"fa-$2? as? 23M
“32:94.1 ragga»
33: V‘Tﬁaﬁi

' W'h
r‘v’“

 

 

 

 

 

 

'13

'4

k-A ' -.-.-. .
."”-'. ‘I‘

L-

n
q» ,.....-»‘ v n ..

 

 

.-. (.

mun"
HM-.«--

 

 

(.2,ng I
.Ezgmki

N. 7.‘
.2 I)- (“I

———————__..... —
o‘- v-—. r.

LI mi I: Y
I’licrtgaa Stye

I
".r,t ; ‘ -""A- ‘4. 0
53“." “1"". I
uni. \t..~l‘.,1g‘pl
J

 

 

 

This is to certify that the
dissertation entitled

HRTEM AND EELS STUDIES OF NANOSCALE
STRUCTURED ELECTRONIC MATERIALS

presented by

JIAMING ZHANG

has been accepted towards fulfillment
of the requirements for the

Materials Science and
Engineering

Ph.D. degree in

 

 

 

ﬂZ/Kﬂ

Major Professor’s Signature

,7//7/7 7

Date

MSU is an affirmative-action. equal-opportunity employer

._—_—-.--—.-

 

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.

 

DAIEDUE

DATEDUE

DAIEDUE

 

AUHI % 01220433

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

6/07 p:lClRC/Date0ue.indd—p.1

 

 

HRTEM AND EELS STUDIES OF NANOSCALE
STRUCTURED ELECTRONIC MATERIALS
By

Jiaming Zhang

A DISSERTATION

Submitted to
Michigan State University
in partial fulﬁlment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY

Materials Science and Engineering

2007

ABSTRACT

HRTEM AND EELS STUDIES OF NANOSCALE STRUCTURED ELECTRONIC
MATERIALS

By
Jiaming Zhang

This thesis focuses on the growth and structure of a number of nanoscale
structured electronic materials characterized using HRTEM and EELS. Both rare earth
silicide nanostructures self-assembled on Si(001) and sputtered GMR multilayers have
been studied by characterizing their crystal structures, interfacial epitaxy, interfacial
chemical nature, and electronic nature, which provide fundamental insights into material
behavior at the nanometer scale.

High aspect ratio nanowires and nanosized islands have been observed to self-
assemble on Si(OOl) for both the Gd and Tm silicide systems. HRTEM results Show that
Gd silicide nanostructures exhibit either the hexagonal GdSiz.x or the orthorhombic GdSiz
crystal structures, with lattice parameters consistent with the bulk phases. In the case of
Tm, the observed nanostructures are likewise either hexagonal or orthorhombic. The
hexagonal phase has lattice parameters consistent with the bulk, while the orthorhombic
does not. For both systems, bi-phasic silicide structures were observed, which may
reﬂect a mechanism for strain accommodation at the interface with the substrate. In the
case of Gd, the phase with lower strain lies at the substrate. For the case of Tm, the
relative mismatches of the two phases predicted from bulk silicide lattice parameters
disagree with that derived ﬁ'om measured lattice constants, and it is a relaxed
orthorhombic phase at the interface that appears to have the lowest mismatch with the

substrate. EELS studies were carried out to compare the electronic structures of metallic

Gd, thin ﬁlm Gd silicide, and Gd oxide in bulk phases and Gd silicide nanostructures.
The results from the three bulk phases are similar, while the intensity ratio of M52M4 in
the GdSiz nanostructures varies from the bulk, which may suggest that a slightly different
spin state exists in the silicide nanostructures.

As-sputtered and annealed F(Co, or Py)/Al multilayers have been studied using
HRTEM and EELS. Although the interfacial intensity proﬁles from EELS spectrum
images suggest some limited intermixing exists in the F/Al interfacial regions, both
HRTEM and diffraction studies Show no obvious intermediate phase formation. In
particular, the annealing treatments do not signiﬁcantly alter the multilayer structures. In
contrast, intermediate phase formation has been observed in both Cu/Al multilayers and
Cu/Al regions in spin valves. Tetragonal AlgCu and bcc AlCu3 are formed in the
Cu(8nm)/Al(10nm) multilayers, while A12Cu and fee Cu are formed in the
Cu(5nm)/Al(3nm) multilayers. For Cu/Al/Cu layers in the spin-valves, evidence of
AlzCu and AlCu3 phase formation in the annealed spin-valve with the 30m Al layer was
found, while A12Cu and Cu were observed in the as-sputtered spin-valve with the lOnm
Al layer. These results are discussed in terms of the balance between interfacial and

volume ﬁ'ee energies in order to rationalize the formation of non-equilibrium structures.

ACKNOWLEDGMENTS

First of all, I wish to express my gratitude to my research advisors, Professor
Martin A. Crimp and Professor Jun Nogami throughout my entire PhD. research. In
2003, I joined Prof. Crimp and Nogami’s team and started my research on rare-earth
silicide nanowires under their supervision. They spent enormous time and energy to help
me to start as a self-motivated graduate student. I am especially grateful for their
patience and encouragement during my Ph.D. study. I feel very lucky to learn various
advanced microscopy techniques from Prof. Crimp, which I believe will help me in my
future career. I am also grateﬁil for Prof. Nogami’s time and effort to come from Toronto
to East Lansing several times after he moved to U of Toronto to help me with my
research. Without Prof. Crimp and Prof. Nogami’s huge contribution, I do not think my
dissertation is possible.

I am grateful to my committee members, Prof. Case, Prof. Norman Birge and Prof.
Tim Hogan for their valuable suggestions and comments on my research and dissertation.
I also express my appreciation to Dr. Xudong Fan (Center for Advanced Microscopy) for
his excellent TEM technical. I am also grateful to Prof. Bass and Prof. Pratt for
providing me with GMR multilayer samples for my Ph.D. research. I feel very lucky that
I have been able to study multiple systems during my study.

I would also like to thank my previous and current co-workers Dr. Gangfeng Ye,
Dr. Yucheng Lan, Yan Cui and Dr. Chigusa Ohbuchi. I enjoyed working with them, and

their help, encouragement, and humor will remain in my memory.

iv

Finally, I would thank my family members, my wife Ran Mu, my parents and my
in-laws. They have spent extensive time and effort to support me and take care of my
newborn daughter Chelsea Zhang. Their immeasurable love and understanding are

overwhelming and go beyond words.

TABLE OF CONTENTS

LIST OF TABLES ............................................................................... viii
LIST OF FIGURES ................................................................................. x
Chapter 1 Overview ................................................................................ 1
Chapter 2 Background ............................................................................. 3
2.1 Fundamental of HRTEM and EELS ...................................................... 3
2.1.1 HRTEM ...................................................... ' ....................... 4

2.1.1.1 Specimen Function ................................................... 4

2.1.1.2 Microscope Effect .................................................... 5

2.1.1.3 Fast Fourier Transform .............................................. 7

2.1.2 EELS .............................................................................. 8

2.1.2.1 Fundamentals ......................................................... 8

2.1.2.2 Energy-Loss Near-Edge Structures in EELSIO

2.2 Self-Assembled Silicide Nanowires .................................................... 12

2.2.] Introduction to Self-Assembled Silicide Nanowires ...................... 12

2.2.2 Crystallography of Epitaxial Silicide Nanowires .......................... 13

2.2.2.1 General Understanding of Bulk RE Silicides Phases ............ 13

2.2.2.2 Self-Assembled RE Silicide Nanostructures. .16
2.2.2.3 Crystal Structure of Bulk Phases in the Gd-Si and Tm-Si

Systems ................................................................ 20

2.2.3 Electronic Structure of Silicide Nanowires..................................24

2.2.3.1 EELS Study of Electronic Structure .............................. 25

2.3 Giant Magnetoresistance (GMR) Multilayers .......................................... 27
2.3.1 Introduction of the GMR Effect .............................................. 27

2.3.1.1 Origin of the GRM Effects ......................................... 27

2.3.1.2 CIP and CPP ......................................................... 29

2.3.1.3 Exchanged-Biased Spin-valves .................................... 30

2.3.1.4 Current-Induced Magnetization Switching ...................... 30

2.3.2 Correlation between GMR and Multilayer Structures ..................... 31

2.3.3 Motivation of This Study ...................................................... 32
Chapter 3 Experimental Approaches and Procedures ................................... .39
3.1 Bulk Gd Oxide, Metallic and Silicide Reference Sample ............................. 39
3.1.1 Bulk Gd203 ....................................................................... 39

3.1.2 Bulk Metallic and Silicide Samples Produced by Dc-magnetron

Sputtering ........................................................................ 40
3.2 RE Silicide Nanostructures Grown in UHV Chamber ................................. 40
3.3 Magnetic Multilayers Synthesis .......................................................... 42
3.4 TEM Characterization ..................................................................... 42

vi

3.4.1 Cross-sectional TEM Sample Preparation ................................... 43
3.4.2 High Resolution TEM45

3.5 EELS Characterization ..................................................................... 46
3.5.1 Design of Electron Energy-Loss Spectrometers ........................... 46

3.5.2 EELS Spectrum Collection ................................................... 47

3.5.3 Gd M-Edge Spectrum Curve Fitting ........................................ 48

3.5.4 Spectrum Image Collection .................................................... 48

3.5.5 Intensity Proﬁles and Simulation ............................................ 59
Chapter 4 Results from RE Silicide Nanostructure Studies ............................. 51
4.1 Gd Oxide, Bulk Metallic and Silicide Crystal Structure Study ..................... 51
4.2 HRTEM Study on Gd and Tm Silicide Nanostructures .............................. 56
4.2.1 Gd Silicide Nanostructures .................................................... 56

4.2.2 Tm Silicide Nanostructures ................................................... 64

4.2.2.1 Tm Silicide Nanostructure Morphologies ....................... 64

4.2.2.2 HRTEM Image Simulation ....................................... 66

4.2.2.3 Results from HRTEM Imaging ................................... 68

4.3 EELS Study on Gd Silicide Nanostructures ........................................... 76
4.4 Conclusions .................................................................................. 80

Chapter 5 Results from Multilayer Structure Studies82

5.1 Co(Py)/Al Multilayers .................................................................... 82
5.1.1 AS-sputtered and Annealed Co/Al Multilayers ............................. 82

5.1.2 As-sputtered and Annealed Py/Al Multilayers ............................. 95

5.2 Cu/Al Multilayers and Cu/Al in Spin-valve Structure .............................. 100
5.2.1 Cu/Al Multilayers ............................................................. 101

5.2.1.1 Cu(8nm)/Al(10nm) Multilayer Structures ...................... 101

5.2.1.2 Cu(3nm)/Al(5nm) Multilayer Structures ....................... 108

5.2.2 Cu/Al in Spin-Valve Structure ............................................... 110

5.2.2.1 As-Sputtered Spin-Valve with 10m Al ........................ 110

5.2.2.2 Annealed Spin-Valve with 30nm A1 ............................ 118

5.3 Discussion and Coclusions ............................................................... 125
Chapter 6 Overall Discussion and Summary ............................................... 128
Chapter 7 Conclusions and Suggestions for Future Works ............................. 136
7.1 Patterned Growth Studies of Silicide Nanostructures ................................ 136

7.2 High-resolution EELS Analysis from Monochromated TEM ....................... 137

7.3 EELS Quantiﬁcation Analysis in Magnetic Multilayers ............................. 138
REFERENCES ................................................................................... 139
APPENDIX A Detailed TEM Sample Description ....................................... 147

vii

Table 2.1

Table 2.2

Table 2.3

Table 2.4

Table 2.5

Table 2.6

Table 2.7

Table 2.8

Table 4.1

Table 4.2

Table 4.3

Table 5.1

Table 5.2

Table 5.3

Table 5.4

LIST OF TABLES

Important edges in EELS associated with their electronic states excitation
and degeneracy .................................................................... 10

A summary of the lattice constants of RE Silicides and their growth
behavior on Si(001) ............................................................... 13

Gd-Si crystal structure data [19] ................................................. 15
Gd disilicide phases and lattice mismatches with Si<110> after [19].. . ...22

Tm silicide crystallographic data and lattice mismatches with Si<110>

after [19] ............................................................................. 23
White lines energy differences and intensity ratios for the investigated rare
earth oxides 26] .................................................................... 26
Co-Al crystal structure data [19] .................................................. 35
Cu-Al crystal structure data [19] ................................................. 38

Lattice parameters of the bulk GdSi2 phases and those measured from the
three imaged nanostructures. (I, II, III refers to the imaged
nanostructures) ...................................................................... 58

Lattice parameters of the bulk Tm silicide phases and those measured from
the imaged nanostructures ........................................................ 71

Summary of N and M peak positions and intensity rati measurements of
the white lines from the metallic Gd, the bulk GdSiz, the GdSiz
nanostructure (NS), and Gd203 powder in comparison to Gd203 from
EELS Atlas [27] ................................................................... 80

Lattice parameters and close-packed interplaner spacing variants for the
multilayers ......................................................................... 88

Summary of the lattice spacings measured from all the phases in the as-
sputtered spin-valve ............................................................. 104

Summary of the lattice spacings measured from all the phases in the as-
sputtered Spin-valve ............................................................. 1 15

Summary of the lattice spacings measured from all the phases in ﬁgure
5.29 ................................................................................ 123

viii

Table 6.1 Table 6.1 Electrical resistivity data of the phases in the Cu/Al multilayers,
after [73] ........................................................................... 135

ix

LIST OF FIGURES

Chapter 2
Figure 2.1 (a) sin x(u) vs. u (without damping), and (b) T(u) vs. II modiﬁed by imposing
an envelope function (dashed line); Af = -100nm, C5 = 2.2 mm ................................ 6

Figure 2.2 Figure 2.2 (a) An HRTEM image of Si along [110] zone (b) corresponding
FFT pattern. The reciprocal spacings of the spots are related to the corresponding Si
atomic planes .......................................................................................... 7

Figure 2.3 A typical EEL spectrum ﬁom an NiO specimen including zero-loss peak,

plasmon peak, and compositional edges for NiO ................................................. 9
Figure 2.4 Gd-Si phase diagram, after [19] ...................................................... 14
Figure 2.5 Tm-Si phase diagram, after [19] ...................................................................... 16

Figure 2.6 Schematic representation of a hexagonal Silicides forming a nanowire on
Si(001) due to the small mismatch along the a axis and large mismatch along the c axis,
after

[20] ..................................................................................................... 17

Figure 2.7 AFM image of GdSiz nanowires and nanoislands grown on Si(OOI) substrate
along two perpendicular Si <110> directions, after [17] ....................................... 18

Figure 2.8 Plot of hexagonal Ale type of RE silicide anisotropic mismatches with
Si(001). With respect to Si(001) two perpendicular <110> directions, the lattice
mismatches of Silicides have a large absolute value between 4% and 10% and a small
lattice mismatch within i2%. (after [11]) ....................................................... 19

Figure 2.9 An AF M image of Tm silicide nanostructures on Si(001) with 0.98 ML Tm
coverage .............................................................................................. 20

Figure 2.10 Crystal models of RE Silicides, (a) hexagonal Ale structure (b)
orthorhombic GdSiz or tetragonal ThSiz structure. (after [20] ) .............................. 21

Figure 2.11 Projections along various axes of the hexagonal and orthorhombic GdSiz
phases ................................................................................................. 22

Figure 2.12 Projections along various axes of the hexagonal Tm3Si5 and orthorhombic
TmSi
phases ................................................................................................. 24

Figure 2.13 Magnetoresistance curves of Fe/Cr multilayers at 4.2 K [32] .................. 28

Figure 2.14 Schematic representations of the scattering electrons in a magnetic multilayer.
The signs + and - are for spins Sz=+l/2 and Sz=-1/2, respectively. F represents a

ferromagnetic layer and N represents a non-magnetic layer, after [33] ...................... 29
Figure 2.15 Schematic representative of a spin-valve structure ............................... 30
Figure 2.16 Co-Al phase diagram [19] ........................................................... 34
Figure 2.17 Cu-Al phase diagram [19] ........................................................... 37
Chapter 3

Figure 3.1 Schematic representation of RE Silicide nanostructure growth experimental
setup ................................................................................................... 41

Figure 3.2 Schematic representation of the process of preparing a cross-sectional TEM
sample ................................................................................................. 45

Figure 3.3 Schematic overview of the in-column Q ﬁlter (a) and post-column Gatan
Image Filter (b) after [2] ........................................................................... 47

Figure 3.4 Zero-loss peak with a full width at half maximum (FWHM) ~1.0 eV shows
that the EELS spectrometer is well aligned ...................................................... 48

Figure 3.5 A typical EELS spectrum of M45 edge (black line) ﬁom GdSiz ﬁtted as a
superposition of a power-law background (straight line), cross sections from Hartree-
Slater model (rising line), and a set of Lorentzian proﬁles (gray lines). The dotted curve
Shows a good curve ﬁtting result ................................................................. 49

Figure 3.6 Simulation of an F(Co, Ni, or Fe) L-edge intensity proﬁle with a 0.5nm

FWHM of Gaussian electron beam proﬁle across idealized F/Al interface ................ 50
Chapter 4

Figure 4.1 X—ray diffraction shows the bcc Gd203 agree with the prototype of Mn203
(Bixbyite) structure ................................................................................. 52

Figure 4.2 (a) Phase contrast TEM image of Gd203 powder particles on a holy carbon
ﬁlm. Insets (b) and (c) Show HRTEM images of a single grain and poly-grain atomic
lattice ﬁ'om two powders respectively .......................................................... 52

Figure 4.3 (a) Phase contrast TEM image and (b) corresponding selected area electron

diffraction pattern of amorphous Si and metallic Gd layers alternately sputtered on a
Si(001) substrate ................................................................................... 53

xi

Figure 4.4 (a) Phase contrast image of post-annealed bulk GdSiz layer formed on the
Si(001) substrate. (b) Selected area diffraction pattern from one grain of the layer with
arrows presenting strong diffraction spots from the orthorhombic phase of GdSiz. (c)
HRTEM image showing the [100] orthorhombic zone axis from the GdSiz. The atomic
lattice symmetry is indicted by the stacking of (001) planes in an aabb manner .......... 55

Figure 4.5 (a) 3D topography STM image of GdSiz nanostructures grown on Si(001) with
(b) a cross-sectional line proﬁle, indicated by the line on (a) ................................. 57

Figure 4.6(a) Cross sectional phase contrast TEM image of a GdSi2 nanostructure grown
on Si(001). (b) Higher magniﬁcation HRTEM image Showing the interface between the
GdSiz and Si. (c) FFT of area shown in (b) ..................................................... 58

Figure 4.7 (a) Cross sectional phase contrast TEM image of a GdSiz nanostructure with
multiple thicknesses. (b) Higher magniﬁcation HRTEM image shows an aabb lattice
stacking, consistent with the orth/tet phase of GdSi2 viewed along the [100] zone axis. (c)
FFT from area (b) conﬁrms the crystal structure of the orth/tet phase of GdSiz .......... 59

Figure 4.8 (a) An HRTEM image of a GdSiz nanostructure with a wavy top. Two
different lattice symmetries are observed in the nanostructure on the two sides of the
dashed horizontal line. Above this line an aabb stacking is displayed, and an FFT from
area 1 shows a tilted orthorhombic GdSi2[100] zone. Below the line, close to the Si
substrate a rectangular symmetry with aaaa stacking is observed. An FFT ﬁom area 2
shows a hexagonal GdSi2-x[100] zone. (b) An enlarged image from area 3 illustrates the
contrast from mismatch dislocations at the interface, as well as a periodic array of strain
contrast marked by arrows ........................................................................ 62

Figure 4.9 (a) An STM image of Tm silicide nanostructures on Si(001) with 2.8 ML Tm
coverage. (b) A cross sectional proﬁle along the line shown in (a) ........................ 65

Figure 4.10 (a) An STM image of Tm silicide nanoisland with a modulation in height. (b)
Cross-sectional proﬁle along the line shown in (a) ............................................ 65

Figure 4.11 A comparison of the projections and corresponding simulated lattice images
of the hexagonal Tm38i5 (a and b) and orthorhombic TmSi (c and d) imaged down
different axes ........................................................................................ 67

Figure 4.12 (a) HRTEM image Shows a Tm silicide nanostructure approximately 10 nm
in width and less than 1 nm in height. (b) Enlarged lattice image shows a rectangular
stacking symmetry. (c) FFT shows streaked diffraction peaks ............................... 68

Figure 4.13 (a) Cross sectional HRTEM image of a Tm Silicide nanostructure on Si(001).
(b) Enlarged atomic lattice shows rectangular symmetry. (c) Corresponding FFT from
area (b). ((1) X-ray energy dispersive spectrum shows chemical nature of the
nanostructure ......................................................................................... 70

xii

Figure 4.14(a) Cross-sectional HRTEM image of Tm silicide nanostructure from an
elongated Tm silicide nanostructure ~90 nm wide and up to 3nm thick between Si
substrate and Can layer. (b) Enlarged lattice image shows a stacking similar with the
simulated image of orthorhombic TmSi viewed along [100]. (0) Corresponding FFT
Shows a TmSi [100] zone. ((1) and (e) shows a comparison of FFT ﬁltered lattice image
from the nanostructure and Si, which conﬁrms the stacking in the nanostructure is
consistent with the simulated image of orthorhombic TmSi viewed along [100] (inset in
(d)) ................................................................................................... 71

Figure 4.15 (a) Cross-sectional HRTEM image of Tm silicide nanostructure on Si(001)
shows a stacking symmetry with a half-atom shift between every two rows of atoms

consistent with the orthorhombic TmSi [001] axis. (b) and (c) are the corresponding
FFTS from areas (1) and (2) ...................................................................... 73

Figure 4.16 Two types of lattice stacking are observed in each of the two nanostructures
(a) approximately 25 nm wide and approximately 2nm thick (b) approximately 10 nm
wide and approximately 2nm thick. Different stacking are marked by arrows and are
compared to the simulated images ............................................................... 74

Figure 4.17 The energy loss near-edge structure (ELNES) of the Gd N edge from metallic
Gd, bulk GdSiz, and Gd203 powder in comparison to the Gd203 spectrum from the
standard EELS Atlas [27] ........................................................................ 78

Figure 4.18 The energy loss near-edge structure (ELNES) of the Gd M45 edge from
metallic Gd, bulk GdSiz, and Gd203 powder in to the Gd203 spectrum ﬁom the standard
EELS Atlas [27]. The line with open squares in (b) demonstrate good curve ﬁtting using
a power-law backgound, Hartree-Slater cross section and a set of Lorentzian proﬁles..79

Figure 4.19 Gd electron conﬁgurations for the ground state neutral gaseous atom ........ 80

Chapter 5

Figure 5.1 Phase contrast TEM image of as-sputtered Co/Al multilayers on a Si substrate.
Inset is a selected area diffraction pattern from both the multilayers and Si substrates
showing strong diffraction from the close-packed growth planes of both the Co and Al,
parallel with Si(001). In the upper right, an inclined gain is illustrated schematically to
Show that the waviness perpendicular to the electron beam is up to 2 nm at the interfaces.
......................................................................................................... 83

Figure 5.2 HRTEM images showing Co/Al atomic lattices with different contrast. The
Co/Al interfaces are poorly deﬁned. A gain boundary in Al is observed in (a). Both
FCC and HCP stacking in the Co are Shown in the enlarged images in (b). Insets in the
upper right show the corresponding FFTS from the images ................................... 86

Figure 5 .3 A through-focus series of images from the same area of the multilayer
indicates that the Al and Co interfaces are inclined along the beam direction .............. 87

xiii

Figure 5. 4 Schematic growth model ofFCC A1 {111} on FCC Co W{l 1 l} with a lattice
mismatch of 13. 7%. ... .. ... ... ... ... . .. . .......88

Figure 5.5 Phase contrast TEM image of an annealed Co/Al multilayer. Inset is a
selected area diffraction pattern ﬁom both the multilayers and Si substrates showing
strong diffraction ﬁ'om the close-packed gowth planes of both the Co and Al ............ 89

Figure 5.6 (a) A dark ﬁeld STEM image of the as-sputtered Co/Al multilayers showing
where the EELS spectrum image was acquired along the white dashed line. The spatial
drift square is the cross-correlation scan area used to correct the drift. (b) A comparison
of line proﬁles of image gey level across a Co/Al/Co layers from both as-Sputtered and
annealed multilayers. (c) A perspective view of an EELS spectrum image showing the
plasmon intensity changing across the Co/Al multilayers ..................................... 91

Figure 5.7 A comparison of the plasmon intensity proﬁle for as-sputtered and annealed
Co/Al multilayers. The Al plasmon peak proﬁles have full width at half maximum of 5-6
nm, similarly for both as-received and annealed samples. The only difference is that the
low variation of Co intensity length is 15 nm for the as-received multilayers and 17 nm
for annealed one92

Figure 5.8 A perspective view of the Co white line Spectrum image scanned across
Co/Al/Co layers over a length of 20 nm of the as-sputtered multilayers ..................... 94

Figure 5.9 Co white line intensity proﬁles for both the as-sputtered and annealed
multilayers. A theoretical proﬁle simulated by using a 0.5nm FWHM Gaussian electron
beam proﬁle across perfect Co/Al interfaces is used to compare with the experimental
data .................................................................................................... 94

Figure 5.10 Phase contrast TEM image of as-sputtered and annealed Py/Al multilayers.
Insets are selected area diffraction patterns from the multilayers. Both morphologies and
diffraction patterns show similar structures in two conditions. 96

Figure 5.11 HRTEM image of an annealed Py/Al multilayer along a Py <110> zone with
corresponding FFT in the inset showing the gowth relationship .............................. 97

Figure 5.12 A comparison of plasmon intensity proﬁle from the as-sputtered and
annealed Py/Al multilayers. The Al plasmon peak proﬁles have full width at half
maximum of 6.5-7 nm, Similarly for both as-sputtered and annealed samples. . . . . .....98

Figure 5.13 Ni (a) and Fe (b) white line intensity proﬁles for as-sputtered and annealed
multilayers are compared separately with the theoretical proﬁle simulated by using a
0.5nm FWHM Gaussian electron beam proﬁle across perfect Py/Al interfaces. (c) A
comparison of line proﬁles of the image gey level from Py/Al/Py layers in the
Z-contrast images of the two conditions ......................................................... 99

xiv

Figure 5.14 Phase contrast TEM image of as-sputtered Cu(8nm)/Al(10nm) multilayers.
Observed layers are in different contrast, exhibiting ~12nm thick Cu-rich layer and ~6nm
thick Al-rich layer, which are different ﬁ'om the nominal thickness ....................... 101

Figure 5.15 HRTEM image shows three types of stacking ﬁinges with different contrast
existing in the Cu/Al multilayers. Five different areas in the image are analyzed using
FFTS, resulting in three crystal zone patterns shown in the insets. Areas 1 and 5 show the
same 0 AlzCu <001> zone patterns, areas 2 and 4 Show the same [3 AlCu3 <111> zone
patterns, and area 3 shows a 0 AlzCu <001> zone pattern .................................... 103

Figure 5.16 Schematic representation of the bee AlCu3 and tetragonal A120u crystal
structures and the orientation relationships of their close-packed planes. (a) bee AlCu3
crystal structure (b) tetragonal AlzCu crystal structure (c) a structural rectangle unit in
AICU3(110) ((1) two orientation relationships with A12Cu(l 10): AlCu3[111] // A12Chr[001]
(the ﬁne dotted lines) and AlCu3[111] // A12Cu[111] (the coarse dotted lines). .....106

