Microstructural and magnetic properties of polycrystalline and epitaxial permalloy (Ni80Fe20) multilayered thin films - CaltechTHESIS
CaltechTHESIS
A Caltech Library Service
About
Browse
Deposit an Item
Instructions for Students
Microstructural and magnetic properties of polycrystalline and epitaxial permalloy (Ni80Fe20) multilayered thin films
Citation
Hashim, Imran
(1994)
Microstructural and magnetic properties of polycrystalline and epitaxial permalloy (Ni80Fe20) multilayered thin films.
Dissertation (Ph.D.), California Institute of Technology.
doi:10.7907/X1XA-W810.
Abstract
NOTE: Text or symbols not renderable in plain ASCII are indicated by [...]. Abstract is included in .pdf document.
Permalloy ([...]) thin films are of great scientific and technological interest because of their unique soft magnetic properties, and applications to magnetic recording. Chapter 1 provides an introduction to magnetic and magnetotransport properties of [...] thin films, and how the film microstructure affects these properties. Chapter 2 discusses the instrumentation used for thin film fabrication, and for magnetic and structural characterization. Further details of instrumentation are discussed in Appendix A.
Typically, the [...] films for magnetoresistive applications are capped with a refractory metal thin film such as Ta to prevent its oxidation and corrosion. We investigated the interdiffusion kinetics of polycrystalline Ta/[...] thin films and found that for 400 [...] T [...] 600'C, there was significant grain-boundary interdiffusion which drastically affected soft magnetic properties of [...]. In Chapter 3, we present details of the microstructural evolution of these multilayers and the subsequent effects on their magnetic properties.
An alternate method for reducing grain-boundary scattering would be to fabricate grain-boundary free epitaxial [...] films. The epitaxy of [...] on MgO, NaCl and Cu had been demonstrated by investigators as early as the 60s. However, none of these substrates are available with as good atomic flatness as Si wafers. Following reports of epitaxial growth of Cu on Si[1], we proposed using it as a seed layer for growing [...] epitaxially on Si. However, there were conflicting reports of Cu epitaxy on Si, as some investigators claimed that Cu epitaxy on Si in UHV was not possible[2]. We were able to resolve some of these controversies (see Chapter 4 for details) and thus fabricate epitaxial Ni80Fe20 films on Cu/Si.
Chapter 5 examines the effect of the lattice mismatch between Cu and [...] and the subsequent strain, on the soft magnetic properties of [...]. To explain these experimentally observed magnetic properties, a micromagnetic model was developed taking into account domain wall interaction with misfit dislocations and film surface roughness especially during the initial stages of epitaxial growth. Finally, epitaxial growth of [...]/Cu on Si suggests the possibility of growing grain-boundary free atomically sharp [...]/Cu multilayers which exhibit recently-discovered "giant" magnetoresistance.
[1] C.A. Chang, Appl. Phys. Lett. 55, 2754 (1989).
[2] B.P. Tonner, J. Zhang, X. Chen, Z-L. Han, G.R. Harp, and D.K. Saldin,
J. Vac. Sci. Technol. B10, 2082 (1992).
Item Type:
Thesis (Dissertation (Ph.D.))
Subject Keywords:
Materials Science
Degree Grantor:
California Institute of Technology
Division:
Engineering and Applied Science
Major Option:
Materials Science
Thesis Availability:
Public (worldwide access)
Research Advisor(s):
Atwater, Harry Albert
Thesis Committee:
Atwater, Harry Albert (chair)
Fultz, Brent T.
Corngold, Noel Robert
Nicolet, Marc-Aurele
Defense Date:
19 May 1994
Record Number:
CaltechETD:etd-11302007-112544
Persistent URL:
DOI:
10.7907/X1XA-W810
Default Usage Policy:
No commercial reproduction, distribution, display or performance rights in this work are provided.
ID Code:
4698
Collection:
CaltechTHESIS
Deposited By:
Imported from ETD-db
Deposited On:
06 Dec 2007
Last Modified:
24 May 2024 18:42
Thesis Files
Preview
PDF (Hashim_i_1994.pdf)
- Final Version
See Usage Policy.
11MB
Repository Staff Only:
item control page
CaltechTHESIS is powered by
EPrints 3.3
which is developed by the
School of Electronics and Computer Science
at the University of Southampton.
More information and software credits
Microstructural and Magnetic
Properties of Polycrystalline and
Epitaxial Permalloy (NiggFe29)
Multilayered Thin Films
Thesis by
Imran Hashim
' Jn Partial Fulfillment of the Requirements
for the Degree of Doctor of Philosophy
California Institute of Technology
Pasadena, California
1994
(Defended May 19, 1994)
Imran Hashim
il
Acknowledgments
It would be extremely difficult for me to give due credit to all the stimu-
lating people I have come across during the course of my graduate studies
at Caltech. I feel I have benefited enormously from interaction with various
members of the Caltech community, which range from my undergrad students
from A.Ph.9 to lab technicians and machinists, fellow graduate students and
of course, members of the faculty.
The energy and enthusiasm for science of the Caltech undergrads in
A.Ph.9, the Solid State Electronics Lab, which I was a TA for, was infec-
tious. Also, I enjoyed working with the undergraduates who worked in our
research group, especially Wurzel Keir who wrote the software for data ac-
quisition of my experiment. I would also like to thank Carol Garland for
teaching me the operation of electron microscopes, TEM sample prepara-
tion, and the importance of being honest and responsible when working in a
multiuser lab facility. It’s also a pleasure to acknowledge Joe Fontana, the
machinist in Keck, who helped me enormously with machining and design of
various parts of my experiment. Without his help, my experiment would not
have survived a day. I would also like to acknowledge Reynold Johnson, the
technician for A.Ph.9 lab, for teaching me how to maintain a sense of humor
when things in lab are not going your way. I would also like to acknowl-
edge the secretary for our research group, Rosalie Rowe, for being extremely
helpful and patient while helping me out with various paperwork. Also, it
is a pleasure to acknowledge Paula Samazan, the librarian of Electrical En-
gineering and Applied Physics, for technical assistance with library research
iv
as well as always providing a welcome break from work by lending advice on
all sorts of worldly matters. Thanks, Paula!
I also enjoyed enormously interaction with various Research Fellows in
our research group, including Winston Saunders, Channing Ahn, Shouleh
Nikzad, Byungwoo Park and Hyun Sung Joo, who taught me how much fun
scientific research can be. I would also like to thank Ruth Brain, Renato Ca-
mata, Heather Frase, Gang He, Ramana Murty, Kirill Schgelov, Pete Sercel,
Jung Shin, Maggie Taylor, Cho-Jen Tsai, Selmer Wong, and Jimmy Yang, for
providing companionship and help during the course of my graduate career
(especially Selmer for proofreading the thesis and reminding me of all the
punctuation and grammar rules that I had forgotten, and Ramana Murty
for help with AFM measurements). I would like to thank Tab Stephens for
helping me setup the Inel power supply for small angle x-ray diffractome-
ter. I would like to thank Jason Reid and Donald Lie for introducing me
to “friendly” GSC basketball as well as help with backscattering and chan-
neling. I would also like to thank Bruce Gurney of IBM Almaden Research
Center for warning me of the perils I would encounter while setting up the
magneto-optic Kerr system and also for making magnetostriction measure-
ments for some of my samples. I would also like to acknowledge Romney
Katti of Jet Propulsion Laboratory for introducing me to magnetism as ap-
plied to magnetic recording.
I would like to thank Prof. Fultz for introducing me to diffraction from
materials, especially x-ray diffraction, Prof. Nicolet for teaching me backscat-
tering spectrometry and workings of semiconductor devices, and Prof. Corn-
gold for making the subject of statistical mechanics and thermodynamics
most interesting as well as intriguing. Finally, I would like to thank my re-
search advisor, Prof. Atwater, for being a great friend, counselor and mentor
at the same time. Not only did I learn most of what I know about electronic
materials and thin film microstructure from him, but he also taught me how
to be innovative in scientific research, how to communicate scientific results
more effectively, and the hookshot in basketball which I’m still working on.
He always seemed to have time to talk to me whether it was related to re-
search, instrumentation, problems in my personal life or job search. It was a
great pleasure to work under his supervision. Thanks, Harry!
Last but not least, I would like to acknowledge my parents for their
support and patience during the course of my thesis. I would not have been
here at Caltech if not for the attention they paid to my education all my life.
Vi
Abstract
Permalloy (NigpFe2o) thin films are of great scientific and technological in-
terest because of their unique soft magnetic properties, and applications to
magnetic recording. Chapter 1 provides an introduction to magnetic and
magnetotransport properties of NigoFe29 thin films, and how the film mi-
crostructure affects these properties. Chapter 2 discusses the instrumentation
used for thin film fabrication, and for magnetic and structural characteriza-
tion. Further details of instrumentation are discussed in Appendix A.
Typically, the NigoFego films for magnetoresistive applications are capped
with a refractory metal thin film such as Ta to prevent its oxidation and corro-
sion. We investigated the interdiffusion kinetics of polycrystalline Ta/NigoFe2o
thin films and found that for 400 < T < 600°C, there was significant grain-
boundary interdiffusion which drastically affected soft magnetic properties of
NigoFe29. In Chapter 3, we present details of the microstructural evolution
of these multilayers and the subsequent effects on their magnetic properties.
An alternate method for reducing grain-boundary scattering would be
to fabricate grain-boundary free epitaxial NigoFe29 films. The epitaxy of
NigpgFeg9 on MgO, NaCl and Cu had been demonstrated by investigators as
early as the 60s. However, none of these substrates are available with as good
atomic flatness as Si wafers. Following reports of epitaxial growth of Cu on
Si[l], we proposed using it as a seed layer for growing NigoFe2o epitaxially
on Si. However, there were conflicting reports of Cu epitaxy on Si, as some
investigators claimed that Cu epitaxy on Si in UHV was not possible[2]. We
were able to resolve some of these controversies (see Chapter 4 for details)
Vil
and thus fabricate epitaxial NiggFe29 films on Cu/Si.
Chapter 5 examines the effect of the lattice mismatch between Cu and
NigoFe9 and the subsequent strain, on the soft magnetic properties of NiggFego.
To explain these experimentally observed magnetic properties, a micromag-
netic model was developed taking into account domain wall interaction with
misfit dislocations and film surface roughness especially during the initial
stages of epitaxial growth. Finally, epitaxial growth of NigoFeo9/Cu on
Si suggests the possibility of growing grain-boundary free atomically sharp
NigoFe29/Cu multilayers which exhibit recently-discovered “giant” magne-
toresistance.
[1] C.A. Chang, Appl. Phys. Lett. 55, 2754 (1989).
[2] B.P. Tonner, J. Zhang, X. Chen, Z-L. Han, G.R. Harp, and D.K. Saldin,
J. Vac. Sci. Technol. B10, 2082 (1992).
viii
List of Publications
Parts of this thesis have been, or will be published under the following titles.
1. “Interdiffusion Kinetics and Magnetic Properties of Ta-Permalloy
Multilayers,” I. Hashim, H.A. Atwater, K.T.Y. Kung, and R.M.
Valletta, Mat. Res. Soc. Proc. Vol. 232, 1991.
2. “Evolution of Structural and Magnetic Properties of Ta-~NiggFe29
Multilayer Thin Films,” I. Hashim, H.A. Atwater, K.T.Y. Kung,
and R.M. Valletta, J. Appl. Phys. 77, 458 (1993).
3. “Orientational and Microstructural Evolution During Epitaxy of
Cu on $i(100),” I. Hashim, B. Park, and H.A. Atwater, Mat. Res.
Soc. Proc. Vol. 280, 1992.
4. “Epitaxial Growth of Cu(001) on $i(001): Mechanisms and Defect
Morphology,” I. Hashim, B. Park, and H.A. Atwater, Appl. Phys.
Lett. 63, 2833, 1993.
5. “In situ Analysis of Structural and Magnetic Properties of Epi-
taxial and Polycrystalline NiggFe29 Thin Films,” I. Hashim and
H.A. Atwater, Mat. Res. Soc. Proc. Vol. 313, 1993.
6. “Epitaxial NiggFe29/Cu Films for Spin-Valve Heterostructures,” I.
Hashim and H.A. Atwater, J. Appl. Phys. 75, 6516 (1994).
7. “Magnetoresistance of Epitaxial NiFe/Cu Spin-Valve Heterostruc-
tures,” H.S. Joo, I. Hashim, and H.A. Atwater, in preparation.
Contents
Acknowledgments iil
Abstract vi
List of Publications viii
1 Introduction 1
1.1 Introduction to Soft Magnetic Materials ............ 1
1.2 Magnetic and Magnetotransport Properties of NiggFex .... 3
1.3. Relationship to Microstructure of Thin Films.......... 6
2 Instrumentation for Thin Film Deposition, Structural and
Magnetic Characterization 13
2.1 Introduction to Ion-Beam Sputtering for Deposition of Thin
Films... 2 13
2.2 UHV Sputtering System .. 2.2... 2.2.2... . 20022-0006, 17
2.3 In Situ Magnetic Characterization using Magneto-Optic Kerr
Effect 2... 18
2.3.1 Theory... 2... 0... 02. ee ee 18
ix
2.3.2 Implementation .....................-. 23
2.4 In Situ Structural Analysis by Reflection Electron Diffraction. 27
2.4.1 Theory... 2... . 0.0.0.0... 02 eee eee 27
2.4.2 Implementation ..................0.008. 28
2.5 X-ray Diffraction Studies... 2... ......20-2.2.20-, 30
2.5.1 Small Angle X-ray Diffraction ........-...0... 30
2.5.2 Large Angle X-ray Diffraction ...........2... 30
2.6 Transmission Electron Microscopy................ 31
2.7 Additional Characterization Tools ................ 32
3 Structural Stability and Magnetic Properties of NiggFeo9/Ta
Multilayers 35
3.1 Motivation... 2... ee 35
3.2 Small Angle X-ray Diffraction ............-..0... 37
3.2.1 Introduction. ................2.0.000. 37
3.2.2 X-ray Reflectivity From Multilayers and Interface Rough-
3.2.3 Interdiffusion in Multilayers and Small Angle X-ray
Diffraction... 2... ee ee eee 43
3.3 Concurrent Grain Growth and Grain Boundary Diffusion... 46
3.4 Interdiffusion Kinetics of Polycrystalline NiggFe29/Ta Multilayers 49
3.5 Evolution of Magnetic Properties ................ 59
3.6 Conclusions .... 2... 2.2.2... 00200 eee ee eee 61
4 Epitaxial Growth of Cu on Si at Room Temperature 66
4.1
4.2
4.3
4.4
4.5
4.6
Xi
Introduction and Motivation ..............-..2.-.. 66
Substrate Preparation and Deposition Conditions ....... 67
In Situ Electron Diffraction Analysis .............. 69
X-ray Analysis .......-2..0- 000-2240 0 ee eee 69
Electron Microscopy Analysis .............2...-. 76
Conclusions .. 2... 2.2.0... ee ee 77
Structural, Magnetic and Magnetotransport Properties of
Epitaxial NiggFe2. Films on Cu/Si 85
5.1 Introduction. .............. 00-2002. 020008. 85
5.2 Strain Relaxation during Heteroepitaxial Growth of Thin Films 89
5.3 Strain Relaxation Measurements using Electron Diffraction . . 93
5.4 Structural Characterization ..........-0.2..2008. 95
5.5 Magnetoelastic Energy and Magnetostriction. ......... 104
5.5.1 Theory... .... 2.000. eee ee ee eee 104
5.5.2 Magnetostriction Measurements ............. 105
5.6 Magnetic Anisotropy and Strain in Thin Films......... 107
5.7 Variation of Magnetic Properties with Coherency Strain. ... 111
5.7.1 Experiment ...............2..00 0808048 111
5.7.2 Theory... .....0- 0002 0 eee ee es 116
5.8 Magnetotransport Properties of Epitaxial and Polycrystalline
Nigg9Feoo Films. 2... 2 ee 124
5.8.1 Instrumentation... .............-2.00048 124
5.8.2 Results... 2... 0.2.0.2... 00002 2 eee 125
5.9
Conclusions .........-. 0 0c ee 128
A More Details of the Instrumentation
A.1 Design and Operation of the MOKE system
A.l.1 Hardware ..........-.....
A.1.2 Software ..............0..
A.1.3 Operation and Suggestions for Improvement ......
A.2 SAXD system: Operation and Current Status
A.2.1 Operation ...........-....
A.2.2 Current Status ............
A.3 Cu Epitaxy: What Worked and What Didn’t
B Domain Walls in Thin Films
Or
xu
135
135
135
141
144
145
145
146
147
152
List of Figures
1.1
1.2
1.3
1.4
2.1
2.2
2.3
Kerr hysteresis loops for 20.0 nm thick NigopFe29/Cu/Si along
easy and hard directions obtained using magneto-optic Kerr
effect. 2. ee
Anisotropic magnetoresistance at room temperature of a poly-
crystalline NiggFe2q film deposited on glass... .........
Geometry used for measurement of anisotropic magnetoresis-
tance. 2. ee ee
Schematic of a polycrystalline thin film illustrating the defects
which can interfere with domain wall motion, and hence cause
higher He. 2... ee
The experimentally measured energy distribution of atoms
sputtered from a Cu target by 600 eV Arions[7]. .......
Schematic of UHV sputtering system with in situ magneto-
optic Kerr and reflection electron diffraction analysis. (a) Side
view and (b) top view. .. 2... ....-.... 000020004
Various configurations for observing the magneto-optic Kerr
effect. (a) Polar. (b) Longitudinal. (c) Transverse... .....
Xi
19
2.4
2.5
2.6
3.1
3.2
3.3
3.4
3.5
3.6
Variation of Kerr rotation with incident angle for (a) s and
(b) p polarizations for various NiFe film thicknesses... ....
Variation of Kerr rotation with NiFe film thickness for s po-
larization. ©... ee ee
Schematic of Kerr effect magnetometer for in situ analysis of
magnetic thin films................-.+-.--000.
(a) Small angle x-ray reflectivity of as-deposited Ta/NigoFeo9
multilayer (15 x 128 A) using Cu K, radiation. (b) Best fit ob-
tained to the experimentally observed reflectivity using optical
multilayer (OM) simulations... .............002.
Characterization of as-deposited multilayer period from satel-
lites observed in small angle x-ray spectrum. ..........
Interface roughness calculated from fit to background x-ray
reflectivity from Ta/NiggFeg9 multilayer. .........2.2..
Schematic showing the evolution of correlated interface rough-
ness in a short period multilayer. ................
Schematic of a columnar-grained thin film shown in (a) cross
section and (b) plan view for modelling of grain-boundary dif-
fusion concurrent with grain growth. ..............
Variation of first satellite intensity with cumulative anneal-
ing time at various temperatures and fit to decay of satellite
intensities using model for concurrent grain growth and grain-
boundary diffusion. ..........0.2..0 0.000 +e eee
Xiv
3.7
3.8
3.9
3.10
3.11
3.12
4.1
4.2
Variation of second and third satellite intensities vs. cumula-
tive annealing time at 375°C. Dashed lines are guides to the
Arrhenius plot of D with temperature to estimate activation
energy and preexponential for interdiffusion. ..........
(a) Bright field cross-sectional transmission electron micro-
graph of as-deposited multilayer (15 x 128 A) and (b) after
annealing at 525°C for4h. . 2... 2.2... 2. ee ee ee ee
High resolution cross-sectional transmission electron micro-
graph of NiggFeo9/Ta multilayer as deposited on Si(111).
(a) High angle x-ray diffraction spectra using Mo K, radiation
of as-deposited Ta/NiggFe29 multilayer and (b) after annealing
at 525°C for4h. 2... ee
Variation of (a) H., (b) normalized 47 M,, and (c) B,/B, for
Ta/NigoFe29 multilayer with interdiffusion lengths correspond-
ing to 4 h anneals at 300, 375, 450, and 525°C. The dashed
lines are guides to the eye... 2 2. 2 ee ee ee ee ee
X-ray diffraction scan of epitaxial Cu (001) film on Si (001)
deposited at room temperature for 6; = 30°. ..........
(a) Grazing incidence (@ < 5°) x-ray diffraction scan of epitax-
ial Cu (001) film on Si (001) along [110] Si azimuth and (b)
the RHEED patterns during film growth for [100] and [110]
Cu azimuths. ........0.0. 02.00 bee ee es
XV
53
70
4.3
4.4
4.5
4.6
4.7
4.8
5.1
5.2
5.3
The intensity profile across RHEED patterns at various stages
of Cu(001) film growth on Si(001). .. 2... 2... ee ee.
X-ray diffraction scan of (111)-textured polycrystalline Cu film
on Si (001) for 86; = 20°. 2 2 ee ee
(a) High resolution cross-sectional transmission electron mi-
crograph (XTEM), along [110] Si zone axis, of Cu film on Si.
The Si [110] and Cu [100] selected area diffraction pattern is
shown as inset. (b) Schematic showing microstructure evolu-
tion as obtained from high resolution XTEM image.......
Lower magnification high resolution cross-sectional transmis-
sion electron micrograph of Cu film illustrating the misfit strain
field extending into the Si substrate. ©... ...........
Bright field cross-sectional transmission electron micrograph
of Cu film on Si, illuminating the defects in the film and the
copper oxide. ©... 2. ee ee ee
Plan view transmission electron micrograph of epitaxial Cu
film with the diffraction pattern shown in inset. ........
Variation of (a) elastic strain and (b) misfit dislocation density
with film thickness as predicted by theory of strain relaxation
for NiggFez9 film growth on Cu(001). ..........0-..
A stressed solid with a square wave profile............
Surface lattice constants as calculated from RHEED measure-
ments of (001) NigoFezo film grown on Cu/Si(001) as a function
of film thickness... 2... ...0.0.0-202-02. 200000.
XV1
5.4
5.9
5.6
5.7
5.8
5.9
5.10
5.11
5.12
5.13
XVH
CoK, x-ray diffraction spectrum of 30 nm NigoFe2go epitaxial
film on Cu(30nm)/Si(001). 2. 2... ee ee 98
High resolution cross-sectional transmission electron micro-
graphs of epitaxial (001) NigoFe29 film on Cu/Si(001). The in-
set shows the diffraction pattern of Si substrate and NiggFeg9/Cu
films along [110] and [100] zone axes, respectively. ....... 100
Plan view dark field transmission electron micrographs of epi-
taxial NigpFe29 film on Cu/Si(001), viewed along [211] zone
axis with g=[I1l]). .. 2... ee ee ee eee 101
Schematic of the microstructure of the Cu film which can cause
increasing roughness with increasing thickness.......... 102
Atomic force microscope images of (a) epitaxial NigoFego film
deposited on Cu(50 nm)/Si(001) and (b) polycrystalline NiggFe29
film deposited on Si02/Si. .. 2.2... 22. ee ee ee 103
Schematic of methods to measure the magnetostriction of a
thin film deposited on a nonmagnetic substrate. ........ 106
Polar Kerr loop for an epitaxial NiggFe29 film 5.0 nm thick
grown on Cu/Si(001).. 2. 2 ee ee 111
Variation of H,. with film thickness for NigpFe29 deposited on
epitaxial Cu and Cu,Ni;_, seed layers, and Si02/Si....... 113
Variation of H.. with thickness for NiggFeg9 films deposited on
epitaxial Cu seed layers of different thicknesses. ........ 115
Schematic of domain wall separating two domains in a mag-
netic thin film. ........... 0.0202 cee ee eee 118
5.14
5.15
5.16
5.17
A.l
A.2
A.3
B.1
B.2
B.3
xviii
Film thickness dependence of H, computed by assuming that
strain fluctuations are caused by (a) misfit dislocations and
(b) coherent islanding. Experimentally observed H, is super-
imposed on the plot in (b)..............2.02.0.0.. 122
Variation of H, due to coherency strain-induced islanding with
NigoFeo film thickness for epitaxial growth on CugsNigs. Ex-
perimentally observed data is superimposed. .......... 123
Schematic of the setup used for magnetoresistance measure-
ments. . 2... 126
Variation of (a) resistivity and (b) magnetoresistance of poly-
crystalline NigpgFe29 film as a function of temperature for 1
hour anneals. 2... 0. ee 129
Electrical schematic of the photodetector and the amplifier.
