. .noqv

 

 

   

:. . 2......
p .

 

 

n... c . .

 

 

 

 

 

 

  

 

 

 
 
 

 

 

 

 

 

 

 

 
 
 

,,. x.‘ . , ‘.
. : . A ;r .
‘ ; ., L. A ,. , (
; . :J... ,; .1 ,... ...
. . H, . U. :wv,_.I<‘ . .
,..“...:...f. . Z... .. ‘
. ‘ my... what." .~m.?.xu
, 3 .i.’

 

|'

THETriS

400?;

 

_ LIBRARY
M'Chigan State
Unlversity

 

 

 

This is to certify that the

dissertation entitled

Perpendicular Giant Magnetoresistance in Thin Films
with Microfabricated Superconducting Top Contacts

presented by

ROBERT IIDUGLAS SLATER

has been accepted towards fulfillment
of the requirements for

Ph D degree in ths ics

 

WW flW

Major professor [

W Pratt

Date _lQ125./_Ql__

MS U it an Affirmative Action/Equal Opportunity Institution 0- 12771

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 cJCIRCIDateDuepss-sz

PERPENDICULAR GIANT MAGNETORESISTANCE IN THIN FILMS WITH
MICROFABRICATED SUPERCONDUCTING TOP CONTACTS

By
ROBERT DOUGLAS SLATER

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
2001

ABSTRACT
PERPENDICULAR GIANT MAGNETORESISTANCE IN THIN FILMS WITH
MICROFABRICATED SUPERCONDUCTING TOP CONTACTS
By
ROBERT D. SLATER

A new method to measure the position of a domain wall in a thin film is proposed.
By fabricating micron-size superconducting top contacts to produce micron size ar-
eas of perpendicular current flow in exchange bias spin valves (EBSVs) of C091Fe9
and Ni34Fe16, the position of a domain wall can be measured using the giant (G)
magnetoresistance (MR) effect.

Exchange bias spin valves have the form: antiferromagnet (AF) /ferromagnet
(F) /non-magnetic (N) /ferromaget (F). For the purposes of this study the AF was
Fe5oMn5o , F was C091Fe9 or Ni84Fe16, and N was Cu. The AF layer serves to fix the
magnetization of the adjoining F layer in one direction. The other F layer is then
free to reverse with applied magnetic field. The GMR effect produces a large change
in resistance when the F layers magnetizations change from parallel to anti-parallel.
As the free F layer reverses, domain walls will move in that layer. If a domain wall
moves past a measuring contact, the change in resistance is related to the position of
the domain wall and can be used to measure its relative location.

To demonstrate this, a series of 1 x 10 mm EBSVs were patterned with micron
size superconducting top contacts of Nb. The results of transport studies showed that
shrinking the area of current flow had no affect on the transport properties of the
samples. Then a new series of EBSVS were shaped into long wires with dimensions
of order 1 x 20 pm by electron-beam and photolithography. Micron size top contacts
were placed on top of the lithographed features and evidence of domain wall trapping

was observed in the MR of the samples.

ACKNOWLEDGEMENTS

I would like to take this space to acknowledge those whose time and effort has
contributed significantly to this dissertation.

First, I would like to thank my advisor Prof. William P. Pratt, Jr. The time
spent in learning the craft of experiential physics was invaluable. In addition, Prof.
Pratt shared the many lessons of being a professional in the world of physics beyond
the basics, including producing professional presentations and technical writing.

Second, I would like to acknowledge the help and support of Dr. Reza Loloee,
the “Man behind the curtain” within the GMR group. Without the support of Dr.
Loloee, much of the research conducted within the GMR group would grind to a halt.
His knowledge and assistance to the users of all the critical systems (and sometimes
how to deal with life in general) is freely given and always given with smile.

There were a variety of people who were very important to the development of this
project. Prof. Anne Reilly and Dr. Juan Caballero were two postdoctoral researchers
with whom I worked very closely on the early stages of this project. Secondly I would
also like to thank Prof. Norman Birge and Prof. Jack Bass with whom I have had
countless discussions about a variety of topics related to this research. Dr. Baokang
Bi, manager of the Keck Microfabrication Facility, has provided guidance for many
of the fabrication techniques.

Next I would like to thank the remaining members of the “GMR Group” with
whom I have worked with during my time at MSU: Khalid Eid, Dr. Jiyeong Gu,
Mustafa Al—Haj Darwish, Huseyin Kurt, Antonio Zambano, Dr. Konstantine Niko-
laev, Dr. Wanjun Park, and Dr. Maxim Tsoi.

I would also like to thank of those who helped me keep my sanity and provide
moments of levity away from work. Fellow Packer supporters Brian Sharpee and Dr.
Barton Pritzl have shared many football Sundays in misery and joy. My partners

in crime in the physics department, Dr. Charles Moreau, Dr. Peter Peterson, Aaron

iii

Lacluyzé, Dr. FTédéric Pierre, Mike Davis, AJ Radar, Dr. Eric Tryggestad, and Chris
Hanley have been indispensable friends. Finally the LAN party crowd has been my
source of relief and relaxation for longer than I can remember. Special thanks to
Defiant Victim, Mr. Armageddon, Mr. Book, MikeDog, Maxie, Gold, Heywood,
‘The Doc,’ Slo-Tek, Get-Funk, Digi-Guy, Fty Me2k, Lunchbox, LeFrank, and Happy
Fun Ball: the best group of l33t-®n|1|\|3.vv4n|<3r5 there is.

Finally I would like to thank my family who have been a source of constant support
throughout my education. I would like to especially mention my wife Jennifer and
my son Jacob who have been at the center of the maelstrom these last several years.

Your patience and support have meant more than words can express.

iv

For my father and mother

It was a dark and stormy night...

Table of Contents

LIST OF TABLES

LIST OF FIGURES

1 Introduction
1.1 Magnetic Domains ............................
1.1.1 Local Energy Contributions ...................
1.1.2 Non-local Magnetic Energy Contributions ...........
1.1.3 Interaction of Energy Contributions ...............
1.1.4 Shape Effects ...........................
1.1.5 Domain Walls ...........................
1.2 Background of GMR ...........................
1.3 CPP Measurements ............................

1.4 Theory of CPP—GMR ...........................

2 Sample Fabrication
2.1 Introduction ................................
2.2 Types of Samples .............................
2.2.1 Type—I Samples ..........................
2.2.2 Type-II Samples .........................

2.2.3 Type-III samples .........................

vi

ix

12

13

16

22

22

22

24

25

26

2.2.4 Type—IV samples ......................... 28

2.3 Procedures ................................. 28
2.3.1 Substrate Cleaning ........................ 28

2.3.2 Sputtering ............................. 29

2.4 Lithographic Procedures ......................... 32
2.4. 1 Photolithography ......................... 32

2.4.2 Electron Beam Lithography ................... 34

2.4.3 Etching .............................. 38

2.4.4 810 Evaporation ......................... 40

2.5 Measurement Equipment ......................... 45
2.5.1 SQUID Based Null Circuit Voltage Measurements ....... 46

2.5.2 Nanovolt measurements ..................... 48

2.5.3 Area Measurements ........................ 49

3 CPP-Multilayers with Micron Size Top Contacts 53
3.1 Introduction ................................ 53
3.2 Initial Work and Tests .......................... 54
3.2.1 Lithography Test ......................... 54

3.2.2 Current Dependence ....................... 56

3.3 GMR Ttansport .............................. 59

4 CPP-Domain Wall Trapping 65

vii

4.1 Initial Trials ................................ 66

4.2 Improved Lithography .......................... 70
4.3 E—beam Lithography Samples ...................... 76
4.4 Results ................................... 80
5 Conclusions 81
LIST OF REFERENCES 84

viii

2.1

2.2

2.3

List of Tables

The side dimension of each square listed for the features on the pho-
tomask that is shown in figure 2.8. All units are in microns. ..... 35

The side dimension of each square listed for the features on photomask
2 that is shown in figure 2.9. All units are in microns. W40 and W20
refer to the wire geometries of 40 microns length and 20 microns length,
respectively, shown in Figure 2.3 ..................... 35

Ion milling rates in A / s for various metals at 10 mA and 20 mA current.
For 10 mA, The Ar flow was 3 sccm and for 20 mA, 6 seem. ..... 40

1.1

1.2

1.3

1.4

1.5

List of Figures

The basis of this thesis research. A domain wall moves in a ferro-
magnetic layer (F) that is free to reverse its magnetization (Top F
layer) . As the wall moves past the top measuring contact the resis-
tance changes in proportion to the location of the wall. The resistance
change is from GMR where parallel F-layer magnetizations will give a
low resistance and antiparallel magnetizations give a high resistance. .

Magnetic hysteresis of a ferromagnet. The rapid changes in magneti-
zation are due to growth of favorably oriented domains by domain wall
propagation. The slower changes near saturation are due to coherent
magnetization rotation of domains. ...................

Domain formation in a single crystal. (a) A single crystal that is uni-
formly magnetized. (b) Dividing the crystal into two magnetic domains
reduces the stray field energy. (c) Further division into domains will
reduce the stray field energy until the energy to create a new domain
is greater than the reduction to the stray field. (d) Closure domains
can reduce the stray field energy to zero, but can contribute to the
anisotr0py energy ..............................

Shape anisotropy is used to trap a domain wall in patterned wires.
In (a) through (c) a domain wall starts in the wider portion of the
wall. (a) As an applied magnetic field is increased in the up direction
the wall will move towards the narrow portion of the wire. (b) The
applied magnetic field does not give a large enough Zeeman energy
contribution to allow the wall to enter the narrow portion of the wire.
The narrow portion of the wire has much larger stray field energy due
to surface and edge effects. (c) The magnetic field is increased enough
for the wall to enter the narrow portion of the wire. (d) through (f)
repeat this sequence with a narrow neck or notch region. The effect is
the same, except once the wall gains enough energy to leave the initial
constriction, it most likely has enough energy to move entirely to the
end of the wire. ..............................

The two most common types of domain walls. The Neel wall has the
magnetization rotating in x—z plane. The Bloch wall has the magneti-
zation rotate vertically in the z-y plane ..................

11

1.6

1.7

1.8

1.9

2.1

2.2

An explanation of the basic GMR effect. On the left, when the mag-
netizations (white arrows) of the magnetic layers, F, are antiparallel,
both the spin up and spin down electrons have equal resistance paths.
When the magnetizations are parallel, the spin up electrons (in this
case) have a lower resistance path compared with spin down electrons.
This leads to a much lower overall resistance ...............

A diagram of the CIP geometry vs. the CPP geometry. In the CIP
geometry the current (black arrow) flows parallel to the multilayer in-
terfaces. In the CPP geometry the current flows perpendicular the
layer interfaces (The diagram dimensions are not to scale, typical layer
thickness (vertical dimension here) is 1 - 10 nm, while the width and
depth of the samples (horizontal and perpendicular out of plane di-
mensions) are 0.1 to 10 mm) . .....................

A simplified diagram of the 4s and 3d energy bands for Co. The up
and down arrows represent the bands for the up and down electrons,
respectively. Since the 3dT band fills with up electrons first, at the
fermi level, the density of states for down electrons is much larger
giving a higher resistivity. ........................

A diagram of the two current series resistor model for the AP state.
The current is divided into two channels: spin up channel (top) and
spin down channel (bottom). One adds the series resistances for each
channel and then combines the series resistances as two resistances in
parallel to produce the overall resistance for both the AP states. . . .

An overview of the fabrication process. A side profile is provided on
the left (only half of the lithographic features are shown to conserve
space), and a complete top view is provided on the right. The dimen-
sions of the system are not to scale. A multilayer is sputtered onto a
cleaned substrate (A). A lithographic pattern / mask is placed onto the
multilayer (B). A portion of the multilayer is etched away, an insult-
ing layer deposited and the lithographic pattern removed (C). The top
contacts are sputtered onto the sample (D). ..............

An overview of the different types of samples. Type-I has contact made
to a Au capping layer. Type-II has a partially etched Au/ Nb/ Au cap.
Type-III has the entire multilayer etched. Type-IV has a separate
contact geometry on top of a patterned multilayer ...........

13

14

17

19

23

25

2.3

2.4

2.5

2.6

2.7

2.8

2.9

A diagram of the mask for fabricated wires with a neck. There were
two sizes of wire designed. Mask b is exactly a factor of two smaller
than a. Each is symmetric about the notch region ............

The difference between a via and a proud column. A via is a trench
or window in the insulating layer through which contact with the top
superconductor (dashed line) is made. A proud column sticks above
the insulating layer. Proud columns become necessary when the feature
size is below 2 pm. ............................

The layout of the mask system. The rotating mask (inner circle) is
rotated to expose a sample to the sputtering plasma. A shadow mask
(not shown) over the substrate defines the actual pattern. The 12 o
clock and 6 o clock positions correspond to a closed position while the
3 o clock and 9 o clock position will expose the sample. The rotating
mask is aligned by lining up the pins of the rotating mask to be in line
with the bridges holding down the samples. The mask here is shown
in the 9 o clock position while the others are represented with dashed
lines .....................................

A exploded view of the sample holder. The bridge clamps the heat
sink / substrate/ shadow mask together. The assembly is attached with
screws to the sample holder through the bridge. Each sample holder
is designed to hold two assemblies. ...................

A layout of the mask system to deposit top Nb leads. (A) The multi-
layer with patterned mask features (exaggerated size, in black). Each
feature is spaced so that when the far right opening in the shadow mask
(B) is lined up with the protrusion in the multilayer (A), the individual
exposed contact areas will each fall within a single Nb contact (C)

A diagram of the layout and dimensions of the first photomask. A
single array of features was arranged as shown. The dimensions of the
features are listed in Table 2.1 ......................

A diagram of the layout and dimensions of the second photomask. On
this mask, the array was repeated multiple times on the mask. The
dimensions of the features are listed in Table 2.2 ............

xii

27

27

30

30

32

35

36

2.10 The bilayer of PMMA and Copolymer after evaporation. The devel-

2.11

2.12

2.13

2.14

2.15

opment removes the copolymer more than the PMMA causing the un-
dercut. The difference in removal is due the copolymer’s stronger sen-
sitivity to the backseattered electrons from the substrate. This allows
the deposited A1 feature, patterned from the PMMA, to be isolated
from the rest of the evaporated material. The bilayer will be removed

in the next step by Acetone, and only the Al on the sample will remain.

The setup of the small chamber. The samples can be rotated to the
desired position on the sample plate. A shutter, independent of the
sample plate, can also be rotated to an open or closed position de-
pending on source to be run. The magnetic arm for loading samples
through a load lock is in the upper right .................

The tantalum boat for SiO evaporation. The SiO is placed in the
loading area. The entire boat is heated using a current source. When
the SiO begins to evaporate it must flow through a series of baffles
that prevent spitting. The evaporated SiO then escapes vertically (side
view) toward the sample. ........................

SiOx deposition. The insulator is evaporated at a 15 degree angle with
respect to the sample—surface normal. This and the rotation help to
eliminate pinholes. ............................

Measurement setup and geometry of the sample. Input current and
voltage leads are attached to the contact to be measured (#1). Output-
current and second-voltage leads are attached to #2 and #3 contacts,
respectively. Current flows between contacts #1 and #2 (via solid
arrows), and the voltage contact at #3 measures the potential of the
bottom superconductor. .........................

The SQUID null feedback circuit. An RF SQUID system (SQUID,
inductor, RF LC circuit, and semiconductor electronics) balances the
circuit so no current flows from point A to point B. The voltage drop
across both resistors is equal in this balanced state. From Ohm’s law,
one can then solve for R1 and in terms of I A, I p3 and RRef. I A is applied
to the system and is known. This applied current causes the SQUID
system to go out of balance. The feedback current from the SQUID
system, IFB, is measured by a resistor in series with the reference
resistor, RRef. RRef is a precision resistor with a known resistance.

xiii

38

41

43

44

46

47

2.16

2.17

2.18

3.1

3.2

The circuit layout of the nanovolt probe. A Keithly 2400 current source
(1) is connected to a Keithly 7011 multiplexer card (2) with two elec-
tronically isolated banks of connections that allow each wire coming
from the probe to be selected as either high or low in combination with
any other wire. A Keithly 2182 nanovoltmeter (3) is connected to the
voltage leads. However, the Keithly 7168 multiplexer card can only
support one voltage connection per voltage lead. Thus half the voltage
leads are hardwired as V+ and the others are V— (4a,4b) The wires are
then connected to the probe (5,6; see Figure 2.18). Finally the leads
are taken to the low temperature portion of the measurement in the
measuring probe and connected to the sample ..............