Figure 5.17 Phase contrast TEM image of annealed Cu(8nm)/Al(10nm) multilayers. The
Cu-rich layers are ~13 nm thick and the Al-rich layers are ~7 nm thick ................... 107

Figure 5.18 HRTEM image with corresponding FFTS from the armealed
Cu(8nm)/Al(10nm) multilayers. The Al-rich phase and the Cu-rich phase are identiﬁed
to be A12Cu and AlCu3 respectively ............................................................. 107

Figure 5.19 Phase contrast TEM image of the as-sputtered Cu(5nm)/Al(3nm) multilayers.
The Cu-rich layers are 3-4nm thick and the Al-rich layers are 4-5 nm thick .............. 109

Figure 5.20 HRTEM image shows two types of lattice fringes existing in the
Or(5nm)/Al(3nm) multilayers. Four different areas in the image are analyzed using FFTS,
resulting two crystal zone patterns shown in the insets. Areas 1 and 3 show the same 0
A12Cu <111> zone patterns, while areas 2 and 4 show the same Cu <110> zone

patterns .............................................................................................. 109

Figure 5.21 Schematic representation of FCC Cu (111) and tetragonal AlzCu (110)
gowth orientation and lattice mismatches ..................................................... 110

Figure 5.22 TEM bright ﬁeld image of an as-Sputtered spin-valve with nominal thickness
in nanometers
Nb(1 50)Cu(5)FeMn(8)Py(24)Cu(l 0)Al(10)Cu(1 0)Py(24)Cu(5)Nb(50) .................. 1 1 2

Figure 5.23 Phase contrast TEM image of Py/Cu/Al/Cu/Py layers in the as-sputtered
spin-valve with x=10nm ........................................................................... 1 13

Figure 5.24 (a) HRTEM image of Py/FeMn/Cu/Nb layers in the as-sputtered spin-valve
with x=10nm and (b) the corresponding FFT from the image .............................. 114

XV

Figure 5.25 (a) HRTEM image of Cu/Al/Cu layers in the as-Sputtered spin-valve with
x=10nm and (b) corresponding F FT from the image ....................................... 116

Figure 5.26 (a) STEM bright ﬁeld image of the multilayers in the as-sputtered spin-valve.
(b) XEDS intensity proﬁles of Cu, Ni, Fe, and Al were collected ﬁom the line across the
Py/Cu/Al/Cu/Py/FeMn/Cu layers in the image (a) .......................................... 117

Figure 5.27 TEM bright ﬁeld image of an annealed spin-valve with nominal thickness in
nanometers Nb(150)Cu(5)FeMn(8)Py(24)Cu(10)Al(30)Cu(10)Py(24)Cu(5)Nb(50)...l 19

Figure 5.28 Phase contrast TEM image of Py/Cu/Al/Cu/Py layers in the annealed spin-
valve with x=30nm .............................................................................. 120

Figure 5.29 HRTEM image of the Al/Cu/Py layers in the annealed spin-valve with
x=30nm. Five different areas in the image are analyzed using FFTS. Areas 1 shows a 0
A120u <100> zone pattern, area 2 shows a B’ AlCu3 <111> zone pattern, area 3 shows a
B’ AlCu3 <001> zone pattern, area 4 shows a BCC Py <111> zone pattern, and area 5
Shows a FCC Py <111> zone pattern ......................................................... 122

Figure 5.30 (a) STEM bright ﬁeld image of the multilayers in an annealed spin-valve. (b)
XEDS intensity proﬁles of Cu, Ni, Fe, Al and O are collected from the line across the
Py/Cu/Al/Cu/Py layers in the image (a) ....................................................... 124

Images in this dissertation are presented in color.

xvi

Chapter 1 Overview

Great achievements have been obtained in nanotechnology in the last decade by
taking advantage of the unique properties of materials structured on the nanometer scale
(normally between 1-100 nanometers). Nevertheless, due to lack of fundamental
understanding of the thermodynamic and kinetic characteristics of nanoscale materials,
there is still a large barrier between rational desigr and controlled synthesis with desired
properties at the nanometer scale [1]. It is well known that bulk material properties are
not critically dependent on Size, but the properties of nanoscale materials are often very
size-dependent. Consequently, the structure, property, and stability of nanoscale
materials must be understood to enable their desigr and application.

The objective of this work is to study the gowth and crystallogaphy of a number
of nanoscale heterostructures by characterizing their crystal structures, interfacial epitaxy,
and interfacial chemical and electronic nature. This will provide basic understanding of
the thermodynamic phase stability and surface energies that affect nucleation, gowth,
and stability. The selected systems in this thesis are two classes of nanoscale electronic
material systems. The ﬁrst type of materials systems are epitaxial rare earth Silicides,
which can self-assemble to form nanowires on silicon(001) substrates under speciﬁc
gowth conditions. Different morphologies, nanowires and nano-sized islands, were
observed during the gowth of the Silicides. The morphologies are correlated with the
different crystal structures of the RE Silicides, which have different lattice mismatches
with the Si substrates. The second type of materials systems are sputtered magnetic
multilayers, including giant magnetoresistance (GMR) multilayers and spin-valves,

which are the basic components for many read heads in hard-disk drives. These

nanometer thick metallic multilayers show interesting magnetoresistance (MR) behaviour,
such as large current-perpendicular-to-plane (CPP) speciﬁc resistances, small interface
scattering asymmetries, and unstable MR in terms of time and temperature. These
interfacial properties and the changes in MR may be affected by individual layer
structures, interfacial roughness, and/or the thermodynamic stability of the structures.
High resolution transmission electron microscopy (HRTEM) has been carried out
to investigate the crystal and interfacial structures of these nanoscale electronic materials.
The phases formed and epitaxial relationships between the phases have been
characterized to obtain an understanding of the gowth mechanisms and stability of the
structures at the nanometer scale. Simultaneously, analytical electron microscopy,
including electron energy-loss spectrometry (EELS) and x-ray energy dispersion
spectrometry (XEDS), have been used to investigate the chemical and electronic nature
of these structures, which can provide useful information for understanding their physical

and chemical properties.

Chapter 2 Background

This chapter deals with the current critical backgound literature for this thesis. It
will start with an introduction of the fundamental theory of high-resolution transmission
electron microscopy (HRTEM) and electron energy-loss spectrometry (EELS). This is
followed by a literature review of the fundamentals of the epitaxial rare earth silicide

nanowires and giant magretoresistance multilayers.

2.1 Fundamentals of HRTEM and EELS

Transmission electron microscopy (TEM) has been demonstrated to be an
efﬁcient technique for investigating the crystal structures and chemical nature of
materials. TEM is becoming more important as the dimensions of many material
structures approach the nanometer scale, which often lie below the spatial resolution of
other experimental methods. Hi gh-resolution transmission electron microscopy
(HRTEM) is a particularly powerful technique for the investigation of local crystal
structures on the atomic scale and is very useful for understanding the details of
nanoscale materials. Complemented with analytical capabilities, including x-ray energy
dispersive spectrometry (XEDS) and electron energy-loss spectrometry (EELS),
HRTEM’S spatial resolution in the nanometer and sub-nanometer range, makes it an
indispensable tool for nanoscale analytical application.

In this thesis, HRTEM and EELS are the primary techniques being applied to the
study of two classes of nanometer scale electronic materials: self-assembled rare earth
silicide nanostructure epitaxial gown on Si(001) substrates and giant magretoresistance

(GMR) multilayers and spin-valves.

2.1.1 HRTEM

TEM image contrast arises from the scattering of the incident electron beam by
the specimen. Both the amplitude and phase of the electron wave can be changed, which
gives rise to image contrast. Conventional TEM, including bright ﬁeld (BF) and dark
ﬁeld (DF) imaging, takes advantage of diffraction contrast (one type of amplitude
contrast). However, high-resolution TEM is essentially phase contrast imaging. While
BF or DF imaging selects a single beam to form an image, a phase contrast image
contains more than one beam. The contrast observed can be described in a quantitative
way that breaks the process into a number of steps. These include a specimen function

and contributions ﬁ'om the microscope.

2.1.1.1 Specimen Function

The specimen interacts with the incident electron beam because of the electronic
static potential of the specimen. Therefore, there is a phase shift of an electron when
passing through a specimen, which is

0 = other (2.1)
where o is the interaction constant, (Do is the mean electrostatic potential, and r is the
thickness [2]. The interaction of the incident electrons with the specimen is described
with the specimen transmission function f(r),

11w) = eXPl-i6¢o(X.y)-u(x.y)]. (2.2)
where (Do(x,y) is the projected potential of the specimen taking account of variations in
the z-direction and u is an absorption term [2]. If the specimen is thin, the absorption part

can be igrored. The resulting wave exiting the specimen is described as

WAX,” = f(X,Y) \P0(X’Y)' (2'3)
which means the amplitude of a transmitted wave function (‘11,) will be linearly related to
the projected potential of the specimen for a very thin specimen. This is also called

weak-phase-obj ect approximation.

2.1.1.2 Microscope Effect

After the electrons leave the specimen, the ability of a lens to pass the electrons is
described with the transfer ﬁmction T(u):

T(u) = 2A(u) sin x(u) (2.4)
where u is a reciprocal space lattice vector, A(u) is the aperture function, and x(u) is the
phase—distortion function, which is expressed as a function of the spherical aberration (Cs),
defocus (AI), and electron wavelength (A):

x(u) = niAfu2+1/2nc,ru“ (2.5)

Plots of x(u) vs. II, and T(u) vs. u are shown in ﬁgure 2.1a and 2.1b [2]. The best
resolution that allows direct interpretation of bright/dark contrast as atom position
corresponds to the ﬁrst zero point of T(u), or sin x(u), which is called the Scherzer
resolution, as indicated in ﬁgure 2.1. The Scherzer resolution is achieved when the
transfer function is optimized by balancing the effect of spherical aberration at a

particular negative value of Af, which is known as Scherzer defocus:

%...... = -1-2 0.1 (2.6)

At this defocus condition, the electrons at different reciprocal lattice positions

are transmitted in a nearly constant manner (similar phase and amplitude) up to the

Scherzer resolution, which is 0.66Cl/4/I3/4 [2]. The TEM used for HRTEM in this

study is a JEOL JEM2200F S working at 200KV, which has a Cs of 0.5 mm, and hence, a
theoretical point-to-point Scherzer resolution of 0.19 nm.

Note however, that information at larger reciprocal space frequencies (ﬁner real
space resolutions) can be passed. Imaging at such higher frequencies allows imaging
structures below the Scherzer limit. Nevertheless, image interpretation becomes more
difﬁcult with such “pass ban ” imaging, because not all information is passed in the same
manner. The ultimate resolution is limited by the spatial coherence spread of the electron
source and chromatic aberration. Hence, T(u) is damped by imposing an envelope
function on it, as Shown in ﬁgure 2.1. It is evident that T(u) is not able to transfer any
information with u larger than a speciﬁc value. This resolution limit is known as the

information limit, as indicated in the ﬁgure.

(3)

511196; 098 oar/I III >u
..-; _____ =\llll
\V II

(b) 2 Scherzer resolution

 

 

 

 

 

T(u) \\information limit

envelope function

 

Figure 2.1 (a) sin x(u) vs. II (without damping) at the Scherzer defocus, and (b) T(u)
vs. II modiﬁed by imposing an envelope function (dashed line); Af = -100nm, C5 = 2.2
mm [2].

2.1.1.3 Fast Fourier Transform

Fast Fourier transform (FFT) analysis is an effective way to interpret the atomic
structure in the HRTEM images. F FT diffractogams are obtained by capturing an area
of interest in the images. The atomic lattice spacings and symmetries (including
interpretable allowed and/or forbidden reﬂections) can been determined from F FT, which
in turn provides useful information for analyzing the crystal structures. Such FFTS can
be used in a manner analogous to selected area diffraction (SAD) patterns, without the
spatial limitation of SAD patterns [2]. Figure 2.2 shows an HRTEM image of single
crystal Si viewed along Si[110] and its corresponding FFT diffractograrn. Calibrated for
the appropriate image magniﬁcation, these Si FFT spot spacings represent the reciprocal
lattice spacings of the Si crystal planes of the imaged zone (i.e. Si {1 1 1} and Si{002} for
the <110> zone axis shown here). Note that additional effects, such as spot spreading can

result from edge effects.

Z.‘ 'iij.‘.l.';l 'suTIn

‘an” ’ "'
_

~ R
)1 Si(lTl)
' ‘~'Sil00l‘l‘ -Si(l)02)

 

Figure 2.2 (a) An HRTEM image of Si along [110] zone (b) corresponding FFT pattern.
The reciprocal spacings of the spots are related to the corresponding Si atomic planes.

2.1.2 EELS
2.1.2.1 Fundamentals

Electron energy-loss spectrometry (EELS) is based on the general idea that as
incident electrons interact with a Specimen, some electrons will lose energy (inelastic
scattering). Speciﬁcally, the primary electron loses some energy by transferring it to the
material, thereby exciting electrons in the specimen to higher energy states.
Measurement of the energy distribution of these inelastically scattered electrons leads to
the so-called electron energy-loss spectrum (EEL spectrum). These Spectra have been
used to analyze the electronic structure of specimen atoms, which in turn reveals details
of the nature of these atoms, their bonding and nearest-neighbor distributions, and their
dielectric response [3].

Figure 2.3 shows a typical EELS spectrum from an NiO Specimen. The most
intense feature seen in the spectrum is the zero-loss peak. This peak is generated from
elastically scattered electrons. It has a certain width (1-2 eV) caused by the non-
monochromaticity of the primary electrons and some instrumental broadening. The
plasmon peak, which occurs because of oscillation of weakly bound electrons, is the
second most dominant feature of the EEL spectrum after the zero-loss peak. The other
features are composition related edges due to the electron excitation in the specimen to
speciﬁc states, which can be used to study the chemical as well as the electronic nature of

the materials.

 

«Zero Loss

 
  

O K edge
A/

gain change

Plasmon Loss

Intensities (arb. units)

 
    
 

A/ NI M edge

 

 

 

 

I ' I I n 1 n
0 200 400 600 800 1000
Energy Loss (eV)

Figure 2.3 A typical BEL Spectrum from an NiO specimen including zero-loss peak,
plasmon peak, and compositional edges for NiO.

EELS edges are indicated by the letters K, L, M, N followed by one or more
numbers. The letter indicates the atomic shell from which the incident electrons excited
bound electrons, while the number indicates the orbital. Table 2.1 gives an overview of
the most important edges. For example, the L1 indicates an excitation of a 25 state, L2,;
indicates an excitation of 3 2p state, which is spin split into 2pm (L2) and 2pm (L3) states.
The spin split states are labeled by their total angular momenttmr J and are (21+1)-fold
degenerate, e.g. a 2pm state can have j=—3/2, -1/2, 1/2, 3/2 and is therefore 4—fold
degenerate [2]. The degeneracy of a state determines the number of electrons that can be
in that state and is important because the more electrons there are in a state, the more
likely an excitation from that state will be. As a result, one would expect that the

intensity ratio between an L3 and an L2 edge to be 2:1 [4].

Table 2.1 Important edges in EELS associated with their electronic states excitation
and degeneracy.

 

Edge State Degeneracy
K 1 s1 ,2 2
L1 251,2 2
L2 2131/2 2
L3 2133/2 4
M1 331,2 2
M2 3pm 2
M3 3133/2 4
M4 3d3,2 4
M5 3d5,2 6

 

2.1.2.2 Energy-Loss Near-Edge Structures in EELS

The energy resolution of EELS (ranging ﬁ'om approximately 0.3 to 3 eV,
depending on instrument) is much better than X-ray spectrometry (approximately 135 eV
for XEDS). Because of this, EELS can be used to determine a wealth of information
about the Specimen in addition to its basic elemental chemistry. Much of this information
is contained in ﬁne-detail intensity variations in the ionization edges, termed energy-loss
near-edge structure (ELNES). From these ﬁne structures, information such as the
bonding state of the ionized atom, the coordination of the atom, and the density of states,
can be obtained. For example, due to different atomic bonding between gaphite (Sp2
bonds in the basal plane with van der Waals bonding between the planes) and diamond
(four directional hybridized Sp3 covalent bonds) [5], one can easily distinguish gaphite

from diamond from the carbon K edges in the EELS Spectra.

10

EELS spectra Show intense peaks for L2; edges of transition metals and M4;
edges of RE metals. These spectral features are called “white lines” because of their
appearance on the photogaphic plates in early experiments with x-ray absorption
spectroscopy (XAS) [2]. The positions and relative intensities of these edges can be
inﬂuenced by local chemistry and the bonding between atoms due to the solid state effect
(i.e. the inelastic scattering of electrons is affected by the atomic crystal periodicity rather
than a single atom [6]). Leapman et a1. [4] have studied these ﬁne structures of 3d
transition metals and their oxides using EELS. They found that there is a difference in
the L3 threshold energy (chemical shift) between the metal and oxide. Furthermore, the
statistical ratio of intensities of L3 and L2 has been found to deviate ﬁom the statistical
value (2.0) based on the occupancy of the 3d states, i.e., with values ranging between 0.8
for Ti and 5.0 for Fe. In the case of metallic copper, the ﬁrll 3d states lead to the absence
of threshold peaks, while in compounds such as CuO electrons are drawn away from the
copper atom, leading to d-level being empty and sharp L3 and IQ threshold peaks. Kurate
et a1. studied EELS of a series of Mn oxides and demonstrated that the intensity ratio of

the white lines is related to the charge state or valency of the elements [7].

ll

2.2 Self-Assembled Silicide Nanowires

The epitaxial gowth of rare-earth (RE) metal Silicides on Si(001) has attracted
interest Since nanowires have been found to self-assemble on Si substrates due to an
anisotropic lattice mismatch [8-14]. These nanowires display metallic conduction [10, 14]
and consequently are promising candidates for ﬁrture nanoscale interconnects and devices.
In addition to this technological interest, detailed study of the crystal structures, defects,
and stoichiometry at the interface of the nanostructures and Si surface should give a basic
understanding of epitaxial thin ﬁlms in lattice mismatched systems.

The motivation of this study is to investigate the gowth mechanisms and
electronic structure of self-assembled RE silicide nanowires using HRTEM and EELS.
In the following sections, backgound and basic information on RE Silicides are

introduced, along with more detailed material on the electronic nature of these nanowires.

2.2.1 Introduction to Self-Assembled Silicide Nanowires

In 1998, Preinesberger et al. [8] ﬁrst reported the gowth of RE disilicide
nanowires formed by evaporating a small amount Dy on to a Si(001) surface followed by
subsequent annealing at high temperature. Later Chen et a1. [9] explained that the self-
assembly of Er disilicide nanowires on Si(001) is due to the anisotropic mismatch

between the hexagonal Ale type structure of the Silicides and the Si(001) surface. The

nanowire is free to gow along ErSiz [1120 ], which has a small mismatch (-1.3%) with
the hexagonal a axis, and is constrained from gowing along ErSiz [0001] where there is a
large mismatch (+65%) with the hexagonal c axis. Besides Dy and Er, other RE metal

Silicides including Sm [l3], Gd [11, 14], Ho[10,12], Yb [15], Se [1 1], Y [16] and Tm

l2

[l7] Silicides are reported to form nanowires on Si(001). A summary of the lattice
constants of RE Silicides and their gowth behavior on Si(001) is shown in Table 2.2.
Although Y and Se are technically not rare-earth metals, their Silicides have similar
properties as the lanthanide metal Silicides, so typically Y and Sc Silicides are also

included as RE Silicides.

Table 2.2 A summary of the lattice constants of RE Silicides and their gowth behavior
on Si(001).

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Lattice parameter (mismatch
Silicides %) Nanowires

a (M!) (%) 6 (nm) (%)
ScSi1'7[1 1] 0.366 (-4.7) 0.387 (+0.8) Yes
YSi1_7[l6] 0.384 (0) 0.414 (+7.9) Yes
NdSil.7 [18] 0.412 (+7.3) 0.444 (+16) No
Sm3Si5 [13] 0.390 (+1.6) 0.421 (+9.6) Yes
GdSi 1'7 [11] 0.389 (+1.0) 0.417 (+8.7) Yes
DySi 1.7 [8] 0.383 (-0.2) 0.412 (+7.3) Yes
HoSi 1.7 [10] 0.382 (-0.6) 0.411 (+7.0) Yes
ErSi 1.7 [9] 0.380 (-l.3) 0.409 (+6.5) Yes
Yb3Si5 [15] 0.378 (-1.5) 0.410 (+6.7) Yes
Tm3Si5 [17] 0.377 (-1.9) 0.407 (+6.0) Yes

 

 

2.2.2 Crystallography of Epitaxial Silicide N anowires
2.2.2.1 General Understanding of Bulk RE Silicide Phases

Before going into the details of RE Silicide nanostructures, it is very important to
gain a general understanding of the known bulk phases of the RE Silicides. In many

cases, the phase equilibrium of bulk RE Silicides are poorly deﬁned, and there is

13

considerable uncertainty regarding the solubility and temperatures where the RE Silicides
exist.
The Gd-Si phase diagarn (ﬁgure 2.4) and crystal structure data of the system

(Table 2.2) [19] show that 7 phases can form depending on the temperatures and
stoichiometries. Since in the gowth of nanostructures, only a small amount of the RE
metals are evaporated onto Si substrates, and the substrate temperature is typically around
600°C (see Chapter 3.2), the Si-rich Silicides are expected to form. On the Si—rich side of
the Gd-Si phase diagam, GdSi2 forms tetragonal or orthorhombic phases over a narrow
stoichiometry range depending on temperature, while at a slightly different stoichiometry,
GdSi2.x forms a poorly deﬁned [3 phase at high temperature and a hexagonal phase below
1000°C. Consequently, any of these phases with similar stoichiometry might form the

RE silicide nanostructures on Si substrates.

 

 

 

 

 

 

 

 

 

2100°C
2100' L 9 ~ -
it ~ 2' ‘ ~
9 'I' ' "‘7' \\
1800- o 1. .t - - - ‘ _
1650120_C ”3;: a ‘.
_ 1313°C ‘. ."u- ..«S ‘ _
1500 ’ I .‘ : - ‘3' . g \\ 1414OC‘
o KTIBGd) 1’ ' ‘5 ~1200°C .
°~ 1200'\ 1’ 1070°C .. fl. """" ~ 735 """
'- ' ?
g \ o 15.3 ?
a 900- 1235 C 1060120°C T (._BGdSi2
N
E 600‘ m .3 _ E o -
«.0... mm 3.. g .. ”aegis _____
0 60 ’64668 (Si)->
0 2'0 4'0 60 8'0 1(0
Gd Si

Atomic Percent Silicon

Figure 2.4 Gd-Si phase diagam, after [19].

14

Table 2.3 Gd-Si crystal structure data [19].

 

 

 

 

 

 

 

 

 

 

 

 

 

Pooooo
Gd5Si3 37.5 P63/mcm MnSSi3 hexagonal
Gd5Si4 44.5 P41212 Ge4Sm, tetragonal
GdSi 50 ana FeB orthorhombic
terSi2_x 60 to 62.5 P6/mmm AlB2 hexagonal
[3GdSi2_x 60 to 62.5
0tGdSi2 64 to 66.8 Imma GdSi2 orthorhombic
[3GdSi2 64 to 66.8 I4lamd ThSi2 tetragonal

 

 

Many RE-Si systems (such as Dy-Si, Y-Si) have similar phase diagams as that of
Gd, with hexagonal, orthorhombic, and tetragonal phases being reported around the
stoichiometry of RESi2 [19].

Unlike Gd and most RE Silicides, which have three possible disilicide phases with
very close stoichiometry, the Tm-Si phase diagam (ﬁgure 2.5) predicts three phases with
very different stoichiometries [19]. Among the three phases Tm5Si3, TmSi, and Tm38i5,

the latter two are phases that may form in a Si rich enviromnent.

15

 

 

 

 

 

 

0 L
1600 ‘é/l545 C 9 9 9 l4l4°C
Do 1200- ? """" .7" """ -
2
E
E" 800- ..
D.
E
h M In
400- :7) 3;) ii '
é” E E
H I" I-
0 . . . .
0 20 40 60 80 100
Tm Si

Atomic Percent Silicon

Figure 2.5 Tm-Si phase diagam, after [19].

2.2.2.2 Self-Assembled RE Silicide N anostructures
The formation of nanowires is attributed to the anisotropic mismatch between the

hexagonal Ale type RE Silicides and the Si substrate, as illustrated in ﬁgure 2.6. The

silicide nanowires are proposed to gow with the silicide [0001] and [11 20] directions
parallel with the (001) surface, lying along the two perpendicular Si <110> directions of
the substrate. Using Dy silicide as a prototype for the RE Silicides that form nanowires
on Si(001), hexagonal DySiz, with lattice constants a=0.383nm and c=0.412nm,
corresponds to mismatches with the Si <110> of -0.2% and 7.3% respectively. As a
result, nanowires gow to arbitrary length in the direction of small mismatch, but are

limited in their ability to gow coherently with the substrate in the lateral direction [9].

16

a C NW growth

 

Figure 2.6 Schematic representation of a hexagonal Silicides forming a nanowire on
Si(001) due to the small mismatch along the 3 axis and large mismatch along the c axis,
after [20].

Besides nanowires, nanostructures with other morphologies, such as nano-size
islands, can be observed during silicide gowth (as seen in ﬁgure 2.7). It is intuitive to
imagine that either tetragonal or orthorhombic Silicides contribute to this type of
morphology. However, these may form with different stoichiometries. Hence,
identifying the silicide structures and relating them to the nanostructure morphology is
important in understanding the gowth mechanisms of the epitaxial Silicides. IRTEM,
combined with fast Fourier transform (FFT) analysis, is an effective way to identify the
structures of these nano-size features. Using both cross-sectional and plan view HRTEM,
Ye et al. demonstrated that both hexagonal and orthorhombic (or tetragonal) structures

exist when Dy Silicides gow epitaxially on Si(001) [21].

500 nm Si [100]

 
     

nanoislands %.‘I'
\.

4.

Si [010]

\. .
v/ I] a n 0“" l I‘CS

r/

a»

0 500 nm

Figure 2.7 An atomic force microscope (AFM) image of GdSi; nanowires and
nanoislands gown on Si(001) substrate along two perpendicular Si <110>
directions, after [17].

It is generally accepted that epitaxial gowth occurs when the lattice constant of
the epilayer closely matches that of the substrate. The lateral gowth will be limited
only if there is a sigiiﬁcant lattice-mismatch (>2%) between the epilayer and the
substrate [11]. For those Silicides that can form nanowires on Si(001), the magnitudes of
lattice-mismatch in the smaller mismatch direction are all within i2%. Figure 2.8 shows
the magnitudes of the anisotropic mismatches in two directions for hexagonal Ale type
of RE silicide nanowires with respect to Si<110>. Different magnitudes of lattice
mismatches for these Silicides may result in different ways to relax the strain (such as
dislocation generation and/or epilayer tilt) between the epilayer and the substrate.
Investigating the interfacial crystallogaphy by HRTEM will reveal important

information for understanding the gowth mechanisms of the RE silicide nanowires.

