You are on page 1of 73

relative to the error necessitates much longer integration

times, leading to much slower adaptation. This problem


can be overcome, and the accuracy improved, by using a
carrier suppression lter before performing the gradient
calculation.
Chen et al. [10] proposed an alternative fast adaptive
algorithm to track the linearizer control parameters, sep-
arating the rapidly varying but known factors (i.e., the
signal statistics) from the slowly varying environmental
factors affecting the power amplier model, resulting in
swift and accurate convergence.
Echeverria et al. [11] demonstrated that very high lev-
els of intermodulation suppression, over 60dB, could be
achieved by tuning the control loops manually for individ-
ual 10 MHz subbands, over a total bandwidth of 19 MHz
(2.012.205 GHz).
4.4. Hybrid Feedforward Ampliers
It is possible to combine feedforward linearization with
other linearization methods for improved overall results.
For example, Horiguchi et al. [12] demonstrated a high-
power 2.12 GHz feedforward power amplier in which the
overall operating efciency was improved by 1% by the
addition of a simple predistortion linearizer to the input of
the main amplier.
Another more radical hybrid feedforward amplier con-
cept has been proposed by Randall et al. [13], in which
part of the rst loop is replaced by a DSP implementation.
The main and reference signals are both generated by
DSP before upconversion to the carrier frequency. The
second loop would operate conventionally. The advantage
of DSP implementation is that phase and amplitude
equalization of both the main and reference signals can
be carried out and adapted as required in DSP software,
rather than in RF analog components, allowing more ac-
curate cancellation.
4.5. MMIC Integration
Integrating the whole feedforward amplier onto a single
MMIC is an attractive proposition from the point of view
of miniaturization and repeatability at higher microwave
frequencies or even millimeter-wave frequencies. Achiev-
ing such integration is, however, a major research chal-
lenge, because of the difculty of creating the necessary
high-isolation couplers, low-loss delay lines, and phase
shifters on a MMIC, along with high linearity, high ef-
ciency, and accurately modeled ampliers. Parkinson and
Paul [14] have carried out initial studies of the possibility
of using a distributed amplier structure as an active cou-
pler within a MMIC feedforward amplier, as a starting
point for full MMIC integration.
BIBLIOGRAPHY
1. H. S. Black, Translating System, U.S. Patent 1,686,792 (1928).
2. H. Seidel, A microwave feedforward experiment, Bell Syst.
Tech. J. 50:28792916 (1971).
3. V. Steel, D. Scott, and S. Ludvik, A 618GHz high dynamic
range MMICamplier using a feedforward technique, IEEE
MTT-S Microwave Symp. Digest, 1990, Vol. 2, pp. 911914.
4. K. Konstantinou and D. K. Paul, Analysis and design of
broadband, high efciency feedforward ampliers, IEEE
MTT-S Microwave Symp. Digest, 1996, Vol. 2, pp. 867870.
5. F. H. Raab, P. Asbeck, S. Cripps, P. B. Kenington, Z. B. Popo-
vic, N. Pothecary, J. Sevic, and N. O. Sokal, Power ampliers
and transmitters for RFand microwave, IEEE Trans. Micro-
wave Theory Tech. MTT-50(3):814826 (2002).
6. N. Pothecary, Feedforward Linear Power Ampliers, Artech
House, Norwood, MA, 1999.
7. K. Konstantinou, P. Gardner, and D. K. Paul, Optimisation
method for feedforward linearisation of power ampliers,
Electron. Lett. 29(18):16331635 (1993).
8. Y. K. Hau, V. Postoyalko, and J. R. Richardson, Design and
characterization of a microwave feed-forward amplier with
improved wide-band distortion cancellation, IEEE Microwave
Theory Tech. MTT-49(1):200203 (2001).
9. J. K. Cavers, Adaptation behavior of a feedforward amplier
linearizer, IEEE Trans. Veh. Technol. 44(1):3140 (1995).
10. J. Chen, H. Tsai, and Y. Chen, Fast adaptive wide-band power
amplier feed-forward linearizer, IEEE Vehicular Technology
Conf., 1998, pp. 21672171.
11. A. Echeverr ia, L. Fan, S. Kanamaluru, and K. Chang, Fre-
quency tunable feedforward amplier for PCS applications,
Microwave Opt. Technol. Lett. 23(4):218221 (1999).
12. K. Horiguchi, M. Nakayama, Y. Sakai, K. Totani, H. Senda,
Y. Ikeda, and O. Ishida, A high efciency feedforward ampli-
er with a series diode linearizer for cellular base stations,
IEEE MTT-S Microwave Symp. Digest, 2001, pp. 797800.
13. R. G. Randall, J. G. McRory, and R. H. Johnston, Broadband
DSP based feedforward amplier lineariser, Electron. Lett.
38(23):14701471 (2002).
14. G. Parkinson and D. K. Paul, Non-reciprocal couplers using
distributed amplier topology, Microwave Opt. Technol. Lett.
38(5):366368 (2003).
FERRITE CIRCULATORS
E. K. N. YUNG
D. X. WANG
City University of Hong Kong
Kowloon, Hong Kong, P.R. China
R. S. CHEN
Nanjing University of Science
and Technology
Nanjing, P.R. China
1. INTRODUCTION
The circulator is one of the elementary building blocks in
radiofrequency and microwave circuits. It is used exten-
sively in making basic devices for communications and
radar systems. The latter are often used in broadband and
high-power rating systems at microwave and millimeter-
wave frequencies. It is, however, seldom found in consum-
er products because some of its functions can be performed
1448 FERRITE CIRCULATORS
Previous Page
by simpler and cheaper alternatives. However, interest in
circulators has been revitalized with the emergence of
mobile communications. As an increasing number of users
are accustomed to the freedom provided by cellular
phones, communication without the constraint of an at-
tached wire is considered as natural as breathing air. In
addressing the ever-increasing public demands, numerous
wireless systems have been launched and more are com-
ing. The most notable ones are wireless local area net-
work, wireless local subscriber loops, and other high-
bit-rate yet low-error-rate systems of data transmission.
For these systems, the overlooked circulator enjoys a com-
petitive edge as most active switches could not deliver the
needed power rating and bandwidth at a reasonable price,
especially those in the ultra-high-frequency band and
beyond.
For engineering students, the circulator is perhaps
the rst multiport device covered in their foundation
course on microwave engineering, and it might be
the rst nonreciprocal passive device encountered [1,2].
The importance of this three-port device in communica-
tion systems can be explained by its functionality. Simply
put, waves entering one of the identical ports of a circu-
lator, say, port 1, are totally transmitted to one of the
output ports, port 2, with none to the remaining one,
port 3. Similarly, those inputted to port 2 are passed on
to port 3 without loss while port 1 is isolated. The cycle
is completed as signals entering port 3 are outputted
to port 1 only. Schematically, a circulator is shown in
Fig. 1.
The scattering matrix of the clockwise circulator shown
in Fig. 1 is
S
clockwise

0 0 1
1 0 0
0 1 0
_

_
_

_
1
Depending on the physical layout of a circuit and ones
point of view, a circulator can also be counter-clockwise
and the relevant scattering matrix is
S
counterclockwise

0 1 0
0 0 1
1 0 0
_

_
_

_
2
Extension of a three-port circulator to an m-port one is
possible, but it is skipped here because it introduces no
new concepts, yet the algebra involved is so tedious that
the logic ow of this introductory article could be derailed.
Note that, the scattering matrices given in (1) and (2) are
valid for lossless three-port circulators with perfectly
matched inputs and outputs. Modications of the theories
and practices presented in this article to low-loss m-port
ones with less than perfect matched input and output
ports have been conducted in many pioneering studies
[3,4]. Again, they are omitted here, and interested readers
are referred to a comprehensive book [5].
The passive trafc control in circulators is facilitated by
an anisotropic element. Isotropy or anisotropy is an in-
trinsic characteristic of all matters that nds its origin in
their molecular structures and atomic dipole moments of
electric or magnetic nature [6]. In general, these dipole
moments are randomly oriented in the absence of an ex-
ternal excitation; therefore the material on the whole has
a negligibly small net dipole moment. The scenario could
change abruptly with the introduction of an applied static
electric eld as the electric dipole moments align them-
selves with the impressed force and a net electric dipole
moment results. Similarly, a spontaneous magnetic dipole
moment is obtained by a biasing magnetic eld. The dipole
alignment is dependent on the strength of the excitation
until all dipoles are almost perfectly oriented. The align-
ment, in fact, depends on the molecular structure of the
material, the ambient temperature, the initial settings,
and other physical conditions. In general, the relationship
between the dipole alignment and the applied eld is non-
linear, and for some materials, it is anisotropic.
The principles of circulator operation, basic theories of
anisotropy, wave propagation in anisotropic materials,
and circulator applications will be briefly presented in
Sections 24, followed by a general description of selected
circulators in Section 5. The most popular ones, especially
waveguide junction circulators, will be analyzed in depth
in Section 6. A few concluding remarks will be presented
in Section 7.
2. ANISOTROPY AND ANISOTROPIC MATERIALS
In theory, both electrical and magnetic materials can be
used to produce anisotropic effects in a circulator. Mag-
netic anisotropy is, however, still used in practically all
circulators over a half-century since its invention. Hence,
the discussion in this section focuses on magnetic anisot-
ropy. Nevertheless, the potential of using electrically an-
isotropic materials in circulators in the future cannot be
dismissed. One of the major candidates for this is the uni-
axial chiral material especially made for microwave fre-
quencies. Although significant progress has been observed
2
1
3
Figure 1. A schematic sketch of a circulator.
FERRITE CIRCULATORS 1449
in this area in recent years, the technologies at our dis-
posal at this moment are far from satisfactory. Thus it has
been decided not to further investigate the use of electri-
cally anisotropic materials in circulators until more ma-
ture technologies are available.
As this is only an introductory article on circulators, we
will not attempt to cover all circulators available on the
market. For similar reasons, only theories of anisotropy
directly relevant to the selected circulators discussed here
will be presented.
2.1. Spinning Electron in Free Space
The angular momentum of an electron in an atom or an
ion arises from two factors: its orbital motion around the
atomic nucleus and its spin. Based on quantum mechan-
ics, the total angular momentum of an electron in free
space is given by
s
1
g
h
2p
3
where h6.626 10
34
J/s is Plancks constant and g is
Landes factor. If the angular momentum of an electron is
due solely to its orbital motion, then g 1. At the other
extreme, a g 2 factor is used to signify a spin-only mo-
mentum. For most materials of interest in microwave sys-
tems, it can be shown that g ranges from 1.98 to 2.01;
therefore g 2 is a good approximation. Based on Bohrs
model, the magnetic dipole moment of an electron is
m
e
m
0
h
4p
9:27 10
24
A=m
2
4
where e 1.6 10
19
C is the charge of an electron and
m
0
9.1 10
31
kg is its mass.
Comparing (3) and (4), it is found that the ratio of the
magnetic dipole moment of a spinning electron to its an-
gular momentum is a constant, the so-called called gyro-
magnetic ratio:
g
m
s