The modular measuring probe used for nanovolt measurements. The
magnet, magnet leads, and persistent switch connections are self-con-
tained in a sheath that connects to the I-V portion of the probe via a
quick connect. This sheath surrounds the probe and the connection is
placed at a pre—determined location to put the sample in the middle of
the magnet. The measurement portion of the probe can be used with
or without the magnet portion. .....................

A layout of the external connections to the measuring probe (one side
only, see items 5 8c 6 in Figure 2.16). For both the current and voltage
leads, each wire is connected to a grounding system individually. The
grounding system is also wired to a switch which chooses between a
direct ground or grounding through a 1 M0 resistor to minimize cur-
rents induced by static charges. The shield for each cable is directly
connected to ground. Outside the probe, all current leads are inside a
single shielded cable and each pair of voltage leads has its own shielded
cable ....................................

A plot of the specific resistance for single layer C091Fe9 samples ver-
sus the C091Fe9 layer thickness. The circles represent samples with a
C1 cap that underwent a cleaning procedure without ultrasonic agi-
tation. The triangles represent samples that did have ultrasonic ag-
itation (C1 cap). The squares represent a Cu cap with the original
(non-ultrasonic) cleaning. ........................

An explanation of vias (black areas with white arrows). The portion of
the Nb (white, black for vias) where the current (block arrows) flows
from the Nb that lies on the SiO (grey) to the portion of Nb in contact
with the multilayer. The current only flows along the edges because
the penetration depth of the superconducting Nb is ~100 nm.

xiv

50

51

52

55

57

3.3

3.4

3.5

3.6

3.7

3.8

3.9

3.10

The current dependence of several of the smallest contact areas for
C091Feg samples. The right hand vertical line represents the expected
Nb limit of 5 x 106 A / cm2 [39]. As the lithography improved over time
so did the amount of current that the superconducting contact could
support. The sample with 28 #1112 was one of the earliest samples while
the 4.5 pm2 was one of the last fabricated samples. Also note the high
contact resistance of the 9.5 pm2 sample and how it will not support
a high current density. ..........................

AAR vs Area for several thickness of C091Fe9 . The variations in AAR
are similar to those seen in previously published data for samples with
areas of current flow of 1 mm2 (far right). Note that while the 30 nm
data has large variations, individual substrates showed more consistent
data, except for substrate 1 which had a poorly defined contact area
for one of the contacts (poor-lift-off) ..................

AAR vs Area for several thickness of Py. The variations in AAR are
similar to those seen in previously published data for samples with
areas of current flow of 1 mm2 (far right hand side) ...........

Minor hysteresis loops of CPP resistance versus magnetic field for one
of the smallest contact-area Py samples compared with a 1 x 1 mm
sample (t Py = 6 nm for both samples). The saturation and coercive
fields have been affected by processing. .................

Minor hysteresis loops of CPP resistance versus field for (a) t Py=15
nm and (b) tCope = 18 nm samples. Here the magnetic properties are
very similar to those at 1 mm2 samples ..................

The effect of the spin diffusion length lsfon AAR. The experimental
results show that l,fis indeed finite. ...................

AAR versus tCope for 1 x 1 mm (solid squares) and micron-size (open
circles) CoFe EBSVs ............................

AAR versus tpy for Py EBSVs with 1 x 1 mm (solid triangles) and
micron-size (open circles) areas. .....................

XV

58

60

60

61

62

63

64

64

4.1

4.2

4.3

4.4

4.5

4.6

4.7

4.8

The layout of different wires used for domain wall trapping. (a) A poor
candidate for domain wall trapping as the ends will not easily produce
domain wall nucleation. Many of the initial 2-pm-wide samples were
shaped like this due to poor contact in the mask aligner. (b) The flat
ends will be better for domain wall nucleation. This was the goal of
the photolithography trials. (c) The best option is to provide a source
of domain walls. The diamond on the end will be multidomain and
nearly always inject a domain wall into the sample. This configuration
was used for e-beam samples. ......................

A Ni84Fe16EBSV (t Py = 60 nm) in the large photolithography geome-
try. There is poor evidence of domain wall trapping ..........

A COglFeg EBSV (tcope = 30 nm) from earlier transport studies (Chap-
ter 3) that gives indications of domain wall trapping behavior. Note
the plateau (in the transitions) between the P and AP states.

The best domain wall trapping sample with the initial lithography.
Although decent results were obtained, they were not reproducible, as
evidenced by these two runs on the same structure. This is a C091Fe9
EBSV With tCoFe = 18 nm. .......................

Removing the buildup of photoresist with acetone and a q-tip greatly
increased the resolution of the photolithography. Before this improve-
ment, the photoresist along the edges was much thicker and hindered
good contact with the mask ........................

A domain wall trapping sample with the smaller of the 2 wire geome—
tries. Although decent results were measured they were not repro-
ducible, as evidenced by these two runs on the same structure. This
data is from a COglFeg EBSV with tCope = 6 nm. ...........

A 5 pm contact on the small wire configuration for a C091Fe9 EBSV
(tCoFe = 12 nm). The large plateaus in between the P and AP states
are likely the domain wall trapping. ARNotch is roughly the same for
both intermediate states. Based on the total AR and the resistance
noise, one can locate the domain wall in this sample to about 10 nm.

An SEM picture of a Type-IV sample. Notice that the top contact
(square) is off center. The location of the notch in top contact region

67

68

69

71

72

73

74

determines where the intermediate state will occur in the MR transition. 75

xvi

4.9

4.10

4.11

4.12

5.1

The pattern for e—beam lithographed samples. A diamond at the end
of each wire is patterned to provide a source of domain wall injection.
The aspect ratio of each wire has also been increased to facilitate single
domain wires. Ideally the wide wire would have 40 pm segments, but
the pattern would exceed the range of the writing microscope.

A single MR curve for a Ni34Fe15EBSV (tpy = 12 nm). Notice the
intermediate state located nearly at the half way point. ........

Repeated MR curves for a single sample 1 pm wide (Py EBSV with
t pg = 12 nm). The trapping appears reproducible. Complete curves

were not obtained in some cases due to problems with the electronics .

The motion of a domain wall (grey) after it becomes pinned by defect
or geometry (black circles). The ends of the wall tend to stay fixed
while the center will bow out. ......................

A proposed top view of the new fabrication to put Nb only in certain
areas of the sample. (a) Photolithography patterned leads are made
by sputtering an entire EBSV onto a substrate photoresist mask. The
mask allows features down to several pm to be patterned. Note: exten-
sions of these patterned leads to mm size contact pads are not shown
here. (b) Using the unattached leads as rough alignment marks, a wire
mask feature is patterned on the center lead by e—beam lithography.
The sample is then ion milled to the bottom Nb (Au coated) leaving
the EBSV structure only underneath the mask (including alignment
marks). (c) The sample is planarized. (d) The alignment marks are
used to align a pattern (photo- or e—beam lithography) that will expose
the notch region and a connection to the remaining leads. Sputtering
Nb into the exposed region completes the sample. ...........

xvii

77

78

79

80

83

Chapter 1: Introduction

One of the most important areas of magnetic technology today is the study of
small features where the size of the ferromagnetic (F) elements used in a device is
comparable to the domain size. As those elements shrink those sizes, the mechanism
of domain switching and domain wall motion are strongly affected, and the physics
of such mechanisms is of great interest.

The goal of this research is to make high resolution studies of domain wall motion.
As the direction of magnetization switches in a magnetic element in the presence of a
changing magnetic field, a domain wall, representing the boundary between favorably
oriented magnetic moments and non-favorable magnetic moments, will move through
the sample. By using a giant (G) magnetoresistance (MR) multilayer with the current
perpendicular to the plane (CPP), it is possible to study the motion of the wall on a
scale of a few nm (Figure 1.1).

The GMR effect is a large change in resistance for a multilayer that depends on the
relative orientation of the magnetizations of the F layers within the multilayer. Al-
though the details are provided later, the basic effect is that when the magnetizations
of the F layers are aligned parallel (P), the resistance is low; when the magnetizations
are aligned anti-parallel (AP), the resistance is high. In these studies the magnetiza-
tion of one layer is fixed, while the other is free to respond to an applied field. When
the switching of the free layer takes place, it typically occurs by the propagation of
a domain wall. As that domain wall moves past the measuring contact, the move-
ment of the wall will be sensed as a resistance change since the regions of P and AP
magnetization are changing.

In order to provide a background for what this research encompasses, a brief
introduction to the subject of magnetic domains and the basic physics of GMR will

be given.

domain wall

    

P state AP state

 

 

Current Flow

Figure 1.1: The basis of this thesis research. A domain wall moves in a ferromagnetic
layer (F) that is free to reverse its magnetization (Top F layer) . As the wall moves
past the top measuring contact the resistance changes in proportion to the location of
the wall. The resistance change is from GMR where parallel F-layer magnetizations
will give a low resistance and antiparallel magnetizations give a high resistance.

1. 1 Magnetic Domains

When placed in a sufficiently strong magnetic field, ferromagnets show a very
large magnetization. In addition, they can remain magnetized even when the applied
field is returned to zero. However, an ordinary piece of ferromagnetic material can
at first be demagnitized. Such a demagnetized initial state is due to the presence of
magnetic domains with randomly oriented magnetizations. When exposed to a slowly
varying magnetic field, the domain structure will change so that the sample becomes
magnetized. Due to the irreversible nature of the domain structure changes, the
magnetization will exhibit hysteresis. Magnetic domains redistribute themselves to
reduce the overall energy of the sample based on several competing energies. There are
two methods by which domains change their structure. First is the growth of favorably
oriented domains. Domain ‘walls’ move to incorporate more moments that have
switched to a favorable orientation. Second is coherent rotation where the magnetic
moments of an entire domain rotate together to a more favorable orientation. Wall
propagation happens at lower fields while coherent rotation occurs at higher fields
(Figure 1.2).

Domains form in ferromagnets due to competing energies. These energies can be
divided into local and non-local contributions. Local terms are derived from energy
densities which depend on the local value of the magnetization. The energy associ-
ated with each can be calculated by a single integral of the form f f (m)dV where
f(m) is a function of the magnetization direction m(r) = J (r) / J5, where J (r) is the
magnetization [1].

N on-local energies are torques on the magnetization vector at each point that
depend on the magnetization vector at every other point. Generally, these energies

cannot be calculated with a single integral [1].

‘ Saturation

   

Domain Rotation

 

 

Coercive
Field, Hc

 

 

l

E/V

Domain Rotation

' Domain Wall
l

Propagation

Figure 1.2: Magnetic hysteresis of a ferromagnet. The rapid changes in magnetization
are due to growth of favorably oriented domains by domain wall propagation. The
slower changes near saturation are due to coherent magnetization rotation of domains.

1.1.1 Local Energy Contributions

The first local energy contribution is the exchange energy. The exchange energy
causes a preference for a constant direction of magnetization. Changes from this

equilibrium produce an energy of the form:
EB, = A [(Vm)2 dV (1.1)

where A is a constant that depends upon the material. A is referred to as an exchange
stiffness constant and gives a measure of the strength of the coupling between the
spins in the material [1].

The second local energy contribution is the anisotropy energy Ea. A ferromagnetic
crystal will have preferred axes of magnetization. These preferred directions arise from
the band structure of the material and the spin-orbit interaction [1,2]. The form of
the anisotropy energy is highly dependent on the crystal symmetry and is discussed
in detail elsewhere [3,4].

The third local energy contribution is the external field or Zeeman energy. This
is the contribution from the interaction of the external applied field Hmp and the
magnetization [1]:

Ez = —J./H.,,, . m dV . (1.2)

1.1.2 Non-local Magnetic Energy Contributions

There is one large and one small non—local energy contribution. The large contri-
bution is the stray field energy or the energy of the magnetic field Hm created by the

magnetization. This usually takes the form [1]:

ES,=%po/Hm2dV=-%/qudv . (1.3)

The other is magnetostriction, which is the energy associated with the elastic motion
due to magnetic fields. This latter energy will not be an important contribution in

this research.

1.1.3 Interaction of Energy Contributions

All of the above energy components combine to give a domain structure. The best
way to explain this is with an example. Consider a square crystal magnetized entirely
in a single direction (Figure 1.3a). By dividing the configuration into two separate
oppositely magnetized domains, the stray field energy of the example has roughly
been reduced by a factor of two (Figure 1.3b). The addition of more oppositely
magnetized domains will continue to reduce the stray field energy until the energy
to produce a new boundary (both the exchange and anisotropy energies) between
domains is greater than the stray field energy reduction in the field (Figure 1.3c).

It is possible to form domain arrangements where the magnetic field is zero and
thus the stray field energy is zero outside the sample. By forming triangular domains
at the ends of the example with magnetization at 90° to the original domains, there
will be no surface poles of magnetization and the stray field energy will then be zero
(Figure 1.3d) externally. These types of domains are known as domains of closure.
The anisotropic energy may hinder the formation of the closure domains, since the

magnetization may fall along a hard axis of magnetization.

1.1.4 Shape Effects

Altering the external shape of a sample can also affect the domain structure of a
sample. In Figure 1.3 the example was a rectangle. It is well known the external shape
can significantly affect the preferred axis of magnetization, and it is also possible to
trap a domain wall using shape effects. For long narrow wires it is well known that the

magnetization prefers to align only along the long axis of the wire. This is so-called

 

 

 

 

 

 

 

 

 

(a) (b)

 

 

u}

H

4-
(C) (d)

 

 

 

 

 

 

 

 

 

Figure 1.3: Domain formation in a single crystal. (a) A single crystal that is uniformly
magnetized. (b) Dividing the crystal into two magnetic domains reduces the stray
field energy. (0) Further division into domains will reduce the stray field energy until
the energy to create a new domain is greater than the reduction to the stray field.
(d) Closure domains can reduce the stray field energy to zero, but can contribute to
the anisotropy energy.

shape anisotropy and is a consequence of the stray field energy [1].

Consider a long narrow wire with a domain wall located in the lower half (Figure
1.4) . As an applied field is reversed, the domain wall will move along the wire
to accommodate the growth of the favorably oriented domain. If a narrow region is
placed along the wire, the wall will enter that region and if the field is being increased
slowly, the wall will remain in the notch area since the wall energy itself is reduced,
and the energy to facilitate domain growth in the narrow region is greater than the
wide region. Extra energy is needed to overcome this barrier. The wall will then stay
in the neck (trapped) until the applied field contributes enough stray field energy to

drive the wall out of the notch.

1. 1.5 Domain Walls

The boundaries between domains are known as domain walls. These are regions
where the magnetic moment changes from one orientation to another. There are
several types of walls that form based on the previously mentioned competing ener-
gies. Before the different types of walls are described, it is important to see why the
competing energies lead to the formation of walls.

As mentioned before, it is energetically favorable to have domains. However, the
exchange interaction energy makes it more favorable for the transitions between the
differently oriented domains to be gradual rather than abrupt transitions. As shown
in standard physics texts [5,6], a gradual transition from one orientation to another
will reduce the overall energy of the system.

