Di Kai PHD Thesis Nus 2015

Experimental and Theoretical Studies of
Spin-Wave Dynamics in Magnetic

Nanostructures
DI KAI
(B.Sc. Sichuan University)
A THESIS SUBMITTED FOR THE DEGREE OF

DOCTOR OF PHILOSOPHY
DEPARTMENT OF PHYSICS
NATIONAL UNIVERSITY OF SINGAPORE
2015
ii
DECLARATION
I hereby declare that this thesis is my original work and it has been
written by me in its entirety. I have duly acknowledged all the sources
of information which have been used in the thesis.
This thesis has also not been submitted for any degree in any university previously.
Di Kai
18 August 2015
iv
Acknowledgement
I would like to thank my supervisor Prof. Kuok Meng Hau for his
continuous support and guidance during my four-year PhD study,
without which the thesis cannot be finished. His knowledge, patience,
enthusiasm and conscientiousness helped me enormously and will benefit me in my future career. I would also like to express my gratitude
to my co-supervisor Associate Prof. Lim Hock Siah for training me
to do calculations and sharing his breadth of knowledge and experience during countless discussions. Under his guidance, I learned how
to tackle complex problems. I would like to thank my co-supervisor
Associate Prof. Seidikkurippu N. Piramanayagam for always encouraging me to attend academic conferences and guiding me when I was
with DSI. Understanding the applications of my research from this
experience has benefited me greatly.
I am sincerely thankful to Prof. Ng Ser Choon for his invaluable suggestions and discussions during the past four years. Many
problems became clear and tractable with his experienced recommendations. A warm thanks I give to Prof. Mike G. Cottam, who gave
me helpful guidance and comments during his visits and via email.
I owe a debt of thanks to Miss Zhang Li for training and helping me
to do the Brillouin measurements. I would like to express my appre-
ciation to Dr. Wong Seng Kai for training me in sample fabrication.

I am also thankful to Dr. Ma FuSheng and Dr. Pan HuiHui for their
helps during my first-year study.
Additionally, I am very grateful to Associate Prof. Yang Hyunsoo
and his group members, Dr. Kulothungasagaran Narayanapillai, Dr.
Qiu Xuepeng, and Miss Yu Jiawei, for providing the samples in my
studies. I would also like to thank Mr. Hou Chenguang, Dr. Luo
Yuan, and Dr. Gong Li for sharing their thoughts on my questions.
Last but not the least, I would like to thank my parents for their
unconditional support. A special thanks goes to my wife Dr. Lin
Jia-Dan for being always caring and encouraging.
Di Kai
August 2015
If all you have is a hammer, everything looks like a nail.

Abraham Maslow
To my parents, my wife and my brother.
iv
CONTENTS
i
Acknowledgement
viii
Summary
xi
List of Tables
xiii
List of Figures
xv
Glossary
1 Introduction
1.1
1.2
LLG Equation & Spin Waves . . . . . . . . . . . . . . . . . . . .
1.1.1
Ferromagnetism . . . . . . . . . . . . . . . . . . . . . . . .
1.1.2
Landau-Lifshitz-Gilbert Equation . . . . . . . . . . . . . .
1.1.2.1
Heisenberg Model . . . . . . . . . . . . . . . . . .
1.1.2.2
Macroscopic Theory of Magnetization Dynamics .
1.1.2.3
Effective Field in the LLG Equation . . . . . . .
1.1.3
Spin Waves in Thin Films . . . . . . . . . . . . . . . . . .
1.1.4
Magnonic Crystals . . . . . . . . . . . . . . . . . . . . . .
10
Brillouin Light Scattering . . . . . . . . . . . . . . . . . . . . . .
12
1.2.1
Physics of BLS . . . . . . . . . . . . . . . . . . . . . . . .
12
1.2.2
Brillouin Spectrometer . . . . . . . . . . . . . . . . . . . .
13
1.2.3
Intensity of Brillouin Light Scattering . . . . . . . . . . . .
14
Bibliography
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
16
CONTENTS
2 Micromagnetic Simulations
19
2.1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
19
2.2
OOMMF Approach . . . . . . . . . . . . . . . . . . . . . . . . . .
19
2.3
Finite-Element Approach Using COMSOL . . . . . . . . . . . . .
21
2.3.1
Brief Introduction to Finite Element Method . . . . . . . .
22
2.3.1.1
Weak Formulation: An Example . . . . . . . . .
22
2.3.1.2
Mesh Descritization & Basis Functions . . . . . .
25
2.3.1.3
LLG Equation in Weak Form . . . . . . . . . . .
28
2.3.2
Frequency Domain . . . . . . . . . . . . . . . . . . . . . .
30
2.3.3
Calculation of Demagnetizing field . . . . . . . . . . . . .
32
2.3.4
Frequency Domain in 2D . . . . . . . . . . . . . . . . . . .
33
Bibliography
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
34
3 Band Structure of a 1D Width-Modulated Nanostripe

Magnonic Crystal
35
3.1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
35
3.2
The Complete Band Structure . . . . . . . . . . . . . . . . . . . .
36
3.3
Band Structure Modulation by a Transverse Magnetic Field . . .
39
3.3.1
Tunability of Band Structure Parameters . . . . . . . . . .
43
3.3.2
Symmetry Assignment and Excitation Efficiency . . . . . .
46
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
50
3.4
Bibliography
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4 Artificial Defects in Magnonic Crystals
50
53
4.1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
53
4.2
Structures of Defect MCs and Experimental Details . . . . . . . .
56
4.3
Band Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . .
58
4.4
Size Dependence . . . . . . . . . . . . . . . . . . . . . . . . . . .
60
4.5
Summary & Discussion . . . . . . . . . . . . . . . . . . . . . . . .
64
Bibliography
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
65
5 SWs in Ferromagnetic Films with Interfacial DzyaloshinskiiMoriya Interactions

69
5.1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
vi
69
CONTENTS
5.2
Dispersion Relation in the Presence of Interfacial DMI . . . . . .
71
5.3
Experimental Results . . . . . . . . . . . . . . . . . . . . . . . . .
5.3.1 Pt/Co/Ni Film . . . . . . . . . . . . . . . . . . . . . . . .
73
73
5.3.2
5.3.1.1
Sample Fabrication & Basic Magnetic Properties
74
5.3.1.2
Details of BLS Experiment . . . . . . . . . . . .
75
5.3.1.3
5.3.1.4
Dispersion Relation . . . . . . . . . . . . . . . . .
Spin-wave Linewidth . . . . . . . . . . . . . . . .
76
80
Pt/CoFeB Film . . . . . . . . . . . . . . . . . . . . . . . .
83
5.3.2.1
Introduction
. . . . . . . . . . . . . . . . . . . .
83
5.3.2.2
5.3.2.3
Sample Fabrication and Characterization . . . . .

Frequency Asymmetry caused by Asymmetric Sur-
84
face Pinning . . . . . . . . . . . . . . . . . . . . .
85
BLS Measurement . . . . . . . . . . . . . . . . .
86
5.3.2.4
5.4
5.3.2.5 Dispersion Relation . . . . . . . . . . . . . . . . . 88

Discussion: Physical Explanation of DMI-induced Asymmetric Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
90
5.4.1
Asymmetric Dispersion Relation . . . . . . . . . . . . . . .
90
5.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
92
92
6 Spin-Wave Nonreciprocity
99
6.1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.2
Nonreciprocity by Interband Magnonic Transitions

6.2.1
6.2.2
99
. . . . . . . . 100
Coupled-Mode Analysis . . . . . . . . . . . . . . . . . . . 102

Micromagnetic Simulation . . . . . . . . . . . . . . . . . . 107
6.3
Nonreciprocity Enhanced by Antiferromagnetic coupling . . . . . 110
6.4
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
Bibliography
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
7 Conclusions
121
8 List of Publications and Conferences
125
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
vii
Summary
In this thesis, the spin-wave dynamics in different magnetic nanostructured systems has been studied experimentally or/and theoretically. Experimentally, Brillouin light scattering (BLS) is the principal
characterization method employed. Analytical and numerical calculations based on the Landau-Lifshitz-Gilbert equation are also used
either to explain the experimental data or to predict new phenomena.
Chapters 1 and 2 outline the theory of spin waves (SWs) of ferromagnets, the principles of the BLS technique, and various micromagnetic simulation methods. These are the fundamental theory and
tools used in the research projects discussed in the remaining chapters.
Chapter 3 presents a theoretical study of a width-modulated nanostripe magnonic crystal (MC). We have commented on a micromagnetic simulation method for the calculation of the band structures of
such MCs. Further, the field-tunability of the band structure of such
MC is investigated.
In Chapter 4, a BLS study of one-dimensional MCs is discussed.
Here, emphasis is placed on MCs with controllable artificial defects,
which offer more flexibility in designing the spin-wave band structures.
Chapter 5 deals with the experimental and theoretical studies of
the interfacial Dzyaloshinskii-Moriya interaction (DMI) in ultrathin
multilayer magnetic thin films. Employing BLS, the first direct observations of interfacial DMI in several structures are made, which
also demonstrate a convenient approach to determining the DMI constant. Specifically, the effect of the DMI on the lifetimes of SWs is
also observed, with measurements agreeing with a simple analytical
theory. The DMI gives rise to a few nonreciprocal properties of SWs,
for which a simple physical insight is provided.
Chapter 6 discusses different mechanisms of spin-wave nonreciprocity, such as conventional nonreciprocity of Damon-Eshbach surface
waves, nonreciprocal magnonic mode transitions, and nonreciprocity
in the presence of DMI.
Finally, Chapter 7 summarizes the findings of the thesis and proposes further possible extensions of the studies.
CONTENTS
List of Tables
3.1
Character tables of symmetry groups (after Ref. [7]) . . . . . . . .
xi
46
LIST OF TABLES
xii
List of Figures
1.1
Ground state of a 1D spin lattice. . . . . . . . . . . . . . . . . . .
1.2
Excited state of a 1D spin lattice showing spin waves. . . . . . . .
1.3
Band structure of a Co(250nm)/Py(250nm) MC measured by BLS. 11
1.4
Brillouin light scattering processes by magnons. . . . . . . . . . .
13
1.5
Schematic of the optical path of Brillouin light scattering.
. . . .
14
2.1
Triangular mesh grid of an irregular 2D structure. . . . . . . . . .
26
2.2
Illustration of 1D piecewise linear basis functions. . . . . . . . . .
26
2.3
Sinc function approximated by linear combination of triangular

basis functions. . . . . . . . . . . . . . . . . . . . . . . . . . . . .
27
3.1
Geometric structure of the MC. . . . . . . . . . . . . . . . . . . .
36
3.2
Band structure and symmetry assignment of the MC. . . . . . . .
38
3.3
Geometric and static properties of the MC. . . . . . . . . . . . . .
41
3.4
Calculation results of magnonic band structure under various transverse field. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
42
3.5
Mode frequencies and sizes of band gaps as functions of transverse

field. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
44
3.6
Illustration of field-tunability of band gaps. . . . . . . . . . . . . .
45
3.7
Eigenmodes of magnetization with phase information. . . . . . . .
47
3.8
Excitation spectra under different transverse field. . . . . . . . . .
48
4.1
SEM image of the MC with 400nm-wide Py defect stripes. . . . .
56
4.2
Brillouin spectra of the MC with and without defects recorded at
4.3
various wavevectors. . . . . . . . . . . . . . . . . . . . . . . . . .
57
Magnon band structures of the perfect and defect MCs. . . . . . .
58
xiii
LIST OF FIGURES
4.4
Dependence of defect mode frequency on defect Py stripe width. .
61
4.5
Mode profiles of confined and propagating modes. . . . . . . . . .
62
5.1
In-plane and perpendicular magnetic hysteresis loops of the Pt/Co/Ni
5.2
film. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Typical Brillouin spectra of the Pt/Co/Ni film. . . . . . . . . . .
74
75
5.3
Brillouin results of the Pt/Co/Ni and control samples. . . . . . .
78
5.4
Spin-wave linewidth of the Pt/Co/Ni sample. . . . . . . . . . . .
81
5.5
5.6
Magnetic hysteresis loops of the Pt/CoFeB sample. . . . . . . . .

BLS configuration and Brillouin spectra of the Pt/CoFeB sample.
85
87
5.7
Measured spin-wave dispersion relation of the Pt/CoFeB film. . .
88
5.8
Illustration of DMI-induced frequency asymmetry. . . . . . . . . .
91
6.1
Nonreciprocal spin-wave mode transitions in the Py nanostripe

waveguide. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
6.2
Forward spin-wave mode transitions in the waveguide. . . . . . . . 105
6.3
Snapshots of out-of-plane components mz under the perturbation

field. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
6.4
Schematic diagram of the magnonic crystal with AFM coupling. . 111
6.5
Calculated dispersion curves of the MCs with different magnetic
6.6
parameters. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
Calculated nonreciprocity and interlayer angle versus applied magnetic field. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
6.7
Band structures, calculated for zero applied field, of various AFM
6.8
magnonic crystals. . . . . . . . . . . . . . . . . . . . . . . . . . . 114

Band structures, calculated for zero applied field, of various AFM
magnonic crystals. . . . . . . . . . . . . . . . . . . . . . . . . . . 116
xiv
Glossary
dimensionless Gilbert damping coefficient
Heff
effective magnetic field of

LLG equation
Hex
effective Heissenberg
change field
hex
dynamic component of effective Heissenberg exchange

field
angular momentum
lS
surface normal of dS
total magnetization vector
dynamic component of magnetization vector
M0
equilibrium
vector
ex-
magnetic moment
torque
magnetic induction field
ei
unit vector along i-th coordinate axis, i = x, y, z
unit vector of total magnetization
H0
applied external magnetic

field
position vector, r = [x, y, z]
effective magnetic anisotropy

field
Hani
spin vector
gyromagnetic ratio
reduced Planck constant
wavevector along the width

of waveguide
decay length
vacuum permeability
del operator
angular frequency (rad/s)
magnetic scalar potential
dynamic component of magnetic scalar potential
hani
dynamic component of effective magnetic anisotropy

field
Hdip
demagnetizing field, magnetic dipolar field
hdip
dynamic component of demagnetizing field
HDM
effective magnetic field arising from DMI
hDM
dynamic component of effective magnetic field arising

from DMI
xv
magnetization
GLOSSARY
incident angle of light in the

back-scattering geometry of
BLS
permittivity
q, k
wavevector
temperature
time
TC
Curie temperature
vg
group velocity
width
1D,2D,3D
one-,
two-,
dimensional
AFM
anti-ferromagnetic,
ferromagnetism
API
application program interface
exchange stiffness constant
lattice constant, period of

MC
DMI constant
film thickness
energy
ordinary frequency (Hz)
Hsat
saturation magnetic field
BLS
Brillouin light scattering
HT
transversely applied magnetic field
BZ
Brillouin zone
DC
direct current
DE
Damon-Eshbach geometry
DMI
Dzyaloshinskii-Moriya interaction
threeanti-
HU
effective magnetic field of

uniaxial anisotropy
2A/(0 MS )
J0
coupling constant of Heisenberg exchange interaction
DW
domain wall
Eq.
equation
magnetic
stant
FDM
dinite difference method
FEM
finite element method
length
FFT
fast Fourier transform
MS
saturation magnetization
Fig.
figure
me
electron mass
FM
Mx , My , Mz
x-,y-,z-component of total
magnetization
ferromagnetic,
netism
FP
Fabry-Prot (FP) (interferometer)
FWHM
full width at half maximum
GMR
giant magnetoresistance
GUI
graphical user interface
mx , my , mz
nx , ny , nz
anisotropy
x-,y-,z-component of
namic magnetization
con-
dy-
x-,y-,z-component of unit
vector of magnetization
xvi
ferromag-
GLOSSARY
permalloy, Ni-Fe alloy
LL
Laudau-Lifshitz equation
Py
LLG
Landau-Lifshitz-Gilbert
equation
Re[C],Im(C) real, imaginary part of a

complex number C
MC
magnonic crystal
Ref.
reference
MOKE
magneto-optic Kerr effect
RT
Object Oriented MicroMagnetic Framework
room temperature
OOMMF
SEM
scanning
scope
electron
micro-
PC
personal computer
PDE
partial differential equation
SkX
skyrmion crystal
PhC
photonic crystal
SW
spin wave
PMA
perpendicular
anisotropy
VSM
vibrating sample magnetometry
PSSW
perpendicular standing spin

wave
YIG
yttrium iron garnet
magnetic
xvii
GLOSSARY
xviii
1
Introduction
1.1
1.1.1
LLG Equation & Spin Waves

Ferromagnetism
Ferromagnetism is an old and new phenomenon. It has been utilized by humans thousands of years ago. The invention of the compass, which relies on the
Earths magnetic field for navigation, made sea travel possible and played an important role in the age of discovery. In modern society, magnets are ubiquitous
and are the functional part of numerous equipment, ranging from electric motors
in trains, electromagnets in container lift cranes, to doors of refrigerators, and
speakers of multimedia systems. The discovery of the giant magnetoresistance
effect is a landmark in the field magnetism. This effect has played a major role
in various magnetic sensors and most importantly, revolutionized data storage in
hard disks. It is said that the use of giant magnetoresistance can be regarded as
one of the first major applications of nanotechnology [1].
Over the centuries, various mechanisms have been proposed to explain ferro-
1. INTRODUCTION
magnetism. However, there is no satisfactory understanding until the establishment of quantum mechanics. The idea that ferromagnetism is a purely quantum
mechanical phenomenon was a breakthrough and set the foundation of modern
research and applications of nanomagnetism. Many novel physical phenomena related to ferromagnetism have been discovered over the past century. Today, due
to its interesting fundamental science and wide-ranging applications, magnetism
is a very active research field, especially in spintronics.
1.1.2
Landau-Lifshitz-Gilbert Equation
1.1.2.1
Heisenberg Model
The classical Heisenberg model is defined by the following Hamiltonian:

H = J0 Si Sj ,
(1.1)
where the nearest-neighbor exchange constant J0 > 0 for ferromagnets (J0 = 0

for non-nearest neighbors), and Si is the spin vector at site i. The macroscopic
magnetic property can be described by the quantity magnetization M , which is
the magnetic dipole moment per unit volume in a magnetic material. At 0 K,
all the spins of a ferromagnet are aligned parallel, known as the ground state
(see Fig. 1.1). The spontaneous magnetization of a ferromagnet becomes zero
if the temperature T > Tc , where Tc is called the Curie temperature. For low
temperatures below Tc , the spontaneous magnetization follows the Blochs law
[2]:
"
M (T ) = M (0) 1
T
Tc
3/2 #
,
(1.2)
1.1 LLG Equation & Spin Waves
Figure 1.1: Ground state of a 1D spin

lattice. Arrows represent classical spin
vectors.
Figure 1.2: Excited state of a 1D spin

lattice showing spin waves.
where M (0) is the spontaneous magnetization at 0 K. The decreased spontaneous

magnetization at higher temperatures arises from thermally excited SWs, wavelike
deviations in the direction of spins from the ground state (see Fig. 1.2). The
energy of SWs is quantized, the energy quanta of which are called magnons,
quasi-particles like phonons.
From the Heisenberg model, it can be shown that the dispersion relation of a
1D spin chain is:

~ = 2J0 S 1 cos(k a) ,
(1.3)
where k and a are the wavevector and lattice constant, respectively. In the long
wavelength limit, Eq. 1.3 becomes
~ = J0 S(k a)2 .
Similar results can be obtained for 3D spin lattices.
(1.4)
1. INTRODUCTION
1.1.2.2
Macroscopic Theory of Magnetization Dynamics
If we are only interested in long-wavelength SWs, such as those detectable by

Brillouin spectroscopy (see Section 1.2), the spin-wave dynamics can be conveniently described by macroscopic continuum models. The magnetic moment of
an atom is related to its angular momentum L by
|e|
gL,
2me
(1.5)
where the factor |e| g/(2me ) is the gyromagnetic ratio [3], e, me being the elementary charge, and electron mass, respectively. From dL/dt = , where the
torque due to a magnetic field H is = 0 H, the equation of motion for
the magnetization can be obtained:
dM
= M H,
dt
(1.6)
|e|g
where = 0 2m
. Henceforth, for brevity, we will call the gyromagnetic ratio.
e
Equation 1.6 describes the precession of M around H without energy loss.

The applicability of Eq. 1.6 is not limited to the precessional motion of magnetic
moments in an applied magnetic field. It can be used to describe any interactions
that impose a torque on a magnetic moment once an effective magnetic field is
defined [4]:
1 E(M )
0 M
"
#
1 E(M )
E(M )
E(M )
=
ex +
ey +
ez ,
0
Mx
My
Mz
Heff =
(1.7)
where E(M ) is the energy of the interaction under consideration, and ex,y,z are
unit vectors along their respective coordinate axis.
By introducing a phenomenological damping term to represent the behavior of
real magnetic materials, L. D. Landau and E. M. Lifshitz proposed the following
Landau-Lifshitz (LL) equation [5]:
dM
= M Heff + M (M Heff ) ,
dt
(1.8)
where is a phenomenological damping parameter, Heff the effective magnetic

field including various interactions. To better describe ferromagnets with large
dampings, Gilbert modified the damping term of the LL equation, resulting in
the Landau-Lifshitz-Gilbert (LLG) equation [4, 6]:
dM
M
dM
= M Heff +
,
dt
Ms
dt
(1.9)
where the dimensionless parameter is the phenomenological Gilbert damping

constant. In this thesis, the LLG equation will be extensively used for both
analytical and numerical calculations.
1.1.2.3
Effective Field in the LLG Equation
Generally speaking, for a realistic material diverse interactions contribute to

the effective field Heff in the above-mentioned LLG equation. In most cases, these
interactions include Zeeman energy, magnetic dipole-dipole interaction, exchange
interaction, and magnetic anisotropic energy.
(1) Zeeman energy
1. INTRODUCTION
Zeeman energy is the energy of a magnetic moment subject to an external
magnetic field with the following energy density term
EZ = 0 M H0 .
(1.10)
The corresponding effective field is simply H0 .

(2) Magnetic dipole-dipole interaction
This classical interaction is also called dipolar coupling. The potential energy
of two magnetic moments m1 and m2 is given by:
Edip =

0
3
(m
e
)
(m
e
)
m
,
1
12
1
12
1
2
4 |r|3
(1.11)
where e12 denotes the unit vector along r, which is the vector joining m1 and
m2 .
For a continuous medium, the magnetostatic dipolar field can be calculated
by solving the Maxwells equations

B = 0 Hdip + M = 0
(1.12)
Hdip = 0
with suitable boundary conditions.
(3) Exchange interaction
The exchange interaction is a purely quantum mechanical effect of identical
particles with a definite parity with respect to particle exchange [2, 7]. For electrons, it is a natural result of the electostatic interaction and Pauli principle,
which requires the total wavefunction of electrons be antisymmetric. Within the
Heisenberg model, the energy between two atomic spins is given by Eq. 1.1. To
describe long wavelength SWs in a crystal, the Heisenberg model can be reformulated in a continuum form with the following energy density expression [3, 8]
A
(M )2 ,
MS2
Eex =
(1.13)
where A = 2nJ0 S 2 /a is the exchange constant, and n = 1, 2, or 4 for sc, bcc, or

fcc lattice, respectively.
Therefore, the effective exchange field can be derived as
Hex =
1 Eex
0 M
=
=
2A
M
0 MS2
!
(1.14)
2A
2 M ,
0 MS2
where the last line is valid for homogeneous materials.