 

10 _ : . .
_ Sm3Si5
Q 8 _ Yb3Si5 . o +vsrz GdSiz
9.. 6 _. . O . DySi' ‘
d: . ErSiz "”512 I
3 PTm3SIs I
I
a 4 P |
a I
.2 2 .. I
2 . :
‘9 0 . ...................... ' ......................
'3 - :'
a - ..
A 2 _ E
0
PE” "‘- i -
..1 -6 - : ScSiz
i. I
-3 _ I
I
_10 L l . l . l I l 1 l . l

 

 

 

-2.0 -l.5 -l.0 -0.5 0.0 0.5 1.0 1.5 2.0
Small Lattice Mismatch (%)

Figure 2.8 Plot of hexagonal AlB2 type of RE silicide anisotropic mismatches with
Si(001). With respect to Si(001) two perpendicular <110> directions, the lattice
mismatches of Silicides have a large absolute value between 4% and 10% and a
small lattice mismatch within :l:2%. (after [1 1])

Among the RE Silicides, it is especially interesting to note that hexagonal Tm
silicide has a large lattice mismatch (-l .9%) in the small mismatch direction, very close
to the generally accepted limit (2%) for free epitaxial gowth [l 1]. Recent work has
shown that Tm Silicides form epitaxial nanowires on Si(001) [17]. Nevertheless, there
are also islands present. An AFM image Shown in ﬁgure 2.9 shows the typical
morphologies of Tm silicide nanostructures: nanowires with high aspect ratio and islands,
typically much lower aspect ratios, are often observed at nanowire intersections. In

addition, there is a very low density of large bright features may be clusters of silicon

l9

carbide. These features are both much wider and much higher than the silicide related
features.

The different silicide morphologies might be correlated with different crystal
structures such as the Ale hexagonal Tm3Si5 (TmSiL7) and BCr orthorhombic TmSi. In
this thesis work, cross-sectional HRTEM was carried out to identify the crystal structures
of these Tm silicide nanostructures. It is also interesting to investigate the relaxation of
the strain between the epilayer and the substrate, since compared to other RE Silicides, a
larger magritude of lattice mismatch must exist in the Tm silicide nanowire gowth
direction. Furthermore, due to the potential large differences in stoichiometry of the
Silicides associated with these two morphologies, it may be that transformation between
morphologies might be more difﬁcult than in the other RE silicide systems, which may

result in the potential to form more stable nanowires.

unknown clusters

II

II

L
l/

Vi- ' ,

nanowires 500 nm
—

 

Figure 2.9 An AFM image of Tm silicide nanostructures on Si(001) with 0.98 monolayer
(ML) Tm coverage.

20

2.2.2.3 Crystal Structure of Bulk Phases in the Gd-Si and Tm—Si Systems

Three potential Gd Silicide phases, including the hexagonal, tetragonal and
orthorhombic phases, may form the silicide nanostructures on Si(001) substrates. Figure
2.10 shows crystal models of the prototype hexagonal Ale structure of GdSiz.x (a) and
the prototype tetragonal ThSiz structure of GdSi; (b). The prototype orthorhombic GdSi2
structure is a slight distortion of the tetragonal ThSiz structure, with a slight variation of
lattice parameters a and b. However, since it is difficult for TEM to identify the slight
lattice differences between the orthorhombic and tetragonal phases, orth/tet will be used
to describe these silicide phases in this thesis. The lattice parameters of these phases and
their lattice mismatches to Si<110> are listed in table 2.4. The hexagonal phase has an
anisotropic mismatch with Si, which may result in the NW formation. Both of
orthorhombic and tetragonal phases have large lattice mismatches in two perpendicular

directions, which suggests they will not form nanowires but instead form nanoislands.

 

c—axis
[001]
. RE
0 Si
c-axrs a-gxis ﬁg b-axis
[00011 [211°] [100] [010]

(a) hexagonal structure (b) orthorhombic or tetragonal

structure

Figure 2.10 Crystal models of RE Silicides, (a) hexagonal Ale structure (b)
orthorhombic GdSiz or tetragonal ThSiz structure. (after [20] )

21

Table 2.4 Gd disilicide phases and lattice mismatches with Si<110>, after [19].

 

 

 

Struct a b c Mismatch(%)
ure

(Hm) (nm) (nm) a b c
GdSi,_x hexagonal 0.388 0.417 1 8.6

 

orthorhombic 0.409 0.401 1.344 6.51 4.43

GdSi
2 tetragonal 0.410 0.410 1.361 6.77 6.77

 

 

 

 

 

 

 

 

 

 

 

To study the crystal structures using cross—sectional HRTEM, it is useful to
examine the different phases projected down various axes. The crystal structures of the
hexagonal and orthorhombic phases, together with the projections along different axes,
are shown in ﬁgure 2.11. The projection along the hexagonal a axis shows rectangular
symmetry with aaaa stacking of the strong scattering RE cations, quite different from the
symmetry seen along the orthorhombic a or b axes, which have aabb stacking of the RE
atoms. The projection along the hexagonal c axis shows 6-fold symmetry. The different
symmetries in the projections along different axes can help to distinguish the crystal

structures.

(a) “0

Hexagonal GdSi“ ‘ . o 0 g ‘ . O . O Orthorhombic GdSi2
. . O o O O O . .
G G 0 e e e
i : z e . 0 ¢ 0
G O 0 I o 0
Along hexagonal ‘ . 0 . o
\ an axis (or [100]) . O K
O O O . O . O
c O 0 Along orthorhombic
.°.'.°g aorbaxis(or[100])
O O O
0 0 00000 A!
O 0 O
‘ G_d Along hexagonal :::::
. Sr c axis (or [001]) .
Along orthorhombic
c axis (or [001])

 

Figure 2.11 Projections along various axes of the hexagonal and orthorhombic
GdSi; phases.

22

Three Tm silicide phases, including hexagonal TmSSi3, orthorhombic TmSi, and
hexagonal Tm3Si5, and their lattice mismatches to Si<110> are listed in table 2.5 [19].
Both the AlB2 hexagonal Tm3Si5 (TmSim) and BCr orthorhombic TmSi have reasonable
anisotropic mismatches with Si<110> and are candidates to form the nanowires on
Si(001). Figure 2.12 shows crystal models of the prototype hexagonal Ale structure (a)
of Tm38i5, the prototype tetragonal CrB structure of TmSi (b) and the projections of the
two phases down various axes. The projection along the hexagonal Tm3Si5[001] shows
6-fold symmetry, which is very similar with the projection along the orthorhombic
TmSi[lOO]. The projection along the hexagonal Tm3Si5[100] Shows rectangular
symmetry with aaaa stacking of the cations, quite different with the symmetry projected
along orthorhombic TmSi[001], which has aabb stacking. The difference of the latter

two projections is useful for distinguishing between the two phases.

Table 2.5 Tm silicide crystallogaphic data and lattice mismatches with Si<110>
after [19]

 

 

 

 

Composition Space Crystal Lattice parameters
Phase at.% Si goup Prototype structure (mismatch %)
= +
TmSSi3 37.5 P63/mcm MnSSi3 hexagonal ac=00.862158((+1610l98))
a= 0.418 (+8.9)
TmSi 50 Cmcm CrB orthorhombic b=1-035(+170)
c= 0.378 (-l.6)
Tm3Si5 62.5 P6/mmm Ale hexagonal :: 84377 ((35%))

 

 

 

 

 

 

 

 

23

(a) (b)

 

828::
f. f. f 0:0:0
Hera onaleS'
g "5 0.. 0.0.0 000.0
0 O O R
[Along TmJSisc ‘0‘“:
axis (or [001]) Along TmSi c
axis (or [001])
”0 0° 0 °.° K
0 ° 0'. o
O 0 O O O
c ' 3 e e e
O 0
Z ° 9 o o e
onng,Srsa 0 0
axis (or [100]) ....0
O‘.°.
AlonngSia
axis(or[100])

Figure 2.12 Projections along various axes of the hexagonal Tm3Si5 and
orthorhombic TmSi phases.

2.2.3 Electronic Structure of Silicide Nanowires

Epitaxial thin ﬁhns of these RE Silicides gown on Si are reported to have
metallic conductivity [10, 14] and form contacts to n-type silicon with low Schottky
barrier height [22]. Consequently, these Silicides have been proposed for applications in
microelectronic devices and interconnections [23]. For the one-dimensional RE silicide
nanowires being considered for electronic applications, it is critical to understand if their
electrical and electronic behavior varies between the bulk (or thin ﬁlm form) and

nanophase. Research on the electronic properties of Dy silicide nanowires [24] and Gd

24

silicide nanowires [23] has been performed using angular resolved photoelectron
spectroscopy measuring a uniformly oriented array of RE silicide nanowires on vicinal
Si(001). The results Show that both Dy and Gd silicide nanowires have one-dimensional
metallic character. However, the electronic structure of individual RE silicide

nanostructures has not been well characterized.

2.2.3.1 EELS Study of RE Electronic Structure

The M4; edge of rare earth metals in the electron-energy—loss spectrum are
dominated by the transition of 3d electrons to unoccupied 4f states [25]. To study the
near edge ﬁne structures, including edge positions and intensity ratios, it is important to
understand the electronic structures of the RE metals. A systematic experimental study
of M45 RE oxide edges done by Manoubi et a1. [26] showed that the separation of M5 and
M4 peaks is prominent and varies between l6eV for lanthanum and 45eV for ytterbium
(Table 2.6), reﬂecting the Spin-orbit splitting of the initial states in the transition. The
white line intensity ratios of RE oxides vary depending on their unoccupied 4f states. It
is interesting to investigate if the ELNES of the RE metals varies with their oxides. To

date, there have been no published results in the literature or standard EELS Atlas [27].

25

Table 2.6 White lines energy differences and intensity ratios for the investigated
rare earth oxides [26].

 

Rare earth A (Ms' 4) (eV) I (M5)3I(NI4)

 

La 16.7i0.3 0.76
Ce 18.5i0.2 0.85
Pr 19.6i0.1 1.20
Nd 20.9i0.1 1.45
Sm 25.6i0.1 1.84
Gd 31.2i0.7 1.73
Tb 33.1 i0.4 2.30
Dy 36.7i0.5 4.05
Ho 38.4i0.9 4.92
Er 41.6;I:O.9 5.80
Yb .. oo
Lu -

 

It is known that thin ﬁlm of RE Silicides on Si(lOO) or Si(l 11) display classic
metallic behavior [28,29]. As a result, it is interesting to compare the ELNES of the RE
silicide thin ﬁlm with that of the RE metals and their oxides. Furthermore, by comparing
the ELNES of M45 edges of the RE silicide nanostructures to those from the bulk
materials, the electronic nature, local chemistry, and coordination information of

individual nanostructure can be well understood.

26

2.3 Giant Magnetoresistance (GMR) Multilayers

This section will introduce some general backgound on giant magietoresistance
(GMR) multilayers. It starts with basic concepts related to the GMR effect, and then
includes a literature review on the correlation between the GMR effects and multilayer

structures. This is followed by the motivation of this study.

2.3.1 Introduction of the GMR Effect

GMR is a quantum mechanical effect observed in multilayered structures
composed of alternating ferromagietic (F) and nonmagnetic (N) metal layers. These
multilayers can display a large relative change in the electrical resistance upon the
application of an external magretic ﬁeld. This effect has been widely used for
commercial applications such as magnetic random access memory (MRAM), GMR read

heads, and magretic position sensors [30].

2.3.1.1 Origin of the GMR Effect

GMR effect was independently discovered in 1988 in Fe/Cr/F e trilayers by a
research team led by P. Griinberg [31] and in F e/Cr multilayers by the goup of A. F ert
[32]. Figure 2.13 shows the GMR effect found in the Fe/Cr multilayers [32]. In these
multilayers, for certain thicknesses of the Cr interlayer, the magnetizations of adjacent Fe
layers are antiparallely oriented (i.e. antiferromagnetically coupled). Upon application of
an external magnetic ﬁeld, the resistances of the multilayers decrease drastically since the
magnetizations of the two layers align in the direction of the applied ﬁeld. The percent

change in the resistance between the two states is deﬁned as the GMR ratio.

27

GMR = (RAP — Rp)/Rp, (2.7)

where Rpwo) is the resistance in the parallel (antiparallel) state.

R/R(H=0)

 
 
     

(Fe3run/Crl .8nm)3o

(Fe3rIm/Crl .2nm)35

(Fe3nm/Cr0.9nm)4o

 

0.
I I I I 1’ I I )
-40 -30 -20 ~10 0 10 2'0 30 40
Magnetic ﬁeld (kG)
Fe . .
Cr
Fe

 

Figure 2.13 Magnetoresistance curves of F e/Cr multilayers at 4.2 K [32].

The GMR is related to the spin dependence of the electron conduction in
ferromagnetic metals and alloys. Electron scattering depends on whether the electron
spins (S,=il/2) are parallel or anti-parallel to the magnetic moment of the magnetic
layers. Figure 2.14 illustrates the GMR mechanism in the two magietic states: parallel
and antiparallel conﬁgurations. The resistance of the magnetic multilayers can be
changed by altering the orientations of the magnetic moments of the F layers to be either

parallel or antiparallel to each other. The GMR is obtained by switching the magnetic

28

moment from the antiparallel to the parallel conﬁguration by applying the external

magnetic ﬁeld.

Parallel Conﬁguration Antiparallel conﬁguration

 

 

 

 

 

 

 

 

F N F W F N F
"”T V

Figure 2.14 Schematic representations of the scattering electrons in a magnetic
multilayer. The signs + and - are for spins Sz=+1l2 and Sz=-1/2, respectively. F
represents a ferromagnetic layer and N represents a non-magnetic layer, after [33].

2.3.1.2 CIP and CPP

GMR effects have been obtained in two geometries. In the ﬁrst one the electric
current is applied in the plane of the layer (therefore denoted as CIP), while in the second
one the cm'rent ﬂows perpendicular to the plane of the layers (therefore denoted as CPP)
[34]. In general, the GMR measured in CPP is larger than that measured in CIP for the
same system. The difference is due to the two different scaling lengths of the process.
The scaling length of the CIP geometry is the mean free path, while the scaling length of
the CPP geometry is the spin diffusion length, which is larger than the mean free path
[34]. CPP-GMR is the major candidate for the next generation of read head structures

since it has the advantage of development to a higher area density over CIP-GMR.

29

2.3.1.3 Exchanged-Biased Spin-Valves

The exchanged-biased spin-valve structure is designed to obtain an antiparallel
conﬁguration in the multilayer [35]. In this case, the multilayer is in the form of
F/N/F/AF, which is composed of two ferromagnetic layers separated by a nonmagnetic
layer and an additional antiferromagretic layer (see ﬁgure 2.15). The antiferromagnetic
layer acts to pin the neighboring ferromagnetic layer. When the magnetic ﬁeld is
increased ﬁ’om negative to positive values, the magnetization of the free layer reverses in
a very small ﬁeld, while the magretization of the pinned layer remains ﬁxed [36]. The
increase of the resistance in a small ﬁeld obtained in this way is now used for many low-
ﬁeld applications such as magnetic sensors, read heads, and Magnetic Random Access

Memories.

Ferromagnetic layer (free layer)
Nonmagnetic layer (spacer)
Ferromagnetic layer (pinned layer)

Antiferromagnetic layer
(pinning layer)

 

Figure 2.15 Schematic representative of a spin-valve structure.

2.3.1.4 Current-Induced Magnetization Switching

Current-induced magnetization switching (CIMS) has attracted technical interest

based on the desire to switch the bits in magnetic memory by local current directly rather

30

than by the ﬁeld of external wires [37]. CIMS in F/N/F metal trilayers has potential
applications for writing nanopillar magnetic memory units and for CPP-MR read heads
[3 8]. Since the CPP-MR is affected by spin polarization, one might expect that the
switching current (15) between the parallel and antiparallel states of the F layers would be
inversely proportional to the change in resistance AR. Indeed, AR was Shown to be

proportional to 1/15 in Py/Cu/Py nanopillars (Py = permalloy = NimFex with x~0.2) [3 7].

2.3.2 Correlation between GMR and Multilayer Structures
In reality, the multilayer gowth mechanisms, atomic structure, and interfacial
structures play important roles in the GMR effects. For example, if GMR multilayers are
deposited with a rough interface or chemically intermixed layers, and/or a high
concentration of crystal defects, electrons will be scattered more at the interface and/or
defects, which in turn will diminish the GMR properties.

A broad range of microanalysis research has been reported on the correlation
between the GMR effects and the multilayer structures, including layer structures,
interfacial roughness, and layer intermixing. For example, Dang and co-workers [39]
studied the inﬂuence of the Co crystal structure and GMR effects in Co/Cu multilayers
using nuclear magnetic resonance (NMR). Their results showed that GMR effect was
enhanced by high quality layers and diminished by increased numbers of stacking faults
in Co. Fullerton and co-workers [40] also showed that in Fe/Cr multilayers, the GMR
was enhanced by the interfacial roughness due to strong Spin-dependent scattering of
electrons. In contrast, Speriosu and co-workers [41] showed that in NiFe/Cu multilayers

or spin-valves, the GMR is strongly reduced by the presence of an intermixed region at

31

the NiFe/Cu interfaces. Hence, while there is general ageement that individual layer
structures will inﬂuence the MR of the multilayers [42], the roles of interface roughness,

crystal defects, diffusion, and phase formation at the interface are not well understood.

2.3.3 Motivation of This Study

Recently, Garcia et al. reported that the room-temperature ARs of Py/Cu/Py
nanopillars were ten times larger than those of Py/Al/Py nanopillars, while the Is
increased only modestly from Py/Cu/Py to Py/Al/Py [43]. These results were interpreted
as suggesting a sigriﬁcant difference in the spin-dependent scattering properties of the
Py/Al interfaces and Py/Cu interfaces.

To get a better understanding of the phenomenon, both Co/Al and Py/Al
multilayers were studied using the CPP-MR measuring technique at MSU [38, 44]. Two
parameters were characterized:

(1) the enhanced speciﬁc resistance 211R} , N = (ARI, ~+ AR}, 100 (A is the area
through which the CPP current ﬂows),

(2) the scattering asymmetry ymv = (ARﬁ/N-ARLN) /( ARI,~+AR;,N) at the F/N
interface.

Comparing the results of this study with the well-characterized F/N systems
Py/Cu, Co/Cu, and Fe/Cr [34, 45], it was found that 2Apr/A1 is unusually large while
'YPy/A] was very small. In addition, the values of R for F/Al multilayers increased by 5-
10% after 6-11 months and then by an additional 2-7% upon annealing to 180°C [46].

Both the interfacial properties and the changes in MR might be affected by the

structures and changes therein of: (1) the individual layers and interfaces, (2) interfacial

32

roughness, and/or (3) by formation of intermediate phases, as a range of intermediate
phases are known to form in such systems [19]. For example, several inter-metallic
phases (listed in table 2.7) are known to form in Co-Al system, as shown in the
equilibrium phase diagam (ﬁgure 2.16). Vovk and co-workers [47] have found
relatively rapid A19Coz and B2-ordered CoAl phase formation in Co/Al bilayer atom
probe tips annealed at 300°C for 5min. Likewise, as both Fe and Ni form a range of
intermetallic compounds with Al [19], one might expect additional phase formation in

Py/Al multilayers.

33

00

£3.00 aaooaom 3534.

2

.8: Ease case 2.8 2a 2&5

 

 

 

 

 

 

 

 

 

 

34

 

c
8" o—a-nn O” G“ On” Ohm DbPPO—vpbybhbhbbFOGr bhbcbmu bbpbbbD—bptbbbbr 0°"
-hIDb I b F b DID thththth _ PPF lb}. Pb bi D F} lb D F. I'D pﬂfht F D D.>I V.
_ . n
AOOHV _ u n
-I ""l“l""“"l"I IIIIII“ IIIIIIII . V
n m
s
. I
g“ - ﬂ .-
n n o n
A u .w n
n - H
In " moon
m " IQI. 8.34.8.
” . . n
A - Q t L
. . $0.2 . n a
A . ~ .I-
II. a s I O8 w
. . . . d
I n! o \ v
u \ Honda—nouns: 03254: n \ n m
\ v
u s . \\ I 9
A ” IHHuI... h i
1.. - S ”Ing n
A u 0.82 x n r
Gog“ . or. ado
. . .188
I“ . r
A - ‘
m OO~< . w
A C v
a n woe:
x u \ v
. . . s .
n . . \s 1— n
003:“ .. \\ n
u I ~ \\\ n
I” I’””"“~.. \5 ‘ \ mgﬁﬂ
U I u u
........ trill...F..I...I.....-
...... b...-. TIL-PI»I_IIIII->bIIPb>PI.rIIdIPIPPI>>>..-—. .b q q _ O—d c
03 cm on or on an ov on ON

 

:38 388m Ewe;

Table 2.7 Co—Al crystal structure data [19].

 

 

 

 

 

 

 

 

 

 

 

 

A19Co2 18. 1 P21/c A19Co2 monoclinic
A113Co4 23.5 Cm AlBCo4 monoclinic
Al3Co 25.6 N/A

A15C02 28.6 P63/mmc A15C02 hexagonal
AlCo ~48 to 78.5 Pm—3m Cle Cubic

 

Furthermore, interesting MR behavior was found in Py/Cu/Py based spin-valve
structures by adding a thin layer of aluminum to the center of the Spacer (Cu) layer
between the two ferromagnetic layers [48]. There is a Sigriﬁcant difference in resistance
between the parallel (P) and anti-parallel (AP) states, while the difference in the
resistance between the P and AP states decreased Sigiiﬁcantly by increasing the thickness
of the Al layer [48]. One would expect that this behavior is attributed to the formation of
an Al-Cu alloy. The Al-Cu phase diagam shown in ﬁgure 2.17 and crystal data listed in
table 2.8 Show that a number of intermediate phases exist in the equilibrium or metastable
states, with some of these phases still poorly deﬁned. To understand the MR behavior
variation that results from adding an Al layer to the spin-valves, both Cu/Al multilayers
in different thickness ratios and selective Py/Cu/Py based spin-valve structures with Al
interlayers have been studied.

To understand the inﬂuence of the multilayer structures on the MR behavior and
their apparent instabilities, all of the multilayer structures, including F/Al (F=Py, Co)

multilayers, Cu/Al multilayers, and spin-valve structures, were characterized using cross-

35

sectional transmission electron microscopy. Individual layer and interfacial region
structures were investigated by high resolution transmission electron microscopy
(HRTEM) with associated fast Fourier transform (FFT). In complement, electron
energy-loss spectrometry (EELS) and X-ray energy dispersive spectrometry (EXDS),
performed in conjunction with scanning transmission electron microscopy (STEM), were

carried out to characterize the composition nature of the multilayers.

36

.8: Same 83% 2-5 A a .N 25mm

 

 

 

 

 

 

 

 

 

 

 

 

 

 

do pounce “Geckos.— 250.5.
a. ... .... .... .... ... ....
<vluu ‘ d
I a
M III#
I a
m 0.33
“ ...
.u .
W F n
.
c
u
I W... "
.... u
...1.
a . u..-
l v
p. .
a
of;
. e
OONQ.§P “ ........ — ... .I .. I. I.:b.p.fnp...»:I.III..-.>IIIpIhp—II.IIIII
8. av mu m» an

 

 

8&8 ...:88m Emma:

30 arnseaeduraj,
37

Table 2.8 Cu-Al crystal structure data [19].

 

 

 

 

 

 

 

 

 

 

 

 

 

Phase Giff/00 823:0'1 233:; Prototype “21::
0A12Cu 31.9 to 33 .3 I4/mcm A12Cu tetragonal
nlAlCu 49.8 to 52.4 Cmmm N/A orthorhombic
nzAlCu 49.8 to 52.4 C2/m N/A monoclinic

C1A13Cu4 55.2 to 59.8 P6/mmm N/A hexagonal
C2 55.2 to 56.3 N/A N/A N/A
82 55.0 to 61.1 P63/mmc N/A hexagonal
5 59.3 to 61.9 R-3m N/A rhombohedra
71A14Cu9 62.5 to 69 P-43m A14Cu9 complex cubic
BAICu3 70.6 to 82 Im-3m w 2:131)?
002 76.5 to 78 N/A N/A N/A
[3’ N/A Fm-3m BiF3 bcc

 

 

 

 

 

38

 

Chapter 3 Experimental Approaches and Procedures

This chapter presents the experimental approaches and procedures of all the
studies throughout this thesis. The preparation procedures for the bulk Gd oxide, metallic
Gd, and GdSiz reference samples for EELS quantiﬁcation studies are ﬁrst described. The
procedures for the gowth of the silicide (including Gd and Tm silicide) nanostructures
using ultra-high vacuum evaporation are then presented. The details of all the magnetic
mutlilayers sputtered on Si substrates are then described. These are followed by a
detailed discussion of the transmission electron microscopy (TEM) characterization,
including cross-sectional TEM sample preparation, high resolution TEM (HRTEM)
imaging conditions, and parameters used in image Simulation. At last, electron energy-
loss spectroscopy (EELS) characterization is described by covering the type of the EELS
spectrometer used in this study, spectrum collection, the curve ﬁtting approach for
quantiﬁcation, Spectrum-image collection, and the simulation method used to interpret

the intensity proﬁles from the spectrum-images.

3.1 Bulk Gd Oxide, Metallic, and Silicide Reference Samples
3.1.1 Bulk Gd203

The bulk Gd oxide TEM reference sample was prepared as follows. Micron
sized Gd203 powder (ﬁom Alfa Aesar) was ﬁrst characterized with x-ray diffraction
(XRD) using a Scintag XDS 2000 diffractometer with Cu-Ka radiation to verify that the
Gd203 was the Mn203 (Bixbyite) crystal structure. The powders were then gound with a

mortar and pestle, and then suspended in methanol using ultrasound. Smaller sized

39

powders were picked up on a holy carbon ﬁlm supported on a 200 mesh Cu TEM gid.

In this way, thin platelets of Gd203 were used to carry out EELS study.

3.1.2 Bulk Metallic and Silicide Samples Produced by DC-magnetron Sputtering

Bulk metallic Gd and Gd Silicides thin ﬁlms were prepared using DC-magnetron
triode sputtering. Si(001) wafers were cleaned using acetone, then methanol, and dried
using dry N2. Alternating layers of Gd and Si were deposited on the Si substrate in a DC-
magnetron triode Sputtering system with a base pressure of 3><10'8 Torr. To produce
stoichiometric GdSi2 (65.4% atomic percent Si in the range of orthorhombic GdSiz, per
the Gd-Si phase diagam), Si 43 nm/ Gd 75nm/ Si 85 nm/ Gd 75 nm/ Si 43nm layers
were deposited. These thicknesses were monitored using a quartz crystal thickness
monitor. The base Ar pressure during Sputtering was 2><10'3 Torr. After sputter
deposition, the selected multilayers were annealed in a high vacuum of 5><106 torr at 900
°C for 1 hour to form Silicides. Off stoichiometric ﬁhns did not form well deﬁned
silicide ﬁlms after annealing (composed of Si and silicide gains). The annealed layers
were faced to un-annealed layers to make TEM samples (section 3.4), allowing

examination of both bulk Silicides and the base metal in the same TEM sample.