e
m
0
1:75910
11
C=kg 5
According to Hunds rule, electrons in a shell would spread
out over the available states with spins in the same direc-
tion until the shell is half-lled, and all subsequent addi-
tions would have spins in the opposite direction. Thus, the
magnetic dipole moment of an atom with a fully lled out-
ermost shell such as found in the inert gases is zero. At the
other extreme, the maximum dipole moment is obtained in
a half-lled shell, and in the 10-state d shell it is 5.
Returning to the familiar Newtonian mechanics, the
magnetostatic torque acting on the dipole moment due to
an impressed static magnetic eld H
DC
is
Tm
0
mH
DC
6
As electrons are negatively charged, its angular momen-
tum is opposite to its dipole moment:
m gs 7
Since torque is equal to the rate of change of angular mo-
mentum, one arrives at
ds
dt
T m
0
gs H
DC
8
The equation of motion for a spinning electron is obtained
by substituting (7) into (8)
dm
dt
m
0
gmH
DC
9
Without loss of generality, the study can be furthered with
a ^ zz-biased H
DC
. Thus, the components of the vector equa-
tion above are
dm
x
dt
m
0
gm
y
H
DC
10
dm
y
dt
m
0
gm
x
H
DC
11
dm
z
dt
0 12
Differentiating (10) and (11) once more, two harmonic
equations are derived
d
2
m
x
dt
2
o
2
0
m
x
0 13
d
2
m
x
dt
2
o
2
0
m
y
0 14
where o
0
is the Larmor frequency of precession, given by
o
0
m
0
gH
DC
15
A solution to both (13) and (14) is
m
x
A cos o
0
t 16
m
x
A sin o
0
t 17
It is observed from (4) that the magnitude of m is xed;
therefore one gets
m
2

gh
4p
_ _
2
m
2
x
m
2
y
m
2
z
A
2
m
2
z
18
In short, m would precess around the ^ zz axis with an angle
y:
sin y
A
m
19
Moreover, the projection of mon the xy plane would trace
a circular path with an angular frequency o
0
as depicted
in Fig. 2. For a standalone atom in free space, m could
process indefinitely. However, for an atom in a crystal, the
electron would experience numerous gravitational and
1450 FERRITE CIRCULATORS
electromagnetic forces. Altogether, these interactions
would constitute a damping force that causes m to spiral
inward, and in due course, m would align with the biasing
eld.
Even after the dipole alignment is completed, the mag-
netic ux density due to magnetization for most materials
is very weak, compared with the biasing magnetic eld.
Hence, the relative permeability in paramagnetic materi-
als is slightly greater than 1, while that of diamagnetic
materials is marginally less than unity.
Here, it is worthwhile to comment on the units used in
this article. Traditionally, studies of magnetism are con-
ducted in CGS units with the magnetic eld strength ex-
pressed in oersteds and the magnetization in gauss. The
units are chosen such that the magnetic ux density B and
the magnetic eld strength H in free space would have
equal numeric values. On the other hand, in line most en-
gineering texts, formulas presented in this article will be
derived in MKS units. Hence, utmost care must be taken
in using some properties of magnetic materials quoted in
CGS units. For easy reference, some of the most frequently
used unit conversions are tabulated in Table 1.
2.1.1. Ferromagnetic and Ferrimagnetic Materials. In
general, magnetization effects are very weak even for at-
oms with a half-lled outermost shell such as chromium
and manganese. A different mechanism is needed to en-
hance the magnetization. It is found in some substances,
called ferromagnetic materials. From a microscopic point
of view, ferrous metals are composed of tiny domains of
linear dimensions of a few micrometers. It is proved in
quantum mechanics that strong coupling forces exist be-
tween atoms in each domain such that all atomic dipole
moments are kept in parallel even when there is no ex-
ternal eld. As the dipole moments of these domains are
randomly oriented, a demagnetized ferromagnetic mate-
rial shows no magnetization effect, from a macroscopic
perspective. Analogous to the ionic dipole moments in
paramagnetic and diamagnetic materials, the magnetic
dipole of every domain in ferrous metals reacts to an im-
pressed eld, except that the responses are much stronger.
Consequently, the permeability of ferromagnetic materials
is notably higher than its paramagnetic and diamagnetic
counterparts, a factor of many orders. Again, the magne-
tization depends on the eld strength until saturation.
Probably as a result of some form of domain deformation
during the magnetization processes, certain domain rota-
tions are not reversible after the external bias is removed,
and a residue magnetic effect results. This remanent mag-
netization plays a key role in subsequent magnetizations
and contributes to the infamous hysteresis loop.
Significant magnetization can also be observed in the
so-called antiferromagnetic materials. Even though the
dipole moments in the latter class of materials align with
the applied eld equally fervently, their directions are
either in line with or opposite to that of the excitation, and
the end result is a complete cancellation of magnetization
effects.
Another class of substances, termed ferrimagnetic ma-
terials, exhibits a behavior intermediate between ferro-
magnetism and antiferromagnetism. The pattern of dipole
alignment in these materials is similar to that of antifer-
romagnetic ones except the number of magnetic dipoles in
opposite polarities is different and their magnitudes are
not equal. Hence, a partial cancellation of dipole moments
is observed. As expected, magnetization of ferrimagnetic
materials is considerably lower than that of ferrous met-
als. The maximum magnetic ux density in ferrimagnetic
materials is seldom greater than 0.3Wb/m
2
, approximate-
ly one-tenth that of ferromagnetic ones. However, a mag-
netization of this magnitude is significantly higher than
that of nonmagnetic matters; therefore they are often used
in picking up spurious waves in electromagnetic compat-
ibility and electromagnetic interference control.
Ferrimagnetic materials can be found in nature in
a variety of forms such as lodestones. They were rst
mentioned in history more than 4500 years ago when nat-
ural magnets played a decisive role in a major battle
in China as a component in compasses. A nonnatural
ferrimagnetic material was rst produced by Hilpert in
1909 [7], and practical ones were developed shortly after
Spinning
electron
H
dc
S
m
z