(4) Magnetic anisotropic energy
Here, the magnetic anisotropy refers to magnetocrystalline anisotropy and surface/interface anisotropy. Shape anisotropy is described by the magnetic dipolar
interaction. The magnetocrystalline anisotropy, which mostly originates from
relativistic spin-orbit coupling in a crystal, tends to orient the magnetization
direction along specific crystalline axes.
In this thesis, the most relevant anisotropy is uniaxial anisotropy, with the
following energy density [3]
EK = K1 sin2 + K2 sin4 ,
(1.15)
1. INTRODUCTION
where is the angle between the magnetization and axis of anisotropy, and K1 ,
K2 the anisotropy constants. If K2 = 0, Eq. 1.15 describes an anisotropy type of
either easy axis (K1 > 0) or easy plane (K1 < 0, hard axis).
Due to broken symmetry, spins at surfaces or interfaces has asymmetric nearest neighbors, resulting in an exchange energy different from those in the bulk [9].
Phenomenologically, the surface energy term usually tends to align the magnetization either parallel or perpendicular to the film plane. By defining an effective
volume anisotropy constant Keff = KS /d, where KS is the surface anisotropy
constant and d the film thickness, surface anisotropy can be described by the uniaxial anisotropy to the lowest order (See Eq. 1.15). To study SWs in thick films,
surface anisotropy can be treated as an effective pinning boundary condition [10].
Numerous other types of magnetic anisotropies [3] will not be encountered in
this thesis and therefore are omitted here.
Besides these four types of interactions, other interactions can also be included
to describe a particular system. For instance, in Chapter 5 we will take the
Dzyaloshinskii-Moriya interaction into consideration, the detail of which will be
discussed therein.
1.1.3
Spin Waves in Thin Films
The spin-wave dispersion relation of a thin film is important for understanding the band structures of magnetic nanowires and magnonic crystals (see Section
1.1.4). For an in-plane magnetized homogeneous film with thickness d, the frequency of a spin-wave mode with wavevector k is dependent on the angle
between k and the magnetization as

s
(k, ) =
2A 2
H0 +
k + (kd)MS sin2
0 MS

2A 2
H0 +
k + MS (kd)MS
0 MS
(1.16)
where magneticrystalline anisotropy has been neglected, and (x) = 1 (1

e|x| )/ |x| is positive. Note that the above dispersion relation is obtained by
assuming that the distribution of the dynamic magnetization of the spin-wave
modes across the film thickness is uniform. If the wave propagation direction is
perpendicular to the magnetization (i.e. = /2), the spin wave mode is call
Damon-Eshbach (DE) wave, as the corresponding dispersion relation, if exchange
interaction is ignored (long-wavelength limit), is very similar to that of the dipolar
surface spin wave, which was first derived by Damon and Eshbach [11]. In the
other case when = 0, the spin-wave is called backward-volume (BV) wave,
because near the Brillouin zone center (k 1), the sign of its group velocity is
opposite to that of its phase velocity. The word volume indicates that the wave
amplitude varies sinusoidally across the film thickness (of which the dispersion
relation describes the lowest order mode), resembling that of volume spin waves
subject to certain boundary conditions, which is different from the exponential
variation of surface waves.
Based on the above discussion, even for homogeneous materials without magnetic anisotropy, the frequencies of SWs are generally dependent on their propagation direction when the magnetization is in-plane. The propagation of SWs
is isotropic if the magnetic films magnetization is perpendicular to the plane, in
which case the magnetization (and the applied magnetic field) does not break the
in-plane rotational symmetry. This mode is call the forward-volume mode, since
1. INTRODUCTION
the phase and group velocity are in the same direction.
1.1.4
Magnonic Crystals
It is well-known in solid state physics that band gaps will form within a medium
with periodically modified properties, such as electrons in crystals. This concept
has long been employed to design artificial crystals to fabricate noval materials with properties that cannot be found in conventional materials. The most
well-known example is the photonic crystals, which are structures with periodic
refractive indices. Based on photonic crystals, many useful and interesting phenomena of light have been realized, such as slow light, guided light propagation at
the nanoscale, photon storage, and all optical switches [12]. Similarly, by varying
the magnetic properties of a material, one can fabricate the so-called magnonic
crystals (MCs), which give rise to forbidden energy gaps for SWs .
Although not as mature as photonic crystals, MCs are believed to be first studied earlier than the emergence of photonic crystals in 1987 [13], as periodically
modulated magnetic materials has been employed in microwave technology for
near 40 years [13]. It was not until the recent advancement in nano-fabrication
technology in the last two decades that MCs attracted enormous research interests. Nanoscale MCs allow the manipulation of microwaves at the nanoscale,
which has tremendous application potentials. Many concepts of photonic crystals
have been applied to MCs, and the operational principles of optical devices based
on photonic crystals generally can be employed to design spin-wave devices based
on MCs. However, MCs exhibit much richer physics due to the unique properties
and merits of SWs, such as strong coupling to an external magnetic field, intrinsic
10
Figure 1.3: Band structure of a Co(250nm)/Py(250nm) MC measured

by BLS. Experimental results are represented by dots. Color shaded areas denote
band gaps.
high nonlinearity, and scalability.

In the following, recent representative experimental progresses of nanosized
MCs will be briefly summarized. In 2007, Gubbiotti et al. measured the band
structures of 1D thin-film MCs consisting of dipolarly coupled nanoslabs [14].
In their results, several dispersive collective modes and band gaps were clearly
observed. Because the nanoslabs are coupled by the magnetic dipole-dipole interaction, the collective behavior of these modes will vanish when the geometrical
parameters of the MCs are scaled down to tens of nanometers, for which the local
exchange interaction will dominate. In 2009, Wang et al. reported the obser-
11
1. INTRODUCTION
vation of similar behavior in a 1D MC composed of alternating and contacting
Ni80 Fe20 /Co nanostripes (See Fig. 1.3) [15]. Later work by Zhang et al. discussing the ferromagnetic and antiferromagnetic alignment between neighboring
nanostripes confirmed that a strong exchange interaction at the Py/Co interfaces is necessary to correctly interpret their observations [16]. As the Py and
Co stripes are coupled by both exchange and dipolar interactions, the results are
thus scalable. Later, efforts have been devoted to fabricating 2D MCs [17, 18].
However, complete band gaps in all directions of the reciprocal space have not
been observed. One interesting application of MCs was reported by Chi et al.
[19], where a line defect was introduced into a 2D MC. The line defect then can
serve as a spin-wave waveguide for frequencies that lie within the band gaps of
the MC.
1.2
1.2.1
Brillouin Light Scattering

Physics of BLS
Brillouin light scattering (BLS) is the inelastic scattering of photons through

annihilating (anti-Stokes process) or creating (Stokes-process) low-energy quasiparticles, such as phonons, magnons and polarons [20]. Based on the conservation of energy and momentum (Eq. 1.17), analysing the resultant energy and
momentum shifts of scattered photons provides information on the energy and
momentum of the elementary excitations studied.
~S = ~I ~
~qS = ~qI ~q
12
(1.17)
1.2 Brillouin Light Scattering
S, qS
S, qS
I, qI
I, qI
, q
, q
Figure 1.4: Brillouin light scattering processes by magnons. left panel:

anti-Stokes process; right panel: Stokes process. I (qI ), S (qS ), (q) are the respective frequencies (wavevectors) of the incident light, scattered light, and SWs.
BLS gives information on a sample that is different from that provided by

Raman scattering. BLS measures a relatively larger scale property, such as elasticity and magnetization. In contrast, Raman scattering provides information on
molecular structures by detecting vibrational and rotational motions of atomic
bond. Due to the different underlying mechanisms, the energy shift in BLS is typically lower than that of Raman scattering by three orders of magnitude. Because
the frequency shift of light is very small (typically < 1000 GHz), a high-resolution
Fabry-Prot (FP) interferometer is usually employed to analyse the scattered light
in BLS.
1.2.2
Brillouin Spectrometer
Figure 1.5 depicts the main typical components of the BLS experimental system in the 180-backscattering geometry . In our experiments, the 514.5nm radiation of single-mode Ar+ laser is used as the excitation source. The laser beam
is focused onto the sample surface by a convex lens, which also serves to collect
scattered light. Usually, the power of the incident light on the sample surface is
around 100 mW, with an irradiated spot about 50 m in diameter. The scattered
13
Ar+ laser
1. INTRODUCTION
photon detector
FP2
incident light
magnet
polarizer
Sample
FP1
magnet
scattered light
Figure 1.5: Schematics of the optical path of Brillouin light scattering.

Green lines represent the incident and scattered laser light.
light then passes through a polarizer in ps polarization, which filters out inelastic
light scattering from magnons, and is analysed by the FP interferometer. The
transmitted light is then detected by a 1024-channel silicon avalanche diode detector. The FP interferometer is also equipped with a control unit and interfaced
to a PC for data acquisition and storage.
1.2.3
Intensity of Brillouin Light Scattering
Microscopically, light scattering by solids originates from the change in the

motion of charges subject to the incident wave, causing emission of the scattered light. For the description of BLS at room temperature (RT), a quantum
mechanical approach will generally give the same results as macroscopic methods.
It has been shown that when the frequency shift of light is relatively small,
i.e. I S , the differential light scattering cross section is related to the
correlation function of the permittivity of the medium by the following relations
14
1.2 Brillouin Light Scattering
[21]
d
h i,q
dd Z Z
=
h (t1 , r1 )(t2 , r2 )iei(tqr) dtdV
Z Z
hm (t1 , r1 )m(t2 , r2 )iei(tqr) dtdV
(1.18)
where h. . . i represents statistical average, the dynamic fluctuation in the permittivity of the medium, t = t1 t2 , and r = r1 r2 . is proportional to
the dynamic magnetization m due to the magneto-optical effect [20], resulting in
the last equation. Equation 1.18 indicates that the differential scattering cross
section is proportional to the Fourier component of the dynamic magnetization
[22] through

2
d
m,q .
dd
(1.19)
where m,q is the eigenmode of the dynamic magnetization.

For light scattering by magnons in magnetic thin films, the translational symmetry in the out-of-plane direction is broken. Therefore, the momentum in Eq.
1.17 is no longer a fully conserved quantity [22]. In other words, only the component that is parallel to the film plane is conserved. Consequently, by varying the
incident angle of light (and therefore the momentum transfer in the film plane) in
the back-scattering geometry, spectrum of SWs with different wavevectors can be
collected, which allows mapping the dispersion relation. This result will be extensively employed in this research for magnetic thin films and MCs with periodic
in-plane structures.
15
BIBLIOGRAPHY
Bibliography
[1] The royal swedish academy of sciences, the nobel prize in physics
2007, http://www.nobelprize.org/nobel_prizes/physics/laureates/
2007/popular-physicsprize2007.pdf (2007). 1
[2] N. W. Ashcroft and N. D. Mermin, Solid State Physics (Harcourt College
Publishers, New York, 1976). 2, 6
[3] S. Chikazumi, Physics of Ferromagnetism (Oxford University Press, New
York, 1997). 4, 7, 8
[4] T. L. Gilbert, IEEE Transactions on Magnetics 40, 3443 (2004). 4, 5
[5] L. D. Landau and E. M. Lifshitz, Phys. Z. Sowjet. 8, 153 (1935). 5
[6] T. L. Gilbert, Physical Review 100, 1243 (1955). 5
[7] D. J. Griffiths, Introduction to Electrodynamics (Prentice Hall, 1999). 6
[8] M. Krawczyk, M. L. Sokolovskyy, J. W. Klos, and S. Mamica, Advances in
Condensed Matter Physics 2012, 14 (2012). 7
[9] A. Aharoni, lntroduction to the Theory of Ferromagnetism (Oxford University Press, 2007). 8
[10] B. A. Kalinikos and A. N. Slavin, Journal of Physics C: Solid State Physics
19, 7013 (1986). 8
[11] R. W. Damon and J. R. Eshbach, Journal of Physics and Chemistry of Solids
19, 308 (1961). 9
[12] J. D. Joannopoulos, S. G. Johnson, J. N. Winn, and R. D. Meade, Photonic
Crystals: Molding the Flow of Light (Second Edition) (Princeton University
Press, Princeton, 2008), 2nd ed. 10
[13] M. Krawczyk and D. Grundler, Journal of Physics: Condensed Matter 26,
16
BIBLIOGRAPHY
123202 (2014). 10
[14] G. Gubbiotti, S. Tacchi, G. Carlotti, N. Singh, S. Goolaup, A. O. Adeyeye,
and M. Kostylev, Applied Physics Letters 90, 092503 (2007). 11
[15] Z. K. Wang, V. L. Zhang, H. S. Lim, S. C. Ng, M. H. Kuok, S. Jain, and A.
O. Adeyeye, Applied Physics Letters 94, 083112 (2009). 12
[16] V. L. Zhang, H. S. Lim, C. S. Lin, Z. K. Wang, S. C. Ng, M. H. Kuok, S. Jain,
A. O. Adeyeye, and M. G. Cottam, Appl. Phys. Lett. 99, 143118 (2011). 12
[17] S. Tacchi, F. Montoncello, M. Madami, G. Gubbiotti, G. Carlotti, L. Giovannini, R. Zivieri, F. Nizzoli, S. Jain, A. O. Adeyeye, et al., Physical Review
Letters 107, 127204 (2011), pRL. 12
[18] S. Tacchi, G. Duerr, J. W. Klos, M. Madami, S. Neusser, G. Gubbiotti,
G. Carlotti, M. Krawczyk, and D. Grundler, Physical Review Letters 109,
137202 (2012), pRL. 12
[19] K. H. Chi, Y. Zhu, and C. S. Tsai, Journal of Applied Physics 115, 17D125
(2014). 12
[20] M. G. Cottam and D. J. Lockwood, Light Scattering in Magnetic Solids (New
York, 1986). 12, 15
[21] L. D. Landau and E. M. Lifshitz, Electrodynamics of Continuous Media
(Pergamon Press, 1984). 15
[22] J. Jorzick, S. O. Demokritov, C. Mathieu, B. Hillebrands, B. Bartenlian, C.
Chappert, F. Rousseaux, and A. N. Slavin, Physical Review B 60, 15194
(1999). 15
17
BIBLIOGRAPHY
18
2
Micromagnetic Simulations
2.1
Introduction
Micromagnetic simulation is used to predict magnetic behaviours or to analyse

experimental results. The typical length scales involved are usually large enough
to ignore the atomic discretization of the material, yet small enough to resolve
the magnetization inhomogeneity. Therefore, micromagnetic simulation is ideal
for predicting and analysing the spin dynamics of magnetic structures in the
nanometer and micrometer scales.
2.2
OOMMF Approach
Throughout this thesis, the object-oriented micro-magnetic framework (oommf)

[1] has been used to solve the LLG equation in the time domain. Being a free
and open source, oommf has gained wide popularity and has become the de
facto standard software for micromagnetic simulations, which has been used as a
19
2. MICROMAGNETIC SIMULATIONS
reference of performance and accuracy by many newly developed micromagnetic
packages. Oommf models a magnetic system by the finite difference method
(FDM), for which the whole space is modelled as a large box that is divided into
many small identical cuboid cells, with the magnetization assumed to be uniform
within each cell.
The average magnetostatic field within a cell is calculated from
Z
Hdip (r) =
N (r r 0 ) M (r 0 )d3 r,
(2.1)
where N (r r 0 ) is the demagnetizing tensor which remains constant for the

same mesh discretization. According to the convolution theorem, calculation of
the convolution given by Eq. 2.1 can be accelerated by a fast Fourier transform
(FFT) [2], a procedure already incorporated in oommf. Since calculation of
the demagnetizing field is the most time-consuming part, for the same number
of degrees, oommf is usually more efficient than finite-element methods (FEM)
based on non-uniform mesh discretization (see Section 2.3). Moreover, finite
difference approaches use much less memory than does FEM.
In this thesis, only linear SWs are of interest, which requires a stable equilibrium state. Therefore, an oommf simulation usually starts with obtaining a
desired equilibrium magnetization. Then an excitation magnetic field is applied
to certain regions of the system studied to excite propagating SWs. Depending
on the purpose of simulation, the function representing the excitation field may
vary. For calculations of the dispersion relation, a magnetic field pulse within a
limited frequency range is usually applied. This is normally obtained with a field
20
2.3 Finite-Element Approach Using COMSOL
of the following form

Hpulse sinc(0 t) = Hpulse
sin(0 t)
,
0 t
(2.2)
where Hpulse is a vector constant. In contrast, if a SW of only a particular

frequency is of interest, a sine function or a sine function with a Gaussian envelope
is applied instead.
Instantaneous magnetizations are saved during the simulation with a fixed time
step. Subsequent data analysis and visualization are implemented in matlab
and Python. To obtain dispersion relations, a Fourier transform of the obtained
magnetization is performed in both space and time. To retrieve mode profiles, the
distribution of dynamic magnetization of a specific eigenmode, Fourier transform
is performed only in the time domain.
2.3
Finite-Element Approach Using COMSOL
COMSOL Multiphysics [3] is a user friendly FEM software, with a well designed graphical user interface (GUI) that offers rich features. The user can create
the geometrical domain with the tools provided, choose the underlying physical
equations from a library, input model parameters, and set the types of problem
to be solved (e.g. eigenvalue, time dependent, or stationary, etc). After the
above preparation, the software takes care of the details in tedious tasks, such
as mesh creation, choosing suitable solver, and performing the calculation. It is
possible to specify the parameters of nearly all aspects of a particular problem,
including mesh size, shape of mesh cells, customized physical equations, boundary
21
conditions (periodic boundary conditions are allowed), algorithms of the solver,
tolerance of numerical error, and the maximum number of iterations. Data visualization tools are also included, which work seamlessly with all other parts of
the software. Being simple to use, comsol also provides application program
interfaces (APIs) for matlab, which enables users to retrieve almost all data
(such as values of physical quantities interpolated at arbitrary coordinates) from
comsol to matlab and thus perform complex data processing using the full
power of the later. Moreover, combined with the comsol server for matlab,
this API allows the variation of model parameters and control over the process of
running a comsol model within matlab. This feature is very useful for automatically running a batch of tasks and changing model parameters on the fly. In
the following, we will demonstrate how to solve the LLG equation using comsol
for both time-evolution and eigenfrequency problems.
2.3.1
Brief Introduction to Finite Element Method
2.3.1.1
Weak Formulation: An Example
Consider the following 1D problem:

00
f (x) = g(x) in [0, a],

0
(2.3)
f (0) = c1 , f (a) = c2 ,
where f (x) is unknown while g(x) is given, a, c1 and c2 being constants. Suppose
f (x) solves the above problem, then for any smooth function t(x) (called test
1
For details, please refer to Ref. [4].
22
function), we must have

Z
00
f (x)t(x)dx.
0
Z a
0
0
0
a
f (x)t (x)dx
= f (x)t(x)|0
0
Z a
0
0
f (x)t (x)dx.
= c2 t(a) c1 t(0)
g(x)t(x)dx =
0
(2.4)
Our objective is to find a solution to f (x) that satisfies Eq. 2.4 for any test
function that makes sense in this equation. Conversely, it can be shown that the
requirement that Eq. 2.4 is valid for all admissible test function t(x) is so strong
that the solution f (x) to Eq. 2.4 is virtually the same as that of the original
PDE, for which the two solutions are called weak solution and strong solution,
respectively. In general, the strong solution satisfies the original PDE exactly at
every point. The weak solution solves the original PDE in an integral sense, i.e.
it satisfies the strong form on average over the domain.
Actually, the weak form can approximate the original problem to any precision.
Let us examine the first line of Eq. 2.4. If we take the test function to be nonzero
only within a very narrow range, for example
c, if x0 < x < x0 +
t(x) =
0, otherwise
(2.5)
where c is a nonzero constant, a very small positive number, and x0 [0, a].
The integral equation then becomes
Z
x0 +
x0 +
g(x)dx = c
x0
x0
23
00
f (x)dx.
(2.6)
This equation approximates the original PDE in an average sense over the interval
[x0 , x0 + ]. If the parameter approaches to 0, Eq. 2.6 becomes equivalent to
the original PDE near x0 . If we take the test function to be Dirac delta function,
we immediately recovers the original PDE from the weak form. Another point
to mention is that the strong solution has to be differentiable to second order,
which is relaxed to first order for the weak solution.
It is noteworthy that in the weak form 2.4, the Neumann boundary condition
(first order derivative) naturally comes into the equation. A little more effort is
required to implement the Dirichlet boundary conditions (fixed boundary condition), which will not be covered here.
Now, let us approximate the unknown function f (x) and test function t(x) by
expanding them in a finite basis function space {i }N
i=1
f (x) =
t(x) =
i=N
X
i=1
i=N
X
ci i (x),
(2.7)
di i (x),
i=1
where the coefficients ci are to be solved, and di are arbitrarily chosen. Substituting Eqs. 2.7 into the weak form 2.4, we have
Z
i=N
aX
j=N
0
ci i (x)
i=1
j=N
dj j (x)dx = c2 t(a) c1 t(0)
g(x)
j=1
dj j (x)dx, (2.8)
j=1
which is
j=N
X
j=1
dj
i=N
X
i=1
j=N
0
i (x)j (x)dx = c2 t(a) c1 t(0)
ci
0
X
j=1
24
Z
dj
0
g(x)j (x)dx. (2.9)
Because dj can take any value, the above equation leads to

i=N
X
i (x)j (x)dx = c2 t(a) c1 t(0)
ci
i=1
g(x)j (x)dx.
(2.10)
Therefore, we arrive at the following linear problem

(2.11)
Kc = g,
where Kij =
Ra
0
i (x)j (x)dx, gi = c2 t(a) c1 t(0)
Ra
0
g(x)i (x)dx, vector c to
be solved for.
2.3.1.2
Mesh Descritization & Basis Functions
In the finite-element approach, a space domain under study is discretized into

sub-domains, i.e., tetrahedra for 3D problems, triangles or quadrilaterals for 2D,
and line segments for 1D. Sizes of these sub-domains are usually unevenly chosen, with relatively dense segmentations at sub-domains with fine structures or
where the objective variables are expected to vary steeply. This feature has the
advantage that structures with irregular or curved edges can be resolved more
accurately than finite difference, provided that the numbers of mesh grids are the
same.
For better computing efficiency, the basis functions are usually chosen to
be piecewise linear functions or piecewise polynomials that are highly localized
around mesh-grid points. In 1D problems, the basis functions can be chosen to
25
Figure 2.1: Triangular mesh grid of an irregular 2D structure. The mesh

can adapt itself to both straight and curved boundaries. If we are more interested
in a physical field around the circle (green area), finer meshes can be used, as shown
in the figure.
k k+1
xk-1 xk xk+1 xk+2
xl
Figure 2.2: Illustration of 1D piecewise linear basis functions. Only

neighboring basis functions (k and k+1 in the figure) that overlap possess nonzero
inner product.
be the following triangular functions
i (x) =
xxi1
,
xi xi1
if xi1 < x < xi
xk+1 x
, if xi < x < xi+1
xi xi1
0, otherwise
(2.12)
The inner product of these basis functions are mostly zero except for nearest
R
neighbours, i.e. i (x)j (x)dx = 0 if |i j| > 1 (see Fig. 2.2 for an example).
26
1
0.8
0.6
0.4
0.2
0
0.2
1
0.5
0.5
1.5
2.5
Figure 2.3: Sinc function approximated by linear combination of triangular basis functions. Dashed blue and solid black lines are the approximating
and real sinc functions, respectively. Scaled triangular basis functions are plotted
in different colors. Note that the mesh grid is not uniform.
Therefore, the matrix elements Kij of Eq. 2.11 are mostly zero and we have a
sparse matrix K, which is typically tridiagonal and can be handled efficiently
using many linear algebra packages. Figure 2.3 shows how a 1D function can be
approximated by linear combinations of piecewise linear base functions.
The main difference between FDM and FEM is summarized here. FDM discretizes domain of a PDE to identical grids, which will face efficiency problems
when dealing with complex geometries. It approximates partial derivatives using Taylor expansion and evaluates functions only at the grid points. Domain
discretization in FEM has more freedom and allow efficient representation of complex irregular structures. By solving PDE in the integral form (weak form), the
order of partial derivatives can be reduced. In FEM, functions are approximated
by linear combinations of basis functions from a finite function space. Therefore,
functions are defined continuously in the whole domain (not just on grid points).
Another consequence is that partial derivatives are evaluated exactly.
27
2.3.1.3
LLG Equation in Weak Form
Solving the LLG equation in time domain using comsol is generally slower
than using oommf. However, it is necessary to obtain the equilibrium magnetization used as input to the frequency-domain problem, as discussed in the next
Section 2.3.2. Rewrite the LLG equation 1.9 to the following form
dn
dn
= n Heff + n
,
dt
dt
(2.13)
where n = M /MS is the unit vector along the magnetization. For brevity, the
function arguments r and t have been omitted, for instance, n n(r, t). Suppose
the test function can be written as
(2.14)
Taking the inner product of Eqs. 2.13 with the test function 2.14 and integrating
over the whole domain gives
Z
dn
d +
dt
Z
(n Heff )d
(n
dn
)d = 0
dt
(2.15)
For brevity, I will assume that the total effective internal field Heff consists of
only exchange field, demagnetizing field, and applied field:
Heff = J2 n + Hdip + H0 ,
28
(2.16)
where the first term on the right hand side is the exchange field written as a
function of the unit vector of the magnetization, and J =
2A
.
0 MS
Although the
dipolar field Hdip is also a function of the magnetization, its calculation will be
conducted in another equation that is coupled to the LLG equation (refer to
Section 2.3.3 for details).
To reduce the order of derivative in the exchange field, we will use the divergence theorem as follows
Z
(n 2 n)d
X Z
2 np (n ep )d
p={x,y,z}
Z

Z
h

i

np (n ep ) d np (n ep ) d
p={x,y,z}
Z

lS np (n ep ) dS

np (n ep ) d ,
p={x,y,z}
(2.17)
where lS is the normal vector of the surface element dS. We immediately find
that the term lS np (=
1 Mp
),
MS lS
proportional to the normal derivative of the
magnetization, corresponds to surface boundary conditions of the magnetization.