Starting with the Heisenberg model for the exchange energy E 5,, we have
E13,, = —2JSi - Sj , (1.4)

where J is the exchange integral. Replacing cos 0 with the small angle approximation

 

 

 

 

 

 

 

 

 

 

 

 

(a) (b) (C)

 

 

 

 

 

 

 

 

>—< > <

 

 

 

 

 

 

 

(d) , (e) (D

Figure 1.4: Shape anisotropy is used to trap a domain wall in patterned wires. In (a)
through (c) a domain wall starts in the wider portion of the wall. (a) As an applied
magnetic field is increased in the up direction the wall will move towards the narrow
portion of the wire. (b) The applied magnetic field does not give a large enough
Zeeman energy contribution to allow the wall to enter the narrow portion of the wire.
The narrow portion of the wire has much larger stray field energy due to surface
and edge effects. (0) The magnetic field is increased enough for the wall to enter the
narrow portion of the wire. ((1) through (f) repeat this sequence with a narrow neck
or notch region. The effect is the same, except once the wall gains enough energy to
leave the initial constriction, it most likely has enough energy to move entirely to the
end of the wire.

gives an exchange energy between two spins of J S 262. Using 1r/ N as the angle between
nearest neighbors of a wall N atoms thick, one obtains that each neighboring pair
contributes JS2(1r/N)2. The total energy of N+1 atoms would then be JSz(7r2/ N)
for large N. If there were no anisotropy energy, the wall would thicken to infinity, but
generally the spins are directed away from the easy axis of magnetization and thus
contribute an addition anisotropy energy [2,5].

Two of the most common walls in thin films are the Bloch and N eel walls (Figure
1.5). Although Bloch walls are energetically favorable in thicker materials, Neel walls
become preferred as the layer thickness decreases, since the Bloch wall will produce
poles at the surface, creating a stray field energy (at the surface) which increases
with decreasing layer thickness. The Neel wall avoids this problem at the cost of
a large stray field energy inside the magnetic region (which makes the N eel wall
unfavorable in bulk materials). For the samples studied in this research, the Neel
wall is energetically more favorable [7].

The width of the wall depends on both the exchange and anisotropy energies.
There is no widespread agreement on what defines the wall “width.” However, for
the current studies a precise definition of the width is not necessary. It has also been
proposed that a geometrically constrained wall in thin films may be a new type of
wall that has Néel tails extending several nm [8]. It is the domain wall propagation
or movement that is of interest to this study. Ono et al. showed by using a current-
in-plane GMR that observing domain wall trapping was possible [9,10]. CPP-GMR
can also be used to study the domain wall propagation down a long narrow (~ 1p m
wide) wire. By using external shape to create pinning sites (Figure 1.4), a domain
wall can be trapped in a narrow neck placed in the above-mentioned wire. For a
GMR sandwich with two F layers and well defined P and AP states, the trapping of a
domain wall in one of the layers will create an intermediate state where the resistance

is between the P and AP resistances. A further advantage of CPP-GMR is that the

10

N éel Wall

 

 

 

 

 

 

 

 

A

 

 

// x\\ ~

V

 

 

 

V

‘/// \\\.~w

N

 

 

V

 

 

 

 

 

 

   

ZA
Easy Axis

 

 

 

 

tum «

 

 

 

)( ~

 

Bloch Wall 4» x

Figure 1.5: The two most common types of domain walls. The Neel wall has the
magnetization rotating in x—z plane. The Bloch wall has the magnetization rotate

vertically in the z-y plane.

11

 

current may be confined to the trapping region instead of flowing through the entire

sample, as occurred for the Ono samples.

1.2 Background of GMR

Magnetoresistance (MR) is the change in resistance in a conductor due to an
applied external magnetic field. Giant magnetoresistance (GMR) is a specific MR
effect in multilayers due to spin dependent scattering. Current in a ferromagnetic /
non-magnetic (F / N) multilayer can be thought as having two current components:
one current composed of up electrons (with spin parallel to the magnetization) and
one composed of spin down electrons (with spin anti-parallel to the magnetization).
Due to spin dependent scattering, the resistivity of one channel of electrons is less
than the resistivity of the other spin channel in the F layers (normally spin up has
a much less resistivity than spin down, but this is not always the case). When the
magnetizations of all F layers in the multilayer are parallel (P), the up electrons have
a much lower total resistance path than the down electrons. The total resistance
is ‘shorted,’ similar to two resistances in parallel, the total circuit resistance will be
approximately the lesser of the two resistances if one resistance is much smaller than
the other. If the F layers have an anti-parallel configuration (AP), both current paths
have identical resistances and the overall resistance is higher. A simple explanation
of GMR is shown in Figure 1.6. This effect can lead to resistance drops by a factor of
two or more in some layer configurations. This large change in resistance has many
practical applications such as magnetic read heads for hard drive storage, magnetic
sensors, and magnetic random access memories.

GMR was discovered in 1988 in a superlattice multilayer with alternating Fe and
Cr layers [11, 12]. This discovery has opened a new field of research in the area of

multilayer structures, which is now known as spin electronics or spintronics. These

12

Strong scattering

  

 

1

High Resistance State Low Resistance State

Figure 1.6: An explanation of the basic GMR effect. On the left, when the magne-
tizations (white arrows) of the magnetic layers, F, are antiparallel, both the spin up
and spin down electrons have equal resistance paths. When the magnetizations are
parallel, the spin up electrons (in this case) have a lower resistance path compared
with spin down electrons. This leads to a much lower overall resistance.

first experiments were done on single crystal samples grown with molecular beam
epitaxy. Later investigations reproduced the results of the Fe / Cr multilayer as well as
Co/ Ru and Co/ Cu multilayers using polycrystalline samples deposited by sputtering
[13]. In addition, the first experiments were done in a geometry where the current
flows parallel to the planes of the multilayer (Figure 1.7).

The geometry where the current flows in the plane (CIP) of the multilayer is dif-
ficult to model experimentally. In order for an electron to show the GMR effect it
must sample several different magnetic layers. Since the current is travelling along
the multilayer, scattering must be responsible to get the electron to move across in-
terfaces and into several different layers. The mean free path is thus a very important

parameter with CIP geometry experiments.

1.3 CPP Measurements

An alternative geometry has been developed [14—22]. By passing the current
perpendicular to the planes (CPP), the electron current must now travel through

every layer. Since scattering is no longer required to move electrons across interfaces,

 

 

 

 

 

 

Figure 1.7: A diagram of the CIP geometry vs. the CPP geometry. In the CIP
geometry the current (black arrow) flows parallel to the multilayer interfaces. In
the CPP geometry the current flows perpendicular the layer interfaces (The diagram
dimensions are not to scale, typical layer thickness (vertical dimension here) is 1 - 10
run, while the width and depth of the samples (horizontal and perpendicular out of
plane dimensions) are 0.1 to 10 mm) .

the mean free path is no longer an important parameter. Instead, the spin diffusion
length, [8,, becomes the dominating length scale. l,, is the distance an electron will
hold its spin direction while it is diffusing in a given material. Once electrons start
to spin flip scatter, the GMR effect is reduced.

CPP-GMR presents several advantageous over CIP-GMR. First, the GMR effect
is larger in the CPP geometry. Second, the CPP geometry allows easier access to the
fundamental parameters of GMR. The disadvantage of CPP-GMR is the very low
resistance of the samples. For example, a typical 100—nm-thick single layer of C091Fe9

, 1 mm wide and 1 cm long in the CIP geometry will have a resistance of

_ p_l _ 7pf2cm x 1cm _ .
R " A _ .lcm x (100 x10‘7cm) " m ’ (1‘5)

 

for the CPP configuration a typical geometry is 1 mm x 1 mm for a 100 nm thick

COglFeg sample, which gives a resistance of

d _ 70pflcm x 100nm _

R=A .lcmx.1cm

 

7119 (1.6)

14

well outside the range of traditional resistance measurements.

There are several methods used to experimentally measure CPP-GMR. Because of
the very small layer thickness, these fall into two major categories: (1) very sensitive
measurement techniques can be used; or (2) the area of current flow can be reduced
to produce resistances measurable by more traditional means.

The first measurements of CPP-GMR were done at Michigan State University
[14] by sandwiching a multilayer between two 1-mm-wide crossed superconducting
strips composed of Nb. This geometry provides a uniform current flow through the
multilayer, and a SQUID based self-balancing potentiometer circuit is used to detect
the nil resistance. Because of the use of superconductors, measurements are generally
done at liquid helium temperatures.

Additional techniques [16—18,22] have been developed which rely on lithograph-
ically microfabricated samples to eliminate the need for superconducting contacts.
This type of sample is several pm in diameter and S 1 pm thick. This type of
measurement is not straightforward since there are significant and not well defined
contributions from the contact resistance and additional resistance from the leads.
Also, the current flow is complex and not necessarily perpendicular to the planes
of the multilayer. Despite these drawbacks the first temperature dependent CPP-
GMR measurements were performed with this method. Later efforts [22] including
superconducting contacts to improve current flow, but this method used many (100)
contacts in series, providing an additional constraint of high yield lithography, since
a single poor contact in the series chain can ruin the measurement.

Another microfabrication method to produce small size samples is to use electrode-
posited nanowires formed in nuclear-track-etched polycarbonate membranes [19—21].
The very high aspect ratio, with a length of ~10 pm and diameter of 20-100 nm,
provides a more perpendicular current flow than the previous microfabricated sam-

ples. It is difficult, however, to determine the number of nanowires being measured in

15

parallel [23]. This makes it difficult to determine the absolute value of the resistance
of an individual wire.

This thesis proposes a new type of CPP—sample measurement. By employing mi-
crofabricated superconducting contacts, micron size areas of current flow can be used
to study initially otherwise macroscopic (1 x 10 mm) samples. This method can
additionally be used to study multiple areas of the same sample. These microfabri-
cated contacts will provide uniform current flow through a small cross sectional area,
combining many advantages of superconducting leads and small areas of current flow,
while avoiding complex lithography. A secondary goal of this research is to produce
samples whose resistance is large enough to measure with a nanovolt system. Finally,
as the sample size is reduced, it will be possible to do high-resolution ‘imaging’ of

domain wall trapping.

1.4 Theory of CPP-GMR

The present theory of CPP-GMR has two forms based on the relative size of Is;
and the layer thickness, t. The first form describes [24] why the two current series
resistor (2CSR) model holds when l3f>> t. The second form was developed [24,25]
for I”: t and includes an extension of the 2CSR model for short ls].

Before going into the details of CPP-GMR, it is useful to provide a simple picture
of the physics occurring in the multilayer. For example, in this simple picture, Co
can be thought of in terms of 4s and 3d parabolic electron bands [24, 26, 27]. It is
the 4s band that will be the primary conduction carrier, since the band has a very
low effective mass. The 3d electron bands can be split into spin up and spin down
bands. Because of the exclusion principle, electrons with spin up will begin filling the
3d-bands of the metals. Once all the spin up states are occupied the spin down states

will become partially occupied. From Fermi’s golden rule, the rate of scattering of the

16

 

 

4s 3d

Figure 1.8: A simplified diagram of the 4s and 3d energy bands for Co. The up and
down arrows represent the bands for the up and down electrons, respectively. Since
the 3d] band fills with up electrons first, at the fermi level, the density of states for
down electrons is much larger giving a higher resistivity.

4s electrons (and thus resistivity) will be proportional to the density of states of the
3d electrons, so spin down electrons, having a higher density of states at the Fermi
level, will have a higher resistivity (Figure 1.8).

In the 2CSR model, the resistance of a multilayer is obtained by dividing the
current into independent spin up and spin down contributions and calculating the
resistance each current encounters in both the parallel (P) and anti-parallel (AP)
states. Each bulk and interface resistance can be thought of as a separate resistor.
To calculate the GMR one adds up the resistance contributions in series for each
current channel and combines them as two overall resistors in parallel. When the
resistances from both the P and AP magnetic states are known a simple calculation
of the specific resistance AAR = Area x (RAP - RP) can be made, where A is the
area of CPP current flow.

Before taking a closer look at the 2CSR model it is important to define several

parameters used to make calculations and model the multilayer. (This review follows

17

the notation and spirit of the original work by Fert as well as several review papers on
the matter [24, 26, 28].) The resistivity of each channel in a F layer is denoted as p],
for electrons with their spins parallel to the magnetization and p]? for electrons with
their spin parallel anti-parallel to the magnetization. A useful parameter for such a
material is

plv + pie

" = 1.
PF 4 ( 7)

which provides a measure of the average resistivity for spin up and down electrons.

In addition, the spin asymmetry variables a, and 3 are also important:

 

l
a=%; (1.8)
l T
— a—l

p}+pi~ 0+1.

The resistance of each channel can then be related to p“ by using 5:
p} = 2pi~(1- fl) ; (1.10)

pt = 2pm + m . (1.11)

It is important to note for non-magnetic materials such as Cu that ,6 = 0, giving

pi = 2va . (112)

There is an interface resistance counterpart to 6. Instead of a resistivity, '7 refers to a
specific resistance AR (= Area x Resistance; A is the area of current flow). 7 relates

the average AREA, to ARp/N for the individual spin up and spin down channels.

ARL/N = 2AR}/N(1— 7) (1.13)

18

F/Nt +ARFi

 

 

 

 

 

Figure 1.9: A diagram of the two current series resistor model for the AP state. The
current is divided into two channels: spin up channel (top) and spin down channel
(bottom). One adds the series resistances for each channel and then combines the
series resistances as two resistances in parallel to produce the overall resistance for
both the AP states.

ARi/N = QAR‘p/NU + 7) (1.14)

Using these relations one can determine the resistivity for a F/ N multilayer in terms
of p3,, p}, '7, 6, and ARE/N where the subscripts denote which layer the parameter
represents.

If one takes the case of a F/ N/ F multilayer with layer thicknesses t p, tN, and
tp respectively (shown in Figure 1.9), the calculation is done in the following way.
Taking the case of the spin up channel in the AP state, one adds up the various

resistances from both the bulk and interface.

ARLP = 2p2~(1—fi)tp+2AR;/~(1—7)+2p7vt~+2AR;~/N(1+v)+2p‘p(1+fi) (1-15)

19

This can be simplified to
ARI“, = 4p}tp + 4.412;”, + 2p;,tN . (1.16)

Then one calculates the spin down channel, which gives the same result as Equation
1.16.

Calculating for the parallel state one gets for the spin up channel:
AR} = 2piv(1-fi)tp+2AR‘p/~(1-7)+Zpivt~+2AR;~/~(1-7)+2pir(1-fi)tp, (1-17)
while the spin down channel gives:
AR], = 2p}(1+,3)tp+2AR}/N(1+7)+2p7vtN+2AR}/N(1+’7)+2p}(l+fl)tp. (1.18)

The overall resistance of each magnetic state is a parallel addition of the spin-up and

spin-down channels resistances. The resistance of the parallel state, ARp, is:

ARLAR],

AR = ,
P AR}.+ARP

 

(1.19)

and the specific resistance of the AP state, ARM», is computed in the same manner.
AAR is then just ARAp - ARp, and the GMR ratio is 5 (AR/”3 - ARp)/ARp.

This model holds as long as l,f>> t, as shown by Fert and Valet [24]. However,
for lg not >> than t p (such as in the material Ni84Fe16) one must take into account
spin relaxation [24,25]. The 2CSR model must then be modified, and for many cases
only numerical solutions exist. However, for a F/ N multilayer with M F / N bilayers,

where M is large, an algebraic solution exists [24]:

AREA}: = M(AR0 + 2.412%pr , (1.20)

with AR0 = (1 - c2)p;.tp + p7,,tN + 2(1 — 72)AR;/N (1.21)

20

and

 

 

pfvlfi 21$ 7

 

Meow [t—N] + 72 cothlti] +...
3

 

 

 

 

 

 

 

 

 

 

 

 

 

ARSI __ PF sf 7213f
P _ ,
1 by; 1 tF
.1. N COth [7:] *—FCOth[ F] +...
lesf 213] pFlsf lef
2
..+ Ag
3 1 1 t m 1 t (122)
...+ ,, 1—coth J— + * coth F
ARF/N {psz} [21,“; Flf, [213]}
_ 2 2
(3* pt [t%]+ :YFcoth[I;,]-i-
Alli]: _ lef 213} ,, pFlsf lef a}
tN 1 [ ]
tanh coth +
.. N N .1 F F
lef [2131] pFlsf lef
2
..+ Ag,
3 1 I t ”N 1 t (123)
. . . + ‘ tanh l + '7— COth F
ARF/N {pj‘vlf} 21$] pplff [Zlff]}

 

Both the two current series resistor model and the extension for short 1,, are supported

by experimental evidence [15,26].