0
Figure 2. Angular momentum and magnetic dipole moment of a
spinning electron.
Table 1. Conversion between CGS and MKS Units
Symbol MKS Units CGS Units
Magnetic ux c 1 weber 10
8
maxwell
Magnetic ux density B 1 tesla1 weber/meter
2
10
4
gauss
Magnetic eld intensity H 1 ampere-turn/meter 4p 10
3
oersted
Magnetization M 1 weber/meter
2
2500/p gauss
FERRITE CIRCULATORS 1451
World War II [8]. Synthetic ferrimagnetic materials are
usually polycrystalline. For some applications, large crys-
tals can be made, but they are seldom used as they are
very expensive.
Humanmade (synthetic) ferrimagnetic materials are
now called ferrites for easy reference. Over the years, nu-
merous ferrites have been synthesized for assorted appli-
cations at different frequencies. Based on their crystal
structures, these ferrites can be grouped into three cate-
gories, namely, spinels, garnets, and hexagonal ferrites.
Mineral spinels have a molecular formula of
(MgAl
2
O
4
)
8
. Over the years, many synthetic spinels have
been developed and they share a structure similar to that
of (MOFe
2
O
3
)
8
, where M represents divalent metal, in-
cluding cobalt, aluminum, cadmium, copper, lithium, iron,
magnesium, manganese, nickel, titanium, zinc, and their
combinations. As metallic ions are smaller than oxygen
ions, the crystal structure of spinels depends solely on the
arrangement of oxygen ions, the face-centered cubic lat-
tice [10] as shown in Fig. 3. Also shown in Fig. 3 are the
sites where metallic ions can be found. It can be shown
that a metallic ion in site A has four oxygen neighbors and
thus is called tetrahedral site; site B is termed octahedral
because it has six adjacent oxygen ions. In a unit crystal of
56 ions, there are only 8 tetrahedral and 16 octahedral
metallic ions; therefore not all sites in the crystal lattices
of 32 oxygen ions are lled. It is also found that the mag-
netic moments in tetrahedral and octahedral sites cancel
one another; the remaining 8 octahedral ones give the
ferrite crystal its net magnetic moment [11].
Common garnet or andradite has a molecular structure
of Ca
3
Fe
2
Si
3
O
12
. The frequently used yttrium iron garnet
(YIG) is obtained by replacing calcium by yttrium and sil-
icone by iron. The net magnetic moment can be adjusted
by substituting some iron ions by aluminum ones (YAG),
or yttrium by a rare earth such as gadolinium, holmium,
or dysprosium. For synthetic garnets, a unit crystal of 40
ions has a molecular formula of (Fe
2
O
3
)
5
(M
2
O
3
)
3
; where
M stands for rare earth. There exist three types of sites
in the crystal structure for housing the metallic ions and
all sites are lled.
Ferrites can be mass-produced as if they were ceramics,
except the demands on purity of raw materials, uniformity
in grain sizes, correctness in mix proportion, homogeneity
of the suspension, precision in casting, timing and tem-
perature control in ring, and other manufacturing pro-
cesses are very stringent and the error margin is very
slim. The raw materials include oxides, carbonates,
nitrates, oxalates, and some metallic compounds. As the
facilities used in making ferrites are very expensive and
the exact procedures are usually company secrets, few
laboratories can make them from scratch. It is, however,
recommended that research centers be equipped with di-
amond wheels and other apparatuses for cutting and
grinding raw ferrites into the desired shapes because the
physical dimensions of ferrites have direct consequences
on the performance of circulators, as are the smoothness
and the cleanliness of their surfaces.
As ferrites are ceramiclike compounds, they are very
hard, brittle, and low in thermal conductivity. Unlike fer-
romagnetic materials, ferrites have low electrical conduc-
tivities; thus, they are frequently used in electrical and
electronic devices for a variety of purposes. However, the
feature that makes ferrites indispensable in microwave
circuit designs is anisotropy, not low conductivity. Of
course, an in-depth study of anisotropy is beyond the
scope of this article, but a brief outline of its origin is
deemed appropriate for a better understanding the nature
of wave circulations in magnetized ferrites.
2.1.2. Magnetization of Ferrites. The conguration of
interest is a ferrite crystal under an internal magnetic
eld H
int
. The excitation exterior to a ferrite specimen is
different from H
int
, and it will be treated later. For a sam-
ple with N unbalanced electron spins per unit volume, the
total magnetization is
M
DC
Nm 20
Similar to (9), the equation of motion is
dM
DC
dt
m
0
gM
DC
H
int
21
As the damping force inside a ferrite crystal is much
stronger than that acting on a standalone atom in free
space, the magnetic dipole moments will align with the
magnetic excitation readily. With an increase in H
int
, an
increasing number of magnetic dipoles line up with the
biasing eld, and eventually the alignment process is com-
pleted and the magnetization is saturated as depicted in
Fig. 4. The saturation magnetization M
s
is a physical
property of ferrite, and typically it ranges within 4pM
s

3005000 G. Unless mentioned otherwise, ferrites operate


in the state of saturation magnetization because they are
very lossy at lower levels of magnetization. Hence, the
equation of motion in (21) could be rewritten as
dM
s
dt
m
0
gM
s
H
int
22
Regardless of whether the magnetization is saturated,
ferrite will be permanently magnetized after the magne-
tization process. The crystal structure of having dipole
moments of opposite polarities in an alternating fashion
may constitute a cementing force that keeps the align-
ment intact after the external bias is removed. In some
B
A
Figure 3. Tetrahedral (A) and octahedral (B) sites in the crystal
structure of oxygen ions.
1452 FERRITE CIRCULATORS
circulators, a premagnetized ferrite is adequate and ex-
ternal magnetization is not needed [32].
It is well known that atoms can pick up thermal energy
from the surrounding environment and vibrate according-
ly. Sometimes, the atomic vibration in ferrites could make
the dipole alignment very difcult even at room tempera-
ture. As a rule of thumb, an increase in ambience temper-
ature is accompanied by a decrease in saturation
magnetization. At a sufciently high temperature, called
the Curie temperature T
c
, the vibrational energy could
overwhelm the electromagnetic energy, and the process of
dipole alignment would break down as illustrated in Fig. 5.
3. WAVE PROPAGATION IN FERRITE
Our attention now turns to wave propagations in magne-
tized ferrites. Without loss of generality, a ferrite specimen
is magnetized in the ^ zz direction and is illuminated by an
incident plane wave of arbitrary orientation and of ampli-
tude H
RF
. The total magnetic eld and magnetization are
HH
int
H
RF
H
x
^ xx H
y
^ yy H
int
H
z
^ zz 23
MM
s
M
RF
M
x
^ xx M
y
^ yy M
s
M
z
^ zz 24
The corresponding equation of motion is
dM
dt

dM
s
dt

dM
RF
dt

dM
RF
dt
m
0
gMH:
25
In its components, the equation of motion is
dM
x
dt
m
0
gM
y
H
int
H
z

m
0
gH
y
M
s
M
z

26
dM
y
dt
m
0
gM
x
H
int
H
z

m
0
gH
x
M
s
M
z

27
dM
z
dt
m
0
gM
x
H
y
m
0
gM
y
H
x
28
Since H
RF
5H
int
, M
x
H
y
and similar products can be
ignored in accordance with the small signal approxima-
tions. As a result, the preceding equations can be simpli-
ed to
dM
x
dt
m
0
gM
y
H
int
m
0
gM
s
H
y
o
0
M
y
o
m
H
y
29
dM
y
dt
m
0
gM
x
H
int
m
0
gM
s
H
x
o
0
M
x
o
m
H
x
30
dM
z
dt
0 31
where o
0
is the Larmor frequency of precession, and
o
m
m
0
gM
s
32
Comparing (15) and (32), o
m
can be expressed as
o
m
o
0
M
s
H
int
33
Differentiating (29) and (30) once more, one gets
d
2
M
x
dt
2
o
2
0
M
x
o
0
o
m
H
x
o
m
dH
y
dt
34
d
2
M
y
dt
2
o
2
0
M
y
o
0
o
m
H
y
o
m
dH
x
dt
35
In a time-harmonic analysis, these equations can be sim-
plied to
o
2
0
o
2
M
x
o
0
o
m
H
x
joo
m
H
y
36
o
2
0
o
2
M
y
joo
m
H
x
o
0
o
m
H
y
37
0 H
int
M
s
M
d
c
Figure 4. Magnetization of ferrites versus magnetic eld inten-
sity.
T
c
M
s
0
Figure 5. Magnetization of ferrites versus ambient temperature.
FERRITE CIRCULATORS 1453
Note that both M
x
and M
y
are dependent on H
x
as well as
H
y
. In other words, a tensor is needed to characterize the
relationship between magnetization M and its driving
force, the magnetic eld intensity H; that is
Mw H
w
xx
w
xy
0
w
yx
w
yy
0
0 0 0
_

_
_

_
H 38
The nonzero components of the susceptibility tensor w are
w
xx
w
yy

o
0
o
m
o
2
0
o
2
39
w
xy
w
yx

joo
m
o
2
0
o
2
40
Consequently, the magnetic ux density B must be mod-
ied to
Bm
0
HM m H: 41
The z-bias Polder permeability tensor m is given by [13]
m m
0
I w
_ _

m jk 0
jk m 0
0 0 m
0
_

_
_

_
42
where I is an identity tensor.
The components of the permeability tensor are
m m
0
1 w
xx
m
0
1
oo
m
o
2
0
o
2
m
_ _
43
k jm
0
w
xy
m
0
o
0
o
m
o
2
0
o
2
m
44
It is observed that an ^ xx-directed or ^ yy-directed Hwould give
rise to both ^ xx and ^ yy components of B with a quadratic
phase shift between them. A material having this type of
properties is called gyrotropic.
If the biasing magnetic eld is ^ xx-directed, the corre-
sponding gyrotropic permeability tensor is
m
m
0
0 0
0 m jk
0 jk m
_

_
_

_
45
3.1. Forced Precession
If the incoming wave in the preceding section is a right-
handed circularly polarized plane wave propagating in the
z direction, H
x
jH
y
H

0
, the relevant magnetization
can be determined by (36) and (37):
M
x

o
m
H

0
o
2
0
o
2
o
0
o
o
m
H

0
o
0
o
46
M
y

jo
m
H

0
o
2
0
o
2
oo
0

jo
m
H

0
o
0
o
jM
x
47
Hence, the magnetization vector is also circularly polar-
ized and rotates in the same direction with an angular
velocity o. It then follows that the magnetic ux density is
circulating synchronously with the driving force and the
effective permeability is given by
m

0
m
0
1
o
m
o
0
o
_ _
48
It is interesting to compare the angle of precession of the
magnetic eld intensity y
H
and that of the magnetization
vector y

M
:
tany
H

0
H
int
H
z

0
H
int
49
tan y
M

M
x
M
s
M
z

o
m
o
0
o
H

0
M
s

o
0
o
0
o
H

0
H
int
50
By invoking (32), we can rewrite (50) as
tan y

M

o
0
o
0
o
H

0
H
int
51
Hence, as long as oo2o
0
, then y

M
> y
H
as shown in Fig. 6.
Since H

0
5H
int
, the angle of precession is usually very
small except when oo
0
, the Larmor frequency. As indi-
cated in (29) and (30), the magnetization could be innitely
large at oo
0
; therefore o
0
is also known as ferrimagnetic
resonance. We will revisit ferrimagnetic resonance later
because circulators usually operate in its vicinity.
If the driving force is a ^ zz-directed left-handed circularly
polarized wave, the magnetization and the magnetic ux
H
int
H
0
+
H
M

+
z

Spinning
electron
0
H
0
M
Figure 6. Force precession with y

M
> y
H
.
1454 FERRITE CIRCULATORS
density are also similarly polarized, and the correspond-
ing effective permeability and angles of precession are
m