The dashed rectangle represents the packaged photodetector
and the op-map whereas the encircled numbers represent ex-
ternal pin connections to the photodetector. .......... 141
Block diagram of the DT2821 board used for MOKE experi-
ment’s data acquisition. ...........-.2.-5200200. 142
Schematic of the Inel tube tower and the GE diffractometer
for small angle x-ray diffraction measurements.......... 148
Cross-section of Bloch and Néel walls according to the approx-
imation of Néel.. 2... 2. ee ee ee 155
Néel wall thickness vs. film thickness for < 20.0 nm thick films.155
Néel wall energy vs. film thickness for < 20.0 nm thick films. . 156
List of Tables
5.1 Magnetostriction values of epitaxial NiggFeo9 thin films on
Cu/Si(001). 2. ee 107
XIX
Chapter 1
Introduction
1.1 Introduction to Soft Magnetic Materials
Humans have been fascinated with magnetic materials since “ancient times”
when the ore magnetite was first discovered. The name of the ore, and hence
that of the whole science of magnetism, is said to be derived from the Greek
province of Magnesia in Thessaly, where magnetite was found as a natural
mineral[1]. The wide variety of magnetic materials can be sharply divided
into two categories, the magnetically soft (easy to magnetize and demagne-
tize) and the magnetically hard (hard to magnetize and demagnetize). The
former category is distinguished by high permeability and low saturating
fields, and typically find applications as core materials for transformers and
magnetic recording heads. On the other hand, the figure of merit for hard
magnetic materials is high coercivity, which is the field required to reduce
the magnetization to zero, so as to be able to resist the demagnetization
action of stray fields as well as its own. Hard magnetic materials are used as
permanent magnets and also as the medium for information storage.
Chapter 1 2
The last few decades have witnessed enormous progress in engineering
the microstructure of magnetic materials for appropriate applications. The
coercivity H, spans an enormous range varying from a few tenths of an
Oersted for 78 Permalloy (Ni7gFe22) to 24,000 Oersteds for SmCo. A famous
example of control of magnetic properties through microstructure is soft Fe-
3%Si used in tranformer cores, whose coercivity was reduced by two orders
of magnitude by control of its microstructure. In 1900, some investigators
stumbled upon the fact that the addition of 3% Si to Fe increases its electrical
resistance and reduced its coercivity, making it suitable for use in transformer
cores{1]. Further progress was made in the 1930s when it was discovered
that cold-rolling and annealing yielded (100)-textured sheets which had lower
coercivity. Even lower H, was obtained in the 1970s by having the sheets
made of large-grained (100)-oriented Fe-Si. As a consequence, there were
fewer defects in the material and its magnetic properties approached those
of single-crystal (100) Fe whiskers[2]. In soft magnetic materials, defects act
as pinning sites for domain walls, and cause higher coercivity[3].
Another extremely important soft magnetic material is Permalloy (NigoFe29)
which is used in magnetic recording heads because of its high permeability
and low coercivity. Also, because of its magnetoresistance (change in resis-
tance upon application of a magnetic field[4]) at low fields, it has been the
material of choice in magnetoresistive sensors. Some of the relevant mag-
netic and magnetotransport properties of Nig9Fe29 are reviewed in the next
section.
Chapter 1 3
1.2 Magnetic and Magnetotransport Prop-
erties of Nigg Feo
Of all the soft magnetic materials, NigoFe29, also known as Permalloy, is
probably the most widely studied. This is owing to the fact that for this
composition of Ni and Fe, the magnetocrystalline anisotropy and the mag-
netostriction are near zero[3]. Magnetocrystalline anisotropy refers to the
fact that for a a single crystal ferromagnetic material, the path to magnetic
saturation in a magnetic field is not the same along different crystallographic
directions. Thus, Ni saturates most easily along < 111 > directions whereas
for Fe, < 100 > are the easy axes. However, for NigoFe29, the magnetocrys-
talline anisotropy energy E, reduces to 3 x 10° ergs/cm* as compared to
~ 10° ergs/cm* for elemental Ni and Fe([3]. Similarly, the magnetostriction
(which is the fractional change AL/L in the dimensions of a magnetic sam-
ple when it is fully magnetized) for Ni and Fe is typically 10-°, whereas for
NiFe alloys, it is in the range 10~°, and goes to zero near the composition 81-
19[3, 5]. As a consequence, around this composition, the magnetic properties
of NiFe alloys are relatively insensitive to stress. The fundamental origin of
both magnetocrystalline anisotropy and magnetostriction is due to spin-orbit
coupling|3] (see Chapter 5 for more discussion on magnetostriction in thin
films). Hence, it is not surprising that both of them vary in an analogous
fashion with NiFe composition.
As a consequence of the above properties, Permalloy has a relatively
high permeablity (hence the name) and is easy to magnetize. Furthermore,
even though it is a random face-centered cubic substitutional alloy, it can
Chapter 1 4
possess directional order amongst nearest neighbors due to the interaction
of different magnetic moments of Ni and Fe with the external field present
during deposition[6]. Directional order implies a preferred orientation of the
axes of like-atom pairs[3]. Usually, an external field of > 20 Oe is sufficient
to induce this directional order[7]. This results in a uniaxial anisotropy such
that along one axis (easy axis) the hysteresis loop is square whereas along
an orthogonal axis (hard axis), the magnetization is almost linear and has
very little coercivity. Figure 1.1 shows typical Kerr hysteresis loops along
easy and hard axes for a 20.0 nm thick NigoFez9 film deposited on Si02/Si.
The coercivity for random polycrystalline films along hard and easy axes is
typically ~ 0.1 and ~ 1 Oe respectively, whereas the uniaxial anisotropy Hy
is ~ 5 Oe. In general, the above parameters, coercivity in particular, depend
sensitively on sample cleanliness and film deposition conditions[7].
NiFe alloys and thin films are also of interest because they exhibit anisotropic
magnetoresistance (AMR)(4] which is different from the Hall effect magne-
toresistance exhibited by most nonferromagnetic materials. The latter is
caused by the deflection of the conduction electrons or holes in a solid by
the Lorentz force due to an applied magnetic field. On the other hand, most
ferromagnetic materials have different resistivities p or p, depending upon
whether the current J is parallel or perpendicular to the magnetization M
of the material. Thus, when the magnetization vector of a ferromagnetic
material is rotated as a single domain with respect to J, there will be a
change in the voltage drop across the sample due to the change in resistivity,
Chapter 1 5
0.0015
0.0010
0.0005
LD
Kerr Rotation (Arb. Units)
fo}
oO
oO
to}
—20 —10 0 10 20
H (Oersteds)
Figure 1.1: Kerr hysteresis loops for 20.0 nm thick NiggFe29/Cu/Si along easy
and hard directions obtained using magneto-optic Kerr effect.
Ap = py — p1. This anisotropy in the resistivity can be expressed as/4]:
(0) = pi + Apcos”6 (1.1)
where 6 is the angle between J and M as illustrated by Fig. 1.3. The result-
ing change in the output voltage is proportional to Ap/p. For ferromagnetic
elements, the AMR is a few tenths of a percent whereas for NiFe and NiCo
alloys and thin films, it is a few percent at room temperature. For the com-
position 80-20, NiFe thin films have 2-3% AMR with a peak value of 5%
at 90-10 composition. The origin of AMR has to do with the anisotropy
of the Fermi surface for these alloys and the dependence of the scattering
probability of conduction s electrons on the magnetic moment orientation of
the d orbital[4]. Hence, magnetoresistance is also related to spin-orbit cou-
Chapter 1 6
pling, which may explain its high value near the NiFe composition for which
magnetocrystalline anisotropy and the magnetostriction are zero. Because of
these properties, NiggFe29 thin films are ideally suited for magnetoresistive
applications such as magnetoresistive read heads for high density magnetic
recording and detectors for bubble memories|8]. Typically, in these applica-
tions, a Permalloy-coated strip is magnetized at 45° with respect to J by a
bias field in order to maximize flux sensitiviy. Figure 1.2 shows the magne-
toresistive signal at room temperature for a NigoF ego film as its magnetization
is rotated by 180° due to a field applied along the hard axis.
More recently, even higher magnetoresistance in relatively low fields has
been reported for NiFe-based heterostructures|9] (see Chapter 5 for an intro-
duction to giant magnetoresistance (GMR) in NiFe-based multilayers). In
particular, spin valves[9], which consist of two ferromagnetic layers (typi-
cally NiFe) separated by an interlayer (typically Cu), with one of the ferro-
magnetic layers exchange-biased by an antiferromagnetic FeMn film[10], are
being considered for future magnetoresistive sensors.
1.3. Relationship to Microstructure of Thin
Films
Like other electronic properties, magnetic properties of thin films are also
very sensitive to their microstructural properties. Deposition conditions
such as substrate temperature, and Ar gas pressure in the case of sputtered
films, can significantly affect the stress in the growing film. Another possible
source of thin-film stress is the lattice mismatch with the substrate in epi-
Chapter 1 7
3.0 oo
2.5
reper,
ng
j rt
pony
on
Loe 9 ek en on ae
AR/R (4%)
Torr an eee ee ee |
H (Oersteds)
Figure 1.2: Anisotropic magnetoresistance at room temperature of a poly-
crystalline NiggFe29 film deposited on glass.
Figure 1.3: Geometry used for measurement of anisotropic magnetoresis-
tance.
Chapter 1 8
taxial growth. Now, the magnetostriction of polycrystalline NigoFe29 films
depends sensitively on its composition and crystallographic texture. If the
magnetostriction is nonzero (albeit small), it can adversely affect magnetic
properties of thin films by interacting with the strain present in the film due
to thermal or lattice mismatch with the substrate. Residual strain may be
present in the film also as a consequence of the growth process. Structural
defects in the film such as surface roughness and voids can also cause the
coercivity to increase by acting as pinning sites for domain walls. Figure 1.4
shows a schematic of a thin polycrystalline film, illustrating the defects in
the film which may act as pinning sites for domain walls. The arrows in the
figure indicate the crystallographic orientation of the grains whereas h(z) and
e(z) refer to the film thickness and film strain, respectively. Furthermore,
the uniaxial anisotropy field H; also is sensitive to annealing at relatively low
temperatures. Thus, a field of 2000 Oe maintained at 300°C for 1 hour is suf-
ficient to rotate the easy and hard directions of a uniaxial NiFe thin film|7].
This indicates that H;, depends not only on directional ordering of Ni and Fe
atoms but also on relatively mobile structural defects in the film. Further-
more, presence of nonmagnetic atoms at the grain boundaries of NigpFeg9 can
exchange decouple the magnetic grains, causing switching behaviour of the
film to be similar to that of ferromagnetic particulate arrays. This was found
to be the case for NiggFe29/Ta multilayers annealed at T > 450°C, causing
accumulation of Ta at NiFe grain boundaries (more discussion in Chapter 3).
The resistivity of ferromagnetic films can be divided into spin-dependent
and spin-independent components. The former, which is also responsible
for the magnetoresistive part in NiFe alloy films, originates due to scatter-
Chapter 1 9
OS
impurities Void €, #0 Grain Orientation
| €, #0
wa Reltecnes:.
Substrate
Figure 1.4: Schematic of a polycrystalline thin film illustrating the defects
which can interfere with domain wall motion, and hence cause higher H..
ing of conduction electrons by magnetic impurity atoms, and is relatively
insensitive to film thickness and structural defects. On the other hand, the
spin-independent component is very sensitive to the microstructure. As the
film thickness h decreases below 20.0 nm, the resistivity increases which was
explained by Fuchs[11] in terms of increased interfacial scattering due to
h being less than the mean free path of electrons. Grain boundaries can
also cause the resistivity of polycrystalline films to be higher than that of
epitaxial single-crystal films. The resistivity dependence on grain bound-
ary density is fit well by a model due to Mayadas and Shatzkes|12]. Since
the output of a magnetoresistive sensor is proportional to the ratio Ap/p,
this suggests that epitaxial films may be more suited for those applications
in which spin-independent scattering effects of grain boundaries are absent.
Chapter 1 10
Chapter 5 discusses properties of epitaxial NigoFe29 films grown on Si utiliz-
ing an epitaxial Cu seed layer. Epitaxial NigoFe29 films have been obtained
for depositions on MgO, NaCl and Cu substrates[13, 14, 15], respectively.
However, none of these substrates are available with as good atomic flatness
or are as technologically useful as Si wafers. The lattice mismatch between
Cu and NigoFeo9 is 1.85%; the strain due to this mismatch can affect the
soft magnetic properties of NiFe films with nonzero magnetostriction. The
magnetic and magnetotransport properties of these films are discussed in
Chapter 5.
Finally, giant magnetoresistance in multilayers depends extremely sensi-
tively on their microstructure, especially on interfacial roughness, crystallo-
graphic texture, and presence of voids and pinholes in the interlayers. The
interface roughness is especially more important as the thickness of the non-
magnetic interlayers is typically 1-2 nm[16]. As a consequence, different
investigators have reported drastically different GMR for multilayers grown,
for example, by sputtering as opposed to thermal evaporation[17]. Epitax-
ial growth of NigpFe29/Cu on Si suggests the possibility of growing grain
boundary-free, atomically sharp NiggFeo9/Cu multilayers, in particular epi-
taxial spin-valve heterostructures. To date, these structures have been com-
posed of polycrystalline films with possibly atomically-rough interfaces|9].
Bibliography
[1] U. Enz, “Magnetism and Magnetic Materials,” in Ferromagnetic Ma-
terials, ed. by E.P. Wohlfarth, v.3, p.3, North-Holland, Netherlands,
1982.
[2] H.J. Williams, Phys. Rev. 52, 747 (1937).
[2] B.D. Cullity, Introduction to Magnetic Materials (Addison-Wesley,
Philippines, 1972), Ch.13.
[4] T.R. McGuire and R.I. Potter, IEEE Trans. Mag. MAG-11, 1018, 1975.
[5] E. Klokholm and J.A. Aboaf, J. Appl. Phys. 52, 2474 (1981).
[6] M. Takahashi, J. Appl. Phys. 33, 1101 (1962).
[7] M. Prutton in Thin Ferromagnetic Films, Butterworth & Co., England,
1964.
[8] D.A. Thompson, L.T. Romankiw, and A.F. Mayadas, IEEE Trans. Mag.
MAG-11, 1039, 1975.
il
Chapter 1 12
[9] B. Dieny, V.S. Speriousu, B.A. Gurney, $.S.P. Parkin, D.R. Wilhoit,
K.P. Roche, S. Metin, D.T. Peterson, and S. Nadimi, J. Magn. Magn.
Mater. 93, 101 (1991).
[10] C. Tsang, N. Heiman, and K. Lee, J. Appl. Phys. 52, 2471 (1981).
[11] K. Fuchs, Proc. Cambridge Phil. Soc. 34, 100 (1938).
[12] A.F. Mayadas and M. Shatzkes, Phys. Rev. B 1, 1382 (1970).
[13] M.H. Kryder and F.B. Humphrey, J. Appl. Phys. 42, 1808 (1971).
[14] R.D. Burbank and R.D. Heidenreich, Phil. Mag. 5, 373 (1960).
[15] U. Gradmann and J. Muller, J. Appl. Phys. 39, 1379 (1968).
[16] S.S.P. Parkin, Appl. Phys. Lett. 60, 512 (1992).
[17] W.P. Egelhoff, Jr. and M.T. Kief, IEEE Trans. Mag. 28, 2742 (1992).
Chapter 2
Instrumentation for Thin Film
Deposition, Structural and
Magnetic Characterization
2.1 Introduction to Ion-Beam Sputtering for
Deposition of Thin Films
The erosion of material by energetic ions in a plasma has been a known prob-
lem for almost a century, since the invention of the light bulb. Currently,
this problem is utilized extensively in industry for etching materials as well
as for depositing thin films, and is known as sputtering. Typically, in an
industrial sputter-deposition system, a plate of the material which needs to
be deposited is connected to a negative voltage supply (dc or rf)[1]. An inert
gas, typically Ar, is introduced into the chamber to sustain a plasma. The
gas pressure is typically a few millitorr. The substrate as well as the walls
of the chamber are kept at ground potential so that the ions in the plasma
interact mainly with the target. In rf sputtering, there is some “leakage” of
13
Chapter 2 14
ions which make it to the substrate and may cause (un)desired modification
of the deposited film{1]. To control this leakage flux, the substrate can be
biased slightly. In the case of dc sputtering, there is no ion bombardment of
the substrate. However, rf sputtering has the advantage that it can be used
to sputter insulators as well because at radio frequencies, voltages can still
be coupled to them by matching impedance[1]. In magnetron sputtering, a
magnetic field is used to enhance the plasma interaction with the target sur-
face, which is particularly important for ferromagnetic materials which have
very low sputtering rates otherwise[1]. Another method of sputtering mate-
rials not used as extensively in industry, but nevertheless quite important,
is ion-beam sputtering. This method of sputtering was originally developed
in 1960 for space propulsion applications as an ion thruster[2]. Since then,
it has evolved into an extremely useful research tool for investigating sput-
tering process as well as obtaining high purity films. The ion gun essentially
consists of a discharge chamber filled with Ar gas and a tungsten filament
as a source of electrons which are accelerated to the anode. The presence
of cylindrical magnets along the walls of the discharge chamber forces the
electrons to follow helical orbits, thus increasing their ionization cross section
and creating an Ar plasma. A screened grid which has the same potential as
the cathode allows some of the ions from the plasma to escape[3]. These ions
are further accelerated by a potential of 1-2 kV while exiting, thus forming
an ion-beam which can be focused at a target some distance away. Owing
to charge repulsion, the ion-beam spreads out with distance and eventually
acquires a Gaussian profile[1]. For this reason, the ion gun should be placed
as close to the target as possible to avoid possible sputtering of the target
Chapter 2 15
holder and the walls of the chamber. A schematic of the UHV sputtering
system utilizing such an ion gun for sputter deposition is shown in Fig. 2.2.
The actual interaction of ions with a target surface which leads to ejection
of atoms is a fairly complex process|4]. For the energy range of 0.1-20 keV,
the ions generate a collision cascade beneath the target surface. In the course
of these collisions, some of the ion energy can be transferred to atoms near
the target surface, leading to their ejection. The ratio of ejected target atoms
to incoming beam ions is known as the sputtering yield. At energies higher
than 20 keV, ions penetrate deeper into the solid resulting in a decrease in
the sputtering yield. The sputtering yield is also dependent on the angle
of incidence of the ions. It increases with angle of incidence away from
normal incidence to a maximum which depends on ion and target species
and energy/1], then decreases to zero at grazing incidence angles.
The energetic ions, after penetrating the target material, lose energy via
Coulombic scattering from the atomic cores and from electronic excitations
in the solid[4]. These two types of stopping mechanisms are referred to as
nuclear and electronic stopping powers. It is the former type of collisions
which leads to ejection of atoms from the target surface, or sputtering. A
universal curve for nuclear stopping power was derived by Lindhard, Scharff
and Schiott[5]. Sigmund extended this theory to correlate the energy distri-
bution of sputtered atoms with the nuclear stopping power(6]. This theory
predicts a log-normal energy distribution with a mean energy of few eV and
a small fraction of sputtered atoms possessing energy as high as 30-60 eV.
This is found to be in good agreement with experimental results such as
those shown in Fig.2.1, which gives the energy distribution of atoms and
Chapter 2 16
gE ae All esected atoms
seosneee Monomers only
Lo.
! i 1. faut i i n
0 2 4 6 8 10 12 14°°30 40 50 60
Energy (eV)
Figure 2.1: The experimentally measured energy distribution of atoms sput-
tered from a Cu target by 600 eV Ar ions(7].
clusters ejected from a Cu target by 600 eV Ar ions[7]. By comparison, the
mean energy of evaporated atoms is of the order of a few tenths of an eV. As
a consequence, the structural and electronic properties of evaporated films
can be significantly different from those of sputtered films. Furthermore, in
sputtering, there can be significant incorporation in the film of the sputtering
gas used (up to a few %) which can alter properties of the sputtered films
as well{4]. This incorporation is higher for lighter. element noble gases such
as Ar which can occupy interstitial sites. Finally, high energy neutral atoms
of the sputtering gas specularly reflected from the target, can also bombard
the substrate during film growth. Their exact fraction depends upon the
deposition geometry and can affect properties of the films as well/4].
Chapter 2 17
2.2 UHV Sputtering System
Figure 2.2 shows the schematic of the ultra-high vacuum (UHV) sputter-
ing system used primarily for deposition of metal films in the experiments
described in this thesis. It uses a 3 cm ion-beam source manufactured by
Ion Tech, with Ar as the sputtering gas. Up to 3 different targets can be
mounted on a triangular target assembly as shown in Fig. 2.2. A 300 1/s
Balzers turbopump with a mechanical pump as the backing pump is used
for achieving UHV in the main chamber. The main chamber is interfaced
to a load lock chamber which allows samples to be loaded without break-
ing vacuum. A separate 60 1/s turbopump is used for pumping the load
lock and differentially pumping the reflection high energy electron diffrac-
tion (RHEED) gun. A base pressure of 4 x 10~!° torr can be achieved in the
main chamber, after a day-long bake-out and if system cleanliness is main-
tained properly. The Ar flow was 3.0 sccm which resulted in an Ar pressure
of 2.5 x 10-4 Torr during sputtering. The purity of the Ar gas was 99.99%
which was further improved to impurity levels of < 1 ppm by passing the
gas through a bakeable Ar filter from Ultrapure Systems. The ion beam was
typically operated at a beam voltage of 1200 V, < 100 V accelerator voltage,
and a discharge voltage of 38.0 V. The cathode current varied from 2.0-4.0
amperes depending upon the filament lifetime. See Appendix A for more
details of the instrumentation and operating parameters. The film thickness
during deposition was measured using a Inficon thickness monitor which uti-
lizes a piezoelectric quartz crystal. The film thickness thus measured was
calibrated for some samples using Rutherford backscattering spectrometry.
Chapter 2 18
The film composition for certain samples was determined after deposition
using electron probe microanalysis (EPMA).
The system is also equipped with a 15 keV Vieetech RHEED gun to mon-
itor the crystallographic texture of the growing film. See Sec.2.4 for more
details of this tool for in situ structural analysis. A horseshoe shaped electro-
magnet was placed close to the sample outside the sputtering chamber during
deposition, to induce uniaxial anisotropy in the NigoFe29 film. The electro-
| magnet was used instead of a permanent magnet inside the chamber because
the strong magnetic field due to the latter interfered with the RHEED mea-
surements. Finally, the system is also interfaced to a quartz tube which
allowed in situ measurement of magnetic properties of thin films using the
magneto-optic Kerr effect (MOKE) as described in the next section.
2.3. In Situ Magnetic Characterization using
Magneto-Optic Kerr Effect
2.3.1 Theory
The magneto-optic Kerr effect (MOKE) refers to the interaction of the po-
larization of light reflected from a magnetic thin film. This is due to the in-
equivalent interactions between the magnetization M of the sample and the
s & p polarizations of the light where s and p refer to the electric field vectors
being perpendicular and parallel to the plane of incidence, respectively. The
origin of the effect is due to the spin-orbit interaction within the magnetic
medium which gives rise to off-diagonal terms in the dielectric tensor govern-
ing the reflectivity|8, 9]. Thus, when plane polarized light containing only s
Chapter 2 19
Ar Gas
Ton Gun
Load Lock -pam le
1 .
Transfer Rod po OT Tn am Kerr
I ysis
Gate 7
Mech Valve
Pump | |
puro / Tub LI
mp / urbo
Target Pump Mech. oo
Holder Pump
(a)
To Load Lock
RHEED ]
gun
lE =] | _ sample RHEED
—- ~ screen
(b)
Figure 2.2: Schematic of UHV sputtering system with in situ magneto-optic
Kerr and reflection electron diffraction analysis. (a) Side view and (b) top
view.
Chapter 2 20
eal Ne NN
c = yu
(a) (b) (c)
Figure 2.3: Various configurations for observing the magneto-optic Kerr ef-
fect. (a) Polar. (b) Longitudinal. (c) Transverse.
or p component, is reflected from a magnetic film, it acquires a component of
polarization which was not present in the incident light. This rotation of the
polarization can be measured using a Wollaston prism which splits the light
into its two components. The MOKE measurements are usually carried out
in one of the three geometries[10] shown in Fig. 2.3; namely, longitudinal (M
parallel to the plane of incidence and in-plane), transverse (M perpendicular
to the plane of incidence and in-plane) and polar (M parallel to the plane of
incidence and perpendicular to plane of sample).
The reflection of light from a magnetic film can be expressed mathemat-
ET Pes Pep EY
s = . . $ 2.1
(a )-(e 2) le) en)
where 7'ss,sp,ps,pp ate the Fresnel reflection coefficients coupling E7, and £%_,,
ically as:
Chapter 2 21
the reflected and the incident electric field vectors for the two polarizations,
respectively. If the incident light is linearly polarized (say s) then Eq. (2.1)
ET Post
$ _ ee Q 2.2
( ES ( PepLo ( )
where E, = E‘. The rotation 6,, in s or p polarizations after reflection
reduces to:
from a ferromagnetic film, known as Kerr rotation, can also be expressed
in terms of the Fresnel reflection coefficients introduced above. Thus, for
s-polarization, the Kerr rotation is:
6, = Re. (2.3)
Tss
Kerr rotation for p-polarization can be similarly expressed. For NigoFeao,
6.» ~ 0.01° which corresponds to a change in polarization of 1 photon out
of 10*. Figure 2.4 shows the variation of 0,,, with incident angle for various
thicknesses of NiggFeo9 film, as computed from expressions for Fresnel coef-
ficients in terms of the dielectric tensor ¢;; for magneto-optic materials[10}
whereas Fig.2.5 shows the variation of Kerr rotation for s polarization with
NigoFe29 film thickness. The minimum in @, near 60° corresponds to the
strong absorption of that component near Brewster’s angle.