21

Chapter 2: Sample Fabrication

2.1 Introduction

All samples were prepared and measured at Michigan State University. Although
there were four separate procedures for lithography, each sample underwent the same
basic fabrication procedure (Figure 2.1). The samples (in sequential processing or—
der) were: cleaned, sputtered with a multilayer, lithographically patterned, etched,
planarized with an insulating layer, cleaned to remove the lithographic pattern, and
finally sputtered with top Nb leads. The details of each specific sample type, or

category, are contained below.

2.2 Types of Samples

Four categories of photolithography sample were made (Figure 2.2). The samples
differed in the top portion of the multilayer or ”cap” and whether the giant magne—
toresistance structure was etched or not. The multilayer consists of a bottom, 250
nm thick Nb layer, followed by an exchange bias spin valve (EBSV) or a hybrid spin
valve (HSV). The exchange bias spin valve structure is (beginning with the previously
mentioned Nb layer): Nb (250) / Cu (10) / FesoMnso (8) / F (x) / Cu (20) / F (x)
/ Cu (10), where F = Nig4Fe160r C091Fe9 , x = 3-40, all units are run. The HSV
structure is: Nb (250) / Cu (10) / F (x) / Cu (10) / F (y) / Cu (10), x = 3-30 and y
is 5—30, but x 7S y. The EBSV uses the Fe5oMn5o layer, which is an antiferromagnet,
to bias the adjacent F layer’s magnetization in a ‘pinned’ direction. This causes the
coercivity of that F layer to be very large. The other F layer is free, or unpinned,
and has a relatively low coercivity. For low magnetic fields the free layer will align

with the applied magnetic field while the pinned magnetization will remain fixed.

22

 

 

 

A Multilayer
'_—:_submte

 

 

 

 

B lithography mask \

 

 

 

 

 

C SiO ——

 

 

 

 

 

 

D top Nb —

 

 

 

 

 

Figure 2.1: An overview of the fabrication process. A side profile is provided on
the left (only half of the lithographic features are shown to conserve space), and
a complete top view is provided on the right. The dimensions of the system are
not to scale. A multilayer is sputtered onto a cleaned substrate (A). A lithographic
pattern/ mask is placed onto the multilayer (B). A portion of the multilayer is etched
away, an insulting layer deposited and the lithographic pattern removed (C). The top
contacts are sputtered onto the sample (D).

23

This setup allows good control of the anti-parallel and parallel states for GMR. The
HSV uses different layer thicknesses to achieve the same results. Since layer thickness
(especially when the lateral layer dimensions are small) will affect the coercivity, by
choosing one layer to be very thin and one very thick, good control of the anti-parallel
state can also be achieved using this method. Each spin-valve was then finished with
one of two types of “caps.” A layer of Au, 15 nm thick (although for some samples
this was increased to up to 150 nm), formed a type-I cap (C;). A type-II cap (Cu)
was formed by a trilayer of An (10 nm)/ Nb (20—250 nm), Au (15 nm). (The final
Au layer on the surface prevents oxidation during lithographic processing.) The C1
cap was easy to process, requiring no etching, but needed extensive cleaning to make
contact with the final Nb layer. The C 11 cap required etching, but more easily pro-
vided low contact resistances with the final Nb layer. Each spin valve and its cap
were processed using techniques described below. A final Nb layer was added to the
top of the multilayer stack to complete the sample. Thus, each final multilayer has
three parts: 1) the initial(bottom) Nb layer followed by the GMR spin valve; 2) the

cap; 3) the top Nb layer.

2.2.1 Type-I Samples

Type-I samples were multilayer with a C 1 cap of 15 nm Au on the surface. Type—I
samples had the photomask placed on the Au, and the insulating layer evaporated
onto the entire sample. Removal of the photoresist exposed a trench, or via, through
the insulating layer to the multilayer. Through this via, sputtered Nb (through a
shadow mask) would make electrical contact with the multilayer below. This type of
sample was simple to process, but could suffer from interface resistance problems if

the Au/ Nb interface was polluted due to insufficient cleaning.

24

Type-1 Type III

2.2.

Type- -II Type-VI

EE

Figure 2.2: An overview of the different types of samples. Type-I has contact made
to a Au capping layer. Type—II has a partially etched Au/ Nb/ Au cap. Type-III has
the entire multilayer etched. Type-IV has a separate contact geometry on top of a
patterned multilayer

2.2.2 Type-II Samples

Type-II samples consisted of multilayers capped with a Cu cap. After placing
a photolithography mask on the sample, the Nb/Au part of Cu was removed by
various methods depending on the thickness of each layer. Initially, Au was wet
etched with a KI / I2 solution. However, this solution also attacked other portions
of the multilayer and therefore an ion milling process was introduced that did not
have these difficulties. The ion mill was then exclusively used for later samples. The
Nb layer, if 20 nm or less, was also ion milled. If thicker than 20 nm, the Nb layer
would take too long to remove using ion milling (up to several hours) and was instead
removed by reactive ion etching (RIE). This entire process left a short column of
Nb(x)/Au(20 run), where x = 20 - 100 nm, underneath the patterned features. An
insulating layer was evaporated and the lithographic features were removed. Last, the
top Nb layer was sputtered onto the sample. This type of sample, in theory, makes

it easier to eliminate contact resistances, since the Nb/Au/ Nb interface should be

25

superconducting by proximity [29,30]. In addition, the important F / Au/ N b interface
is in effect ‘buried,’ and not exposed to lithographic processing. Type-II samples were
proposed before type-I samples, but due to problems with Reactive Ion Etching (RIE),
type-I samples were developed to eliminate the RIE step. These problems were later
solved (the details can be found in the reactive ion etching section). Both type-I and
—II samples were integral in the development of the type-III and -IV samples. Type-I

and -II samples were used exclusively in CPP-transport measurements to study the

AAR behavior.

2.2.3 Type-III samples

The type-III and type-IV geometries were used to detect domain wall trapping.
Whereas type-I and -II samples always employed square lithographic patterns, type-
III samples had a more complex wire geometry (Figure 2.3) . Typically, the type-III
samples were multilayers with a C1 cap for photolithography and a modified C1
cap (with cap Au now 100-150 nm thick) for e-beam lithography. The increase in
Au thickness was needed because electrical contact into a ‘trench’ or via becomes
difficult or impossible once the dimensions are reduced below 2 pm. The thicker
Au creates a ‘proud’ column (Figure 2.4) that, after ion milling, exceeds the level
of the insulating layer and avoids the contact problem with vias (Figure 2.2 shows
a photolithgraphy sample and thus shows a via instead of a proud column). After
the lithographic mask was placed onto the sample (this was done by both e-beam
and photolithography), the entire GMR spin-valve was ion milled to the bottom Nb
superconductor. An insulating layer was then evaporated onto the sample. Following
removal of the mask, Nb was sputtered onto the sample. This left the entire in—plane
geometry of the patterned multilayer in contact with the top superconductor. In
order to study only part of the wire, it was necessary for the superconducting contact

and the spin-valve to have different geometries.

26

 

4pm 2pm

l__ l—

 

 

2 [Im 40 um I um 20 pm
2 pm 1 pm
V

 

a b

Figure 2.3: A diagram of the mask for fabricated wires with a neck. There were two
sizes of wire designed. Mask b is exactly a factor of two smaller than a. Each is
symmetric about the notch region.

 

Figure 2.4: The difference between a via and a proud column. A via is a trench or
window in the insulating layer through which contact with the top superconductor
(dashed line) is made. A proud column sticks above the insulating layer. Proud
columns become necessary when the feature size is below 2 11m.

27

2.2.4 Type-IV samples

Type—IV samples underwent type-III processing up to and including the removal
of the lithographic mask, and then a second lithography step was used to produce a
top contact of different geometry (a square) than the original patterned multilayer
feature from the first set of lithographic procedures. This small square type-I contact
allowed study of particular regions of the previously patterned wire structure. The
second lithographic pattern was placed on the previously patterned features, SiO
was deposited, the lithographic mask was then removed and finally the top Nb layer
was deposited. In this dual procedure, a variety of different multilayer and contact

geometries could be produced.

2.3 Procedures

2.3.1 Substrate Cleaning

The sample fabrication process began by cleaning sample substrates in a class
100 cleanroom (measured to be class 100, although certified class 1000) . Either
silicon (100) or c-axis oriented sapphire substrates were used. Silicon was used for
the majority of samples. Sapphire was employed only for samples that needed to
be reactive ion etched (i.e., that had a Nb layer in the cap that was thicker than
20 nm). The substrates were first immersed in Alconox® at 40°C. An ultrasonic
cleaner agitated the substrates for approximately 15 minutes. All samples were then
rinsed in water to avoid a reaction between the Alconox® and the acetone used in
the next step. When mixed, Alconox® and acetone produce a dark film that hinders
lithography because it is difficult to remove. After rinsing with water, substrates were
placed in acetone at 40°C, agitated in an ultrasonic cleaner for 15 minutes, dried, and

visually inspected using a Olympus BX60 optical microscope at 20-100X. Samples

28

with defects in the central area (specifically, where the multilayer would be deposited)
were scrubbed in acetone by rubbing the surface with a Q-tip and inspected again.
If non-removable defects such as scratches were present the substrate was discarded.
Substrates that passed inspection were placed in isopropyl alcohol (IPA) at 40°C and
agitated in the ultrasonic cleaner for 15 minutes. Finally, samples were immersed in
deionized water at 80°C for at least 15 minutes and left in water until placed in the
ultra high vacuum system for sputtering (a time of minutes to a few hours at most).

Thus samples were not stored for long periods of time in this cleaned state.

2.3.2 Sputtering

Sputtering was done in an ultra high vacuum system equipped with a cryopump
capable of producing pressures of 10“8 Torr after 24 hours of pumping [31]. Sample
substrates were loaded onto a position and movement assembly (SPAMA) plate that
contained eight holders, each holder having the capability to hold two samples. The
holders sat in holes in the SPAMA plate and were attached to the plate via two
screws. In addition, a circular stainless steel rotating mask was attached to the base
of the sample holder which allowed the user to expose or protect the substrate by
rotating the mask to an open or closed position (Figure 2.5). A separate shadow
mask (stainless steel) was placed in the holder first, allowing the formation of a
pattern during sputtering. The sample substrate was placed on top of the shadow
mask. Copper heat sinks were placed on top of the substrates, and the entire stack
was held in place by removable “bridges” that attached to the holder with screws, to
apply pressure to the top of the heat sink (Figure 2.6). Any alignment of the shadow
mask with a previously made pattern on the substrate had to be done visually and
before the entire stack was secured.

After the substrates were loaded onto the SPAMA plate, the plate was placed

above a system of movable chimneys that allows the user to control sample exposure

29

 

 

 

 

  
 

Sample Holder

 

 

 

Figure 2.5: The layout of the mask system. The rotating mask (inner circle) is
rotated to expose a sample to the sputtering plasma. A shadow mask (not shown)
over the substrate defines the actual pattern. The 12 o clock and 6 o clock positions
correspond to a closed position while the 3 o clock and 9 o clock position will expose
the sample. The rotating mask is aligned by lining up the pins of the rotating mask
to be in line with the bridges holding down the samples. The mask here is shown in
the 9 o clock position while the others are represented with dashed lines.

I‘l'l”

Heat Sink

L—
_I.. .l-

Sputtered Material

Figure 2.6: A exploded view of the sample holder. The bridge clamps the heat
sink/ substrate/ shadow mask together. The assembly is attached with screws to the
sample holder through the bridge. Each sample holder is designed to hold two as-
semblies.

30

to the sputtered material. The system was pumped down to 4 x 10‘8 Torr. A
overnight baking of the chamber sped up the pump down process. Before sputtering,
a cold trap using liquid nitrogen was activated to reduce the overall pressure to 2 x
10‘8 Torr.

Sputtering was done at an Ar pressure of 2.5 mTorr. Samples were cooled to
-30° C by a capillary system using cold compressed flowing N 2 gas. The pressurized
gas is allowed to expand in a network of capillary tubes, causing the system to cool.
A purifier produces ultra-pure Ar gas for sputtering. Samples were sputtered at
temperatures between ~30 and 30°C. The sample to be made was exposed by rotating
the bottom mask to an open position using a wobble stick. The holder was then
rotated, via the SPAMA plate, over the gun containing the material to be sputtered;
and, when in place, the system of chimneys would rotate to allow deposition for a
specified time. When the desired thickness was reached, the chimneys were rotated to
their previous position to prevent deposition. The SPAMA plate was then rotated so
that the sample was over the next sputtering gun. This was repeated until the sample
was finished, and then the rotating mask positioned to prevent further deposition on

the sample.

Final Top Nb

When depositing the top Nb layer on a sample with completed microfabricated
features, a special alignment step was necessary. As shown in Figure 2.7, the protru—
sion on the sample strip had to line up with the first hole in the top contact shadow
mask. The layout of the lithographic features was intentionally designed to have one

sample contact within each top Nb window of the shadow mask.

31

A B c

substrate multilayer .
protrusion shadow mask (stainless steel)

     

top Nb leads

 

multilayer lithography mask SiO

Figure 2.7: A layout of the mask system to deposit top Nb leads. (A) The multilayer
with patterned mask features (exaggerated size, in black). Each feature is spaced
so that when the far right opening in the shadow mask (B) is lined up with the
protrusion in the multilayer (A), the individual exposed contact areas will each fall
within a single Nb contact (C)

2.4 Lithographic Procedures

The lithographic procedures described here employ a patterned resist mask on the
sample. Depending on the geometry and processing of the mask, the same exposure
procedure can create different types (Type-I,-II, etc...) of samples. Lithography
defines small features from 75 to .3 pm

2.4.1 Photolithography

Photolithography began by placing a sample with a. sputtered multilayer on a
vacuum spinner. The vacuum in the spinner chuck holds the sample in position on
top of a chuck while the sample is spun. A large drop, ~.1mL, of Shippley 1813 or
1805 photoresist is placed by a pipette on the surface of the sample, and the sample
is immediately spun for 60 seconds. The rate of spinning, in addition to the type of

photoresist, affects the thickness of the photoresist layer. For initial samples, Shippley

32

1813 resist was spun at a rate of 4000 rpm to give a photoresist thickness of 1.3 to 1.5
pm. (Measured by exposing large features, developing the photoresist and scanning
the features with a profile stylus.) Later samples had Shippley 1805 photoresist spun
at 5000 rpm to give a thickness of 500 nm. The samples were then baked for 40
minutes at 95°C to harden the photoresist. Feature sizes from 10 to 75 mm were
processed using the thicker photoresist. When smaller 1-um features were made, the
thinner 500 nm thick photoresist was used. For a contact aligner, diffraction is the
limiting factor, and the critical dimension, or smallest feature size, can be estimated
as:

d=kasz (2.1)

where k is 1.6 (index of refraction for the photomask), /\ is the wavelength of radiation,
and z is the distance between the mask and substrate. Using A of 365 nm and z of
1.5'pm gives (1 z 1 pm. A z of .5 pm gives a d z .5 pm. Physical limitations,
such as buildup of photoresist on the corners of the substrate and the inability of the
substrate to flex which allows better contact with the mask, limited the resolution of
the 1.5 pm photoresist layer to 7-10 pm and the 500 nm photoresist layer to 1 pm.
After baking, the samples were placed in a contact mask aligner (ABM, INC). Using
a photomask from Align-Rite corporation, the samples were exposed in selected areas
to ultraviolet radiation of wavelength 365 to 400 nm. Both of the photoresists were
‘positive,’ meaning that the photoresist forms strong chemical bonds (crosslinked).
When exposed to ultraviolet light, the chemical bonds of the positive photoresist break
down, making the exposed resist soluble in certain chemicals known as developers.
For these photoresists the developer was KOH (Shippley 452 developer). The portion
of resist underneath the mask is still crosslinked and insoluble in the developer. Two
different photomasks were used to expose patterns on the photoresist for our samples.
Both photomasks were clear field masks. That is the majority of the mask is clear

or transparent to ultraviolet light, while the features are dark chrome. The first

33

photomask (layout shown in Figure 2.8 and Table 2.1) had only square patterns
ranging from 75 x 75 to 6 x 6 u m. The second photomask (layout shown in Figure
2.9 and Table 2.2) had square features from 75 x 75 to 2 x 2 p m in addition to
two wire geometries (Figure 2.3). The change in photomasks was made to produce
dimensions closer to the theoretical limit of the contact alignment system and give the
possibility to pattern wire geometries of appropriate size in the multilayer for domain-
wall trapping. In addition, the features on the second mask were produced with a
smaller spot size (the masks are produced with dedicated electron-beam writers)
which made the smaller features at higher resolution

After aligning the mask so that a pillar was aligned with the protrusion of the 1
x 10 mm strip, the samples were exposed to ultraviolet light for a fixed period of
time. The samples with 1.5 pm thick photoresist were exposed for 6 seconds. The
samples with 500 nm thick photoresist layer were exposed for 1.5 seconds. The ex-
posed photoresist was removed using Shippley 452 developer (KOH solution). Each
sample was placed in a bath of developer at room temperature for 25 seconds, next
immediately placed in a second bath of developer for an additional 25 seconds, and
then rinsed in a bath of deionized water. The two developer steps were used to avoid
contamination by removed photoresist redepositing on the surface of the multilayer.
Ideally, this process will reproduce the mask pattern on the photoresist. The pho-
toresist is now ready to be used as a mask for selecting which areas of the sample
will be processed. Any area underneath the photo resist is protected from processing

while the photoresist remains.