0
m
0
1
o
m
o
0
o
_ _
52
tan y

o
0
o
0
o
H

0
H
int
53
Here, y

M
oy
H
and the magnetic dipole moment rotates
in a direction opposite to its free precession as indicated in
Fig. 7. In short, the external excitation sets up a prefer-
ential pattern of precession and leads to nonreciprocal
characteristics in wave propagation.
3.2. Damping
As mentioned previously, circulators are usually magne-
tized such that they operate in the vicinity of ferrimag-
netic resonance. Based on (47), both M
x
and M
y
could be
innitely large with y
M
901. To stabilize the magnetiza-
tion vector at resonance, a damping term must be added
into the equation of motion in (25). Landau and Lifshitz
(as cited by Helszajn [12]) have shown that the damping
force pulling the magnetization vector toward the driving
force H
int
is
F
damping
/ M
dM
dt
54
The damping force acting on the precession is illustrated
in Fig. 8. The equation of motion with the damping force
included is
dM
dt
m
0
gMH
am
0
jMj
M
dM
dt
55
where a is a dimensionless constant.
Using small argument approximations and following
procedures presented in deriving (39) and (40), the com-
ponents in the susceptibility tensor can be derived as
w
xx
w
yy

o
m
o
0
jao
o
0
jao
2
o
2
56
w
xy
w
yx

joo
m
o
0
jo
2
o
2
57
It appears that the susceptibility with damping factors
can be obtained by replacing o
0
in the loss-free formulas
with o
0
jao. Watch out, the conventional way in deriving
damping effects is to replace o by ojao! Usually, the
susceptibility tensor is expressed in its real and imaginary
parts; that is
w
xx
w
0
xx
jw
00
xx
58
w
xy
w
0
xy
jw
00
xy
59
where
w
0
xx

o
0
o
m
o
2
0
o
2
a
2
o
0
o
m
o
2
o
2
0
o
2
1 a
2

2
4a
2
o
2
0
o
2
60
w
00
xx

ao
m
oo
2
0
o
2
1a
2

2
o
2
0
o
2
1 a
2

2
4a
2
o
2
0
o
2
61
w
0
xy

o
m
oo
2
0
o
2
1a
2

2
o
2
0
o
2
1 a
2

2
4a
2
o
2
0
o
2
62
w
00
xx

2ao
0
o
m
o
2
o
2
0
o
2
1 a
2

2
4a
2
o
2
0
o
2
63
H
int
H
0

M
z

Spinning
electron

H
Figure 7. Forced precession with y

M
oy
H
.
Spinning
electron
H
0

M

M x

dm
dt

dm
dt

0
M
Figure 8. Effect of damping force on the precession.
FERRITE CIRCULATORS 1455
Note that all components reach a common peak at
o
0;max
o

1 a
2
_
% o 1
a
2
2
_ _
64
Since o
0
is given in terms of the impressed driving force
H
int
, the magnetic eld needed to attain a ferrimagnetic
resonance at a given frequency is
H
fr

o
0;max
m
0
g
%
o
m
0
g
1
a
2
2
_ _
65
The susceptibility component of particular interest is w
00
xx
,
whose maximum is
w
00
xx;max

o
m
2ao
66
Variation of w
00
xx
with the impressed magnetic eld is
sketched in Fig. 9, which shows the so-called linewidth
of ferrites. Linewidth is the parameter provided by ferrite
manufacturers to describe magnetic losses at resonance. It
is dened as the difference between the magnetic elds
where w
00
xx
is one-half of w
00
xx;max
for a given frequency.
Assuming that w
00
xx
is changing quadratically near res-
onance, the linewidth can be derived as
DH %
2ao
m
0
g
67
In general, circulators are operated in the region either
above resonance H
int
4H
fr
or below resonance H
int
oH
fr
as
indicated in Fig. 9.
3.3. Demagnetization
As mentioned previously, the biasing eld exterior to a
ferrite sample is different from the interior one. This re-
lationship is dependent on the shape of the specimen as
well as the orientation of the external excitation. For ex-
ample, if the impressed eld B
ext
is normal to the surface
of a ferrite slab or innite extent, the continuity of the
normal component of magnetic ux density dictates that
B
ext
m
0
H
ext
m
0
H
int
M
s
68
Hence
H
int
H
ext
M
s
69
However, if the applied eld is parallel to the slabs sur-
face, the continuity of the tangential magnetic eld inten-
sity demands that
H
ext
H
int
70
In general, the magnetic eld intensity in ferrite can be
determined in terms of a demagnetization factor x:
H
int
H
ext
xM
s
71
The argument is valid for all directions; therefore there
are three x values, one for each coordinate. Moreover, the
demagnetization factors are dependent on the shape of the
ferrite sample as well as the direction of the external eld,
relative to the specimen. It can be proved that the sum of
three demagnetization factors in Cartesian coordinates is
equal to 1:
x
x
x
y
x
z
1 72
For a circular ferrite disk of negligible thickness, the de-
magnetization factors can be determined as if it were a slab
such as that just mentioned, that is, x
z
1 and x
x
x
y
0.
On the other hand, if a ferrite post of innite length is
magnetized axially, the demagnetization factor in the z
direction is zero, and the x values on the xy plane can be
determined by twofold symmetry; therefore x
x
x
y

1
2
.
Similarly, on the basis of threefold symmetry, the demag-
netization factors for a sphere excited in any coordinate
are x
x
x
y
x
z

1
3
.
3.4. Propagation Transverse to Magnetization
In the study of forced precessions, a plane wave propa-
gates in the direction of the static magnetic excitation.
Preferential treatment of some waves and ferrimagnetic
resonance are established. However, wave propagations in
most circulators are transverse to the biasing magnetic
eld. To illustrate the latter phenomenon, the simplest
conguration is used. It is a rectangular ferrite slab of in-
nite width and depth but of negligible thickness. For
consistency with the notations used in waveguides and
microstrip lines, the external magnetic eld is ^ xx-directed
and the permeability tensor given in (45) is applicable.
The ferrite slab is illuminated by an incident plane
wave propagating in the ^ zz direction:
E ^ xxE
x
^ yyE
y
e
jb
0
z
73
H ^ xxH
x
^ yyH
y
e
jb
0
z
74
Above
Resonance
Below
Resonance
H
1
H
2
H
int
H
fr
H / 2
X"
max
1/2X"
max
Figure 9. Variation of w
00
xx
with H
int
.
1456 FERRITE CIRCULATORS
In a gyrotropic medium, the Maxwell equations are
rE jomH 75
rHjoeE 76
r
.
D0 77
r
.
B0 78
It is found that (75) can be satised with a nonzero E
z
or a
nonzero H
z
, but not both. Hence, a more general form for
waves in ferrite is selected:
E ^ xxE
x
^ yyE
y
^ zzE
z
e
jbz
79
H ^ xxH
x
^ yyH
y
^ zzH
z
e
jbz
80
According to this postulation, @/@z jb; by symmetry,
@/@y 0; and by the negligible thin approximation, @/@x
0. Thus, the constituents of the curl equations in (58) can
be reduced to
jbE
y
jom
0
H
x
81
jbE
x
jomH
y
jkH
z
82
0 jojkH
y
mH
z
83
Similarly, the curl equation in (76) can be simplied to
jbH
y
joeE
x
84
jbH
x
joeE
y
85
0 joeE
z
86
Based on (86), E
z
is zero; therefore H
z
must be nonzero.
Moreover, from both (84) and (85), the intrinsic impedance
of the medium is found:
Z
E
x
H
y

E
y
H
x

b
oe
87
Substituting (87) into (81), the propagation constant is
derived:
b
2
o
2
m
0
e 88
It is also observed in (83) that H
z
can be evaluated in
terms of H
y
:
H
z

jk
m
H
y
89
This formula can be used to eliminate H
z
from (82):
bE
x
o m jk
jk
m
_ _
H
y

o
m
m
2
k
2
H
y
90
Combining (87) and (90), another solution for the propa-
gation constant is obtained
b
2
e

o
2
e
m
m
2
k
2
o
2
m
e
e 91
where b is denoted by b
e
to distinguish it from the other
solution given in, and m
e
is an effective permeability of the
medium, given by
m
e

m
2
k
2
m
92
If the electric eld of the incident wave is ^ yy-directed, the
continuity of the tangential electric eld intensity at the
air-ferrite interface implies that E
x
0; so that H
y
is also
zero, and by (89) H
z
also vanishes. Hence, the only solution
for b is (88) and the wave is called ordinary, which means
that wave propagation in ferrite is not affected by the
magnetization. However, if the electric vector is ^ xx-direct-
ed, the only viable b is that given in (91). With a nonzero
H
y
, H
z
is also nonzero, and the resultant wave is called
extraordinary. Thus, wave propagation in ferrite is po-
larization-dependent. On the other hand, if the incident
eld has both ^ xx and ^ yy components, it is treated as a com-
bination of two linearly independent entities, and each
propagates independently with a different phase velocity.
As a result, two images are created, the so-called birefrin-
gence effect. It is also noticed that k is frequency-depen-
dent; therefore the effective permeability for a given H
int
and M
s
could become negative if the frequency is high
enough, and an incident wave will be totally reected from
ferrite. This scenario can be viewed from a different angle
in which the frequency is xed and evanescent effects can
be obtained by changing the external magnetic eld
intensity.
3.5. Wave Circulation around a Ferrite Cylinder of
Innite Length
The conguration of interest in this subsection is a ferrite
cylinder of radius a and of innite length. It is placed on
the ^ zz axis and magnetized axially; therefore the perme-
ability tensor given in (42) is appropriate. Wave propaga-
tion in ferrite can be determined by treating the cylinder
as a dielectric resonator. It is assumed that the cylindrical
surface would behave as if it were a magnetic wall; that is,
E
j
0 at ra. So are the surfaces on top of and at the
bottom of the cylinder; therefore E
r
0 for z 7N. Since
the cylinder is innitely long, @/@z 0; consequently E
r
0
everywhere. Using standard coordinate transformations,
the nonzero components of the magnetic ux density can
FERRITE CIRCULATORS 1457
be determined by (41):
B
r
B
x
cos jB
y
sin j
mH
x
jkH
y
cos j
jkH
x
mH
y
sin j
mH
r
jkH
j
93
B
j
B
x
sinjB
y
cos j
mH
x
jkH
y
sin j
jkH
x
mH
y
cos j
jkH
r
mH
j
94
The components of the curl equation in (75) in cylindrical
coordinates can be expressed as
1
r
@E
z
@j
jomH
r
jkH
j

jomH
r
okH
j
95

@E
z
@r
jojkH
r
mH
j

okH
r
jomH
j
96
Hence, the magnetic eld components can be written in
terms of E
z
:
H
r