The two polarizations of the reflected light can be separated using a Wol-
laston prism and detected by a photodiode pair each of which gives out a
voltage signal proportional to the intensity incident upon it. Besides the
Kerr rotation, the reflected light also gets elliptically polarized due to the
fact that the s and p components are not in phase after reflection. For s
Chapter 2 22
polarization, this ellipticity, known as Kerr ellipticity, can be defined as
é, = Im_. (2.4)
This ellipticity can be eradicated by inserting a quarter-wave plate (with
proper orientation of its slow and fast axes) before the Wollaston prism|[11].
This elimination of the phase difference between the s and p components re-
sults in an enhancement of the Kerr rotation. Figure 2.6 shows the schematic
of optical components used for Kerr effect magnetometry of thin films. If I,
and Ig correspond to the intensity of the s and p components respectively,
then they can be expressed in terms of Fresnel coefficients and E, as given
by:
In =|ES|’ = F672,E0
” = Rept BP (2.5)
sp’“o°
Ip = |Ez
Using Eq. (2.3) & (2.5), it can be shown that the difference of the two in-
tensities, normalized by the total incident intensity, is related to the Kerr
rotation as follows:
7 7 2 = 2sin@,cos6, (2.6)
or for small 6,,
I, — Ip
6,. 2.7
latlp G7)
Now the dependence of 6,, on the magnetization M , although complex,
nevertheless is quasilinear as shown by the plot of 0,, vs. NigoFeso film
thickness in Fig.2.5. Hence, the difference between the output voltage signals
of the photodiode pair, V4 and Vg normalized by the sum i.e., Ane will
Chapter 2 23
vary quasilinearly with the magnetization M of the film. Thus, if this signal
is plotted against the applied field H for a ferromagnetic film, a hysteresis
loop known as the “Kerr” loop can be obtained with the ordinate being Kerr
rotation instead of the magnetization M. Using an appropriate expression
for 8, in terms of M or a table of computed values, this “Kerr” loop can be
converted to a regular hysteresis loop. Nevertheless, the coercivity H, and
the magnetic anisotropy H;, for the uniaxial magnetic film can be obtained
directly from the “Kerr” loop without conversion of 6,, to M.
2.3.2 Implementation
Magnetic properties were measured in situ using the magneto-optic Kerr
effect as described above, after sample transfer to a quartz tube interfaced to
the UHV-sputtering system as shown in Fig. 2.2. A 670 nm semiconductor
diode laser beam s-polarized by a Glan-Thompson polarizer was used as
the incident beam with the angle of incidence being approximately 70° from
sample normal. This incidence angle was chosen to exploit the maximum in 0,
as shown in Fig.2.4. Two pairs of Helmholtz coils were used to magnetize the
sample such that “Kerr” loops could be observed for longitudinal and polar
geometry. The rotation in polarization after reflection from the magnetic
sample was detected using a Wollaston prism and a photodiode pair. The
sum and difference voltage signals from the photodiodes were amplified and
acquired on a 386 computer using a Data Translation data acquisition board
(Model #2821) and customized software. The computer was also used to
control a 20V/10A Kepco power supply which provided the current to the
Helmholtz coils. The applied field H can be computed from this current, the
Chapter 2 24
__ Incident Angle
20 40 60 80
“DP ~-0.0002
on
= 0004
eT -0.
=~
— -0.0006
Ww
ap)
-0.0008
(a)
0.003}
0.002
-001
Q. (radians)
Pp
-0.001
Figure 2.4: Variation of Kerr rotation with incident angle for (a) s and (b)
p polarizations for various NiFe film thicknesses.
Chapter 2 25
0.0090 mp ay
0.0008 + _
_ L
3 L A
Cc
oO
> 0.0006 + -
.o]
Fa a
et
. fi J
a 5
0.0004 =
: 1 l L L L 1 | i 1 i 1 | er cee ree i
0.0002 50 100 150 200
Nie (A)
Figure 2.5: Variation of Kerr rotation with NiFe film thickness for s polar-
ization.
number of turns and the geometry of the Helmholtz coil. Thus, by plotting
the voltage signal from the photodiodes vs. the ac applied field H, a “Kerr”
hysteresis loop can be graphed on the computer. The driving ac frequency
was typically 1 Hz; signal-to-noise ratio was improved by averaging over more
than one cycle. The sample holder was designed such that the sample could
be rotated about its surface normal thus allowing in-plane easy and hard axis
longitudinal “Kerr” loops to be measured. Using this setup, “Kerr” loops
were obtained for NigoFeg9 films as thin as 2.0 nm. Appendix A lists some
suggestions which might improve the resolution of the system further. Figure
1.1 shows such loops obtained for easy and hard axes using this setup for a
20 nm thick NigoFe29 film on 5SiO2/Si. See Appendix A for more details of
the software and hardware employed for data acquisition.
Chapter 2 26
— Quartz Tube
all Evacuated to UHV
Helmholtz Coils | Photodiodes
‘> and Pre-amps
| |
Na
No ISN toa
| ~
670 nm
Diode Laser
Polarizer
Power Supply
for Coils
Figure 2.6: Schematic of Kerr effect magnetometer for in situ analysis of
magnetic thin films.
Chapter 2 27
2.4 In Situ Structural Analysis by Reflection
Electron Diffraction
2.4.1 Theory
The crystallographic structure of the surface of the substrate or growing film
was studied by the technique of reflection high-energy electron diffraction
(RHEED). The basis of this analysis technique is the same as that employed
by Davisson and Germer who first demonstrated diffraction of low energy
electrons from a Ni crystal[12]. The electron diffraction is a consequence of
the wave-like behaviour of the electrons and the wavelength \ of the electrons
is given by the famous de Broglie relationship. Typically, the electron beam
energy F ranges from 10 to 30 keV for RHEED applications. For these
energies, the nonrelativistic version of the de Broglie relationship can be
used, so that the electron wavelength in A is given by:
12.
y= 3 (2.8)
E(eV)
At 15 keV, which was the energy of the RHEED beam utilized in this exper-
iment, \ = 0.1 A. Thus, for grazing incidence angles, the wave-vector k of
the electrons will be comparable to the interatomic spacings in a solid and
can be used to obtain information about long range order in a single crystal.
Depending upon the RHEED geometry, the elastically scattered electrons
penetrate a very short distance into the solid[13]. As a consequence, the
diffraction pattern is determined mostly by the two-dimensional surface of
the solid and can be used to ascertain whether the crystalline surface is rough
or smooth, reconstructed or unreconstructed. In particular, when the sur-
Chapter 2 28
face is ordered and clean, i.e., has atomic steps, and the electron beam is
properly oriented with respect to the sample, its diffraction pattern consists
of discrete lines, known as Bragg rods. This condition is satisfied when
k-d=2nn (2.9)
where n is an integer, and d is the lattice vector in the plane of the surface.
Most of the qualitative features of RHEED patterns can be understood
using the kinematical theory of diffraction, just as in case of x-ray diffraction
and transmission electron microscopy|14]. However, dynamical theory of
diffraction is needed to calculate the RHEED intensity and its dependence
on the defects in the surface.
2.4.2 Implementation
A Vieetech electron gun operating at 15 keV was used to produce an electron
beam with a current of about 20 uA. The reflected beam was allowed to hit
a viewport coated with fluorescent material. The angle of incidence of the
beam was < 5°. Using the sample manipulator, the sample geometry can
be adjusted such that the desired diffraction pattern is observed for single
crystal substrate or film; typically Si substrates are viewed along [100] or
[110] azimuths. At an energy of 15 keV and an incidence angle of 5°, the
electron penetration depth is about 15 A from the surface[13]. Hence, a
-RHEED beam is extremely sensitive to the crystallographic orientation of
the surface. Furthermore, for this scattering geometry, the scattering vector
k of the electron beam is in-plane so that it is sensitive to the in-plane
crystallographic orientation and lattice constant of the film.
Chapter 2 29
During film growth, the RHEED gun was differentially pumped by a 60
1/s turbopump to allow its operation at sputtering gas pressures. For certain
film growths, the RHEED patterns were recorded using a video camera and
then frame-grabbed onto a MacII computer. Great care was taken to ensure
that the RHEED camera length did not change during these measurements,
i.e., the sample and the RHEED beam position were kept fixed. The resulting
images were analyzed using Image software (courtesy of National Institute
of Health Sciences) to find the intensity profile across the Bragg rods in
the RHEED pattern. These intensity profiles were used to calculate the
separation of the Bragg spots or the reciprocal lattice spacing K, which can
be related to the surface lattice constant a in the plane of the film or the
substrate, as given by:
Ka (2.10)
The surface lattice constant measurements thus obtained were used to get
information about the evolution of the crystallographic texture of the film
and the in-plane strain ¢ in the film, as defined by:
€ = “film ~ “substrate | (2.11)
“substrate
See Chapter 5 for strain measurements for epitaxial growth of NigoFe2o films
on Cu, and comparison with the strain relaxation as predicted by theory.
Chapter 2 30
2.5 X-ray Diffraction Studies
2.5.1 Small Angle X-ray Diffraction
The diffractometer used for small angle x-ray diffraction (SAXD) consisted
of a GE x-ray power supply and tube tower equipped with a Cu source,
6 — 20 goniometer, EG&G proportional counter and detector electronics, and
a sample holder mounted on high resolution rotation and translation stages
to allow accurate alignment of the sample with respect to the main beam (for
an introduction to the theoretical basis for SAXD, please refer to Sec.3.2).
The last addition was necessary as the higher satellites were quite sensitive
to accurate alignment of the sample and the diffractometer. A Ni filter was
placed in front of the x-ray detector to absorb Cu-Kg radiation. Later, the
GE x-ray power supply was replaced by a 2.5 kW Inel XRG-2500 power
supply and a tube tower. The new tube tower was equipped with a graphite
monochromator which led to narrower x-ray peakwidths. See Appendix A
for more details of hardware and procedures used for maximizing the SAXD
satellite intensities.
2.5.2 Large Angle X-ray Diffraction
Crystallographic texture of the thin films on Si was assessed by x-ray diffrac-
tion utilizing an Inel Thin Film Diffractometer. Most of the measurements
were performed using a Co-K, x-ray source with fixed incident beam an-
gles in the range from 5°-45° in 5° intervals, along both the [100] and [110]
azimuths of the Si(100) substrate. Scattered x-rays were measured using a
parallel detection and data acquisition instrument simultaneously over an
Chapter 2 31
angular range of 120° with 0.05° angular resolution. While not a complete
pole figure analysis, these measurements do probe a wide range of reciprocal
space, and thus provide a more comprehensive picture of film texture than
can be obtained from a single @ — 20 diffraction measurement.
2.6 Transmission Electron Microscopy
Transmission electron microscopy was performed on a Phillips EM430 elec-
tron microscope. The point resolution of the microscope is 2.3 A. Also avail-
able with the microscope are analytical instrumentation such as an energy
dispersive x-ray detector, an electron energy loss spectrometer, and a scan-
ning transmission electron microscopy unit. For cross-sectional transmission
electron microscopy (XTEM) analysis, a double-tilt goniometer was used to
align the film/substrate interface along the [110] zone axis for $i(100) sub-
strates. Plan view sample preparation was done by back-etching of the Si
substrate using HF:HNO3:H20 solution. For XTEM analysis, two 2.5 mm by
1.0 mm pieces of the samples were glued together using adhesives which did
not require high-temperature curing. These were mechanically lapped and
polished and then mounted on a 2.5 mm diameter Cu grid using epoxy. Next,
the samples were dimpled until a small hole was made near the film/substrate
interface. Finally, the samples were further thinned by ion-milling using 5
keV Art ions and inserted into the microscope immediately afterwards for
analysis.
Chapter 2 32
2.7 Additional Characterization Tools
Besides the above mentioned characterization techniques, electron probe mi-
croanalysis was used to determine the composition of the thin films deposited.
In particular, the amount of metallic impurities and Ar incorporated in the
film, as a consequence of sputter-deposition process, was estimated. The
resolution of this technique is approximately 10-20 ppm for 20.0 nm thin
films, within a spot size of 1 micron. Rutherford backscattering spectrome-
try (RBS) was also utilized for composition analysis, as well as for calibrating
the thickness and uniformity of some of the samples. Details of this tool for
thin film analysis are given in Ref[15]. Atomic force microscopy (AFM) was
also used to evaluate the surface roughness of some of the NigoFezo epitaxial
thin films on Cu/Si(100).
Bibliography
[1] J.L. Vossen and W. Kern, Thin Film Processes, (Academic Press, Or-
lando, 1978) pp. 175-206.
[2] H.R. Kaufman and P.D. Reader, Am. Rocket Soc. [Pap.] No. 1374-60
(1960).
[3] Ion Source Manual, Ion Tech, Inc., Fort Collins, Colorado.
[4] J.C. Pivin, J. Mat. Sci. 18, 1267 (1983).
[5] J. Lindhard, M. Schraff, and M.E. Schiott, Kgl. Danska Videnskab. Sel-
skab. Mat. Phys. Medd. 33, (1963).
[6] P. Sigmund, Phys. Rev. 184, 383 (1969).
[7] B.J. Garrison, N. Winograd, and D.E. Harrison, J. Chem. Phys. (1978).
[8] M.J. Freiser, IEEE Trans. Mag., MAG-4, 152 (1968).
[9] T. Yoshino and S. Tanaka, Jap. J. Appl. Phys. 5, 989 (1966).
[10] J.H. Judy, “Magneto Optics Theory,” in Proceedings, Conference on
Advances in Magnetic Recording, N.Y. Acad. of Science 189, 239 (1972).
33
Chapter 2 . 34
[11] P. Wolniansky, S. Chase, R. Rosenvold, M. Ruane, and M. Mansuripur,
J. Appl. Phys. 60, 346 (1986).
[12] C.J. Davisson and L.H. Germer, Phys. Rev. 30, 705 (1927).
(13] L.C. Feldman and J.W. Mayer, Fundamentals of Surface and Thin Film
Analysis, (Elsevier Science, New York, 1986) p. 129.
[14] B. Fultz, Diffraction Theory with Applications to X-ray Diffraction and
Electron Microscopy, to be published.
[15] W.K. Chu, J.W. Mayer, and M.A. Nicolet, Backscatiering Spectrometry,
Academic Press, Orlando, 1978.
Chapter 3
Structural Stability and
Magnetic Properties of
NiggFes9/Ta Multilayers
3.1 Motivation
Thin films of NiggFego are of considerable interest to magnetoresistive sensors
owing to their favorable soft magnetic properties such as low magnetocrys-
talline anisotropy, low magnetostriction, and high permeability. However,
these properties depend sensitively on the microstructure of the film which is
related to the film deposition conditions, as well as interfacial stability during
post-growth thermal annealing. The objective of this study was to relate the
evolution of microstructure to soft magnetic properties during thermal an-
nealing of multilayer films which were composed of sequences of NigoFe2o/Ta
bilayers. Ta was chosen as it is typically used as an encapsulation material
for NiggFe29 thin films for magnetoresistive sensor applications.
As discussed in Sec.1.2, the magnetoresistance ratio (Ap/p) of a thin film
35
Chapter 3 36
varies inversely with its resistivity o which for a polycrystalline NiggFe29 film
is affected by grain boundary and interfacial scattering[1]. Thus, achievement
of larger grain sizes compared to the mean free path for electron scattering
is desirable. One approach to reducing grain boundary scattering is by ther-
mally induced grain growth. For NigoFe2o thin films, two activation energies
for grain growth have been estimated from changes in resistivity with ther-
mal annealing; they are 0.70 +0.05 and 1.86 +0.15 eV, accounting for grain
growth in two temperature regimes([2]. However, annealing treatments have
been reported to adversely affect the soft magnetic properties of NigoFe29;
in particular, the coercivity H, increases[3]. A possible cause for the in-
crease in coercivity is the phenomenon of thermal grooving which occurs at
grain boundaries in polycrystalline films[4]. During thermal annealing, grain
boundary grooves deepen and eventually lead to the formation of voids be-
tween grains, and the transport of matter away from the grooves often takes
place via surface diffusion. However, if the polycrystalline film is encapsu-
lated, groove formation may be significantly retarded, presumably due to
inhibition of surface diffusion [5].
In multilayer films, interdiffusion into NiggFeg9 layers from adjacent films
during annealing can dramatically affect its magnetic properties. The ther-
mal stability of Ti/NigoFe2o films has been investigated using diffusion couples
and high angle x-ray diffraction[6]. However, quantitative conclusions about
kinetics of interdiffusion could not be obtained from that study because of
insensitivity of the characterization method used. In this study, we did a
detailed analysis of interdiffusion kinetics of Ta/NigoFeo9 using small angle
x-ray diffraction (SAXD) from multilayers, which is an extremely sensitive
Chapter 3 37
probe of interdiffusion at low temperatures [7]. From the Ni-Ta phase dia-
gram, it can be observed that Ta has relatively high solubility in Ni (7 at. %
at 600°C[8]) whereas solubility of Ni in Ta is much lower ( < 0.14 at. % at
600°C). Thus, some Ta solubility in NiggFe29 is anticipated from thermody-
namics upon annealing at sufficiently high temperatures. The kinetics of Ni
diffusion into Ta might be expected to be rapid as in case of Ni-Zr system
which undergoes a solid-state amorphization reaction[9]. However, at lower
temperatures and smaller grain sizes, grain-boundary diffusion, which may
not correlate at all with bulk diffusion kinetics, should dominate. The pres-
ence of Ta in Ni is known to lead to a reduction in magnetic moment (20 at.
% Ta in Ni reduces its magnetic moment to zero [10]). This suggests that the
Ta/NiggFeo9 system can be used to investigate very quantitative correlations
between interdiffusion and changes in soft magnetic properties of NiggFe2o.
Finally, SAXD has not been used as commonly for detailed analysis of
interdiffusion in polycrystalline films as for amorphous or epitaxial films[7].
In view of that, a model to enable the application of SAXD to analyze grain-
boundary diffusion with concurrent grain growth in polycrystalline multilay-
ers is discussed in Sec.3.3.
3.2 Small Angle X-ray Diffraction
3.2.1 Introduction
The wavelength of x-rays (A ~ A) makes them an ideal probe of atomic
structure in solids. However, they can also be used to study artificial struc-
tures such as periodic multilayers. The Bragg condition for diffraction from
Chapter 3 38
multilayer of period d can be written as:
kd = 2xm (3.1)
where m is an integer and k is the wave vector defined by:
k= = sind (3.2)
and @ is the incidence angle of the x-ray beam. Using Eq. (3.2.1), it can be
shown that for d ~ 100 A, the Bragg satellites corresponding to the artificial
periodic structure would be observed for 0 < @ < 5°. An example of such
a diffraction spectrum using Cu K, radiation is shown in Fig. 3.1(a) for an
as-deposited NiggFeo9/Ta multilayer of period d = 128.7 A. The character-
istic oscillations superposed on the higher satellites are probably due to an
aperiodicity in multilayer which can be observed in the cross-sectional trans-
mission electron microscope (XTEM) images of the multilayer as well. Figure
3.1(b) shows the best fit to the experimentally observed reflectivity using op-
tical multilayer (OM) theory (more details of OM theory are discussed in the
next section). The exact position of these higher satellites are obtained by
modeling the refractive indices n for individual layers in a bilayered periodic
multilayer as
n=1-—é-2i8, (3.3)
where 6 and @ represent the multiple scattering and absorption of the x-rays
in matter[12] and are given by:
Nr»? ,
6 = 50 (2 + f')
Nr»? .,,
g = Any (3.4)
20
Chapter 3 39
with N being the number density of the electrons, r. the classical radius of
the electron, Z the atomic number, and f' & f" the Honl corrections found
in standard x-ray tables(13]. Typically, 6,8 < 1 and furthermore, @ is an
order of magnitude smaller than 6. Hence, it will be neglected in subsequent
analysis. The condition for diffraction of x-rays from such a bilayered periodic
multilayer of period d = d, + dy becomes
MA = 2Zngdgsinb, + 2npd,sinOy. (3.5)
Since the difference in 6 for the individual layers is small, ¢, ~ #. Asa
result, Bragg’s law, modified for periodic multilayers, can be written as
[ma] -e <2
(3.6)
2sin0 sin’
where < 6 > represents a weighted average of the real part of the x-ray re-
fractive index in the multilayer[12]. By doing a linear fit to (m\/2sin@)? vs.
1/sin?@ for the Bragg satellites, the multilayer period d can thus be calcu-
lated. Figure 3.2 shows such a plot used to calculate a period of 128.7 A for
the as-deposited NigpFe29/Ta multilayer. For details of the instrumentation
used for SAXD, please refer to Sec.2.5.1.
3.2.2 X-ray Reflectivity From Multilayers and Inter-
face Roughness
The reflectivity of x-rays from a multilayered film can be calculated analo-
gously to that for optical coatings, utilizing the Fresnel reflection coefficients
and the refractive indices for x-rays, with the condition of continuity of the
parallel component of the electric field vector across the interfaces[14]. This
Chapter 3 40
10) et
we wees
10°b
105
Ln
Intensity (arb. units)
10*
Try rire
Tred ee a ae ro on Ol me 1
0.0 60.5 1.0 1.5 2.0 25 3.0 3.5
2© (deg)
(a)
10) ere
~—_ sl
2 10°E 3
c E 4
an) C =
Te’ .
108
2 10°F 3
= L.
bal —— a
= 'CF 3
5 m=4
40° ee a ee ee ee Ast po
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
20 (deg)
(b)
Figure 3.1: (a) Small angle x-ray reflectivity of as-deposited Ta/NigoFe2o
multilayer (15 x 128 A) using Cu K, radiation. (b) Best fit obtained to the
experimentally observed reflectivity using optical multilayer (OM) simula-
tions.
Chapter 3 41
1.8x10*
1.6
oh
aN
~_s
oO
oy
(ma/2sin0)?
0.6
OF
fe)
Ny
0.4 0.6 .
(1/sine)? — xT08
Figure 3.2: Characterization of as-deposited multilayer period from satellites
observed in small angle x-ray spectrum.
Chapter 3 42
approach to x-ray reflectivity from multilayered thin films is known as optical
multilayer (OM) theory, and is readily adaptable to computation because of
its recursive nature. One such version written by Lee Goldman and David Wu
of Harvard University([15] was used to calculate the reflectivity of Ta/NigpFe29
multilayers and compare with the experimentally observed reflectivity (see
Fig. 3.1). A comparison of ratio of satellite peak intensities was used to es-
timate the average thickness of Nig9Fe29 and Ta layers, which were found to
be 84412 and 45+9 A, respectively.
The reflectivity of x-rays from a multilayer can also be used to estimate
its interface roughness. The satellites in a periodic multilayer are superim-
posed on a monotonically decreasing reflectivity background, which depends
on its interface roughness o;. Assuming o; is a random variable (see discus-
sion at the end of this section), the decrease in x-ray reflectivity due to the
interface roughness can be modelled by a Debye-Waller factor, analogous to
the decrease in diffracted beam intensity due to thermal disorder(16], i-e.,
I(k) = I,(k)exp(—k’o?) (3.7)
where I(k) and [,(k) are background reflectivities from a rough and a per-
fectly smooth multilayer. It can be shown that these reflectivities vary as
1/k*[15] where k is the x-ray wave-vector. Thus, if log k*I(k) is plotted ver-
sus k?, the slope of the straight line fit to the background will be o?. Using
this method, the interface roughness of the Ta/NiggFeo9 multilayer was esti-
mated to be 13 A, as shown in Fig. 3.3. This was in good agreement with
the interface roughness estimated from XTEM analysis.
The above analysis is based on the assumption that the interface rough-
Chapter 3 43
ness of multilayers is uncorrelated. However, frequently in short period (<
5.0 nm) multilayers, the roughness is cumulative as illustrated by a schematic
shown in Fig.3.4. In such cases, o; cannot be treated as a random variable.
The effects of correlated interface roughness on SAXD spectrum have been
modelled by Payne and Clemens/17]. Specifically, their analysis predicts
broadening of the diffraction features due to roughness correlations. How-
ever, in our NiggFe29/Ta multilayers, the period is much larger. As a result,
correlations in the roughness are unlikely and are not evident in XTEM anal-
ysis of as-deposited multilayers. Hence, the above analysis may be justified
for estimating the interface roughness of the as-deposited multilayers.