Now, both derivatives of the test function and the unit vector of magnetization
29
are first order. We may further simplify the second integrand of Eq. 2.17 as
X

np (n ep )
p={x,y,z}
np [(n ep ) ]
xq
xq
p={x,y,z} q={x,y,z}
!
X
X np

n ep + n ep
=
xq
xq
xq
p={x,y,z} q={x,y,z}
X
X np

n X np
ep + n
ep
=
xq
xq
xq
xq
p={x,y,z}
q={x,y,z}
p={x,y,z}
!
X
n
=
.
n
xq
xq
=
q={x,y,z}
(2.18)
2.3.2
Frequency Domain
In this thesis, much effort has been devoted to the calculation of dispersion
relations of linear SWs in ferromagnetic thin films, nano-stripes, and magnonic
crystals. This is achieved by solving the linearized LLG equation for a system
subject to certain boundary conditions. The magnetization can be decomposed
into two components, the static and dynamic magnetizations:
M (r, t) = M0 (r) + m(r)eit ,
(2.19)
where |m| |M0 |, the static component satisfies |M0 | = MS , MS being the
saturation magnetization. The dynamic magnetization m(r)eit is generally
a rotating vector, with an elliptical polarization due to magnetic anisotropies.
30
Therefore, writing the dynamic magnetization as a product of spatial and temporal components as in Eq. 2.19 requires that m(r) be a vector of complex
numbers, the real components of which represent the actual values of dynamic
magnetizations. Similarly, we can decompose the demagnetizing field and the
exchange field as follows:
Hdip (r, t) = Hdip,0 (r) + hdip (r)eit ,
(2.20)
and

J
Hex (r, t) =
M (r, t) = Hex,0 (r) + hex (r)eit .
MS
(2.21)
Substituting Eq. 2.19-2.21 into the LLG equation 1.9 without damping term,
we have the following linearized eigenfrequency equation [5]

im
J
+M0 hdip +M0

m+mH0 +mHdip,0 +mHex,0 = 0,
MS
(2.22)
where we have assumed the equilibrium condition

M0 H0 + Hex,0 + Hdip,0 = 0
(2.23)
and have neglected second order terms of the dynamic magnetization, i.e.
m (hex + hdip ) = 0.
(2.24)
Both Hex,0 and Hdip,0 of Eq. 2.22 are independent of the dynamic magneti-
31
zation m and are calculated prior to solving the eigenvalue problem. Concretely,
the same system is first relaxed to a ground state in the time domain to obtain the
equilibrium magnetization M0 , wherefrom the fields Hex,0 and Hdip,0 are evaluated. These quantities will serve as input to the above-mentioned eigenfrequency
problem. In contrast, being dependent on m, the calculation of hdip should be
calculated by equations that are coupled to Eq. 2.22, as discussed shortly in
Section 2.3.3.
2.3.3
Calculation of Demagnetizing field
For a complete description of the above comsol method, we discuss the calculation of the demagnetizing field in this subsection. Because the typical propagation velocity of SWs is much lower than that of electromagnetic waves, the
problem should be treated within the quasi-magnetostatic approximation [10000].
The demagnetizing field is expressed as
Hdip (r) = (r),
(2.25)
where is the magnetic scalar potential. Together with B = 0 (M + Hdip ) and

B = 0, we have (see also Eq. 1.12)
2 = M .
(2.26)
Therefore, for a time-domain problem, each instantaneous magnetization at

a time step creates a demagnetizing field, which is then substituted to the LLG
equation to solve magnetization at the next time step. This process is repeated
32
iteratively, evolving the magnetization forward in time. For the eigenfrequency

problem, the solver treats Eq. 2.26 and the linearized LLG equation in a fully
coupled way.
2.3.4
Frequency Domain in 2D
In Section 2.3.2, solving eigenfrequency problems in general 3D cases is discussed. In this Section, I will focus on the same topic in special systems that could
be reduced to 2D problems. I will take the MC in Chapter 4 as an example.
With a finite thickness along the z-direction, the MC exhibits a 1D periodicity
in the x-direction and a continuous translational symmetry (homogeneous and
infinitely long) along the y-direction (see Fig. 4.1). Suppose both the applied field
H0 and the magnetization M0 are along the y-axis, then we have the following
relations:
Hdip,0 = 0
Hex,0 = 0
m ey = 0, meaning m = (mx , 0, mz )
(2.27)
hex ey = 0
hdip ey = 0
where ey is the unit vector along the y-axis. Using equations 2.27, equation 2.22
can be simplified to be
i
mx + J2 mz + MS hd,z = 0
i
mz J2 mx MS hd,x = 0
33
(2.28)
BIBLIOGRAPHY
where the equation corresponding to the y-component becomes identity.

Reducing the system to 2D has a few obvious benefits. First, the number
of degree of freedom is vastly decreased, thus reducing the need on both CPU
time and memory by many folds. Second, spin-wave modes that has variations
along the y-axis will all be suppressed. This greatly simplifies the analysis of the
theoretical results when compared with BLS experiments. This is because only
modes with zero wavevector along the y-axis can be detected by BLS when the
incident plane of laser is in the xz-plane, which is the most interesting case.
Bibliography
[1] M. Donahue and D. G. Porter, OOMMF Users Guide, Version 1.0, Interagency Report NISTIR 6376 (NIST, Gaithersburg, MD, USA) (1999). 19
[2] N. Hayashi, K. Saito, and Y. Nakatani, Japanese Journal of Applied Physics
35, 6065 (1996). 20
[3] COMSOL Multiphysics, COMSOL AB, Stockholm, Sweden. (Version 4.0). 21
[4] O. Zienkiewicz, R. Taylor, and J. Zhu, The Finite Element Method: Its Basis
and Fundamentals (Elsevier, Singapore, 2013). 22
34
3
Band Structure of a 1D
Width-Modulated
Nanostripe Magnonic
Crystal
3.1
Introduction
In this Chapter, results of simulations of the band structure of a 1D nanostripe MC will be presented. Its geometrical structure is depicted in Fig. 3.1.
Phononic and photonic crystals with this width-modulated-stripe structure have
been studied. Lee et al. [1] were the first to theoretically investigate the magnonic
dispersion spectra, which showed interesting properties and triggered experimental research [2]. The oommf-based simulation method for calculating the band
structures of MCs reported by Lee et al. was widely adopted by the magnonic
community. However, this method, if not properly implemented, will lead to in-
35
3. BAND STRUCTURE OF A 1D WIDTH-MODULATED

NANOSTRIPE MAGNONIC CRYSTAL
Figure 3.1: Geometric structure of the MC, where a = 18 nm, P1 = P2 = 9 nm,

t = 10 nm, w1 = 24 nm, and w2 = 30 nm.
complete energy band diagrams of MCs and hence result in incorrect conclusions
[3]. Unfortunately, this consequence is present in many theoretical papers based
on this method [46], including Lee et al.s paper . As an example, we will consider the width-modulated MC of Lee et al. We will reveal the errors in their
results and present the complete magnonic dispersion relations, as well as three
different approaches to calculate them. Finally, ways to tune the band structure
of this MC will be explored.
3.2
The Complete Band Structure
We will focus on the calculation of the dispersion relation of the MC with a

period of 18 nm. The full band structure was calculated based on three different
theoretical approaches, namely, a microscopic approach, oommf simulations, and
a method based on the linearized Landau-Lifshitz (LL) equation, which yield
consistent predictions. Further, we provide a physical interpretation of the origin
of the incomplete band structure using group theory [7].
36
3.2 The Complete Band Structure
In all the calculations, the same magnetic parameter values were used as those
in Ref. [1], with the saturation magnetization MS , exchange constant A and
gyromagnetic ratio of Permalloy set to 8.6 105 A/m, 1.3 1011 J/m and
2.21 105 Hzm/A, respectively, magnetocrystalline anisotropy and surface pinning neglected. The oommf approach and the linearized LL method using FEM
has already been discussed in Chapter 2. It is to be noted that calculation in the
latter approach was done within only one unit cell for a wavevector kx with the
following Bloch-Floquet boundary conditions
(x + a) = (x)eikx a
(3.1)
ikx a
m(x + a) = m(x)e
where and m are the dynamic component of the magnetic scalar potential and
magnetization, respectively. In obtaining the equilibrium magnetization state,
application of the periodic boundary conditions (x+a) = (x) and M (x+a) =
M (x) automatically guarantees the periodicity of the exchange field. In the
microscopic method, the waveguide structure was represented in terms of a large
number of cubic cells, each with an effective spin, and a Hamiltonian approach
was adopted for the interacting spin system including the exchange and dipoledipole terms. Discrete-lattice dipole sums were performed, instead of utilizing
Maxwells equations, with the cell size being less than the exchange length. For
details of the microscopic approach developed by M. G. Cottam, please refer
to [810]. Both these two methods actually solve the problem in the frequency
domain and thus do not depend on using an excitation field.
Interestingly, the dispersion curves (see Fig. 3.2) contain branches, labeled
37

Figure 3.2: Band structure and symmetry assignment of the MC. (a)
Dispersion relations for the waveguide obtained from linearized Landau-Lifshitz
equation method (red dashed lines), microscopic calculations (blue dotted lines)
and oommf simulations (green lines). The symmetry assignment of the modes is
indicated. (b) Mode profiles of mz for SW branches (labeled with 1-5) at , and
X points. (c) Instantaneous spatial profiles (y-z cross-section) of the excitation
field, with different symmetries (A, B, and both). The shaded area denotes the
section of the waveguide where the field is applied.
2 and 4, not reported in [1]. The presence of these additional modes, which
have A symmetry magnetization profiles (Fig. 3.2b), drastically reduces the first
bandgap width from the reported 11 to 3 GHz and the second bandgap from
16 to 2.2 GHz. As the magnetization is an axial vector, the symmetry of the
magnetic ground state is the C2h group [7]. The groups of the wavevectors at the
, (a general point in the first Brillouin zone) and X points are C2h , C2 , and
C2h , respectively. The symmetry assignment of the modes, based on their profiles
(Fig. 3.2b) of the out-of-plane component mz of the dynamic magnetization, is
presented in Fig. 3.2a.
In our oommf simulations, based on a cell size of 1.5 1.5 10 nm3 , a sinc
38
3.3 Band Structure Modulation by a Transverse Magnetic Field
magnetic field pulse was applied in the y-direction to a 1.5 15 10 nm3 central
section of a 4m-long waveguide (shaded area of bottom panel of Fig. 3.2c). As
the field can be expressed as a linear combination of the basis functions that transform as the A and B irreducible representations of the C2 group, it can therefore
simultaneously excite modes of both A and B symmetries. The dispersion curves,
obtained by performing the Fourier transform of mz in space and time (summing
contributions from all cells), are included in Fig. 3.2a. Further calculations confirm that only A (B) symmetry modes can be excited by an excitation field of A
(B) symmetry (Fig. 3.2c). Hence, the use of an excitation field of B symmetry
by Lee et al. precluded their observation of A symmetry modes [1, 11] and led to
misleading conclusions for the bandgaps. Consequently we find that the reported
numbers of bandgaps of most of the waveguides studied are wrong. For instance,
the [P1 , P2 ] = [15nm, 15nm] structure has only three rather than the stated five
bandgaps for the same frequency range, as well as significantly narrower bandgap
widths than claimed.
3.3
Band Structure Modulation by a Transverse

Magnetic Field
Studies have shown that magnonic band gaps, the most essential feature of
MCs, are controllable by tuning the geometrical dimensions [1, 1214], lattice
symmetry [14, 15], or constituent materials [13, 16] of the MCs. In this section,
effects of a transversely applied magnetic field on the band structure parameters
of the above-mentioned MC will be discussed.
39

The magnetic field dependence of band structures is of particular interest.
Both theory and experiment reveal that magnonic band gaps can be monotonically shifted and resized in frequency by simply varying the strength of an external
applied magnetic field [17]. The great majority of studies were restricted to cases
where the equilibrium magnetization is nearly uniformly aligned along an external field in either the backward-volume or Damon-Eshbach geometries. There is
to date, however, very little knowledge of the band gaps of MCs for intermediatefield cases where the magnetization is arbitrarily oriented with respect to the SW
wavevector, which makes the field dependence more complex.
Here, employing the three independent theoretical methods mentioned in the
previous section, we present a detailed study of the SW band structure of the
MC with a static magnetic field HT applied transversely (+y direction) to the
long axis of the waveguide system (see Fig. 3.3a). The band structure exhibits
different behaviors depending on whether the field is below or above a critical
value. The widths and center frequencies of magnonic band gaps are shown to
be tunable in a non-monotonic fashion by varying the applied field. Interestingly,
some bandgaps attain their maximal width at a field below the critical value,
when the equilibrium magnetization is in an intermediate configuration. Finally,
results for the excitation efficiency of the SWs by an applied oscillating magnetic
field are also presented and interpreted using group-theory analysis.
Under an increasing transverse field HT applied in the y direction, the equilibrium magnetization of the waveguide is gradually oriented from the x to the
y direction due to the competing demagnetizing and applied field. The simulated average My /MS versus HT plot presented in Fig. 3.3(c), indicates that the
waveguide is almost completely magnetized in the y direction above a critical
40
(a)
y
30
HT
nm
m
10 n
24
nm
P1
(b)
P2
odd
even
odd+even
(c)
<My / MS>
0.5
0
0
0.1
0.2
HT (T)
(d)
0.3
0.4
y
x
0T
0.2 T
0.3 T
Figure 3.3: Geometric and static properties of the MC. (a) Schematic view
of the magnonic crystal waveguide. The shaded blue block represents the computational unit cell used in the finite-element and microscopic calculations. The green
bar indicates the region where the excitation field is applied in the oommf simulations. (b) Instantaneous cross-section profile of the excitation field for oommf.
(c) Average value of the normalized equilibrium transverse magnetization My /MS
calculated as a function of the transverse field strength HT . (d) Ground state
magnetization of one unit cell subjected to various HT .
field of about HC = 0.29 T. This value accords well with the analytically estimated average demagnetizing field of 0.32 T, based on the assumption that the
41

(a)
(b)
(c)
80 Au
80 Au
Frequency (GHz)
60 B
g
A
B
A
Au
B
Bg
1st Bandgap
Bg
HT = 0 T
(d)
20
/a
kx
min
max
z
x
X
Ag
A'
Au
Bu 1st Bandgap
Bg
A"
20
Ag
Bg
HT = 0.2 T
/a
kx
(e)
A'
Bu
Ag
2/a
Bg
A"
40
Au 1st Bandgap
Au
Bg
Au
40
A"
60
Ag
Au
Ag
60 A
g
2nd Bandgap
Bu
40
20
Ag
80 B
u
Bg
min
max
2/a
z
y
/a
kx
(f)
x
X
HT = 0.3 T
min
max
2/a
z
y
x
X
Figure 3.4: (a-c) Magnonic dispersion curves under various transverse magnetic
fields. The blue solid, red dashed and green bold lines are data obtained from
comsol, microscopic and oommf calculations, respectively. The wavevector at
the point is 0.13 nm1 . (d-f) Comsol-simulated mode profiles of mz for HT
= 0, 0.2 and 0.3 T, respectively.
infinitely-long waveguide has a uniform average width of 27 nm and is uniformly

magnetized in the transverse (y) direction. For all values of HT considered, the
magnetization is nearly uniform, since the exchange interaction is dominant due
to the relatively small size of the waveguide.
42
3.3.1
Tunability of Band Structure Parameters
The calculated dispersions of magnons in the MC waveguide under transverse

magnetic field HT = 0, 0.2 and 0.3 T are shown in Fig. 3.4. It is clear that, for
all HT values, the MC waveguide exhibits complete magnonic band gaps arising
from Bragg reflection and anti-crossing between counter-propagating modes. The
frequencies of each branch, at the and X points, as functions of field strength
are presented in Figs. 3.5 (a) and (b), respectively. At the point, instead
of being a monotonic function, the frequencies of all five branches first decrease
for HT < HC and then increase for HT > HC , where HC 0.29 T. Such a
behavior under a transverse magnetic field has been experimentally observed in
ferromagnetic nanowires [18, 19] of rectangular or circular cross-section, for which
the SW wavevector component along the wire axis is zero. Fig. 3.5(b) shows that
at the X point, the curves feature an obvious dip only for branches labeled 2 and
4, and the fields at which a frequency minimum occurs are lower than HC .
Of special interest is how the complete magnonic band gaps change with magnetization as HT is varied. Interestingly, Figs. 3.5 (c) and (d) reveal that the overall variation of the bandgap parameters with increasing field is non-monotonic,
which contrasts with previous reports of MCs [4, 12, 15, 17]. With increasing
field, the width of the first band gap first increases and then decreases. It remains constant after the equilibrium magnetization is totally aligned along the
y direction. We note that a maximum value of 7.1 GHz is attained at about
HT = 0.2 T, when the average angle between M0 and the x axis is about /4.
By comparison, the second band gap decreases monotonically and vanishes above
HT = 0.25 T. The center frequency of the first (second) band gap monotonically
43

(b)
80 4
Frequency (GHz)
Frequency (GHz)
(a)
60 3
40
20
0
0
80
60
40
20
0
0
0.2
0.4
H (T)
T
10
1st gap
5
nd
2 gap
0
0
(d)
Gap center (GHz)
Gap width (GHz)
(c)
0.2
0.4
H (T)
80
60
40
20
0
0
0.2
0.4
HT (T)
2nd gap
1st gap
0.2
0.4
HT (T)
Figure 3.5: The transverse field dependences of (a) mode frequencies at the
point, (b) mode frequencies at the X point, (c) complete band gap widths, and (d)
center frequencies of band gaps.
increases (decreases) as the field increases.