1
om
e

k
m
@E
z
@r

j
r
@E
z
@j
_ _
97
H
j

j
om
e

@E
z
@r

j
r
k
m
@E
z
@j
_ _
98
Similarly, the curl equation in (59) can be reduced to a
scalar one:
1
r
@rH
j

@r

@H
r
@j
_ _
joeE
z
99
Substituting (97) and (98) into (99), a wave equation of E
z
is obtained:
@
2
E
z
@r
2

1
r
@E
z
@r

1
r
2
@
2
E
z
@j
2
b
2
e
E
z
0 100
This wave equation is the characteristic equation for nd-
ing the admissible a. It is identical to that of a cylindrical
dielectric waveguide whose solutions are
E
z;n
A

n
e
jnj
A

n
e
jnj
J
n
b
e
r; n1; 2; . . . 101
where J
n
is a Bessel function of the rst kind of the nth
order. The magnetic eld of interest is H
j
, which can be
found by substituting (101) into (98):
H
j

j
Z
e
A

n
e
jnj
J
0
n
b
e
r
n
b
e
r
k
m
J
n
b
e
r
_ _ _
A

n
e
jnj
J
0
n
b
e
r
n
b
e
r
k
m
J
n
b
e
r
_ __
102
The resonance can be obtained by enforcing that H
j
0 at
r a:
J
0
n
b
e
a
n
b
e
a
k
m
J
n
b
e
a 0; n1; 2; . . . 103
Note that there are two possible roots for each n, associ-
ated with waves circulating in the clockwise direction e
jnj
and the other in the counterclockwise direction, e
jnj
.
In most cases, the n1 mode is the dominant one, and
the corresponding solution of (103) is 1.84. Hence, b
e
is
derived as
b
e
a1:84 104
The waves circulating in opposite directions could be made
in phase at one of the output ports and out of phase at the
remaining port. Assume that port 1 at j0 is the input,
the electric eld at r a can be obtained by (101) for n1:
E
z;1
E
0
A

1
A

1
J
1
b
e
a 105
At the isolated port at j 1201, the electric eld inten-
sity is
E
z;1
A

1
e
j120

1
e
j120

J
1
b
e
a 106
If the output at the isolated port is made zero, the un-
known constants can be determined by solving the simul-
taneous equations in (105) and (106):
A

1

1 j=

3
p
2J
1
b
n
a
E
0
107
A

1

1 j=

3
p
2J
1
b
n
a
E
0
108
It is interesting to nd that the electric eld at the output
port at j1201 is
E
z;1
A

1
e
j120

1
e
j120

J
1
b
e
a E
0
109
By substituting (107) and (108) into (102), the tangential
magnetic eld at every port of the circulator can be deter-
mined in terms of E
0
. After a lengthy yet straightforward
algebraic manipulation, the results are
H
j
r a
E
0
=Z
e
; j0

E
0
=Z
e
; j120

0; j240

_
110
1458 FERRITE CIRCULATORS
where Z
e
is the effective impedance of ferrite, given by
Z
e

m
e
e
_
111
Sketched in Fig. 10 is the standing-wave pattern in a
magnetized ferrite cylinder. It is clearly shown that ener-
gy inputted to port 1 is outputted to port 2 with none to
port 3.
For comparison, the eld pattern in a cylindrical
dielectric resonator is shown in Fig. 11, where the signals
at ports 2 and 3 are equal in both amplitude and phase.
3.6. Typical Ferrites Used in Circulators
In circulator designs, the selection of an appropriate fer-
rite element is of paramount importance. On one hand, it
must satisfy the electromechanical requirements on size,
weight, power handling, and temperature range. On the
other hand, it must meet the stipulated performance elec-
tronically, including center frequency, bandwidth, inser-
tion loss, and linearity.
Besides the physical dimensions and shape of a ferrite
specimen, the characteristics of the material could make a
major difference in the performance of a circulator. The
most important ferrite parameters are saturation magne-
tization (4pM
s
), Curie temperature (T
c
), and linewidth
(DH). For reference, the characteristics of some spinels
and garnets are listed in Tables 2 and 3.
For example, the magnetic eld needed to establish
ferrimagnetic resonance at f 4GHz is
H
fr
%
o
m
0
g
1:13710
5
A=m1429 Oe
For a specimen of zincnickel spinel with 4pM
s
5000 G,
the Cuire temperature is 3751C, and the damping factor is
a %
m
0
g
2o
DH
1
2H
RF
DH0:17 Np=m
Input
Output
Isolation
Figure 11. Standing-wave pattern in a demagnetized ferrite
cylinder.
Input
Output
Isolation
Figure 10. Standing-wave pattern in a magnetized ferrite cylin-
der.
Table 2. Major Characteristics of Spinels
Composition 4pM
s
(G) T
c
(1C) DH (Oe)
MgAl 650 100 115
1700 225 120
2420 310 180
MgMn 1130 175 180
1900 280 350
2800 300 300
MgMnAl 750 90 120
1300 140 135
1750 225 225
MgMnZn 2500 275 520
3000 240 190
NiAl 1000 400 320
2500 570 490
NiZn 4000 500 270
5000 375 160
LiTi 1000 330 300
2000 490 400
2900 600 550
Table 3. Major Characteristics of Garnets
Composition 4pM
s
(G) T
c
(1C) DH (Oe)
Y 1800 280 45
Yal 250 100 40
550 160 40
1000 210 40
1600 265 40
YgdAl 210 110 65
550 185 65
800 260 75
1400 265 50
YGdAlDy 500 225 95
800 245 70
1200 260 60
1600 280 75
YgdAlHo 550 180 100
700 240 90
800 240 110
CaVIn 600 200 25
1200 220 10
1850 240 15
FERRITE CIRCULATORS 1459
If the circulator is designed to operate in the abovemen-
tioned resonance region, the external magnetic eld re-
quired must be greater than
H
ext
H
int
M
s
> H
RF
M
s
1827 Oe
4. APPLICATIONS
4.1. Isolator
The principal application of circulators is load isolation, a
low transmission loss in one direction, and a relatively
high attenuation in the reverse path. At microwave fre-
quencies, generators are vulnerable to frequency shifting
due to load variations; therefore isolation is needed to
protect them from waves reected from the less than per-
fectly matched loads. Reasonable isolation can be imple-
mented by inserting a circulator between the source and
its load with the third port terminated at a matched load
as depicted in Fig. 12.
For better protection, the isolation can be obtained by a
series of circulators. However, if ampliers are used, ad-
ditional isolations as shown in Fig. 13 are recommended
because ampliers are seldom unconditionally stable for
all loads.
Shown in Fig. 14 is a circulator inserted between a re-
ceiving antenna and a low-noise amplier. In satellite re-
ceiving antennas, isolation is needed to prevent waves
leaking from the low-noise amplier from interfering with
the weak signals picked up by the antenna.
Good isolation between the source and the load can be
obtained by using a one-port negative-resistance amplier
as illustrated in Fig. 15. It is noted that many IMPATT
(impact ionization avalanche transit time) and BARITT
(barrier-injected transit time) diodes exhibit a negative
input impedance of Z
d
. The output signal is that reected
from the diode as a result of impedance mismatch, and the
gain of the amplier is equal to
r
2