3.2 RE Silicide Nanostructures Grown in UHV Chamber

RE silicide nanostructures were gown by evaporating the RE metals onto Si (001)
substrates in an ultrahigh vacuum (UHV) chamber (base pressure 2.0X10'10torr), as
illustrated in ﬁgure 3.1. Prior to deposition, the Silicon substrates were pre-cleaned by

washing in acetone and isopropyl alcohol in ambient atmosphere, pre-oxidized in a UV-

40

Ozone chamber, then transferred into the vacuum chamber and repeatedly heated up to
1175 °C to remove the surface oxide. The temperature of the substrates was measured
using an optical pyrometer. The cleanliness of the substrates was checked by scanning
tunneling microscope (STM) imaging of the Si(001) surface, where defect densities were
typically well below a few percent of a monolayer.

During the deposition of the RE metals, the Si substrates were maintained at 600
°C. The RE metal was sublimed when heated by a current passing through a tungsten
wire support basket and then condensed on the Si substrate. The condensation rate was
calibrated by a quartz crystal thickness monitor and the metal coverage was determined
by timed exposure to the source. The RE metal coverage ranged from 1 to 3 monolayers
(ML) [1ML=6.78><10l4 atoms/cm2 =surface atomic density on Si(001)]. After the
samples were cooled down to room temperature, they were transferred to an in-situ STM

chamber for study.

  
  

600C // /'

i\ W//:;‘ evaporated
8' metal

Figure 3.1 Schematic representation of RE silicide nanostructure gowth
experimental setup.

41

3.3 Magnetic Multilayer Synthesis

All the magnetic multilayers were gown on Si(001) substrates at low temperature
(from -30 to 0°C) using a dc-magnetron triode sputtering system with a base pressure of
2x 10'9 torr.

Ten Co(or Py)/Al bilayers consisting of a 20 nm Co (or Py) layer and a 10 nm Al
layer were sputtered on Si(100) substrates. For comparison with the as-sputtered samples,
some multilayers were annealed for less than 5 min at 180°C after sputtering in the same
chamber.

Alternating Cu/Al multilayers were gown in the same dc-magnetron triode
Sputtering system with similar conditions. These multilayers were sputtered in different
thickness ratios: (1) Cu(8nm)/Al(10nm) with an overall composition of 52.9at. % Cu,

(2) Cu(5nm)/Al(3nm) with an overall composition of 70.1at. % Cu.
Some of the Cu(8nm)/Al(10nm) multilayers were annealed for less than 5 min at 180°C
for comparison with samples in the as-sputtered condition.

The two spin valves studied had the following conﬁgurations, where all the
values are given in nanometers:

Nb(1 50)Cu(5)FeMn(8)Py(24)Cu(l 0)Al(x)Cu(l 0)Py(24)Cu(5)Nb(50),

where x=10 for the as-sputtered sample and x=30 for the annealed sample.

3.4 TEM Characterization
Most of the Specimens in this study were characterized using cross-sectional TEM.
The cross-sectional TEM sample preparation is introduced in this section. This is

followed by a description of HRTEM imaging conditions and image simulation analysis.

42

3.4.1 Cross-sectional TEM Sample Preparation

TEM cross-sectional samples were prepared using the sandwich technique by
joining the thin ﬁlms on the Si substrates (including bulk thin ﬁlms, silicide
nanostructures, and magnetic multilayers) face-to-face and sectioning along a Si<110>
direction lying in the surface plane of the Si substrate. The samples of the sputtered thin
ﬁlms and the silicide nanostructures were separately joined using G1 epoxy (Gatan),
which can be cured at 100 °C for 5 min. The samples of the magnetic multilayers were
joined using M-bondTM 610 epoxy (SP1) and cured at room temperature for a relatively
long time (i.e. 24h) to prevent the samples from experiencing high temperature during
sample preparation. The cross sections were glued onto molybdenum (Mo) rings, pre-
thinned by mechanical thinning and dirnpling, and then low angle ion milled to electron
transparency. Figure 3.2 shows a schematic representation of the entire process of the
sample preparation. The details of the procedures to form cross-sectional TEM samples
are as follows [49]:

1. Section thin ﬁlm sample on substrate into about 4 mm X 7 mm slabs with a
diamond scribing pen.

2. Clean slabs in acetone, followed by a rinse with ethanol.

3. Glue two slabs of the thin ﬁlm face-to-face with G1 epoxy. The epoxy was cured
by heating to 100 °C for 5 min on a hot plate. (The magretic multilayers were
joined using M-bondTM 610 epoxy and cured at room temperature for a relatively
long time i.e 24 h.)

4. Slice the slab perpendicular to the gowth plane into 800~1000 um thin slabs

using a diamond saw (Struers Accuton-5 operated at 3000-rpm wheel speed with

43

a 0.05 mm/sec feed rate).

5. Glue Slices onto Mo rings with G1 epoxy and cure by heating to 100 °C for 5 min
again. (The cross sections of the multilayer samples were bonded to M0 rings
using M-bondTM 610 epoxy and cured at room temperature for a relatively long
time i.e 24 h.)

6. Mount the Slab and Mo ring to the center of a sapphire ﬂat with a thin layer of
CrystalbondTM mounting wax. The CrystalbondTM wax is heated to 130 °C on a
hot plate for 5 min. The melted wax is transferred using a toothpick onto the
backside of the Mo ring (which is brieﬂy held on a glass slide on the edge of the
hot plate. Temperature measurements Show this area to be at ~ 80 °C.). A
sapphire ﬂat, preheated on the edge of the hot plate (~ 80 °C), is placed onto the
ring and then removed with the sample and ring. This entire bonding process
takes place less than 30 seconds. The crystal bond solidiﬁes in a matter of
seconds.

7. Grind and polish the open side of the sample assembly using a series of (15, 6, 3,
l um) diamond lapping ﬁlms (Allied) until the slice thickness is around 100 um.

8. Attach (using a screw ﬁtting) and center the sample assembly on a VCRTM Group
D500i Dimpler platen using an eccentric platen assembly. A 15.5 mm stainless
steel dimpling wheel and 1 urn diamond Slurry are used. The dimpling is done
until a center thickness of about 20 pm is reached.

9. Remove the sample assembly from sapphire ﬂat with acetone to dissolve the
crystal bond.

10. Rinse the sample assembly in ethanol and air-dry.

ll. Ion mill using a Gatan M691 PIPSTM ion mill with a Gatan DuoPostTM support.
The center of sandwich interface is ion milled at ﬂ°~ :l:4° beam incident angle

with 4 keV Ar+ ions until perforation.

    

/ glue
cut 1n half face-to—face
cross-sectional
view
mount on

Ar‘ gun dimple aﬁer

grinding

   

Mo ring

 

I, Low angle
ion milling
{-—

  

(———

 

cross-sectional
TEM sample

Figure 3.2 Schematic representation of the process of preparing a cross-sectional TEM
sample.

3.4.2 High Resolution TEM

HRTEM was carried out using a JEOL JEM2200FS ﬁeld emission gun
TEM/STEM working at 200KV, which has a C5, of 0.5 mm, and hence, a theoretical
point-to-point Scherzer resolution of 0.19 nm. During observation, Si substrates were
tilted to Si<110> zones to orient the thin ﬁlm (either bulk Silicides, silicide
nanostructures, or multilayers) edge-on. HRTEM images were collected using the Gatan

DigitalMicrogaphTM microscope control and imaging software. Fast Fourier transform

45

(F FT) diffractogams were obtained and analyzed using ImageJTM. By comparison with
the diffraction pattern simulated using J EMS (developed by Pierre A. Stadelmann at
Center Interdisciplinaire de Microscopic Electronique in Switzerland [50]), crystal zones
and structures were identiﬁed. Some of the structural features were too small to obtain
good quality F FTs. In this case, HRTEM image simulation using I EMS was performed
to interpret the crystal structures. The selected microscope parameters are as follows:
accelerating voltage 200 keV, Spherical aberration Cs 0.5 mm, chromatic aberration Cc

1.0m, with thickness and defocus as variables.

3.5 EELS Characterization
3.5.1 Design of Electron Energy-Loss Spectrometers

Two types of EELS spectrometers are presently used in TEM: the omega (9)
ﬁlter (LEO) and the magnetic prism spectrometer (Gatan). The {2 ﬁlter disperses the
electrons in the TEM column. It is placed between the intermediate and projector lenses,
as Shown in ﬁgure 3.3a [2]. The 0 ﬁlter is used mainly for energy-ﬁltered imaging
although spectra can be obtained. It is a specialized spectrometer that has to be built into
the microscope column rather than an added option. The Gatan Image Filter (GIF)
(shown in ﬁgure 3.3b) is mounted under the microscope column without changing the
optical path in the microscope [2]. The JEOL JEM2200FS transmission electron
microscope in the Center for Advanced Microscopy at MSU is equipped with an in-

column Q energy ﬁlter allowing EELS and energy-ﬁltered TEM (EFTEM).

46

f To projector lens crossover
object plane of spectrometer

     

 

 

 

 

.... 8L rrj‘)‘

Sextupoles

 

 

Figure 3.3 Schematic overview of the in-column (2 ﬁlter (a) and post-column Gatan
Image Filter (b) after [2].

3.5.2 EELS Spectrum Collection

Electron energy-loss spectra were collected in STEM mode with a ~0.5nm size
probe. The zero-loss peak was collected before each analysis to conﬁrm the spectrometer
had good energy resolution (between 1.0 to 1.5 eV FWHM, see ﬁgure 3.4). This was
done with a minimum exposure time (i.e. 0.01s) to avoid damaging the CCD camera.

The Gd M45-edge was selected for quantiﬁcation analysis to understand the
electronic structures of Gd in the samples. Since it is a high energy-loss edge (~1185eV),

this edge requires a 10 sec acquisition time to achieve suitable counting statistics.

47

 

 

Intensity (arb. units)

 

 

 

 

 

 

 

. I ‘ ‘
5 I
I
' l
‘ . l i .
i i
u..- :L i - a 4
. . = 9 .
I : -
.' ‘ i
__ l 2
I 17 l l I 1

20 40 60 80 100 120 I40 100 180
Figure 3.4 Zero-loss peak with a full width atelilalf maximum (FWHM) ~1.0 eV shows
that the EELS spectrometer is well-aligned.

3.5.3 Gd M-Edge EELS Spectrum Curve Fitting

For quantitative analysis of the Gd M45-edge EELS spectrum, spectrum model
ﬁtting techniques developed by Manoubi et al. [51] were used to extract useful
information from experimental spectra.

The experimental spectra of M45-edges were ﬁtted as a superposition of a pre-
edge power-law backgound, with a set of Lorentzian proﬁles to account for the white-
lines, and a continutun term corresponding to the transitions to empty states in the

continuum modeled by a Hartree-Slater function of cross sections extracted from Gatan

DigitalMicrogaph [52] (ﬁgure 3.5).

3.5.4 Spectrum Image Collection
To analyze the interfacial chemical nature of the F/Al multilayers, a series of Spatially
collected EELS spectra, referred to as a EELS Spectrum image, were collected by

scanning an ~05 nm diameter (FWHM) probe across the interfaces of the F/N

48

multilayers. Each spectrum was composed of 1024 EELS channels of 0.11 eV width.

Low loss spectrum images with plasmon peaks were acquired over scans of ~50nm

across F/N/F/N/F layers. Core loss spectrum images with white lines for Co (or Ni, Fe)

were acquired using scans of ~20nm across the F/N/F layers. During the acquisition

process, specimen drift caused by thermal effects was corrected using Gatan

DigitalMicrogaphTM by periodically scanning over a cross-correlated reference region in

the survey images.

Intensity (arb. units)

 

1

I'l

r

 

GdSi2 EELS spectrum curve ﬁtting

M

5.

 

 

 

l' \—~~
I l I I l 1 I
l 160 l 180 1200 1220 1240
Energy loss (eV)

Figure 3.5 A typical EELS spectrum of M45 edge (black line) from GdSi; ﬁtted as a
superposition of a power-law background (straight line), cross sections from Hartree-
Slater model (rising line), and a set of Lorentzian proﬁles (gay lines). The dotted

curve shows a good curve ﬁtting result.

3.5.5 Intensity Proﬁles and Simulation

49

The acquired spectrum images were processed in the Spectrum Imaging package
in Gatan DigitalMicrogaphTM. Each spectrum in the image (core loss spectrum image)
was extracted and then normalized by subtracting a pre-edge power-law backgound.
The intensities of the peaks (plasmon peaks and core loss peaks after backgound
subtraction) were then plotted as a function of distance. The intensity proﬁles of the as-
sputtered and annealed F /Al multilayers were then compared.

To facilitate an understanding of the experimental F L-edge intensity proﬁles,
theoretical intensity proﬁles across perfect F/Al/F interfaces (ﬁgure 3.6) were simulated

using an electron beam with a 0.5 nm (FWHM) Gaussian proﬁle.

Normalized intensity

rjmvﬁrf'vvv' v 'u v'

 

 

 

 

2.5 5 7.5 10 I25
Distance (nm)

Figure 3.6 Simulation of an F (Co, Ni, or Fe) L-edge intensity proﬁle with a 0.5nm
FWHM of Gaussian electron beam proﬁle across idealized F/Al interface.

50

Chapter 4 Results from RE Silicide N anostructure Studies

In this chapter, the results of the HRTEM and EELS studies of Gd and Tm
silicide nanostructures are presented. The crystal structure study of the bulk phases,
including Gd oxide, metallic Gd and GdSlz in thin ﬁlm form, are presented ﬁrst. This is
followed by cross-sectional HRTEM study of Gd and Tm silicide nanostructures. EELS
quantiﬁcation study of the Gd silicide nanostructures compared to the bulk phases will be

presented last.

4.1 Gd Oxide, Bulk Metallic and Silicide Crystal Structure Study

To understand the extent that the nanostructures vary from the bulk materials,
bulk samples, including Gd oxide, metallic Gd, and GdSi; in thin ﬁlm form, were used as
references for crystal and electronic structure study.

The crystal structure of the as-received Gd203 powders was studied by 0-20 x-ray
diffraction (XRD) scan. The scan is presented in ﬁgure 4.1, which indicates that the
Gd203 has a lattice constant a=10.813 A, space goup Ia3, and is consistent with the
prototype of Mn203 (Bixbyite) structure [50].

Figure 4.23 Shows three gound GdzO3 powder particles on a carbon holy ﬁlm.
These particles are in the range of a hundred nanometers in diameter. Insets (b) and (c)
Show that the individual particles can be either single crystals or polycrystals. The thin

areas of the Gd203 particles are suitable for EELS study.

51

60

 

50 -

Counts (CPS)

 

 

 

 

20 25 30 35 40 45 50 55 60 65 70
2 0 (degree)

Figure 4.1 X-ray diffraction Shows the bcc Gd203 agee with the prototype of
Mn203 (Bixbyite) structure.

 

—
100 nm

Figure 4.2 (a) Phase contrast TEM image of Gd203 powder particles on a holy
carbon ﬁlm. Insets (b) and (c) Show HRTEM images of a Single gain and poly-
gain atomic lattice from two powder particles respectively.

52

Figure 4.3 shows a cross-sectional TEM image (a) and a corresponding selected
area electron diffraction (SAD) pattern (b) of the sputtered Si/Gd layers before annealing.
The Si is amorphous and the Gd displays ﬁne crystalline gains. The selected area
electron diffraction pattern shows polycrystalline diffraction rings corresponding to the
hexagonal crystal structure of metallic Gd (with c/a ratio of 1.59). It is known that this
RE metal is reactive and oxidizes readily. Nevertheless, subsequent EELS examination
of this layer revealed no obvious oxygen K edge. As a result, metallic Gd EELS spectra

can be collected from the Gd layers.

   

102
100 Amorphous 110
Si

.

Figure 4.3 (a) Phase contrast TEM image and (b) corresponding selected area electron
diffraction pattern of amorphous Si and metallic Gd layers alternately sputtered on a
Si(001) substrate.

Upon annealing at 900 °C for 1 hour in vacuum, the Gd/ Si layers transformed to
GdSi; by inter-diffusion of Si and Gd. Figure 4.4a presents a low magniﬁcation phase
contrast TEM image of the post-deposition annealed bulk GdSi; layer formed on the
Si(001) substrate. The GdSi; layer ranged in thickness from 100 to 200 nm and formed
as a relatively large gained structure, with individual gains traversing the entire layer.
A SAD pattern from one gain of the bulk GdSi; layer shows a single crystal [100] zone

(seen ﬁgure 4.4b). The HRTEM image in ﬁgure 4.4c shows the GdSi; layer imaged

53

along this [100] zone axis. The imaged atomic lattice is consistent with the projection
viewed along the orthorhombic GdSi2[100], with a half-atom shift between every two
horizontal rows of atoms (aabb stacking in ﬁgure 2.11). It has been reported that both the
hexagonal GdSi” and the orthorhombic GdSi; can be formed by annealing Gd thin ﬁlms
on Si substrate, depending on the thickness of Gd layer and the Si orientation (Si(001)
and Si(111)) [53, 54]. In our case, only the orthorhombic phase GdSi; was found to exist,
which indicates that the orthorhombic phase is energetically favorable under these
conditions. This bulk GdSi; will be used as a reference for comparison with the Gd

silicide nanostructures.

54

.7 4“"
.35»!
, . .5055”
~ .9”

Bulk GdSi2 layer

Si Substrate

 

55

 

Figure 4.4 (a) Phase contrast image of
post-annealed bulk GdSiz layer formed
on the Si(001) substrate. (b) Selected area
diffraction pattern from one gain of the
layer with arrows presenting strong
diffraction spots from the orthorhombic
phase of GdSiz. (c) HRTEM image
showing the [100] orthorhombic zone
axis from the GdSiz. The atomic lattice
symmetry is indicted by the stacking of
(001) planes in an aabb manner.

4.2 HRTEM Studies of Gd and Tm Silicide Nanostructures

This section will present results from cross-sectional HRTEM study of both Gd
and Tm silicide nanostructures. To correlate the silicide phase crystal structures with
their morphologies, some STM topogaphic images are presented before showing the
HRTEM results. The Gd silicide nanostructures imaged are from one TEM sample,
which means all the nanostructures were gown with the same metal coverage and
heating process. In contrast, the Tm silicide nanostructures imaged were from different
samples with different gowth conditions, which are described separately. Cross-
sectional HRTEM images, canied out in conjunction with FFT, are assessed with image

simulations to interpret the experimental images and identify the crystal structures.

4.2.1 Gd Silicide Nanostructures

The Gd silicide nanostructures imaged were from one TEM thin foil. The
nanostructures were gown on a substrate heated to 650°C with a metal coverage of
1.5ML. This was followed by post-deposition annealing for 20 min at the gowth
temperature. An STM image of this speciﬁc sample rendered in 3-D (ﬁgure 4.5) shows a
large number of islands, mostly compact, with a small fraction of high aspect ratio
nanowires. The line proﬁle shows the dimension of an island with varying thickness up
to 3nm from the bottom of adjacent trenches (these trenches are due to local Silicon

depletion during the formation of the silicide).

56

0))

“height (nm)

   

distance (nm)

Figure 4.5 (a) 3D topogaphy STM image of GdSi2 nanostructures gown on
Si(001) with (b) a cross-sectional line proﬁle, indicated by the line on (a).

The cross-sectional phase contrast image in ﬁgure 4.6a shows the entirety of a
GdSi; nanostructure, approximately 70 nm in width and 3-4 nm in height. The atomic
resolution HRTEM image in ﬁgure 4.6b exhibits atomic planes with aabb stacking, which
is consistent with the orth/tet phase of GdSi2 viewed along the [100] zone axis (shown in
ﬁgure 2.11b). An FFI‘, which includes information from both the GdSi: and Si (ﬁgure
4.6c), shows that there is an angle of approximately 2° between the GdSi2(020) and the
Si(220), suggesting that the epitaxial relationship deviates slightly from a perfect
GdSiz(001) // Si(001) and GdSi2[100]// Si[110]. This tilt most likely reﬂects a strain
accommodation mechanism. There is a lattice mismatch of approximately 5% between
the GdSi2(020) and Si(220) measured by the FFT. The lattice parameters measured from
this nanostructure are a (and/or b) = 0.40 nm, c = 1.35 nm, which agee well with the
expected values for bulk silicide lattice parameters within experimental error (see table

4.1).

57

- 8 >4 -.--. .P .
‘1} iv I'K'.'”

Ia.‘ V";’.A{,I‘V\"‘ ""4"“.f’ "hr ' ‘ In
~ . a" If '31?» ' '
13:1.”"93‘ .5:-

I

(idSi2(020)i-I

t .

All
" suzztijf
‘ffsiroogi

' lSi2(00-l)/7'
‘ Nﬁﬁ‘OStructure

._'.-‘1Ir--

 

Si substrate ' l _.

Figure 4.6(a) Cross sectional phase contrast TEM image of a GdSiz nanostructure
gown on Si(001). (b) Higher magriﬁcation HRTEM image showing the interface
between the GdSiz and Si. (c) FFT of area shown in (b).

Table 4.1 Lattice parameters of the bulk GdSi2 phases and those measured from
the three imaged nanostructures. (I, II, III refers to the imaged nanostructures)

 

 

 

 

 

 

 

Structure a(nm) b(nm) c(nm) Mismatch with Si (001)
a(%) b(%) C(%)
Bulk Hex 0.388 0.417 1.0 8.6
phases
Tetra 0.410 1.361 6.77 6.77
OIth 0.409 0.401 1.344 6.51 4.43
Measured Hex 0.39III 0.42 1.6 8.0
Tetra/orth 0.40I 1.35 5
0.41II 1.36 6.8
0.39m 1.34

 

 

 

 

 

 

 

 

 

 

Figure 4.7a shows a cross-sectional phase contrast TEM image of another GdSi;

nanostructure, approximately 70 nm in width with varying thicknesses up to 7 nm. The

58

atomic resolution HRTEM image in ﬁgure 4.7b again exhibits an aabb stacking, which is
consistent with the orth/tet structure viewed along the [100] zone axis (shown in ﬁgure
2.11b). An FFT diffractogam in ﬁgure 4.7c shows the Si[110] zone and the orth/tet
phase of GdSi2[100] zone. Similarly, the epitaxial relationship identiﬁed deviates
slightly from a perfect GdSi2(001) // Si(001) and GdSi2[100]// Si[110], with a tilting
angle of ~2° between GdSi2(020) and Si(220). The lattice mismatch parallel to the
interface measured by the FFT is approximately 6.8%. The lattice parameters measured
from this nanostructure are a (and/or b) = 0.41 nm, e = 1.36 nm, which agee with the
expected values for bulk silicide lattice parameters within experimental error (see table

4.1).

' " Nanostructure

. T l-(lnin'

Si (220)

I

. /ﬂ
,cdsrzmzo

t

I
I

 

1
(.(lSIﬂlﬁ-I) Si 002

Figure 4.7 (a) Cross sectional phase contrast TEM image of a GdSi; nanostructure
with multiple thicknesses. (b) Higher magniﬁcation HRTEM image shows an aabb
lattice stacking, consistent with the orth/tet phase of GdSi; viewed along the [100]
zone axis. (c) F FT from area (b) conﬁrms the crystal structure of the orth/tet phase of
GdSiz.

59

In both cases, the fact that these relatively thick islands are entirely orthorhombic
phase of GdSi; is consistent with previous work [20]. Since the sample experienced post-
deposition annealing, this observation suggests that the orthorhombic phase of GdSi; is
the energetically favorable silicide phase.

Figure 4.8a shows an HRTEM image of a third GdSi; nanostructure,
approximately 42 nm in width with a wavy top with thickness up to 10m. The structure
is tilted relative to the Si(001) surface. The structure contains two types of atomic
stacking, separated near the dashed horizontal line Shown at the left hand side of the
image. The top part of the structure exhibits an aabb atomic stacking. The FFT from
area 1 (in inset) Shows a tilted orth/tet GdSi2[100] zone. The bottom part shows a
different atomic stacking with rectangular symmetry. This stacking is consistent with the
hexagonal phase of GdSiz.x viewed along the [100] zone (shown in ﬁgure 2113). An
FFT diffractogam from area 2 (in inset) shows a hexagonal GdSiz.x [100] zone. The FFT
also shows that there is an approximately 4° angle between the GdSi2-x(002) and Si(220),
which suggests that the epitaxial relationship deviates from hexagonal GdSi2-x(OlO) //
Si(001) and GdSi2-,.[100]// Si[110]. There is a lattice mismatch of approximately 8%
between the hexagonal GdSi2-x(002) and Si(220), measured by the FFT ﬁ’om area 2. The
lattice parameters measured from the hexagonal phase are a = 0.39 nm and c = 0.42 nm,
and the lattice parameters measured from orth/tet phase are a (and/or b) = 0.39 nm and c
= 1.34 nm, both of which agee with the expected value for bulk silicide lattice
parameters within experimental error (see table 4.1).

The relatively large mismatch (8%) between the hexagonal GdSi2(002) and

Si(220) requires a large strain accommodation at the interface. A periodic array of strain

60

contrast features along the interface (marked by arrows) is clearly resolved in ﬁgure 4.8b
(enlarged from are 3 in ﬁgure 4.8a), which is consistent for an array of misﬁt dislocations.
The spacing of the strain contrast (a pair of dark and bright constrast features) is
approximately 2 nm (i.e. 5-6 Si(l 10) planes). This spacing is very close to the spacing
between the thinner nanowires that gow in bundles at lower metal coverages (which is
reported in the range 5-7 unit cells, i.e. 2-3 nm) [55]. The Burger circuit shows that there
are approximately 9 GdSi2-x(001) planes on every 10 Si(l 10) planes, resulting in a semi-
coherent interface. The Burger vector b measured is 1/2<110> in Si, existing in every
two contrast features. However, the 9:10 mismatch, with the observation of
approximately 5-6 Si(110) planes suggests that the observed dislocations are partial
dislocations. It should be noted that if these are in fact partials, they would have
b=l/4<1 11> in the Si, with displacements out of the interface plane. Based on the
measured lattice mismatch between Si and Silicides, the spacing between the 1/2<110>

dislocations should be 4.48nm which covers two pairs of dark and bright contrast periods.

61

,‘- ."
I

A.--.',Nanostruct'ure.o ' ' _ . (“5‘2
A (020)

84002), i P05521016)
. . . 2 ‘\‘ ’/

I

I

I

I

I

I

I
,l
.I
I

 

Figure 4.8 (a) An HRTEM image of a GdSi; nanostructure with a wavy top. Two
different lattice symmetries are observed in the nanostructure on the two sides of the
dashed horizontal line. Above this line an aabb stacking is displayed, and an FFT ﬁ'om
area 1 shows a tilted orthorhombic GdSi2[100] zone. Below the line, close to the Si
substrate a rectangular symmetry with aaaa stacking is observed. An FFT from area 2
shows a hexagonal GdSi2_x[100] zone. (b) An enlarged image from area 3 illustrates the
contrast from mismatch dislocations at the interface, as well as a periodic array of strain
contrast marked by arrows.

62

According to the analysis from these GdSi; nanostructures, both hexagonal GdSiz-
x and orth/tet GdSiz can self-assemble on Si(lOO). The F FT analysis illustrates that the
lattice parameters of the nanostructure phases do not deviate signiﬁcantly from the bulk
phases. The strain caused by the lattice mismatch between the Silicides and Si substrates
is accommodated by tilting of the orth/tet phase or the introducing dislocations between
the Si and the hexagonal phase.