2.4.2 Electron Beam Lithography

When the limits of the photolithography equipment were reached, the procedures
employed were adopted to electron beam (e-beam) lithography. The basic principle

of the photolithography employed for this research was that the photoresist itself was

34

 

 

 

Rowzlllll
2IIII
3'...
4|..-
5....

f 4in

——_E!

 

 

 

 

 

Row 1
Row 2
Row 3
Row 4
Row 5

 

 

75
75
75
75
75

 

75
40
20

75

75
40
20

6
40

6

 

 

75
40
20

6
20

 

75
40
20
6
6

75
75
75
75
75

 

 

 

Figure 2.8: A diagram of the layout and dimensions of the first photomask. A single
array of features was arranged as shown. The dimensions of the features are listed in

Table 2.1: The side dimension of each square listed for the features on the photomask
that is shown in figure 2.8. All units are in microns.

 

 

Row 1
Row 2
Row 3
Row 4
Row 5
Row 6
Row 7

 

75
75
75
75
75
75
75

 

75
10

5

2

20
W40
W40

75
10

5

2

10
W40
W40

 

 

75
10

5

2

5
W20
W20

 

75
10

5

2

2
W20
W20

 

75
75
75
75
75
75
75

 

35

Table 2.2: The side dimension of each square listed for the features on photomask 2
that is shown in figure 2.9. All units are in microns. W40 and W20 refer to the wire
geometries of 40 microns length and 20 microns length, respectively, shown in Figure

 

 

 

.—s
M

_l
7;
[+1

 

I Rowzllllll
92IIIIII
3IIIIII|

4IIIIII
SIIIIII

\6 IIIIII
\7:IIIIII

 

 

 

1 Sin

 

 

 

Figure 2.9: A diagram of the layout and dimensions of the second photomask. On
this mask, the array was repeated multiple times on the mask. The dimensions of the
features are listed in Table 2.2

the mask that protected the multilayer or contact area. Because the nature of e-beam
lithography is one of much smaller exposed areas and since a reliable negative resist
process is difficult to develop, the e-beam mask instead patterns a secondary mask of
Al that protects the sample in a similar manner as the photoresist. The actual limits
of e-beam lithography are a function of the spot size which was much smaller (5 to 10
nm in diameter) than the feature size (500 to 2000 nm). Even though the spot size was
much smaller than the pattern, there are still parameters which need to be adjusted
to produce well-defined exposed areas. This was done by varying the parameters
of the exposure, such as the line to line and center to center spacing as well as the
dosage, and adjusting or tweaking the layout of the pattern. The e-beam exposure
has a proximity effect in that regions near the electron beam also get exposed. In
order to establish a well defined pattern, a small amount of trial and error is needed.
To select the correct exposure setting a line to line distance was chosen. Then a series
of identical patterns was made using dosages from 200 to 500 [1C / cm2. This process

was repeated for the next line to line spacing to cover all the combinations of line

36

to line spacing and dosage. In this manner the combination of 247 A line to line
spacing and 300 iiC/cm2 dosage were selected as producing the best pattern. Since
the pattern itself was simple and relatively large compared to the beam spot size, no
further modification of the writing pattern was needed to correct for proximity effects.
To create this mask, a layer of 9% copolymer in chlorobenzene was spun at 3000 rpm
onto the sample and baked for at least 1 hour at 155°C. (Copolymer is polymethyl
methacrylate (PMMA), 495k molecular weight, and 8.5% methacrylic acid (MAA).)
The sample was spun again, this time with 2% PMMA (in chlorobenzene) at 4900
rpm, and again baked at 155°C for at least 1 hour. The bilayer resist was then exposed
in a JOEL 840 electron microscope with a dose of 300 [LC/(31112 for features that were
in the .5 - 2 micron range at a 5 pA current. Since the current varied from session to
session, it was measured first, and the writing system would then adjust the timing
to produce an exposure of 300 pC/cm2 at 1000X. The sample was then developed for
65 seconds in a 1:3 methyl isobutyl ketone (MIBK) to IPA mixture. The copolymer
is much more sensitive to secondary electrons from the multilayer and substrate than
the PMMA, and thus more of the copolymer is exposed, producing a large undercut
in the resist system after developing (Figure 2.10). After developing, the sample was
rinsed in IPA for 25 seconds and water for 30 seconds. The samples were then dried
and placed in a Edwards evaporator to have aluminum deposited.

Aluminum was evaporated at setting 2.2 A, giving a deposition rate of 6-10 A per
second. An Al layer, 150 nm thick, was deposited onto the sample. .After removal
from the deposition system, the sample was placed in acetone at 40°C to remove the
Al that was not in contact with the surface. After approximately 1 hour, the excess

Al was removed and the mask was complete for further processing procedures.

37

 

Sample on Si Substrate

Figure 2.10: The bilayer of PMMA and Copolymer after evaporation. The devel-
opment removes the copolymer more than the PMMA causing the undercut. The
difierence in removal is due the copolymer’s stronger sensitivity to the backscattered
electrons from the substrate. This allows the deposited Al feature, patterned from
the PMMA, to be isolated from the rest of the evaporated material. The bilayer will
be removed in the next step by Acetone, and only the Al on the sample will remain.

2.4.3 Etching

Once a sample had a lithographic mask in place portions of the thin film were
removed using one or both of the following techniques. Reactive Ion Etching (RIE)
was the only practical choice to remove thick (> 200 nm) Nb layers. Ion milling

removed thin (3 20mm) Nb and any other thin metal film.

Reactive Ion Etching

Samples to be reactive ion etched were taken to the Plasma-Therm BatchTop
Reactive Ion Etcher. This machine is capable of producing flows of SF6 from 0 to 50
sccm and had a base pressure of 20 mTorr. Before the etching run, the entire chamber
was cleaned to remove debris from previous samples. The cleaning procedure began
by pumping down the vacuum chamber of the RIE to 20 mTorr. SF6 was then bled
into the chamber at 5 seem. The RIE was turned on with a power of 50 W. This
initial cleaning etch was performed for 10 minutes to remove any leftover debris from
previous SF6 etches. After this, the chamber was purged with N2 gas and again
pumped to the base pressure of 20 mTorr. 5 sccm of SE; and 02 each were bled
into the chamber. The plasma cleaning was started with a power of 50 W and again

allowed to run for 10 minutes to remove hardened debris that SFG alone could not

38

remove. The chamber was then purged and bled with pure 02 at 5 sccm. This final
02 etch was designed to remove any organic material left in the chamber and was
done with a power of 20 W for 10 minutes. This entire procedure was successful
at providing a clean system, although an ideal finishing step would be a physical
bombardment with Ar-plasma. (Ar gas was not installed on this system.) (Although
developed independently, a more detailed discussion of why this cleaning procedure
works is described elsewhere [32].)

After purging the chamber with N2 gas and raising the pressure back to an at-
mosphere (again with N 2 gas), the sample was placed in the vacuum chamber of the
RIE . The top Nb layer was removed by using a 5 sccm flow of SF6 at a pressure of 60
mTorr with a power of 50W. The etch was run until the multilayer was viewed to have
changed color—a sign the Nb was completely removed. The actual rate of etching
was not calculated, because when the Nb was removed, the etch will not continue;
SF5 will not etch copper or Au under these conditions. Since SF6 also reacts with
Si, SiO and SiOg, sapphire substrates had to be substituted for silicon to prevent
redeposition of Si etch byproducts on top on the multilayer.

This redeposition of etched Si onto the substrate, including the multilayer, led to
the development of type-I samples where no processing of the cap was needed. After
fine tuning the cleaning process and switching to sapphire substrates, successful type-
11 samples were made using RIE. However, the RIE still did not give consistent results,
so the process of etching Nb was moved to ion milling, which gave more consistent

results.

Ion Milling

Samples which were ion milled were placed in holders consisting of a mask / sample
holder, copper heat sink, and magnetic disk. The holders were placed in a vacuum

chamber containing an evaporation boat for SiO and a 3 cm ion source for ion milling

39

]] Nb Au Cu CoFe Py FeMn

10mA 1.72 9.75 7.75 2.70 3.92 5.20
20mA 3.92 38.5 25.8 9.0 13.1 17.2

 

 

 

 

 

 

 

 

Table 2.3: Ion milling rates in A/s for various metals at 10 mA and 20 mA current.
For 10 mA, The Ar flow was 3 sccm and for 20 mA, 6 seem.
(Commonwealth Scientific Corporation). The system was continuously pumped down
to < l x 10‘6 Torr. Samples were loaded through a load lock with a magnetic arm.
Ion milling was performed at a base pressure < 5 x 10‘6 Torr. To start the ion
milling, Ar gas was bled into the chamber at 3 or 6 sccm giving a pressure of 3 x
10"4 Torr. A shutter shielded the samples while the ion gun was started. The current
was first increased to the desired level, either 20 or 10 mA and the voltage was then
increased to 500V. When ready, the samples were rotated via the sample plate over
the ion mill with the shutter open (Figure 2.11). Although sample heating took place,
the temperature of the substrates was measured to remain below 40°C for 3 sccm of
Ar flow and 70°C for 6 sccm Ar flow. Ion milling rates were calibrated beforehand,
and the sample exposure was timed to mill to the desired thickness. Ion milling rates
were measured by etching single layers of material patterned with a photo mask and
then etched for a predetermined amount of time. After etching, the photomask was
removed and the resulting profile height measured with a profile stylus. This was
done for multiple time intervals for each material, and then a linear regression was
plotted to calibrate the milling rate [33]. Those rates for specific conditions are listed
in Table 2.3.

2.4.4 SiO Evaporation

Once the samples were etched, an insulating layer of 810 was evaporated to pla-
narize and insulate the microfabricated features from the as-yet undeposited top
Nb-layer. (The material is only nominally SiO; it actually is SiOx, with x=1 to 1.5,

measured via its index of refraction [34].) If one uses the same deposition rate and

40

 

 

' Load Lock

Feedthrough for SiO

r

 

 

 

Sample Plate
\
GT 5
E " j

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Shutter :3
:2 SiO
Ion Mill
(*4 - ...)

 

Figure 2.11: The setup of the small chamber. The samples can be rotated to the
desired position on the sample plate. A shutter, independent of the sample plate, can
also be rotated to an open or closed position depending on source to be run. The
magnetic arm for loading samples through a load lock is in the upper right.

41

pressure, fairly reproducible results can be achieved.) Deposition of SiO took place
in the same chamber as ion milling (Figure 2.11 ) at pressures < 2 x 10"5 Torr.
A special tantalum boat (Figure 2.12) was slowly heated at a rate of .15 A per 30
seconds (primary current) using a variac. When the primary current reached 2.66
A, the boat was sufliciently hot to evaporate SiO. The SiO was 99.99% pure (metals
basis) 3 to 6 mm diameter rocks from Alfa lEsar. The SiO was placed in one end of
the boat. Between the SiO and the chimney of the boat were a series of baffles which
prevent spitting and provide a uniform deposition rate. (Much of the background for
this setup can be found in an article by Blevis [35].)

Heating the boat by the above method gave a deposition rate of 15 to 20 A/s.
Above the boat chimney was an additional 10 cm long chimney to prevent contami-
nation of the other devices in the vacuum chamber. The same shutter protecting the
samples during ion milling was also used to start and finish the SiO deposition. Each
sample had a deposited SiO layer thickness of 200 to 350 nm to provide insulation
from top contacts. During deposition, the sample was spun via the magnetic arm at
a rate of 60-100 rpm, and the evaporation source was offset at a 15 degree angle with
respect to the normal of the substrate to prevent the formation of pinholes (Figure
2.13).

Lift Off

Once the microfabracited features are planarized, the mask is removed by a pro-
cedure known as hft-ofl. Lift off or removal of the photoresist (Al for e-beam samples)
was done in a class 100 cleanroom. The samples with photoresist for a mask were
placed in an acetone bath, while samples with an Al mask were placed in a bath
of KOH (Shippley 452 developer). Each sample with photoresist was individually
cleaned with a acetone dipped Q-tip to remove the hardened photoresist. (It was

sufficient to immerse samples with an Al mask in KOH for 5 minutes.) The sample

42

Top View

source Source loading
chimney Area w/ lid in place

 

 

 

 

 

 

 

 

 

Connections to

| “‘ current (heating)
— sourcel y-

II I |_

 

 

 

 

 

 

Baffles Source material

Side View

Figure 2.12: The tantalum boat for SiO evaporation. The SiO is placed in the
loading area. The entire boat is heated using a current source. When the SiO
begins to evaporate it must flow through a series of baffles that prevent spitting. The
evaporated SiO then escapes vertically (side view) toward the sample.

43

. fl

Magnetic Arm
.. 60-100 rpm

 
 

 

 

Sample r“

15" 0
SiO

 

Figure 2.13: SiO,c deposition. The insulator is evaporated at a 15 degree angle with
respect to the sample-surface normal. This and the rotation help to eliminate pin-
holes.

44

was then visually inspected using an optical microscope at 20-100X to ensure that
the mask was removed. After confirmation that the bulk of the mask was removed,
the samples were again immersed in 40°C acetone or room temperature KOH and ag-
itated in an ultrasonic cleaner for 15 minutes. Samples with an Al mask were rinsed
with water and immersed in acetone (40°C) for 15 minutes. All samples were then
rinsed (with IPA) and immersed in 40°C IPA. Again, the samples were agitated using
an ultrasonic cleaner. Finally, the samples were rinsed in deionized water and placed
in 80°C deionized water, agitated for 10 minutes in the ultrasonic cleaner and allowed
to cool to room temperature. After this, the samples were taken for a final sputtering
deposition of Nb superconductor. Since the contact of the Nb to the multilayer was
critical, the Nb deposition was done as soon as possible after lift-off to help prevent

contamination processes which might increase the contact resistance.

2.5 Measurement Equipment

For a fully prepared sample, MR measurements were made by connecting three
separate pillars to the measuring apparatus. Although both a nanovolt and SQUID
null detection system were employed, the connections from the sample to both systems
were identical. The contact (and thus area of current flow) to be measured (#1) was
connected to both a voltage (V+) and current lead (1+). A second contact (#2) was
connected to the other current (I—) lead. The third contact (#3) was connected to the
final voltage lead (V-). This allows current to pass from contact 1 to contact 2 (via the
bottom Nb superconductor) without creating a potential drop in contact 3 (Figure
2.14). Placing two leads on each contact allows one to permute the leads connected to

the voltage and current measuring devices and thus measure each individual contact.

45

V+ 1+ I- V-

#1 l I #2 I #3 I SiO
Nb /
lrll [III II '

. \\\\\\ \
\ Multilayer

” Nb Superconductor

 

 

 

 

 

 

 

 

 

 

 

 

 

Figure 2.14: Measurement setup and geometry of the sample. Input current and
voltage leads are attached to the contact to be measured (#1). Output-current
and second-voltage leads are attached to #2 and #3 contacts, respectively. Current
flows between contacts #1 and #2 (via solid arrows), and the voltage contact at #3
measures the potential of the bottom superconductor.