Z
d
Z
0
Z
d
Z
0
_ _
2

Z
0
jZ
d
j
Z
0
jZ
d
j
_ _
2
112
where Z
0
is the characteristic impedance of the transmis-
sion line.
In fact, greater isolation can be obtained by inserting
an additional circulator in the isolator in Fig. 15 as shown
in Fig. 16.
4.2. Duplexer
A duplexer is a device that allows an antenna to serve as a
transmitter as well as a receiver. Many duplexers made of
hybrids junctions, solid-state switches, and others are
available on the market, but simple ferrite duplexers are
preferred because they allow a single antenna to carry out
both functions simultaneously. The outgoing waves are
coupled to the transmitting antenna through a circulator,
and signals picked up by the antenna are fed to the
receiving amplier via a circulator such as that shown
in Fig. 17.
Load
Source
Z
o
Figure 12. Isolation between the source and its load.
Source
Load
Diode
Figure 15. A negative-resistance isolator.
Z
o
LNA Receiver
Figure 14. Isolation between a receiving antenna and a low
noise amplier.
Load
Source
Amp
Z
o
Z
o
Figure 13. Isolation between the source, the ampliers, and the
load.
Source
Load
Z
o
Diode
Figure 16. An enhanced negative-resistance isolator.
1460 FERRITE CIRCULATORS
For a poorly matched antenna, part of the outgoing
signal is reected by the antenna and is directed by the
circulator to the amplier with the incoming signal. In
order to enhance the transmitterreceiver isolation and to
prevent damage to the amplier, a limiter is usually added
at the output port of the circulator as shown in Fig. 18.
4.3. Reciprocal Ferrite Switch
As wave circulation in a circulator is dependent on the
polarity of its excitation, the direction of circulation can be
reversed by reversing the biasing magnetic eld. It is also
noted that ferrite retains its magnetization even after the
external driving force is removed and waves can circulate
in the same course until the permanent magnetization
effect wears off. If the magnetic eld is generated electro-
magnetically by a current coil, magnetization can be re-
versed by a large current pulse in an appropriate
direction. On the basis of this idea, a reciprocal ferrite
switch has been designed and is sketched in Fig. 19.
4.4. Multiplexing
A multiplexer is a device that carries many channels of
signals in a given bandwidth. The simplest one is the dip-
lexer of two channels shown in Fig. 20. Signals output
from the lowpass lter are directed by the circulator to the
highpass lter, where they are rejected. The reected
waves are then forwarded to the common channel with
the output of the highpass lter via the same circulator.
In general, multiplexers can be built by simply inter-
connecting the lters together, but utmost care must be
taken to reduce interactions among various lters. Multi-
plexers with circulators, on the other hand, can accommo-
date almost any lter because the circulator serves as an
isolator as well as an integrator. Naturally, this concept
can be extended to design a three-channel multiplexer as
sketched in Fig. 21. As circulators have insertion losses,
this conguration is not recommended for building a mul-
tiplexer of many channels.
5. TYPES OF CIRCULATORS
Of all the available circulators, the junction circulator is
by far the dominant one. Over the years, many types of
junction circulators of different power ratings have been
developed for use at a wide range of frequencies, from VHF
Receiver
Transmitter
Figure 17. A ferrite duplexer.
Limiter
Transmitter
Receiver
Figure 18. A ferrite duplexer with a limiter.
Source
Switched
circulator
Unswitched
circulator
Load
Electromagnet
Figure 19. A reciprocal ferrite switch.
Channel 1
Channel 2
Figure 20. A diplexer.
Channel 1
Channel 2
Channel 3
Figure 21. A three-channel multiplexer.
FERRITE CIRCULATORS 1461
to millimeter waves. Depending on the transmission
medium, circulators can be sorted into categories of wave-
guides, microstrip lines, striplines, and coaxial lines. Com-
mon to all junction circulators, the magnetization is
transverse to the direction of wave propagation.
For comparison, waves propagate in the direction of
magnetization in differential phase shift circulators and
Faraday rotation circulators. Hence, the principle of oper-
ation is drastically different from that of junction circula-
tors. These four-port circulators are skipped in this article
simply because they are seldom used in modern systems.
5.1. Waveguide Y-Junction Circulators
At microwave and millimeter-wave frequencies, the only
feasible circulator is a waveguide junction circulator, con-
stituting a Y junction of three H-plane rectangular wave-
guides as shown in Fig. 22 [13,14]. A ferrite cylinder
spanning the height of the waveguide is mounted normal
to the oor of the waveguide junction. It is placed at the
center of the Y junction for symmetry. Moreover, a metal
plate with three symmetric spikes pointing toward the
ports of entry is added between the ferrite specimen and
the waveguide oor for better impedance matching as
shown in Fig. 23. Also shown in the same gure is the
glue that serves as an insulating layer separating the fer-
rite post and the impedance-matching plate. The ferrite
post is magnetized axially by a permanent magnet or an
electromagnet outside the waveguide.
Even though the connecting waveguides are operating
in TE
10
modes, higher-order modes, including the evanes-
cent ones, exist in the junction. Note that the ferrite cyl-
inder itself is a dielectric resonator whose resonant
frequency is dependent on its length and its radius a
f
.
To bring about wave circulations in the waveguide junc-
tion, a
f
is obtained by solving the characteristic equation
given in (103). However, the result obtained from (104) is
adequate for most cases:
a
f

1:84
b
e

1:84l
2p

m
e
e
p 113
The leverage in adjusting does provide us a mechanism
to generate higher-order modes [33] such as HE
11
for im-
proving the performance of the circulator, including im-
pedance matching, bandwidth, quality factor, and
insertion loss [15]. To this end, a partial-height ferrite cyl-
inder on an elevated oor as shown in Fig. 24 is developed.
For stronger mechanical support, the space above and be-
low the shortened post is lled by two dielectric cylinders
of similar shape. Moreover, it is observed that the electric
eld tangential to the central plane of the waveguide is
either maximum or zero due to symmetry. If the post is cut
into two halves, a zero tangential eld is obtained. Hence,
an additional option is gained by cutting the ferrite post
into two with one remaining on the oor and the other
attached to the ceiling as shown in Fig. 25.
Ferrite posts of circular and triangular cross sections are
the popular choices, but hexagonal and other shapes
can also be used. Since disks can be made by slicing a cyl-
inder and their principles of operation are identical, it is
Ferrite
R
0
Figure 22. A Y-junction waveguide circulator.
Glue
Ferrite
Waveguide
wall
Impedance
matching plate
Figure 23. Side view of a ferrite post in a waveguide Y junction.
Ferrite
Dielectric
2a
f
l
Figure 24. Side view of a partial ferrite post in a waveguide Y
junction.
l/2
l/2
Figure 25. Side view of a divided partial ferrite post in a wave-
guide Y junction.
1462 FERRITE CIRCULATORS
considered as a cylinder. It has been shown that circulator
with a triangular ferrite cylinder might have a marginally
lower insertion loss [16], but its overall performance is
comparable to that of a circular one. However, in cases
where the tolerance on physical dimensions is very slim,
the circular post prevails because grinding and polishing
the superhard and brittle ferrite cylinder to a cross section
of a perfect circle is much easier than making one with a
faultless equilateral triangle. The situation is especially
critical for millimeter-wave circulators because the diame-
ter of the relevant ferrite posts is less than a millimeter.
Along this line of thought, a new design is proposed in
which the cylinder is replaced by a sphere as shown in Fig.
26. Spheres are chosen because they can be mass-produced
with the highest precision as if they were ball bearings
produced by the time-honored technologies [17]. In fact, the
major gain in this maneuver is ease in implementation. As
the base of a ferrite cylinder is very small, mounting it
normally on a at surface is easier said than done; there-
fore erecting one properly on the oor of an equally minis-
cule Y-junction circulator is very tricky. This agony is
totally resolved in the present design because a sphere is
always normal to the oor. The performance of this circu-
lator is at least on par with the traditional one. Rigorous
analysis and experimental studies of waveguide circulators
with ferrite spheres will be examined in later sections.
5.2. Stripline Y-Junction Circulators
At ultrahigh frequencies, stripline circulators are more
popular because waveguides are seldom used in this fre-
quency range [18]. The junction where three striplines
meet is a conducting disk of radius a
c
. Anisotropic effects
are provided by two ferrite disks of radius a
f
above and
below the center conductor. For better performance, a
f-
Za
c
, and both of them are much larger than the width of
the stripline w. Preferably, the ferrite disks are thin
enough such that the sandwich can be squeezed between
the upper and lower conducting plates of the original
striplines as shown in Fig. 27. The junction circulator is
magnetized externally such that the magnetic vector is
normal to the surface of the ferrite disks. The ferrite disks
and the center conductor may be in any shape as long as
the threefold symmetry is retained.
The striplines carry TEM modes only; therefore wave
propagations are transverse to the magnetization vector.
As w5a
c
, the tangential magnetic eld is constant over
the widths of the striplines and is zero on the circumfer-
ence of the center conductor. Since a
f
Ea
c
, the magnetic
eld strength on the peripheral of the ferrite disk is ob-
tained by enforcing the continuity of tangential magnetic
elds, a constant at the input ports and zero elsewhere.
The normal electric eld in ferrite E
z
satises the Helm-
holtz equation given in (100), and the magnetic elds are
expressed in terms of E
z
as given in (97) and (98).
The required wave equation is derived by expressing
E
z
as
E
z
j
_
p
p
Gj; j
0
H
j
j
0
dj
0
114
where G(j;j
0
) is the Green function [18], given by
Gj; j
0
j
Z
e
2p
J
0
b
e
a
f