3.2.3 Interdiffusion in Multilayers and Small Angle
X-ray Diffraction
Measuring the decay of the satellites in SAXD spectrum as a function of
annealing time is one of the most sensitive methods for calculating diffusion
coefficients at low temperatures|7]. The electron density p.(z) in a periodic
multilayer can be represented as a Fourier series:
pe(z) = >. Amsin(kmz + m) (3.8)
m=0
where k, = 2xm/d is the “reciprocal” lattice vector of the artificial periodic
multilayer. Now, for Fickian diffusion, the diffusion equation is[20]:
OC (z,t) _ po Cat)
Bi 522” (3.9)
where C'(z,t) is the time-dependent concentration profile of the multilayer
and D is the diffusion constant. The standard solution by Fourier decompo-
Chapter 3 44
3 5 a t i Li t | ay 7 t i I i ' t t { i T i t J
r i \ 7
r 1 \ _ 4
3.0 F i \ o = 11.64 -
r 1 \ _ . 1
ask ; k=4msin@/r E
r | :
— r A
@ 2.0F ~
NZ Z -
ron 7
1.06 4
0.5 F
0.0 : 1 i] l I | l j l l | I rn l 1 | 1 1 1 I :
0.000 0.005 0.010 0.015 0.020
k?(A-?)
Figure 3.3: Interface roughness calculated from fit to background x-ray re-
flectivity from Ta/NigpFe29 multilayer.
Chapter 3 45
MMU LL Ves
Figure 3.4: Schematic showing the evolution of correlated interface roughness
in a short period multilayer.
sition of the above partial differential equation yields:
C(z,t) = S> Ame *"P*sin(kmz + bm); (3.10)
m=0
Since the electronic density p.(z,t) is linearly proportional to C(z,t), the
expression for p.(z,t) would also be of the same form as Eq. (3.10). By the
kinematic theory of diffraction, the intensity of the mth satellite [,,(t) in
the SAXD spectrum will be proportional to the square of the mth Fourier
component of p.(z,t) which can be expressed mathematically as:
Tia(t) oc A? e72hmDE, (3.11)
Thus, the rate of decay of the log of satellite intensities with annealing time
would be given by:
OlnIm 822m?
Tp Dz (3.12)
Chapter 3 46
Analysis leading to Eq. (3.12) assumes Fickian diffusion which does not
include contributions to free energy from the concentration gradients at the
interface. If the period of the multilayer is small enough[21], the diffusion
constant for the multilayer may not necessarily be the same as the bulk
diffusion constant. When gradient energy contributions are important, it is
wavelength dependent as given by(21]:
Dy m= D ( + 13 , (3.13)
where « is the gradient energy coefficient, g” is the second spatial derivative
of the free energy per unit volume, and D is the bulk diffusion constant. Both
« and g” can be estimated using the regular solution model([21]. Using the
heat of mixing for the Ni/Ta and Fe/Ta systems, as calculated by Miedema
et al.[22], we found that Dy, = 0.995D for d = 100 A. Thus, Fickian
diffusion is a good approximation for the NigpFe29/Ta system for multilayers
with periods d> 100 A.
3.3 Concurrent Grain Growth and Grain Bound-
ary Diffusion
The standard analysis for interdiffusion in multilayers as discussed in the pre-
vious section is appropriate when the diffusion mechanism is dominated by
bulk lattice diffusion. However, in polycrystalline films, grain boundary dif-
fusion is generally the dominant diffusion mechanism at lower temperatures.
Consider a thin film of composition A with a columnar-grained microstruc-
ture, which is modeled for simplicity as an array of close-packed hexagons so
that the hexagon extremal cross section is twice the grain radius r and height
Chapter 3 A7
equal to film thickness d, as shown in Fig. 3.5. If the grain-boundary width
is 6, and the film of composition A has interfaces to films of composition B
on top and bottom, then the grain-boundary area in the plane of the film
per grain, defined as the grain-boundary density, to first order is 6rd/3r?.
If temperature is low enough such that lattice diffusion is negligible com-
pared to grain-boundary diffusion, then the appropriate diffusion constant D
of Eq. (3.12) is the grain-boundary diffusion constant Dj, weighted by the
grain-boundary density:
D= Dye. (3.14)
However, the analysis is further complicated by the possibility of concur-
rent grain growth which decreases the grain-boundary density. Grain growth
kinetics in thin films typically obey the semiempirical relationship[11]:
r(t) — r7(0) = ae Catm/FTE = A(T, (3.15)
where r(t) is the mean grain size at time t, r(0) is the initial mean grain size, a
is a weakly temperature-dependent constant, n is typically in the range of 2-4,
and Qgbm is the activation energy for grain boundary migration. The kinetic
models for grain growth typically imply a value of 2 for the exponent n in the
case of two-dimensional normal grain growth. However, in thin films, grain
growth is typically saturated, i.e., undergoes a rapid reduction in growth
rate as the grain size approaches the layer thickness[11]. The specific causes
for this saturation are complex and are often material-dependent but may
be significantly related to grain-boundary grooves in the columnar structure.
As a consequence, the experimental data fits better to an exponent in the
Chapter 3 48
Film B
Film A
2r
—~i<«— FilmB
y—<
Figure 3.5: Schematic of a columnar-grained thin film shown in (a) cross sec-
tion and (b) plan view for modelling of grain-boundary diffusion concurrent
with grain growth.
Chapter 3 49
range of n =3-4 at later times. Assuming that the grain-boundary density
decreases due to grain growth as given by Eq. (3.15), the appropriate diffusion
equation for this problem is:
[r"(0) + A(T)t]/" 8C(z, t) 8C(z,t)
26 at az?
Equation (3.16) can be solved using separation of variables and Fourier de-
= Dy (3.16)
composition in the spatial variable z, analogously to Eq. (3.9). The solution
can be written as:
C(z,t)= xu Am€Xp Ee ~ 7) rca + acryey4| cos(kmz +m).
(3.17)
The time dependence of the resulting mth satellite intensity [,, will then be:
2 2 2D_6, 4 1-L
. _ ame Lda nl, wl
In(t) « exp | 2Rim = “TV AT) (r™(0) + A(T)t) | (3.18)
Hence, the rate of decay of In I,,(t) becomes:
Ololin _ 167?m? Dad
Ot d r(t)-
Comparing Eq. (3.19) with Eq. (3.12), note that D has been replaced by
(3.19)
2D,46/r(t) which is a time-dependent diffusion constant.
3.4 Interdiffusion Kinetics of Polycrystalline
NiggFeo)/Ta Multilayers
Small angle x-ray diffraction was used to study interdiffusion in multilayers
which were annealed at various temperatures in the range 300 — 600°C. As
Chapter 3 50
the decay rate of the log of satellite intensity is proportional to m?, as given
by Eq. (3.12), the higher satellites are more sensitive to interdiffusion at
lower temperatures. The decay of higher satellites for short period multilay-
ers can be used to measure interdiffusion lengths of a few angstroms. Figure
3.6 shows the integrated first satellite intensity versus cumulative annealing
time at various temperatures. No change in any of the satellite intensities
was observed at 300°C. As shown in Fig. 3.6, there was no significant change
in the first satellite intensity also at 375°C. However, we were able to char-
acterize interdiffusion at this temperature from the decay of the higher order
satellites (see Fig. 3.7). As can be seen from Fig. 3.6, the satellite intensities
do not have a constant decay rate, as predicted by analysis of Ref.[7]. A bet-
ter fit to the data, superposed as dashed lines in Fig. 3.6, was obtained using
the model described above for concurrent grain growth and grain boundary
diffusion. Quantative analysis of grain growth in polycrystalline multilayer
films is complicated by the superposition of the microstructure of the indi-
vidual layers when imaged in plan view or cross-section electron microscopy.
As a result, evidence for grain growth in the temperature range 300 — 600°C
was also obtained from investigation of single layer NiggFe29 thin films which
were annealed in high vacuum and analyzed by plan view TEM. The mean
grain size in these studies was calculated by averaging over up to 100 grains
contained in bright field TEM micrographs. Although this type of grain
growth analysis is not as desirable as analysis of the multilayers themselves,
the results for grain growth in single layer films are qualitatively consistent
with observations of increased grain size in the thin regions of multilayer cross
sections as well as the grain size analysis being significantly more straight-
Chapter 3 51
T=375°C
i oe —— e 9
ae
pe Tae _._T=450°C
Lod ee ee
Zp .
7-3 - |, ~~ _
= rte ~~ ~
= fh . ~~ _T=525°C
oS \ Te
—4b | ~.- 4
\ 0,
, T=600°C
0 50 100 150 200 250
Annealing Time (min)
Figure 3.6: Variation of first satellite intensity with cumulative annealing
time at various temperatures and fit to decay of satellite intensities using
model for concurrent grain growth and grain-boundary diffusion.
forward in single layer than in multilayered thin films.
Assuming parabolic grain growth kinetics (n=2), Eq. (3.18) becomes
mo = Oa | + 2) (3.20)
Im(t) __ 16n?m? _ r?(0) f ; An -
where J, is the constant of integration chosen so that [,,(t = 0) computed
from the above equation matches the experimentally observed mth satellite
intensity at t=0. Using Eq. (3.20), a fit was done to the decay of observed
satellite intensities, as shown in Fig. 3.6, to calculate D. The initial grain
size r(0) and a were calculated from grain growth analysis of single layer
NigoFe2o films and were found to be 60.37 A and 5.24 x 1075 cm?/s, re-
spectively. Based on this analysis, the activation energy for grain boundary
migration, Qgim, for single layer NiggFe29 films was found to be 0.55 + 0.11
Chapter 3 52
“54 T T TT
~6 i 4
fo
ra —7 aN ~4
—_ r" N
= -8 aN a ee 4
PONS Second Satellite intensity” ° :
§ oN J
i Third Satellite Intensity _ 7
aT) | ee cr a
is) 20 40 60
Annealing Time (min)
Figure 3.7: Variation of second and third satellite intensities vs. cumulative
annealing time at 375°C. Dashed lines are guides to the eye.
eV. An Arrhenius plot of D, shown in Fig. 3.8, yielded the following diffusion
coefficient and activation energy:
D, = 1.70 x 10-° cm?/s Q = 1.28 + 0.26 eV.
No information on Ta lattice diffusion in NiggFeo9 or vice versa was avail-
able. However, the following diffusion constants for lattice self-diffusion in
Ni[18] and Ta[19] are known from previous studies:
Ni: D, = 1.30 cm?/s, Q; = 2.27 eV,
Ta: D, = 0.124 cm?/s, Q) = 4.2 eV.
Typically, the activation energy for grain boundary diffusion is significantly
less than that for lattice diffusion (Qj, ~ 0.4 — 0.7Q; for most metals([20]).
Chapter 3 53
Temperature (°C)
600 400
r t T T T T T T t
on
Oo
T | E T J i t i t | qv Lj 1 1 1 Li T 1 7
i l i i Ll I rT Ll i i i i lL l i i 1 I i
Log of Diffusion Constant (cm?/s)
1.4 1.6 1.8
1000/T (°K)
—_
Oo
—_
NO
Figure 3.8: Arrhenius plot of D with temperature to estimate activation
energy and preexponential for interdiffusion.
Chapter 3 54
Comparing activation energies for lattice diffusion to the experimentally ob-
served activation energy agrees with our assumption that the interdiffusion
mechanism is grain-boundary diffusion. The relatively small value of the
pre-exponential found experimentally can be explained by the fact that D
has been weighted by the grain-boundary density.
Depending upon deposition conditions, different investigators have re-
ported the as-deposited Ta film to be amorphous, bcc (a) or metastable
tetragonal (8) phase. TEM analysis of the as-deposited Ta/NiggFe29 mul-
tilayers by us suggested that the Ta film was nanocrystalline with possibly
some amorphous regions (see Fig. 3.9(a) and 3.10). However, presence of
many bcc Ta crystalline peaks in the high angle x-ray diffraction spectrum
shown in Fig. 3.11(a) suggests that the Ta film is not amorphous.
Comparison of high angle x-ray diffraction spectra of as-deposited and
annealed multilayers, shown in Fig. 3.11, indicate that the bcc Ta crystalline
peaks were broadened after a 4 h anneal at 525°C, which might suggest a
diminishing crystal size due to grain boundary mediated solid-state amor-
phization. However, the persistence of bcc Ta crystalline peaks in the x-ray
spectrum and the polycrystalline nature of the NiggFeo film, as evidenced by
the micrograph of Fig. 3.9, suggest that a solid-state amorphization reaction
may have begun but has not resulted in transformation of a significant frac-
tion of the Ta film. In this temperature regime, a solid-state amorphization
reaction (SSAR) between Ni and 6-Ta multilayers was indeed reported by
Hollanders et al.[23] for Ni/Ta multilayered films. However, in case of their
multilayers, the Ta layer had a greater thickness than the Ni layer. It is
plausible that led to a faster SSAR than in case of our Ta/Nigg9Fe29 multilay-
Chapter 3 55
ers. Solid-state amorphization typically precedes nucleation of a crystalline
intermetallic compound phase[9] whose presence was not observed in high
angle x-ray diffraction spectra after annealing the Ta/NigoFeo9 multilayers at
525°C. It is possible that some regions in the NigoFeg9 film have begun to be
amorphized due to Ta accumulation at the grain-boundaries and at higher
temperatures would have been replaced by the intermetallic compound Ni3Ta
whose presence was reported by Hollanders et al.[23].
Comparison of the bright field micrographs of an as-deposited multilayer
and another after annealing at 525°C for 4 h, as shown in Fig. 3.9, indicates
that the NigoFe29-Ta interface is roughened and more diffuse after annealing
at 525°C. No quantitative information about the diffusion constants for the
different constituents could be obtained from just this analysis. However,
evolution of magnetic properties, as discussed below, suggests that there is
accumulation of Ta at NigpFe2o grain boundaries. This is consistent with
our diffusion kinetics analysis which suggests grain boundary diffusion as
the mechanism for interdiffusion. The enhanced contrast at NiggFe29 grain
boundaries in the bright field micrographs of multilayer annealed at 525°C,
might suggest that NiggFee9 grain boundaries are Ta rich. As the magnetic
properties of the multilayer may not be as sensitive to outdiffusion of Ni
and Fe into Ta as to Ta diffusion into Ni-Fe, that possibility cannot be
ruled out. However, interdiffusion studies of Ta/Ni bilayers using Rutherford
backscattering (RBS) by Kellock et al.[24] also lends supports to the grain
boundary diffusion of Ta into Ni. In particular, they observed a Ta surface
peak in the RBS spectrum after an anneal at 350° for 32 hours, when the Ta
layer was located beneath 100 nm of Ni. This was not observed to be the case
Chapter 3 56
(b)
Figure 3.9: (a) Bright field cross-sectional transmission electron micrograph
of as-deposited multilayer (15 x 128 A) and (b) after annealing at 525°C for
4h.
Chapter 3 57
ee ne
: Scere ees : ; Fee Ra iy
ee See
pee acs a ee
oe oo
Figure 3.10: High resolution cross-sectional transmission electron micrograph
of NiggFeo9/Ta multilayer as deposited on Si(111).
58
Chapter 3
2000
ee a ee as ee a ee as wa a a ee t 8 \ i ee ae Si
L 1 r 9
r (ossy(orzor € FS i (ogs)(orzon. & 7S
: (009)'(ZP+)PL 4Io— r (009)‘(Zy)OL 43
| (OLS)oL | 9 L (OLS)OL ;
- (Zp)°41N 482 > FP (Zrv)e4iN 48
I (oee)oL | @ ~ | (oge)oL 1
i (Z2Z) OL i r ;
r 17 - (Licjewn @ 49
} (LLZ)OL : | (1iz)eL }
(i 41)e4IN a PCL L)@aiN
f Fee Se ne SO CU OT TO os eS 4 ] r
oO oO oO oO Nevandleermeh meal mos Aedhmatynece mma cemdbeored em SareBemal armdacemabeoone
8 8 iB 8 8 8
(s}iun ‘quo) Aysuazu| S) = =
(sun *qup) Aysuazuj
of as-deposited Ta/NigoFe29 multilayer and (b) after annealing at 525°C for
Figure 3.11: (a) High angle x-ray diffraction spectra using Mo K, radiation
4h.
Chapter 3 59
for the Ni peak when 100 nm of Ta was located on top, indicating that at
lower temperatures Ta diffusion (presumably due to grain-boundaries) was
higher than that of Ni.
3.5 Evolution of Magnetic Properties
B-H loop measurements were made for the as-deposited Ta/NigoFe29 mul-
tilayer after thermal annealing at 300, 375, 450 and 525°C for 4h. The
coercivity of the as-deposited multilayer was 0.5 Oe, and increased to 6.5 Oe
after annealing at 525°C. The coercivity is plotted vs interdiffusion length,
which is calculated directly from the decay of the first satellite with anneal-
ing, in Fig. 3.12(a). The 47M, measurement of the as-deposited multilayer
indicated an average magnetic thickness of 53.9 A/period, implying that
30.1 A/period was nonmagnetic. This suggests the presence of a nonmag-
netic NiggFeg9 with an average thickness of 15 A at each interface possibly
due to intermixing with Ta during the deposition process. The B-H loops
for the as-deposited multilayer were found to be isotropic, so H, could not
be measured. However, single layer NigoFez9 films with thicknesses equal to
or greater than 100 A, under the same deposition conditions, were found to
have well-defined easy and hard axes. The disappearance of induced uniaxial
anisotropy for ultrathin NiggFeo films has also been reported elsewhere[25).
After annealing at 300°C, the change in 41M, of the multilayer was less
than 0.5%. This is consistent with the interdiffusion data, as no change
in satellite intensities was observed at 300°C. At higher temperatures, 47M,
dropped sharply, as shown in Fig. 3.12(b). For multilayers annealed at 450°C
Chapter 3 60
and higher temperatures, the field required to saturate the magnetization in-
creased dramatically. A field greater than 100 Oe was required to saturate
the multilayer annealed at 525°C. This is also illustrated by the variation
of B,/B, with interdiffusion length in Fig. 3.12(c). The transition from
soft to hard magnetic properties can also be understood in terms of our
analysis of interdiffusion data, if it is assumed that accumulation of Ta at
NigoFe29 grain boundaries results in a microstructure consisting of ferromag-
netic NiggFea9 particles embedded in a nonmagnetic Ta-NigoFeo9 matrix. As
the thickness of this nonmagnetic matrix increases, the coupling between the
magnetic particles decreases, thus requiring higher saturating fields. Thus
the magnetic microstructure evolves from soft, magnetically continuous lay-
ers of NigoFego to isolated NiggFe29 nanometer-scale particles. It is interesting
to note that similar B-H loop characteristics have been reported in work on
a photolithographically defined array of micron-sized NiggFegq particles as
the interparticle spacing increases [26], although it should be pointed out
that the magnetic length scales in these two experiments are quite different.
More recently, short period Ag/NiFe multilayers were annealed to prepare
granular films of Ag with embedded NigoFeo9 particles, for magnetoresistive
applications[27]. Silver, having a positive heat of mixing with both Ni and
Fe[22], accumulated along the NigoFe29 grain-boundaries after annealing but
did not diffuse farther into the NiggFeg9 grains unlike the case of Ta/NigoFe29
multilayers. As a consequence, the magnetic moment of NiggFe29 was not re-
duced drastically and the magnetic properties remained relatively soft. The
above work and our investigations illustrate that the evolution of film mi-
crostructure can be used to modify magnetic properties at the scale of the
Chapter 3 61
grain size in NigpF ego films.
3.6 Conclusions
Using small angle x-ray diffraction, interdiffusion lengths ranging from 5
to 40 A were measured for Ta/NigoFeo9 multilayers in the temperature range
300 —600°C. The interdiffusion kinetics suggest grain boundary interdiffusion
concurrent with grain growth in NiggFee9. Evolution of soft magnetic proper-
ties is consistent with accumulation of Ta at the grain-boundaries of NiggFego.
Since soft magnetic properties of NiggFeg9 are not as sensitive to outdiffusion
of Ni or Fe into Ta, that possibility cannot be excluded. Significantly, the
change in saturation magnetization of the Ta/Nig9Fe29 multilayer upon an-
nealing at 300°C for 4 h was less than 0.5%. However, at temperatures higher
than 375°C, the saturation field H,.: increased dramatically whereas 47M,
decreased due to interdiffusion of Ta in Nigg9Feo9. Grain-boundary diffusion
of Ta into NiggFeg9 possibly led to a microstructure consisting of isolated
NiggFeo9 particles embedded in a nonmagnetic matrix, thus increasing the
field required to saturate the multilayer. The above hypothesis might be
confirmed by high resolution analytical electron microscopy techniques in
cross-section such as scanning transmission electron microscopy or electron
energy loss spectroscopy work.
Chapter 3 62
Interdiffusion Length L=[Dat]’ (4)
10 perros oer , :
ab J
; e peresy ;
3 °F i 7
oO L. f 4
= L /
5 / 4
4r ‘
[ /
2r / 4
L ve
fe
ole i lL L i
0 10 15 20 25 30
(a)
2 Lj E + i i 4
1.0phe ~~. 0 =
os ‘\ 7
Z N 7
< a \ 4
5 08F 8 1
f \ q
3 ot N j
2 O66 x 4
rs) - \ 4
E iL N 4
c . oN J
2 O4PE ane 4
<= i ~ 4
Ea Sy ~~ =
= a ~~ oe 4
= 0.25 4
oo Ut i ore a Ll pt |
10 45 20 25 30
(b)
1.0 Sree eee
0.8 be ~ J
5 “ey j
oor ‘\ «
06+ x 4
att \
x ot S 4
a r 4
0.4- a 4
L N q
bon se
0.2- ~ 4
eee Sees eee ee ee rere
0.0 10 15 20 25 30
(c)
Figure 3.12: Variation of (a) H., (b) normalized 47M,, and (c) B,/B, for
Ta/NigpFe29 multilayer with interdiffusion lengths corresponding to 4 h an-
neals at 300, 375, 450, and 525°C. The dashed lines are guides to the eye.
Bibliography
[1] A.F. Mayadas and M. Shatzkes, Phys. Rev. B 1, 1382 (1970).
[2] S. Krongelb, A. Gangulee, and G. Das, IEEE Trans. Magn. 9, 568
(1973).
[3] R.L. Anderson, A. Gangulee, and L.T. Romankiw, J. Electron. Mater.
2, 161 (1973).
[4] W.W. Mullins, J. Appl. Phys. 28, 333 (1957).
[5] D.J. Srolovitz and C.V. Thompson, Thin Solid Films 139, 133 (1986).
[6] L.G. Chow, $.K. Decker, D.J. Pocker, G.C. Pendley, and J. Papadopou-
los, IEEE Trans. Magn. 15, 1833 (1979).
(7| F. Spaepen, Mater. Res. Soc. Symp. Proc. 37, 207 (1985).
[8] P. Hansen, Constitution of Binary Alloys, (McGraw-Hill, New York,
1958), p. 1045.
[9] R.W. Johnson, C.C. Ahn, and E.R. Ratner, Phys. Rev. B 40, 8139
(1989).
63
Chapter 3 64
[10] R.M. Bozorth, Ferromagnetism, (Van Nostrand, Amsterdam, 1957), p.
823.
[11] C.V. Thompson, Annu. Rev. Mater. Sci. 20, 245 (1990).
[12] L.M. Goldman, H.A. Atwater, and F. Spaepen, Proc. Mater. Res. Soc.
160, 571 (1990).
[13] B.E. Warren, X-ray Diffraction, (Dover Publications, New York, 1990)
p. 372.
[14] J.H. Underwood and T.W. Barbee, Jr., Applied Optics 20, 3027 (1981).
[15] L.M. Goldman, Ph.D. Thesis, Harvard University, 1990.
[16] D.G. Stearns, J. Appl. Phys. 65, 491 (1989).
[17] A.P. Payne and B.M. Clemens, Phys. Rev. B 47, 2289 (1993).
[18] R. Hoffman, F. Pickus, and R. Wood, Trans. AIME 206, 483 (1956).
[19] N.L. Peterson, in Solid State Physics, edited by F. Seitz, D. Turnbull,
and H. Ehrenreich, (Academic, New York, 1968), Vol. 22.
[20] P.G. Shewmon, Diffusion in Solids, (J. Williams Book Company, Jenks,
1983), pp. 164-175.
(21) J.W. Cahn and J.E. Hilliard, J. Chem. Phys. 28, 258 (1958)
[22] A.R. Miedema, F.R. de Boer, and R. Boom, CALPHAD, (Permagon,
New York, 1977), Vol. 1, pp. 341-359.
Chapter 3 65
[23] M.A. Hollanders, C.G. Duterloo, B.J. Thijsse, and E.J. Mittemeijer, J.
Mater. Res. 6, 1862 (1991).
[24] A.J. Kellock, J.E.E. Baglin, K.R. Coffey, J.K. Howard, M.A. Parker,
and D.L. Neiman, Mater. Res. Soc. Symp. Proc. Vol. 343, in press.
[25] M. Goto, H. Tange, and T. Kamimori, J. Magn. Magn. Mater. 62, 251
(1986).