2.5.1 SQUID Based Null Circuit Voltage Measurements

Most field dependent resistance measurements were done on a low temperature
probe containing a supercondcuting quantum interference device (SQUID) null circuit
and a superconducting magnet. Samples were cooled to 4.2 K while connected to a
SQUID feedback circuit. This potentiometer circuit allows measurements of resistance
in the nil range (Figure 2.15). A known current, I A is applied through the unknown
resistor Rt. The SQUID, coupled to an inductor between points A and B provides a
very sensitive feedback current I p3 which balances the circuit so that points A and
B are at an equipotential. The feedback current through the reference resistor R38,
(= 100 p52) is then measured with a 10 k (2 feedback resistor (not pictured) in series
with the reference resistor. The equipotential balance gives R; x I A = R33, x I 173.
R, can then be calculated. Because of the sensitive nature of the measurement, all
leads connecting R, and Rae; are superconducting.

The applied magnetic field in the probe can range over :i: 1.5 Tesla. This was

46

to SQUID electronics
A A
I L
| F
(mm

 

 

 

 

E

ref

IA IFB

To SQUID
electronics

Figure 2.15: The SQUID null feedback circuit. An RF SQUID system (SQUID,
inductor, RF LC circuit, and semiconductor electronics) balances the circuit so no
current flows from point A to point B. The voltage drop across both resistors is equal
in this balanced state. Fi'om Ohm’s law, one can then solve for R, and in terms of
IA, IFB and R38}. IA is applied to the system and is known. This applied current
causes the SQUID system to go out of balance. The feedback current from the SQUID
system, I F3, is measured by a resistor in series with the reference resistor, RRef. R38]:
is a precision resistor with a known resistance.

47

accomplished by applying a known current through a superconducting coil operating
in persistent mode (the sample sat inside the magnetic coil). The magnetic field was

calculated by using the known coil constant (533.2 Gauss/ A) of the magnet.

2.5.2 Nanovolt measurements

A new multiplexer system with 12 current channels and six voltage channels was
built for four point MR measurements on samples with resistances > than 100 p0
(Figures 2.16, 2.17, and 2.18). A Keithly 2400 current source supplied current while
a Keithly 2182 nanovoltmeter measured voltage. The sample was measured in liquid
He (4.2 K), and measurements were taken over a wider range of currents than was
possible with the SQUID system. The multiplexer system was designed to allow
different samples to be measured on a single substrate without removing the probe
from the liquid He dewar.

The probe consisted of six current and voltage Cu leads (12 total) that were con-
nected to the external measuring equipment at room temperature. Each. connection
included shorting and grounding (including direct to ground and ground through a
1 M0 resistor) switches for each individual wire. These switches were installed for
future applications where static charges could have enough energy to burn out narrow
wires or features made by e-beam lithography. Each lead was then connected to the
multiplexer system. The current leads were connected via a Fisher seven-pin connec—
tor, while three all-Cu fl-Engineering thermocouple connectors were used to connect
the voltage leads (2 leads per connector, plus a ground connection). This all-Cu wire
and connection system minimizes voltage offsets and drifts due to the thermoelectric
effects at room temperature.

The Keithly 7001 multiplexer with 7011 card was used to switch the current
leads, while the voltage leads were connected to a Keithly 7168 low thermal offset

card. Due to the layout of the 7168 card only eight channels with low thermal offsets

48

were available for voltage multiplexing which meant that half of the channels were
hardwired as voltage high (or V+) connections and half hardwired as voltage low
(V-). The current connections had 40 channels and each wire could be connected to
the current high (1+) or low (I-) channel.

As shown in Figure 2.17, the measuring probe fits inside a separate sheath for low
temperature measurements with a superconducting solenoid of coil constant 506.6
Gauss / A. For room temperature measurements the probe can be separated from the
sheath and inserted into a standard electromagnet.

A LabView program was written to automate the measurement process. Using
the already existing computer code from the SQUID system to control the persistent-
mode magnet, seven measurements were made at each field setting by measuring
DC voltage versus DC current, where the current was reversed each time to null-
out thermoelectric offsets. In addition, each sample was given an overall current

dependence check to ensure that the resistance was not current dependent

2.5.3 Area Measurements

The areas of contacts were measured using a JOEL 840 electron microscope. Sam-
ples were photographed at an appropriate magnification and areas measured using
IP Lab software or a vernier caliper to measure dimensions and compare with the
distance marker placed by the microscope. (N o self-consistent measurements were
possible.) These measurements were only good to 10% due to operational parameters
of the microscope [36]. Accuracy of the microscope is dependent on keeping parame-
ters such as the beam current, aperture size and magnification identical from image
to image. For our purposes this was not feasible. Every effort was made to minimize

these effects and obtain reproducible results.

49

 

+— Low Temperature

   

 

 

Figure 2.16: The circuit layout of the nanovolt probe. A Keithly 2400 current source
(1) is connected to a Keithly 7011 multiplexer card (2) with two electronically isolated
banks of connections that allow each wire coming from the probe to be selected as
either high or low in combination with any other wire. A Keithly 2182 nanovoltmeter
(3) is connected to the voltage leads. However, the Keithly 7168 multiplexer card can
only support one voltage connection per voltage lead. Thus half the voltage leads
are hardwired as V+ and the others are V- (4a,4b) The wires are then connected to
the probe (5,6; see Figure 2.18). Finally the leads are taken to the low temperature
portion of the measurement in the measuring probe and connected to the sample.

50

 

 

E :l

 

 

 

 

 

 

 

 

 

 

   

1 connections V connections
Quick connect
hda net . .
g :]Pers18tant sw1tch
leads . .
Connections (w1res
not shown)
Persistant ”A
Switch
L Sample holder

 

 

 

 

 

 

 

Magnet

 

Figure 2.17: The modular measuring probe used for nanovolt measurements. The
magnet, magnet leads, and persistent switch connections are self-contained in a sheath
that connects to the I-V portion of the probe via a quick connect. This sheath
surrounds the probe and the connection is placed at a pre-determined location to put
the sample in the middle of the magnet. The measurement portion of the probe can
be used with or without the magnet portion.

51

Connections

 

 

 

 

i

 

 

 

J

 

+++ii+
__J
I.

 

|

 

 

 

 

 

 

 

 

 

Shield

 

 

\J/
V

To Sample

 

 

Figure 2.18: A layout of the external connections to the measuring probe (one side
only, see items 5 & 6 in Figure 2.16). For both the current and voltage leads, each
wire is connected to a grounding system individually. The grounding system is also
wired to a switch which chooses between a direct ground or grounding through a 1
MO resistor to minimize currents induced by static charges. The shield for each cable
is directly connected to ground. Outside the probe, all current leads are inside a
single shielded cable and each pair of voltage leads has its own shielded cable

52

Chapter 3: CPP-Multilayers with

Micron Size Top Contacts

3. 1 Introduction

Since the initial work of Pratt et al. [14], CPP multilayers have been the source of
intense study. As mentioned previously several different methods have been developed
to cope with the inherently low resistance of the multilayer due to the very short
current path. These typically fall into two major categories: samples which employ
superconducting contacts; and those which employ lithography.

Recently Cyrille et al. [22] made initial attempts to join lithography and super-
conducting contacts in hopes of combining the advantages of both. However, the
complex lithography involved gave a very poor yield of useful samples. Part of this
difficulty was due to the geometry—many (100) lithographed areas were connected
in series, and a single bad contact could render the entire chain useless.

A new geometry was developed here [37], that combines simple lithography with
superconducting contacts. In addition to a simplified geometry, the size of the con-
tacts can be reduced to the physical limits of the lithography equipment. This should
increase the resistance of a single contact to a point were a nanovolt system may
be employed to perform magnetoresistance measurements. The process employed to

fabricate these samples (type-I & -II) was described in Chapter 2.

53

3.2 Initial Work and Tests

3.2. 1 Lithography Test

In order to test the initial fabrication procedure, several trial multilayers were
made of the form: Nb(250) / Cu(100) /C091Fe9 (x) / Cap / Nb(250) , all units in
run, x = 10-60, Cap was type C; ( Au (15 nm)) or C” (Au (10 nm) / Nb (20-50 nm)
/ Au (15 nm)). The thickness of the CoglFeg layers was varied in order to measure
both the resistivity of the C091Feg layer and Nb/C091Fe9 interface resistance, since
the initial assumption is that the interlayers of Au will be superconducting by the
proximity effect.

The results of the test are shown in Figure 3.1 [37]. From the 2CSR model, one
can model the resistance as twice the Nb/C091Feg interface specific resistance plus

the specific resistance of the C091Fe9 single layer:

AR = 2ARNb/CoFe+ARCoFe (31)

= ZARNb/CoFe + PaoFetcoFe (3.2)

The resistivity of C091Fe9 can be found by plotting the specific resistance versus the
layer thickness where the resistivity is the slope of the line and the Nb / CoFe interface
resistance is half the intercept of the y-axis.

The resistivity of C091 F69 in all cases agrees well with previously reported data [38].
Here the value is calculated from a linear regression for the two sets of samples with
multiple values of COglFeg layer thickness. For the type-I samples, the resistivity was
71 :l: 5 nflm. Type-II samples gave a value of 61 :l: 8 nflm. The previously reported
value of 70 :l: 10 nflm was performed using Van der Pauw techniques.

The N b / C091 Feg interface resistance is also in agreement with previously reported

results. Type-I samples give 2ARNb/Cope = 8.0 :i: 0.1 fflmz. The data for ultrasoni-

54

 

147'tvj'l-I'I'rrf' '1"'."""""t

 

-0— Type I
+ Type II
A Type I w/ ultrasonic lift-off

 

 

0 jqu-l-Ill-I'll-llllllluljl-lr

0 10 20 30 40 so 60
ComFe9 Thickness (nm)

Figure 3.1: A plot of the specific resistance for single layer C091Fe9 samples versus the
C091Fe9 layer thickness. The circles represent samples with a C; cap that underwent
a cleaning procedure without ultrasonic agitation. The triangles represent samples
that did have ultrasonic agitation (C 1 cap). The squares represent a C 1 1 cap with
the original (non-ultrasonic) cleaning.

cally agitated type-I samples extrapolate (using the slope for previous type—I samples)
to ZARNb/Cope = 6.4 :E .4 ffimz. Both type-I values are not significantly different
than the previously published value of 7.0 d: 1 film2 for the 2 x Nb/COglFeg inter-
faces [38]. However, it appears that the ultrasonic agitation lowered the Nb/C091Feg
interface value, indicating that the final cleaning step is most likely very important
to produce a clean-interface resistance [37].

It was concluded from earlier studies [29,30] that the middle Nb layer (in Cu) is su-
perconducting at 4.2 K, and thus most of the CPP resistance will come from the metal-
lic multilayer stack that was sputtered under UHV conditions. Since ZARNb/Cope
= 5.1 :i: 0.2 film2 (solid squares in Figure 3.1) for such type-II structures, it appears
that direct contact between CoFe and An (10 nm) (in place of 10-nm-thick Cu in 1
x 1 mm samples) lowers the effective Nb/CoFe interface resistance [37].

In addition, as previously mentioned, the ability to perform self-consistent mea-

55

surements in the electron microscope was limited, and thus the accuracy of our mea-
surements is at best 10 % (verified by the manufacturer) [36]. However, because
most of the measurements were taken with similar setups: constant aperture size;
constant working distance (focal length); similar magnification; and similar current,

the precision of the measurements is probably much better than 10%.

3.2.2 Current Dependence

For superconducting contacts, the CPP current density J 0 pp will likely be limited
to a maximum of ~5 x 10° A/cm2 (observed in our laboratory for sputtered Nb
films [39]). Below this range it is expected that there will be no current dependence
in the resistance of the sample. In order to confirm this, the resistance of the sample
was measured using our nanovoltmeter-based 4 point V-I technique. Only the smallest
samples have resistances large enough to be measured with this method. Because of
the limited nature of the SQUID feedback. system, a large range of currents was not
available in that setup, especially at the high current densities.

The highest the current density will most likely occur is in the vias of the samples—
between the top Nb contact and the multilayer. (Figure 3.2). Because of the ~100-um
penetration depth of our Nb [39], current will only flow along the perimeter of the
vias. Thus the cross sectional area for current flow is roughly given by the product
of the penetration depth and the perimeter of the contact (feature size x 0.4 pm”).

The measurements for several COglFeg EBSV samples are shown in Figure 3.3.
A secondary judge of the lithography is how close to the critical JCPP of Nb can a
contact sustain a superconducting current. The 9.5 pm2 and 28 pm2 samples exhibit
a significantly lower critical J 0 pp of 105 A / cm2, but these samples were fabricated
in the earlier stages of the lithographic processing. In contrast, the 4.5 pm2 data
do approach the expected critical J 0p p of the Nb without significant increases in

resistance. This shows that the lithography again is not significantly affecting our

56

 

 

Figure 3.2: An explanation of vias (black areas with white arrows). The portion of
the Nb (white, black for vias) where the current (block arrows) flows from the Nb that
lies on the SiO (grey) to the portion of Nb in contact with the multilayer. The current
only flows along the edges because the penetration depth of the superconducting Nb
is ~100 nm.

ability to perform CPP transport studies with a nanovoltmeter.

Ideally, one would like to take advantage of the ultra-sensitive SQUID based mea-
suring system mentioned in Chapter 2 (Figure 2.15) for these small—contact samples.
However, the electronics of the SQUID setup limit the feedback current to ~ 1 mA.
The current through a 1 mil sample (a typical resistance for the smallest contacts)
will then be 0.1 mil x lma / 1 m9 = 100 11A. If one increases the reference resistor
to 1 m9, the sample current will then be increased to 1 m9 x 1mA / 1 m9 = 1 mA.
For a 1 pm2 contact this gives J CPP = 105 A/cmz, an acceptable current density even
for the worst of the samples shown in Figure 3.3. Since the voltage across the sample
is now 10 times larger than before, the precision of the measurement is significantly

increased.

57

 

 

 

 

 

 

 

 

 

6
10 : WTTFIIIIIF TITIIIIII I IIIIIII! I IIIIWT! I IIIIIIII I IIIIIIII I II III:
; 2 E E I
—- . I : -1
9.5 mm Area, EBSV tC Fe= 18 run an: Damn”;
— o . : -.
5 D 28 2Ar i
10 :- 11m ea, -------- "j
I Single Layer t = 30 nm ‘ ' 1
: CoFe . A 1
A >- A 2 E ’ . —:
(:1 45pm Area, EBSVt =30nm ; : .
v '- COFC : i A
8 4 ’ 5
c 10 _. -------- -:
m : : : Ii
.... )- . . _,
.‘L’ - ° : -
a) - : . -
o L : I -
I :0 5
. ........ _
1000 E ‘ O b o o o o o {- E 3
.— . 0 ' j
_ 15’ ; 1
. a = I
a W
_ c1 DUAEIIIIIIID: A g .i
1m 1 11111111 1 11111111 1 11111111 111111111 L11111111 111111111 1 11 111
4 s 6 7
l 10 100 1000 10 10 10 10
Current Density (A/cmz)

Figure 3.3: The current dependence of several of the smallest contact areas for C091 Feg
samples. The right hand vertical line represents the expected Nb limit of 5 x 106
A / cm2 [39]. As the lithography improved over time so did the amount of current that
the superconducting contact could support. The sample with 28 pm2 was one of the
earliest samples while the 4.5 nm2 was one of the last fabricated samples. Also note
the high contact resistance of the 9.5 pm2 sample and how it will not support a high
current density.

58

3.3 GMR Transport

Using the information gathered from these first experiments, exchange bias spin
valves (EBSVs) of permalloy and COglFeg were fabricated: Nb(250) / Cu(10) /
Fe5oMn5o (8) / X(t) / Cu(20) / X(t) / Cap; with X = COglFeg (= ‘CoFe’) or Ni34Fe16
(= Permalloy or ‘Py’). The majority of these are type—II samples with a 20 nm thick
Nb layer in the cap. This change was made to allow ion milling of a thin Nb layer
as opposed to reactive ion etching of a thick Nb layer. Initially, RIE was a difficult
process to control and gave inconsistent results due to a leak in the vacuum system,
and the reactivity between the gas used, SF6, and Si which caused redepositing of Si
on the multilayer. Switching to sapphire substrates eliminated the second problem,
but ion milling gave a yield rate of nearly 100% while RIE gave only 60-7 5%.