J
0
0
b
e
a
f


Z
e
p

1
n1

k
m
nJ
n
b
e
a
f

b
e
a
f
sin nj j
0
jJ
0
n
b
e
a
f
cos nj j
0

J
0
n
b
e
a
f

k
m
nJ
n
b
e
a
f

b
e
a
f
_ _
2
J
n
b
e
a
f

115
Except for simple congurations, the wave equation in a
Y-junction stripline circulator stated in (103), (114), and
(115) cannot be solved analytically in a closed form. How-
ever, the electric and magnetic elds can be accurately
and efciently determined by numerical means. In terms
of the eld quantities, other characteristics of the junction
circulator such as input impedance, reection coefcient,
transfer function, isolation, bandwidth, quality factor, and
insertion loss can be derived.
However, for most applications, the radius of the ferrite
disks required for resonance can be obtained by (113)
and the eld pattern in ferrite is similar to that sketched
in Fig. 10. Moreover, stripline junction circulators are
usually biased far above resonance in the UHF region:
H
int
bH
fr
. It then follows that m and k can be approximated
by
m
e
% m1
4pM
s
H
int
116
k
4pM
s
H
fr
H
2
int
117
a
f
Ferrite sphere
Resin
Figure 26. Side view of a ferrite sphere in a waveguide Y
junction.
Ferrite
Ferrite
Center conductor
Ground
plane
Ground
plane
w
a
f
a
c
Figure 27. A stripline junction circulator.
FERRITE CIRCULATORS 1463
It is also recommended that the minimum radius required
be
a
f
>
4w
3
118
5.3. Microstrip-Line Y-Junction Circulators
One reason why circulators are seldom used today is the
lack of effective circulators for microstrip lines, the most
popular media of wave transmission at UHF and micro-
wave bands. Traditionally, a microstrip-line junction cir-
culator is made by laying three metallic strips on a ferrite
disk as shown in Fig. 28. For better impedance matching,
the center conductor may assume any three-way symmet-
ric shape such as the equilateral triangle shown in Fig. 28.
Unlike its stripline counterpart, a microstrip-line junction
circulator has only one ferrite disk on a large ground
plane. For the microstrip line on a substrate of ferrite, its
characteristic impedance is quite different from that on a
dielectric substrate. Hence, an additional impedance
matching is needed at every port of entry.
In order to improve impedance matching and to reduce
cost, a simpler version is developed. Instead of putting the
microstrip lines and the center conductor on a ferrite disk,
they are printed on a dielectric substrate as illustrated in
Fig. 29. Wave circulation is made possible by covering the
junction with a ferrite disk of radius a
f
. Similar to a wave-
guide Y-junction circulator, the disk can be replaced by a
hemispherical one. The major advantage gained in this
maneuver is exibility because the frequency of a given
circulator can easily be altered by replacing its ferrite by a
different material or by changing its physical shape or di-
mension. Unfortunately, the performances of the afore-
mentioned circulators are not very satisfactory and
further improvements are needed.
One improvement proposed is a microstrip-line circu-
lator with a ferrite sphere as shown in Fig. 30 [19]. Even
though a better performance is observed, it remains infe-
rior to those waveguide and stripline circulators. With the
increasing popularity of microstrip lines and other copla-
nar waveguides, development in this direction is desper-
ately needed.
5.4. Lumped-Element Circulators
Common to all junction circulators, as mentioned previ-
ously, the size of the ferrite specimen is proportional to
wavelength. This means that the circulator could become
prohibitively large in HF and VHF regions. To this end, a
simple circulator is designed by winding three coils
around a ferrite disk as shown in Fig. 31 [20]. The coils
in these lumped-element circulars are oriented 1201 from
one another, so are the magnetic elds created by these
coils. Two ferrite disks of 1015mm in diameter and 1
2 mm thick are packed inside a grounded metal box. The
so-called coil in each port in fact consists of two one-turn
coils in parallel; one returns through the bottom of the
shielding box and the other via its top, as illustrated in
Fig. 32. In order to distribute the magnetic eld more
evenly in ferrite, the current is divided into multiple l-
aments as shown in Fig. 33. The magnetic eld is parallel
to the surface of the ferrite disk. The magnetization vector
is normal to the magnetic elds due to the coils and the
ferrite disk is usually biased above resonance.
Ground plane
Figure 28. A junction circulator with microstrip lines on a sub-
strate of ferrite disk.
Ferrite
Sphere
Figure 30. A microstrip-line junction circulator with a ferrite
sphere.
Substrate
Ground plane
Figure 29. A microstrip-line junction circulator with a ferrite
disk on top. Figure 31. A lumped constant circulator.
1464 FERRITE CIRCULATORS
Applying the concept of a balanced three-phase power
circuit, the neutral or return line can be obliterated, pro-
vided the three branches are 1201 out of phase from one
another. The combined length of each coil is much shorter
than a wavelength, so it is essentially an inductor. To re-
store resonance at the junction, each branch is shunted to
ground via a capacitor, and the resultant star circuit is
depicted in Fig. 34a. Alternatively, a series capacitor C
s
can
be added in series with the coil as shown in Fig. 34b. It can
be shown that the waves rotating in opposite directions in
ferrite can be characterized by two inductances, namely
L

L
0
m k 119
L

L
0
m k 120
where L
0
is the inductance of each phase with the ferrite
disk removed. For an ideal circulation, these inductances
are given by
L

2
o
2
C
s
121
L

1:156
Z
0
o
122
The gyrotropic parameters can be determined on the basis
of (119)(122):
m
1
o
2
L
0
C
s
123
k
Z
0
1:73oL
0
124
6. IN-DEPTH ANALYSIS OF SOME SELECTED
CIRCULATORS
The formulas presented in Section 4 are derived for
circulators with clear-cut geometries, and they are
based on many approximations, including the over sim-
plied ones. It is obvious that these simple formulas
are not accurate enough for computer-aided designs of
miniature circulators at microwave and millimeter-wave
frequencies. In-depth analyses of circulators, on the
other hand, do exist, but the number is limited because
the topic itself is commonly considered as old-fashioned,
compared with mobile communications, networking,
nanotechnology, and other trendy ones. Even if one exists,
it is customarily developed for waveguide circulators.
Nonetheless, it is worthwhile to present some of the
techniques in detail in this section because they can be
modied to treat other types of circulators. Most impor-
tant of all, the rationale behind the development of these
techniques and the insights gained during the process are
invaluable in the design of innovative circulators in the
future.
All in all, four techniques for analyzing typical circu-
lators will be covered. Note that the emphasis is not placed
on the techniques themselves, but on how the technique is
applied to solve the problem and the characteristics of the
circulators. Through the examples, readers may pick up
the skill for identifying an appropriate method for ana-
lyzing their circulators.
Ferrite
Center conductors
Shielding box
Figure 32. Side view of a lumped constant circulator.
Parallel
Conductors
Magnetic
flux lines
Magnetization
Ferrite
Figure 33. Magnetic eld lines in ferrite and around the current
coil.
(a) (b)
Figure 34. Equivalent circuit of the lumped
constant circulator: (a) shunt; (b) series.
FERRITE CIRCULATORS 1465
6.1. Finite-Difference Method
The conguration of interest is an H-plane waveguide
Y-junction circulator with a full-length ferrite post
as sketched in Fig. 22. Consistent with the assumption
that the only mode propagating in the waveguide in
the dominant TE
10
mode, all eld quantities in the
circulator are independent of z, a direction normal to
its oor. Hence, the problem is reduced to a two-dimen-
sional one. It can be shown that the transverse compo-
nents of the electromagnetic wave in ferrite can
be determined in terms of E
z
and H
z
. In the cylindrical
coordinate system, the characteristic equations in
ferrite are
@
2
E
z
@r
2

1
r
@E
z
@r

1
r
2
@
2
E
z
@j
2
b
2
e
E
z
0 125
@
2
H
z
@r
2

1
r
@H
z
@r

1
r
2
@
2
H
2
@j
2
b
2
H
z
0 126
where b and b
e
are as given in (88) and (91), respectively.
Equations similar to (125) and (126) can be used to
determine both E
z
and H
z
exterior to the ferrite
post in the junction, provided b
e
in (125) is replaced
by b.
The major difculty in solving the characteristic equa-
tions is to resolve the conict due to the rectangular wave-
guides and the cylindrical junction. The simplest yet
sufciently in-depth method for solving this boundary
problem is the nite-difference method proposed by Yung
and his researchers [21]. In order to keep the boundary of
a waveguide Y junction cylindrical in shape, a truncation
boundary as illustrated in Fig. 35 is introduced. With ev-
anescent modes ignored, the eld values at the truncation
boundary can be expressed in terms of the TE
10
modes at
the input and output ports; therefore all boundary values
are dened. It is readily recognized that discretization of
(125) requires the eld values at ve nodes. A unique so-
lution is obtained by enforcing that the tangential electric
and magnetic elds are continuous on the surface of the
ferrite post.
The nite-difference method is applied to analyze
a waveguide Y circulator with a G1002 ferrite post
(e
r
15.4, 4pM
s
1000G, DH20 Oe, and a
f
3mm).
The width of the waveguide is 22.86 mm and the internal
magnetic eld is H200Oe. The reection coefcient,
the insertion loss, and the isolation are plotted in Figs.
36 and 37, and 38, respectively. Also shown in these g-
ures are the experimental data published in Ref. 22. Ex-
cellent agreement between the analytical results and
measurements is seen. It is also observed from the plots
that the nite loss of ferrite has negligible effects on the
general performance of a circulator except when the fre-
quency is close to ferrimagnetic resonance.
6.2. Finite-Difference Time-Domain Method
Other than full-height ferrite posts, the eld quantities in
the waveguide junction are usually dependent on z, even
though the dominance of TE
10
remains valid in the input
and output waveguides. For a three-dimensional case, the
nite difference method just mentioned is no longer ade-
quate, but the nite-difference time-domain (FDTD) meth-
od is. Yees discretization scheme [23] is chosen because it
does not require the time-consuming matrix inversion and
convergence is assured. In terms of Yees mesh, the FDTD
j - Plane
Truncation
boundary node
waveguide
wall
Figure 35. The truncation boundary and the nodal pattern in
the waveguide junction.
lossless
lossy
experimental
R
e
f
l
e
c
t
i
o
n