It should be noted that such band gap tunability is mainly a result of the fieldtunable separation between center frequencies of different gap openings induced
by coupling between counter-propagating modes [1], instead of the tunable widths
of the gap openings. We illustrate this with respect to the first band gap (see Fig.
3.6). In Ref. [3], it is reported that without a static magnetic field, only evensymmetry branches 1, 3, and 5 (odd-symmetry branches 2 and 4) will be excited
by an even (odd) excitation field, meaning that there is no coupling between
the even and odd modes. Therefore, gap openings appears between bands with
the same symmetry, e.g., between 1 and 3 or 2 and 4. When HT = 0 T, the
gap opening is 9.7 GHz between branches 1 and 3, and 21.5 GHz between
44
(b)
Frequency (GHz)
(a)
60
20
10
0
0
40
(c)
20
Frequency (GHz)
Frequency (GHz)
80
30
HT = 0 T
0.2
0.4
HT (T)
50
40
30
0
0
/a
kx
2/a
0.2
0.4
HT (T)
Figure 3.6: Illustration of field-tunability of band gaps. (a) Formation

of the first complete band gap as the frequency overlap between the first (green
hatched area) and second (red hatched area) gap openings. (b) Widths and (c)
center frequencies of the first (green circle line) and second (red triangle line) gap
openings.
branches 2 and 4, the frequency overlap of which gives the first complete band
gap of 2.8 GHz. It is clear from Fig. 3.6(b) that the relatively small change
in the widths of the gap openings cannot explain the large variation of the first
complete band gap width for HT < 0.2 T. Fig. 3.6(c) indicates that for HT < 0.2
T, the center frequency of the first (second) gap opening shifts up (down) sharply,
leading to a decreasing center-to-center distance of the gap openings, which should
be the main reason for the observed band gap tunability. A similar discussion
shows that when HT increases, the two gap openings responsible for the second
complete band gap shift away from each other, causing the band gap to decrease
and finally to close fully.
45

Table 3.1: Character tables of symmetry groups (after Ref. [7])
C2h
Ag
Au
Bg
Bu
C2
A
B
Ci
Ag
Au
C1h
A0
A00
C1
A
3.3.2
E
1
1
1
1
C2
1
1
-1
-1
h
1
-1
-1
1
E
1
1
E
1
1
E
1
1
i
1
-1
1
-1
C2
1
-1
i
1
-1
h
1
-1
E
1
Symmetry Assignment and Excitation Efficiency
The magnon modes can be classified based on their mode profile symmetry
of the dynamic magnetization, as indicated in Fig. 3.4. The respective symmetries of the magnetic ground state [see Fig. 3.3(d)] for HT = 0, 0.2, and 0.3 T
correspond to the respective C2h , Ci , C2h groups, whose character tables [7] are
presented in Table 3.1. The lower symmetry associated with the ground state for
0 < HT < HC precludes the labeling of the branches as even or odd symmetry.
For HT = 0 or HT > HC , the ground state has a higher symmetry as the magnetization is completely saturated along either the x or y direction, respectively. The
symmetry group of the wavevector at the and X points is the same as that of
the ground state, while that at a general point is usually a subgroup, leading to
a lower symmetry. For HT = 0, 0.2, 0.3 T, the symmetry groups corresponding to
46
(a) 0
max
1
2
3
2
(b)
z
y
x
X
Figure 3.7: Eigenmodes of magnetization with phase information. The

(a) absolute values and (b) phases of the dynamic magnetization mz for the respective modes, at the , and X points, of the five dispersion branches in Fig.
3.4. The mode profiles were calculated for HT = 0.2 T.
the wavevector at are C2 , C1 , and C1h , respectively. In all the cases considered,
the axis of rotational symmetry, if one exists, is coincident with the direction of
the transverse field.
Using time-domain oommf simulations, we have earlier established that [3]
(see Section 3.2), for the same MC in the absence of an external field, only A (B)
symmetry branches can be excited by an A (B) symmetry magnetic field [corresponding to odd (even) field in Fig. 3.3(b)] for time-domain oommf simulations.
However, for a transverse field below HC , the above conclusion becomes inapplicable as no point symmetry operation (besides the identity operation) exists
for modes at a general point in the Brillouin zone. This is illustrated by the
absence of symmetry for the corresponding mode profiles. Fig. 3.7 indicates that,
47

15
10
Intensity (arb. unit)
0.20
0.25
0.30
0.00
0.05
0.10
0.15
2
14
10
4
13
10
12
10
10
20
30
40
50
60
70
Frequency (GHz)
Figure 3.8: Magnon spectra, at the point, excited by the same even-symmetry
field and calculated for various HT values. Groups of peaks corresponding to the
five branches are labeled 1, 2, 3, 4 and 5, respectively.
although the amplitude of mz at the point has inversion symmetry, its phase,
on the other hand, has no such symmetry. In the linear excitation regime, the
excitation efficiency for a particular mode m(r, t) by an excitation field h(r, t)
is proportional to the overlap integral [20] of m(r, t) and the torque exerted by
h(r, t):
Z
hm, M0 hi =
m (M0 h) dV = M0
h m dV
(3.2)
where M0 is regarded as effectively uniform since the lateral size of the MC lies
within the exchange-dominated regime, m the complex conjugate of m, and
the integration over volume V is done within regions where the excitation field is
nonzero.
For the regimes of HT = 0 and HT > HC , only modes with the same symmetry
as that of the excitation field have non-vanishing excitation efficiency. In contrast,
48
in the 0 < HT < HC range, there is no simple correspondence between the

symmetry of a mode and that of the excitation field. In this last case, an excitation
field with an even distribution [middle panel of Fig. 3.3(b)] in the y-z plane can
excite modes of all branches (i.e. of both even and odd symmetries). Figure
3.8 presents the oommf-simulated excitation spectra excited by the same even
excitation field for HT ranging from 0 to 0.3 T. The modes of branches 2 and 4
can only be excited for 0 < HT < HC , with the maximum efficiency occurring at
HT 0.2 T. It is to be noted that Lee et al. [21] claimed that the A + B field
[corresponding to the odd + even field in Fig. 3.3(b)] in Ref. [3] is not sufficiently
general to generate complete magnonic band structure. However, based on Eq.
3.2 and micromagnetic simulations, we have shown that the A + B field can indeed
excite all the modes. Additionally, this is true even for the single-width nanostripe
considered in Ref. [21], because the demagnetizing-field-induced effective pinning
[22] of the dynamic magnetization along the y axis results in a relatively small yet
nonzero excitation efficiency by the A + B field. Although the relative excitation
efficiency generally decreases with increasing node number in the y-direction, our
simulations show that the high-order m = 3 mode can be excited by the A + B
field, which can be identified under logarithmic scale (not shown). Finally, for
the oommf simulations, using just one layer of cells across the thickness provides
sufficient accuracy for the frequency range considered. For higher frequencies,
the above discussion can be trivially extended to include perpendicular standing
spin waves (pssw).
The dynamic magnetic-field tunability of the bandgap has advantages over
that based on structural dimensions or material composition. It is not feasible
to reshape band structures in the latter case by tuning corresponding parame-
49
BIBLIOGRAPHY
ters once the MCs are fabricated. In contrast, we have demonstrated here that
bandgaps can be dynamically tuned by applying a transverse field. More importantly, by changing the direction of magnetization, some bandgaps can be reversibly and dynamically switched on and off, which may be utilized for nanoscale
SW switches.
3.4
Summary
In this Chapter, we have addressed a possible issue in calculating the band

structure of a width-modulated MC. Specifically, the symmetry of the excitation
field in an oommf simulation may affect the excitation efficiency of different
spin-wave modes, which in turn has a bearing on the completeness of the band
structures obtained. Additionally, we have presented a solution to this problem,
which is corroborated by two independent theoretical methods. Our methods are
generalizable to other MCs of different structures.
The tuning of the band structure parameters of the width-modulated MC with
a transversely applied magnetic field was also discussed. Our simulations revealed
some interesting properties, such as the dynamically switching on/off of magnonic
band gaps by changing the strength of the applied magnetic field.
Bibliography
[1] K.-S. Lee, D.-S. Han, and S.-K. Kim, Physical Review Letters 102, 127202
(2009). 35, 37, 38, 39, 44
[2] A. V. Chumak, P. Pirro, A. A. Serga, M. P. Kostylev, R. L. Stamps, H.
50
BIBLIOGRAPHY
Schultheiss, K. Vogt, S. J. Hermsdoerfer, B. Laegel, P. A. Beck, et al., Applied

Physics Letters 95, 262508 (2009). 35
[3] K. Di, H. S. Lim, V. L. Zhang, M. H. Kuok, S. C. Ng, M. G. Cottam, and H.
T. Nguyen, Physical Review Letters 111, 149701 (2013). 36, 44, 47, 49
[4] F. S. Ma, H. S. Lim, Z. K. Wang, S. N. Piramanayagam, S. C. Ng, and M.
H. Kuok, Applied Physics Letters 98, 153107 (2011). 36, 43
[5] K. Dheeraj, D. Oleksandr, P. Sabareesan, and B. Anjan, Journal of Physics
D: Applied Physics 45, 015001 (2012).
[6] F. S. Ma, H. S. Lim, V. L. Zhang, Z. K. Wang, S. N. Piramanayagam, S. C.
Ng, and M. H. Kuok, Journal of Applied Physics 111, 064326 (2012). 36
[7] M. Dresselhaus, G. Dresselhaus, and A. Jorio, Group Theory: Application to
the Physics of Condensed Matter (Springer, 2008). xi, 36, 38, 46
[8] H. T. Nguyen and M. G. Cottam, Journal of Physics D: Applied Physics 44,
315001 (2011). 37
[9] H. T. Nguyen and M. G. Cottam, Journal of Applied Physics 111, 07D122
(2012).
[10] K. Di, H. S. Lim, V. L. Zhang, S. C. Ng, M. H. Kuok, H. T. Nguyen, and M.
G. Cottam, Journal of Applied Physics 115, 053904 (2014). 37
[11] S. K. Kim, Journal of Physics D-Applied Physics 43, 264004 (2010). 39
O. Adeyeye, ACS Nano 4, 643 (2010). 39, 43
[13] S. Mamica, M. Krawczyk, M. L. Sokolovskyy, and J. Romero-Vivas, Physical
Review B 86, 144402 (2012). 39
[14] J. W. Klos, M. L. Sokolovskyy, S. Mamica, and M. Krawczyk, Journal of
Applied Physics 111, 123910 (2012). 39
51
BIBLIOGRAPHY
[15] J. W. Klos, D. Kumar, M. Krawczyk, and A. Barman, Sci. Rep. 3, 2444

(2013). 39, 43
[16] C. S. Lin, H. S. Lim, Z. K. Wang, S. C. Ng, and M. H. Kuok, Ieee Transactions
on Magnetics 47, 2954 (2011). 39
O. Adeyeye, Applied Physics Letters 94, 083112 (2009). 40, 43
[18] C. Bayer, J. P. Park, H. Wang, M. Yan, C. E. Campbell, and P. A. Crowell,
Physical Review B 69, 134401 (2004). 43
[19] Z. K. Wang, M. H. Kuok, S. C. Ng, D. J. Lockwood, M. G. Cottam, K.
Nielsch, R. B. Wehrspohn, and U. Gosele, Physical Review Letters 89,
027201 (2002). 43
[20] M. Bolte, G. Meier, and C. Bayer, Physical Review B 73, 052406 (2006). 48
[21] K.-S. Lee, D.-S. Han, and S.-K. Kim, Physical Review Letters 111, 149702
(2013). 49
[22] K. Y. Guslienko and A. N. Slavin, Physical Review B 72, 014463 (2005). 49
The following MATLAB codes are for Fig. 3.1 (3D image of the width-modulated MC).
HIDE:
A = imread(commentGeometry2.bmp);
B = imread(rom2.bmp);
[x,y,z] = size(B);
mask = uint8( 3*ones(size(B)) );
data1 = bitand( bitshift(B,0),mask);
data2 = bitand( bitshift(B,-2),mask);
A = bitshift(bitshift(A,-2),2) + [data1,data2;data3,data4];
imwrite(A,commentGeometryHide.png,png);
RETRIEVE:
A = imread(commentGeometryHide.png);
[x,y,z] = size(A);
A = bitshift(bitshift(A,6),-6);
X = 2^0*A(1:x/2,1:y/2,:) + 2^2*A(1:x/2,y/2+1:y,:) + ...
2^4*A(x/2+1:x,1:y/2,:) + 2^6*A(x/2+1:x,y/2+1:y,:);
Y = uint8(zeros(y/2,x/2,z));
for n = 1:3
Y(:,:,n) = fliplr(X(:,:,n));
end
imwrite(Y,romRetrived.png,png)
52
4
Artificial Defects in
Magnonic Crystals
Chapter 3 addresses the numerical investigation of a particular MC, which is
free from any defect. In this chapter, we will go one step further and consider
imperfect MCs, crystals with artificial defects.
4.1
Introduction
The controlled introduction of defects in crystals can radically alter their properties in desired ways. This is perhaps best exemplified by the enhanced electrical
conductivity of semiconductors, as well as the color and luminescence of certain
crystals, such as ruby, arising from trace amounts of chemical impurities in them.
Over the past half century, the constantly progressing technology of p- and ntype doping of silicon has nurtured the semiconductor industry that has a market of hundreds of billions of dollars, in which most of the computing chips and
flash memories are made of silicon crystals with controlled defects. While still
53
4. ARTIFICIAL DEFECTS IN MAGNONIC CRYSTALS

under extensive study for conventional crystals, the concept of defect has also
been introduced to artificial crystals, especially photonic crystals.
The properties of artificial crystals, in the form of periodic arrays of elements,
can also be controlled or enhanced by the presence of defects. The defects studied
here are deviations from the artificial crystal periodicity which result in imperfect crystals. For instance, a structural or compositional defect could be a stripe
element, of a different width or material, in an otherwise perfect periodic multicomponent array of stripes. Generally, defects in artificial crystals offer more engineering flexibility than those in atomic crystals. For example, artificial crystals
could be made of a larger collection of array elements (e.g. various ferromagnetic materials for MCs) than the number of elements in real crystals available
as dopant atoms. Moreover, the size of both perfect and defect array elements
in artificial crystals are continuously adjustable. These advantages impose fewer
constraints on the applications and might lead to new physics.
Defect-induced properties of photonic crystals have attracted much attention
owing to their interesting physics and wide potential applications. Indeed, various
photonic devices based on defects already serve as building blocks in integrated
photonic circuits [1]. It is found that microcavities which are isolated defects in
photonic crystals can be employed to control and manipulate the spontaneous
emission of light [2, 3], a functionality which allows their application in lasers and
quantum optics. Other defect-induced phenomena include the all-optical control
of light propagation based on enhanced optical nonlinearity in photonic-crystal
nanocavities demonstrated recently [4, 5]. Also, defects in photonic crystals have
been shown to slow down, trap and guide light in the nanoscale [68].
In the case of MCs, although extensive studies have been undertaken on defect-
54
4.2 Structures of Defect MCs and Experimental Details
free MCs [915], relatively little research, especially experimental ones, has been
done on defect MCs [9, 1622]. For defect-free MCs, forbidden magnonic gaps
have been experimentally observed in 1D and 2D MCs [1215, 23]. Nikitov et
al. [9] calculated the band structures and spin wave (SW) propagation properties
of multilayer-type lattices with defect, while Yang et al. [18] theoretically investigated the band structure and coupling between point defects of a 2D MC. In
their microwave transmission and reflection study of magnetostatic SWs in defect
magnetic structures, Filimonov et al. [16] observed resonance peaks which they
ascribed to defect states. The structures studied were linear 150micron-period
arrays of etched grooves, on an yttrium iron garnet (YIG) film, each of which
contained only either a single defect groove or crest. Chi et al. [17] showed that
a line defect in a 2D MC can give rise to, as well as guide confined magnetostatic
waves. In analogy to defect photonic crystals, it is expected that the incorporation of defects in MCs opens the way to new physics and new functionalities
for application in data communication, processing, and magnetic storage devices.
However, to date, there is no experimental observation of the band structures of
nanostructured defect MCs, a crucial information for the understanding of SW
propagation in these metamaterials. Here, we will present the first experimental
observation of the band structures of 1D bi-component MCs with periodic defects.
55
Figure 4.1: SEM image of the MC with 400nm-wide Py defect stripes.

Darker regions represent Py stripes while lighter regions, the Co stripes. The red
arrow indicates the direction of equilibrium magnetization.
4.2
Structures of Defect MCs and Experimental

Details
All the 30nm-thick MCs, with a surface area of 100m 100m, were fabricated by Prof Yang Hyunsoo of the ECE Department (NUS) using electron-beam
lithography, magnetron sputtering and lift-off processes [24]. The perfect MC
consists of alternating equal-width 250nm-wide Py and Co stripes, with a periodicity a = 500 nm. The defect MCs are otherwise perfect arrays with one Py
stripe, having a width other than 250 nm, periodically located in every 10 periods.
Crystals containing Py defect widths of 300, 400, and 500 nm were fabricated.
Figure 4.1 shows the SEM image of a defect MC, with a defect stripe width of
400 nm and a supercell period ad = 5150nm, which will hereafter be referred to
as the 400nm defect MC.
Prior to the BLS measurements, the samples were saturated along the long axis
of the stripes in a magnetic field 0 H = 0.4 T, which was subsequently reduced
to zero. Using the = 514.5nm radiation of an argon-ion laser and a (3+3)-pass
56
4.2 Structures of Defect MCs and Experimental Details
Figure 4.2: Brillouin spectra of the (a) defect-free MC and (b) 400nm
defect MC recorded at various wavevectors about their respective first
BZ boundaries of q = /a, and q = 10/ad. Measured spectra are represented
by dots. All spectra were fitted with Lorentzian functions (blue dotted curves),
and the resultant fitted spectra are shown as red solid curves. Defect mode peaks
are shown shaded in blue.
tandem FP interferometer, BLS spectra were recorded in the 180 back-scattering

geometry and in ps polarization. With the scattering plane lying perpendicular to
the long axis of the stripes, SWs propagating in the x-direction were probed (Fig.
4.1). Figure 4.2 presents typical Brillouin spectra of the perfect and 400nm defect
MCs, recorded at various wavevectors about their respective first Brillouin zone
(BZ) boundaries of q = /a, and q = 10/ad . Wavevectors q were selected by
setting the laser light incident angle to the corresponding values of sin1 (q/4).
All the observed Brillouin peaks shifted under the applied magnetic field and were
thus ascribed to SW modes, with the middle peak assigned to a defect mode in
the defect MC (see discussion below).
57
Figure 4.3: Magnon band structures of (a) the perfect MC and (b) the
400nm defect MC. The dashed vertical lines indicate their respective BZ boundaries at q = /a, and q = 10/ad , where a = 500 nm and ad = 5150 nm. Measured
Brillouin data are indicated by red squares, while finite-element and micromagnetic
simulated data are shown as black circles and green lines, respectively. The circle size and the green shade intensity are proportional to the calculated Brillouin
intensities.
4.3
Band Structures
The measured magnon dispersion relations of the defect-free MC and the

400nm defect MC are presented in Fig. 4.3. For the latter, besides the two
prominent branches corresponding to those of the defect-free MC [12, 23], an additional dispersionless branch, which is symmetrically located ( 9.5 GHz) about
the former two, appears.
Time-domain micromagnetic simulations were performed by solving the LLG
equation [25] based on the oommf package [26], for which 50m 50m 30nm
MC models and 5nm 50m 30nm(= x y z) computational cell size
were used. In the calculations, the saturation magnetizations and the exchange
58
4.3 Band Structures
stiffness constants of Co and Py were taken to be MS,Co = 1.1 106 A/m, ACo =
2.5 1011 J/m, MS,Py = 7.3 105 A/m, APy = 1.1 1011 J/m, while the
gyromagnetic ratio and the dimensionless Gilbert damping coefficient were set
at = 195 GHz/T and = 0.0001 respectively. As only SW modes in the
x-direction were measured, modes with magnetization variation along the long
axes of the stripes were neglected in the simulations, since only one cell in the
y-direction was considered. A pulsed sinc magnetic field in the z-direction, Hz
sin[2f (t t0 )]/[2f (t t0 )] where f = 20 GHz, was applied within one unit cell
to excite spin waves. The calculated dispersion relations, obtained by a Fourier
transform of the dynamic magnetization with respect to time and x, are presented
in Fig. 4.3 as green lines.
Frequency-domain finite-element simulations based on the above magnetic parameters were also carried out, by solving the 2D linearized Landau-Lifshitz equation implemented in the comsol multiphysics software [27], with the BlochFloquet boundary condition applied. A supercell comprising nine Py/Co unit cells
plus a defect Py/Co cell was used in the calculation of the dispersion relations.
To compare theory with experiments, the Brillouin scattering intensities I of SW
modes mq,f (r) were estimated as Fourier transforms of the out-of-plane compoR
2
nents of their dynamic magnetization mz namely, I unit cell eiqr mq,f (r)d3 r ,
where q and f are the mode wavevector and frequency, respectively. Finiteelement calculated SW dispersion relations are presented in Fig. 4.3 as black
circles, the sizes of which give an indication of the Brillouin intensities. Because
the BZ of the supercell of the defect MC is smaller than that of the perfect MC
unit cell, the theoretical band structure of the former comprises more dispersion
curves. However, only modes of the branches corresponding to the two branches
59

of the perfect MC have significant Brillouin intensities, a feature borne out by
Brillouin measurements. As seen from Fig. 4.3, this observation is also consistent
with the micromagnetic-simulated modes, whose calculated Brillouin intensity is
proportional to the intensity of the shade of green representing them. It should
be noted that unlike the arrays of micron-size YIG elements studied by Filimonov
et al. [16] where only the magnetostatic dipolar interaction is taken into account,
consideration of the exchange interaction for our nano-size structures is essential
for good agreement with BLS experiments.
Reference to Fig. 4.3 reveals that except for the defect mode branch, the main
features of the measured and simulated band structures of the defect MC are
very similar to those of the defect-free MC. This is expected as the influence of
low-concentration defects on the collective magnetic excitations is weak. In the
limiting case of zero defects, the defect MC should have a dispersion relation
identical to that of a perfect one. It is interesting to know how the presence
of low-concentration periodic defects affects the extended SW branches. Our
calculations reveal that the modified periodicity of the defect MC induces additional bandgaps at the BZ boundaries corresponding to a primitive supercell, as
shown in Fig. 4.3. These bandgaps, having narrow widths of about 0.2 GHz,
were not resolved by BLS measurements because the linewidth ( 1 GHz) of the
inelastically scattered signal is much broader than these bandgaps.
4.4
Size Dependence
The dependence of defect modes on defect Py stripe widths was also experimentally and theoretically investigated. BLS spectra of MCs with various defect
60
4.4 Size Dependence
Figure 4.4: Dependence of defect mode frequency on defect Py stripe

width. The red squares and curve represent measured and calculated frequencies of
the defect modes at the BZ boundary q = 10/ad , while the blue ones, the measured
and calculated band edge frequencies, with the shaded regions representing the
allowed magnonic bands.
stripe widths were recorded at BZ boundary wavevectors q = 10/ad , where ad

are their respective periods. Measurements reveal that the frequency of the defect mode decreases with increasing defect stripe width, as shown in Fig. 4.4.
In contrast, the frequencies of the bandgap edges are not very sensitive to the
defect stripe width. As SWs with frequencies within the bandgap cannot propagate in the non-defect regions of an MC, the defect mode will be confined by the
perfect stripes sandwiching the defect ones [9]. Because of the confinement, the
narrower the defect width, the shorter would be the mode wavelength and hence,
the higher its frequency. This effect is consistent with the calculations presented
in Fig. 4.4. As the width of the defect stripe approaches the 250 nm width of
a defect-free one, a transition from a localized to an extended mode occurs (see
discussion below).
61
Figure 4.5: Mode profiles of confined and propagating modes. Coloured

curves represent distributions of the x-component mx of the instantaneous dynamic
magnetizations along respective supercells, at q = 10/a, of (a) defect mode of the
400nm defect MC, (b) defect mode of the 300nm defect MC, and (c) the lower
(blue line) and higher (red line) extended branches of the 400nm defect MC at the
bandgap edges. Py and Co stripes are represented by respective grey and white
bands.
The simulated defect mode profile, across one supercell, for the 400nm defect
MC is displayed in Fig. 4.5(a). It is clear that the amplitude of the defect mode
is maximal within the defect Py stripe and decays rapidly outside it, with the
mode penetrating into the non-defect portions by only about four stripes. This
localization behavior of the defect mode in the vicinity of the defect is consistent
with earlier calculations [9, 28, 29]. Additionally, the defect mode profile of
the 300nm defect MC, shown in Fig. 4.5(b), indicates that the localization of a
defect mode is more pronounced, the closer its frequency is to the bandgap center
frequency ( 9.4 GHz), and it becomes more extended the closer its frequency is
62
4.4 Size Dependence
to those of the extended branches (see also Fig. 4.4).