One of the most important predictions of the 2CSR model is that AAR should
be unaffected by area of current flow. The results of the C091Fe9 and permalloy
EBSVs [37] are shown in Figures 3.4 and 3.5. The variance of AAR with area is
similar to that of sputtered samples with current flow areas of 1 mm2 (data points
along the far right hand side of the graph) . The variances in the 30 nm data in
Figure 3.4 correspond to pillars on different substrates. Each single substrate has
pillars that exhibit only small changes in AAR as A decreases.

Notice the Py data appear to have smaller variations in AAR (especially the 3
nm data). This is due to the implementation of refined lithography process and a
new mask (described in the Chapter 2). The introduction of the higher resolution
photomask greatly stabilized contact area and gave much better lift-ofl results. This
resulted in much better estimations of the area since irregular breaks or rounding of
the edges were much less common.

Most of the variations in AAR can be attributed to 2 sources. First, as can be seen
from the 1 x 1 mm data included at the far right of each figure, AAR sometimes varies

from sample to sample by more than the 5% uncertainty of the areas. Such variations

59

 

I v v vvvv'

30nm(Subctrate l)

 

 

 

 

 

 

o
o 30nm(Substntc 2) .‘
0 3.0111115111131131: 3) 1
A anm .
X 15 nm
e.
b . 9 V 12nm
1.5 ~ 0 I 8 nm(micron). lOnm (mm) ‘«
+ Guru
0
O O o
4‘ 3 ‘ g’
E ‘ 4
g l __ v xv . '-
3 v F
I I I
+.
0.5 r i
+ .
+
+ .
o ....-1 1 . ‘ A...11
1 10 100 moo 10‘ io’ 10‘
Contact Am (mm)

Figure 3.4: AAR vs Area for several thickness of C091 Feg . The variations in AAR are
similar to those seen in previously published data for samples with areas of current
flow of 1 mm2 (far right). Note that while the 30 nm data has large variations,
individual substrates showed more consistent data, except for substrate 1 which had
a poorly defined contact area for one of the contacts (poor-lift-off)

 

 

 

2.5

E I

: i 0:

2 .' 8 8 A'.

A ' S . :

"a '. .‘

g 1" - o 311mest 1

v i A 6nmPyEBSV § .

3 11A 1211mest A.‘

: O 24nmPyEBSV :

0.5 E O . Q Q g“:

0 WM

1 10 100 1000 10‘ 10’ 10‘
Contact Area (pmz)

Figure 3.5: AAR vs Area for several thickness of Py. The variations in AAR are
similar to those seen in previously published data for samples with areas of current
flow of 1 mm2 (far right hand side).

60

18.2 . . . . . . 0.37s

. . . , . . . ,
I—Il—lxlmmContact I
-—.-5x5meontact

 

 
  

  

u
N

b
I
I

a
Q
be

   

§§§

CPP-Resistance (mil)

8
01

5

in
."
&

 

CPP-Resistance (n0)
3
a

in:
Q
h

       

I I I I I I I I I I I I I I J I

-100 . -50 0 so 160
Field (06)

p-s
\I

Figure 3.6: Minor hysteresis loops of CPP resistance versus magnetic field for one
of the smallest contact-area Py samples compared with a 1 x 1 mm sample (t p" =
6 nm for both samples). The saturation and coercive fields have been affected by

processing.

are probably due to subtle changes in the sputtering conditions during deposition.
Second, irregular holes or improper lift-off makes the contact area difficult to measure
for the lithographed samples. When the lift-ofl of the resist is clean, the area is very
easy to measure as it is a square with well-defined borders.

During the study of Area versus AAR, it was discovered that while AAR was
reasonably constant the magnetic properties were affected by the fabrication process
[37]. Ion milling increased the saturation and coercive fields of Py EBSVs when t F”
was less than 12 nm. (Figure 3.6). The change was not as significant in samples that
were: (1) reactive ion etched; (2) C091Fe9 samples; and (3) Py samples with Py layer
thickness greater than 12 nm. Better heat sinking of the sample during ion milling
also reduced this change. An extra piece of silicon substrate was placed in the sample
holders (Chapter 2) to be sure the sample was thermally anchored to the copper heat
sink.

A final test of the transport properties was to demonstrate that the spin diflusion

length remained unchanged in the new samples [37]. Previous work on Ni84Fe16and

61

18.2 .....,...,...,..... 48

 

H
a

 

 

 

 

 

 

 

 

 

 

q A
V 17.8 463
3 8
g 17.6 45 g
:3 8
95 17.4 4405
g, In
g. Q:
O 17.2 430
17 _________ .11. n L J n n 14 A 7 42
.40 .20 . o 20
F1eld(0e)
16.8~...f,r.--,..-. ....118
16.6 -: 1.78A
g 16.4 E ”6%,
0 I
g 162 .: 1.74%l
g 16 .: 172 '8
J
D! 15.8 1 17 E
m d
1.684.
E3 15.6 1, 0
+1 x1 mmContact (20 nm Co 9lFeg) . 1.66
15.4 -o-2x2meomact(18nmCo :Fe,) 1
. n n I n l j n n I l n n n n l L L ‘ I.“
.100 .50 so 100
Field (0e)

Figure 3.7: Minor hysteresis loops of CPP resistance versus field for (a) t py=15 nm
and (b) toope = 18 nm samples. Here the magnetic properties are very similar to
those at 1 mm2 samples.

62

 

2.57'Vitjviiirriii‘irrf'I‘

Infinite spin diffusion length—‘7’

Finite spin diffusion length (l‘! = 12 run)

I 1x1 mm2 EBSV

 

o I I I I I I I I I I I I L I l I I I I I I I

o 10 20 30 4o
CoflFe9 layer thickness (nm)

Figure 3.8: The effect of the spin diffusion length lsfon AAR. The experimental
results show that l,fis indeed finite.

C091Feg [38,40] showed a finite diffusion length in both materials. This was shown by
comparing AAR and the layer thickness. If the spin diffusion length is infinite, AAR
will continually increase with increasing layer thickness. If la, is finite, then after a
certain layer thickness, AAR will no longer increase. This effect is shown in Figure
3.8 for 1 x 1 mm samples [38].

The results from the lithographed spin valves [37] and those from previous work
[38, 41](1 x 1 mm samples) show good agreement. (Figures 3.9 and 3.10). The
finite spin diffusion length for both ferromagnetic alloys is demonstrated again by
the micron size data where the areas range from ~5100 to 4 pm2. In addition the
variation of AAR for fixed thickness is the same as or better than that for the 1 x 1
mm EBSVs. This improvement is likely due to the fact that multiple-size contacts are
available on the same multilayer (fabricated under identical conditions), while each
1 mm2 data point represents a separate multilayer deposition. These figures clearly

demonstrate that CPP-GMR transport is unaffected by the lithographic processing.

63

 

 

 

 

2.5 . I I I— 1 I I I f I’ ‘firfif f ‘7' I I I I I T .
2 r 1
. I I I ;
..A L I I 1
a 15 . I I o 1
h d
g : I ' ll :
l r- ! O -
a ,, o. d
P I 2 II
0.5 I. I 1x1 mm EBSV :
b . .

r 5' O Micron-Sized EBSV j
o b I I I I l I I I I l I I L I l I I I I .1 I IJ

0 10 20 30 40

CofiFe, layer thickness (nm)

Figure 3.9: AAR versus toope for 1 x 1 mm (solid squares) and micron-size (open
circles) CoFe EBSVs.

 

 

 

 

’ "I""I""I""I""l“"I"
. ‘ I I
2'. o 2 ‘ A Q;
. 6 .
A h I
"a . f .
g 15." A0 1
v I i I
E I: 1
, . -. :
0.5:, O 0 Micron :ized samples 1
z A lxlmmsamples :

o ...In.n.lnnnnjnnnnl-nnalnnnnln

0 5 10 15 20 25 30

Py layer thickness (nm)

Figure 3.10: AAR versus tpy for Py EBSVs with 1 x 1 mm (solid triangles) and
micron-size (open circles) areas.

64

Chapter 4: CPP-Domain Wall
Trapping

Ono et al. showed [9], using a trilayer of NigoFego (5) /Cu(10) /Nl30F€20 (20) and a
CIP geometry, that domain wall trapping was possible in the narrow neck of a wire of
sub micron width. Based on MR measurements as the field was swept from negative
to positive, an intermediate “state” between the P and AP state was achieved where
a domain wall was trapped in a narrow neck of the long wire. Further work showed
that a source of domain walls was needed to provide a reliable method of injecting a
domain wall into the narrow wire [42]. Using a diamond shape F-pad [43] at the end
of a long wire helped to facilitate this injection and the eventual trapping of a wall
at the neck region.

By transferring this experiment into the CPP geometry, it may be possible to
study specific regions of the wire by using localized superconducting contacts. The
contacts in the CPP geometry can be made to lie only in the region of interest. This
allows a localized current to probe the area of trapping, and the resistance changes
from such a measurement as a domain wall sweeps past the contact would allow one
to study the relative position of the wall based on the size of the GMR in the AP and
P states.

The initial geometry of the sample was a long wire with flat ends (Figure 4.1) .
Continuing studies on domain wall formation show that flat ends of the wire are more
likely to produce domain wall nucleation. As mentioned before, a study by the Shinjo
group showed that placing a domain wall ‘source’ [42] at the end of the wire further
improves the likelihood of domain wall propagation. Otherwise it is uncertain where
the wall nucleates in the sample. Additional study showed that a diamond-shaped

pad on one end of the wire is most successful at injecting a domain wall into the

65

sample [43]. This work was not published at the time of the initial photolithography

study but was incorporated into e—beam lithography samples fabricated later.

4. 1 Initial Trials

A series of Ni34Fe16exchange bias spin valves was fabricated as type-III samples.
Applying the previous photolithography work of superconducting contacts, the EBSV
was shaped into a wire geometry using the previously described methods for the high
resolution mask. This time, as mentioned in the fabrication section, all of the EBSV
layers were ion milled (excepting the portions under the photolithography mask).
After planarization with SiO, the entire wire was covered with a superconducting
contact. When measured, these Ni34Fe16EBSVs did not show conclusive evidence
of domain wall trapping (Figure 4.2). Only the larger of the two photolithography
geometries were successfully fabricated at this time. The smaller samples were poorly
defined due to the buildup of photoresist along the edges of the sample which limits
the resolution of the mask aligner. These small samples had ends that resembled
Figure 4.1a; poor choices to study for domain wall trapping.

Previously measured C091 Feg EBSVs, for the study of superconducting contacts
on macroscopic multilayers (Chapter 3), did show some evidence for domain motion
at small contact sizes. Very small, and unrepeatable states were visible at the smaller
contact sizes (Figure 4.3). Since this is comparable to the size of the wire geometry,
the decision was made to use C091Fe9 in the EBSVs for photolithography samples
with both pattern sizes.

Using C091 Feg in the experiment was more successful. An initial experiment with
the ComFeg EBSVs showed some evidence of domain wall trapping. Although not
entirely reproducible, the experiment left a hint of the direction to go (Figure 4.4).

Clearly the domain wall was becoming trapped, not only in the notch but in other

66

 

 

 

 

>< >< >8

 

 

 

 

 

 

\/ _

Figure 4.1: The layout of different wires used for domain wall trapping. (a) A poor
candidate for domain wall trapping as the ends will not easily produce domain wall
nucleation. Many of the initial 2-um-wide samples were shaped like this due to
poor contact in the mask aligner. (b) The flat ends will be better for domain wall
nucleation. This was the goal of the photolithography trials. (0) The best option is
to provide a source of domain walls. The diamond on the end will be multidomain
and nearly always inject a domain wall into the sample. This configuration was used
for e-beam samples.

67

 

      

86 I I I I I I I I I I I I I I [ I I

[I- O
: 4x40wirewith2pmneck
85 - Nb top contact everywhere 5 E . ,, ..

I
l.‘. I
I ' .o‘. . I .
I ' A'O
I I 3“ I
s‘.‘ ' ' " ‘ ................

I I '0. ' -

I I ‘
O [I .
I ' I‘l) ... ‘l
I ' .

82

CPP-Resistance (p.52)

81

 

 

 

 

 

Applied Field (Oe)

Figure 4.2: A Ni34Fe15EBSV (t Py = 60 nm) in the large photolithography geometry.
There is poor evidence of domain wall trapping

68

33.5

 

  

 

 

 

 

 

 

 

__ I F I I I I I I I I I I I I I I 1 T _‘
. ~4OO umz coniact 3 5 AP State .
’- (ZOXZOM) ' .(:;‘_;_"::‘.22127.21'.'.'.;2)){1€:‘.3.";:;" ,':::: ;:.:: :;;;l :3. tt."' ._ . “
33 r """""""""""" """""" :- -------- . . ,
L I I I
32.5 - ---------------------
a 32 :— ---------------------
0 +
O p 1
a 31.5 :- ------------------- ‘
a? 5
a 31 j """"""""""""""" i
U .
30.5 :— ---------------------
30 '_...P. State ........ n
29.5 ' . . 1 1 '
-100 -50 O 50 100

Applied Field (0e)

Figure 4.3: A C091Feg EBSV (tCope = 30 nm) from earlier transport studies (Chapter
3) that gives indications of domain wall trapping behavior. Note the plateau (in the
transitions) between the P and AP states.

69

places along the wire as well. This happened because the photolithography process
left the wire with edges that were not sharp, and the wire was populated with many
small “necks.” This occurrence of multiple intermediate states was evidence that
either the photolithography had to be refined further, or a transfer to an electron-

beam process was needed.

4.2 Improved Lithography

An attempt to improve lithography was made by reducing the buildup of photore-
sist along the edges of the square substrate. Using acetone and a Q-tip to remove
the excess buildup along the edge reduces the contact distance and ensures that the
resist is of uniform thickness over the entire sample (Figure 4.5).

This process improved the overall resolution by a roughly a factor of 2. The
smallest photoresist patterns (2 microns wide with a 1 micron neck) were now better
defined and produced better results (Figure 4.6). Still the results were not as clean
as the original CIP work by One et al. Several non-reproducible intermediate steps
were still observed in—between the P and AP states.

Type-IV samples combine all of the above mentioned techniques with the addition
of a second lithography process that independently defines the top contact area. The
first lithography procedure now shapes the multilayer feature into the wire geometry.
All of the previously described methods are used (unchanged) to produce the shaped
wire. Instead of a final step of sputtering top Nb contact, the sample is again processed
using the same procedures; this time, however, a square photolithographic mask is
used. The square features are placed directly over the notch area. A type-I procedure
is employed to shape a ‘window’ in the notch area, while the rest of the wire is
insulated from electrical contact. Now, only the area in the window is in contact with

the final sputtered Nb layer. This arrangement uses the current flow properties of

70

CPP-Resistance (um

CPP-Resistance (no)

102

101

100

 

lr‘IY—TYY—IVTYV

 

   

I'IVTUVYj'IYIUIIIIIlTTTT‘l‘leIUVU'I'I‘U
I
a ' '

    

4x4dwimwitiazwnnafck . . ';;:=::c::.-;.
..thp.mmactgvctywhete ......... ........... 5 '

nnnnnnnn

 

 

 

    

 

  

99 ...........
98 ---------------------------------------------------------------------------------
97 ..................................................................................
96 ...............................................................................
9s _
94LlLJALLLlLLlJliLALLLillllilgiillLliLlll‘

-200 -150 —100 -50 0 50 100 ISO 200

AppliedFicld(Oe)
1mht.III'Y—IfiTITijIU'IYIFUTflj—FU'!111'!TTTI
E 4x4pwirewigh2mnmcj k g : 4:3”: .. .;-.-.

101 -...hlbtcipcontactcvcrywhpm ....... i. ......... ......

100’

98

95 _.:

 

 

 

...........

...........................................................

 

 

. . , . 1
94 ILALLilllLilllllllllilllLillllllllllllll-J
-200 -50 0 50 100 150 200
Applied Field (0c)

Figure 4.4: The best domain wall trapping sample with the initial lithography. Al-
though decent results were obtained, they were not reproducible, as evidenced by
these two runs on the same structure. This is a COglFeg EBSV with toope = 18 nm.