[26] J.F. Smyth, S. Schutz, D. Kern, H. Schmid, and D.Yee, J. Appl. Phys.
63, 4237 (1988).
[27] T.L. Hylton, K.R. Coffey, M.A. Parker, and J.K. Howard, Science 261,
1021 (1993).
Chapter 4
Epitaxial Growth of Cu on Si
at Room Temperature
4.1 Introduction and Motivation
The growth of single-crystal copper films on silicon is potentially of great
technological interest both as a convenient template for the preparation of
other metallic films in magnetic thin film applications, and as an interconnect
material for future large-scale integrated circuits[1]. Recently, considerable
controversy had surrounded reports of growth of epitaxial Cu films on Si.
Successful epitaxy of Cu (001) on Si (001) was reported for growth in vac-
uum systems with base pressures in the high-vacuum range[2, 3], but the
analysis of microstructure was limited to x-ray diffraction in conventional
6 — 26 geometry and ion channeling. On the other hand, surface science
investigations employing photoelectron holography[4] indicated the absence
of (001) epitaxial growth on $i(001) 2 x 1 structures under ultrahigh vacuum
‘ conditions/4]. Furthermore, there were conflicting reports of silicide forma-
tion upon deposition of Cu on Si at low temperatures (0-100°C). Walker
66
Chapter 4 67
et al.[5, 6] reported 7-Cu3Si formation upon deposition at 100°C of Cu on
Si(111) which aided the establishment of an epitaxial relationship between
Cu and Si. Ichinokawa et al.[7] observed an intermixed layer of Cu and Si
at room temperature deposition on Si(001) 2 x 1 surface. Also, Sosnowski
et al.[8], who used ionized cluster beam deposition at room temperatures,
did not find any evidence for an interfacial reaction between Cu and Si. The
kinetics of silicide phase formation in Cu-Si diffusion couples in the tem-
perature range 200 — 260°C was investigated by Hong et al[9]. Harper et
al.[10] reported an epitaxial relationship between CugSi and $i(001) which
was formed upon deposition of Cu on $i(001) at 200°C by dc magnetron
sputtering. This silicide phase catalyzed a remarkably rapid oxidation of
Si at room temperature, leading to the formation of over one micron thick
buried $i02/5i(001). The purpose of this investigation was to resolve some of
these controversies, and to show how orientation and microstructure develop
during epitaxial growth of Cu on Si (001).
4.2 Substrate Preparation and Deposition Con-
ditions
Prior to sample insertion into the vacuum system, samples were cleaned
by sequential chemical oxidation in a 5:1:1 solution of H,O:HCl:H2O2 at
80°C, followed by etching in a buffered HF solution. Upon insertion into
the sputtering system chamber, reflection high energy electron diffraction
(RHEED) was used to confirm (1 x 1) surface reconstruction of the (001) Si
that is commonly observed on HF-dipped Si (001), and which corresponds to
Chapter 4 68
dihydride termination of the surface[11]. This surface is stable in air for up
to an hour and for 10 hours or so in vacuum, depending upon the residual
gases present in vacuum. Upon heating to ~ 370°C, this surface transforms
to a monohydride (2 x 1) surface. Futhermore, the H-terminated Si(1 x 1)
surface has the same structure as bulk Si and is different from the (2 x 1)
reconstructed $i(001) surface obtained after desorbing the native oxide in situ
at 1200°C. It is plausible that the different film morphologies obtained for
Cu growth on Si by different investigators may be due to different 51 surface
structure and reconstruction prior to deposition of Cu. The as-inserted Si
(001) substrates were heated to T= 200°C for approximately two hours and
cooled to room temperature at a cooling rate of 3°C/minute, prior to sputter
deposition of Cu. This pre-deposition bake is carried out to desorb any
possible residual moisture and hydrocarbons on the surface, the effectiveness
of which is supported by reflection electron energy loss spectrometry work
done in our group[12]. See Appendix A for more details of the substrate
cleaning and deposition conditions which affected the Cu epitaxy.
Films were grown at room temperature in the ultrahigh vacuum ion-
beam sputtering system described in Chapter 2. The deposition rate was
approximately 0.10 nm/s and the background Ar pressure during sputtering
was 2 x 1074 torr. The film thickness, estimated during growth using an
oscillating-crystal thickness monitor, was confirmed afterwards with Ruther-
ford backscattering spectometry. The purity of the Cu film was confirmed
using electron microprobe analysis which indicated less than 0.2 at. % Ar
and no other metallic impurities.
Chapter 4 69
4.3 In Situ Electron Diffraction Analysis
In situ RHEED measurements indicated that the [100] and [110] in-plane
azimuths were parallel to [110] and [100] azimuths of Si, respectively, and
are shown in Fig. 4.2(b). The lattice mismatch for the orientation Cu[100]
|| Si[110] is 6% as opposed to 33% when the two unit cells are parallel.
Information about the evolution of the film microstructure during growth
was obtained using in situ RHEED measurements which were recorded on
video and later analyzed to calculate the intensity profile across the Bragg
rods (see Sec.2.4 for more details of this technique). As indicated by RHEED
measurements shown in Fig. 4.3, during growth of the first 2 nm of copper,
broad and diffuse Bragg rods corresponding to a nominally (001)-textured
Cu film were observed. For thicknesses in the range 5-10 nm, the Bragg
rod width gradually decreased whereas its intensity gradually increased with
increasing Cu thickness. No significant changes in the RHEED features were
observed for thicknesses greater than 20 nm. The change in the relative peak
intensities of the 01 and 01 satellites in Fig. 4.3 for thicknesses greater than
5 nm were due to variations in the sample alignment with respect to the
electron beam during growth.
4.4 X-ray Analysis
Crystallographic texture of the Cu films on Si was assessed by the Inel thin
film x-ray diffractometer as described in Chapter 2. Measurements were per-
formed with incident beam angles in the range from 5°-45° in 5° intervals,
along both the [100] and [110] azimuths of the Si surface. Fig. 4.1 shows
Chapter 4
pony
os
INTENSITY (Arb. Units)
uw —
oy
oO
BS
oe Ppenay
Figure 4.1:
deposited at
70
a Uryte
ry Pypony
20 40 60 80 100 120
20 (Degrees)
X-ray diffraction scan of epitaxial Cu (001) film on Si (001)
room temperature for 6; = 30°.
Chapter 4 71
10°
E ij ' 3 qT E Ey a | € 6 v if a f ¥ i 7 a é i E
on, m a
#2 105k “s
Cc = =
> FE 7
Fe} LC :
Som
Siote at ges om on NS
a = oS =i —- Ww J
a 7 = S R 5 ay
%) I SJ Ss =) > 3 |
2 7 56 6 D Oo 6 3
bul 3
E 10°p 3
<= 3 3
107 I j i | 2 i L | 3 L i | i i L i] i L n ]
20 40 60 80 100 120
26 (Degrees)
Figure 4.2: (a) Grazing incidence (6 < 5°) x-ray diffraction scan of epitaxial
Cu (001) film on Si (001) along [110] Si azimuth and (b) the RHEED patterns
during film growth for [100] and [110] Cu azimuths.
Chapter 4
01
50 A Cu/Si
01
10 A Cu/Si
72
Figure 4.3: The intensity profile across RHEED patterns at various stages of
Cu(001) film growth on Si(001).
Chapter 4 73
10 a a Ey t { 1 i F | t fi t i t F EI i E T a I 3
—~ + :
= 5 el
pm
> 10° =
a Ff all S ea:
a - = (oo) AN = N 7
< a) & x 2&2)
— i J = ~~ = 3
vam L ©O Oo © Oo Oo 3
” 2
D 10 ‘aie
Lil 7
— =
= 5
10! L k n i i in n | i i i j i L j H i i i
20 40 60 80 100 120
26 (Degrees)
Figure 4.4: X-ray diffraction scan of (111)-textured polycrystalline Cu film
on Si (001) for 6; = 20°.
Chapter 4 74
the x-ray diffraction spectrum for 6; = 30° which is equivalent to the more
conventional 9 — 26 x-ray scattering geometry. For Co Ka, 26 for Cu(200)
reflection is close to 60°. Therefore, this reflection should be much stronger
than other reflections of fec Cu, which is indeed the case, as shown in Fig. 4.1
for 0; = 30°. Other textures of fcc Cu, notably (111), are detectable though
their intensity is 2-3 orders of magnitude smaller than (200). Fig. 4.2(a)
illustrates a diffraction spectrum for grazing angle incident beam (0; < 5°)
along the [110] and [100] Si azimuths. In this geometry, the x-ray scatter-
ing vector is close to being in the plane of the sample, and hence, sensitive
to the in-plane orientation of the film texture. For near-zero incident angle
and Cu[100] azimuth, a strong Cu(220) reflection should be observed from
an epitaxial Cu(001) film for Co-K, radiation. As shown in Fig. 4.2(a), a
strong Cu(220) reflection is indeed observed for this azimuth whereas for
5i[100] azimuth, the reflection completely disappears, thus confirming the
orientation Cu[100] || Si[110]. No reflections corresponding to copper sili-
cide phases were detected for films deposited at room temperature. Room
temperature-deposited Cu(001) films with very strong texture (i.e., com-
peting texture fraction < 0.005) exhibited no silicide formation even after
annealing at 120°C for 2 hours. This is in disagreement with the kinetics of
silicide formation reported by Hong et al.[9], as based on their kinetics for
silicide growth, the thickness of the silicide would be 180 nm after annealing
at 120°C for 1 hour. However, the kinetics they reported are for polycrys-
talline Si-Cu thin film couples. It is plausible that a kinetic barrier for silicide
nucleation exists in well-oriented Cu(001) films on $i(001). Furthermore, a
silicide was formed immediately upon deposition of Cu at 80°C. This was ev-
Chapter 4 75
idenced by the absence of a Cu(001) RHEED pattern, the film being unlike
in appearance to that of Cu, and identification of CugSi peaks in its x-ray
diffraction spectrum. This further supports the the possibility of a kinetic
barrier for well-oriented Cu(001)/Si(001) films and that this barrier is either
absent or smaller during the initial stages of film growth. For certain Cu
films where trace metallic impurities such as Fe, Cr and Ni were detected by
electron microprobe analysis in amounts 0.1-2%, polycrystalline films with
strong (111) texture were obtained as indicated by x-ray diffraction spectrum
of one such sample for 6; = 20° shown in Fig. 4.4. Cross-sectional transmis-
sion electron microscopy studies of these films revealed an interfacial silicide
less than 1 nm thick separating the columnar grains of (111)-oriented Cu
film and Si substrate. In addition, Cu,Ni,_, alloyed films grown directly on
Si, with x=0.45 had a similiar (111)-textured columnar microstructure. It is
plausible that the presence of impurities such as Ni, Fe and Cr catalyzes a
thin interfacial silicide formation, which then promotes (111)-textured poly-
crystalline film growth. Furthermore, in another experiment a 50 nm thick
Cu film was annealed at 230°C until it was completely consumed to form
CugSi as evidenced by RHEED. This was followed by deposition of another
50 nm of Cu film at room temperature. The film subsequently deposited
was found to be polycrystalline with strong (111) texture. The above in-
vestigations suggest that if the Cu film growth proceeds with the formation
of an interfacial silicide due to presence of impurities or due to deposition
at higher temperatures, then (001) epitaxial growth might be inhibited. It
is possible that some of the investigators did not observe Cu epitaxy on Si
due to one of the above reasons, in particular, possibly because typically in
Chapter 4 76
vacuum systems, the true substrate temperature is not accurately measured
due to bad thermal contact.
4.5 Electron Microscopy Analysis
Transmission electron microscopy was done using a Philips EM430 micro-
scope operating at 300 keV, as described earlier and sample preparation was
done using adhesives which did not require any high-temperature curing.
The samples were ion-milled using Art at 77 K, 10 minutes prior to in-
sertion into the microscope. A high resolution cross-sectional transmission
electron micrograph of the Cu/Si interface along the [110] zone axis of Si,
for room temperature-deposited sample, is shown in Fig. 4.5(a). The inset
shows the [110] and [100] selected area diffraction pattern due to Si and Cu,
respectively. The spots due to Cu are broader than those due to $1 and have
a mosaic spread of +5° about Si [110] zone axis. Furthermore, the interface
is atomically sharp with no evidence of any interfacial silicide. Interestingly,
Fig. 4.5(a) illustrates that grains with orientations other than (001) can be
observed within the first 5 nm of Cu film, which is in agreement with x-ray
analysis and RHEED observations. These grains are occluded during the
early stages of Cu growth by grains of (001) orientation, eventually leading
to a single crystal Cu(001) film after about 20 nm. This evolution of mi-
crostructure by occlusion is summarized in a schematic diagram (Fig. 4.5(b))
showing the distribution of orientations as obtained from the cross-sectional
transmission electron micrograph shown in Fig. 4.5(a).
A nearly periodic strain field can be observed extending into the first 2-3
Chapter 4 77
nm of Cu film as well as the Si substrate, as can be seen in the lower magni-
fication cross-sectional electron micrograph shown in Fig. 4.6. The period of
this strain field (= 3 nm) agrees well with the periodicity of misfit dislocations
corresponding to the 6% mismatch between Cu and Si for Cu[100] || Si[110].
Furthermore, this illustrates that some of the misfit strain is accomodated
in the substrate as well, inconsistent with the assumption often made in the
coherent thin film growth analysis that the substrate is perfectly rigid[13].
Extension of misfit strain fields into the substrate have also been observed for
coherent island growth of Ge on Si[14]. See Chapter 5 for a more elaborate
discussion of coherent film growth. It is possible that each strained region
near the interface corresponds to a growing Cu island at the early stages of
growth and these islands coalesced to form a columnar-grained film. Bright
field XTEM analysis indicated a high density of defects in the film, as well
as a polycrystalline copper oxide up to 20 nm thick at the Cu film surface, as
shown in Fig. 4.7. As a consequence, when the Cu film is viewed in plan view
(see Fig. 4.8), it appears polycrystalline due to the copper oxide grains. The
faint circular rings due to the polycystalline copper oxide and the diffraction
spots due to the Cu film can also be seen in the diffraction pattern shown in
the inset of Fig. 4.8.
4.6 Conclusions
In conclusion, we have defined conditions for which Cu(001) epitaxy is possi-
ble on H-terminated $i(001) in ultra-high vacuum conditions. The RHEED,
x-ray diffraction and electron microscopy results suggest that the mechanism
Chapter 4 78
"is
Be GB BS,
Ba Re
Figure 4.5: (a) High resolution cross-sectional transmission electron micro-
graph (XTEM), along [110] Si zone axis, of Cu film on Si. The Si [110] and
Cu [100] selected area diffraction pattern is shown as inset. (b) Schematic
showing microstructure evolution as obtained from high resolution XTEM
image.
Chapter 4 19
Figure 4.6: Lower magnification high resolution cross-sectional transmission
electron micrograph of Cu film illustrating the misfit strain field extending
into the Si substrate.
ES
Chapter 4
icrograph of
ide.
ion electron m
in the film and the copper ox
iss
ional transm
sect
ing the defects
nn
wm
fe)
he
oO
‘od
2 &
oO
OA
"6
o &
he “J
3°46
00
m ©
Chapter 4
Figure 4.8: Plan view t
with the diffraction pattern shown in inset.
ransmission electron micrograph of epitaxial C
u film
Chapter 4 82
for orientation development is an occlusion of grains misoriented with re-
spect to (001) grains, eventually leading to a reduction in the mosiac spread
of the columnar-grained structure. No evidence suggesting an interfacial sili-
cide was found for room temperature (001) Cu film growth. However, for
(111)-textured polycrystalline Cu films, an interfacial silicide was observed
which inhibited Cu(001) epitaxy on 5i(001). It is possible that a nucleation
barrier for silicide formation exists for Cu(001) films grown epitaxially on
Si(001) which might explain the absence of a silicide in these films, contrary
to kinetics of silicide formation reported by other investigators[1, 9].
Bibliography
[1] J. Li, Y. Shacham-Diamand and J.W. Mayer, Mater. Sci. Rep. 9, 1
(1992).
[2] C.A. Chang, Appl. Phys. Lett. 55, 2754 (1989).
[3] J. Li and Y. Shacham-Diamand, J. Electrochem. Soc. 139, L37 (1992).
[4] B.P. Tonner, J. Zhang, X. Chen, Z-L. Han, G.R. Harp, and D.K. Saldin,
J. Vac. Sci. Technol. B10, 2082 (1992).
[5] F.J. Walker, E.D. Specht, and R.A. McKee, Phys. Rev. Lett. 67, 2818
(1991).
[6] F.J. Walker, J.R. Conner, and R.A. McKee, Mat. Res. Soc. Symp. Proc.
187, 249 (1990).
[7] T. Ichinokawa, T. Inoue, H. Izumi and Y. Sakai, Surface. Sci. 241 416
(1991).
[8] M. Sosnowski, H. Usui and I. Yamada, J. Vac. Sci. Technol. A&, 1470
(1990).
83
Chapter 4 84
[9] S.Q. Hong, O.M. Comrie, S.W. Russell and J.W. Mayer, J. Appl. Phys.
70 3655 (1991).
[10] J.M.E. Harper, A. Charai, L. Stolt, F.M. d’Heurle, and P.M. Fryer, Mat.
Res. Soc. Symp. Proc. Vol. 187, 107 (1990).
[11] J.A. Schaefer, Physica B170, 45 (1991).
[12] S. Nikzad, $.S. Wong, C.C. Ahn, A.L. Smith, and H.A. Atwater, Appl.
Phys. Lett. 63, 1414 (1993).
[13] J.W. Matthews and A.E. Blakeslee, J. Cryst. Growth. 27, 118 (1974).
[14] C.J. Tsai, Ph.D. Thesis, California Institute of Technology, 1992.
Chapter 5
Structural, Magnetic and
Magnetotransport Properties
of Epitaxial NiggFeo9 Films on
Cu/Si
5.1 Introduction
Permalloy (NigoFe29) heterostructures are of great interest for magnetore-
sistive devices based on anisotropic magnetoresistance (AMR) and recently
discovered “giant” magnetoresistance (GMR) in magnetic multilayers[1, 2].
In particular, all investigations to date of magnetic multilayers which exhibit
GMR have employed polycrystalline films with atomically rough interfaces
where the interface roughness is of the same order as the nonmagnetic spacer
layer thickness[2, 3]. Giant magnetoresistance in magnetic multilayers was
first observed for Fe/Cr multilayers for Fe layers < 100 A and Cr layers
< 50 A[i]. The Fe layers were antiferromagnetically(AF)-coupled with the
coupling strength depending sensitively on the Cr interlayer thickness. As a
85
Chapter 5 86
function of Cr thickness, this AF-coupling exhibited an oscillatory behaviour.
The cause of this anomalous coupling and the oscillations is a source of great
interest and debate[4]. Furthermore, when the magnetizations of the AF-
coupled Fe layers are made parallel by application of an external field, a
significant change in the resistance of the multilayer occurs, which is referred
to as GMR. Since then, both AF-coupling and GMR have been observed for
many magnetic multilayered systems including Co/Cu[5] and NigoFe29/Cul[2].
The AF-coupling can neither be explained in terms of exchange interactions,
as the exchange interaction length (typically a few A) is much smaller than
the interlayer thicknesses, nor in terms of magnetostatic interactions, as the
dipolar coupling is much weaker than that observed experimentally. A com-
plete explanation of AF-coupling remains a challenge to theorists. Currently,
the two major contenders are (i) Ruderman-Kittel-Kasuya-Yosida (RKKY)
coupling aligning the antiparallel magnetic moments via conduction electrons
and (ii) quantum-well states near the fermi level, arising from the periodicity
of the multilayer[4]. On the other hand, the most common mechanism for
explanation of GMR is a spin-dependent interfacial scattering of conduction
electrons in the ferromagnetic layers of the multilayer|[6, 7]. These theories
assume that nonmagnetic metal atoms dissolved in the ferromagnetic lay-
ers near the interfaces act as spin-dependent impurity scatterers which are
known to contribute to magnetoresistance of bulk ferromagnetic materials|8].
This suggests that the magnetotransport properties of atomically-abrupt epi-
taxial grain-boundary free films should more clearly elucidate the underlying
physics(9]. Furthermore, with atomically-abrupt film interfaces, it would be
possible to fabricate thinner nonmagnetic spacer layers without the interface
Chapter 5 . 87
roughness being the limiting factor[2]. Growth of epitaxial multilayers of
NiFe/Cu as well as other heterostructures such as spin-valves for studying
GMR was one of the motivations for work described in this chapter.
Furthermore, in the last decade, a great deal of attention has been di-
rected to the study of coherency strain and its effects on the electronic and
mechanical properties of strained multilayers, or superlattices. However, the
growth of epitaxial magnetic films and the effects of coherency strain on
magnetic properties, especially anisotropy and magnetic moment, have only
recently attracted attention[10]. In particular, magnetic properties of epi-
taxial Fe and Ni thin films on Cu(001) have been studied recently[11, 12]
whereas heteroepitaxy of metal films has been of interest since 1970 when
Matthews and Crawford[13] studied strain relaxation in epitaxial Ni(001)
films on Cu(001).
In this chapter, we report structural, magnetic and magnetotransport
properties of epitaxial NigpFe29(001) films grown on Cu(001) as well as on
nearly lattice-matched CugsNigs films, 10-50 nm thick, oriented epitaxially
with respect to $i(001). Permalloy (NigoFe29) has a fee structure with a lat-
tice parameter of 3.549 A, which has a misfit of -1.85% with that of Cu and
-0.2% with CugsNigs. This is assuming that Cu and Ni form a solid solution
(even though Cu and Ni have a miscibility gap at room temperature[14], it
is unlikely that elemental segregation takes place, because film growth by
sputtering is a process, which occurs far from thermodynamic equilibrium).
The NiggFe29 as well as the Cu3sNigs films had the same crystallographic ori-
entation as the Cu seed layers, as verified by RHEED and x-ray diffraction.
The epitaxy of Cu(001) at room temperature on H-terminated $i(001) is dis-
Chapter 5 88
cussed in Chapter 4. Epitaxial NiFe films have been grown in the past as well
by researchers on MgO, NaCl and Cu substrates[15, 16, 17]. However, high
quality substrates of Cu, MgO and NaCl are not as cheaply available as high
quality Si substrates, and also have higher dislocation densities. This sug-
gests that magnetoresistive Permalloy devices can be integrated with other
electronic devices grown on Si. The growth of NigoFe29 and Ni-rich Cu, Nij_2
films directly on H-terminated S$i(001) resulted in highly (111)-textured poly-
crystalline films. This suggests the possibility of controlling crystallographic
texture in these multilayers and studying AF-coupling and GMR in multi-
layers as a function of texture[18, 19].
Surface lattice constant variations of NiggFe29 as a function of thickness
were also studied during its epitaxial growth using RHEED, as explained in
Sec.2.4. See the next two sections for comparison of experimental results
with the theory of strain relaxation in epitaxial films[13]. X-ray diffrac-
tion, atomic force microscopy (AFM) and transmission electron microscopy
(TEM) were also utilized for ex situ structural characterization of these films.
Magnetostriction measurements were performed to estimate the magnetoe-
lastic anisotropy in these films as a consequence of coherency strain. Mag-
netic properties of these films were measured in situ as a function of film
thickness and were correlated with coherency strain. Finally, magnetic and
magnetoresistive properties of these epitaxial films were compared with those
of polycrystalline films deposited on 5102/Si.
Chapter 5 89
5.2 Strain Relaxation during Heteroepitax-
ial Growth of Thin Films
It is well known that for growth of an overlayer B on substrate A and for
small misfit f (< 10%) between the lattice constants of A and B, the lattice
of B may expand or contract to form a coherent interface with that of A.
However, formation of such an interface is associated with strain energy in
the overlayer B. Using linear elasticity theory, it follows that the energy per
unit area tco, due to coherency strain ¢ in a film of thickness h, is{20, 21):
eon = 2 (F—~) he? (5.1)
l-yvy
where uw and v are the shear modulus and the Poisson’s ratio for the overlayer,
respectively. For h greater than a critical thickness h,, it becomes energeti-
cally more favorable to relieve this coherency strain energy by generation of
misfit dislocations at the interface. As a result, the misfit is shared between
elastic strain €, and misfit dislocations with density Pmg and Burger’s vector
b. If the dislocation makes an angle a with the plane of the interface, then
the component which relieves misfit strain is b = bcosa, so that
f = €+ Pmabcosa. (5.2)
Using Eq. (5.1) and (5.2), ucon can be expressed in terms of pma as
=] CF = pray) (5.3)
l-v
Ucoh = 2
On the other hand, the energy per unit area ugis. required to create misfit
dislocations of density pma, is given by[22]:
a ub? [1 — vcos?B 4h
Chapter 5 90
where @ is the angle between the Burger’s vector and the dislocation line.