We can understand the above behavior by expanding the upper extended dispersion curve f (q) to lowest order near the bandgap [30]: q f (q) f (/a)
c(q /a)2 , where c > 0 is a constant. For a defect mode with a frequency f (q)
p
smaller than f (/a), f < 0, and thus q /a i f /c. The imaginary
part of q causes the mode to decay exponentially in the non-defect portions of the
crystal. A defect mode frequency that is closer to the gap center frequency will
have a larger imaginary q component, resulting in a faster decay of its dynamic
magnetization and hence, a more pronounced mode localization. This behaviour
can be exploited to control the degree of coupling between a defect mode and
other modes in the design of functional magnonic devices based on defects.
It is noteworthy that the mode profiles of the extended modes themselves
are also affected by the presence of defects. For the 400nm defect MC, the mode
profiles of the lower and higher extended branches at the point and the bandgap
edges are shown in Fig. 4.5(c). In contrast to the defect mode, the propagating
modes possess smaller amplitudes in the vicinity of the defect stripe.
Both the micromagnetic and finite-element simulations indicate that the defect
mode within the bandgap can be observed by BLS only within a narrow range of
wavevectors ( 5.5 to 8.5 m1 ) about the first BZ boundary. This is because
the distribution of the dynamic magnetization mz of the defect mode is nearly
independent of wavevector. Therefore, its BLS intensity is higher for wavevectors
in the vicinity of the first BZ boundary, where the Fourier transform of mz has
larger components.
63
4.5
Summary & Discussion
The defect-induced phenomena in 1D bicomponent MCs with structural defects have been investigated by BLS and numerical simulations. The presence
of the Py defect stripes is principally manifested as a dispersionless branch lying within the bandgap of the experimental magnonic band structure. The two
dispersive branches observed correspond to those of the defect-free MCs. Calculations also reveal that the magnon spectra of the defect crystals comprise
additional branches and narrow-width Bragg bandgaps arising from the periodicity of the supercell. The defect mode frequency was observed to be tunable by
varying the defect stripe width, a feature that would enhance the functionality of
MC-based devices, e.g. tailoring the single-frequency passband of SW filters. Just
as the incorporation of defects offers novel physics and functionalities for photonic
crystals, the same is expected for defects in their magnetic analogues, MCs. Our
work represents the first observation of the band structures of nanostructured
MCs with periodic defects and would be of use for a better understanding of
their underlying physics and their applications in MC-based devices.
Our work here has considered only one approach of creating artificial defects in
MCs. There are many other ways for achieving this. For instance, the thickness of
certain stripes can be changed to create defects. Also, instead of studying defects
with geometrical variations, it would of interest to study MCs with defects of a
(magnetic or nonmagnetic) material different from those of the pure MC. Comparing the properties of these variations with those of our present observations
would also be of interest. Moreover, as only a basic investigation of defects has
been conducted here, there is much more work to be done in the direction of de-
64
BIBLIOGRAPHY
signing functional devices based on these defects. These devices could be similar
to their photonic counterparts, since both deals with waves. However, it would
be more interesting to exploit the unique properties of magnons. One approach is
to dynamically control propagating SWs by an external applied field or another
spin-wave pulse [31]. The former is rather clear as spins are strongly coupled to
external fields via the Zeeman effect, while the second is due to the fact that the
LLG equation is intrinsically nonlinear. Note that achieving similar behaviors
of light has proven to be very demanding [32], because Maxwells equations are
linear in nature and observable nonlinear effect usually necessitates very strong
light intensities.
Bibliography
[1] K. J. Vahala, Nature 424, 839 (2003). 54
[2] D. Englund, D. Fattal, E. Waks, G. Solomon, B. Zhang, T. Nakaoka, Y.
Arakawa, Y. Yamamoto, and J. Vukovi, Physical Review Letters 95,
013904 (2005). 54
[3] S. Ogawa, M. Imada, S. Yoshimoto, M. Okano, and S. Noda, Science 305,
227 (2004). 54
[4] M. SoljaCiC and J. D. Joannopoulos, Nature Materials 3, 211 (2004). 54
[5] K. Nozaki, T. Tanabe, A. Shinya, S. Matsuo, T. Sato, H. Taniyama, and M.
Notomi, Nature Photonics 4, 477 (2010). 54
[6] H. Altug and J. Vuckovic, Applied Physics Letters 86, 111102 (2005). 54
[7] H. Gersen, T. J. Karle, R. J. P. Engelen, W. Bogaerts, J. P. Korterik, N. F.
van Hulst, T. F. Krauss, and L. Kuipers, Physical Review Letters 94, 073903
65
BIBLIOGRAPHY
(2005).
[8] J. Scheuer, G. T. Paloczi, J. K. S. Poon, and A. Yariv, Opt. Photon. News
16, 36 (2005). 54
[9] S. A. Nikitov, P. Tailhades, and C. S. Tsai, Journal of Magnetism and Magnetic Materials 236, 320 (2001). 55, 61, 62
[10] A. N. Kuchko, M. L. Sokolovskii, and V. V. Kruglyak, Physica B: Condensed
Matter 370, 73 (2005).
[11] A. V. Chumak, A. A. Serga, S. Wolff, B. Hillebrands, and M. P. Kostylev,
Applied Physics Letters 94, 172511 (2009).
O. Adeyeye, Applied Physics Letters 94, 083112 (2009). 55, 58
[13] V. L. Zhang, H. S. Lim, C. S. Lin, Z. K. Wang, S. C. Ng, M. H. Kuok, S.
Jain, A. O. Adeyeye, and M. G. Cottam, Applied Physics Letters 99, 143118
(2011).
[14] S. Tacchi, G. Duerr, J. W. Klos, M. Madami, S. Neusser, G. Gubbiotti,
G. Carlotti, M. Krawczyk, and D. Grundler, Physical Review Letters 109,
137202 (2012).
[15] R. Zivieri, S. Tacchi, F. Montoncello, L. Giovannini, F. Nizzoli, M. Madami,
G. Gubbiotti, G. Carlotti, S. Neusser, G. Duerr, et al., Physical Review B
85, 012403 (2012). 55
[16] Y. Filimonov, E. Pavlov, S. Vystostkii, and S. Nikitov, Applied Physics Letters 101, 242408 (2012). 55, 60
[17] K. H. Chi, Y. Zhu, and C. S. Tsai, Journal of Applied Physics 115, 17D125
(2014). 55
[18] H. Yang, G. Yun, and Y. Cao, Journal of Applied Physics 112, 103911 (2012).
66
BIBLIOGRAPHY
55
[19] A. Barman, Journal of Physics D: Applied Physics 43, 195002 (2010).
[20] J. W. Klos and V. S. Tkachenko, Journal of Applied Physics 113, 133907
(2013).
[21] M. Madami, G. Gubbiotti, S. Tacchi, G. Carlotti, and S. Jain, Physica B:
Condensed Matter 435, 152 (2014).
[22] M. Krawczyk and D. Grundler, Journal of Physics: Condensed Matter 26,
123202 (2014). 55
O. Adeyeye, ACS Nano 4, 643 (2010). 55, 58
[24] V. L. Zhang, C. G. Hou, H. H. Pan, F. S. Ma, M. H. Kuok, H. S. Lim, S. C.
Ng, M. G. Cottam, M. Jamali, and H. Yang, Applied Physics Letters 101,
053102 (2012). 56
[25] T. L. Gilbert, IEEE Transactions on Magnetics 40, 3443 (2004). 58
[27] COMSOL Multiphysics, COMSOL AB, Stockholm, Sweden. (Version 4.0).
59
[28] V. V. Kruglyak, M. L. Sokolovskii, V. S. Tkachenko, and A. N. Kuchko,
Journal of Applied Physics 99, 08C906 (2006). 62
[29] H. Yang, G. Yun, and Y. Cao, Journal of Applied Physics 111, 013908 (2012).
62
[30] J. D. Joannopoulos, S. G. Johnson, J. N. Winn, and R. D. Meade, Photonic
Crystals: Molding the Flow of Light (Second Edition) (Princeton University
Press, Princeton, 2008), 2nd ed. 63
67
BIBLIOGRAPHY
[31] A. V. Chumak, A. A. Serga, and B. Hillebrands, Nature Communications 5,

4700 (2014). 65
[32] T. Baba, Nature Photonics 2, 465 (2008). 65
68
5
SWs in Ferromagnetic Films
with Interfacial
Dzyaloshinskii-Moriya
Interactions
Chapters 3 and 4 mainly focus on the spin-wave properties of various MCs.
In this Chapter, our attention will be on the effects of the Dzyaloshinskii-Moriya
interaction (DMI) on SWs in multilayer magnetic films.
5.1
Introduction
DMI, an anti-symmetric exchange interaction, was initially proposed to explain

the weak ferromagnetism of antiferromagnets both phenomenologically [1] and
microscopically [2]. Recently, DMI has attracted renewed interest due to its
fundamental role in various novel phenomena such as the magnon Hall effect [3],
69
5. SWS IN FERROMAGNETIC FILMS WITH INTERFACIAL

DZYALOSHINSKII-MORIYA INTERACTIONS
molecular magnetism [4] and multiferroicity [5]. Also, DMI is now known to be
responsible for chiral spin textures such as magnetic skyrmions [68], which show
promise in spintronics and high-speed, ultra-high density storage technology due
to their unique properties like propagation under ultralow current densities [9, 10]
and rewritability by spin-polarized currents [11].
DMI can arise from inversion symmetry breaking at surfaces or interfaces
between a ferromagnetic layer and a nonmagnetic one having a strong spin-orbit
coupling [1215] due to 3-site indirect exchange mechanism [16, 17]. Such an
interfacial DMI may exist regardless of the crystal symmetry of the component
materials [18], and is expected to be much stronger than the bulk DMI [12, 13].
In this Chapter, we will focus on the interfacial DMI at two interfaces: Pt/Co
and Pt/CoFeB. While the momentum-resolved spin dynamics measurement techniques of spin-polarized electron energy loss spectroscopy [15] and inelastic neutron scattering [19] can offer direct evidence of the presence of DMI, they are not
suitable for multilayers with buried metal/ferromagnet interfaces [18]. Magnons
probed by these methods generally have energies, lifetimes and attenuation lengths
of the order of 10 meV, 10 femtoseconds, and 1 nm [20], respectively. However,
for magnonics and spintronics applications, SWs in the GHz range (0.01 to 0.1
meV), which can be readily excited by microwaves or spin transfer torque [21, 22],
are of greater relevance due to their longer lifetime in the nanosecond range and
micron-scale coherence length. Within this energy scale, Brillouin light scattering (BLS) with its sub-GHz resolution and high surface sensitivity ( few nm)
for metals is ideally suited for studying both the interfacial DMI and spin-wave
dynamics of Pt/ferromagnet multilayers. DMI can induce strong nonreciprocal
propagation of SWs (see Section 5.2), resulting in asymmetric spin-wave disper-
70
5.2 Dispersion Relation in the Presence of Interfacial DMI
sion relations, which can be detected by BLS. Thus the existence and strength of
DMI in multilayer film stacks can be conveniently investigated by the technique
of BLS.
5.2
Dispersion Relation in the Presence of Interfacial DMI
In this section, the dispersion relation of SWs in uniform ferromagnetic films

with interfacial DMI will be derived. Although interfacial DMI exists only near
the interfaces of a ferromagnet, it is a good approximation to treat this interaction
as an effective bulk DMI for ultra-thin films. For the lowest magnon mode, the
variation of magnetizations across the film thickness will also be neglected.
Phenomenologically, the DMI energy density of our sample is given by [13, 23,
24]
EDM
D
= 2
MS

My
Mz
My
Mx
My
Mx
+ My
Mz
,
x
x
z
z
(5.1)
where D is the DMI constant, and Mi the i-component of the magnetization. The
magnetization of linear SWs can be expressed as M (x, t) = MS n(x, t). For the
coordinate system and magnetic parameters shown in Fig. 5.2, the unit vector
along the magnetization direction is
n(x, t) = nx ex + ny ey ez = nx0 ei(tkx) ex + ny0 ei(tkx) ey ez ,
(5.2)

where |nx0 |, ny0 1, and ex , ey , ez are unit vectors codirectional with the
coordinate axes. For SWs traveling along the x-axis, the effective field due to the
71

DMI is given by [24, 25]
HDM
1 EDM
2D
=
=
0 M
0 MS

ny
nx
ex
ey .
x
x
(5.3)
The total field is

Heff = H0 ez + J2 n + Hdip + Hani + HDM ,
(5.4)
where J = 2A/0 MS , A is the exchange stiffness constant, the dipolar field in

the film [26] is Hdip = MS (kd)nx ex MS 1 (kd) ny ey , d (= 3.2 nm) is the
magnetic film thickness, (x) = 1 (1 e|x| )/ |x|, the perpendicular uniaxial anisotropic field Hani = 2K/(0 MS )ny ey , and K is the anisotropy constant.
Substituting Eqs. 5.2 and 5.4 into the LLG equation
dn
dn
= n Heff + n
dt
dt
(5.5)
yields the following spin-wave dispersion relation

= 0 + DM
q

2
= H0 + Jk 2 + (kd)MS H0 HU + Jk 2 + MS (kd)MS
Dk
0 MS
(5.6)
where the Gilbert damping constant has been set to zero, is the gyromagnetic
ratio, and HU = 2K/ (0 MS ). The angular frequency in the absence of DMI is
denoted by 0 , while the DMI-induced frequency shift by DM .
It is apparent that the DMI-induced frequency difference of counter-propagating
72
5.3 Experimental Results
SWs, given by
f (k) =
(k) (k)
2
=
Dk,
2
0 MS
(5.7)
is linear in the DMI constant and the spin-wave wave vector k, but independent
of the applied in-plane magnetic field.
5.3
5.3.1
Experimental Results
Pt/Co/Ni Film
In Section 5.1, we mentioned that interfacial DMI may arise at the interfaces
between a ferromagnet and a metal with strong spin-orbit coupling. Among these
multilayer structures, Pt/Co is of enormous scientific and technological interest
because of its strong perpendicular magnetic anisotropy (PMA), which results in
high thermal stability and reduced threshold current for spin-transfer switching
[27] and current-driven domain wall (DW) motion [28, 29].
The interfacial DMI at the Pt/Co interface can be tuned, by interface engineering, to change the type (Nel or Bloch), chirality, and velocity of the DWs
[3032]. Because of the significance of Pt/ferromagnet contacts in spintronics [33],
it is critical to study the spin dynamics of such interfaces in the presence of DMI,
especially when spin waves (SWs) serve as spin-current carriers. However, no
direct experimental evidence of the interfacial DMI in Pt/ferromagnet films has
been reported. In this Section, we will present a BLS study the interfacial DMI
in an as-grown Pt/Co/Ni multilayer film possessing an in-plane magnetization.
73

H
H
SiO2 (3 nm)
MgO (2 nm)
Ni (1.6 nm)
Co (1.6 nm)
Pt (4 nm)
MgO (2 nm)
Si / SiO2
Figure 5.1: In-plane and perpendicular magnetic hysteresis loops of the

Pt(4)/Co(1.6)/Ni(1.6) film measured by VSM. Inset: Schematic of the multilayer film and directions of the applied magnetic fields.
5.3.1.1
Sample Fabrication & Basic Magnetic Properties
The sample studied was deposited on a thermally oxidized silicon wafer by

both direct-current and radio-frequency magnetron sputtering at room temperature. Specifically, the unannealed film stack is substrate/MgO(2)/Pt(4)/Co(1.6)
/Ni(1.6)/MgO(2)/SiO2 (3) (hereafter referred to as sample Pt/Co/Ni), where the
figures in parentheses are the nominal thicknesses in nm. Argon gas ( 2.3 mTorr)
was used during the sputtering process with a background pressure of 2 109
Torr, and the deposition rates for MgO, Pt, Co, Ni and SiO2 were 0.026, 0.54, 0.14,
0.21 and 0.10 /s, respectively [34]. The in-plane and out-of-plane magnetic hysteresis loops of the sample, measured by vibrating sample magnetometry (VSM),
as shown in Fig. 5.1, reveal that the in-plane saturation field and the saturation
magnetization MS are approximately 50 mT/0 and 1160 kA/m, respectively.
74
H0
k
x
Figure 5.2: Brillouin spectra recorded at a fixed incident angle of =

45 (|k| = 17.3 m1 ) under oppositely-oriented external magnetic fields
H0 = 97 mT/0 . Inset: Schematic of the Cartesian coordinate system and 180
back-scattering geometry. The incident and scattered light beams lie in the x-y
plane and are at an angle to the y-axis. The magnon wave vector is denoted by
k.
5.3.1.2
Details of BLS Experiment
All BLS measurements were performed in the 180back-scattering geometry

(see inset of Fig. 5.2) and in ps polarization using the 514.5nm radiation of
an argon-ion laser and a six-pass tandem FP interferometer. The in-plane DC
magnetic field H0 was applied perpendicular to the incident plane of light, corresponding to the Damon-Eschbach (DE) geometry. As the metallic film is very
thin, counter-propagating surface waves localized at the top and bottom interfaces
were simultaneously observed in the BLS experiments. At room temperature,
thermal magnons with all possible values of wavevectors could contributed to
Brillouin scattering for both Stokes (magnon creation) and anti-stokes (magnon
annihilation) processes. However, for the light scattering geometry shown in Fig.
75

5.2, in-plane momentum change of photons must be in the direction of positive x.
As total momentum is conserved along the film surface, the scattering geometry
selects processes involving either the annilation of magnons with +x momentum
or the creation of magnons with x momentum. The other two cases, i.e. the
annilation of magnons with x momentum and the creation of magnons with
+x momentum, contribute a negative change in the momentum of light along the
x-axis and therefore are not measured by the experimental set-up. The above discussion concludes that there is a one-to-one correspondence between the signs of
the photon energy shift and the magnon wavevector: the Stokes and anti-Stokes
peaks arise from SWs propagating in the x and +x directions, respectively. The
respective frequencies of counter-propagating SWs, having the same momentum,
are simultaneously presented in the same spectrum.
Figure 5.2 shows that the magnon peaks in the Stokes and anti-Stokes portions
of a typical spectrum are asymmetric in terms of both frequency and intensity.
It is noteworthy that on reversing the direction of the applied magnetic field H0 ,
the respective center frequencies, linewidths and intensities of the Stokes and
anti-Stokes peaks were also interchanged. This is a consequence of the reversed
spin-wave propagation direction, as reversing the magnetization is equivalent to
a time-reversal operation [20].
5.3.1.3
Dispersion Relation
To obtain the spin-wave dispersion relation, the BLS measurements were performed with the magnetization saturated along the z direction under magnetic
fields H0 > 50 mT/0 . The momentum transfer in the film plane was varied by
changing the incident angle of the laser beam, thus allowing the recording of BLS
76
spectra at different wavevectors [35]. The resulting spin-wave dispersion relations

measured under various applied magnetic fields are presented in Fig. 5.3(a). The
most prominent feature of the dispersion curves is their asymmetry with respect
to the wave vector k, which is attributed to the lifted chiral degeneracy arising
from the interfacial DMI [14]. For the same wavelength, the frequency and group
velocity of SWs propagating in the x direction are larger than those of SWs
propagating in the +x direction.
To simplify the analysis, the Pt/Co/Ni film is assumed to be an effective
homogeneous medium [36, 37], which is reasonable since the Co and Ni layers are
coupled by strong exchange interactions and only long-wavelength acoustic SWs
are of interest. Fitting the experimental data to Eq. 5.6, based on the VSMmeasured value of MS , yielded the following values of the magnetic parameters:
/0 = 194 2 GHz/T (corresponding to a g-factor of 2.2), HU = 670 20
kA/m, D = 0.44 0.02 mJ/m2 , and A = 17 1 pJ/m. Assuming a second-order
perpendicular uniaxial anisotropy, the out-of-plane saturation field was estimated
to be Hsat, = MS HU = 0.61 T/0 , which is of the order of the VSM measured
value of about 0.7 T/0 . Consideration of the anisotropic field in the calculations
is necessary for ultrathin Pt/Co structures, which usually possess perpendicular
magnetic anisotropy.
Figure 5.3(b) presents the frequency difference of counter-propagating SWs as
a function of wavevector. It is apparent that f is linear in k, consistent with
Eq. 5.7. It is noteworthy that the frequency difference is more pronounced the
shorter the wavelength, which is the case for exchange SWs [15]. Such a feature
would be desirable in the nano-miniaturization of integrated magnonic circuits
based on Pt/Co structures.
77

(a)
Pt/Co/Ni
Frequency f (GHz)
14
200 mT
12
150 mT
10
97 mT
8
54 mT
6
-20
0
Wave vector k (m-1 )
20
(b)
Df (GHz)
Pt/Co/Ni
Co/Ni
0
Co/MgO/Ni
200 mT
150 mT
97 mT
-1
-20
0
54 mT
54 mT
54 mT
20
Figure 5.3: Brillouin results of the Pt/Co/Ni and control samples. (a)
Spin-wave dispersion relations of the Pt/Co/Ni film at various applied magnetic
fields. (b) Frequency difference of counter-propagating SWs as a function of wave
vector. Symbols denote measured data. The symmetrical dotted lines in (a) represent the theoretical dispersion curves in the absence of DMI, while solid lines in
(a) and (b) represent respective fits to Eqs. 5.6 and 5.7.
We now discuss why the observed asymmetry is not due to the nonreciprocity
of surface SWs. For the scattering geometry shown in Fig. 5.2, the dynamic magnetizations of +k SWs tend to localize at the top surface, while those of k SWs at
the bottom surface [38], which is consistent with the higher BLS intensities of the
former. Consequently, asymmetric distributions of magnetic parameters across
78
the samples thickness could lead to a frequency shift for counter-propagating

