, p 175 (1995).

[103] J.A. Borchers, J .A. Dura, C.F. Majkrzak, S.Y. Hsu, R. Loloee, W.P. Pratt, Jr., and
J. Bass, Physica B 283, 162 (2000).

[104] J .A. Borchers, P.M. Gehring, R.W. Erwin, J .F . Ankner, C.F. Majkrzak, T.L.
Hylton, K.R. Coffey, M.A. Parker, and J .K. Howard, Phys. Rev. B 54, 9870 (1996).

[105] K. Eid, D. Portner, R. Loloee, W.P. Pratt Jr, and J. Bass, J. Magn. Magn. Mat. 224
L205 (2001).

[106] K. Eid, M. Tsoi, D. Portner, R. Loloee, W.P. Pratt Jr., and J. Bass, J. Magn. Magn.
Mat. 240, 171 (2002).

[107] A.C. Reilly, W.-C. Chiang, W. Park, S.Y. Hsu, R. Loloee, J. Bass, and W.P. Pratt
Jr., IEEE Trans. Magn. 34, 939 (1998).

222

[108] K. Eid, D. Portner, J. Borchers, R. Loloee, M. AlHaj Darwish, M. Tsoi, H.Kurt,
K.V. O’Donavan, W.P. Pratt Jr., and J. Bass, Phys. Rev. B 65, 054424 (2002).

[109] G. Suran, F. Machizaud, and M. Naili, Rhys. Rev. B 47, 15007 (1993).
[110] R. Mohammed Idrus and P J Grundy, J. Phys. D: Appl. Phys. 19, 1245 (1986).

[111] K. Bouziane, J. Ben Youssef, M. El Harfaoui, O. Koshkina, H. Le Gall, J. M.
Desvignes, M. El Yamani, and A. Fert, J. Magn. Magn. Mat. 165, 284 (1997).

[112] K. Bouziane, M. Mamor, and M. Maaza, Solid State Communications 107, 107
(1998).

[113] M. El Harfaoui, M. Paris, A. Qachaou, J. Ben Youssef, H. Le Gall, and D.
Meziane. Mtalsi, J. Magn. Magn. Mat. 223, 81 (2001).

[114] S. Wang, J .I. Guzman, and M.H. Kryder, J. Appl. Phys. 67, 5114 (1990).

[115] E. Jedryka, M. Wojcik, S. Nadolski, T. Stobiecki, and M. Czapkiewicz, J. Magn.
Magn. Mat. 156, 38 (1996).

[116] W.P. Pratt, Jr., S.D. Steenwyk, S.Y Hsu, W.-C. Chiang, A.C. Schaefer, R. Loloee,
and J. Bass, IEEE Trans. Mag. 33, 3305 (1997).

[117] J. Bass, K. Eid, R. Loloee, and W. P. Pratt Jr., Bull. Am. Phys. Soc. 45, 223 (2000).

[118] L.L. Henry, Q. Yang, W.-C. Chiang, P. Holody, R. Loloee, W. P. Pratt, Jr., and
J. Bass, Phys. Rev. B 54, 12336 (1996).

[119] M. Pinarbasi, S. Metin, H. Gill, and M. Parker, J. Appl. Phys. 87, 5714 (2000).

[120] J. Chen, S.N. Mao, J. Femandez-de-Castro, T.S. Choy, and S. Hershﬁeld, IEEE
Trans. Mag. 36, 2885 (2000).

[121] K. Meguro, H. Hoshiya, K. Watanabe, Y. Hamakawa, M. Fuyama, H. Fukui, IEEE
Trans. Mag. 35, 2925 (1999).

[122] A. Dinia and K Rahmouni, Phys. Lett. A 264, 482 (2000).

[123] K. Rahmouni, A. Dinia, D. Stoeffler, K. Ounadjela, H.A.M. Van den Berg, and H.
Rakoto, Phys. Rev. B 59, 9475 (1999).

[124] K. Eid, R. Fonck, M. AlHaj Darwish, W.P. Pratt, Jr., and J. Bass, J. Appl. Phys.
91, 8102 (2002).

223

[125] M. Wojcik, E. Jedryka, S. Nadolski, D. Kubinski, M. Parsons, and H. Holloway,
Acta Physica Polonica A 97, 551 (2000).

[126] W.J. Antel, Jr., F. Perjeru, and GR. Harp, Phys. Rev. Lett. 83, 1439 (1999).

[127] B.Y. Wong, C. Mitsumata, S. Prakash, D.E. Laughlin, and T. Kobayashi, J. Appl.
Phys. 79, 7897 (1996).

[128] F. Taddei, S. Sanvito, J .H. Jefferson, and OJ. Lambert, Phys. Rev. Lett. 82, 4938
(1999)

[129] F. Taddei, S. Sanvito, and OJ. Lambert, Phys. Rev. B 63, 012404 (2001).

224