The fact that two phases coexist in the same silicide island is not surprising, as
similar behavior has been seen for Dy silicide islands [56]. The fact of one phase
overgows another suggests a strain relaxation mechanism is acting in these
nanostructures. We can deﬁne a two dimensional (2D) mismatch as the product of the
mismatches in the orthogonal directions. Taking the bulk data from Table 4.1 for the hex
phase:

1.01 x 1.086 = 1.097 -) 9.7% for 2D mismatch.

In a Similar manner, the 2D mismatch between the substrate and the orthorhombic phase
is 11%, and the 2D mismatch between the two Silicides is 1.3%. It can be seen that the
mismatch between the hexagonal and orthorhombic phases is very small, and Should
result in a coherent or semi-coherent interface. A planar, semi-coherent interface is likely
lower in energy than a more incoherent interface associated with tilting, which will
require a combination of dislocation and/or atomic ledges to accommodate the mismatch.
As the hexagonal phase has a smaller 2-D lattice mismatch with the Si substrate than the
orthorhombic phase, having the hexagonal phase act as a buffer layer at the interface
between the orthorhombic phase and the substrate may avoid complex interfaces and

provide strain relief.

63

Another possible explanation for the observed overgowth of the orthorhombic
phase on hexagonal may be that it results from the intersection of two gowing structures,
since other work has Shown that the orthorhombic phase can nucleate at the intersections
of intersecting hexagonal islands [20]. Thus, it is possible that this cross section reveals

the overgowth of the orthorhombic phase over a hexagonal island.

4.2.2 Tm Silicide Nanostructures
4.2.2.1 Nanostructure Morphology

The Tm silicide nanostructures were gown at approximately 600°C with a metal
coverage of l to 3 ML, either with or without post-deposition annealing. Figure 4.9a
shows an STM image of nanostructures gown with 2.8 ML metal coverage without post-
deposition annealing (Sample Tm-A). There are two basic morphologies of
nanostructures gown on the Si surface: nanowires and islands. The nanowires are ~ 10
nm wide and 0.3 to 1.3 nm high. The islands typically have much lower aspect ratios and
are often observed at nanowire intersections. It is difficult to discriminate islands and
nanowires on the basis of height, but the island widths are typically 20 nm or more.
Figure 1b shows the cross sectional proﬁle along the line displayed in the image. Three
nanostructures are seen: (1) a multiple layer thick nanowire, (2) a one layer high
nanowire ~0.34 nrn high and 10 nm wide, and (3) an island ~30 nm wide. Typical
features exhibited by islands include a minimum height of 1 nm, a curved top, and
adjacent trenches that indicate local depletion of Si.

Some of the islands Show a periodic modulation in height. One of these islands,

from a sample gown at 1 ML metal coverage with a post-deposition annealing at 600°C

for 30 min, is shown in ﬁgure 4.10 (Sample Tm-B). This island shows a modulation of
about 0.1 nm in amplitude and a lateral period of about 10nm. This modulation period in
diﬂerent islands ranges from 7 to 15 nm. The modulation is seen in both ﬁlled and empty

states STM images, strongly suggesting that it is structural rather than electronic in origin.

 

(a) 10.5 nm

(2)

 

 

 

Figure 4.9 (a) An STM image of Tm silicide nanostructures on Si(001) with 2.8 ML
Tm coverage. (b) A cross sectional proﬁle along the line shown in (a).

 

 

 

 

 

 

Figure 4.10 (a) An STM image of Tm silicide nanoisland with a modulation in height.
(b) Cross-sectional proﬁle along the line shown in (a).

A third sample (Tm-C) was gown under very similar conditions to Tm-B: 1 ML
Tm deposited at 600°C with 20 minutes post annealing. However, this sample was also
capped with a layer of amorphous Can in an attempt to decrease the amount of post

gowth oxidation that occurs once the samples were removed from the UHV system.

The most sigriﬁcant difference between samples Tm-A and Tm-B was that Tm-A
was gown to a higher metal coverage and consequently had a tendency to Show a higher

ﬁ'action of silicide islands in the STM images.

4.2.2.2 HRTEM Image Simulation

HRTEM image simulations were carried out to interpret the stacking symmetries
of the projections along various axes of the potential Tm silicide phases. These the
simulated images were then compared with the experimental HRTEM images so that the
crystal structures could be identiﬁed.

Figure 4.11 shows a comparison of the projections and simulated images of the
hexagonal Tm3Si5 and orthorhombic TmSi along different axes. If either of these two
structures forms nanowires, ﬁgures 4.11 a and c would be expected to be the side view of
the nanowires, while ﬁgures 4.11 b and (1 would be expected to be the “end on” views,
based on the anisotropic gowth mismatch with these phases.

The simulated image along the hexagonal Tm3Si5[001] (ﬁgure 4.113) displays a
six-fold symmetry, while the simulated image along the orthorhombic TmSi[lOO] (ﬁgure
4.11c) shows an arrangement of ﬂattened and distorted hexagons, and also a clear band in
dark contrast between every two atomic layers (note that the high Z RE cations dominate
the contrast). The simulated image along the hexagonal Tm3Si5[100] (ﬁgure 4.1 lb)
Shows a rectangular symmetry with aaaa stacking, while the simulated image along the
orthorhombic TmSi[OOl] (ﬁgure 4.11d) shows a aabb stacking. These differences

facilitate discrimination of the two phases in HRTEM cross-sectional images.

66

 

3 '3 '3
. . 2 . 2 .

. . 0 .
131919911011 along Image simulation
Tm3Sls [001] along TmJSis [001]

 

Projection along Image simulation
TmSi [100] along TmSi [100]

   

Projection along Image simulation
Tm3Si5 [100] along T1113815 [100] Projection along Image simulation

TmSi [001] along TmSi [001]
Figure 4.11 A comparison of the projections and corresponding simulated lattice
images of the hexagonal Tm3$i5 (a and b) and orthorhombic TmSi (c and d) imaged
down different axes.

4.2.2.3 Results from HRTEM Imaging

Figure 4.12a shows a cross-sectional HRTEM image of a Tm silicide
nanostructure from sample Tm-B, approximately 10nm in width and less than 1 nm in
height. The enlarged lattice image Shows a rectangular stacking symmetry with only two
atomic layers clearly resolved. This makes it impossible to determine the crystal
structure since it does not have a complete unit cell: the rectangular symmetry with only
two cation layers appears in both ﬁgures 4.11 b and d. The corresponding FFT (ﬁgure

4.12c) shows streaked diffraction peaks since very limited periodicities are present.

67

 

Figure 4.12 (a) HRTEM image shows a Tm silicide nanostructure approximately 10 nm
in width and less than 1 nm in height. (b) Enlarged lattice image shows a rectangular
stacking symmetry. (c) FFT shows streaked diffraction peaks.

Figure 4.13a shows a cross—sectional HRTEM image from sample Tm-A of a Tm
silicide nanostructure approximately 35 nm in width and approximately 1 nm in height.
Two sections of the nanostructure are noted. To the left, a section approximately 6 nm in
width is observed. This section is separated by approximately 4 nm of Si from a much
broader section approximately 26 nm in width. This nanostructure has similar
dimensions to the island seen in the STM image in ﬁgure 4.10. The rectangular
symmetry of the lattice stacking (aaaa stacking) seen in the enlarged atomic lattice image
(ﬁgure 4.11b) and the corresponding FFT (ﬁgure 4.130) are consistent with those of the
hexagonal Tm3Si5 viewed along [100] (i.e. ﬁgure 4.11b). FFT analysis reveals that the
epitaxial relationship is hexagonal Tm3Si5(100) // Si(001) and Tm38i5[001] // Si[110]
with an angular deviation as much as 5°. Because the nanostructure is so thin, only

lateral periodicities in one direction result in good diffraction peaks, and only the lattice

68

parameter c, measured to be 0.40 nm, can be determined. This value is consistent with
the bulk hexagonal silicide value 0.407 nm (listed in table 4.2). Figure 4.13d shows an x-
ray energy dispersive spectrum that conﬁrms the nanostructure contains signiﬁcant Tm.
Figure 4.14a shows a cross-sectional HRTEM image from sample Tm-C of an
elongated Tm silicide nanostructure approximately 90 nm wide and up to 3nm thick, with
a Can capping layer. The enlarged lattice image in ﬁgure 4.14b shows stacking is
similar to the simulated image of the orthorhombic TmSi viewed along [100]. The
corresponding FFT (ﬁgure 4.14c) shows a TmSi [100] zone. The lattice parameter c is
measured to be 0.38nm, which is consistent with the bulk value 0.378nm (listed in table
4.2). To get clearer contrast in the lattice image, the FFT was ﬁltered by selecting the
TmSi and Si diffraction peaks respectively and ﬁltered images were obtained from the
nanostructure (ﬁgure 4.14d) and Si (ﬁgure 4.14e). Figure 4.14d shows the stacking has a
roughly six-fold symmetry, but more signiﬁcantly, a dark contrast band between every
two atomic layers, which is consistent with the simulated image (inset) viewed along the
orthorhombic TmSi viewed along [100]. Thus, this nanostructure can be interpreted as a

nanowire with the orthorhombic TmSi crystal structure viewed on side.

69

‘- .4 (c) A
.' Sl‘lii38i5(ltm,‘/Si(tl(|2)
' < ' t 5
l . ‘ . I
, /‘. s
‘ f _'Ttii35i5(001) -‘ “(22“,

Si substrate

 

 

 

 

 

C
"5 _ Si
=3
3
3
=
5 -
O
U
0 Tm Tm Tm
1 l l A l l J _L [y MA 1““ l
0 1 2 3 4 5 6 7 8 9 10

Figure 4.13 (a) Cross sectional HRTEM image of a Tm silicide nanostructure on
Si(001). (b) Enlarged atomic lattice shows rectangular symmetry. (c)
Corresponding FFT from area (b). ((1) X-ray energy dispersive spectrum shows
chemical nature of the nanostructure.

70

Table 4.2 Lattice parameters of the bulk Tm silicide phases and those measured
from the imaged nanostructures.

 

Structure a (nm) b (mm) 0 (11m) Mismatch with Si (001)

 

 

 

 

 

 

 

 

 

 

 

 

a(%) b(%) c (%)
Bulk Orth 0.418 1.035 0.378 +8.9 +170 -l.6
phases Hex 0.377 0.407 -l.9 +6.0
Orth 0.40 1.32 0.38 +4.2 -1.0
Measured
Hex 0.40 +4.2

 

 

 

CaF7

VIA-Nanostructure

_ (10' nm

.‘

TmSii {LU}
‘/

V

 

CLO-‘4’vlfmno-0t-04s -

0

Figure 4.14(a) Cross-sectional HRTEM image of Tm silicide nanostructure from an
elongated Tm silicide nanostructure ~90 nm wide and up to 3nm thick between Si
substrate and Can layer. (b) Enlarged lattice image shows a stacking similar with the
simulated image of orthorhombic TmSi viewed along [100]. (c) Corresponding FFT
shows a TmSi [100] zone. ((1) and (6) shows a comparison of FF T ﬁltered lattice
image from the nanostructure and Si, which conﬁrms the stacking in the nanostructure
is consistent with the simulated image of orthorhombic TmSi viewed along [100]
(inset in (d)).

71

Figure 4.15a shows an HRTEM image of another Tm silicide nanostructure from
sample Tm-C, which is approximately 30 nm in width and approximately 2 nm in height.
The lattice in the left in the image (area 1) shows a stacking symmetry with a half-atom
horizontal shift between every two rows of atoms, which is consistent with the schematic
projection along the orthorhombic TmSi[OOI] in ﬁgure 4.11d. The corresponding FFT
(ﬁgure 4.15b) is also consistent with the orthorhombic TmSi[OOI] zone pattern. Since
the Tm silicide lattice is limited in height, the diffraction peaks for TmSi {040} are
streaked in the FFT. It is noted that the lattice in the right of the image (area 2) shows
different apparent symmetry. However, the FF T (ﬁgure 4.15c) reveals a tilted
orthorhombic TmSi[OOI] zone, which indicates the structure is still orthorhombic but
slightly tilted. This suggests some strain accommodation is taking place in the
nanostructure and the epitaxy on Si(001) is not ideal. The lattice parameters measured
ﬁom the FFT are a=0.40 nm, b=1.32 nm, which vary from the bulk values a=0.418 nm,
b=1.035 nm to some extent (listed in table 4.2). As a result, the large lattice mismatch
expected with the orthorhombic a axis (8.9%) turn out to be ~4.2%, which may suggest a
lattice distortion occurred. This magnitude of lattice mismatch may not constrain the
lateral growth as expected from the bulk orthorhombic phase, which may result in a
relatively broader lateral dimension. Based on the analysis from ﬁgure 4.14,
orthorhombic TmSi may be the phase that forms nanowires on Si. As a result, this
nanostructure in ﬁgure 4.15 may be a nanowire viewed end on. The relatively small
lattice mismatch (4.2%) may not signiﬁcantly conﬁne the lateral growth, which results in

a width of approximately 30 nm for this nanowire.

72

'll 1..

pppppp

' :‘,"-:\‘1t::1)l
+\~. ' '
..., ,. _ limit—111M". ‘ " U ,
. v I1 a I " VT\ I n ‘

. . lmMHHm wool)

......

  

, \\ ‘

" ,. v

’ lmxn‘lllm
/

 

Figure 4.15 (a) Cross—sectional HRTEM image of Tm silicide nanostructure on Si(001)
shows a stacking symmetry with a half-atom shift between every two rows of atoms
consistent with the orthorhombic TmSi [001] axis. (b) and (c) are the corresponding
FFTS from areas (1) and (2).

A signiﬁcant portion (2 out of 3) of nanostructures (in all the three samples) were
observed to contain more than one type of stacking lattice, which may suggest they are
made up of different crystal structures. Figure 4.16 shows two examples of this type of
nanostructure, which are from sample Tm—C. Figure 4163 shows an HRTEM image of a
Tm silicide nanostructure approximately 25 nm wide and approximately 2nm thick. Two
different lattice stackings are observed. The bottom two layers of the structure show
stacking consistent with the simulated image of the orthorhombic TmSi viewed along
[100], while the upper stacking is consistent with simulated image of hexagonal Tm3Si5
viewed along [100]. Figure 4.16b shows an HRTEM image of a Tm silicide
nanostructure approximately 10 nm Wide. Again, the bottom part of the nanostructure
has stacking consistent with the simulated image of the orthorhombic TmSi viewed along
[100], while the upper two layers are consistent with simulated image of hexagonal

Tm3Si5 viewed along [100]. FFTS cannot be used to determine the crystal structures due

73

to the limited periodicities. The poor contrast between the different stacking may suggest

these nanostructures exhibit two different crystal structures.
C a F2
Nanostructure
Si substrate
5 nm

€an

_ , ._ .7 ,_ , ./ Nanostructure ,_
195032-“-‘*““W’-"* J?" ' ' ‘ ‘ ‘.<"“>‘\\\\\s¢\s{\\.\‘"i:\mt.

Si substrate?) - “ ‘
' Stun 31...

 

Figure 4.16 Two types of lattice stacking are observed in each of the two
nanostructures (a) approximately 25 nm wide and approximately 2nm thick (b)
approximately 10 mm wide and approximately 2nm thick. Different stacking are
marked by arrows and are compared to the simulated images.

These results show that all of the Tm silicide nanostructures have rectangular
cross-sections, which is also consistent with the STM topography. Both the hexagonal
Tm3$i5 and the orthorhombic TmSi are observed to grow epitaxially on Si(001). For the

hexagonal ngsis, only the lattice parameter c was measured, but it did not vary from the

74

bulk phase. Nevertheless, for the orthorhombic TmSi phase, the lattice parameters vary
ﬁ'om the bulk to a signiﬁcant extent.

The relatively high number of nanostructures consisting of two phases, with the
hexagonal phase overlying the orthorhombic phase, may represent a new mechanism for
strain relief in this system. As was noted in section 2.2.2.1, the hexagonal and
orthorhombic phases have very different compositions, which is different than for Gd
silicide, and for most of the other RE silicide systems. Therefore, the coexistence of two
silicide phases means the stabilization of an extreme compositional gradient within this
nanostructure. Furthermore, the measured crystal structures imply that the orthorhombic
phase richer in Tm lies at the interface with the Si substrate.

If we consider the 2D lattice mismatch with the substrate, the hexagonal phase
has 4%, and the orthorhombic phase has 7%, based on the bulk silicide lattice constants.
However, the fact that the orthorhombic phase lies at the interface with the Si substrate,
although it is richer in Tm, may suggest that it has a relatively small strain with Si(001).
In fact the 2D lattice mismatch based on the measured lattice constants for the
orthorhombic phase is 3% which is less than the hexagonal phase.

Since there is poor contrast between the two phases in these structures, no clearly
resolved mismatch can be observed. Nevertheless, this overgrowth pattern may suggest
that the orthorhombic phase has a smaller 2D mismatch with Si substrate, while the
hexagonal phase is more stable than the orthorhombic phase in term of volume free

energy.

75

4.3 EELS Study of Gd Silicide Nanostructures

To study the electronic nature of the Gd silicide nanostructures, the metallic Gd,
thin ﬁlm Gd Silicides, and Gd oxides in bulk form were studied systematically as
references. Comparing the near-edge ﬁne structures provides a way to investigate how
the electronic nature of the nanostructures varies from the bulk phases.

Both the N45 and M45 edges are detectable edges in the Gd EELS spectrum. Both
of these edges represent transitions from initial (1 levels (respectively 4d and 3d) to ﬁnal
unoccupied 4f orbitals. Figure 4.17 shows the energy loss near-edge structure (ELNES)
of the Gd N edge (transition of 4d to 40 ﬁ'om the metallic Gd (a), the bulk GdSiz (b), the
GdSi; nanostructure (c) and the Gd203 ((1) powder in comparison to Gd203 spectrum
published in the standard EELS Atlas (e) [27]. The position and shape of the N peaks are
similar in the four spectra, except for the presence of the Si L edge in bulk GdSiz (b) and
GdSi2 nanostructure (0). Comparison of these spectra shows no difference between Gd
oxide and metallic Gd in terms of ELNES of the N edge.

Figure 4.18 shows the ELNES of the Gd M45 edge from metallic Gd (a), bulk
GdSiz (b), GdSi2 nanostructure (c) and Gd203 powder ((1) in comparison to published data
for Gd203 (e) [27]. Two peaks, M4 and M5, are well separated and correspond to
unoccupied 4f states with different angular moments (M5 : 3d5/2—-> 4f, M4 : 3d3;2—» 40.
Each experimental spectrum has been ﬁtted using a power-law background, Hartree-
Slater cross section and a set of Lorentzian proﬁles. The line with open squares
demonstrates a good curve ﬁtting result. The M peak positions and intensity ratios of the

white lines are measured and listed in table 4.3. All four of the spectra were similar in

76

character. Within an energy resolution of 1.5 eV obtained from the full width of half
maximum (FWHM) of the EELS zero loss peak, there are no measurable chemical shifts.

Figure 4.19 shows the Gd electron conﬁguration for the ground state neutral
gaseous atom. It shows that 4f and 5d electrons have the highest energy states (65 has a
lower energy state). In the metal, Gd has the nominal valence occupancy of 6s2 5d1
hence it tends to be trivalent in compounds such as Gd oxide. Nelson and coworkers
state that the Gd in GdCOB has a 4f peak maximum located at 9.2 eV below the valence
band edge whereas Gd 5d electrons contribute at the crystal valence band edge [57].
Photoemission measurements of Gd metal, Gd on silicon, and Gd3Si5 also show strongly
localized 4f peaks about the same binding energy, whereas all states with about 5 eV of
the Fermi level are either 5d or 63 derived (with some rehybridization with Si valence
states[5 8]. Inverse photoemission measurements show that unoccupied 4f states lie
roughly between 4 and 5 eV above the Fermi level [59].

Since Gd in nontrivalent states is not chemically stable [25], comparing these ﬁne
structures combined with the spectra of the other lanthanides in literature with various
valences can give some indication of the Gd valence states. For instance, Thole and
coworkers [60] have observed a chemical shiﬁ of approximately 3eV between divalent
Eu and trivalent Eu. Similarly, an approximately 4eV in chemical shiﬁ has been
observed in Tb3+ and Tb4+ [60, 61]. Therefore, the similarity of the M edge in the
metallic Gd with the Gd oxide found in the present result conﬁrms the assumption that
the lanthanides are trivalent in the metallic solid state [62]. The similarity of the bulk
silicides and the Gd oxide spectra is expected since Gd is also trivalent in the silicide

form [63].

77

The slight difference of intensity ratio between M5 and M4 in the GdSi;
nanostructures with respect to the silicide is not readily explainable. Effects such as a
surface shift in the unoccupied 4f states, which is seen in the metal and might be present
in the silicide, might affect peak position but not branching ratio [59]. Any selection rule
that might be present between 3d state and the multiplet structure of the empty 4f states
would not easily translate into a branching ratio that varies with the chemical

environment of the Gd atoms.

 

GdN

     
   

(a) Metallic Gd

     
 
 
  
   

3; (b) Bulk GdSi2
3° .

a \ g) GdSi2

3.? /Si L anostructure
E

0

E (d) Gd203 powder

 

(e) Gdzos [27]

a

 

 

 

 

 

1 L 1 FL I 1 i I I :41 r l r l n l r
100 110 120 130 140 150 160 170 180 190 200

Energy loss (eV)

 

Figure 4.17 The energy loss near-edge structure (ELNES) of the Gd N edge from
metallic Gd, bulk GdSiz, and Gd203 powder in comparison to the Gd203 spectrum
from the standard EELS Atlas [27].

78

 

  
    
    
 
     
  

 

(a) Metallic Gd

(b) Bulk GdSi2 j "-

(c) GdSi2
nanostructure

Intensity (arb. units)

(d) GdZOJ
powder

 
 

(e) Gdzo3
[EELS Atlas]

 

 

l l L I 1 l n l I 1 I a l

1150 1160 1170 1180 1190 1200 1210 1220 1230 1240
Energy loss (eV)

 

Figure 4.18 The energy loss near-edge structure (ELNES) of the Gd M4; edge ﬁom
metallic Gd, bulk GdSiz, and Gd203 powder in comparison to the Gd203 spectrum
from the standard EELS Atlas [27]. The line with open squares demonstrate good
curve ﬁtting using a power—law background, Hartree-Slater cross section and a set
of Lorentzian proﬁles.

79

 

 

4f D ID 1D ID ID ID 1E Dll H—lﬂﬂ 5d

IEE 5
”mam“ "

 

[EII 4P

3d tiﬂlEEEE 4s
3p IEI

3S

2pIIlE

25 E
Is.

55

'65

Figure 4.19 Gd electron conﬁgurations for the ground state neutral gaseous atom.

Table 4.3 Summary of N and M peak positions and intensity ratio measurements of
the white lines ﬁom the metallic Gd, the bulk GdSiz, the GdSi; nanostructure (NS),

and Gd203 powder in comparison to Gd203 ﬁom EELS Atlas [27].

 

 

 

 

 

 

 

 

 

 

 

 

 

. . Branchin
Peak posrtron (eV) ratio g
N M5 M4 Ms-M4 Ms/M4
Metallic Gd 149.3 1186.4 1214.6 28.2 2.02i0.06
Bulk GdSi2 148.9 1186.1 1214.6 28.5 1.97i0.02
GdSi; NS 148.7 1186.0 1215.5 29.8 1.74'_1'0.02
Gd203 powder 149 1186.2 1215.5 29 2.03i0.02
Gd203 149 1188 [27] 1217 [27] 29 1.73 [26]

 

4.4 Conclusions

HRTEM was used to characterize epitaxial silicide nanostructures of two different

RE metals: Gd and Tm. In the case of Gd, the silicide nanostructures are either the

hexagonal or the orthorhombic phase, consistent with prior studies. However, some bi-

 

phasic silicide structures are seen, with layers of the orthorhombic phase overlying the
hexagonal. In the case of Tm, the nanostructures are either hexagonal or orthorhombic.
Once again, some nanostructures are seen with one silicide phase overlying the other.
For Tm, the orthorhombic phase lies at the interface between the hexagonal and the
orthorhombic. For both systems, bi-phasic silicide structures may reﬂect a mechanism
for strain accommodation at the interface with the substrate. In the case of Gd, the phase
with lower strain lies at the substrate. For the case of Tm, the relative mismatches of the
two phases predicted from bulk silicide lattice parameters disagree with that derived from
measured lattice constants, and it is a relaxed orthorhombic phase at the interface that
appears to have the lowest mismatch with the substrate. This could also reﬂect the fact
that this interfacial phase is only two or three atomic layers of Tm silicide thick.

EELS studies were carried out to compare the electronic structures of metallic Gd,
thin ﬁlm Gd silicide, and Gd oxide in bulk phases and Gd silicide nanostructures. The
results ﬁ'om the three bulk phases are similar, while the intensity ratio of M5:M4 in the
GdSi2 nanostructures varies from the bulk, which may suggest that a slightly different

spin state exists in the silicide nanostructures.

81

Chapter 5 Results from Multilayer Structure Studies

This chapter presents the HRTEM and EELS results from the various metallic
multilayers examined in this study. The ﬁrst section will cover the Co/Al and Py/Al
multilayers, followed by the results from the Cu/Al multilayers and Cu/Al in the Py-

based spin-valve structures.

5.1 Co(Py)/Al Multilayers

The Co(Py)/Al multilayers were composed of ten Co(or Py)/Al bilayers with a 20
nm Co (or Py) layer and a 10 nm Al layer sputtered on Si(lOO) substrates at room
temperature. To compare with the as-sputtered samples, some multilayers were annealed
for less than 5 min after sputtering at 180°C in the same chamber. The results ﬁom

Co/Al and Py/Al multilayers are presented separately.

5.1.1 As-sputtered and Annealed Co/Al Multilayers

A phase contrast cross-sectional TEM image of the as-sputtered Co/Al
multilayers on Si is shown in ﬁgure 5.1. The well-layered Co and A1 are both
polycrystalline. The thicknesses of the individual layers are consistent with the nominal
20 nm Co and 10 nm A1, with in-plane grains sizes in the range of ~20 nm. Signiﬁcant
layer roughness, in the form of non-planar interface perturbations, is observed, resulting
in the interfacial planes lying as much as 15° from the growth plane. A selected area
diffraction (SAD) pattern ﬁ'om both the multilayers and Si substrates (see inset) shows
strong diffraction from the close-packed planes of the Co and Al in the growth direction,

which indicate the layers are highly textured.

82

7 _ Si(OOZ)
‘0 O
‘Q‘Amm
Cotilll)

 

Figure 5.1 Phase contrast TEM image of as-sputtered Co/Al multilayers on a Si
substrate. Inset is a selected area diffraction pattern from both the multilayers and Si
substrates showing strong difﬁaction ﬁom the close-packed growth planes of both
the Co and Al, parallel with Si(001). In the upper right, an inclined grain is
illustrated schematically to show that the waviness perpendicular to the electron
beam is up to 2 nm at the interfaces.