SWs. First, the PMA is asymmetric as it mainly originates from the bottom
Pt/Co interface. Due to the nonreciprocal localization of SWs, the +k SW generally experiences an average anisotropy field weaker than that felt by the k
SW. However, according to Eq. 5.6, the frequency of the former should be higher
than that of the latter, which contradicts our observation. Thus, asymmetric
PMA cannot explain the observed asymmetry. Second, the saturation magnetization MS is asymmetric due to the Co/Ni bilayer structure. To study its
effect, two more control samples were fabricated under the same conditions as for
the Pt/Co/Ni sample. They are substrate/MgO(2)/Co(1.6)/Ni(1.6)/MgO(2)/
SiO2 (3) and substrate/MgO(2)/Co(1.6)/MgO(0.4)/Ni(1.6)/MgO(2)/SiO2 (3), referred to as samples Co/Ni and Co/MgO/Ni respectively. The measured of the
Co/MgO/Ni sample [see Fig. 5.3(b)] is very small, indicating that the contribution of the asymmetric distribution of MS is insignificant. This also justifies the
use of the above effective medium theory. We have also to consider the asymmetry of MS arising from the proximity-induced magnetic moment in the Pt layer
of the Pt/Co/Ni sample. However, because the magnetization of Pt is highly
localized within the first few atomic layers adjacent to the Pt/Co interface and
is much weaker than those of Co and Ni [39], the resulting frequency asymmetry should be much weaker than that arising from the asymmetric MS due to
the Co/Ni structure. Therefore, such a proximity effect cannot account for the
observed asymmetry of the Pt/Co/Ni sample. Interestingly, the measured of the
Co/Ni sample [see Fig. 5.3(b)] is fairly large, suggesting the existence of asymmetric surface anisotropy and/or DMI at the Co/Ni interface, similar to the case
of Fe/Ni [40]. However, since the of the Co/Ni sample has a sign opposite to
79

that of the Pt/Co/Ni sample, the factors contributing to dispersion asymmetry
in the former sample are not the same as those for the latter. Hence, we conclude
that the observed strong asymmetric dispersion curve of the Pt/Co/Ni sample is
principally due to the existence of DMI at the Pt/Co interface.
5.3.1.4
Spin-wave Linewidth
The effect of the interfacial DMI on the magnon lifetime was also investigated
by analyzing the spin-wave linewidth, namely the full width at half maximum
(fwhm) of the Brillouin peaks. For the linewidth study, additional spectra were
recorded at various wave vectors in the DE geometry under H0 = 54 mT/0 .
Three factors contribute to the measured linewidth: the finite lifetime of magnons
due to damping, the instrumental broadening due to the limited resolving power
of the optics, and the finite size of the collection aperture [41]. To obtain the true
spin-wave linewidth (lineshape assumed to be Lorentzian), both instrumental
broadening and aperture effects were considered in the spectral fitting, which
was carried out following the procedure of Ref. [41]. The instrumental function
of our optical system used was a Gaussian function with a fwhm of 0.17 GHz,
which was obtained by measuring the elastic Rayleigh peak and assuming the
laser line, with a linewidth of a few MHz, to be a delta function. The aperture
effect was reduced by using an f/7 lens, which served as the focusing and collection
lens. Each spectrum was recorded for a Brillouin peak intensity of at least one
thousand counts, which typically entailed a scanning duration of 10 hours for a
laser-light power of 90 mW incident on the sample.
The deconvoluted linewidths of the counter-propagating SWs, displayed in Fig.
5.4, indicate that their lifetimes are asymmetric with respect to their propagation
80
Linewidth G/2p (GHz)
(a)
1.4
-x propagation
+x propagation
1.3
1.2
1.1
-20
Linewidth G/2p (GHz)
(b)
1.4
-10
0
10
20
-x propagation
+x propagation
1.3
1.2
1.1
6
7
8
Frequency f (GHz)
Figure 5.4: Linewidth (fwhm) of spin waves in the Pt/Co/Ni film as a

function of its (a) wave vector and (b) frequency for H0 = 54 mT/0
applied along the z direction. Experimental linewidths are denoted by triangles, with the error bars representing one standard deviation of the spectral fitting.
The solid and dotted curves represent the calculated linewidths in the presence and
absence of DMI, respectively.
direction. This is attributed to the DMI at the Pt/Co interface. From Fig. 5.4(a),
one observes that for the same wavelength, the linewidth of SWs propagating in
the x direction is wider than that of SWs in the +x direction. To eliminate the
contribution of the frequency difference to the linewidth asymmetry, the same
set of measured linewidth data are re-presented in Fig. 5.4(b) by plotting them
against frequency. We see that for a given frequency, the +k SW generally has a
narrower linewidth and hence longer lifetime than those of the k SW. Also, the
81

linewidth difference between counter-propagating SWs increases with increasing
frequency. For H0 = 54 mT/0 , when the wave vector approaches zero and the
frequency reaches a minimum of about 6 GHz [see Fig. 5.4(a)], the linewidth
difference vanishes. Of note is that, in the scattering configuration of Fig. 5.2,
when the direction of the applied magnetic field was reversed, the linewidths of
the k and +k magnons were interchanged.
To quantitatively explain the observed behavior of the magnon linewidth, the
linearized LLG equation (see Eq. 5.5) was solved with the Gilbert damping term

present. Neglecting high-order terms O 2 , the imaginary part of the angular
frequency is

DM (k)
Im[] = H0 + Jk HU /2 + MS /2 1 +
,
0 (k)
2
(5.8)
with the real part Re[] given by Eq. 5.6. The Brillouin intensity is [42]

2
I( 0 ) m( 0 , k)
1
,
2
Re[] + Im[]2
(5.9)
where m( 0 , k) is the Fourier transform of m(x, t), and thus, the linewidth of
the spin-wave peak is = 2Im[]. Eq. 5.8 reveals that, in the presence of DMI,
the magnon linewidth is modified by the term DM /0 , which is antisymmetric
in the wave vector and thus accounts for the observed asymmetric linewidths.
The calculated linewidths, as well as those in the absence of DMI, are plotted in
Fig. 5.4. Good agreement with Brillouin measurements was obtained by setting
= 0.055, which is consistent with literature values for Pt/Co thin films [43].
Interestingly, due to the presence of the interfacial DMI, the calculated linewidth
82
has an unusual non-monotonic dependence on frequency for SWs propagating in

the +x direction [see Fig. 5.4(b)].
5.3.2
Pt/CoFeB Film
5.3.2.1
Introduction
In Section 5.3.1, interfacial DMI at the Pt/Co interface has been experimentally verified through measuring the spin-wave dispersion relation in the Pt/Co/Ni
sample by BLS. In the BLS methodology, interfacial DMI is manifested as asymmetric dispersion relations and the nonreciprocal propagation of DE SWs in multilayer film structures possessing broken inversion symmetry. While the propagation of DE SWs exhibits nonreciprocal surface localizations, it does not result
in asymmetric dispersion relations for symmetrical magnetic films due to 2-fold
rotational symmetry [38]. However, asymmetric multilayers could be subject to
asymmetric surface pinning which would also result in the nonreciprocal dispersion relations of surface SWs. Indeed, such a pinning at opposite surfaces of a
ferromagnetic film have been found to cause a significant difference in the frequencies of counter-propagating SWs [44]. An evaluation of the contribution from this
effect to the asymmetric magnon dispersion is complicated, as it entails detailed
knowledge of the pinning conditions of surface spins on various interfaces. However, this contribution can be made insignificant by making the magnetic film as
thin as possible.
Due to the above complicacies, the Pt/Co/Ni structure is not the ideal structure to demonstrate the existence of interfacial DMI. First, the thickness of the
magnetic layers is 3.2 nm, which is still not thin enough to neglect the nonrecip-
83

rocal localization of DE SWs. Second, the magnetic layer (Co/Ni bilayer) breaks
inversion symmetry, further complicating the issue. To avoid these problems,
we studied a different system with an optimized structure in this Section. The
sample studied is a Pt/CoFeB bilayer, where the CoFeB film is only 0.8nm thick.
As the magnetic film is only about three atomic layers thick, the contribution
from the above-mentioned asymmetric surface pinning to the observed asymmetric spin-wave frequency is negligible compared with that from interfacial DMI.
Hence, a relatively accurate value of the DMI constant was obtained from the
measured asymmetric magnon dispersion.
For spintronics devices and applications, CoFeB is one of the most studied and
widely used materials [34, 45, 46]. The highest tunneling magnetoresistance value
at room temperature (RT) has been reported for a CoFeB magnetic tunneling
junction [47]. Realization of a large DMI constant in CoFeB is highly anticipated
due to its technological implication for chiral domain wall motion [4850] and
formation of helical spin spirals and magnetic SkXs [17, 51].
5.3.2.2
Sample Fabrication and Characterization
The Pt/CoFeB sample, which has a structure of MgO(2)/Pt(2)/Co4 0Fe4 0B2 0(0.8)
/MgO(2)/SiO2 (3) [numbers in parentheses are the nominal thicknesses in nm, see
also inset of Fig. 5.5(a)], was deposited on thermally-oxidized Si substrate by high
vacuum magnetron sputtering (base vacuum < 2 109 Torr) at room temperature [34]. Post-annealing of the film was performed at 240C for 1 hour in high
vacuum. The magnetic hysteresis loops of the sample (see Fig. 5.5), obtained
separately from vibrating sample magnetometry (VSM) and magneto-optic Kerr
effect (moke) measurements, indicate that the sample possesses partial perpen-
84
H
H
SiO2 (3)
MgO (2)
CoFeB (0.8)
Pt (2)
MgO (2)
Si / SiO2
-20
Magnetization (10 6 A/m)
(b)
H0 out-of-plane
MOKE (a. u.)
(a)
1.5 H0 in- plane
0.0
-1.5
0
20
H0 (mT/0)
-1
0
H0 (T/0)
Figure 5.5: (a) Out-of-plane and (b) in-plane magnetic hysteresis loops
of the Pt/CoFeB film measured by moke and VSM, respectively. Inset:
Schematics of the film structure and orientations of the applied magnetic field H0 .
dicular magnetic anisotropy (PMA) with an in-plane saturation field Hsat 0.5
T/0 . The saturation magnetization was found to be MS = 1.6 106 A/m
from the VSM measurement, which is within a similar range of reported values
[46, 50, 52]. Assuming a second-order uniaxial anisotropy, the effective anisotropy
field is estimated to be HU =
2KU
0 MS
= Hsat + MS = 2.5 T/0 , corresponding to
total surface anisotropy energy of about 1.6 mJ/m2 .
5.3.2.3
Frequency Asymmetry caused by Asymmetric Surface Pinning
As the top and bottom interfaces of the CoFeB film are different, the surface
pinning may be asymmetric, resulting in the frequencies of counter-propagating
SWs on opposite interfaces being different. By considering the case of highest
asymmetric surface pinning, in which the surface anisotropy is assumed to be
localized at one interface while the other is unpinned, we can estimate the upper
limit of its contribution to the frequency asymmetry. Micromagnetic simulations
85

using the oommf code were performed [53, 54], in which a lx ly lz = 4m
4m 0.8nm film was used. The film was discretized into 5nm 4m 0.27nm
mesh cells with three layers across the thickness to simulate the atomic lattice
discretization. The magnetic parameters are from our experiment measurements
and fitting (see below). To calculate the highest frequency shift, we assume
only the bottommost atomic layer possesses a uniaxial PMA with an anisotropy
constant KU three times larger than the effective volume value from our measurement. Results reveal that the frequency difference f due to asymmetric
surface pinning is less than 0.02 GHz, even for the largest wavevector k ( 24
m1 ) that can be attained in our BLS experiment. Importantly, this value is
much smaller than the observed value of f 2 GHz (see below), and hence can
be neglected. This result is expected, as the film is only some three-atom-layer
thick, and hence, nonreciprocal localization of surface SWs is insignificant.
5.3.2.4
BLS Measurement
Brillouin measurement was performed with a set-up similar to that explained

in Section 5.3.1.2. With 30mW of laser power incident on the sample surface and
an irradiated spot size of about 50 m, heating effect of the sample due to light
absorption was negligible. In the experiments, the equilibrium magnetization was
oriented in-plane by applying a magnetic field H0 = 0.7 T/0 in the DE geometry
to align the magnetization in-plane [see Fig. 5.6(a)]. Due to in-plane momentum
conservation of the light scattering processing, SWs travelling in the x and +x
directions appear as peaks in the respective Stokes and anti-Stokes spectra.
Typical Brillouin spectra, recorded at various wavevectors and external fields
H0 , are presented in Fig. 5.6(b). Pairs of Stokes and anti-Stokes peaks have asym-
86
(a)
y
H0
a
sc
dl
re
tte
ci
in
H0
k Stokes
ht
ig
ht
ig
l
nt
de
k anti-Stokes
Intensity (arb. units)
(b)
k=21.2m-1
k=12.2m-1
k=2.1m-1
-30
-15
0
15
Frequency (GHz)
30
Figure 5.6: (a) Schematics of Brillouin light scattering geometry, with

scattering plane (in blue), and Cartesian coordinates. (b) Brillouin spectra of the Pt/CoFeB film measured at various spin-wave wavevectors.
Green and red dots represent spectra recorded under applied field H0 = 0.7 T/0
along +z and z directions, respectively.
metric frequencies, with the frequency difference f being more pronounced for
larger spin-wave wavevectors. For a laser light incident angle of 60(corresponding
to wavevector k = 21.2 m1 ), the frequency difference f is almost as large as
2 GHz. Clearly, based on the above-mentioned numerical calculations, this large
difference cannot be fully accounted for by asymmetric surface pinning, and hence
it principally arises from interfacial DMI.
87

Frequency f (GHz)
14.5
(a)
H0 = 0.7 T/0
14.0
13.5
D =0
13.0
12.5
Stokes
-20
anti Stokes
0
Wavevector k (m-1 )
20
Df (GHz)
(b)
H0 = 0.7 T/0
-2
-20
0
Wavevector k (m-1 )
20
Figure 5.7: (a) Spin-wave dispersion relation of the Pt/CoFeB film measured under an in-plane applied field H0 = 0.7 T/0 along the z direction. (b) Frequency difference of counter-propagating SWs as a function
of wavevector. Open circles denote measured data. The symmetrical dotted lines
in (a) represent the dispersion curve calculated for zero DMI, while solid lines in
(a) and (b) represent respective fits to experimental data.
5.3.2.5
Dispersion Relation
The measured spin-wave dispersion relation is shown in Fig. 5.7(a). Unlike

the typical V-shaped dispersion curves of surface SWs in thicker magnetic thin
films, the measured dispersion curve of the Pt/CoFeB sample is monotonic and
88
basically linear for the range of k studied. It is noteworthy that if D = 0, the

magnon propagation, based on Eq. 5.6, will be almost dispersionless [see Fig.
5.7(a)], which is a consequence of the ultra-thinness of the film studied. The
DMI-induced second term on the right hand side of Eq. 5.6 causes the dispersion
curve to become an inclined straight line. This is indicative of the presence of
very strong spin-wave nonreciprocity as the detected SWs possess a near-constant
group velocity, irrespective of the sign of their wavevectors k. Interestingly, due to
the DMI-induced linear term, the +k SWs exhibit negative group velocities. The
D = 0 dispersion curve of Fig. 5.7(a) shows that for small positive wavevectors,
the SWs possess a negative group velocity, unlike typical DE-type SWs. It can
easily be seen from Eq. 5.6 that this is a consequence of the PMA (and hence
MS < HU ) in our film.
From Eq. 5.7, the difference in the frequency of counter-propagating SWs
is linear in D and k. The experimental variation of f with wavevector k is
presented in Fig. 5.7(b). A linear fit to the experimental data, based on /0 =
190 GHz/T and the VSM-measured value of MS = 1.6 106 A/m, yields a DMI
constant value of D = 1.0 0.1 mJ/m2 , where the error is estimated from Eq.
5.7 taking into account the experimental uncertainties of f , , MS , and k.
This value represents the effective bulk DMI constant, i.e. the interfacial DMI
constant averaged over the film thickness [24]. This effective bulk DMI constant
is relatively large compared with those of other similar systems, such as Pt/NiFe
(D = 0.1 0.6 mJ/m2 ) and Hf/CoFeB (D = 0.5 mJ/m2 ) [50, 55]. The /0 value
[190 109 rad/(sT)] used for our CoFeB film is reasonable as it is close to the
respective typical values of 194 and 185 109 rad/(sT) for Co and Fe [56]. The
fitted exchange constant A is about 5 pJ/m, which is smaller than typical values
89

for CoFeB. The estimated 95% confidence interval of A from the fitting is also
very large (> 100%), meaning that an accurate value cannot be extracted from
our data. However, from Eq. 5.7 it should be noted that D is independent of A.
5.4
Discussion: Physical Explanation of DMIinduced Asymmetric Properties
In Section 5.2, the derived spin-wave dispersion relation (Eq. 5.6) has an
antisymmetric term arising from the interfacial DMI. This feature has been corroborated by the experimental results of the Pt/Co/Ni and Pt/CoFeB samples
in Section 5.3. Further, asymmetric linewidths for counter-propagating SWs have
also been observed, consistent with our analytical analysis (Eq. 5.8). Nonetheless, more intuitive understanding of these properties, which gives a clear physics
picture, is still desired.
5.4.1
Asymmetric Dispersion Relation
We first discuss the underlying physics behind the DMI-induced nonreciprocal

frequency shift. For simplicity, we assume that the dynamic magnetization of the
SWs is circularly polarized and can be expressed as m = mx + my = ex sin(t
kx) + ey cos(t kx), where ex and ey are unit vectors along the +x and +y
directions, respectively (see Figs. 5.2 and 5.6). Based on Eq. 5.3, the effective
field arising from the DMI is [25, 57]
hDMI = D
m
ez
x
90
= D km,
(5.10)
5.4 Discussion: Physical Explanation of DMI-induced Asymmetric

Properties
(a)
(b)
z
Heff
z
Heff
MHeff
MHeff
MhDMI
precession
precession Mh
DMI
M
k>0
k<0
Figure 5.8: Illustration of DMI-induced frequency asymmetry. Precession

of the magnetization M under the total effective field Heff + hDMI for (a) k > 0
and (b) k < 0. All the vectors in the xy-plane are labelled blue.
where D = 2D/0 MS . Equation 5.10 shows that the dynamic effective DMI
field is proportional to the wavevector k and is either parallel or antiparallel
to m. As Fig. 5.8 shows, the magnetization vector M precesses under the
influence of the total effective field Heff + hDMI according to the Landau-Lifshitz
equation dM /dt = M (Heff + hDMI ), where > 0, and the damping term
has been neglected. Therefore, the magnitude of M (Heff + hDMI ) determines
the frequency of the precession. For SWs propagating along the +x direction
[k > 0, see Fig. 5.8(a)], M hDMI is antiparallel to M Heff , and hence the
precession frequency will be lower than that for zero DMI. In contrast, for SWs
travelling in the x direction [k < 0, see Fig. 5.8(b)], M hDMI is parallel
to M Heff , resulting in a faster precession and hence higher frequencies.
Furthermore, because M hDMI is proportional to k, the above arguments also
account for the experimental observation that f is larger for larger wavevectors.
91
BIBLIOGRAPHY
5.5
Summary
Using BLS, we have made the first direct observation of interfacial DMI in
ultrathin multilayers, such as Pt/Co/Ni and Pt/CoFeB films. The measured
asymmetric dispersion relations, arising from interfacial DMI, agree well with
theory. Our measurement shows that the linewidth of SWs in the former film
is also asymmetric and depends on their propagation direction, consistent with
our calculated linewidths based on the linearized LLG equation. Moreover, the
asymmetric phenomena studied have been qualitatively explained from analysing
the direction of the torque exerted by the effective DMI field on the magnetization,
which provides a more intuitive picture of the underlying physics involved.
Bibliography
[1] I. Dzyaloshinsky, Journal of Physics and Chemistry of Solids 4, 241 (1958).
69
[2] T. Moriya, Physical Review 120, 91 (1960). 69
[3] Y. Onose, T. Ideue, H. Katsura, Y. Shiomi, N. Nagaosa, and Y. Tokura,
Science 329, 297 (2010). 69
[4] J. Z. Zhao, X. Q. Wang, T. Xiang, Z. B. Su, and L. Yu, Physical Review
Letters 90, 207204 (2003). 70
[5] K. F. Wang, J. M. Liu, and Z. F. Ren, Advances in Physics 58, 321 (2009).
70
[6] S. Mhlbauer, B. Binz, F. Jonietz, C. Pfleiderer, A. Rosch, A. Neubauer, R.
Georgii, and P. Bni, Science 323, 915 (2009). 70
92
BIBLIOGRAPHY
[7] M. Bode, M. Heide, K. von Bergmann, P. Ferriani, S. Heinze, G. Bihlmayer,

A. Kubetzka, O. Pietzsch, S. Blugel, and R. Wiesendanger, Nature 447, 190
(2007).
[8] P. Ferriani, K. von Bergmann, E. Y. Vedmedenko, S. Heinze, M. Bode, M.
Heide, G. Bihlmayer, S. Blgel, and R. Wiesendanger, Physical Review Letters 101, 027201 (2008). 70
[9] X. Z. Yu, N. Kanazawa, W. Z. Zhang, T. Nagai, T. Hara, K. Kimoto, Y.
Matsui, Y. Onose, and Y. Tokura, Nature Communications 3, 988 (2012).
70
[10] J. Iwasaki, M. Mochizuki, and N. Nagaosa, Nature Nanotechnology 8, 742
(2013). 70
[11] N. Romming, C. Hanneken, M. Menzel, J. E. Bickel, B. Wolter, K. von
Bergmann, A. Kubetzka, and R. Wiesendanger, Science 341, 636 (2013).
70
[12] A. Crpieux and C. Lacroix, Journal of Magnetism and Magnetic Materials
182, 341 (1998). 70
[13] A. N. Bogdanov and U. K. Rler, Physical Review Letters 87, 037203
(2001). 70, 71
[14] L. Udvardi and L. Szunyogh, Physical Review Letters 102, 207204 (2009).
77
[15] K. Zakeri, Y. Zhang, J. Prokop, T. H. Chuang, N. Sakr, W. X. Tang, and J.
Kirschner, Physical Review Letters 104, 137203 (2010). 70, 77
[16] A. Fert and P. M. Levy, Physical Review Letters 44, 1538 (1980). 70
[17] A. Fert, V. Cros, and J. Sampaio, Nature Nanotechnology 8, 152 (2013). 70,
84
93
BIBLIOGRAPHY
[18] F. Garcia-Sanchez, P. Borys, A. Vansteenkiste, J.-V. Kim, and R. L. Stamps,

Physical Review B 89, 224408 (2014). 70
[19] M. Mena, R. S. Perry, T. G. Perring, M. D. Le, S. Guerrero, M. Storni, D.
T. Adroja, C. Regg, and D. F. McMorrow, Physical Review Letters 113,
047202 (2014). 70
[20] K. Zakeri, Y. Zhang, T. H. Chuang, and J. Kirschner, Physical Review Letters
108, 197205 (2012). 70, 76
[21] M. Tsoi, A. G. M. Jansen, J. Bass, W. C. Chiang, M. Seck, V. Tsoi, and P.
Wyder, Physical Review Letters 80, 4281 (1998). 70
[22] M. Jamali, J. H. Kwon, S.-M. Seo, K.-J. Lee, and H. Yang, Scientific Reports
3, 3160 (2013). 70
[23] A. Bogdanov and D. Yablonskii, Sov. Phys. JETP 68, 101 (1989). 71
[24] S. Rohart and A. Thiaville, Physical Review B 88, 184422 (2013). 71, 72, 89
[25] J.-H. Moon, S.-M. Seo, K.-J. Lee, K.-W. Kim, J. Ryu, H.-W. Lee, R. D.
McMichael, and M. D. Stiles, Physical Review B 88, 184404 (2013). 72, 90
19, 7013 (1986). 72
[27] S. Mangin, D. Ravelosona, J. A. Katine, M. J. Carey, B. D. Terris, and E. E.
Fullerton, Nature Materials 5, 210 (2006). 73
[28] S. Fukami, T. Suzuki, H. Tanigawa, N. Ohshima, and N. Ishiwata, Applied
Physics Express 3, 113002 (2010). 73
[29] S. Emori and G. S. D. Beach, Applied Physics Letters 98, 132508 (2011). 73
[30] G. Chen, T. Ma, A. T. NDiaye, H. Kwon, C. Won, Y. Wu, and A. K. Schmid,
Nature Communications 4, 2671 (2013). 73
[31] A. Thiaville, S. Rohart, E. Ju, V. Cros, and A. Fert, Europhysics Letters
94
BIBLIOGRAPHY
100, 57002 (2012).