The equilibrium misfit dislocation density for h > h, is found by minimizing
the total energy tro: = Ucoh + Uist With respect to pmq using Eq. (5.3) and
(5.4), ie., by setting
OUtot
= 0. 5.5
OPmd ( )
This condition yields
_ 2 h
Oma f 1 - VCOS | in (=) (5.6)
_ bcosa 8 hcos2a l+uv
An expression for elastic strain as a function of film thickness can be obtained
as well using Eq. (5.2) and (5.6):
h < he
<={ a [2=4S02"4] In (4) heh, 67)
8rhCOSa l+v
The variation of misfit dislocation density and elastic strain as given by
the above equations is plotted in Fig. (5.1) for epitaxial growth of NigoFeao
on Cu(001). The critical epitaxial thickness h, can be obtained by setting
Pmd = 0 in Eq.(5.6), giving:
i) 1 — vcos*B 4h,
he = 8x fcosa | l+v | In( 5 ) (5.8)
which can be solved transcendentally to calculate h, for a given misfit f.
Substituting the lattice constant of NigoFe29 for 6, a = 8B = 60° which is often
the case for fcc crystals, v of 0.31 for Ni (as elastic properties of NiggFe29 are
very similar to those of Ni[23]) and misfit f for NiggFe29/Cu, h, was found to
be 4.1 nm by solving Eq. (5.8). On the other hand, for NigopFe29 growth on
CugsNigs for which the misfit is -0.2%, Eq. (5.8) yielded a critical thickness
of 62.8 nm. As mentioned earlier, the accommodation of misfit for epitaxial
Chapter 5 91
growth of Ni on single crystal Cu(001) substrate was investigated in 1970 by
Matthews and Crawford[13], and more recently by Inglefield et al.[12]. In
both studies, the strain relaxation observed experimentally agreed well with
that predicted by the above analysis. Strain relaxation measurements for
epitaxial growth of NiggFe29 on Cu/Si(001) is discussed in next section.
The above analysis assumes that there are no thermodynamic barriers
to nucleation of misfit dislocations which is a reasonable assumption for
metals[13]. However, if the film has not attained thermal equilibrium, resid-
ual strain may persist in the film for thicknesses greater than h,. In the
case of NiggFe29 growth on Cu/Si(001) films, it is likely that threading dis-
locations, which can give rise to strain-relieving misfit dislocations at the
heterointerface by kink formation, are already present in the Cu film due to
the mismatch between Cu and Si.
Furthermore, besides misfit dislocations, other mechanisms for strain re-
lief in epitaxial thin films include formation of inclusions[24] or coherent
islanding|25|. In particular, the latter mechanism has been observed ex-
perimentally in case of epitaxial growth of SiGe alloys on 5i[26] and InGaAs
on GaAs/27]. XTEM and AFM observations by these investigators clearly
show a sinusoidal surface profile along with oscillatory strain contrast in both
the film and the substrate. This phenomenon is explained well by a stability
analysis of stressed solid surfaces([25]. It is shown that small surface undula-
tions in a band of wavelengths A, < A < Amaz will grow exponentially, where
Ac and Amaz are the critical and the maximum wavelengths, respectively for
coherent islanding to occur. A crude estimate of the critical wavelength can
be made by comparing the change in energy AF associated with a transition
Chapter 5
92
0.020
0.015
w 0.010
0.005
re a a ee ea a ee a ae a aa
0.000
| ee ee ee ee ee ee
0.10
°o
0.02
0.00
ae ee es ee Oe ee
aie
emt
fo)
h (nm)
(b)
Figure 5.1: Variation of (a) elastic strain and (b) misfit dislocation density
with film thickness as predicted by theory of strain relaxation for NigoFe29
film growth on Cu(001).
Chapter 5 93
from a flat surface to a rough surface with wavelength A and amplitude h, as
shown in Fig.5.2 assuming that the stress in the interior of the square protru-
sion is zero[25]. The change in energy would then be the energy associated
with the increased surface area minus the strain energy gained i.e.,
of hA
where o} is the elastic stress, Ysurz is the surface energy, and Y is the modulus
of elasticity. Setting AZ = 0 in above equation gives an expression for the
critical wavelength:
BY sur fp V
ve = . .
=? (5.10)
Substituting Ysu.7 = 1800 erg/cm* from experimentally reported values|28]
and o| corresponding to a fully coherent NiggFeg9 film on Cu, the critical
wavelength for this case was estimated to be 93.6 nm. It is plausible that
with increasing film thickness, these coherent islands coalesce, giving rise to
misfit dislocations at their boundaries, which further relieve coherency strain.
This is supported by XTEM analysis of Cu films on Si (see Fig.4.6), which
shows periodic strain modulations in the substrate as well as the film.
5.3 Strain Relaxation Measurements using
Electron Diffraction
Strain measurements were made using surface lattice constants calculated
from electron diffraction patterns as discussed in Sec.2.4. Surface lattice
constant measurements using RHEED are shown in Fig. 5.3 for growth of
NigoFeo9 films on Cu/Si(001). For these studies, the Cu film was at least
Chapter 5 94
Figure 5.2: A stressed solid with a square wave profile.
50 nm thick to ensure that it had relaxed to its bulk lattice parameter,
and did not have any coherency strain present due to its mismatch with Si.
The RHEED pattern from the H-terminated Si was used to calibrate the
camera length for the RHEED geometry and thus convert the separation be-
tween RHEED spots to surface lattice constants for NiggFeg9 and Cu. During
the course of film growth, great precautions were taken to ensure that the
RHEED beam did not move and thus change the camera constant. The
large error bars for these surface lattice constant measurements are mainly
due to the limited resolution of the video camera used for recording RHEED
patterns. Typically, for an electron energy of 15 keV, the separation be-
tween the electron diffraction spots due to an epitaxial Cu(001) film along
[100] azimuth was = 3 cm. Thus, for NiggFeo film growth on Cu which has a
mismatch of 1.85%, the change in the spot positions would be 0.05 cm. How-
Chapter 5 95
ever, the resolution of the video camera used for recording RHEED patterns
was = 10 cm/600 pixels = 0.017 cm/pixel assuming it had 600 pixels over
a screen length of 10 cm. As a consequence, it was difficult to accurately
calculate the strain in the NigoFe9 film using this technique. Nevertheless, it
can be observed from these studies that the NiggFeg9 is coherent with the Cu
film up to 2 nm and the film relaxes to its bulk lattice constant after 4 nm,
which is in reasonable agreement with the critical thickness predicted by the
strain relaxation theory of Matthews|13] as discussed in the previous section.
The resolution of this technique might be improved further by using a higher
resolution camera such as a CCD array or a carefully apertured analog pho-
todiode for recording RHEED patterns. The other techniques for measuring
the thin film strain are based on wafer curvature and x-ray diffraction. How-
ever, it is difficult to implement these techniques for in situ measurements.
In particular, the latter can be implemented in a practical manner, for in
situ measurements only if access to synchrotron radiation during thin film
growth is possible[29].
5.4 Structural Characterization
The tools employed for structural characterization of these films were in situ
RHEED and ez situ x-ray diffraction and TEM analysis. AFM was also used
to study surface roughness of < 5 nm epitaxial NigoFeo9 films on Cu/Si. No
change in the RHEED intensity profiles was observed during epitaxial growth,
suggesting that the NigoFe2o film had the same crystallographic structure as
the Cu film on $i(001). Furthermore, no change in the Bragg rod width
Chapter 5 96
3.65 Ly 1 ¢ T LJ | t ¥ qT | v
3.55 PO eee dE. : Quan
Surface Lattice Constant(&)
Bs 0 Cae ee ee ae eT ee EN LON
16] 2 4 6 8 10
Nnire (nm)
Figure 5.3: Surface lattice constants as calculated from RHEED measure-
ments of (001) NigoFez9 film grown on Cu/$i(001) as a function of film thick-
ness
Chapter 5 97
was detectable implying that there was not significant roughening during
this epitaxial growth. However, for the geometry used, the RHEED beam
penetration into the film is about 2.0 nm. Thus, if NiggFe29 film growth
occurs via an islanding mechanism below that thickness, it would not be
detectable using the present RHEED configuration.
Large angle x-ray diffraction using the Inel thin film diffractometer de-
scribed in Sec.2.5.2, was used to confirm the crystallographic orientation of
the film just as in the case of Cu films on Si. Figure 5.4 shows a diffrac-
tion spectrum for an x-ray beam incidence angle of 30° for a 30 nm thick
NigoFeoo film deposited on Cu(30 nm)/Si(001) which indicates that the (100)
texture in the films is 2-3 orders of magnitude more pronounced than other
fee textures, notably (111). Furthermore, the lattice constants calculated
from the diffraction spectrum confirm that the Cu and NigoFe29 films are
strain-relieved; i.e., have the bulk lattice constants.
TEM studies were employed to obtain further information about the na-
ture of epitaxy and the sharpness of Cu/NiFe interface. Figure 5.5 shows
a high resolution XTEM image of a 50 nm thick epitaxial NigoFe29 film on
Cu(50nm)/Si(001), along the [110] Si zone axis. In this image, lattice fringes
extending from the Cu/Si interface to the NigoFe29 film surface can be seen.
The inset shows the selected area diffraction pattern due to (001) NigpFeo9/Cu
films and the Si substrate. Furthermore, a mosiac spread of up to + 6° in the
(001) texture can be observed in the lattice fringes as well as the diffraction
pattern due to the film just as in the case of epitaxial Cu films on Si, discussed
in Chapter 4. The NigoFeo9/Cu interface cannot be clearly seen in Fig. 5.5
due to lack of Z-contrast between NiggFeo9 and Cu. However, it is discernible
Chapter 5 98
Sa
E ~e E
: Sa 4
. NS 7
So Si
‘7c 10%R ee oe ee ee Og
cs F- 1
i. = . es
5 > J
Lid +
— 2 re Ore, Cire
B 10
41
10
40 50 60 70 80
20 (Degrees)
Figure 5.4: Cok x-ray diffraction spectrum of 30 nm NigoFea9 epitaxial film
on Cu(30nm)/Si(001).
Chapter 5 99
by the presence of Moiré fringes and misfit dislocations, which indicate a
semicoherent interface between Cu and NiggFeo9. From the atomic sharp-
ness of Cu/Si interfaces shown in XTEM pictures in the previous chapter, it
might be expected that the NiFe/Cu interface is sharp as well. Furthermore,
from XTEM analysis, the surface roughness of the NiggFegq films deposited
on thin Cu seed layers (10-50 nm) was found to be < 1.0 nm whereas for
much thicker Cu seed layers (320 nm) the rms roughness was 9.3 nm, with
a wavelength of about 200 nm. It is possible that this increased roughness
for thicker Cu seed layers is due to the mosiac spread in the epitaxy and
the columnar structure of the Cu film, which can give to a microstructure as
illustrated in Fig.5.7.
Plan view dark field TEM of the NiFe/Cu film is shown in Fig. 5.6 for
which g and the zone axis were along [111] and [211], respectively. The
diffraction pattern is shown in the inset. The micrograph suggests that the
film is polycrystalline; however, slight tilting of the goniometer reveals rectan-
gular domain-like structures which change from strongly diffracting to weakly
diffracting and vice versa. This suggests that not all the regions of the film
diffract strongly for the same orientation because of the mosiac spread in the
(001) epitaxy.
AFM measurements were made of < 5.0 nm thick polycrystalline NiggFea9
films grown on 5i02/Si and epitaxial films grown on 10-50 nm thick Cu
seed layers on Si. A H-terminated $i(001) sample was used to calibrate the
instrumental resolution of the AFM and was found to have a rms roughness
of less than 0.2 nm. For surface roughness measurements, the scan size was
typically 300 nm x 300 nm and scans were made of many different areas of the
100
Chapter 5
Figure 5.5: High resolution cross-sectional transmission electron micrographs
of epitaxial (001) NigoFez9 film on Cu/$i(001). The inset shows the diffraction
pattern of Si substrate and Nig9Fe29/Cu films along [110] and [100] zone axes,
respectively.
Chapter 5 101
Figure 5.6: Plan view dark field transmission electron micrographs of epitax-
ial NigoF eo film on Cu/Si(001), viewed along [211] zone axis with g =[111].
Chapter 5 102
OU ONIFE OR
Cu
Si
Figure 5.7: Schematic of the microstructure of the Cu film which can cause
increasing roughness with increasing thickness.
sample. The rms roughness for the epitaxial films was 0.215 nm and 0.568 nm
for Cu underlayers 10 nm and 50 thick. On the other hand, the polycrystalline
films of the same thickness had a rms roughness of 0.118 nm (see Fig.5.8).
Furthermore, the wavelength of the roughness for the epitaxial films was
much larger than that for the polycrystalline films, as can be observed from
Fig.5.8. The larger roughness of the epitaxial films may be due to kinetic
roughening, which is a consequence of film growth by an island mechanism for
Cu on Si and NigoFe29 on Cu, as discussed in Sec.5.2. This kinetic roughening
has been observed for systems with comparable misfit, e.g., growth of SiGe
alloys on $i[26]. In particular, the larger surface roughness for thicker Cu
underlayers might be a consequence of the columnar structure of the Cu film
as discussed earlier and illustrated in Fig.5.7.
Chapter 5 103
Figure 5.8: Atomic force microscope images of (a) epitaxial NiggFego film de-
posited on Cu(50 nm)/Si(001) and (b) polycrystalline NigoFe29 film deposited
on 5i02/Si.
Chapter 5 104
5.5 Magnetoelastic Energy and Magnetostric-
tion
5.5.1 Theory
Magnetoelastic energy arises from the interaction between the magnetization
and the mechanical strain of the lattice. In particular, it is a consequénce of
magnetocrystalline anisotropy of ferromagnetic materials. When magnetized
along a certain crystallographic direction, the crystal lattice will sponta-
neously deform to lower the magnetic anisotropy energy. The elastic energy
associated with this lattice deformation is referred to as magnetoelastic en-
ergy. Mathematically, the dependence of the magnetic anisotropy energy K
on strain components e,; is expressed in terms of a Taylor expansion|[30]:
OK
Kime = K° + 30 ej +. (5.11)
sy O45
where ge are related to magnetoelastic coupling coefficients, denoted by
B,;, and the direction cosines a; of the magnetization vector M. For cubic
lattices, there are only two magnetoelastic constants, B, and Bo, such that
the above equation becomes:
3 1 3 3
Kme = K°+ 8B; > ejay + 3 Pe S> Ss; C4704; +... (5.12)
i=l tl j= 1,742
The magnetoelastic coefficients are often defined in terms of experimentally
more accessible saturation magnetostriction coefficients, A113; and 2490/30].
While the magnetoelastic coefficients have the dimensions of stress, the mag-
netostriction coefficients have the dimensions of strain and are related to B,;
and By via the elastic moduli ¢;;[30]. For single crystal Ni, A113 = —24.x 107°
Chapter 5 105
and A199 = —46 x 10~® whereas for Fe, they are +21 x 107 and —21 x 107,
respectively. In the case of random polycrystalline samples, the magnetostric-
tion is well represented by a single coefficient ,(31]:
2 3
As = 5100 + Buu (5.13)
For NigoFe29, the magnetostriction is an order of magnitude smaller than
that for either Ni or Fe (A, = 0 near the composition NigiFei9), and further-
more, it is isotropic, i.e., Aino & A111. Nevertheless, its exact value is rather
sensitive to film composition, crystallographic texture, and method of film
preparation|32].
5.5.2 Magnetostriction Measurements
The common techniques employed for measurement of magnetostriction can
be divided into two categories, the direct and the indirect methods. The
latter involves application of a known stress and measuring the resulting
change in the anisotropy field H,. This change in H;, is related to the change
in the applied tensile stress AS, by/[33]:
Xs
AH, = 357-45. (5.14)
The direct method measures the small bending of the thin film upon applica-
tion of a magnetic field. Assuming that the magnetic film is thin compared
to the substrate, the deflection d and the local slope ¢ at the measurement
point at a distance / from the origin O (see Fig. 5.9) are given by[33]:
3t ,l2é
— 2d 6tslé
P= Tp
Xs
(5.15)
Chapter 5 106
Direct Measurement
Bending Cantilever
Indirect Measurement
Stress-induced Anisotrophy (Ar, )
Figure 5.9: Schematic of methods to measure the magnetostriction of a thin
film deposited on a nonmagnetic substrate.
Here t;,¢, are the thicknesses of the film and the substrates, respectively and
€ is defined as:
¥,(1 + vy)’
where Y;, Y,, and vy,v, are Young’s modulus and Poisson’s ratio of the film
(5.16)
and the substrate, respectively.
A magnetostriction measurement tool at IBM Almaden Research Center
based on the direct method described above was used to measure 4, for epi-
taxial NigoFeo9 films of various thicknesses on Cu/Si. The sample dimensions
were 35 x 10 cm and a relatively thin Si wafer of thickness 300 + 254:m was
used as the substrate. The following elastic constants for Si and NiggFeoo
were substituted in Eq.(5.15) to obtain A,:
Y; = 2.14 x 10?dynes/cm?, vy = 0.31,
Chapter 5 107
Thin Film Structure As
NiFe(7.5 nm)/Cu(50nm)/Si(001) | 1.32 x 10-° £ 10%
NiFe(15 nm)/Cu(50nm)/Si(001) | 0.62 x 10-°+7%
NiFe(20 nm)/Cu(50nm)/Si(001) | 0.72 x 10°°+7%
Table 5.1: Magnetostriction values of epitaxial NigpFeog thin films on
Cu/Si(001).
Y, = 1.67 x 10dynes/cm’, v, = 0.25.
The values of A, thus obtained for NiFe film thicknesses of 7.5, 15.0, and 20.0
nm are listed in Table 5.1. The uncertainty of 1.5x 107’ in the measurements
results from noise and the ultimate resolution of the tool. These values of
A; agree well with those reported by Klokholm and Aboaf[32] for NiFe thin
films as a function of composition. As mentioned earlier, 1, = 0 for composi-
tion 81-19 whereas for increasing (decreasing) Fe content, , increases above
(decreases below) zero. An increase in A, for film thicknesses less than 10.0
nm has been observed by other investigators as well, including O. Song et
al.[23] and is attributed to an enhanced effect of the interfaces for ultrathin
films.
5.6 Magnetic Anisotropy and Strain in Thin
Films
Using the definition of magnetostriction introduced in the last section, we
evaluate below the effect of strains on magnetic properties, especially mag-
netic anisotropy of thin films for two important cases: (i) uniaxial strain due
Chapter 5 108
to an applied uniaxial stress, and (ii) biaxial strain in coherent epitaxial thin
films. In both cases, we assume that the magnetostriction is isotropic. If an
applied uniaxial in-plane stress o makes an angle @ with the magnetization M
(also in-plane) then the magnetoelastic contribution follows from Eq. (5.12):
Kme = Becos*@ + const = —5A,0c0s%8 + const (5.17)
where the definition B = —3),Y has been used, with Y and e being the
modulus of elasticity and the strain induced due to the applied stress, re-
spectively. From Eq. (5.17), it follows that an applied uniaxial stress can
lead to a uniaxial anisotropy similar to the induced uniaxial anisotropy in
NiFe alloys.
For biaxial coherency strain es, = éyy = 1, there will be an out-of-plane
Poisson’s strain due to the distortion of the lattice[21], given by
l+v
l—-v
el). (5.18)
€, = €2z =
If M makes an angle ¢ with the normal to the film plane, and using the
above strain components, Eq. (5.12) becomes
2v
Kme = Beas sin? 3 cos*g| (5.19)
For NigoFe2o as well as Ni, v = 0.31, so that 24 & 1. Using this approxima-
l-v
tion, Eq. (5.19) reduces to
Kme = 2Beysin’¢ (5.20)
or in terms of magnetostriction,
Kime = —35Y eysin’¢. (5.21)
Chapter 5 109
The above equation implies a coherency strain-induced uniaxial anisotropy
out of the film plane whereas for in-plane magnetization, there is an isotropic
contribution given by —3\,Ye). Substituting e; = f, the misfit for NiggFe29
between Cu, the elastic modulus of Ni for that of NiggFe9, and the measured
value of A, for NigoFeo9, the magnetoelastic energy Eme contribution due to
coherency strain was estimated to be 5.2 x 10* ergs/cm*. On the other hand,
the magnetoelastic contribution due to coherency strain is reduced to 8.0x 10°
ergs/cm? in the case of NiggFeo9 growth on CugsNigs. This can be compared
with other sources of magnetic anisotropy in NiggFegp, i.e., the induced uni-
axial anisotropy energy which for NiggFego is typically 1 x 104 ergs/cm* and
magnetocrystalline anisotropy which is 2.7 x 10° ergs/cm?. The fact that the
magnetoelastic energy associated with coherency strain is significantly higher
than other sources of magnetic anisotropy for coherent films of NigoFe29, sug-
gests that it would be an important factor in governing magnetic anisotropy.
Indeed, it is well known that a perpendicular anisotropy can be induced
in ultrathin films due to interaction with strain[34], and this is utilized in
magneto-optic media of the future employing Pt/Co multilayers[35].
Besides the magnetoelastic contribution to the perpendicular anisotropy
in thin films as discussed above, another contribution comes from the re-
duced symmetry at the interfaces of a thin film and is known as the surface
anisotropy term[34, 36]. This term has the dimensions of energy per unit
area, and when multiplied by the film thickness has the same dimensions as
the other anisotropy energies. In the following discussion, we will assume
that the film is cubic < 100 > so that there is no magnetocrystalline contri-
bution to the perpendicular anisotropy. For magnetization M of a thin film
Chapter 5 110
normal to its plane, there is an associated demagnetization energy given by
27 M?, as the magnetostatic energy is much lower if M lies in the plane of
the film. Based on the above discussion, we can define an effective anisotropy
energy Eezy:
2E sur
Ee ss = Eme + et + Edemag; (5.22)
or employing Eq.(5.21):
2 sur
Bess = —33e¥ + f _ on M?. (5.23)
The above equation has been found to be in close agreement with ex-
periments for epitaxially grown ultrathin films of fcc Fe[11] and Ni films on
Cu(001)[12]. In the case of fec Fe which has a misfit of 0.9% with Cu, the
effective anisotropy switched from out-of-plane to in-plane at a thickness of
1.2 nm[11] whereas in the case of Ni growth on Cu which has a misfit of
2.6%, this transition occurred at a thickness of 6.0 nm[12]. The higher tran-
sition thickness for Ni is probably due to the combination of demagnetization
energy for Ni, being 10 times smaller than that for Fe, and the magnetoelas-
tic contribution for Ni being positive (because of A199 being negative), thus
favoring out-of-plane magnetization. Using Eq. (5.23), the thickness below
which the magnetization would be out-of-plane for NiggFe29 was estimated
to be 1.0 nm. As this thickness was beyond the resolution of 2.0 nm for our
Kerr system, this could not be verified. However, a polar Kerr loop in which
the sample is magnetized normal to its plane, for 5.0 nm thickness, is shown
in Fig. 5.10. The linear shape of the loop suggests that the magnetization M
for the 5.0 nm film was in-plane and rotated to out-of-plane upon applica-
tion of a large magnetic field. On the other hand, the longitudinal Kerr loops
Chapter 5 111
Kerr Rotation (Arb. Units)
Oo
| ee a oe
L i 2. ] L .3 d. i. oe L
~—40 —20 ie) 20 40
H (Oersteds)
Figure 5.10: Polar Kerr loop for an epitaxial NigoFe29 film 5.0 nm thick grown
on Cu/Si(001).
for in-plane application of a magnetic field were square along the easy axes
for all thicknesses greater than 2.0 nm. This suggests that if perpendicular
magnetization for thin NiggFeg9 films on Cu does occur, it must for less than
2.0 nm. The magnetic properties reported in the following section are those
observed using the longitudinal Kerr effect for in-plane magnetization.
5.7 Variation of Magnetic Properties with
Coherency Strain
5.7.1 Experiment
In this section, we report magnetic properties of 2.0-20 nm thick epitaxial
and polycrystalline Nig9Fe29 films, measured using in situ MOKE magne-
Chapter 5 112
tometry. Most of these films were deposited in the presence of an external
magnetic field, and thus had a uniaxial anisotropy as a consequence. The
magnetic properties measured were the coercivity along easy and hard axes
H.. and H,»,, respectively, and the uniaxial anisotropy field H,. In particular,
the variation of H.,. with epitaxial film thickness was studied. Magnetization
along this axis takes place by domain wall motion and hence, is sensitive to
defects in the film which can act as pinning sites for domain walls. The coer-
civity for films < 10 nm was found to be as high as 15 Oersteds whereas for
greater thicknesess, it decreased to less than 2 Oersteds (see Fig. 5.11). This
coercivity is unusually high for NigoFeo9, since for random polycrystalline
films deposited under the same conditions on $iO2/Si, H.. was between 1-2
Oersteds. However, the thicker epitaxial films on Cu/Si, had relatively soft
magnetic properties (H..=2.04 Oc, H.,=0.22 Oe, and H,=12.3 Oe for 20
nm thick film), which were comparable to those of random polycrystalline
NigoFe2o films of the same thickness. The higher H, for the thinner films sug-
gests the domain wall motion is impeded, possibly due to coherency strain-
related defects in the film. Theoretical models for interaction of domain wall
motion with defects in the film are discussed in the next section.