It is well known that Co can exist as either FCC or HCP [19]. In many cases [64]
these two phases effectively co-exist as heavily stacking faulted permutations of either

perfect structure, which results from the low stacking fault energy that in turn results

83

ﬁom the very small difference in formation energy of these allotropes [65]. Stacking
faults can inﬂuence the magnetic behavior of Co, with higher fault densities decreasing
the MR [39]. Thus, it is important to characterize the crystal structures and faulting of
the Co in the multilayers in this study. Figure 5.2a shows a Co/Al/Co HRTEM image in
which the Co and Al layers are clearly evident. Two Al grains are clearly resolved in the
light contrast layer, revealing an ~7° angle between the two Al(111) growth planes.
Within the Co layers, the atomic structure is resolved, but somewhat obscured by moiré
fringes, which are caused by overlapping grains‘in the beam direction. The FFT of the
image (see inset) shows two strong Al diffraction spots ﬁom the {111} close-pack planes
of the two A1 grains, but a variety of FCC and HCP Co difﬁ‘action spots. Since the image
is not taken ﬁ'om a well-aligned Co crystal zone (as indicated by the FFT), the Co close-
packed plane stacking is not clearly deﬁned. Figure 5.2b shows another area from this
multilayer with both the Al and Co clearly resolved. Most of the Co in this image
displays a <110>FCC zone axis with ABCABC stacking of the close-packed {111} planes
(see upper inset). However, some HCP Co, with ABAB stacking (see lower inset), is
also observed. Analysis from different areas in the multilayers shows that FCC stacking
is more dominant than HCP stacking. An interesting feature of these multilayers is that
although both the Achc and Com: grow on their close-packed {111} planes, Al and Co
do not form coherent {111} interfaces. Instead, a signiﬁcant angular deviation of ~7°
between the {111}A1 and {111}co plane normals is found, as illustrated in the FFT inset in
ﬁgure 5.2b. Although this was most evident in grains imaged in the <110> orientation,

such deviations were consistently observed along the interfaces.

84

 

It is noted that the interfaces of Co/Al layers are obscure in ﬁgure 5.2, which
makes it difﬁcult to identify any inter-metallic phases that may have formed by diffusion
at the interfaces. A through-focus series of images in ﬁgure 5.3 can help illustrate the
interfacial structures along the electron beam. The four images were taken from the same
area, but at defocuses of 0, 32, 64, and 96 nm. For a defocus of 0, the Al lattice ﬁinges
extend through the interfacial region and to the Co at the upper interface. As the defocus
is changed, this distinct stacking of Al planes becomes obscured at the upper interface,
but extends closer to the lower interface, until it ﬁnally reaches the Co layer at a defocus
of 96 nm. HRTEM image simulations have shown that at different defocuses lattice
ﬁinges from a Ni3Al crystal will terminate at different positions relative to an inclined
grain boundary [66, 67]. Thus, this through-focus series suggests that what appears to be
an interfacial region between the Co and Al is actually the result of the interfaces being
inclined to the electron beam direction. Given the observed interface and layer roughness
perpendicular to the electron beam of approximately 2 am over a width of 20 nm
illustrated in ﬁgure 5.1, one would expect the interfaces to also exhibit some waviness in
the electron beam direction. With a TEM foil thickness of 10 nm and an observed
interfacial region thickness of 1 nm, the interfacial waviness in the electron beam
direction is consistent with the interfacial waviness perpendicular to the electron beam
direction. Figure 5.4 shows a schematic model of Al{111} growing on Co{111}. The
mismatch between the two lattices, calculated based on the lattice parameters of bulk
FCC Al and Co listed in Table 5.1, is 13.7% for A1{1 1 1} growing on Co{111}. Clearly,

it is unlikely that these layers will directly hetero epitax on each other [1 l].

85

,Co (

,‘ ll

 

Figure 5.2 HRTEM images showing Co/Al atomic lattices with different contrast. The
Co/Al interfaces are poorly deﬁned. A grain boundary in Al is observed in (a). Both
FCC and HCP stacking in the Co are shown in the enlarged images in (b). Insets in the
upper right show the corresponding FFTS from the images.

86

Defocus 0 CO Defocus 32

Defocus 64

 

Figure 5.3 A through-focus series of images from the same area of the multilayer indicates
that the Al and Co interfaces are inclined along the beam direction.

Instead, it appears the mismatch is accommodated by signiﬁcant tilting between those
lattices, as shown in ﬁgure 5.4, which results in the inclined interfaces and apparent
roughness. The resulting interfaces would be expected to be complex, with atomic ledges

and dislocations with Burger vector both in and normal to the interface.

87

 

Figure 5.4 Schematic growth model of FCC A1 {111} on FCC Co {111} with a lattice
mismatch of 13.7%.

Table 5.1 Lattice parameters and close-packed interplaner spacing variants for the

 

 

 

 

 

multilayers.
Materials Lattice parameters Interplaner spacings Interplaner spacings for
(nm) measured for (111) (111) in theory
Al ao= 0.405 0.235i0.002 mm 0.234 nm
ao=0.251,
Co co=0.407 (HCP) N/A N/A
ao=0.356 (FCC) 0.205i0.002 nm 0.206 nm
Py (NiFe) ao= 0.355 0.204i0.002 mm 0.205 nm

 

 

 

 

 

To study the effect of annealing on the Co/Al multilayers, both TEM and spatially

*dSiooz= 0.272nm is used as reference

resolved EELS were carried out on both the as-sputtered and annealed samples to study

the structures and chemical nature of the Co/Al interfacial region. Figure 5.5 shows a

phase contrast TEM image of an annealed Co/Al multilayer. The morphology and

88

 

corresponding SAD pattern demonstrate that the layers maintain well-deﬁned Co and Al

layers and no obvious intermediate phases.

0
Si (002')

%‘AHHH
XI‘OUH)

 

Figure 5.5 Phase contrast TEM image of an annealed Co/Al multilayer. Inset is a
selected area diffraction pattern from both the multilayers and Si substrates

showing strong diffraction from the close-packed growth planes of both the Co
and A1.

Figure 5.6a shows a high angle annular dark ﬁeld (HAADF) STEM image of the
as-sputtered Co/Al multilayers. The image contrast is proportional to Z2, so the Co

(Z=27) layers are much brighter than the Al (Z=13) layers. As a result, the analysis of

89

the image contrast can give some indication of the compositional information at the
interfacial region by comparing the as-sputtered and annealed samples. A comparison of
line proﬁles of image grey level across a Co/Al/Co layers ﬁ'om both the as-sputtered and
annealed multilayers is shown in ﬁgure 5.6b. Although some differences in image grey
levels in the Co and A1 layers are shown, there are no signiﬁcant differences in terms of
contrast change in interfacial regions. A series of EELS spectra were collected by line
scanning the electron probe across layers. Figure 5.6c shows a perspective view of an
EELS spectrum-image with 61 regularly spaced low loss spectra obtained along the 49
nm line indicated in ﬁgure 5.63. The square frame labeled “Spatial Drift” indicates the

area used as the reference to correct the specimen drift during the spectra collection.

90

Spectrum Image

 

(b) 250 _ annoaied

200 ~

150 >

 

 

 

 

7a
5
3 100 . , as-sputtered
Spatial (3
Drift Co AI
50 » c°
t
o I l ‘ l I l
0 5 10 15 20 25 30 35 40
A1 plasmon Distance (nm)

  
   

Energy loss (e '

Figure 5.6 (a) A dark ﬁeld STEM image of the as-sputtered Co/Al multilayers
showing where the EELS spectrum image was acquired along the white dashed line.
The spatial drift square is the cross-correlation scan area used to correct the drift. (b)
A comparison of line proﬁles of image grey level across a Co/Al/Co layers from both
as-sputtered and annealed multilayers. (c) A perspective view of an EELS spectrum
image showing the plasmon intensity changing across the Co/Al multilayers.

It is known that Al has a very intense plasmon peak compared to many metals,
including Co, Fe, and Ni [27]. Thus, the low loss spectrum of plasmon intensities can
give an indication of the elemental distribution across the spectrum image. Figure 5.7

shows a comparison of the plasmon intensity proﬁles for the as-sputtered and annealed

91

Co/Al multilayers. Both of the plasmon peak proﬁles across the Al-rich layers are 5-6

am in full width at half maximum (FWHM), which indicate similar Al distribution in the
as-sputtered and annealed samples. Within the Co-rich layers, relatively constant regions,
approximately 15 nm wide for the as-sputtered multilayer and 17 nm wide for the
annealed multilayer, suggests there are no signiﬁcant differences in the Co distribution in

the two conditions.

 

 

 

Normalized Intensity

 

 

 

 

 

 

 

 

 

 

 

 

 

 

1.0- 1.0-
0.8 E’ 0.8 -
g .
0.6 g 0.6-
0.4 '7. 0.4-
F.
O
0.2 2 0.2
0.0...1 ................... 0.01:. .I.L.I.LII.I.I. .
04812182024283236404448 048121620242832364044

Figure 5.7 A comparison of the plasmon intensity proﬁle for as-sputtered and annealed
Co/Al multilayers. The Al plasmon peak proﬁles have full width at half maximum of
5-6 nm, similarly for both as-received and annealed samples. The only difference is
that the low variation of Co intensity length is 15 nm for the as-received multilayers
and 17 nm for annealed one.

An alternative way to characterize interfacial compositions using EELS is to
investigate the Co L23 edge intensity proﬁle across Co/Al interface, which can provide
sufﬁcient insight in the chemical nature at the Co/Al interfacial region. Figure 5.8 shows
a perspective view of the EELS spectrum image recorded around the Co L23 edge
(L3z779eV, L22794) across the Co/Al/Co layers of the as—sputtered sample. Upon

removing a power law background for each spectrum, the intensities of L3 peaks were

92

analyzed and plotted in ﬁgure 5.9 (shown as open circles). The intensity proﬁle is
compared to a similarly processed intensity proﬁle for the annealed layers (solid squares
in ﬁgure 5.9). To facilitate an understanding of the two experimental proﬁles, a
theoretical Co intensity proﬁle across perfect Co/Al/Co interfaces (see in ﬁgure 5.9) has
been simulated using an electron beam with a 0.5 nm (FWHM) Gaussian proﬁle. The
experimental data reveals signiﬁcantly broader interfacial mixing than the theoretical
proﬁle. While some of this broadening will be associated with the inclination of the
interfaces described earlier, and possibly from beam spreading in the TEM thin foil (both
ignored in the simulation),the deviation from the theoretical proﬁle may suggest some
mixing across the interfaces. Nevertheless, the intermixing is not sufﬁcient to initiate

additional phase formation.

93

.th

\tff'
tttw
42111:“

\p‘llq‘l ['9 N
it] “ii?!“ 1.1, ,,
l ltidy“ ""1'1‘”'stii'a'it'ktiii ' a»

ti ‘52:le {15%

1M") I' , .
“‘1‘ ll” W \.' 13.11:} w! ( “\ ”all!“
ﬁll}; \ ll ‘ “51‘1” ‘5'";
\tttt ‘~'~'

.-.*~I t‘l‘tl
. \l.‘

.u,’ (,‘v

immmmw

all

Probe position (nm)

 

Figure 5.8 A perspective view of the Co white line spectrum image scanned across
Co/Al/Co layers over a length of 20 nm of the as-sputtered multilayers.

Normalized EELS Intensity

 

Comparison of Co Intensity Proﬁles

I r I ' I ' l ' l ' I

  

... Annealed data

 

 

— Simulation curve

C39 0 00 O As—sputtered data

 
 
   

l I

l

 

 

0 2 4 6 8 10 12 l4 16

Distance lnml

18

20

Figure 5 .9 Co white line intensity proﬁles for both the as-sputtered and annealed
multilayers. A theoretical proﬁle simulated by using a 0.5nm FWHM Gaussian

electron beam proﬁle across perfect Co/Al interfaces is used to compare with the
experimental data.

94

5.2 As-sputtered and Annealed Py/Al Multilayers

This section presents HRTEM and EELS results ﬁom the Py/Al multilayers
sputtered on Si(lOO) substrates. The multilayer was composed of ten bilayers with a 20
nm Py layer and a 10 mm Al layer Phase contrast TEM images of both as-sputtered and
annealed Py/Al multilayers grown on Si are shown in ﬁgure 5.10. Py (permalloy = Nil-
xFex with x~0.2) has an L12 fcc crystal structure. The images reveal the layers are
polycrystalline, while the strong diffraction peaks for the close-packed Py and A1 {111}
planes in the growth direction in the SAD patterns (see the insets in ﬁgure 5 .10)
demonstrate that Py and A1 are highly textured. The thicknesses of the individual layers
are consistent with the nominal 20 nm Py and 10 nm Al. The Py had grains with sizes in
the range of ~20 nm, and similar to the Co/Al multilayers, some of the interface normals
are tilted. The observed structures and interface roughness do not reveal any obvious
differences between the as as-sputtered and annealed samples.

Figure 5.1] shows an HRTEM image of an annealed Py/Al multilayer with most
of the Py and some of the Al clearly resolved. ABCABC stacking of the close-packed
{111}pcc planes are displayed in both Py and Al. The FFT (see lower inset) shows that
the grains are imaged along <110>pcc zones of Py and A]. Again, a signiﬁcant angular
deviation of 7° between the {1 11}A1 and {11 1}py plane normals is found, which will lead
to complex interfaces. Similar to the observation in the Co/Al multilayers, the lattice
fringes are obscured near the interfaces of the Py/Al layers in the image, which can be

interpreted as resulting of inclined interfaces.

95

Py 002 : -
Atooz’” , g
. “ P), 1 11
o “»
v. Ahl-H

s: 002

 

 

Figure 5.10 Phase contrast TEM image of as-sputtered and annealed Py/Al
multilayers. Insets are selected area diffraction patterns from the multilayers. Both
morphologies and diffraction patterns show similar structures in two conditions.

96

 

Figure 5.11 HRTEM image of an annealed Py/Al multilayer along a Py <110> zone
with corresponding FFT in the inset showing the growth relationship.

It is known that both Fe and Ni can form a range of intermediate phases with Al,
including several ternary intermediate phases [1?]. To study if there is any intermixing at
the Py/Al interfacial region in the as-sputtered and annealed multilayers, spatially
resolved EELS was carried out to study the composition across the interfaces. EELS
spectrum images that included the low energy plasmon peaks and Ni and Fe L edges (Ni
L328556V, Fe L3:708eV) have been collected across the interfaces for both the as-

sputtered and annealed samples. Figure 5.12 shows a comparison of the normalized

97

Normalized Intensity

plasmon intensity proﬁles of across the Py/Al/Py/Al/Py layers, which reveal layer
thicknesses consistent with the nominal thicknesses. Figures 5.13 a and b show the
comparisons of intensity proﬁles of Ni and Fe at the interfaces of the as-sputtered and
annealed Py/Al/Py layers with the simulated theoretical proﬁles. Figure 5.13c shows the
comparison of line proﬁles of the image grey level from across Py/Al/Py layers in the Z-
contrast images of the two conditions. No signiﬁcant differences were observed between
the annealed and as-sputtered Py/AJ/Py proﬁles, which suggests the annealing treatments
do not signiﬁcantly alter the multilayer structure. Nevertheless, the intensity proﬁles do

suggest that some limited compositional mixing may occur during the initial growth.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

As—sputtered Plasmon Loss Intensity Across Interfaces Annealed Plasmon Loss Intensity Across Interfaces
l-o _ I I I I I l I I I I - I I I I I I ' I I I I I
.- I‘ 1.0 _ II .-
I I I I I
- I I -
0-3 Py Al Py LA. I Pr =:‘ o s Py .Al. Py IAI- 1’! -
----- I---u-----------: -- -- 5 '"'i"'"""""""lL'ﬂ'"
0 6 > I a . _
I 7nm l 23m l 7mm .— 0.6 _ . 6.5 nir- 22 5 um I 6.5m!-
“ y I E I .
o 4 '- T- o 4 # uh. I _
2g 17 ‘
o 2 ' 16nm I ‘ 03- ' l ——-|nm '
0.0 0.0

 

 

I | I l l I I l l I I l I I I I I I I I l I
0 4 8 l2 l6 20 24 28 32 36 40 44 48 52 0 4 8 12 16 20 74 28 32 36 40 44 48 52
Distance (nm) Distance (nm)

Figure 5.12 A comparison of plasmon intensity proﬁle from the as-sputtered and
annealed Py/Al multilayers. The A1 plasmon peak proﬁles have full width at half
maximum of 6.5-7 nm, similarly for both as-sputtered and annealed samples.

98

(3) Comparison of Ni Interface Proﬁles

 

 

 

 

 

I 2 I I I T I I I
E 1.0 —
a
E
.5 0.8
m
:3
H 0.6
E
= 0.4
E
g 0.2
0.0 I l l J n 1 a I;
0 4 8 I2 16 20 24 28
Distance (nm)
(b) Comparison of Fe Interface Proﬁles

 

I r I r

    

Normalized EELS Intensity

 

 

 

Distance (nm)

 

 

 

(e)
250 -
as- \
sputtered
200 —
T; 150 —
2
g annealed
(J 100 r-
Py AI Py
50 e
.
0 a l L J A a 4.. 1 L4L L I

 

0 4 8 l2 16 20 24 28
Distance (nm)

Figure 5.13 Ni (3) and Fe (b) white line intensity proﬁles for as-sputtered and
annealed multilayers are compared separately with the theoretical proﬁle simulated
by using a 0.5nm FWHM Gaussian electron beam proﬁle across perfect Py/Al
interfaces. (c) A comparison of line proﬁles of the image grey level from Py/Al/Py
layers in the Z-contrast images of the two conditions.

99

5.2 Cu/Al Multilayer and Cu/Al Spin-Valve Structures

To study the effects of Al interlayers on Py/Cu/Py based spin-valves, structures of
alternating Cu/Al multilayers have been investigated by HRTEM. Since increasing the
thickness of the Al layer in the spin-valves can cause a decrease in the resistance [48],
different thickness ratios in Cu/Al multilayers were synthesized to see if the structures
vary with different conditions. Multilayers made up of Cu(8nm)/Al(lOnm) with an
overall composition of 52.9 at.% Cu and Cu(5nm)/Al(3nm) with an overall composition
of 70.1% at. Cu were selected for HRTEM studies. These multilayers were examined in
a number of post sputtering states, speciﬁcally as-sputtered and annealed, to investigate
thermal instability of these multilayers. TEM samples in both states were prepared using
speciﬁc approaches to minimize sample heating (described in 3.4). Results from
HRTEM imaging with associated F FT analysis show that different intermediate phases
form in the Cu/Al multilayers with different thickness ratios.

In order to determine if the intermediate phase formation observed in multilayers
also occurs in a single Al layer bounded by Cu in Py based spin-valves, selected spin-
valves were characterized. The spin-valves have the following conﬁguration, where all
the values are given in nanometers:
Nb(l50)Cu(5)FeMn(8)Py(24)Cu(10)Al(x)Cu(10)Py(24)Cu(5)Nb(50). In this study, spin-
valves with Al(x=10) for as-sputtered and Al(x=30) for annealed were selected. The
spin-valves were studied using HRTEM and EDX line scans to characterize the crystal

structures and chemical nature.

100

5.2.1 Cu/Al Multilayers
5.2.1.1 Cu(8nm)/Al(10nm) Multilayer Structures

A phase contrast cross-sectional TEM image of the as-sputtered
Cu(8nm)/Al(10nm) multilayer is shown in ﬁgure 5.14. Well-deﬁned alternating layers of
different contrast are observed. The layers are polycrystalline and some moiré ﬁinges are
observed. The Cu-rich layers exhibit dark contrast and are ~12 nm thick, while the Al-
rich layers exhibit bright contrast and are ~6 nm thick. By comparing these measured
layer thicknesses to the nominal, it appears that more Al has diffused into the Cu rather
than the opposite. In addition, this change in layer thickness ratio suggests that

intermediate phase formation may have occurred.

Cu rich layer
Al rich layer

Growth direction

20 nm

 

Figure 5.14 Phase contrast TEM image of as-sputtered Cu(8nm)/Al(10nm)
multilayers. Observed layers are in different contrast, exhibiting ~12nm thick Cu-
rich layer and ~6nm thick Al-rich layer, which are different from the nominal
thickness.

HRTEM imaging with associated FFT analysis were can'ied out to characterize

the structures of the intermixed Cu and Al. Figure 5.15 shows an HRTEM image of the

101

same Cu/Al multilayer. Five layers are shown in the image: Al-rich layers labeled 1, 3,
and 5; Cu-rich layers labeled 2 and 4. This image show different fringe contrast and
symmetries in the different Al-rich layers. The Al-rich layers are identiﬁed to be
tetragonal AlzCu (lattice parameters a= =0.604nm, c=0.486nm). The FFTS (see insets)
from the Al-rich layers 1 and 5 show the same A12Cu <001> zone pattern, while the FFT
from Al-rich layer 3 shows an A12Cu <11 l> zone pattern. All the lattice spacings
measeasured are summarized in table 5.2. The FFTS from the Cu-rich layers (2 and 4)
show three-fold symmetry, which suggests the phase has a cubic structure. However, the
lattice spacings are different (by ~60%) than that expected for fee Cu. Instead, the F FT
pattern is consistent with the <111> zone axis for the bee [3 solid solution of AICU3. The
[3 phase is known to exist at high temperature in the Cu-Al system, and the [3 phase can be
retained as a metastable phase due to the sluggishness of the eutectoid reaction B—+ Cu +
71 [17]. Both the A12Cu and AlCu3 are orientated with their close-packed planes of (110)
and (110), respectively, normal to the growth direction. Based on the FFT analysis, two
types of growth relationships are found: A12Cu(110) // AlCu3(110), A12Cu[001] //

AlCu3[lll], and A12Cu(110)//A1Cu3(110), A12Cu[lll] //A1Cu3[111].

102

0 AIZCir <111> " B AlCu,<lll>

 

Figure 5.15 HRTEM image shows three types of stacking fringes with different contrast
existing in the Cu/Al multilayers. Five different areas in the image are analyzed using
FFTS, resulting in three crystal zone patterns shown in the insets. Areas 1 and 5 show
the same 9 A12Cu <001> zone patterns, areas 2 and 4 show the same B AlCu3 <111>
zone patterns, and area 3 shows a 9 A12Cu <001> zone pattern.

103

Table 5.2 Summary of the lattice spacings measured from all the phases in the Cu/Al
multilayers.

 

 

 

 

 

 

 

 

Theoretical Measured
Layer Plane . .
spacrng (nm) spacmg (nm)
(110) 0.427 0.42
A12Cu (011) 0.378 0.38
(020) 0.302 0.30
(112) 0.211 0.21
AlCu3(bcc) (1 10) 0.205 0.20
1 1 1 0.209 0.21
Cu(fcc) ( )
(002) 0.181 0.18

 

 

 

 

 

 

Figure 5.16 shows a schematic representation of the two phases and their growth
orientation relationships. Both bcc AICU3 (ﬁgure 5.16a) and tetragonal AlZCu (ﬁgure
5.16b) will grow on their close-packed {110} planes, but these planes can be aligned in
two different directions. The dashed rectangle in ﬁgure 5.16c shows a structural unit in
the AlCu3(110) plane that can line up in two different ways on the A12Cu, as shown in
ﬁgure 5.16d. When AICU3[111] // A12Cu[001] (the ﬁne dotted lines on ﬁgure 5.16d),
there is a very small mismatch (0.5%) in one direction and a large mismatch (-16.6%) in
the perpendicular direction. When AlCu3[l l 1] // A12Cu[l 1 1] (the coarse dotted lines on
ﬁgure 5.16d), there is a relatively small mismatch (2.4%) in one direction and a even
larger mismatch (-1 7.6%) in the perpendicular direction. Consequently, there are two
rotation alternatives for AlCu3 growing on AlzCu. This can result in a variety of grain
orientations throughout a columnar grain.

To study the thermal stability of these Cu/Al multilayers, some of the

Cu(8nm)/Al(10nm) multilayers were annealed at 180°C for 5 min. Figure 5.17 shows a

104

phase contrast TEM image of the annealed Cu(8nm)/Al(10nm) multilayers. The Cu-rich
layers exhibit dark contrast and are ~13 nm thick, while the Al-rich layers exhibit bright
contrast and are ~7 nm thick. The layer thicknesses are different with the nominal, but
similar with those of the as-sputtered multilayers. Again, the Cu-rich layers are thicker
than the nominal and the Al-rich layers are thinner. This suggests more diffusion of the
Al into the Cu than the Cu in the Al, and possible intermediate phase formation in the
multilayers. An I-[RTEM image from these annealed Cu/Al multilayers is shown in
ﬁgure 5.18. The FFT analysis again shows that the Cu—rich phase is AlCu3 and the Al-
rich phase is A12Cu. All of the measured lattice spacings are summarized in table 5.2. In
conclusion, similar structures with the same intermediate phases are observed in the as-
sputtered and annealed Cu(8nm)/Al(10nm) multilayers. The observed difference is that
the interfaces are wavier in the annealed multilayer compared to those in the as—sputtered

multilayer.

105

   

(a) AlCu3 crystal structure
(b) AlzCu crystal structure

0.243nm (-l 7.6%)
*_ __

      

vow
9V
0.427nm (2.4%)

   

ﬂ
AlCu3 [111]

(c) AlCu3 (110)

(d) A12Cu (110)

Figure 5.16 Schematic representation of the bee AICU3 and tetragonal A12Cu crystal
structures and the orientation relationships of their close-packed planes. (a) bcc AICU3
crystal structure (b) tetragonal A12Cu crystal structure (c) a structural rectangle unit in
AlCu3(110) (d) two orientation relationships with AlzCu(110): AlCu3[11 1] //
A12Cu[001] (the ﬁne dotted lines) and AlCu3[l l 1] // A12Cu[111] (the coarse dotted
lines).

106

« «~er %“W «We.
Aﬁcmhld30£am® . Wﬁmnm

Growth direction

 

Figure 5.17 Phase contrast TEM image of annealed Cu(8nm)/Al(10nm) multilayers.
The Cu-rich layers are ~13 nm thick and the Al-rich layers are ~7 nm thick.

AIZ'Cu(-l10)

Cu—rich

.AlClrj(-l,l(|)

 

Figure 5.18 HRTEM image with corresponding FFTS from the annealed
Cu(8nm)/Al(10nm) multilayers. The Al-rich phase and the Cu-rich phase are identiﬁed
to be A12Cu and AlCu3 respectively.

107

5.2.1.2 Cu(5nm)/Al(3nm) Multilayer Structures

Figure 5.19 shows a phase contrast TEM image of the as-sputtered
Cu(5nm)/Al(3nm) multilayers. The Cu-rich layers are 34 nm thick, while the Al-rich
layers are 4-5 nm thick. The layer thicknesses are different with the nominal, which
again suggests that intermixing, and possibly intermediate phase formation, have
occurred. An HRTEM image from these Cu/Al multilayers is shown in ﬁgure 5.20.
FFTS (insets) from 4 areas were used to identify the phases. The FFTS from the Al-rich
layers 1 and 3 show the same AlzCu <111> zone pattern, while F FTs from Al-rich layer 2
and 4 show the same FCC Cu <110> zone pattern. All the lattice spacings measured are
surmnarized in table 5.2. The growth orientation relationship is identiﬁed to be
AlzCu(110) // Cu(l 1 1), AlzCu[1 11] // Cu[1 10]. Compared with the Cu(8nm)/Al(10nm)
multilayers, Cu did not form a bee solid solution with Al in the Cu-rich layer in the
Cu(5nm)/Al(3nm) multilayers. This may result from the difference in the overall
stoichiometry of the Cu and Al, but the reasons are not entirely clear.