[32] K.-S. Ryu, L. Thomas, S.-H. Yang, and S. Parkin, Nature Nanotechnology
8, 527 (2013). 73
[33] S.-Y. Huang, X. Fan, D. Qu, Y. Chen, W. Wang, J. Wu, T. Chen, J. Xiao,
and C. Chien, Physical Review Letters 109, 107204 (2012). 73
[34] X. Qiu, P. Deorani, K. Narayanapillai, K.-S. Lee, K.-J. Lee, H.-W. Lee, and
H. Yang, Scientific Reports 4, 4491 (2014). 74, 84
[36] F. C. Nrtemann, R. L. Stamps, R. E. Camley, B. Hillebrands, and G. Gntherodt, Physical Review B 47, 3225 (1993). 77
[37] B. Heinrich, S. T. Purcell, J. R. Dutcher, K. B. Urquhart, J. F. Cochran, and
A. S. Arrott, Physical Review B 38, 12879 (1988). 77
[38] R. E. Camley, Surface Science Reports 7, 103 (1987). 78, 83
[39] M. Suzuki, H. Muraoka, Y. Inaba, H. Miyagawa, N. Kawamura, T. Shimatsu,
H. Maruyama, N. Ishimatsu, Y. Isohama, and Y. Sonobe, Physical Review B
72, 054430 (2005). 79
[40] G. Chen, J. Zhu, A. Quesada, J. Li, A. T. NDiaye, Y. Huo, T. P. Ma, Y.
Chen, H. Y. Kwon, C. Won, et al., Physical Review Letters 110, 177204
(2013). 79
[41] W. F. Oliver, C. A. Herbst, S. M. Lindsay, and G. H. Wolf, Review of Scientific Instruments 63, 1884 (1992). 80
[42] J. Jorzick, S. O. Demokritov, C. Mathieu, B. Hillebrands, B. Bartenlian, C.
Chappert, F. Rousseaux, and A. N. Slavin, Physical Review B 60, 15194
(1999). 82
95
BIBLIOGRAPHY
[43] S. Mizukami, E. P. Sajitha, D. Watanabe, F. Wu, T. Miyazaki, H. Naganuma,

M. Oogane, and Y. Ando, Applied Physics Letters 96, 152502 (2010). 82
[44] B. Hillebrands, Physical Review B 41, 530 (1990). 83
[45] T. Kawahara, K. Ito, R. Takemura, and H. Ohno, Microelectronics Reliability
52, 613 (2012). 84
[46] S. Ikeda, K. Miura, H. Yamamoto, K. Mizunuma, H. D. Gan, M. Endo, S.
Kanai, J. Hayakawa, F. Matsukura, and H. Ohno, Nature Materials 9, 721
(2010). 84, 85
[47] S. Ikeda, J. Hayakawa, Y. Ashizawa, Y. M. Lee, K. Miura, H. Hasegawa, M.
Tsunoda, F. Matsukura, and H. Ohno, Applied Physics Letters 93, (2008).
84
[48] P. P. J. Haazen, E. Mur, J. H. Franken, R. Lavrijsen, H. J. M. Swagten, and
B. Koopmans, Nature Materials 12, 299 (2013). 84
[49] S. Emori, U. Bauer, S.-M. Ahn, E. Martinez, and G. S. D. Beach, Nature
Materials 12, 611 (2013).
[50] J. Torrejon, J. Kim, J. Sinha, S. Mitani, M. Hayashi, M. Yamanouchi, and H.
Ohno, Nature Communications 5 (2014). 84, 85, 89
[51] S. Heinze, K. von Bergmann, M. Menzel, J. Brede, A. Kubetzka, R. Wiesendanger, G. Bihlmayer, and S. Blugel, Nature Physics 7, 713 (2011). 84
[52] G. Malinowski, K. C. Kuiper, R. Lavrijsen, H. J. M. Swagten, and B. Koopmans, Applied Physics Letters 94, 102501 (2009). 85
[53] M. Donahue and D. G. Porter, OOMMF Users Guide, Version 1.0, Interagency Report NISTIR 6376 (NIST, Gaithersburg, MD, USA). (1999). 86
[54] K. Di, H. S. Lim, V. L. Zhang, S. C. Ng, M. H. Kuok, H. T. Nguyen, and M.
G. Cottam, Journal of Applied Physics 115, 053904 (2014). 86
96
BIBLIOGRAPHY
[55] H. T. Nembach, J. M. Shaw, M. Weiler, E. Ju, and T. J. Silva, arXiv preprint

p. 1410:6243 (2014). 89
[56] S. Chikazumi, Physics of Ferromagnetism (Oxford University Press, New
York, 1997). 89
[57] K. Di, V. L. Zhang, H. S. Lim, S. C. Ng, M. H. Kuok, J. Yu, J. Yoon, X. Qiu,
and H. Yang, Physical Review Letters 114, 047201 (2015). 90
97
BIBLIOGRAPHY
98
6
Spin-Wave Nonreciprocity
6.1
Introduction
Besides being of scientific interest, the phenomenon of the nonreciprocal propagation of waves is of great technological importance. It may exist in structures
that are both asymmetric and nonlinear [1], or systems with broken time-reversal
symmetry [2, 3]. From the perspective of applications, nonreciprocity provides
an extra degree of control in molding the flow of waves in fields ranging from mature microwave technology [4] to the rapidly expanding areas of photonics [57],
magnonics [811] and phononics [1, 12]. Furthermore, it is crucial for the unidirectional propagation of waves and for the stabilization of integrated circuits
as well as the suppression of undesirable crosstalk and interference arising from
imperfection-induced scattering [6].
While much work has been undertaken on optical nonreciprocity, relatively
little has been done on nonreciprocity involving spin-wave propagation. In the
area of magnonics, nonreciprocity of SWs is an important requisite property for
99
6. SPIN-WAVE NONRECIPROCITY
various functionalities. Thus far, studies of magnonic nonreciprocity [811] have
focused on the long-wavelength surface DE waves with propagation direction perpendicular to the applied magnetic field. Such a nonreciprocity originates from
the lower symmetry at sample surfaces and dipole-dipole interactions rather than
from isotropic exchange interactions [2]. However, in the sub-100nm wavelength
range, SWs are exchange-dominated and nonreciprocity due to dipole-dipole interaction is less significant. Hence, nanoscale miniaturization of magnonic devices
necessitates the realization of nonreciprocity of exchange-dominated SWs.
In Chapter 5, we have shown that the presence of interfacial DMI will lead to
asymmetric dispersion relations and lifetimes of counter-propagating SWs. This
phenomena could be exploited in the design of nonreciprocal spin-wave devices. In
this Chapter, two more theoretical approaches, namely, nonreciprocal interband
magnonic transitions and enhanced nonreciprocity by antiferromagnetic (AFM)
coupling, will be introduced to realize strong nonreciprocity of SWs with short
wavelengths (few nm to 100 nm).
6.2
Nonreciprocity by Interband Magnonic Transitions
We will show that nanoscale linear nonreciprocity of exchange-dominated SWs

in a nanostripe waveguide can be achieved by applying a time-reversal and spaceinversion symmetry-breaking magnetic field. Depending on their propagation direction, SWs travelling in the waveguide may undergo interband magnonic transitions. It is to be noted that the well-studied nonreciprocity arising from magnetic
100
6.2 Nonreciprocity by Interband Magnonic Transitions
(a)
H0
32 nm
4 nm
(b)
Figure 6.1: Nonreciprocal spin-wave mode transitions in the Py nanostripe waveguide. (a) The perturbation magnetic field is applied in the y-direction
within the blue region of the waveguide. The enlarged view shows the instantaneous profile of the perturbation field vector represented by arrows. (b) Dispersion
relations of SWs with blue and red lines representing the respective even and odd
symmetry bands. The solid (dashed) green arrow indicates shifts of frequency and
wavevector in the forward (backward) propagation direction due to the perturbation field. Resonant coupling exists between modes [1 , k1 ] and [2 , k2 ] in the
forward (+x) direction, while coupling between modes [1 , k1 ] and [2 , k2 ] in
the backward (-x) direction is insignificant.
dipole-dipole interaction is completely different from our proposed nonreciprocity. First, our proposed nonreciprocity arises from interband magnonic transitions
and is present even when the magnetic dipolar interaction is negligible. Second,
frequency and wavevector are reciprocal quantities in uniform thin films for the
former case [2], which is no longer so in our case.
The waveguide investigated is an 8m-long Ni80 Fe20 (Permalloy, Py) nanos-
101
tripe of rectangular cross section, with a width w of 32 nm and a thickness of 4
nm, as depicted in Fig. 6.1(a). A bias external magnetic field H0 = 0.3 T/0 ,
where 0 is the vacuum permeability, is applied in the +y-direction. Figure 6.1(b)
shows the lowest and second-lowest energy bands, with respective even and odd
symmetry mode profiles across the width, of the magnonic dispersion relations of
the waveguide calculated from micromagnetic simulations [13] (see Chapter 2).
For brevity, a spin-wave mode with angular frequency n and wavenumber kn
will be labeled [n , kn ], where n = 1, 2.
6.2.1
Coupled-Mode Analysis
For the linearized LLG equation, the dynamic magnetization of SWs [14, 15]
is given by

mx,n 1
mn =
= fn (y)exp i(kn x n t) ,
i
mz,n
(6.1)

where the mode profile fn (y) = an cos n (y + w/2) , and kn is the wavenumber
along the x-axis. In the definition of fn , the effective transverse wavevector
n = pn /w, where pn = 0, 1, 2, . . . is the quantization number of mode mn
(free-spin boundary condition [15]), an the normalization constant determined by
R w/2
f dy = w, and w the width of the waveguide. Note that fn constitutes
w/2 n

a complete orthogonal basis [14] over the interval w2 , w2 , and that fn = 1

and fn = 2 cos (y + w/2)/w for the lowest and second-lowest energy bands
respectively.
In analogy to indirect electronic transitions in semiconductors and indirect
102
photonic transitions in photonic structures [5, 7] (indirect transitions involve simultaneous shifts in energy and momentum), indirect transitions between two
spin-wave modes within the waveguide can be induced by a perturbation magnetic field h0 , of frequency and wavevector Q, applied along the y-direction
h0 (x, y, t) = h0 cos(Qx t),
(6.2)
where the amplitude of the field is defined by h0 (y) = 0.05 T/0 for y < 0, and
h0 (y) = 0 T/0 for y > 0 [see Fig. 6.1(a)]. Significant transition between two
spin-wave modes will occur only if the phase-matching conditions [16], namely
Q = k2 k1 and = 2 1 , where 1 and 2 denote the respective initial and
final modes, are approximately satisfied.
The analytical coupled-mode theory explaining the exchange magnonic mode
transition dynamics is developed as follows. As the considered spin-wave wavelengths (around 30 nm) lie within the sub-100nm range where the exchange interaction dominates [17], it is a reasonable approximation to disregard magnetic
dipole-dipole interactions in the analysis. Therefore, neglecting the Gilbert damping term, the linearized LLG equation for the Py waveguide under field h0 can be
written as

1 d
ex + H
0 + H
0 m,
m= H
(6.3)
dt

2
2
2
2
x
where the dynamic magnetization m = m
,
mz , Hex = J /x + /y

0 = H0
0 = h0
H
, H
,
= 10 1
0 , and J = 2A/ (0 MS ). In the following
calculations, the saturation magnetization MS , exchange constant A and gyromagnetic ratio of Py, were set to 8 105 A/m, 1 1011 J/m and 2.211 105
Hzm/A, respectively.
103
We assume that the perturbation field h0 approximately satisfies the phasematching conditions k2 k1 Q = k 0, and 2 1 = 0, where k2 =
0.149 nm1 , k1 = 0.170 nm1 , 2 /2 = 35.5 GHz, and 1 /2 = 36.6 GHz. In the
0 H
ex + H
0 ), which propagates
presence of the weak perturbation field h0 (i.e. H
in the +x direction, mn are no longer eigenmodes, and the SWs is expressible as
a linear combination of the basis mn , i.e.,
m=
(6.4)
Cn (x)mn ,
where n spans all the eigenmodes, and Cn are coefficients that reflect the mode
coupling along the waveguide. Using Eqs. 6.1 to 6.4 and the slowly varying
amplitude approximation [16] for Cn in which terms containing d2 Cn (x)/dx2 are
neglected, one obtains

X
dCn (x)
0
mn h Cn (x)mn = 0.
2ikn J
dx
n
(6.5)
Taking the inner product of the above equation with m1 the complex conjugate
of m1 , and integrating over y, yields

dC1 (x) X
4ik1 Jw
=
Cn (x) eikn+ x ein+ x + eikn x ein x
dx
n
w/2
h0 (y)f1 fn dy,
w/2
(6.6)
where kn = kn k1 Q, and n = n 1 . Although Eq. 6.6 contains an

infinite number of terms on the right, imposition of the phase-matching conditions
reduces the number to just one. First, due to energy conservation, the occurrence
of mode transition necessitates exact fulfillment of the condition [5] n = 1 .
Second, we need only retain the phase-matching terms with kn 0 on the right-
104
(a)
(b)
1,114
x (nm)
2,228
1,120
x (nm)
2,240
Figure 6.2: Forward spin-wave mode transitions in the waveguide. (a)

Intensities of coupled magnon modes [1 , k1 ] (blue curves) and [2 , k2 ] (red curves)
obtained from the coupled-mode model (solid curves) and micromagnetic simulations (dashed curves), respectively. (b) x-y view of instantaneous dynamic magnetization mz based on the coupled-mode theory (top panel) and micromagnetic
simulations (bottom panel). Dashed boxes denote the regions where the perturbation field is applied. Transitions between the two modes result in the mode profiles
(across the waveguide width) to gradually evolve, with increasing x, from being
even to odd and back to even symmetry. The coupling lengths are determined
to be 1114 and 1120 nm based on the coupled-mode theory and the micromagnetic simulations, respectively. The magnitudes of mz are color-coded, with red,
white and blue representing positive maximum, zero and negative maximum, respectively. The arrows indicate the propagation of the two modes in the forward
(+x) direction.
hand side, and neglect the phase-mismatch terms, because the integral of the
latter, over sufficiently long distance in the x-direction, vanishes [16], indicating
zero contribution. The phase matching conditions reduce Eq. 6.6 to a two-state
105
problem, namely,
0
ie
d C1 (x)
=
dx
C2 (x)
ieixk K/2
where the overlap integral K =
R w/2
w/2
ixk
K/1 C1 (x)
,
0
C2 (x)
(6.7)
h0 (y)f1 f2 dy and n = 4kn Jw (n {1, 2}).
It should be emphasized that mode transition is possible for any two modes
irrespective of their symmetry, provided that the phase-matching conditions are
satisfied, and the coupling coefficient K has a nonzero value. It is to be noted
that to fulfill the latter condition, the perturbation field can be of a form different
from that of Eq. 6.2, which is chosen for simplicity of calculations.
Given the initial conditions C1 (0) = 1 and C2 (0) = 0, the solutions to Eq. 6.7
are
"
#

p

p
k/2
sin x a + (k/2)2
C1 (x) = eixk/2 cos x a + (k/2)2 + i p
a + (k)2
r
p

1
ieixk/2
sin x a + (k/2)2 ,
C2 (x) = p
1 + k 2 /4a 2
(6.8)
where a = K 2 /(1 2 ). If the phase-matching conditions are exactly satisfied
(i.e. k = 0), the mode transition will be complete at integral multiples of the
coupling length lc = /(2 a), defined as the propagation distance over which
one mode undergoes a complete transition to the other. Note that lc is inversely
proportional to the strength of the field h0 . If k 6= 0, then |C1 | never vanishes,
indicating incompletion of the mode transition and energy transfer. However,
for strongly phase-mismatching cases (k a), e.g., when SWs propagate

in the reverse (x) direction, we have C2 C1 , indicating that no significant
106
mode transition will occur. Hence, only for certain pairs of Q and values
will the phase-matching conditions be simultaneously satisfied for transitions between modes [1 , k1 ] and [2 , k2 ], but not for those between mode [1 , k1 ] and
any other mode [see Fig. 6.1(b)]. This means that mode transitions will occur
only for one propagation direction, indicating that spin-wave propagation in the
waveguide is nonreciprocal. Of note is that this nonreciprocity phenomenon is
due to the breaking of time-reversal and spatial-inversion symmetry induced by
the perturbation magnetic field.
The coupling length was calculated to be 1114 nm. The intensities of modes
[1 , k1 ] and [2 , k2 ], evaluated from Eq. 6.8, are presented in Fig. 6.2(a). Due to
Gilbert damping, the intensities of modes along the x-direction are modified by
a factor ex/ , where the spin-wave decay length [18] = vg /(2), vg being the
spin-wave group velocity and the Gilbert damping coefficient. The corresponding instantaneous profile of the z-component mz of the dynamic magnetization is
presented in Fig. 6.2(b).
6.2.2
Micromagnetic Simulation
Micromagnetic simulations were performed to investigate the dynamics of the

predicted spin-wave nonreciprocity by solving the LLG equation using the oommf
package [19] based on a unit cell size of 2 2 4 nm3 and = 5 104 , with
both dipolar and exchange interactions taken into account and neglecting magnetocrystalline anisotropy. Unwanted SW reflections were eliminated by setting
the Gilbert damping to 1.0 at both ends of the waveguide.
To determine the coupling length lc , the magnetic-field perturbation region was
107
first chosen to be sufficiently long along the x-direction, i.e., for the occurrence
of two complete mode transitions. A continuous SW of mode [1 , k1 ] was excited
at the left end of the waveguide. The simulated intensities |C1 |2 and |C2 |2 of
the respective modes [1 , k1 ] and [2 , k2 ] as a function of interaction distance
within the perturbation region, obtained by performing a fast Fourier transform
of the out-of-plane component mz of the dynamic magnetization in time, are
displayed in Fig. 6.2(a). The simulated coupling length is about 1120 nm, which
accords well with the coupled-mode theory in terms of both the relative mode
intensities and the coupling length. Both methods indicate that the energy is
transferred back and forth between the coupled modes. It it noteworthy that the
simulated coupling length is found to be inversely proportional to the strength
of the perturbation field, consistent with the coupled-mode theory. Figure 6.2(b)
shows that the simulated instantaneous dynamic magnetization reproduces well
the theoretically calculated one. It also reveals that the forward-propagating
mode undergoes a transition, with the symmetry of its mz profile progressively
evolving from even to odd and finally reverting to even.
To demonstrate the concept of nonreciprocal mode transitions, the length of
the perturbation region along the x-axis was then set to the simulated coupling
length lc . The frequencies and propagation directions of the initial wave were varied by changing the frequency and location (left or right end of the waveguide) of
the excitation field. In order to excite SWs with the desired (even or odd) symmetry, the applied excitation field should possess the same symmetry, namely, even
or odd, respectively [13]. The snapshot of mz , presented in Fig. 6.3(a), reveals
that the symmetry of the mode profile of the right-propagating SW is completely
converted from even to odd. The right panel of Fig. 6.2(a) shows that the cor-
108
(a)
[ , k ]
1
[ , k ]
(b)
[ , k ]
1
[ , k ]
(c)
[ , k ]
2
[ , k ]
1
(d)
[2 , k 2 ]
[2 , k 2 ]
1,120 ( lc )
0
x (nm)
35
36
/2 (GHz)
37
Figure 6.3: Snapshots of out-of-plane components of the dynamic magnetization under the perturbation field. (a) Spin wave of mode [1 , k1 ] propagating forward (from left to right) undergoing a complete transition to mode
[2 , k2 ]. (b) Spin wave of mode [1 , k1 ] propagating backward (from right to
left) remaining unaltered. (c) Spin wave of mode [2 , k2 ] propagating forward undergoing a complete transition to mode [1 , k1 ]. (d) Spin wave of mode [2 , k2 ]
propagating backward remaining unaltered. The corresponding mode frequency
changes are indicated by panels on the right. Dashed boxes denote regions where
the perturbation field is applied.
responding frequency spectrum changes from 1 to 2 after passing through the

perturbation region. In contrast, on reversal of the propagation direction, i.e.,
when SWs of mode [1 , k1 ] were considered, no changes in their mode profiles
or frequencies were found [see Fig. 6.3(b)], indicating the occurrence of negligible
mode transition. Thus, the waveguide exhibits time-reversal symmetry breaking,
resulting in drastic contrasting effects for the two opposite propagation directions.
The same nonreciprocal effects were also observed for the initial modes [2 , k2 ]
and [2 , k2 ] as illustrated in Figs. 6.3(c), (d).
The magnonic nonreciprocity presented here has the following merits. First,
being valid for exchange-dominated SWs, it could in principle permit the real-
109
ization of nanoscale integrated magnonic circuits. Second, the linearity of the
phenomenon makes its realization more straightforward and energy-efficient than
that involving nonlinear effects. Recently, it was reported that ultrafast dynamic
control of photonic [20, 21] and magnonic [22, 23] structures offers approaches
for achieving novel functionalities, such as the frequency shift of waves, which
usually involves nonlinear effects in static artificial materials, i.e. those with
time-invariant structural arrangements. We have shown, from another perspective, that the dynamic modulation of a magnonic waveguide can give rise to
hitherto unreported nonreciprocal propagation of SWs.
6.3
Nonreciprocity Enhanced by Antiferromagnetic coupling
In this Section, instead of seeking a different mechanism, we propose an alternative method to enhance the dipolar-interaction-induced spin-wave nonreciprocity in MCs via synthetic AFM coupling, which changes the symmetry property
of dipolar interaction. In the absence of AFM coupling, the nonreciprocity is
found to be negligible. By studying the magnetic field dependence of the nonreciprocity, we found that the enhancement is closely correlated with the AFM
alignment of the magnetizations. Finally, a modification of the AFM MCs by
magnetic anisotropy is proposed to facilitate miniaturization of devices based on
them.
The MC considered is a 2m-long, 30nm-wide ferromagnetic nanostripe decorated with a periodic array of nanocuboids, as shown in Fig. 6.4 [2426]. The
110
6.3 Nonreciprocity Enhanced by Antiferromagnetic coupling
top layer
x
H0
10 nm
10 nm
bo!om layer
10 nm
30 nm
Figure 6.4: Schematic diagram of the magnonic crystal with AFM coupling. The external magnetic field H0 is applied transverse (along y-axis) to the
long axis of the nanostripe. Green arrows represent the respective directions of the
magnetizations of the top and bottom layers under an external field H0 = 0.1 T/0
along the y-axis.
crystal has a lattice constant a = 20 nm and consists of two layers of equal

thickness, between which the exchange coupling can be of either the AFM or
ferromagnetic (FM) type. As is usually done in recording media applications
[2729], the AFM interlayer coupling can be realized by separating the two FM
layers by a nonmagnetic (e.g. ruthenium) spacer layer, which could be thinner
than 1 nm, and as such has been neglected in our study [30, 31]. For simplicity,
we assume that the MC is made of only one material (Permalloy), with a saturation magnetization MS = 800 kA/m, and an exchange constant A0 = 13 pJ/m.
The AFM coupling between the two layers is taken to be A1 = 3 pJ/m. An
external magnetic field H0 is applied transverse, in the y-direction, to the long
axis of the crystal.
Micromagnetic simulations based on oommf (see Chapter 2) were adopted to
investigate the magnonic dispersion relations in the MCs. Figure 6.5 shows the
calculated dispersion relations in the first Brillouin zone (BZ) of the MCs with
respective FM and AFM couplings under applied transverse fields of 0.06 and 0.2
T/0 . It is obvious that the dispersion curves of the MC with FM coupling are
111
almost symmetric with respect to the spin-wave wavevector k. This is consistent
with the argument that since the spin-wave wavelengths studied are in the sub100nm range, nonreciprocity is expected to be insignificant. To rely on such a
weak nonreciprocity in MC with FM coupling, devices based on it necessarily
have to have very large footprints. In contrast, the AFM-coupled MC possesses
highly asymmetric dispersion curves. It is obvious that for the same wavevector,
counter-propagating SWs exhibit very different group velocities and frequencies,
and generally, the AFM MC has higher group velocities than those of the FM
MC. Both crystals exhibit a few band gap openings due to Bragg scattering in the
periodic structures. It is interesting that the band gap openings of the AFM MC
are not at the BZ boundaries anymore, because the dispersion curves of counterpropagating waves no longer intersect at the BZ boundary due to their different
slopes [32]. Note that this property is a natural property of gyrotropic periodic
materials [33, 34]. Another interesting property of the band structure of the MC
with AFM coupling is that the SWs now acquire nonzero group velocities when
k = 0.
We next study the dependence of the spin-wave nonreciprocity on the applied transverse field H0 . For brevity, nonreciprocity is defined as n = [f (k)
f (k)]/[f (k) + f (k)], where f (k) is the frequency of the lowest magnon branch
at k = 0.05 nm1 . Figure 6.6 shows that nonreciprocity is most pronounced for
0 T 0 H0 0.3 T approximately, and is insignificant for 0 H0 > 0.3 T. This
can be qualitatively explained by the different relative orientations of the equilibrium magnetizations of the top and bottom layers under different external fields.
For 0 T 0 H0 0.3 T, the angle between the two magnetizations is nearly
180, i.e. the magnetizations are almost perpendicular to the x-axis, and the
112
-/a
50
/a
-/a
50
Frequency (GHz)
(a)
/a
(b)
40
40
30
30
20
20
10 FM
0.06T/0
0
0.15
0
-/a
50
10 AFM
0.06T/0
0
0.15 0.15
0
/a
-/a
50
(c)
0.15
/a
(d)
40
40
30
30
20
20
10 FM
0.2T/0
0
0.15
0
10 AFM
0.2T/0
0
0.15 0.15
0
0.15
-1
Wavevector k (nm )
Figure 6.5: Calculated dispersion curves of the MCs (a = 20 nm) with (a, c)
FM and (b, d) AFM coupling under transversely applied fields H0 = 0.06 and 0.2
T/0 . Insets: Corresponding directions of the equilibrium magnetizations of the
top (red arrows) and bottom (blue arrows) layers.
180
120
0.1
60
0.0
Angle (deg)
Nonreciprocity
0.2
0
0.0
0.2
0.4
0.6
Applied Transverse Field (T/0)