To further investigate the dependence of magnetic properties on coherency
strain, NiggFee9 was grown on nearly lattice matched alloys of Cu,Nij_z. For
the composition 35% Cu (65% Ni), CuNi is nonmagnetic and has a lattice
mismatch of only 0.2% with NiggFeo9. The coercivity for these NiggFe29 films
was much smaller than the ones grown directly on Cu, as shown in Fig. 5.11.
This further suggests that coherency strain-related defects may be responsi-
ble for the higher coercivity in epitaxial films deposited on Cu/Si.
Chapter 5 113
25 i Lj 1 Li t I} T T EJ a | t UJ a rt { Vi LJ t t | t .
TO Epitaxial NiggFeo9/Cu(50nm)/Si J
20 L tEpitaxial NiggFeao/CussNigs(25nm) /Cu(25nm)/SL
r A Polyerystalline NiggFe29/Si02/Si /
@ 15 ° 5
o i fo) .
e 1
1 10+ fo) a
* oO ad
5 LP re) _
- A oO 4
L + °°
r * + m
0 i 1 i n | i I l I { i i 1 i | n l t i T i
0 5 10 15 20
hyire(nm)
Figure 5.11: Variation of H.. with film thickness for NiggFeo9 deposited on
epitaxial Cu and Cu,Ni,_, seed layers, and SiO2/Si.
Chapter 5 114
Furthermore, H.. was also found to be correlated with the thickness of the
Cu seed layer. Figure 5.12 shows H,, for various Cu seed layer thicknesses.
As can be seen from this figure, H,. was higher for 50 and 310 nm thick
Cu underlayers whereas it was lower for 10 nm thick underlayer. This is
consistent with an increasing surface roughness of Cu film with thickness as
suggested by by AFM and XTEM analysis discussed in Sec. 5.4. The cause
of this increasing surface roughness is probably due to the mosaic spread
in the Cu grains and the subsequent columnar structure of these grains as
illustrated by the schematic shown in Fig. 5.7 . This suggests that surface
roughness can also impede domain wall motion and hence increase H,¢.
For some thinner films for which no external magnetic field was used
during deposition, a biaxial anisotropy of = 7 Oe was observed, and con-
firmed by vibrating sample magnetometer measurements. For the biaxial
anisotropy, the easy axis was along the [110] crystallographic direction of
NigoFegq whereas the hard axis was parallel to the [100] direction. At first,
it might seem plausible that this biaxial anisotropy is due to crystalline
anisotropy of NigoFey, as Ky is zero only for the composition 73-27(31].
Dispersion in induced uniaxial anisotropy of epitaxial NiFe films due to its
crystallographic orientation with NaCl substrates was observed by investi-
gators at Caltech as early as 1967(37]. Furthermore, Ky has a negative
value for greater Fe contents which would lead to magnetically hard be-
haviour along [100] crystallographic directions. However, for 80-20 composi-
tion, K, = 2.7 x 103, implying an anisotropy field of 1.65 Oe which is four
times smaller than that experimentally observed. This biaxial anisotropy
was not observed when an external field was used during deposition and was
Chapter 5 115
25 a t q t ' I T UJ qT t 1 qT T i i i é Lj T qT i T 4
[ =O Epitaxial NiggFe2,/Cu(50nm)/Si J
20; A Epitaxial NiggFeg/Cu(10nm)/Si a
[ =—s + Epitaxial NiggFeg9/Cu(300nm) /Si
[ ° :
@ ISP 7
& ~ o + _
e J
x 10F fe) a
5 Oo wd
5b re) =
i O i
rp °°
¢) i i } I L | I lL I 1 | L y i I | L i i i | i 7
0 5 10 15 20
hyire(nm)
Figure 5.12: Variation of H,. with thickness for NiggFe29 films deposited on
epitaxial Cu seed layers of different thicknesses.
Chapter 5 116
replaced by a uniaxial anisotropy. It is possible that an anisotropic growth
morphology during the early stages of epitaxial film growth led to an en-
hancement of the magnetocrystalline anisotropy as also observed by some
other investigators[38].
5.7.2 Theory
Coercivity in magnetic films results from the interaction of domain walls
with defects and variations in anisotropy and exchange constants in a soft
magnetic thin film. Vast quantities of experimental data for variation of H,
for NiFe films with thickness and deposition parameters exist [38]. Some of the
earlier theoretical models to explain the experimentally observed behaviour
were made by Néel[39] and Middelhoek/40]. In this section, we extend these
models to explain effects of coherency strain on H,. We start with a domain
wall of length | separating two domains in a film of thickness h, as shown in
Fig.5.13. If under the action of an applied field H, the domain wall moves
by distance dz, then the work done by the field is 2H M,lhdz. If y is the
domain wall surface energy, the change in energy due to inhomogeneities in
the film would be d(yhl). Now, wall motion occurs when the applied field
equals H., in accordance with definition of coercivity. Energy balance then
implies that
2H.M,lhdz = d(vyhl) (5.24)
or
— 1 adyhi)
He = Sapth de
Chapter 5 117
1 Id dh dl
dy, ydh | y dl
9M, |\dz | hdz | Idz|° (5.25)
The three terms on the right-hand side of the above equation correspond
to variation of coercivity due to changes in wall energy possibly caused by
anisotropy energy variations, film thickness variations possibly due to surface
roughness, and domain wall length variations, e.g., due to pinholes in the film,
respectively. The spatial derivative of the wall energy can be expressed in
terms of spatial variations of anisotropy K as follows:
dy __dydK
dz dK dz’ (5.26)
The dependence of domain wall energy 7 as well as of domain wall thickness
6 on anisotropy K is discussed in Appendix B. Using Eq.(B.6), it can be
shown that ;
rg = 56. (5.27)
Now, as discussed in Sec.5.5, the magnetoelastic anisotropy contribution
for epitaxial NigoFe29 films is much higher than other anisotropy contribu-
tions. Hence, we can assume that the anisotropy variations are due to the
strain fluctuations in an epitaxial film caused by coherent islanding and/or
misfit dislocation arrays as discussed in Sec.5.2. As these fluctuations are
periodic, we can write:
Kine = Kp. (1 + csin +...) (5.28)
where ¢;(< 1) and , are the amplitudes and wavelengths of the various
perturbations in anisotropy, respectively. Considering only the first harmonic
Chapter 5 118
ZZ ———p> (|x 1 om
Figure 5.13: Schematic of domain wall separating two domains in a magnetic
thin film.
at this point, Eq.(5.26) can be written as:
d 1 2 2
oY _ ~€,6K? a cos (=) . (5.29)
Thus, the barrier due to strain fluctuations which the applied field will have
to overcome, can be written as:
strain 3 é 2n
where Eq.(5.21) has been utilized to express Kf, in terms of \,, Y and ey.
If these strain fluctuations are caused by the presence of misfit dislocations
to relieve misfit strain, then we can substitute 4; = 1/pma where Pma is
the misfit dislocation density calculated and plotted in Fig.5.1 for epitaxial
growth of NigoFeo9 on Cu. Assuming that the strain fluctuations are due to
the strain fields of the misfit dislocations and are small except near the core
Chapter 5 119
of the dislocations, we choose €, = 0.1% somewhat arbitrarily. Substituting
the values of \,, Y and M, for NigoFe2o, e and pmg as predicted by theory
of strain relaxation, and the variation of wall thickness 5 as calculated in
Appendix B, we obtain the dependence of H, on film thickness as plotted
in Fig.5.14(a). The coercivity in that plot initially increases to a maximum
with the nucleation of misfit dislocations after the critical thickness is ex-
ceeded to a maximum value of 16 Oersteds. However, even though the misfit
dislocation density keeps increasing with thickness, the coercivity decreases
after reaching the maximum because of its dependence on the elastic strain
and the domain wall thickness, both of which decrease with increasing film
thickness. No such increase in coercivity for NiggFe29 for growth on Cu/Si,
due to nucleation of misfit dislocations was observed. One possible reason
may be that the domain wall thickness is much greater than misfit dislo-
cation period in the film thickness range 1-20 nm. As shown in Appendix
B, the domain wall thickness varies from 1.4 wm to 0.6 wm whereas misfit
dislocation density pmg as calculated in Fig.5.1 implies a misfit dislocation
period of ~ 5 nm after the initial stages of strain relaxation. This suggests
that the domain wall motion may not be sensitive to the strain fluctuations
due to misfit dislocations.
Next, we evaluate the effect of surface roughness on coercivity. There
are many factors which can lead to surface roughness in thin films. Here we
examine effects of surface roughness caused by coherent island formation to
relieve the misfit strain in thin epitaxial films. Even though we have not di-
rectly observed the presence of coherent islands using RHEED for epitaxial
growth of NiggFe29 on Cu, AFM measurements -indicating greater surface
Chapter 5 120
roughness of thinner epitaxial NigoFezo films than polycrystalline ones- sug-
gest that islanding occurs in the early stages of film growth. Furthermore,
it has been shown to occur in systems with comparable misfit in the initial
stages of epitaxial growth|26, 27]. Assuming that, coherent islanding leads
to a sinusoidal surface profile h(x) defined by
h(x) = ho + Asin (=) (5.31)
where A and are the amplitude and wavelength of the fluctuations in the
thickness of the film due to islanding, respectively, and h, is the average
thickness of the film. As demonstrated in Eq. (5.25), H, depends on the
spatial derivative of the film thickness fluctuations, i.e., it depends on the
wavelength of these fluctuations as well. Hence, from the above equation, it
follows that:
(5.32)
= —Acos (——
dx r
The relevant macroscopic quantity in esitmating H, corresponding to these
dh 2n ( ae)
microscopic variations, would be the rms spatial average of the above equa-
(Z) _ ee (5.33)
Substituting this expression for < dh/dz > in Eq. (5.25), we obtain:
tion, 1.e.,
y A 2x
Hrough _ —
‘ Mh J2 Od
(5.34)
The dependence of Néel wall energy y on film thickness is calculated in
Appendix B, whereas the wavelength of coherent islanding was shown to
scale as 8E surf /o; as shown in Sec. 5.2. For film thickness in the range
Chapter 5 121
of 0 < h < 20 nm, the Néel wall energy increases almost linearly with film
thickness and can be approximated as 7h M? as shown in Appendix B (also
see Fig. B.3). With this approximation, Eq. (5.34) becomes:
a2 M,Agj mn? M,Aey
Hrwh = ; 5.35
. 8/2 Ysur f¥ 8/2 Yeur¢ (1 _ vy)? ( )
where use has been made of the relation o) = el from linear elasticity the-
ory. Substituting A = 0.5 nm as typically indicated by AFM measurements,
M, = 800 emu/cm? and = 100 nm as the coherent islanding wavelength
for NigoFep film growth on Cu, in Eq. (5.35), the contribution to H, due to
surface roughening was estimated to be:
prouh we 70-5 X 800 w 58 Oe, (5.36)
e = 2 100
A more accurate estimate of HT™“9" was made, taking into account the exact
value of 7 as calculated in Appendix B and is plotted in Fig. 5.14(b) along
with the experimentally observed coercivity. The reasonably good agreement
between experiment and theory suggests that this mechanism is more likely
to be responsible for the higher H, observed experimentally. This roughen-
ing due to coherent islanding might be suppressed by decreasing the misfit
between NiggFe29 and the underlayer such as CugsNigs, as discussed in the
last section. Figure 5.15 shows the variation of H?9" as calculated for misfit
between CuzsNigs and NigoFeo9 with experimentally observed values super-
imposed.
Chapter 5 122
20 nt
15
aes 2 ae ee a es ne
oO
o Experimental Data for NiFe/C
— Theory
Ln A
rs
fe]
potttottrtrpritiir lisii fii]
Oo
Oo eo a
an
Lo)
~s
or
wl
Oo
So
oO
Figure 5.14: Film thickness dependence of H. computed by assuming that
strain fluctuations are caused by (a) misfit dislocations and (b) coherent
islanding. Experimentally observed H, is superimposed on the plot in (b).
Chapter 5 123
YO ay
j o Experimental Data for NiFe/CuNi ]
Sr 8 Theory 7
6b J
wo f 1
x 4P :
f° 1
2 ° 4
r ° o]
OF =
5 10 15 20
h (nm)
Figure 5.15: Variation of H, due to coherency strain-induced islanding with
NigoFe29 film thickness for epitaxial growth on CugsNigs. Experimentally
observed data is superimposed.
Chapter 5 124
5.8 Magnetotransport Properties of Epitax-
ial and Polycrystalline NiggFeo) Films
5.8.1 Instrumentation
The sheet resistance as well as magnetoresistance measurements were made
at room temperature using a four point probe geometry. This setup was
assembled by Dr. Hyun Sung Joo and its schematic is shown in Fig. 5.16.
The probe spacing was 1/16 inch and corrections were made to account for
the finite sample size in calculating the sheet resistance. These measure-
ments were calibrated using a 250 nm thick Pt film deposited on $iO2/Si
whose sheet resistance was independently measured with a microprobe. The
current source consisted of a 25 V/1 A power supply with a current stabil-
ity of ~ 10 nA. The voltage was measured using the data acquisition board
installed in a model 386 personal computer which was also used for MOKE
data acquisition. For magnetoresistance measurements, a Helmholtz coil pair
consisting of 250 turns and coil separation of 12 cm was utilized to generate a
magnetic field up to 100 Oe at the sample. Precautions were taken to ensure
that there were no components of the four point probe which were magnetic.
The power supply used for MOKE Helmholtz coils was used for sweeping the
magnetic field for magnetoresistance measurements as well. The sweeping
frequency was kept below 0.1 Hz to ensure that there were no eddy cur-
rents in the sample. Magnetoresistance measurements were typically made
parallel, perpendicular and at 45° to the easy axis of the sample. Data acqui-
sition software similar to the one used for MOKE measurements was used to
measure the voltage drop across the sample as a function of magnetic field.
Chapter 5 125
Figure 1.2 shows a plot of the magnetoresistance measurements made with
this tool for a NigpFe29 film deposited on SiO2/Si(001). Magnetoresistance
of some samples was also measured at IBM Almaden using a similar setup.
5.8.2 Results
Analogously to the resistivity of thin films, the magnetoresistance of thin
films is very sensitive to structural properties as well as purity of the film.
Furthermore, NigoFeo9 gets oxidized readily upon exposure to air. To prevent
that, the NiggFeoo films for magnetoresistance measurements were capped
with 2 nm of Ta or 1 nm of Cu. The resistivity of the as-deposited poly-
crystalline capped NigoFe29 films with thickness > 10-100 nm was typically
25-40 w0-cm whereas the magnetoresistance ratio (MR=AR/R) varied from
0.5-1.5%. The resistivity and MR of bulk NiggFe29 are 14 wQ-cm and 2.1%,
respectively. The higher resistivity and lower MR for the polycrystalline films
may be due to the incorporation of Ar in the film as a consequence of the
sputtering process. Annealing of these films can lower the resistivity and
enhance the MRJQ, 41].
The epitaxial NiggFeoo films were deposited on 10-50 nm thick Cu seed lay-
ers. The resistivity of the Cu films ranged from 2.5-9.5 wQ-cm, depending on
thickness and deposition parameters (the resistivity of bulk Cu is 1.8 u0Q-cm).
As aresult, it was difficult to extract the resistivity and magnetoresistance of
NigoFe9 overlayers due to some of the current shunting through the Cu film.
However, in this section, we compare the resistivity and magnetoresistance
of a 100 nm thick epitaxial NiggFe29 film on a 10 nm Cu underlayer of known
resistivity with those of a polycrystalline film of the same thickness. The
Chapter 5 126
SL
386
- Data Board
Easy |
Axis, 27 V VV V
73 wy
ad e
Helmholtz Coil at
Power Supply
Figure 5.16: Schematic of the setup used for magnetoresistance measure-
ments.
Chapter 5 127
resistivity of the 10 nm thick Cu underlayer was measured using a different
sample and was found to be 9.5 wQ-cm. Taking the resistance of the Cu un-
derlayer into account, the resistivity of the epitaxial film was estimated to be
25.1 ~Q-cm which is significantly lower than that of the polycrystalline film
(31.7 wQ-cm). On the other hand, the magnetoresistance of the epitaxial and
the polycrystalline films were 0.49% and 0.48%, respectively. Thus, the mag-
netoresistance of these films is relatively insensitive to the microstructure of
these films. The cause of lower MR in these films than bulk NigoFe9 is prob-
ably due to the presence of Ar and other defects incorporated in these films
during film growth. Figure 5.17 shows the magnetoresistance and resistivity
of a polycrystalline NiggFez9 film as a function of annealing temperature. The
lowering of the resistivity and increase in magnetoresistance with annealing
supports the above hypothesis. The annealing of the epitaxial NiggFeo9 film
at 200°C for 1-hour also led to an increase in the magnetoresistance from
0.49% to 0.89%. X-ray diffraction of the annealed sample indicated low-
ering of the Cu(200) peak intensity compared with the unannealed sample
whereas the NiFe(200) peak intensity was unaffected. This suggests partial
consumption of the Cu underlayer to form a silicide without any effect on the
crystallographic structure of the overlying NiggFe29 film. However, Cu and
Ni as well as Cu and Fe, form a solid solution and interdiffuse at relatively
low temperatures|42]. Significant lattice diffusion starts at 400°C and it is
likely that grain boundary-mediated interdiffusion occurs at even lower tem-
peratures. Thus, there may be undesirable interdiffusion during annealing of
Cu/NiggFe9 bilayers as well as multilayers.
To reduce the shunting current due to the Cu underlayer in epitaxial
Chapter 5 128
NigoFe2o films, underlayers as thin as 5.0 nm might be used. Furthermore,
as discussed in Sec. 5.7.1, Cu,Nij_, alloyed films may be used in conjuntion
with a Cu underlayer even thinner than 5.0 nm. The Cu,Ni,_, films, besides
providing a better lattice match to NiggFe29, also have higher resistivity than
NiggFeg9, as Cu and Ni do not form a common d-band|8]. This would reduce
the shunting current in the epitaxial NigoFe2o films on Si, and result in lower
resistivity as well as higher MR.
5.9 Conclusions
Epitaxial (001) NigoFezo films with atomically-abrupt interfaces were grown
on $1(001) utilizing epitaxial growth of Cu on Si. Magnetic properties of
these films were investigated in situ and were found to be sensitive to the
misfit strain between the NigoFeoo film and the Cu seed layer. H, was ob-
served to be significantly higher for < 10 nm thick epitaxial NigoFe29(001)
films deposited on a 50 nm thick Cu seed layer on Si, as compared with poly-
crystalline films of the same thickness. However, relatively thicker (> 20 nm)
epitaxial NiggFe29 films or those grown on lattice matched Cu,Nij_z alloys
had soft magnetic properties comparable to those of polycrystalline films.
This suggests that the reduction of coherency strain due to the mismatch
between NiggFe29 and Cu improves its soft magnetic properties. Theoretical
models to explain the variation of H. with thickness suggest that the higher
H, is more likely to be caused by coherency strain-induced surface roughness
rather than misfit dislocations. Finally, the magnetoresistances of the poly-
crystalline and the epitaxial films were comparable and lower than that for
Chapter 5
p (uQ-cm)
wl Ww Ww a w
to Gt > ao a
er)
—_
Ww
Oo
2.0
1.5
1.0
MR (%)
0.5
0.0
Se ee
Fo 4
E °
: o 4
ee re ee ee ee
0 50 100 150 200 250 300 350
T (°C)
(a)
r ee i
I. a 4
a ]
i A -
ares rere re ee
fe) 50 100 150 200 250 300 350
T (°C)
129
Figure 5.17: Variation of (a) resistivity and (b) magnetoresistance of poly-
crystalline NiggFeo9 film as a function of temperature for 1 hour anneals.
Chapter 5 130
bulk Nig9Feo9 possibly due to Ar incorporated during film growth. However,
annealing of these films at relatively low temperatures (200-300°C) can en-
hance the MR to its bulk value. Furthermore, the resistivity of the epitaxial
films was significantly lower than that for the polycrystalline films, suggest-
ing that the epitaxial films had lower grain-boundary density which might
be an important consideration for magnetoresistive devices applications.
Bibliography
[1] P. Grunberg, A. Schreiber, Y. Pang, M.B. Brodsky, and H. Sowers, Phys.
Rev. Lett. 57, 2442 (1986).
[2] S.S.P. Parkin, Appl. Phys. Lett. 60, 512 (1992).
[3] B. Dieny, V.S. Speriosu, B.A. Gurney, $.S.P. Parkin, D.R. Wilhoit, K.P.
Roche, S. Metin, D.T. Peterson, and S. Nadimi, J. of Magn. and Magn.
Mater. 93, 101 (1991).
[4] M. van Schilfgaarde and W.A. Harrison, Phys. Rev. Lett. 71, 3870
(1993).
[5] M.T. Johnson, R. Coehoorn, J.J. de Vries, N.W.E. McGee, J. aan de
Stegge, and P.J.H. Bloemen, Phys. Rev. Lett. 69, 969 (1992).
[6] P.M. Levy, S. Zhang, and A. Fert, Phys. Rev. Lett. 65, 1643 (1990).
[7] B. Dieny, Europh. Lett. 17, 261 (1992).
[8] I.A. Campbell and A. Fert, “Transport Properties of Ferromagnets,” in
Ferromagnetic Materials, ed. by E.P. Wohlfarth,
(9] A.F. Mayadas and M. Shatzkes, Phys. Rev. B 1, 1382 (1970).
131
Chapter 5 132
[10] R.F.C. Farrow, 5$.S.P. Parkin, and V.S. Speriosu, J. Appl. Phys. 64,
5315 (1988).
[11] J. Thomassen, F. May, B. Feldman, M. Wuttig, and H. Ibach, Phys.
Rev. Lett. 69, 3831 (1992).
[12] H.E. Inglefield, C.A. Ballentine, G. Bochi, $.5. Bogomolov, R.C.
O’Handley, and C.V. Thompson, Mater. Res. Soc. Proc. Vol. 308,
(1993).
[13] J.W. Matthews and J.L. Crawford, Thin Solid Films 5, 187 (1970).
[14] P. Hansen, Constitution of Binary Alloys, (McGraw-Hill, New York,
1958), p. 1045.
[15] M.H. Kryder and F.B. Humphrey, J. Appl. Phys. 42, 1808 (1971).
[16] R.D. Burbank and R.D. Heidenreich, Phil. Mag. 5, 373 (1960).
[17] U. Gradmann and J. Muller, J. Appl. Phys. 39, 1379 (1968).
[18] W.F. Egelhoff, Jr. and M.T. Kief, IEEE Trans. Mag. 28, 2742 (1992).
[19] H.S. Joo, I. Hashim, and H.A. Atwater, work in progress.
[20] J.W. Matthews and A.E. Blakeslee, J. Cryst. Growth. 27, 118 (1974).
[21] J.Y. Tsao, Materials Fundamentals of Molecular Beam Epitazy (Aca-
demic Press, San Diego, 1993), Ch.5.
(22) D. Hull and D.J. Bacon, Introduction to Dislocations, 3rd Ed., (Per-
magon Press, Oxford, U.K., 1984).
Chapter 5 133
(23] R.C. O’Handley and Ohsung Song, submitted to J. Appl. Phys., Nov.
1993.
[24] J.D. Eshelby, Proc. R. Soc. A241, 376 (1957); A252, 561 (1959).
[25] D.J. Srolovitz, Acta. Metall. 37, 621 (1989).
[26] A.G. Cullis, D.J. Robbins, A.J. Pidduck, and P.W. Smith, Mat. Res.
Soc. Symp. Vol. 280, 383 (1993).
[27] P.R. Berger, K. Chang, P. Bhattacharya, J. Singh, and K.K. Bajaj,
Appl. Phys. Lett. 53, 684 (1988).
[28] A. Zangwill, Physics at Surfaces, (Cambridge University Press, Cam-
bridge, U.K., 1988).
[29] A.P. Payne, B.M. Clemens, and S. Brennan, Rev. Sci. Instrum. 63, 1147
(1992).
[30] C. Kittel, Rev. Mod. Phys. 21, 541, 1949.
[31] B.D. Cullity, Introduction to Magnetic Materials, (Addison-Wesley,
Philippines, 1972), Ch.8.
[32] E. Klokholm and J.A. Aboaf, J. Appl. Phys. 52, 2474 (1981).
[33] A.C. Tam and H. Schroeder, IEEE Trans. Mag. 25, 2629 (1989).