Figure 5.21 shows a schematic representation of the close packed planes of the
two phases, fcc Cu (111) and tetragonal A12Cu (110), and the way they epitax with each
other. There is an equilateral triangular unit existing in the Cu (111) planes and a
similarly orientated isosceles triangular unit existing in the AlgCu (110) planes.
Matching the two units results in a -4% mismatch in two directions and a -5% mismatch

in the third direction. When Cu is epitaxed on to the AlzCu, there is effectively only one

108

 

Figure 5.19 Phase contrast TEM image of the as—sputtered Cu(5nm)/Al(3nm) multilayers.
The Cu-rich layers are 3-4nm thick and the Al-rich layers are 4-5 mm thick.

 

..., .
'.';¢ N

-‘-’II!;:;;.¢I

. t
. .,.-otvaaov

....~aetv-:¢-

 

 

Figure 5.20 HRTEM image shows two types of lattice fringes existing in the
Cu(5nm)/Al(3nm) multilayers. Four different areas in the image are analyzed using
FFTS, resulting two crystal zone patterns shown in the insets. Areas 1 and 3 show the
same 0 A12Cu <111> zone patterns, while areas 2 and 4 show the same Cu <110> zone
patterns.

109

orientation that the Cu can assume. However, the AlzCu can align in three different ways
on the Cu as illustrated in ﬁgure 5.21. This may result in signiﬁcant layer to layer

orientation variation in a columnar multilayer grain.

,0.
t7".

—» I
c [110)
“ .----QW,

FCCCu(111)

 

A12Cu (110)

Figure 5.21 Schematic representation of FCC Cu (111) and tetragonal Alzcu (110)
growth orientation and lattice mismatches.

5.2.2 Cu/Al in Spin-Valve Structures
5.2.2.1 As-Sputtered Spin-Valve with 10nm Al

Two different Cu/Al spin-valve structures were selected for study. They have the
following conﬁgurations, where all the values are given in nanometers:
Nb(150)C\1(5)FeMn(8)Py(24)Cu(l0)Al(x)Cu(10)Py(24)Cu(5)Nb(50), In this study, a
sample with Al(x=10) was examined in the as-sputtered state and a second sample with
Al(x=30) was examined in the annealed state. The spin-valves were characterized using

HRTEM and EDX line scans to determine their crystal structures and chemical nature.

110

While this study focused on the Cu/Al layer structures, other layers in the spin valves
were imaged and used as references (i.e. Py atomic structures were used as reference to
measure the lattice spacings).

Figure 5.22 shows a TEM bright ﬁeld image of all the layers in the as-sputtered
spin-valve with the nominal conﬁguration with Al (x=10nm). The thickness of some of
the individual layers is not clearly resolved in this bright ﬁeld image. In fact, the
Cu/Al/Cu portion of the spin valve appears as one layer. Nevertheless, the image does
show that the spin valve is layered with polycrystalline grains. Twins and stacking faults
are evident in some of the grains. Figure 5.23 shows a phase contrast TEM image of the
PyCuAl region with only the Al layer in clear contrast. While it is difficult to measure
the thickness of the Cu and Py layers due to the similar Z, some region shows the ~10 nm
thick Cu layer, marked by the arrows in the image. The observed thickness of the
resolved layers is consistent with the nominal.

To study the crystal structures of the Cu/Al layers, Py and Nb, which have known
structures, were ﬁrst studied to establish references. Figure 5.24 shows an HRTEM
image (a) of Py/FeMn/Cu/Nb layers in the as-sputtered spin-valve and the corresponding
FFT (b) from the image. Moiré fringes, caused by overlapping grains at the interfacial
region, make it difﬁcult to resolve the FeMn and Cu structures. However, the Py and Nb
layers show clear atomic lattices and the FFT (b) shows strong diffraction from the
Py(l 11) and Nb(110), which were used to as reference to measure the lattice spacings

(listed in table 5.3).

111

§_ \-

I. Cu/Alglrguexs. ’

Py

Cu, FeMn\

50mm
-—"' Si substrate

 

Figure 5.22 TEM bright ﬁeld image of an as-sputtered spin-valve with nominal

thickness in nanometers
Nb(l 50)Cu(5)FeMn(8)Py(24)Cu(l0)Al(10)Cu(l0)Py(24)Cu(5)Nb(50).

112

 

Figure 5.23 Phase contrast TEM image of Py/Cu/Al/Cu/Py layers in the as-sputtered
spin-valve with x=10nm.

113

FeMn/Cu ‘4

..-
."t"

(h)

P}(lll)

X|l(llll)

 

Figure 5.24 (a) HRTEM image of Py/FeMn/Cu/Nb layers in the as-sputtered spin-valve
with x=10nm and (b) the corresponding FFT from the image.

114

Table 5.3 Summary of the lattice spacings measured from all the phases in the
as-sputtered spin-valve.

 

 

 

 

 

 

 

Layer Plane Theoretical Measured
structure spacmg (nm) spacmg (nm)
Py (l l 1) 0.205 As reference
Nb (110) 0.233 0.23
Cu(fcc) (l 1 1) 0.205 0.2
AlzCu (220) 0.214 O. 22
A1 (111) 0.234 N/A

 

 

 

 

 

Figure 5.25 shows an HRTEM image with the Cu and Al region exhibiting
different contrast. The corresponding FFT shows two diffraction peaks along the grth
direction, very close to each other, and diffraction at half the distance in the reciprocal
space, which suggest a structure with a superlattice formed. This observation is
consistent with the double periodicity observed in the Al-rich region of the HRTEM
image. These observations suggest an intermediate phase has formed. Indeed,
measurements on the FFT shows that the closely spaced diffraction peaks are from
CU(111)f¢c and AlzCu(220), (see table 5 .1). This analysis is consistent with that carried
out in other Cu/Al regions of the spin valve. No fee Al was observed in the Cu/Al region
throughout the multilayer.

Complementary XEDS line scan were carried out to investigate the chemical
nature across the multilayer structure. Figure 5.26a shows a STEM bright ﬁeld image of
the as-sputtered spin-valve. XEDS intensity proﬁles of Cu, Ni, Fe, and Al were collected

from the line across the Py(24)/Cu(10)/Al(lO)/Cu(10)/Py(24)/FeMn(8)Cu(5)

115

1' "111.; a

ti-
,

A '
’2“. 3
Al melt

in
“lift“. ‘~

(b)

Al,Cu(220)

it

A

Cu(111)

 

Figure 5.25 (a) HRTEM image of Cu/Al/Cu layers in the as-sputtered spin-valve
with x=10nm and (b) corresponding FFT from the image.

116

 

 

   

o 50 100

nm

COppCfKﬂlJiukrrlbéiL ' ~ 1‘ AlumnmmKat

Figure 5.26 (a) STEM bright ﬁeld image of the multilayers in the as-sputtered spin-
valve. (b) XEDS intensity proﬁles of Cu, Ni, Fe, and A] were collected from the line
across the Py/Cu/Al/Cu/Py/FeMn/Cu layers in the image (a).

117

layers. The FWHM for adjacent Py, Cu and Al layers are 25, 12, and 14 nm respectively. I
The line proﬁles indicate that there may be some intermixing taking place between the
Cu and Al, while the other regions remain well segregated in their metallic structures (i.e.
the measured FWHM of 25nm for Ni and 6nm for Cu compare very well with their
nominal thickness).

Both HRTEM and XEDS analyses suggest that there is some intermixing in the

Cu/Al region in the as-sputtered sample with Al 10nm.

5.2.2.2 Annealed Spin-Valve with 30nm Al

Figure 5.27 shows a TEM bright ﬁeld image of all the metallic layers in the
annealed spin-valve with Al(x=30), i.e. with a nominal conﬁguration of
Nb(150)Cu(5)FeMn(8)Py(24)Cu(l0)Al(30)Cu(l0)Py(24)Cu(5)Nb(50). All the layer
structures are polycrystalline. The thickness of some of the individual layers is not
clearly resolved in this bright ﬁeld image. However, the Al layer shows clear bright
contrast compared to the adjacent Cu layers. Figure 5.28 shows a phase contrast TEM
image of the PyCuAl region that shows the layer more clearly resolved. The Py layers
display a thickness of approximately 24nm consistent with the nominal. However, the
Cu-rich layers, approximately 14nm thick, are signiﬁcantly thicker than the nominal,
while the Al-rich layer is much thinner, approximately 18nm thick, than the nominal
thickness. This again suggests signiﬁcant intermixing and possible intermediate Cu-Al
phase formation. Interestingly, the total Cu/Al/Cu thickness of ~46nrn is relatively close

to the nominal total. In addition, the interfaces between Py-rich and Cu-rich layers are

118

more planer than the ones between Al-rich and Cu-rich layers, which display

considerable waviness.

Cu, FeMn

Si Substrate

5011 m

 

Figure 5.27 TEM bright ﬁeld image of an annealed spin-valve with nominal thickness in
nanometers Nb(1 5 0)Cu(5)F eMn(8)Py(24)Cu(l 0)Al(30)Cu( l 0)Py(24)Cu(5)Nb(50).

119

 

Figure 5.28 Phase contrast TEM image of Py/Cu/Al/Cu/Py layers in the annealed spin-valve
with x=30nm.

Figure 5.29 shows an HRTEM image in the Al/Cu/Py region. Five areas with
clearly resolved lattice images were analyzed using FFT. Area 1 in the Al-rich layer
shows an AlzCu <100> zone pattern. The entire Cu-rich layer was identiﬁed as the [3’
ordered AlCu3 phase. FFTS from this Cu-rich layer show different zone patterns: area 2
shows an AlCu3 <111> zone pattern and area 3 shows an AlCU3 <001> zone pattern. The
observations of A12Cu and AlCu3 suggest signiﬁcant intermixing of Al and Cu have taken
place. The Py layers have been found to form metastable bcc structures in the spin-

valves. This is illustrated in the FFT analysis from area 4, which shows a Py <111> zone

120

pattern with three-fold symmetry (<111> zones in fee Py have lattice spacings that cannot
be resolved with the JEOL 2200FS). However, most of the Py in the spin-valves exhibit
fee structure. A twinned grain of fee Py is observed in area 5, with the corresponding
FFT shown at the bottom of ﬁgure 5.29. It is clear that all the phases have grown on their
close packed planes. The lattice spacings measured from different phases are
summarized in table 5.4. This table shows that these nano-scaled phases exhibit lattice

parameters consistent with the bulk phases.

121

2-20 ‘_ ' '1“
15‘ um, [our] I=('(' l’_\> <1 10>

 

Figure 5.29 HRTEM image of the Al/Cu/Py layers in the armealed spin-valve with
x=30nm. Five different areas in the image are analyzed using FFTS. Areas 1 shows a
0 AlzCu <100> zone pattern, area 2 shows a [3’ AlCu3 <111> zone pattern, area 3
shows a B’ AlCu3 <001> zone pattern, area 4 shows a BCC Py <111> zone pattern,
and area 5 shows a FCC Py <111> zone pattern.

122

Table 5.4 Summary of the lattice spacings measured from all the phases in ﬁgure 5.29.

 

 

 

 

 

 

 

 

 

 

 

 

 

Theoretical Measured
Layer Plane . .
spacrng (nm) spacrng (nm)
(111) 0.205 0.20
Py (fcc)
(200) 0.178 0.18
Py (bcc) (110) 0.205 0.20
(110) 0.205 0.20
AlCu3
(200) 0.145 0.14
(020) 0.302 0.30
AlzCu (002) 0.243 0.24
(022) 0.189 0.19

 

 

Complementary XEDS line scans were carried out to investigate the chemical
nature in the Al/Cu/Py region. Figure 5.30a shows a STEM bright ﬁeld image of the
Py/Cu/Al/Cu/Py layers in the annealed spin-valve. XEDS intensity proﬁles of Cu, Ni, Fe,
Al and 0 were collected ﬁom the line shown across the Py/Cu/Al/Cu/Py layers. The line
proﬁles indicate that the Al intensity proﬁle extends through the adjacent Cu layers while
Py (Ni and Fe) intensity proﬁles do not. This observation suggests that there is no
obvious intermixing between the Py (Ni and Fe) and Cu region, while signiﬁcant
intermixing taking place between the Cu and Al region. This chemical nature analysis is
consistent with the results from the HRTEM analysis.

Both HRTEM and XEDS analyses show that intermediate Cu-Al phases were

conclusively identiﬁed in the Al(30nm) annealed sample.

123

(a)

 

 

 

Copper Ka1 , Nickel K81 , Iron Ka1 , Oxygen K31

Figure 5.30 (a) STEM bright ﬁeld image of the multilayers in an annealed spin-
valve. (b) XEDS intensity proﬁles of Cu, Ni, F e, Al and O are collected from the
line across the Py/Cu/Al/Cu/Py layers in the image (a).

124

5.3 Discussion and Conclusions

Although the interfacial intensity proﬁles for the as-sputtered and annealed F(Co,
or Py)/Al multilayers suggest some limited intermixing exists, both HRTEM and
diffraction studies show no obvious intermediate phase formation. In particular, the
annealing treatments do not signiﬁcantly alter the multilayer structures.

The equilibrium phase diagram of the Co—Al system shows that the B2-ordered
CoAl phase has a very high melting temperature, which suggests this phase is very stable
in terms of volume free energy. However, there is no obvious evidence showing the any
intermediate phase formed in the Co/Al multilayers. It is well known that when a
material’s dimension decreases to the nanometer scale, the surface and interfacial energy
become more signiﬁcant and their contributions to the total free energy cannot be ignored.
Consequently, the absence of any intermediate phase in the Co/Al multilayers suggests
the surface energy of the intermediate phase should be high and /or the interfacial energy
between the Co and Al, and the CoAl phase should be high, which inhibits nucleation of
the B2 phase. However, rapid formation of the A19C02 and B2-ordered CoAl phases were
observed in Co(20nm)/Al(50nm) bilayer atom probe tips annealed at 300°C for 5min [47].
This may suggest the surface energy/volume energy balance of these intermetallic phases
alters depending on the ﬁlm thickness. For example, it is lcnown that Co and Cu are
insoluble in bulk systems [19]. However, at the nanometer scale, solid solutions of Co
and Cu have been observed by X-ray diffraction and calorimetric measurements [68]. In
addition, thermodynamic calculations show that the solubility in Co/Cu multilayers varies
depending on the layer thickness [69]. Consequently, the phase stability in different layer

thickness can give valuable insight for rational design of multilayers structures.

125

HRTEM and XEDS microanalysis of the as-sputtered and annealed Cu/Al
multilayers show obvious intermediate phase formation. A12Cll and AlCu3 are formed
when the overall stoichiometry is 52.9 at.% Cu in the Cu(8nm)/Al(10nm) multilayers,
while AlgCu and Cu are formed when the overall stoichiometry is 70.1 at.% Cu in the
Cu(5nm)/Al(3nm) multilayers. For Cu/Al/Cu layers in the spin-valves, evidence of
AlzCu and AlCU3 phase formation in the annealed spin-valve with the 30nm Al layer was
found, while AlzCu and Cu were observed in the as-sputtered spin-valve with the 10nm
Al layer. The overall Al/Cu stoichiometry is 48.4 at.% Cu in the as-sputtered spin-valve
with Cu(10nm)/Al(10nm)/Cu(10nm) layers and 73.7 at.% Cu in the annealed spin-valve
with Cu(lOnm)/Al(30nm)/Cu(10nm) layers. It is interesting that the intermediate phases
formed in these multilayers are similar in some extent if the overall stoichiometries are
close. This reinforces the critical concept that phase formation in small system is
associated with stoichiometries, volume energies, and interfacial energies.

Kung and co-workers [70] have observed both fee and bcc Cu in the CW
multilayers. That study found that the variation of the Cu structures depended on the
layer thickness between 1.2 and 100 nm with constant volume ﬁaction of Cu of 50%.
For layer thickness above 2.5nm, fcc Cu and bcc Nb formed a Kurdjumov-Saches
orientation relationship. However, for layer thickness under 1.2, a slight distortion of the
bee Cu was observed. Of critical note, they found that the lattice mismatch play a more
important role when the layer thickness decreased below a critical value. No
intermediate phases were observed, consistent with the Cu-Nb phase diagram, which
shows that these metals are insoluble. Geng and co-workers [71] have observed

nonequilibrium bcc Cu structures in Py/Cu spin valves with Cu layer adjacent Nb contact

126

layer. HRTEM and FF T analyses from that study show that in some columnar grains in
the spin valve bcc Cu grew on {110} planes and was epitaxial with the Nb contacts.

In the case of Cu/Al layers observed in the present study, interfacial energies of
the intermediate phases will vary with the overall stoichiometry of the Cu and Al in the
multilayers. Not only will the mismatch contribution to the interfacial energy be a
function of the stoichiometry of the phase (which in turn can be a function of the overall
stoichiometry) due to lattice parameter effects, but the interfacial energy will also be a
function of the changes in chemical bonding associated with stoichiometry changes.

In summary, no intermediate phase formation is observed in both as-sputtered and
annealed Co/Al and Py/Al multilayers. Any interfacial mixing, if it exists, is limited in
scale. Large lattice mismatches result in the interfaces tilting signiﬁcantly to
accommodate the strain. Different intermediate phases were observed in the Cu/Al
multilayers and Cu/Al regions in the spin-valves in both as-sputtered and annealed
conditions. The phases formed vary with the overall stoichiometry of the Cu and Al.
Nevertheless, the intermediate phases always grow on their close-packed planes and

orient in the direction with the minimum lattice mismatches.

127

Chapter 6 Overall Discussion and Summary

This dissertation presents an HRTEM and EELS study of a number of nanoscale
structures. The critical issues that have been focused on include crystal structures,
stoichiometries, phase stabilities, interfacial epitaxy, interfacial intermixing, and the
electronic nature of these systems. It is well known that materials can behave far from
their equilibrium states when their dimensions decrease to the nanometer scale. At this
scale, the surface and interfacial energies become non-negligible factors in the behavior
of materials. The selected systems in this study provide basic insights into the balances
in thermodynamic volume ﬁee energies and interfacial energies that affect nucleation,
growth, and stability of materials at the nanoscale. The understanding of the structure
and stability of such nanoscale materials are critical to realize rational design and
synthesis with desired properties.

In the studies of Gd silicide nanostructures, it has been found that both hexagonal
GdSiz.x and thet GdSi; can self-assemble on Si(lOO). FFT analysis shows that the
lattice parameters of the nanostructure phases do not deviate signiﬁcantly from the bulk
values. In the equilibrium Gd-Si system, the orth/tet GdSiz phase has a high congruent
melting temperature, signiﬁcantly higher than the pertectoid temperature that forms the
hexagonal phase (Gd-Si phase diagram shown in ﬁgure 2.4), suggesting the orth/tet phase
is more stable than the hexagonal GdSiz.x phase in terms of volume free energy. This is
consistent with the observation in this study that the orth/tet Gd silicide thin ﬁlms form
from the annealing of alternating Gd/Si layers. The observation that Gd silicide
nanostructures with the orth/tet structure form after extended post deposition annealing

also suggests that this phase has a lower volume free energy. Despite the orth/tet phase

128

being the lower energy phase, the hexagonal phase is found to form in the early stages of
growth. This is consistent with the interfacial energy dominating during the initial
growth stage, which is far from equilibrium. While the isotropic lattice mismatch of the
hexagonal phase with Si(lOO) is deemed responsible for the formation of nanowires, the
overall surface energy of the hexagonal/Si(l 00) interface associated with this isotropic
mismatch must be lower than that of the orthorhombic/Si( 1 00) interface in order to
overcome the larger volume free energy of the hexagonal phase. The observation of
overgrowth of the orth/tet phase over the hexagonal phase shows that the hexagonal
phase is energetically favorable at ﬁne scale and can act as a buffer layer between the
orth/tet phase and Si substrate. The reason for this is that the hexagonal phase has much
lower 2D lattice mismatch than the orth/tet phase with Si substrate which in turn results
in a lower interfacial energy. A similar balance between volume ﬁ'ee energy and
interfacial energy is found in the Tm-Si system, as discussed in section 4.2.2.

ELS spectroscopy of Gd, Gd oxide and Gd silicide failed to provide insight into
the differences in chemical environment of the Gd in these materials, other than to verify
that Gd is indeed trivalent in all three cases. In this respect, the three materials selected
were a poor test case to explore the capabilities of a newly installed ELS system to
provide detailed chemical bonding information. Nevertheless, the data presented here are
a reference point for future studies. On the other hand, the sensitivity of the ELS spectra
to plasmon related features provides useful information on metallic multilayer systems.

In the case of metallic multilayers, the interfacial energy/volume energy balance
also determines the resulting nanoscale structures. This is revealed in the different

systems by varying types of phase equilibria.

129

The Co-Al binary phase diagram indicates four different intermetallic compounds
that might be expected to form by interdiffusion of the pure Al and Co layers at the
growth and annealing temperature. Of these, the B2-ordered CoAl phase has a high
congruent melting temperature of ~1 648 °C , much higher than the melting temperatures
of Al and Co at 660 °C and 1494 °C respectively [19], suggesting this CoAl phase should
be very stable. Surprisingly, the HRTEM and diffraction studies show no evidence of
formation of this B2 phase, or any other intermetallic phase formation. This lack of
phase formation persists with annealing. This absence of any intermediate phase
formation in the Co/Al multilayers suggests the interfacial energy between the Co and/or
Al and the potential intermediate phases is high, which inhibits nucleation of the
intermediate phases. This high interfacial energy may be a result of signiﬁcant mismatch
between the base components and the intermetallic phases. Indeed, the bee derivative B2
structure will not form a simple coherent interface with either fcc Al or hcp Co, and
might be expected to require formation of complex variation of a typical fcc/bcc interface
such as a Kurdjumov-Sachs or Nishiyama-Wassermann relationship. Likewise, the other
intermetallic phases in this system, monoclinic A19C02, monoclinic A13Co4, and
hexagonal A15C02 (listed in table 2.7) would be expected to form complex interfaces with
Al and Co as these phases have relatively complex crystal structures.

Similarly, the equilibrium binary (F e-Al, Ni-Al) [l9] and ternary (Ni-Fe-Al) [72]
phase diagrams show that a number of intermetallic phases including B2 F eAl (with
melting temperature of ~1310 °C), B2 AlNi (with melting temperature of ~1 638 °C), L12
AlNi3 (with melting temperature of ~1 395 °C) [19], monoclinic FeNiA19 (with melting

temperature of ~809 °C), and F e3NiAllo (with a poorly deﬁned crystal structure and a

130

melting temperature of ~1 050 °C) [72] might be stable and expected to form in the PyAl
multilayers. However, the relatively complicated crystal structures of these intermetallic
phases would be expected to form more complex interfaces with high surface energies.
Consequently, the complex interfaces between the Ni (and/or Fe) and Al intermediate
phases and the Py and Al might be expected to inhibit nucleation of the intermediate
phases.

An important objective of this study has been to understand the structure of the
metallic multilayers in an effort to correlate their observed magnetoresistance behaviors

with the nanoscale phase and interface structure. The idealized design of multilayers for

devices requires both a large value of the enhanced speciﬁc resistance (ZAR; , M) and a

large scattering asymmetry (y) (deﬁned in section 2.3.3). For the case of the F(Co,
Py)/Al multilayers, the measurements show similarly large values of 2AR* ~10 fﬂmz,

but small values of y _<_ 0.1 for both systems, which suggests they are not completely
idealized for devices [46]. It is also observed that these parameters vary with aging and
annealing, speciﬁcally with the AR increasing by 5-10% after 6-11 months and further by
2-7 % on annealing at 180°C for less than 5 min, suggesting subtle changes in the
structures [46]. However, the TEM observation in this study found no obvious
intermediate phase formation and interfacial structure change. Consequently, the
abnormal MR behavior changes observed in the F(Co or Py)/Al multilayers cannot be
attributed to changes in interfacial structure or intermediate phase formation. Other
changes, such as (1) variations in the stacking fault densities in the multilayers, and (2)
changes in electronic nature of the transition metals (Co, Ni and Fe) at the interfacial

region caused by interfacial ordering changes, may affect the multilayer MR behavior. In

131

order to attribute the observed changes in MR behavior to the changes in stacking fault
density, a thorough statistical analysis measuring the stacking fault probability across the
multilayers is necessary. Since XRD is more effective than TEM in determining the
stacking fault densities across larger volumes, further analysis combining extensive TEM
measurements with x-ray diffraction (XRD) may be useful for understanding the MR
changes with aging and annealing. The electronic nature of the transition metals (Co, Ni
and Fe) at the interfacial region can be experimentally investigated by studying the near-
edge ﬁne structures of these elements for chemical shifts and relative intensities changes
of the L-edges. However, such detailed information requires higher resolution EELS
capabilities which are not available at MSU.

Although the Cu/Al multilayers were not originally synthesized for GMR devices,
these multilayers were produced to obtain fundamental data on the CPP interface speciﬁc
resistance (area A times resistance R). Such data is critical for interpreting the large AR*
in F/Al multilayers. However, the TEM results indicate that correlating the data of Cu/Al
multilayers to F/Al multilayers is not straightforward because the HRTEM and XEDS
microanalyses of both the as-sputtered and annealed Cu/Al multilayers show obvious
intermediate phase formation.

The formation of these different intermediate phases is quite interesting and can
again be related to the balance between the volume free energies and surface and/or
interfacial free energies. Based on the equilibrium phase diagram of Cu and Al (ﬁgure
2.17), one might expect that 112 AlCu (monoclinic) and C2 (poorly deﬁned) phases (see
table 2.8 for the crystal structures in the Al/Cu system) will form when the overall

stoichiometry is 52.9 at.% Cu in the Cu(8nm)/Al(10nm) multilayers (based on a simple

132

lever rule calculation, the resulting structure would be expected to be 47.3% 112 and
52.7% C2). Similarly, 69% 71 A14Cu9 (complex cubic) and 31% a2 (poorly deﬁned) will
form when the overall stoichiometry is 70.1 at.% Cu in the Cu(5nm)/Al(3nm) multilayers.
However, our results show that AlzCu (tetragonal) and AlCu; (disordered bcc) are formed
when the overall stoichiometry is 52.9 at.% Cu (Cu(8nm)/Al(10nm) multilayers), while
AlzCu and Cu are formed when the overall stoichiometry is 70.1 at.% Cu
(Cu(5nm)/Al(3nm) multilayers). The observed Cu-Al intermetallic phases were grown
with preferred orientation relationships and relatively small lattice mismatches,
suggesting the interfacial energy is small. Other intermetallic phases such as monoclinic
AlCu, hexagonal A13Cu4, and cubic A14CU9 have not been observed in the multilayers
because more complex interfaces are expected to form. Consequently, the high surface
energy prohibits the nucleation of these phases. In addition, different phases are
observed in Cu/Al multilayers with different layer thickness and overall stoichiometry
suggesting that the balance between thermodynamic volume ﬁ'ee energies and interfacial
energies of the phases depends on the layer thickness and overall stoichiometry.