Figure 6.6: The calculated (a) nonreciprocity (denoted by red triangles) and (b)
angle (denoted by blue squares) between the respective magnetizations of the top
and bottom layers as functions of applied transverse field of the AFM crystal (a =
20 nm). The solid lines only serve as a guide to the eye.
113
-/a
50
/a
Frequency (GHz)
(a)
-/a
50
(b)
40
40
30
30
20
20
10 a=20nm
10 a=20nm
KU=0
0
0.15
-/a
30
/a
0
0.15 0.15
/a
-/a
30
(c)
KU=40kJ/m3
0.15
/a
(d)
20
20
10
10
y
a=60nm
KU=0
0
0.05
a=60nm
KU=40kJ/m3
0
0.05 0.05
y
x
0.05
Wavevector k (nm-1)
Figure 6.7: Band structures, calculated for zero applied field, of various AFM
magnonic crystals for (a) lattice constant a = 20 nm, magnetic anisotropy KU
= 0 kJ/m3 , (b) a = 20 nm, KU = 40 kJ/m3 , (c) a = 60 nm, KU = 0 kJ/m3
and (d) a = 60 nm, KU = 40 kJ/m3 . Insets show the corresponding directions of
the equilibrium magnetizations on the top (red arrows) and bottom (blue arrows)
layers of the samples.
counter-propagating surface waves are subjected to highly asymmetric effective

fields. For 0 H0 > 0.3 T, the angle gradually decreases with increasing transverse
field, resulting in the lowering of the asymmetry of the effective field, and thus
the nonreciprocity. Hence, the enhanced nonreciprocity is a consequence of the
antiparallel alignment of the respective magnetizations of the top and bottom
layers of the MC. Interestingly, our calculation shows that the spin-wave nonreciprocity (H0 = 0.1 T) is dependent on the interlayer AFM exchange coupling
parameter, with a maximum value obtained at A1 1 pJ/m.
It is noteworthy that the nonreciprocity of the AFM MC is significant even in
114
the absence of an applied field. This is because, under zero field, the equilibrium
magnetizations of the crystal are oriented at a large angle to the long crystal axis
[see inset of Fig. 6.7(a)], due to the shape anisotropy. Another AFM MC with
the same magnetic parameters but a larger lattice period of a = 60 nm was also
studied. As shown in Fig. 6.7(c), its band structure is symmetric, indicating
an absence of nonreciprocity. This is because the equilibrium magnetizations, in
zero external field, are either parallel or antiparallel to the MCs long axis and the
wavevector. Hence, we deduce that the AFM configuration of the magnetizations
is a necessary but not sufficient condition for enhanced nonreciprocity. To achieve
nonreciprocity in this crystal, an external transverse field should be applied to
favor the propagation of surface waves. This is consistent with the finding of
Ref. [35], that a non-zero out-of-plane component (y direction in our case) of the
static magnetization is necessary for emergence of nonreciprocity.
Devices based on MCs that rely on the application of a magnetic field for their
operation are necessarily bulkier than those that do not, and do not allow for
easy integration into integrated magnonic circuits. To overcome this problem,
the magnetization of the bottom layer can be pinned along the y-direction (perpendicular to the long axis of the crystal) by either suitable magnetocrystalline
anisotropy, or effective surface pinning through exchange bias [36] by another
contacting antiferromagnetic layer. In the following simulations, we assume that
the bottom layers of the a = 20 nm and a = 60 nm MCs have modest secondorder uniaxial magnetic anisotropies. The anisotropies have easy axes along the
y-direction and an anisotropy constant KU = 40 kJ/m3 . In the absence of an
external magnetic field, the equilibrium magnetizations of the bottom layers of
both crystals are aligned along the y-direction. The simulated magnon disper-
115
Figure 6.8: The simulated effective internal

field (indicated by cones) Heff = Hdip + Hex
within two unit cells of the a=20nm magnonic
crystals in an external field H0 = 0.2 T/0 .
Magnonic crystal with (a) FM and (b) AFM
interlayer-coupling.
The magnitude and ycomponent of Heff are represented by the size
and color of the cones (see bottom color bar),
respectively.
sions are presented in Fig. 6.7. Figure 6.7(b) shows the highly asymmetric nature
of dispersion of the a = 20 nm MC even for H0 = 0. Under zero external field,
its nonreciprocity is higher than that of the MC with KU = 0 [see Fig. 6.7(a)].
Interestingly, the a = 60 nm MC has drastically different dispersion relations [see
Figs. 6.7(c,d)] depending on the value of KU . Even in the absence of an applied
field, its dispersion is highly asymmetric for KU = 40 kJ/m3 . As external fields
are unnecessary, this effect makes possible the on-chip integration of magnonic
devices.
116
6.4 Summary
6.4
Summary
Two methods of realizing strong spin-wave nonreciprocity at the nanoscale

have been proposed. One is based on nonreciprocal interband magnonic transitions, which necessitate a perturbation magnetic field. The other relies on
enhancing the nonreciprocity arising from magnetic dipolar interaction. In the
former approach, it is very challenging to realize the perturbation field at such
a small scale. However, the proposal is also applicable to structures of micron
dimensions. Further, the mode transition will occur as long as the perturbation
field satisfies K 6= 0 (in Eq. 6.7). In contrast, the latter approach is much easier
to realize with state-of-the-art fabrication techniques.
Bibliography
[1] B. Liang, X. S. Guo, J. Tu, D. Zhang, and J. C. Cheng, Nature Materials 9,
989 (2010). 99
[2] R. E. Camley, Surface Science Reports 7, 103 (1987). 99, 100, 101
[3] K. Di, H. S. Lim, V. L. Zhang, S. C. Ng, and M. H. Kuok, Applied Physics
Letters 103, 132401 (2013). 99
[4] B. K. Kuanr, V. Veerakumar, R. Marson, S. R. Mishra, R. E. Camley, and
Z. Celinski, Applied Physics Letters 94, 202505 (2009). 99
[5] Z. Yu and S. Fan, Nat Photon 3, 91 (2009). 99, 103, 104
[6] S. J. B. Yoo, Nat Photon 3, 77 (2009). 99
[7] H. Lira, Z. Yu, S. Fan, and M. Lipson, Physical Review Letters 109, 033901
(2012). 99, 103
117
BIBLIOGRAPHY
[8] P. K. Amiri, B. Rejaei, M. Vroubel, and Y. Zhuang, Applied Physics Letters

91, 062502 (2007). 99, 100
[9] K. Sekiguchi, K. Yamada, S. M. Seo, K. J. Lee, D. Chiba, K. Kobayashi, and
T. Ono, Applied Physics Letters 97, 022508 (2010).
[10] T. An, V. I. Vasyuchka, K. Uchida, A. V. Chumak, K. Yamaguchi, K. Harii,
J. Ohe, M. B. Jungfleisch, Y. Kajiwara, H. Adachi, et al., Nature Materials
12, 549 (2013).
[11] M. Kostylev, Journal of Applied Physics 113, 053907 (2013). 99, 100
[12] B. Liang, B. Yuan, and J.-C. Cheng, Physical Review Letters 103, 104301
(2009). 99
[13] K. Di, H. S. Lim, V. L. Zhang, M. H. Kuok, S. C. Ng, M. G. Cottam, and H.
T. Nguyen, Physical Review Letters 111, 149701 (2013). 102, 108
19, 7013 (1986). 102
[15] K. Y. Guslienko, S. O. Demokritov, B. Hillebrands, and A. N. Slavin, Physical
Review B 66, 132402 (2002). 102
[16] A. Yariv and P. Yeh, Optical Electronics in Modern Communications (Oxford
University Press, 2007). 103, 104, 105
[17] B. Lenk, H. Ulrichs, F. Garbs, and M. Mnzenberg, Physics Reports 507,
107 (2011). 103
[18] M. Madami, S. Bonetti, G. Consolo, S. Tacchi, G. Carlotti, G. Gubbiotti, F.
B. Mancoff, M. A. Yar, and J. Akerman, Nat Nano 6, 635638 (2011). 107
[20] M. F. Yanik and S. Fan, Physical Review Letters 92, 083901 (2004). 110
118
BIBLIOGRAPHY
[21] P. Dong, S. F. Preble, J. T. Robinson, S. Manipatruni, and M. Lipson, Physical Review Letters 100, 033904 (2008). 110
[22] A. D. Karenowska, J. F. Gregg, V. S. Tiberkevich, A. N. Slavin, A. V. Chumak, A. A. Serga, and B. Hillebrands, Physical Review Letters 108, 015505
(2012). 110
[23] A. V. Chumak, V. S. Tiberkevich, A. D. Karenowska, A. A. Serga, J. F.
Gregg, A. N. Slavin, and B. Hillebrands, Nat Commun 1, 141 (2010). 110
[24] J. P. Parekh and H. S. Tuan, Magnetics, IEEE Transactions on 13, 1246
(1977). 110
[25] I. Lisenkov, D. Kalyabin, S. Osokin, J. W. Klos, M. Krawczyk, and S. Nikitov,
Journal of Magnetism and Magnetic Materials 378, 313 (2015).
[26] H. T. Nguyen and M. G. Cottam, Journal of Applied Physics 111, 07D122
(2012). 110
[27] S. N. Piramanayagam, J. P. Wang, X. Shi, S. I. Pang, C. K. Pock, Z. S. Shan,
and T. C. Chong, IEEE Transactions on Magnetics 37, 1438 (2001). 111
[28] S. N. Piramanayagam, K. O. Aung, S. Deng, and R. Sbiaa, Journal of Applied
Physics 105, 07C118 (2009).
[29] M. Ranjbar, S. Piramanayagam, R. Sbiaa, K. Aung, Z. Guo, and T. Chong,
Journal of Nanoscience and Nanotechnology 11, 2611 (2011). 111
[30] W. F. Egelhoff and M. T. Kief, Physical Review B 45, 7795 (1992). 111
[31] S. S. P. Parkin, Applied Physics Letters 60, 512 (1992). 111
[32] M. Mruczkiewicz, M. Krawczyk, G. Gubbiotti, S. Tacchi, Y. A. Filimonov,
D. V. Kalyabin, I. V. Lisenkov, and S. A. Nikitov, New Journal of Physics
15, 113023 (2013). 112
[33] A. M. Merzlikin, M. Levy, A. A. Jalali, and A. P. Vinogradov, Physical Re-
119
BIBLIOGRAPHY
view B 79, 195103 (2009). 112

[34] A. I. Ignatov, A. M. Merzlikin, and M. Levy, Journal of the Optical Society
of America B 28, 1911 (2011). 112
[35] R. Verba, V. Tiberkevich, E. Bankowski, T. Meitzler, G. Melkov, and A.
Slavin, Applied Physics Letters 103, 082407 (2013). 115
[36] J. Nogus and I. K. Schuller, Journal of Magnetism and Magnetic Materials
192, 203 (1999). 115
120
7
Conclusions
This doctoral thesis mainly focuses on the spin-wave dynamics of magnetic
nanostructures.
Improvements were made in the simulation methods for calculating the band
structures of MCs, including a general means for 3D MCs with arbitrary equilibrium magnetizations. These approaches, combined with BLS experiments, were
employed to explore novel properties of various MCs. In particular, MCs with
artificial defects demonstrate new possibilities of band structure engineering and
functionality design.
We have shown that, combining BLS and theory, the physics of SWs gets more
complex and interesting if interfacial DMI is brought into play. The DMI, which
exists in non-centrosymmetric magnetic systems such as asymmetric multilayers,
can give rise to asymmetric spin-wave dynamics. More importantly, DMI plays a
crucial role in the formation of magnetic skyrmions, which because of their unique
properties, have enormous applications potential in areas such as spintronics.
Much investigation has been done on spin-wave nonreciprocity, which is inter-
121
7. CONCLUSIONS
esting in itself and has wide application potential as well. A method based on
synthetic antiferromagnetic coupling has been proposed to enhance asymmetry
in the dispersion relations of MCs. Further, a mechanism to realize nonreciprocal magnonic mode transitions has been established, which is different from
and independent of the well-known nonreciprocity arising from magnetic dipolar
interaction. Besides, the asymmetric properties of SWs due to DMI represent
another way of realizing strong nonreciprocity at the nanoscale.
The research topics covered here could serve as the building blocks for a common aim of magnonics, namely to utilize SWs as efficient information carriers for
computing. Moreover, travelling SWs carry spin angular momentum, thus making them an ideal candidate for pure spin current carriers. Therefore, the results
in this research should also be of use in the field of spintronics.
The following are possible extensions to this doctoral work. First, while we
have mapped the band structures of 1D MCs with artificial defects, the physics of
their dependence on the geometry and structure has not been explored, such as
2D defect MCs (see also Section 4.5). It will also be interesting to design and test
novel magnonic devices based on the novel properties induced by the artificial
defects. Second, there should be new physics to be uncovered in investigating
the effects of interfacial DMI on SWs in complex structures other than uniform
films, such as magnetic dots, nanowires, and MCs. Moreover, it will be exciting
to realize RT-stable magnetic skyrmions based on strong interfacial DMIs. The
theories of magnetic skyrmions, e.g. their dynamic modes, how skyrmions propagate under spin-polarized currents, and how they interact with SWs and external
magnetic fields can therefore be tested.
Considering that magnonics is still a relatively new field compared with pho-
122
tonics and that many novel spin and spin-wave phenomena have been discovered
only recently, there should be plenty of opportunities in this field for further
studies.
123
7. CONCLUSIONS
124
8
List of Publications and
Conferences
[1] Huihui Pan, Vanessa Li Zhang, Kai Di, Meng Hau Kuok, Hock Siah Lim,
Ser Choon Ng, Navab Singh, and Adekunle Olusola Adeyeye. Phononic and
magnonic dispersions of surface waves on a permalloy/BARC nanostructured
array. Nanoscale Research Letters, 8:115, 2013.
[2] K. Di, S. X. Feng, S. N. Piramanayagam, V. L. Zhang, H. S. Lim, S. C.
Ng, and M. H. Kuok. Enhancement of spin-wave nonreciprocity in magnonic
crystals via synthetic antiferromagnetic coupling. Scientific Reports, 5,
2015.
[3] K. Di, H. S. Lim, S. C. Ng, V. L. Zhang, and M. H. Kuok. Defect modes in a
magnonic crystal and their applications in energy-efficient spin-wave devices.
In E-MRS 2014 Spring Meeting, Lille France 2014 (Poster Presentation).
125
[4] K. Di, H. S. Lim, S. N. Piramanayagam, V. L. Zhang, S. C. Ng, and M.

H. Kuok. Micromagnetic simulations and design of magnonic crystals. In
ICMAT 2013, SunTec Singapore 2013 (Oral Presentation).
[5] Kai Di, Hock Siah Lim, Vanessa Li Zhang, Ser Choon Ng, and Meng Hau
Kuok. Spin-wave nonreciprocity based on interband magnonic transitions.
Applied Physics Letters, 103:132401, 2013.
[6] Kai Di, Hock Siah Lim, Vanessa Li Zhang, Ser Choon Ng, Meng Hau Kuok,
Michael G. Cottam, and Hoa T. Nguyen. Comment on Physical origin
and generic control of magnonic band gaps of dipole-exchange spin waves in
width-modulated nanostrip waveguides. Physical Review Letters, 111:
149701, 2013.
[7] Kai Di, Hock Siah Lim, Vanessa Li Zhang, Ser Choon Ng, Meng Hau Kuok,
Hoa T. Nguyen, and Michael G. Cottam. Tuning the band structures of a
one-dimensional width-modulated magnonic crystal by a transverse magnetic
field. Journal of Applied Physics, 115:053904, 2014.
[8] Kai Di, Vanessa Li Zhang, Meng Hau Kuok, Hock Siah Lim, Ser Choon Ng,
Kulothungasagaran Narayanapillai, and Hyunsoo Yang. Band structure of
magnonic crystals with defects: Brillouin spectroscopy and micromagnetic
simulations. Physical Review B, 90:060405(R), 2014. (Rapid Communication).
[9] Kai Di, Vanessa Li Zhang, Hock Siah Lim, Ser Choon Ng, Meng Hau Kuok,
Xuepeng Qiu, and Hyunsoo Yang. Asymmetric spin-wave dispersion due to
Dzyaloshinskii-Moriya interaction in an ultrathin Pt/CoFeB film. Applied
Physics Letters, 106:052403, 2015.
[10] Kai Di, Vanessa Li Zhang, Hock Siah Lim, Ser Choon Ng, Meng Hau Kuok,
126
Jiawei Yu, Jungbum Yoon, Xuepeng Qiu, and Hyunsoo Yang. Direct observation of the Dzyaloshinskii-Moriya Interaction in a Pt/Co/Ni film. Physical
Review Letters, 114:047201, 2015.
[11] Vanessa Li Zhang, Kai Di, Hock Siah Lim, Ser Choon Ng, Meng Hau Kuok,
Jiawei Yu, Jungbum Yoon, Xuepeng Qiu, and Hyunsoo Yang. In-plane
angular dependence of the spin-wave nonreciprocity of an ultrathin film
with Dzyaloshinskii-Moriya interaction. Applied Physics Letters, 107(2):
022402, 2015.
127

Di Kai PHD Thesis Nus 2015

Transféré par

Informations du document

Titre original

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

Di Kai PHD Thesis Nus 2015

Transféré par

Droits d'auteur :

Formats disponibles

Experimental and Theoretical Studies of

Spin-Wave Dynamics in Magnetic

A THESIS SUBMITTED FOR THE DEGREE OF

ciation to Dr. Wong Seng Kai for training me in sample fabrication.

If all you have is a hammer, everything looks like a nail.

LLG Equation & Spin Waves . . . . . . . . . . . . . . . . . . . .

Macroscopic Theory of Magnetization Dynamics .

Effective Field in the LLG Equation . . . . . . .

Spin Waves in Thin Films . . . . . . . . . . . . . . . . . .

Brillouin Light Scattering . . . . . . . . . . . . . . . . . . . . . .

Intensity of Brillouin Light Scattering . . . . . . . . . . . .

Finite-Element Approach Using COMSOL . . . . . . . . . . . . .

Brief Introduction to Finite Element Method . . . . . . . .

Weak Formulation: An Example . . . . . . . . .

Mesh Descritization & Basis Functions . . . . . .

LLG Equation in Weak Form . . . . . . . . . . .

Calculation of Demagnetizing field . . . . . . . . . . . . .

3 Band Structure of a 1D Width-Modulated Nanostripe

The Complete Band Structure . . . . . . . . . . . . . . . . . . . .

Band Structure Modulation by a Transverse Magnetic Field . . .

Tunability of Band Structure Parameters . . . . . . . . . .

Symmetry Assignment and Excitation Efficiency . . . . . .

4 Artificial Defects in Magnonic Crystals

Structures of Defect MCs and Experimental Details . . . . . . . .

Summary & Discussion . . . . . . . . . . . . . . . . . . . . . . . .

5 SWs in Ferromagnetic Films with Interfacial DzyaloshinskiiMoriya Interactions

Dispersion Relation in the Presence of Interfacial DMI . . . . . .

Sample Fabrication & Basic Magnetic Properties

Details of BLS Experiment . . . . . . . . . . . .

Sample Fabrication and Characterization . . . . .

5.3.2.5 Dispersion Relation . . . . . . . . . . . . . . . . . 88

Asymmetric Dispersion Relation . . . . . . . . . . . . . . .

Nonreciprocity by Interband Magnonic Transitions

Coupled-Mode Analysis . . . . . . . . . . . . . . . . . . . 102

Nonreciprocity Enhanced by Antiferromagnetic coupling . . . . . 110

8 List of Publications and Conferences

Character tables of symmetry groups (after Ref. [7]) . . . . . . . .

Ground state of a 1D spin lattice. . . . . . . . . . . . . . . . . . .

Excited state of a 1D spin lattice showing spin waves. . . . . . . .

Band structure of a Co(250nm)/Py(250nm) MC measured by BLS. 11

Brillouin light scattering processes by magnons. . . . . . . . . . .

Schematic of the optical path of Brillouin light scattering.

Triangular mesh grid of an irregular 2D structure. . . . . . . . . .

Illustration of 1D piecewise linear basis functions. . . . . . . . . .

Sinc function approximated by linear combination of triangular

Geometric structure of the MC. . . . . . . . . . . . . . . . . . . .

Band structure and symmetry assignment of the MC. . . . . . . .

Geometric and static properties of the MC. . . . . . . . . . . . . .

Calculation results of magnonic band structure under various transverse field. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Mode frequencies and sizes of band gaps as functions of transverse

Illustration of field-tunability of band gaps. . . . . . . . . . . . . .

Eigenmodes of magnetization with phase information. . . . . . . .

Excitation spectra under different transverse field. . . . . . . . . .

SEM image of the MC with 400nm-wide Py defect stripes. . . . .

Brillouin spectra of the MC with and without defects recorded at

Magnon band structures of the perfect and defect MCs. . . . . . .

Dependence of defect mode frequency on defect Py stripe width. .

Mode profiles of confined and propagating modes. . . . . . . . . .

In-plane and perpendicular magnetic hysteresis loops of the Pt/Co/Ni

Brillouin results of the Pt/Co/Ni and control samples. . . . . . .

Spin-wave linewidth of the Pt/Co/Ni sample. . . . . . . . . . . .

Magnetic hysteresis loops of the Pt/CoFeB sample. . . . . . . . .

Measured spin-wave dispersion relation of the Pt/CoFeB film. . .