71

Uniform photoresist Cleaned Area

|
l

 

   

 

 

 

Thicker photoresist

Figure 4.5: Removing the buildup of photoresist with acetone and a q-tip greatly
increased the resolution of the photolithography. Before this improvement, the pho-
toresist along the edges was much thicker and hindered good contact with the mask.

72

161

2 x 20 with 1
Nb top

159

CPP-Resistance (1.10)
5

.—
LII
\I

156

 

155
~150 -100 -50 0 50 100 150

Applied Field (0e)

2 x 20 with 1
Nb top

CPP-Resistance (no)

 

-150 -100 -50 o 50 100 150
Applied Field (0e)
Figure 4.6: A domain wall trapping sample with the smaller of the 2 wire geometries.
Although decent results were measured they were not reproducible, as evidenced by

these two runs on the same structure. This data is from a 0091Fe9 EBSV with tcope
= 6 nm.

73

 

3.62 I r I T ‘l I I I I I I I I I I r T I I I j i T I I I I I r
I 2 x 20 wire with 1 mngneck i 5 3 1
- 5x5 umthop contact : -
3 -6 f' in 'rio't'c'h' region only ' '§ """

3.58 r ............. ........... .............. ,,,,,,,,,,,,,, s

3.56 " ............. ............ ........ AR—Jmé .............. .............

3.54 .............. ........

 

CPP-Resistance (m 9)

3.52 T— .............

 

 

35 r- ----------- . .....

 

 

I ' . a
I ' . , o
348 railipiiLl—LiliiiiiingiILiiim
.

-100 -50 O 50 100 150 200
Applied Magnetic Field (0e)

 

 

 

Figure 4.7: A 5 pm contact on the small wire configuration for a C091 Feg EBSV (toope
= 12 nm). The large plateaus in between the P and AP states are likely the domain
wall trapping. ARA/etch is roughly the same for both intermediate states. Based on
the total AR and the resistance noise, one can locate the domain wall in this sample
to about 10 nm.

the superconductor to study only the notch area.

The results from this modified sample were again better, but still unsatisfactory.
Again, non-reproducible results were shown. However, there were indications that the
samples did exhibit some behavior of domain wall trapping. As shown in Figure 4.7,
there is a wide plateau located in equivalent spots on opposite sides of the hysteresis
loop. There are still smaller plateaus located seemingly at random. The largest steps,

though, are located in proportion to where the square contact is situated in the notch.

74

 

Figure 4.8: An SEM picture of a Type-IV sample. Notice that the top contact
(square) is off center. The location of the notch in top contact region determines
where the intermediate state will occur in the MR transition.

Note that the location of the large intermediate state is not half way in between
the P and AP states. This is in agreement with the observation that the top contact
is not perfectly centered over the notch, as shown in Figure 4.8. Furthermore one can
estimate from the data of Figure 4.7 that the standard deviation of the resistance
noise is ~ 240 mil. As the wall sweeps through the region of the 5—pm contact, the
overall resistance change in R is AR = 0.12 1110. Thus one can resolve the motion
of the wall to ~ 10 nm (= 5 pm x 240 nQ / 0.12 m9). Of course, further averaging
of R measurements would improve the ability to resolve small motions in the average
position of the wall. Also making the contact smaller should improve the resolution.
In Figure 4.7, the quantity ARM“), (= 0.12 m9) provides an estimate of the average
motion of the wall while it is trapped. One obtains ~ .5 pm ( = 5 pm x 0.012
m!) / 0.12 m9) which is 50% of the neck length in the sample (Figure 2.3), a very

reasonable result.

75

4.3 E-beam Lithography Samples

Since photolithography had been pushed to its resolution limits with the contact
aligner, the patterning process was adapted to an electron beam procedure. Now a
bilayer resist was used to pattern an Al mask on the sample. That mask shaped
the multilayer by ion milling, and the entire substrate was planarized with SiO. In
addition, since the size of the samples was approaching sub-micron dimensions, the ge-
ometry of the sample was changed to eliminate vias. The sample was instead initially
sputtered with a very thick top Au layer (150 run) that protrudes from the insulating
layer. Contact will then be much easier to achieve with the top superconducting Nb.

Initially samples exhibited significantly increased coercive fields, similar to the
EBSV transport study of Chapter 3. However, this problem was significantly re-
duced by improving the thermal contact between the substrate and its copper heat
sink during ion milling and SiO deposition. Thermal grease was placed between the
substrate and a blank Si substrate in the sample holder to ensure good thermal an-
choring. Also, the blank Si substrate had grease applied between it and the Cu heat
sink. This arrangement was successful in reducing the damage to the samples. The
grease was removed by ultrasonic agitation in an acetone bath.

Also the layout of the sample was changed (Figure 4.9). E—beam lithography
allows flexibility in the geometry of the exposure, and the previously mentioned im-
provements in domain wall injection [42,43] were incorporated into the design. In
addition, the size of the sample has now shrunk to a size where experiments done
before have shown domain wall trapping in Ni34Fe16, so that material may be used
again in an EBSV.

This procedure was successful in producing 1 and 2 mm wide geometries. The
results were promising. As shown in Figure 4.10 and 4.11, a single major plateau
is usually observed in each transition from the P to AP state and vise versa in a

Ni84Fe15EBSV with t p3,: 12 nm. The plateaus are now more reproducible and appear

76

T—bfiH—l um -f—>—<—2 um
20 um 35 um

*—»<<—.5 um *—»(<—lllm

20 um 35 pm

\2/ \9/
A /<

2pm 4pm

 

 

 

 

Figure 4.9: The pattern for e-beam lithographed samples. A diamond at the end
of each wire is patterned to provide a source of domain wall injection. The aspect
ratio of each wire has also been increased to facilitate single domain wires. Ideally
the wide wire would have 40 pm segments, but the pattern would exceed the range
of the writing microscope.

77

 

1.31

 

1.3

1.29

1.28

1.27

1.26

Resistance (m9)

 

 

1.25

 

 

 

 

Applied Field (Oe)

Figure 4.10: A single MR curve for a Ni34Fe16EBSV (1313” = 12 nm). Notice the
intermediate state located nearly at the half way point.

at a point halfway between the P and AP state, as expected. Note that the top
Nb contact covers the whole sample, so the CPP-MR is sensitive to domain wall
motion everywhere. In spite of this, it is gratifying to see plateaus at the expected
intermediate position. Also note the area of the diamond-shaped pad is only 10% of
the total area of the narrow wire.) It appears that this particular size of sample will
yield the results desired (Figure 4.11). However, only by restricting the measurements

to the notch region can fully realize the potential of this experiment be fully realized.

78

 

 

 

dill.

11“

‘1‘

‘111

Cddd Ildd

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

b}:

 

Viki

 

 

PP

 

DIP-P

 

 

 

 

IPh

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

I

'

1

r

I

I

I

IDPPD bib :PP hP-rh PPDP

II ‘1“? “Mid “dun: “undid quid“- Mahdi Iqu-

1
I
l
l

I

I

V

1

'

1

t

T

f

'

I

I

PP?» PP?-

 

 

 

 

 

 

 

 

 

 

 

 

.111

1.11

If
(II.

11-.

l I J l

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

bib. Ph-

 

 

I‘D-DI

 

 

 

 

 

 

 

 

 

Figure 4.11: Repeated MR curves for a single sample 1 pm wide (Py EBSV with t py
= 12 nm). The trapping appears reproducible. Complete curves were not obtained

in some cases due to problems with the electronics .

79

Figure 4.12: The motion of a domain wall (grey) after it becomes pinned by defect
or geometry (black circles). The ends of the wall tend to stay fixed while the center
will bow out.

4.4 Results

For all the results one thing is clear. When the intermediate state is achieved, this
state is not ‘flat,’ but instead the resistance continues to change in same direction as
before, usually at a lesser rate as the field is changed. This could be due to bowing
of the domain wall. When a domain wall becomes trapped in the notch, the ends are
very strongly pinned while the center will move forward (Figure 4.12). This will lead
to the effect described above.

For the larger samples there also appeared to be multiple domains. Co is likely to
become single domain only at large aspect rations [44]. From the information in this
paper, it is unlikely that the photolithography samples patterned for this study were
single domain. However, the e-beam samples are of the proper size and aspect ration
to be single domain. Given the problems associated with photolithography at this
resolution, it is probably best to continue the e-beam study and develop a method to

place top Nb contact only in certain regions.

80

Chapter 5: Conclusions

A successful attempt was made to fabricate samples with areas of CPP current
flow down to the micron size using superconducting contacts, but with macroscopic
sizes for the ferromangetic layers. The sample properties were relatively unchanged.
Although the coercive and saturation fields for some samples were increased, these
changes can be minimized by refined techniques described here. The CPP GMR for
both C091Fe9 and Ni34Fe16, two materials important for applications, is unchanged
in the samples with micron size contacts. The samples showed excellent P and AP
states down to 4 pm2 top contact sizes. Although it may be possible to push the
lithography further by adopting a larger wafer size to ensure a more uniform resist
when spinning, the results may only give an improvement in resolution of less than
50%. It would be better, if further studies are needed, to adopt the electron beam
techniques described for the domain wall trapping to produce top Nb contact sizes
down into the sub—micron range.

The study of domain wall trapping is only in its infancy. Now that a reliable
method for producing samples exits, further studies of the behavior of the domain
wall can be done. From the initial C091Feg work done with photolithography, it
is clear that domain wall trapping was occurring to varying degrees. For the final
electron-beam fabricated sub-micron wires, it is fairly clear that more reproducible
trapping is occurring. I

The final step in this process in to perfect a technique to place the Nb only in
the notch region using e—beam lithography. Then it would be possible to gain further
insight into domain wall trapping by controlling the shape and position of the notch.
It has also been proposed to look for telegraph noise in the CPP-MR while the domain
wall is trapped, a possible signature of domain wall tunnelling between pinning sites.

Now that a procedure exists to produce samples, these studies can be done. So far in

81

this project no magnetic force microscopy (MFM) studies of the trapped walls have
been made. This will be an important study to do for the future.

To accomplish this goal a new sample design (Figure 5.1) is proposed. By using a
new photomask layout along with e-beam alignment, one should be able to select the
areas of the wire to be studied and have only ~ 1 rim-wide Nb top contacts (Figure
5.1.)

82

 

 

 

 

 

 

 

    
     

 

 

(b)
alignment
I marks
B
I Wire
(C) (d)
‘ eatures
(Insulator l/ . [Final top Nb layer
I

 

 

 

 

 

 

Figure 5.1: A proposed top view of the new fabrication to put Nb only in certain areas
of the sample. (a) Photolithography patterned leads are made by sputtering an entire
EBSV onto a substrate photoresist mask. The mask allows features down to several
pm to be patterned. Note: extensions of these patterned leads to m size contact
pads are not shown here. (b) Using the unattached leads as rough alignment marks, a
wire mask feature is patterned on the center lead by e—beam lithography. The sample
is then ion milled to the bottom Nb (Au coated) leaving the EBSV structure only
underneath the mask (including alignment marks). (c) The sample is planarized. (d)
The alignment marks are used to align a pattern (photo- or e—beam lithography) that
will expose the notch region and a connection to the remaining leads. Sputtering Nb
into the exposed region completes the sample.

83

Bibliography

[1] A. Hubert and Rudolf Schafer, Magnetic Domains (Springer, Berlin, 2000).
[2] C. Kittel, Rev. Mod. Phys. 21, 541 (1949).

[3] W. P. Mason, Phys. Rev. 96, 302 (1954).

[4] M. I. Darby and E. D. Isaac, IEEE Trans. Magn. 10, 259 (1974).

[5] C. Kittel, Introduction to Solid State Physics, 7 ed. (John Wiley &, Sons, Inc,
New York, 1996).

[6] N. W. Ashcroft and N. D. Mermin, Solid State Physics (Saunders College Pub-
lishing, Fort Worth, 1976).

[7] S. Chikazumi, in Physics of Magnetism, edited by S. H. Charap (John Wiley &
Sons, Inc, New York, 1964).

[8] P. Bruno, Phys. Rev. Lett. 83, 2425 (1999).

[9] T. Ono, H. Miyajima, K. Shigeto, and T. Shinjo, Appl. Phys. Lett. 72, 1116
(1998)

[10] T. Ono et al., J Appl. Phys. 85, 6181 (1999).
[11] M. N. Baibich et al., Phys. Rev. Lett. 61, 2472 (1988).

[12] G. Binasch, P. Griinberg, F. Saurenbach, and W. Zinn, Phys. Rev. B 39, 4828
(1989)

[13] S. S. P. Parkin, N. More, and K. P. Roche, Phys. Rev. Lett. 64, 23042307 (1990).

84

[14] W. P. Pratt, Jr. et al., Phys. Rev. Lett. 66, 3060 (1991).
[15] S.-F. Lee et al., Phys. Rev. B 52, 15426 (1995).

[16] M. A. M. Gijs, J. B. Giesbers, S. K. J. Lenczowski, and H. H. J. M. Janssen,
Appl. Phys. Lett. 63, 111 (1993).

[17] W. Vavra et al., Appl. Phys. Lett. 66, 2579 (1995).
[18] J. A. Katine, F. J. Albert, and R. A. Buhrman, Appl. Phys. Lett. 76, 354 (2000).
[19] L. Piraux et al., Appl. Phys. Lett. 65, 2484 (1994).

[20] A. Blondel, J. P. Meier, B. Doudin, and J .-P. Ansermet, Appl. Phys. Lett. 65,
3019 (1994).

[21] K. Liu, K. Nagodawithana, P. C. Searson, and C. L. Chien, Phys. Rev. B 51,
7381 (1995).

[22] M. C. Cyrille et al., Phys. Rev. B 62, 3361 (2000).

[23] J.-E. Wegrowe et al., Phys. Rev. Lett. 82, 3681 (1999).

[24] T. Valet and A. Fert, Phys. Rev. B 48, 7099 (1993).

[25] A. Fert, J. Duvail, and T. Valet, Phys. Rev. B 52, 65136521 (1995).
[26] J. Bass and WP. Pratt, Jr., J. Magn. Magn. Mater. 200, 274 (1999).
[27] M. A. M. Gijs and G. E. W. Bauer, Adv. Phys. 46, 285 (1997).

[28] A Barthélémy, A. Fert, and F. Petroff, Giant Magnetoresistance in Magnetic
Multilayers (Elsevier, New York, 1999).

[29] C. Fierz et al., J. Phys.: Condens. Matter 2, 9701 (1990).

[30] J. Y. Gu, R. D. Slater, and William P. Pratt, Jr., unpublished.

85

[31] J. M. Slaughter, W. P. Pratt, Jr., and P. A. Schroeder, Review of Scientific
Instruments 60, 127 (1989).

[32] D. J. Adelerhof et al., Physica C 209, 477 (1993).

[33] K. Eid, R. D. Slater, and J. A. Caballero, unpublished.

[34] L. I. McCann, unpublished.

[35] E. H. Blevis, R. D. Mathis Comapny .

[36] R. D. Slater, personal conversations with JOEL Tech Support.

[37] R. D. Slater, J. A. Caballero, R. Loloee, and W. P. Pratt, Jr., J Appl. Phys.
Accepted, December (2001).

[38] A. Reilly et al., J. Magn. Magn. Mater. 195, L269 (1999).

[39] M. Oonk, Master’s thesis, Michigan State University, 1997.

[40] S. D. Steenwyk et al., J. Magn. Magn. Mater. 170, L1 (1997).

[41] L. Vila et al., J Appl. Phys. 87, 8610 (2000).

[42] K. Shigeto, T. Shinjo, and T. Ono, Appl. Phys. Lett. 75, 2815 (1999).

[43] K. S. T. Okuno, T. Shinjo, Y. Suzuki, and T. Ono, J Appl. Phys. 88, 6636
(2000)

[44] E. Seynaeve et al., J Appl. Phys. 89, 531 (201).

[45] J. A. Caballero, unpublished notes.

86

111]]!1[]]l]]]]][1]]][11[1]]]11