The observed intermediate phases are expected to have signiﬁcantly different
ARs than the base Cu and Al. Table 6.1 lists the electrical resistivity data of pure Al,
pure Cu (fee), and Cu/Al alloys Al(67%)/Cu(33%) and Al(25%)/Cu(75%) that have
stoichiometries close to AlzCu and A1Cll3 phases respectively [73]. Nevertheless, it
should be noted that the bee solid solution AlCu3 is not a stable phase at room
temperature and its electrical resistivity is not practically measurable. Consequently, the
alloy Al(25%)/Cu(75%) shown in the table should only be considered as an

approximation for the resistivity of this phase. As both the 8/10 and 5/3 Cu/Al

133

multilayers contain the tetragonal A12Cu phase in similar proportions (roughly 2:1- Al-
rich phase to Cu-rich phase), it would be expected that differences in the Cu-rich phase
will have a strong inﬂuence on the measured AR. When the sputtered Al is relatively
thick, the AlCu; bcc solid solution is observed with the tetragonal AlzCu phase: Since
AlCu3 stoichiometry has large resistivity, the overall AR will be large. When the
sputtered Al is relatively thin, fcc Cu is observed: Since Cu has a small resistivity, the
overall AR will be low. Since the intermediate phases formed in both Al/Cu multilayers
and the Cu/Al/Cu region of the Py/Cu/Al based spin-valves are similar to some extent if
the overall stoichiometries are close (discussed in section 5.3), the relationship between
the transport and possible intermediate phase formation should be considered when

designing GMR devices containing Cu/Al layers [48].

Table 6.1 Electrical resistivity data of the phases in the Cu/Al multilayers, after [73].

 

 

 

 

 

 

 

 

Electrical resistivit
Alloy (LI 9 cm) Y
Al 2.49
Cu (fcc) 1.54
Al(67%)Cu(33%) ~6
Al(25%)Cu(7 5%) ~20

 

In summary, a number of nanoscale structured electronic materials have been
studied using a broad range of microanalytical techniques, including HRTEM,
complimented with XEDS, EELS, and STEM. Characterization of the crystal structures,

interfacial epitaxy, interfacial chemical nature, and electronic nature has lead to a

134

comprehensive understanding of the balance between volume free energy and interfacial
energy determining the phase formation and phase stability on the nanometer scale.
These fundamental structural studies are correlated with their physical properties in an
effort to realize rational design and applications of these nanoscale electronic materials.
The possibility of intermediate phase formation, in equilibrium and nonequilibium states,
should be taken into account when designing electronic materials structured at the

nanometer scale.

135

Chapter 7 Suggestions for Future Work
7.1 Patterned Growth Studies of Silicide Nanostructures

The epitaxial growth of rare-earth (RE) metal silicides can result in nanowires and
nanoislands self-assembling on Si(001) due to the mismatch between the silicides and the
Si(001) surface. However, these nanostructures tend to grow in random locations (and
unless a vicinal substrate is used, in two directions) instead of at speciﬁc locations, since
the nucleation of the silicides on the Si substrate is random. In order to prompt these
nanostructures to grow in desired patterns and locations, one would have to imagine a
way to design patterns for controlled nucleation ﬁrst. There are advanced techniques
available, such as focused ion beam (FIB) nanopatterning, which may realize nucleation
in a designed pattern. FIB (with a high resolution of ~2nm) can be used to mill trenches
or holes in the substrate in a pattern before the metals are deposited, and in this way, the
nucleation might be energetically favorable at these trenches or holes. In addition, one
can also take advantage of FIB for TEM sample preparation. STM and TEM study (plan-
view and cross-sectional) may be more closely correlated by making TEM samples of
speciﬁcally identiﬁed nanowires and structures using FIB. In particular, specimens of
both orientations (plan-view and cross-sectional) can be obtained from the same growth
sample, which is not possible with the sandwich technique described in this thesis. In
this way, the nanowire and island morphologies of the silicide nanostructures can be
more clearly correlated with the crystal structures. Speciﬁc features, such as
nanostructure junctions, can be studied in detail. Consequently, the nucleation and

growth mechanism of the nanostructures can be more clearly understood in the

136

combination of plan-view and cross-sectional TEM. Moreoever, the yield of usable

samples from a small number of growth samples would greatly increase.

7.2 High-resolution EELS Analysis Using Monochromated TEM

Monochromated transmission electron microscopes are now commercially
available and have been reported to have full-width at half-maximum (FWHM) energy
resolutions of 0.1 - 0.25 eV with a cold ﬁeld-emission electron gun (CFEG) [74]. One of
the advantages of the monochromator is that it allows much higher resolution EELS
analysis to be carried out, which can be used to more thoroughly explore the band
structure of electronic materials. Although X-ray absorption has a better energy
resolution (0.03 eV), its spatial resolution (50 um) is not suitable for investigation of the
electronic nature of individual nanoscale structures.

Lazar and co-workers [75] have studied the near-edge ﬁne structure of Ga L2,;
and N ionization edges in zinc—blend GaN and wurtzite GaN using high-resolution EELS
(energy resolution: 0.2 eV) in a monochromated TEM. The experimental spectra were
compared with calculations based on the density functional theory using the Wien2k code.
Different electronic behaviors in the two crystal structures were correlated with the
density of states in different bonding environments. In a similar manner, by studying the
near-edge ﬁne structures of RE silicide nanostructures with different phases using HR-
EELS, the electronic nature of the different phases and at the interfaces could be more

clearly understood.

137

7 .3 EELS Quantiﬁcation Analysis in Magnetic Multilayers

The abnormal MR behavior observed in the F (Co or Py)/Al multilayers can not be
attributed to the intermediate phase formation based on the observation in this thesis.
However, the electronic nature of the transition metals (Co, Ni and Fe) can be affected by
the changes in bonding in the interfacial region, which in turn may affect the entire
multilayer MR behavior. This information can be experimentally investigated by
studying the near-edge ﬁne structures of the atoms, such as the chemical shifts and
relative intensities of the L-edges (white lines in the Co, Ni, and Fe).

Similar near-edge ﬁne structure characterization could also be carried out on
Cu/AJ multilayers. Kung and co-workers [70] have observed a 2 eV shift in the Cu L3
edge in fee and bcc Cu using EELS characterization. This result suggests that bcc Cu has
a lower Fermi energy than fcc Cu. Using this approach to study the Cu/Al multilayers,
the difference of the electronic structure of Cu in the fcc Cu and the intermediate Cu/Al
phases (AlzCu, ordered and disordered AlCu3) could be determined.

Thus, using a monochromated TEM with high energy resolution EELS to study
the near-edge ﬁne structure of the transitional metals in the multilayers using spatially

resolved EELS may give some clues of the origin of the abnormal MR behaviors.

138

REFERENCES

. J. Solomon, Chemical Industry R8D Roadmap for Nanomaterials by Design:
From Fundamentals to Function (2003);
http://www.chemicalvision2020.org/nanomaterialsroadmap.html

. D. B. Williams and C. B. Carter, Transmission Electron Microscopy: A Textbook
for Materials Science, ISBN-10: 030645324X (Plenum, New York, 1996).

. R. F. Egerton, Electron Energy-Loss Spectroscopy in the Electron Microscope,
2'” ed., ISBN-10: 0306452235 (Plenum, New York, 1996).

. R. D. Leapman, L. A. Grunes, and P. L. Fejes, Study of the L23 edges in the 3d
transition metals and their oxides by electron-energy-loss spectroscopy with
comparisons to theory, Phys. Rev. B, 26, p.614-635 (1982).

. R.F. Egerton and M. J. Whelan, Electron energy-loss spectra of diamond,
graphite and amorphous carbon, J. Electron Spectrosc., 3, p.232-236 (1974).

. D. Pine, Elementary Excitations in Solids, ASIN: BOOOIG3ZO8 (W .A. Benjamin,
New York, 1963).

. H. Kurata and C. Colliex, Electron-energy-loss core-edge structures in
manganese oxides, Phys. Rev. B, 48, p.2102-2108 (1993).

. C. Preinesberger, S. Vandré, T. Kalka, and M. Dahne-Prietch, Formation of
dysprosium silicide wires on Si(001), J. Phys. D, 31, p.L43-L45 (1998).

. Y. Chen, D.A.A. Ohlberg, G. Medeiros-Ribeiro, Y.A. Chang, and RS. Williams,
Self-assembled growth of epitaxial erbium disilicide nanowires on silicon (001),
Appl. Phys. Lett., 76, p.4004-4006 (2000).

10. J. Nogami, B.Z. Liu, M.V. Katkov, C. Ohbuchi, and NO. Birge, Self-assembled

rare-earth silicide nanowires on Si(001), Phys. Rev. B, 63, p.233305-8 (2001).

139

11. Y. Chen, D. A.A.Ohlberg, and R. S. Williams, Nanowires of four epitaxial
hexagonal silicides grown on Si(001), J. App. Phys., 91, p.3213-321 8 (2002).

12. C. Ohbuchi and J. Nogami, Holmium growth on Si(001): Surface reconstructions
and nanowire formation, Phys. Rev. B, 66, p.165323 (2002).

13. R. Ragan, Yong Chen, D. A.A.Ohlberg, G. Medeiros-Ribeiro, and RS. Williams,
Ordered arrays of rare-earth silicide nanowires on Si (001 ), Journal of Crystal
Growth, 251, p.657-661 (2003).

14. D. Lee and S. Kim, Formation of hexagonal Gd disilicide nanowires on Si(100),
Applied Physics Letters, 82(16), p.2619-2621 (2003).

15. M. Kuzmin, P. Laukkanen, R.E. Perala, R.-L. Vaara, and 1.]. Vayrynen,
Formation of ytterbium silicide nanowires on Si(001), Applied Surface Science,
222, 394 (2004).

16. M.V. Katkov and J. Nogami. Yttrium growth on Si(001). in American Physical
Society (APS) March Meeting (2002).

17. J. Zhang, M. A. Crimp, Y. Cui, and J. Nogami, Self-assembled Tm silicide
nanowires grown on Si(001) studied by ST M and T EM, in preparation.

18. A. Vantomme, M.F. Wu, S. Hogg, U. Wahl, W. Deweerd, H. Pattyn, G.
Langouche, S. J in, and H. Bender, Stabilisation and phase transformation of
hexagonal rare-earth silicides on Si(I I I), Nucl. Instrum. Meth. Phys. Res. B,
147, p.261-266 (1999).

19. TB. Massalski, H. Okamoto, P.R. Subramanian, and L. Kacprzak, Binary alloy
phase diagrams. Vol. 2. 1990, Metals Park, OH: ASM International.

20. G. Ye, Thesis of Rare-earth Silicide Nanostructure Epitaxial Growth and

Crystallography, Ph.D. dissertation, Michigan State University, East lansing,
Michigan (2006).

140

21. G. Ye, J. Nogami, and M.A. Crimp, Dysprosium disilicide nanostructures on

Si(001) studied by scanning tunneling microscopy and transmission electron
microscopy, Thin Solid Films, 497(1-2), p.48-52 (2006).

22. K.N. Tu, R.D. Thompson, and B.Y. Tsaur, Low Schottky barrier of rare-earth
silicide on n-Si, Applied Physics Letters, 38(8), p.626-628 (1981).

23. H.W. Yeom, Y.K. Kim, E.Y. Lee, K.-D. Ryang, and P.G. Kang, Robust One-
Dimensional Metallic Band Structure of Silicide Nanowires, Physics Review
Letters, 95, p.205504 (2005).

24. C. Preinesberger, G. Pruskil, S.K. Becker, M. Dahne, D.V. Vyalikh, S.L.
Molodtsov, C. Laubschat, and F. Schiller, Structure and electronic properties of

dysprosium-silicide nanowires on vicinal Si(001), Applied Physics Letters, 87(8),
p.083107 (2005).

25. K Suenaga, S. Iijima, H. Kato and H. Shinohara, F ine-structure analysis of Gd
M45 near-edge EELS on the valence state of Gd@C82 microcrystals, Phys. Rev. B,
62, p. 1627-1630 (2000).

26. T. Manoubi, C. Colliex and P. Rez, Quantitative electron energy loss

spectroscopy on M 45 edges in rare earth oxides, J. Electr. Spectr. and Rel. Phen.,
50, p.1-18 (1990).

27. C. C. Ahn, ed., Transmission electron energy loss spectrometry in materials
science and the EELSATLAS, 2'“ ed., ISBN-10: 3527405658 (Wiley, Weinheim,
2004).

28. A. Travlos, P. Aloupogiannis, E. Rokofyllou, C. Papastaikoudis, A. Traverse, G.
Weber, Growth, characterization and electrical properties of gadolinium silicide
thin layers, Philosophical Magazine B, 67, p.485-495 (1993).

29. R. G. Long, M. C. Bost, and J. E. Mahan, Metallic behavior of lanthanum
disilicide, Applied Physics Letters, 53, p.1272-1273 (1988).

30. A. Barthélémy, A. Fert, J-P. Contour, M. Bowen, V. Cros, J. M. De Teresa, A.
Hamzic, J. C. Faini, J. M. George, J. Grollier, F. Montaigne, F. Pailloux, F.

141

31.

32.

33.

34.

35.

36.

37.

38.

39.

Petroff, and C. Vouille, Magnetoresistance and spin electronics, Journal of
Magnetism and Magnetic Materials, 242-245, p.68-72 (2002).

G. Binash, P. Griinberg, F. Saurenbach, and W. Zinn, Enhanced
magnetoresistance in layered magnetic structures with antiferromagnetic
interlayer exchange, Phys. Rev. B, 39, p.4828-483O (1989).

M. Baibich, J .M. Broto, A. Fert, F.N. Guyen Van Dau, F. Petroff, P. Etienne, G.
Creuzet and A. F riederich, Giant Magnetoresistance of (001 )F e/(001 )Cr
Magnetic Superlattices, Phys. Rev. Lett., 61, p.2472-2475 (1988).

R. Loloee, Growth and Characterization of Epitaxial Films and Magnetic
Multilayers Grown by DC Sputtering, Ph.D. dissertation, Michigan State
University, East lansing, Michigan (2000).

J. Bass and W. P. Pratt Jr., Current-perpendicular (CPP) magnetoresistance in
magnetic metallic multilayers, Journal of Magnetism and Magnetic Materials, 200,
p.274-289 (1999).

B. Dieny, V.S. Speriosu, S.S.P. Parkin, B.A. Gurney, D.R. Wilhoit and D. Mauri
Giant magnetoresistive in soft ferromagnetic multilayers, Phys. Rev. B, 43,
p.1297-13OO (1991).

V.S. Speriosu, B. Dieny, J .-P. Nozieres, B.A. Gurney, P. Humbert, H. Lefakis, P.
Baumgart and D. R. Wilhoit, Magnetoresistive properties of spin valves, Digest of
the INTERMAG’92, Conference, paper CA-04 (1992).

S. Urazhdin, N. O. Birge, W. P. Pratt, Jr., and J. Bass, Switching current versus
magnetoresistance in magnetic multilayer nanopillars, Appl. Phys. Lett., 84,
p.1516-1518 (1995).

N. Theodoropoulou, A. Sharma, R. Loloee, W.P. Pratt, Jr., J. Bass, A. Fert, and H,
J afﬁ'es, Interface Speciﬁc Resistance and Scattering Asymmetry of Permalloy/Al,
J. Appl. Phys., 99, 08G502 (2006).

K. Le Dang, P. Veillet, E. Velu, S.S.P. Parkin and C. Chappert, Inﬂuence of
crystal structure on the magnetoresistance of Co/Cu multilayers, Appl. Phys. Lett.,
63, p.108-110 (1993).

142

40.

41.

42.

43.

45.

E. E. Fullerton, D. M. Kelly, J. Guirnpel, I.K. Schuller, and Y. Bruynseraede,
Roughness and giant magnetoresistance in F e/Cr superlattices, Phys. Rev. Lett.,
68, p.859-862 (1992).

V. S. Speriosu, J. P. Nozieres, B. A. Gurney, B. Dieny, T. C. Huang and H.
Lefakis, Role of interfacial mixing in giant magnetoresistance, Phys. Rev. B, 47,
p.11579-11582 (1993).

J. Kudmovsky, V. Drchal, I. Turek, C. Blaas, P. Weinberger, and P. Bruno, The
inﬂuence of defects on electrical transport in magnetic multilayers, J. Mater.,
52(7), p.29 (2000).

A. G. F. Garcia, 1. N. Krivorotov, P. M. Braganca, D. C. Ralph, and A. R.
Buhnnan, Bull. Am. Phys. Soc., 50, p.237 (2005).

. N. Theodoropoulou, A. Sharma T. Haillard, R. Loloee, W.P. Pratt Jr., J. Bass, J.

Zhang and M.A. Crimp, Specific Resistance, Scattering Asymmetry, and some
Thermal Instability, of Co/Al, F e/Al, and C091Fe9/Al Interfaces, IEEE
Transactions on Magnetics, 43(6), p.2860-2862 (2007).

A. Zambano K. Eid, R. Loloee, W.P. Pratt Jr., and J. Bass, Interfacial Properties
of F e/Cr Multilayers n the Current-Perpendicular-to-Plane Geometry, Journal of
Magnetism and Magnetic Materials, 253, p.51-55 (2002).

46. A. Sharma, N. Theodoropoulou, T. Haillard, R. Loloee, W. P. Pratt. Jr., J. Bass, J.

Zhang and M. A. Crimp, Current-perpendicular to plane (CPP)
magnetoresistance of Ferromagnetic (F )/Al interfaces (F = Py, Co, Fe and
C091Fe9) and structural studies of Co/Al and Py/Al. (in preparation)

47. V. Vovk, G. Schmitz, and R. Kirchheim, Nucleation of product phase in reactive

diﬂusion ofAl/Co, Phys. Rev. B, 69, p.104102 (2004).

48. A. Shanna, N. Theodoropoulou, R. Loloee, W. P. Pratt. Jr., J. Bass, J. Zhang and

M. A. Crimp, Current-perpendicular to plane (CPP) Resistance of Cu/Al
Interfaces. (in preparation)

143

49. J .C. Bravman and R. Sinclair, The Preparation of Cross—section Specimens for
Transmission Electron Microscopy, Journal of Electron Microscopy Technique, 1,
p.52-61 (1984).

50. RA. Stadelmann, EMS - a software package for electron diﬁ'action analysis and
HREM image simulation in materials science, Ultramicroscopy, 21, p.131-145

(1987)

51. T. Manoubi, C. Colliex and P. Rez, Quantitative electron energy loss
spectroscopy an M4 5 edges in rare earth oxides, J. Electr. Spectr. and Rel. Phen.,
50, p.1-18 (1990).

52. H. Kurata and C. Colliex, Electron-energy—loss core-edge structures in
manganese oxides, Phys. Rev. B, 48, p.2102-2108 (1993).

53. I. Gerocs, G. Molnar, E. Jaroli, E. Zsoldos, G. Peto, J. Gyulai and E. Bugiel,
Epitaxy of orthorhombic gadolinium disilicide on <100> silicon, Applied Physics
Letters, 51(25), p.2144-2145 (1987).

54. G. Molnar, I. Gerocs, G. Peto, E. Zsoldos, J. Gyulai and E. Bugiel, Epitaxy of
GdSig 1.7 on <111> Si by solid phase reaction, Applied Physics Letters, 58(3),
p.249-250 (1991).

55. D. Lee and S. Kim, Formation of hexagonal Gd disilicide nanowires on Si(100),
Applied Physics Letters, 82(16), p.2619-2621 (2003).

56. G. Ye, J. Nogami, and M.A. Crimp, Crystallographic study of self-assembled
dysprosium silicide nanostructures on Si(001), Phys. Rev. B, 74, p.033104 (2006).

57. A.J. Nelson, J .J . Adams, and KL Schaffers, Electronic structure of gadolinium
calcium oxoborate, Applied Surface Science, 252(5), p.1228-1232 (2005).

58. E. Puppin, J. Nogami, C. Carbone, Z. X. Shen, I. Lindau , B. B. Pate, I. Abbati
and L. Braicovich, A photoemission study of the Si(111)/Gd interface: A
comparison with the bulk silicides, Journal of Vacuum Science & Technology B,
5, p.1083-1086 (1987).

144

59. A. V. Fedorov '1, E. Arenholz, K. Starke, E. Navas, L. Baumgarten, C. Laubschat,
and G. Kaindl, Surface Shifts of 4f Electron-Addition and Electron-Removal
States in Gd(0001), , Phys. Rev. Lett. 73, p.601-604 (1994).

60. B. T. Thole, G. van der Laan, J. C. Fuggle, G. A. Sawatzky, R. C. Kamatak and
J .-M. Esteva, 3d x-ray-absorption lines and the 3d94f+1 multiplets of the
lanthanides, Phys. Rev. B, 32 (8), p.5107-5118 (1985).

61. Z. Hu, G. Kaindl, B.G. Muller, Core-level spectroscopy of the tetravalent
lanthanide compounds M 3LnF 7 (with M = Cs, Rb; LN = Ce-Pr,Nd,Tb,Dy), J.
Alloys Compd., 246, p.177-185 (1997).

62. A. Iandelli, A. Palenzona and G. L. Olcese, Valence ﬂuctuations oy iterbium in
silicon-rich compounds, J. Less- Common Metals, 64, p.213-220 (1979)

63. J. Sugar, Potential-Barrier Eﬂ'ects in Photoabsorption. II. Interpretation of
Photoabsorption Resonances in Lanthanide Metals at the 4d-Electron Threshold,
Phys. Rev. B, 5, p.1785-l792 (1972)

64. Ch. Rath, J. E. Prieto, S. Muller, R. Miranda, and K. Heinz, hcp-to-fcc stacking
switch in thin cobalt ﬁlms induced by Cu capping, Phys. Rev. B, 55, p.10791-
10799 (1997).

65. A F. Guillermet, Critical evaluation of the thermodynamic properties of cobalt,
Int. J. Thermophys., 8, p.481-510 (1987).

66. M. J. Mills, Determination of compositional ordering at grain boundaries in
boron-doped Ni 3A1, Scripta Met, 23, 2061 (1989).

67. H-J. H. Kung, The Structure and Chemistry of Grain Boundaries in Nickel
Alumnide, Ph.D. dissertation, Cornell University, Ithaca, New. York (1991).

68. C. Gente, M. Oehring, R. Bormann, Formation of thermodynamically unstable
solid solutions in the Cu-Co system by mechanical alloying, Phys. Rev. B, 48,
p.13244-l3252 (1993).

69. W. Cao, J. Zhu, Y. Yang, W.A. Oates, Y.A. Chang, Thermodynamic stability of
Co/Cu multilayered nanostructures, Scripta Materialia, 53, p.1379-1382 (2005).

145

70.

71.

72.

73.

74.

75.

H. Kung, Y-C. Lu, A. J. Grifﬁn, Jr., M. Nastasi, T. E. Mitchell, and J. D. Embury,
Observation of body centered cubic Cu in Cu/Nb nanolayered composites, Appl.
Phys. Lett., 71, p.2103-2105 (1997).

H. Geng, J .W. Heckrnan, W.P. Pratt Jr., J. Bass, F.J. Espinosa, S.D. Conradson, D.
Lederman, and M.A. Crimp, Occasional ”long-range" nonequilibrium body-
centered-cubic structures in NiFe/Cu spin valves, J. Appl. Phys., 86, 4166 (1999).

L. Eleno, K. F risk, and A. Schneider, Assessment of the F e-Ni-Al system,
Intermetallics, 14(10-11), p.1276-1290 (2006).

K. Schré'rder, CRC Handbook of Electrical Resistivities of Binary Metallic Alloys,
ISBN-10: 0849335205, p.79 (CRC Press, Inc. 1983).

K. Kimoto, G. Kothleitner, W. Grogger, Y. Matsui, F. Hofer, Advantages of a
monochromator for bandgap measurements using electron energy-loss
spectroscopy, Micron,. 36(2), p.185-189 (2005).

S. Lazar, C. Hébert and H. W. Zandbergen, Investigation of hexagonal and cubic
GaN by high-resolution electron energy-loss spectroscopy and density functional
theory, Ultramicroscopy, 98, p.249-257 (2004).

146

APPENDIX A Detailed TEM Sample Description

. Gd oxide powder (in ﬁgure 4.1and 4.2): micron sized Gd203 powder (from
Alfa Aesar) were grounded to smaller size.

. Sputtered Gd/Si layers (in ﬁgure 4.3): Si 43 nm/ Gd 75nm/ Si 85 nm/ Gd 75
nm/ Si 43nm layers were deposited in a DC-magnetron triode sputtering
system at room temperature.

. Bulk GdSi; layers (in ﬁgure 4.4): the alternating Gd/Si layers (2) were
annealed in a high vacuum of 5><1045 torr at 900 °C for 1 hour to form silicides.
. GdSiz nanostructures (STM reference # 041105) (in ﬁgure 4.5, 4.6, 4.7, and
4.8): the nanostructures were grown on a Si substrate heated to 650°C with a
metal coverage of 1.5ML. This was followed by post-deposition annealing
for 20 min at the growth temperature.

. Tm silicide nanostructure sample Tm-A (STM reference # Tm060619) (in
ﬁgure 4.9 and 4.13) was grown with 2.8 ML metal coverage without post-
deposition annealing.

. Tm silicide nanostructure sample Tm-B (STM reference # Tm082006) (in
ﬁgure 4.10 and 4.12) grown with 1 ML metal coverage with a post-deposition
annealing at 600°C for 30 min.

. Tm silicide nanostructure sample Tm-C (STM reference # Tm070215) (in
ﬁgure 4.14, 4.15, and 4.16) was grown under very similar conditions to Tm-B:
1 ML Tm deposited at 600°C with 20 minutes post annealing. However, this

sample was also capped with a layer of amorphous Can.

147

10.

ll.

12.

l3.

14.

15.

As-sputtered Co/Al multilayers (GMR reference 1693-4b) (in ﬁgure 5.1, 5.2,
5.3, and 5.6): alternating Co and Al layers with 20 nm Co and 10 nm Al were
sputtered on Si(lOO) substrates.

Annealed Co/Al multilayers (GMR reference l693-4b) (in ﬁgure 5.5): as-
sputtered multilayers were annealed for less than 5 min at 180°C after
sputtering in the same chamber.

As-sputtered Py/Al multilayers (GMR reference 1693-4a) (in ﬁgure 5.10):
alternating Py and Al layers with 20 nm Py and 10 nm Al were sputtered on
Si(lOO) substrates.

Annealed Py/Al multilayers (GMR reference 1693-5b) (in ﬁgure 5.10, 5.11):
as-sputtered layers were annealed for less than 5 min at 180°C after sputtering
in the same chamber.

As-sputtered Cu(8nm)/Al(10nm) multilayers (GMR reference 1669-3b) (in
ﬁgure 5.14 and5.15)

Annealed Cu(8nm)/Al(10nm) multilayers (GMR reference l669-3b) (in ﬁgure
5.17 and5 .18): as-sputtered layers were annealed for less than 5 min at 180°C
after sputtering in the same chamber.

As—sputtered Cu(5nm)/Al(3nm) multilayers (GMR reference 1669-5b) (in
ﬁgure 5.19 and 5.20)

As-sputtered

Nb(1 50)Cu(5)FeMn(8)Py(24)Cu(l0)Al(10)Cu(10)Py(24)Cu(5)Nb(50) spin-

valve (GMR reference l689-3a) (in ﬁgure 5.22, 5.23, 5.24, 5.25, and 5.26)

148

1 6. Annealed

Nb(1 50)Cu(5)FeMn(8)Py(24)Cu(10)Al(30)Cu(]0)Py(24)Cu(5)Nb(50) spin-
valve (GMR reference 1689-6a) (in ﬁgure 5 .27, 5 .28, 5.29, and 5.30): as-

sputtered layers were annealed for less than 5 min at 180°C after sputtering in

the same chamber.

*‘149

   

IIi]ll]lli][iil][ilii]lwill]I