[34] J.G. Gay and Roy Richter, Phys. Rev. Lett. 56, 2278 (1986).
[35] D. Weller, W. Reim, K. Sporl, and H. Brandle, J. Magn. Magn. Mater.
93, 183 (1991).
Chapter 5 134
[36] P. Bruno and J.-P. Renard, Appl. Phys. A 49, 499 (1989).
[37] T. Suzuki and C.H. Wilts, J. Appl. Phys. 38, 1356 (1967).
[38] M. Prutton in Thin Ferromagnetic Films, Butterworth & Co., England,
1964.
[39] L. Néel, J. Phys. Radium 17, 250 (1956).
[40] S. Middelhoek, Ph.D. Thesis, Amsterdam, 1961.
[41] S. Krongelb, A. Gangulee, and G. Das, IEEE Trans. Magn. 9, 568
(1973).
[42] C.A. Mackliet, Phys. Rev. 109, 1964 (1958).
Appendix A
More Details of the
Instrumentation
A.1 Design and Operation of the MOKE sys-
tem
A.1.1 Hardware
As soft magnetic materials like Permalloy require only a few Oersteds to
magnetize, great care was taken to isolate the MOKE quartz tube from
stray magnetic fields due to magnets in the ion gun and the rest of the
sputtering system. In particular, the glass-to-metal seals utilized to interface
the quartz tube to the sputtering system contain paramagnetic impurities
which can contribute to the depolarization of the light. Furthermore, they
are magnetostrictive, i.e., magnetic fields can lead to development of stresses
in them. At least 4” of quartz separated these seals on all sides from the
center of the tube where the MOKE analysis was done.
A Helmholtz coil pair was used to magnetize the sample. The radius of
the coils and their separation was chosen to be sufficiently larger than the
135
Appendix A 136
sample size so that the field over the sample surface was fairly uniform. Two
hundred turns of 18 AWG magnet wire from MWS Industries were wound
around two hollow cylinders made of phenolic. The cylinders had a 10 cm
outer diameter and were 2” long. The magnetic field H in Oersteds at the
center of an ideal Helmholtz coil pair with separation 2R is given by(1]:
2x NI
H= OR (A.1)
where N is the number of turns, J is the current in amperes, and R is also
the radius of the coils in cm. In reality, the wires have a finite thickness,
and not all the turns are located at the same radius. To account for these
nonidealities, Eq. (A.1) is replaced by more sophisticated versions which
typically require calculations on a computer. Furthermore, the separation
of our Helmholtz coils was 20% greater than the ideal separation due to
constraints imposed by the quartz tube thickness. As a result, the actual
field was smaller than that predicted by Eq. (A.1). However, a sensitive Hall
probe was used to calibrate the actual magnetic field obtained. For a current
of 5 amperes, a field of 75 Oersteds was obtained, which is 50% less than the
field predicted by Eq. (A.1).
On the other hand, the self-inductance L of the coils is given by
L= Ne (A.2)
where ¢ = f B- dA = BrR? is the magnetic flux passing through the coils.
For a current of 5 amperes, Eq. (A.2) implies a self-inductance of 0.025 Henry
and hence a reactance (wL) of 0.78 Q at a frequency of 10 Hz. Now, the DC
resistance of the two hundred turns of magnet wire is 2.64 2 so that for a
Appendix A 137
frequency of 10 Hz, the impedance of the coils is 2.75 { which implies a
voltage drop of 5 x 2.75 = 13.75 volts. The Kepco power supply used was
rated at 20V/10A so that it was adequate for these loads. However, at a
frequency of 100 Hz, the impedance of the coils would be 8.2 2. implying
a voltage drop of 41 volts at 5 amperes. Thus, it would not be possible
to use this power supply at frequencies higher than 50 Hz. Furthermore,
the maximum current that can be passed through the #18 wire is about 5
amperes so that the maximum field that can be obtained with these coils is
about 75 Oersteds.
In Kerr magnetometry, H. is obtained by converting the coil current to
an applied field H. Thus, it is extremely important that the coil current
is in phase with the magnetic flux measurements, i.e., Kerr rotation mea-
surements. In particular, phase difference between current and voltage due
to the inductance of the coils can give rise to as much as 50% error in H,.
This phase error was compensated in the software by taking into account
the inductance of the coils as calculated above. An alternate method to take
care of that would be having a resistor in series with the coils and using
the current in the resistor to calculate the applied field. Vibrating sample
magnetometry (VSM) was employed to confirm the H. measurements made
using our Kerr system. In particular, H, was measured along at least three
crystallographic directions for (100) NigoFe29 films on Cu/Si (along [100],
[110] and [010]) using both VSM and MOKE for up to three samples. The
MOKE 4H, measurements were thus found to be accurate to within 10%.
The resolution of our in situ MOKE system was limited by noise arising
from lack of proper vibration isolation of the system. In particular, low fre-
Appendix A 138
quency building noise (at a few Hz) was noticeable. It might be eliminated
by installing the sputtering system on a proper vibration-isolated table. The
noise due to mechanical pumps was reduced significantly by physically sep-
arating the pumps from the rest of the system and using vibration isolators
for mechanical pumps. Some of the vibration noise was eliminated by signal
averaging in the software which is discussed in the next section. Another
source of noise was the ubiquitous 60 Hz noise coming mainly from the fluo-
roscent lights in the room. This noise was reduced by switching off most of
these lights during data acquisition and might be further reduced by building
a light-tight enclosure for the photodetectors. A slightly different configu-
ration for making MOKE measurements involving a photoelastic modulator
and a lock-in amplifier, might have less noise and better resolution[2]. How-
ever, the cost of that MOKE configuration may be significantly higher due to
the additional cost of the photoelastic modulator and the lock-in amplifier.
Finally, the laser, photodetectors and the circuitry for their amplifiers
can also be a source of noise in the MOKE system. The 670 nm 10 mW
semiconductor diode laser used was found to be sufficiently stable and to
have enough power. However, the semiconductor laser beam has some in-
herent astigmatism and divergence due to the geometry of the edge-emitting
lasers. This astigmatism was reduced by adjusting the lens already packaged
with the laser and installing another cylindrical lens after the reflected beam.
Planar diffused silicon photodiodes packaged with a built-in amplifier and an
active area of 20 mm? (model UDT-020D) from United Detector Technol-
ogy were used for detection of the rotation in polarization after reflection.
These photodetectors are typically operated in two modes, photovoltaic (PV)
Appendix A 139
and photoconductive (PC). In the latter mode, the photodetector is reverse-
biased, resulting in smaller junction capacitance and hence faster response
time (~ ps). However, this results in increased noise level due to the higher
leakage current in reverse bias. In PV mode, no external bias is applied
to the photodetector so that noise level is lower at the expense of slower
response time. Since our application required operation at relatively low
frequencies and lower noise levels, PV mode was used instead of PC (see
Fig. A.1 which shows the electical schematic of the photodetector and the
amplifier circuitry). The amplifier consisted of an op-amp which required a
DC voltage supply of +15 V. The gain of the amplifier was controlled by a
feedback resistor Ry and a feedback capacitor Cy connected externally. An
appropriate value of 250 nF was chosen for C's to allow operation at < 100 Hz
whereas a 1-10 k{) variable resistor was used as the gain resistor. This gave
a voltage signal of 1-10 V corresponding to the maximum input laser power
of 10 mW. A variable resistor of 100 kQ was connected externally to adjust
the null offset of the photodetector (typically 0.5 mV with a maximum of 3
mV) to zero. The 0.1 uF capacitors connected in parallel to the + 15 V dc
power supply were for shunting any ac noise in the dc power supply from the
op-amp.
The output from the photodetectors was directly connected to two of the
eight analog input channels of the DT2821 board, each with 12 bit resolu-
tion (see Fig. A.2 for a schematic of the board). The voltage range that
is measured by the board with the highest programmable gain, is + 1.25
V. Thus, the smallest voltage signal that can be resolved by the board is
2.5V/2'? = 0.6 mV. The board also had a digital-to-analog converter (DAC)
Appendix A 140
which provided an analog output used as the input control signal to the coil
power supply. The control signal from the DAC to the power supply consisted
of a sinusoidal signal of appropriate amplitude and frequency. This configu-
ration worked out better than controlling the IEEE-488 board installed in the
power supply, via the 16 digital I/O lines also available on the board. That
would have involved installation and programming of another board IEEE-
488 board in the computer to allow communication/handshaking between
the power supply and the computer.
Furthermore, the board also had a programmable gain amplifier for ana-
log inputs which can be chosen to have gains of 1, 2, 4 or 8. This allowed the
amplifier gain to be changed in the software rather than in the photodetector
gain circuit, for different sample reflectivities and hence different photodetec-
tor outputs. The maximum throughput of this board was 50 kHz; hence, it
would not be feasible for higher data acquisition frequencies. Another limita-
tion of this board is that the analog inputs were not read simultaneously into
the computer. In other words, the time elapsed between the two input signals
being read into the board, was 1/50 kHz = 20 us. Hence, this board might
not be appropriate for data acquisition at frequencies higher than 1 kHz. A
better board choice for faster data acquisition would be model DT2831 which
simultaneously “grabs” signals from all the 8 input data channels. However,
the highest frequency for data acquisition was limited by the coil inductance
or the voltage-limitations of the coil power supply, as discussed above. Fur-
thermore, the response time of the power supply is 0.01 ms. Thus, if there
were 100 points in one loop, this would imply a maximum data acquisition
frequency of 1 kHz. Due to reasons mentioned above, the MOKE data was
Appendix A 141
—— Vo
UDF-020D, *_
UDT-020UV Fo
Ameren ewan ar ae eal
~tbV
Figure A.1: Electrical schematic of the photodetector and the amplifier.
The dashed rectangle represents the packaged photodetector and the op-
map whereas the encircled numbers represent external pin connections to
the photodetector.
acquired at < 10 Hz.
A.1.2 Software
The software for data acquisition was written in C, Version 6.0. It consisted
of three subprograms, the main program am.c, dt2821.c which controlled
the DT2821 board, and amgraph.c which contained the graphic routines for
the program. Originally, we thought that ATLAB Software, purchased from
Data Translation along with the board, might save us trouble because it
contained routines which can perform lower-level functions on the board and
can be directly accessed from a higher level language such as C, Pascal or
Fortran. However, the routines available were not specific enough for our
Appendix A
142
. Cock
= —
12- of 16-Bit itch .
| D/A Converter peat :
Anal 12- or 16-Bit Simultaneous
inputs WD —p Glock Analog
Converter 12-0 1681 |_| Deglich Outputs
—~—-—_ 2 =T D/A Converter | | Grevitry
| [Ghonnel-Goin List | | a -
— | F 3 | Line Some
Multiplexer iT — Digit i/o
Contra Logic |_| POR Me 16 tines of
|] 2-Chonnel DMA | | Femme} Digital 1/0
| : > +15V
rite {| Interface Logic | y. Rogulted bc/oc
] Data Transceiver > —15V
Tigger as ee +5
v +
IBM Personal Computer AT Bus
Figure A.2: Block diagram of the DT2821 board used for MOKE experi-
ment’s data acquisition.
Appendix A 143
application so we ended up writing our own routines for data manipulation
on the board. Interrupts generated by the computer signalled the board
to transfer the data from the I/O channels to the buffer on the board and
then to the computer memory. These interrupts were spaced as closely as
possible (i.e., within 1 clock cycle) to read the analog signals from the two
photodetectors. The board then waited for the next input signal until the
current in the coil power supply was changed to its next value. There were
typically 1000 points in a B-H loop, and the data acquisition frequency was
0.1-1 Hz in order to avoid the higher frequency noise.
The graphics routines used were those of DOS to minimize the time re-
quired for plotting the data. In fact for f < 10 Hz, real time display of the
hysteresis loop was possible. However, the raw data for less than 10 nm thick
films typically did not have a very high signal-to-noise ratio but that noise
was reduced considerably by averaging over up to 100 loops. At a frequency
of 1 Hz, this would imply acquiring data for 100 s. Some efforts were made
to eliminate the flickering of the screen as it was cleared after each hysteresis
loop cycle. However, this “bug” still remains undebugged.
The main program am.c was responsible for controlling the interactive
menu for the program as well as the data acquisition parameters which in-
cluded the data acquisition frequency, coil current amplitude, and the pro-
grammable gain. As mentioned earlier, the phase error due to the inductance
of the coils was calculated using expressions analogous to Eq. (A.1) and (A.2)
and compensated for during hysteresis loop plotting. Finally, as pointed out
in Sec.2.3, the hysteresis loop obtained using MOKE is actually a Kerr loop
rather than a B-H loop, as the Kerr rotation is only quasilinear with the mag-
Appendix A 144
netization of the sample. It might not be too difficult to implement a routine
which could convert the Kerr rotation to the actual magnetization. How-
ever, the relationship between the two is fairly complex and requires lengthy
matrix calculations on Mathematica. Therefore, in order to cut down com-
putation during data acquisition, it might be easier to implement a lookup
table with values of the Kerr rotation and actual magnetization for every 10
nm of the film thickness and interpolating for intermediate values.
The software also had a routine which displayed the output of the two
photodetectors as bars. This allowed adjustment of the quarter-wave plate
to get the desired polarization of the reflected light (see next section for more
details).
A.1.3 Operation and Suggestions for Improvement
For reasons mentioned in Sec.2.3, the incident polarization of the light was
chosen to be s. After reflection, a quarter-wave plate was used to split the
light into equal amounts of s and p polarizations, making use of the MOKE
software routine which displays the detector output as bars. Furthermore,
the quarter-wave plate allowed elimination of the Kerr ellipticity and en-
hancement of the Kerr rotation for this orientation, as discussed in Sec.2.3.
However, due to the divergence of the laser beam, the beam spot on the de-
tector was almost the same size as the active area of the detector. This led to
somewhat noisier Kerr loops when the sum-and-difference signal of the two
photodetectors, i.e., a8 was plotted. Less noisy loops were obtained when
output from only one photodetector was plotted while keeping the intensity
of the light on that photodetector relatively small. When the output of only
Appendix A 145
one photodetector is used to obtain the Kerr loop, care should be taken to
convert the Kerr rotation to magnetization as it will be dependent on the
gain of the pre-amp as well as the incident laser power. Using photodetectors
with larger active area may make it feasible to use the sum-and-difference
signal from both the detectors. As some of the divergence of the laser beam
is a consequence of scattering from the quartz tube, it may not be easy to
compensate for that divergence using simple lenses.
For a typical Kerr measurement, the coil current amplitude and frequency
were 3 A and 0.1 Hz, respectively, with signal averaging over 100 loops. Kerr
loops were obtained for both easy and hard axes of the NiFe films. In the
case of epitaxial (100) films, Kerr loops were measured along the [100], [110]
and [010] crystallographic directions.
A.2 SAXD system: Operation and Current
Status
A.2.1 Operation
As mentioned in Sec. 2.5.1, the higher satellites in the small angle x-ray
spectrum were quite sensitive to the alignment of the sample and the diffrac-
tometer with respect to the main x-ray beam. To align the sample accurately,
the following procedure was used:
1. The goniometer was set to 0° and the x-ray power supply was set
to the lowest voltage and current levels for which x-ray emission
could be obtained. This was done to prevent the x-ray detector
from damage due to an excessively strong x-ray beam.
Appendix A 146
A.2.2
. Using the high resolution translation stage, the sample was trans-
lated forward until the x-ray beam was partially blocked by the
sample.
. The high resolution rotation stage was adjusted to maximize the
x-ray intensity.
. Steps (2) and (3) were iterated until the x-ray intensity could not
be maximized any further.
. The sample was translated forward until the main beam was re-
duced to half its intensity.
. Finally, a quick scan (with a short counting time of 3 s and a
larger step size of 0.1°) was made to compare the SAXD satellite
positions with those expected from OM simulations. If at least
three SAXD satellites could be observed in such a scan, a longer
scan with a smaller step size was made to obtain the final SAXD
spectrum.
Current Status
As the age of the GE x-ray power supply exceeded the age of the untenured
faculty members in the materials science department, it was thought best to
replace it with a solid-state power supply from Inel. Furthermore, the tube
tower was also replaced by one from Inel which had a graphite monochro-
mator. Figure A.3 shows the schematic of the tube tower, monochromator,
the sample holder, the diffractometer and the x-ray detector. The graphite
monochromator can be adjusted with respect to the tube tower for the ap-
Appendix A 147
propriate x-ray radiation (typically Mo, Co or Cr). Highest x-ray intensity
was obtained with the Mo tube whereas the Cr intensity was the lowest. The
tube tower was mounted on translation stages and a lab jack which allowed
accurate alignment of the main beam with respect to the goniometer. A x-
ray fluorescent screen was used to ensure that the x-ray optics was properly
aligned. Only with the Mo tube, strong fluorescence could be observed on
the fluorescent screen as well as diffraction peaks from a powdered Si sam-
ple. It is possible that the diffraction peak intensities were low because the
tube tower was not located at the diffractometer circle[3]. However, a SAXD
spectrum could be obtained with all the radiation sources as SAXD does not
require very high intensity levels.
A.3 Cu Epitaxy: What Worked and What
Didn’t
It would be unfair not to mention that obtaining Cu epitaxy on Si repro-
ducibly from run to run was not easy. There was certainly an element of
luck involved, as we were able to achieve epitaxial growth of Cu on Si on our
first attempt but thereafter we were not able to reproduce it for up to 6 at-
tempts. However, since then, we have acquired expertise in what deposition
parameters the epitaxy is most sensitive to.
Residual moisture or hydrocarbons on the Si surface were definitely a fac-
tor, as the success rate for achieving Cu epitaxy was improved after installing
a substrate heater in the sputtering chamber. A pre-deposition bake at 200°
for 2 hours allowed the residual moisture and hydrocarbons to be desorbed
Appendix A 148
Diffractometer Circle
Graphite
Monochromator
Inel Tube
Tower
X-ray Detector
Figure A.3: Schematic of the Inel tube tower and the GE diffractometer for
small angle x-ray diffraction measurements.
Appendix A 149
from the substrate. However, baking the sample at higher temperatures was
sometimes deleterious as that led to outgassing from parts of the heater and
the substrate holder. Ensuring that all parts of heater and the substrate
holder are made from Mo or ceramic rather than stainless steel might solve
the above problem. Furthermore, particulates on the 5i surface were also a
factor as the substrate cleaning was not carried in a clean room environment
by any standards. Pieces of Si (usually 1” x 0.5”) cut from $i(100) wafers
were inspected under an optical microscope for particulates and only the
relatively particulate-free ones were used.
The purity of the Cu film was also very crucial to achieving epitaxial
growth on Si. Presence of < 1% Fe and/or C disrupted the epitaxy leading
to (111)-textured polycrystalline film growth. The Fe impurities entered
the Cu film when the ion gun was moved farther away from the sputtering
system in an effort to isolate the RHEED beam from the magnets in the ion
gun. As discussed in Sec.2.1, the ion beam expands into a Gaussian profile
with distance due to charge repulsion. Thus, a fraction of the ion beam
was possibly sputtering the walls of the chamber behind the target. This
problem was solved by replacing the original 3 cm grid with a customized 2
cm ion-beam grid pattern, thus reducing the beam size. At another stage,
the graphite grids in the ion gun were misaligned due to a crack in the screen
grid which is the grid closer to the ion source than the accelerator grid. This
resulted in erosion of the accelerator grid near the crack in the screen grid,
leading to incorporation of C in the film and disruption of epitaxy.
Furthermore, the epitaxy was also very sensitive to the deposition tem-
perature. For deposition at T > 80°C, a CugSi film was obtained immediately
Appendix A 150
upon deposition, as indicated by absence of the RHEED pattern observed for
an epitaxial Cu film. This suggests that it is vital for the sample to be cooled
to room temperature after the pre-deposition bake, and the temperature to
be measured accurately. Typically, in vacuum systems, cooling rates are very
slow owing to poor thermal contacts and lack of cooling by convection. In
order to increase the cooling rate, a LN2 finger was added to the substrate
holder. This consisted of a Cu braid wrapped around the sample holder and
connected via an electrical feedthrough to a Cu braid outside vacuum dipped
in LN2.
Finally, at this point, we cannot clearly establish any correlation be-
tween Cu epitaxy and the chemical cleaning procedures for obtaining a H-
terminated Si surface. However, it is possible that the concentration of hy-
droflouric acid used may also be a factor, as higher concentrations than 10%
may contribute to greater etching of the Si, resulting in a rough surface.
Bibliography
[1] B.D. Cullity, Introduction to Magnetic Materials, (Addison-Wesley,
Philippines, 1972), Ch.1.
[2] J.F. Dillon, Jr., E.M. Gyorgy, F. Hellman, L.R. Walker, and R.C. Fulton,
J. Appl. Phys. 64, 6098 (1988).
[3] B.D. Cullity, Elements of X-ray Diffraction, (Addison-Wesley, Philip-
pines, 1978), Ch.6.
151
Appendix B
Domain Walls in Thin Films
The domain wall energy and thickness in bulk magnetic materials are gov-
erned mainly by exchange and anisotropy energy contributions. However,
for thin films (< 50 nm for NiFe), the rotation of magnetic moments in the
domain wall takes place in the plane of the film as opposed to out of the
film plane. The former type of domain wall, known as Néel wall, has lower
magnetostatic energy than the latter type, known as Bloch wall. In the ab-
sence of magnetostatic contributions, the domain wall thickness is found by
minimizing the exchange and anisotropy contributions to the total energy of
the domain wall. Assuming that ¢, the angle between neighboring spins in
the domain wall, rotates linearly through 180°, we have
ganz foros
ia
(B.1)
where 6 is the domain wall width. Assuming that the exchange energy exists
only between neighboring spins, the exchange contribution becomes:
Ee, = J (2) =J (z). (B.2)
Appendix B 153
where J is the exchange stiffness constant. On the other hand, contribution
due to uniaxial anisotropy is given by:
Ex = K,cos’¢. (B.3)
To get the total energy of the domain wall, the exchange and anisotropy
energy contributions are integrated over the domain wall width, i.e.,
Biot = [ (Een + Ex)de. (B.4)
Making use of Eq. (B.1), (B.2) and (B.3), the above equation can be written
as:
5 ¢m\? n/2 (5 ; Jn? 5K,
Ba= [I(G) +f, (=) Kucos'odg =~ +> (B.S)
Minimizing the above expression with respect to 6 yields the following ex-
pressions for the equilibrium domain wall width and domain wall energy,
respectively:
y = rV2WVIK
= vay]. (B.6)
For a thin film of thickness h, the magnetostatic contribution to a domain
wall energy can be estimated by modeling the domain wall as an elliptic
cylinder as shown in Fig.B.1[1]. The magnetostatic energy is then evaluated
using the expression:
1 1
Ems = —5HaM, = —5NaM}, (B.7)
Appendix B 154
where H, is the self-demagnetizing field and is related to the demagnetization
factor Nz according to Hz = NaM,. For this elliptical geometry, it can be
exactly evaluated using[2]
276
Na= iah (B.8)
so that the surface energy of a Néel wall is given by:
Jn? 1 nroh
=—+-6k, 2 .
YN 5 + 5 + ix 7 Ms (B.9)
Minimizing with respect to 6 yields
dyn _ Jn? 1 moh? 2
which has been solved numerically by Middelhoek[3] to find the domain wall
thickness 6 as a function of film thickness h. For the film thicknesses of
interest here, h ~ 10 nm and 6 ~ 100 nm so that the magnetostatic contri-
bution in Eq. (B.10) can be approximated as ah Using this approximation,
Eq. (B.10) can be solved analytically for 6 as a function of h and its solution
is plotted in Fig.B.2. Substituting the domain wall width thus calculated
into Eq. (B.10) yields the equilibrium Néel wall energy yy, which is plotted
as function of film thickness for 0 < kh < 20 nm in Fig.B.3.
Appendix B
<_— § —>
Bloch Wall
155
m I
® OO | ©
~+ re) >
Néel Wall
Figure B.1: Cross-section of Bloch and Néel walls according to the approxi-
mation of Néel.
1.6
1.4
5 (um)
0.8 F
Pe 1 1 ‘ | 4 ‘ rl
Figure B.2: Néel wall thickness vs.
0.6 1 1. i 1 ! 1 L rl n
ie)
10 15 20
h (nm)
film thickness for < 20.0 nm thick films.
Appendix B 156
yw (ergs/cm*)
uw rs
NS
aan eee
TT
aa i J. i L i i d, | i J, L I. L H iL 7
QO 5 10 15 20
h (nm)
oO
Figure B.3: Néel wall energy vs. film thickness for < 20.0 nm thick films.
Bibliography
[1] L. Néel, J. Phys. Radium 17, 250 (1956).
[2] B.D. Cullity, Introduction to Magnetic Materials, (Addison-Wesley,
Philippines, 1972), Ch.2.
[3] S. Middelhoek, Ph.D. Thesis, Amsterdam, 1961.
157