(
d
B
)
0
5
10
15
20
25
7.5 8.5 9.5 10.5 11.5 12.5
Frequency (GHz)
Figure 36. Variation of the reection coefcient
at the input port with frequency.
1466 FERRITE CIRCULATORS
approximations of six eld components in (75) through
(78) are
H
n1=2
x
i; j
1
2
; k
1
2
_ _
H
n1=2
x
i; j
1
2
; k
1
2
_ _

m
22
m
2
0
m
2
pa
2

tat
1
taz
_
E
n
y
i; j
1
2
; k1
_ _ _
E
n
y
i; j
1
2
; k
_ __

1
tay
E
n
z
i; j 1; k
1
2
_ _
E
n
z
i; j; k
1
2
_ _ _ __

m
12
m
2
0
m
2
pa
2

tat
1
tax
E
n
z
i
1
2
; j
1
2
; k
1
2
_ _ _ _
E
n
z
i
1
2
; j
1
2
; k
1
2
_ __

1
taz
E
n
z
i; j
1
2
; k 1
_ _
E
n
z
i; j
1
2
; k
_ _ _ __
127
H
n1=2
y
i
1
2
; j; k
1
2
_ _
H
n1=2
y
i
1
2
; j; k
1
2
_ _

m
11
m
2
0
m
2
pa
2

tat
1
tax
E
n
z
i 1; j; k
1
2
_ _ _ _
E
n
z
i; j; k
1
2
_ __

1
taz
E
n
x
i
1
2
; j; k 1
_ _
E
n
x
i
1
2
; j; k
_ _ _ __

m
21
m
2
0
m
2
pa
2

tat
1
taz
E
n
y
i
1
2
; j; k1
_ _ _ _
E
n
y
i
1
2
; j; k
_ __

1
tay
E
n
z
i
1
2
; j
1
2
; k
1
2
_ _ _
E
n
z
i
1
2
; j
1
2
; k
1
2
_ ___
128
lossless
lossy
experimental
7.5 8.5 9.5 10.5 11.5 12.5
Frequency (GHz)
I
n
s
e
r
t
i
o
n

l
o
s
s

(
d
B
)
0
5
10
Figure 37. Variation of the insertion loss at the
output port with frequency.
lossless
lossy
experimental
I
s
o
l
a
t
i
o
n

(
d
B
)
0
5
10
15
20
25
7.5 8.5 9.5 10.5 11.5 12.5
Frequency (GHz)
Figure 38. Variation of the isolation at the
third port with frequency.
FERRITE CIRCULATORS 1467
H
n1=2
z
i
1
2
; j
1
2
; k
_ _
H
n1=2
z
i
1
2
; j
1
2
; k
_ _

tat
m
0
1
tay
E
n
x
i
1
2
; j 1; k
_ _ _ _
E
n
x
i
1
2
; j; k
_ _

1
tax
E
n
y
i 1; j
1
2
; k
_ _ _
E
n
y
i; j
1
2
; k
_ ___
129
E
n1
x
i
1
2
; j; k
_ _
E
n
x
i
1
2
; j; k
_ _

tat
epsilon
r
1
tay
H
n1=2
z
i
1
2
; j
1
2
; k
_ _ _ _
H
n1=2
z
i
1
2
; j
1
2
; k
_ __

1
taz
H
n1=2
y
i
1
2
; j; k
1
2
_ _ _
H
n1=2
y
i
1
2
; j; k
1
2
_ ___
130
E
n1
y
i; j
1
2
; k
_ _
E
n
y
i; j
1
2
; k
_ _

tat
epsilon
r
1
taz
H
n1=2
x
i; j
1
2
; k
1
2
_ _ _ _
H
n1=2
x
i; j
1
2
; k
1
2
_ __

1
tax
H
n1=2
z
i
1
2
; j
1
2
; k
_ _ _
H
n1=2
z
i
1
2
; j
1
2
; k
_ ___
131
E
n1
z
i; j; k
1
2
_ _
E
n
z
i; j; k
1
2
_ _

tat
epsilon
r
H
n1=2
y
i
1
2
; j; k
1
2
_ _ _ _
H
n1=2
y
i
1
2
; j; k
1
2
_ __

1
tay
H
n1=2
x
i; j
1
2
; k
1
2
_ _ _
H
n1=2
x
i; j
1
2
; k
1
2
_ ___
132
where Dx, Dy and Dz are the widths of the Yee cell, and
m
11
m
22
m
xx
and m
12
m
21
m
xy
are as given in (56) and
(57), respectively.
As the FDTD approximations above are expressed in
Cartesian coordinates, the algebra involved in solving a
waveguide Y-junction circulator will be very tedious and
the resultant time of computation prohibitively long. To
better depict the nature of the FDTD scheme, it is applied
to solve a waveguide T-junction waveguide as shown in
Fig. 39. Although a partial-height ferrite post is shown,
the FDTD scheme is, in fact, applicable for any body of
arbitrary shape, including a sphere. Moreover, it is noted
that some values in (127)(132) are not available directly
from Yees scheme, but fortunately, they can be generated
by a linear interpolation of the neighboring elds. Details
are not given here; interested readers are referred to the
paper by Schneider and Hudson [24].
The waveguide T-junction circulator under investiga-
tion consists of three X-band waveguides of dimensions
22.86 10.16 mm, and the radius of its full-height ferrite
post is a
f
3.5 mm. Variations of the reection coefcient
at the input, the insertion loss at the output, and the
isolation at the isolated port are respectively plotted in
Figs. 40 and 41, and 42 as a function of frequency. It is
readily seen that excellent agreement with the results
published in Ref. 25 is observed.
R
0
Figure 39. A waveguide T-junction circulator.
25
20
15
10
5
0
7.5 8.5 9.5 10.5 11.5 12.5
Frequency (GHz)
R
e
f
l
e
c
t
i
o
n

(
d
B
)
FDTD
[25]
Figure 40. Variation of the reection coefcient at the input port
with frequency.
1468 FERRITE CIRCULATORS
To better depict the characteristics of wave circulation,
the power density in the ferrite post is plotted in Fig. 43 as
a function of j at f 10.3GHz for three different r values.
The input and output power densities in a closed path at a
given r are always conserved as ferrite is assumed to be
lossless and no power is dissipated in the waveguide
junction area. It is also observed that the power entering
the input port at j901 is transmitted to the output port
at j2701.
For comparison, the variations of power densities out-
side the ferrite post are plotted in Fig. 44, where R
0
is the
radius of the largest circle inside the T junction. The
nature of power transfer to the output port is clearly
demonstrated.
Although we have absolute faith in the accuracy of our
results computed in the numerical analysis, it is always
nice to know that they compare well with those obtained
in experimental studies. In fact, readings in measurement
are vulnerable to numerous errors and constraints, in-
cluding purity of the materials, uniformity of the mixture,
tolerance of the physical cuttings, stability of the fre-
quency generation, imperfect matching of the intercon-
nects, loss due to spurious radiations, and precision of the
measuring instrument. There is no reason to believe that
measurements are more reliable than computational data.
However, most engineers and researchers will feel more
comfortable whenever they see a reasonably good agree-
ment between the analytical and the experimental results.
In this connection, the FDTD study is repeated for a
waveguide circulator with a NiZn ferrite sphere (a
f

1.0mm, e
r
2.25, 4pM
s
5000G, H
ext
1700Oe, and DH
120 Oe). The numerically obtained reection coefcients,
insertion losses, and isolations are plotted against the
experimental data in Fig. 45. Given the complexity of
the conguration and the gyrotropic nature of the pro-
blem, the agreements between the analytical results and
the measurement are more than acceptable.
6.3. Finite-Element Method
The nite-element method (FEM) is known for its cap-
ability to model complex structures. Since it requires only
information about the geometry of the device to be ana-
lyzed, the method can be developed into a general-purpose
software package, for example, Ansoft HFSS. With the use
of edge-based vector elements, the FEM can handle dis-
continuity interfaces between different materials without
invoking spurious solutions. When combined with the
25
20
15
10
5
0
7.5 8.5 9.5 10.5 11.5 12.5
Frequency (GHz)
I
n
s
e
r
t
i
o
n

l
o
s
s

(
d
B
)
FDTD
[25]
Figure 41. Variation of insertion loss at the output port with
frequency.
20
15
10
5
0
7.5 8.5 9.5 10.5 11.5 12.5
Frequency (GHz)
I
s
o
l
a
t
i
o
n

(
d
B
)
FDTD
[25]
Figure 42. Variation of the isolation at the other output port
with frequency.
1.4
1
0.6
0.2
0.2
0.6
1.4
P
o
w
e
r

d
e
n
s
i
t
y
1
0 60 120 180 240 300 360
j/R
0
=0.3
j/R
0
=0.6
j/R
0
=0.9

Figure 43. Variation of the power density inside the ferrite post
with j.
1
2
3
4
0
1
2
3
4
0 60 120 180 240 300 360
P
o
w
e
r

d
e
n
s
i
t
y
j/R
0
=0.3
j/R
0
=0.9
j/R
0
=0.6

Figure 44. Variation of the power density outside the ferrite post
with j.
FERRITE CIRCULATORS 1469
asymptotic waveform evaluation (AWE) and complex fre-
quency-hopping (CFH) techniques, the FEM can generate
frequency responses over a wide band very efciently.
Although the low-order FEM suffers from the problem of
numerical dispersion, which tends to accumulate as the
wave propagates, this problem has been largely overcome
with the development of higher-order nite elements [28].
The major remaining difculty is the relatively long com-
puting time and the associated memory requirements for
the resulting nite-element system. More recently, the
preconditioned iterative solutions [29] and numerical
de-embedding techniques [30] have been developed to
accelerate solutions and reduce memory.
Consider a microwave ferrite device with multiple
ports, inside the ferrite device, where the eld satises
the vector partial differential equation
rfm
r

1
.
rEg k
2
0
e
r
.
E0 133
The boundary condition on the surface of port i can derives
as
^ n n
i
rEP
i
E U
inc
i
134
where ^ nn
i
denotes the outward unit vector normal to the
surface S
i
, the cross section of port i, and
P
i
E

m
k
2
0
jk
m
e
TM
e;m
__
S
i
e
TM
e;m
.
Eds

n
jk
n
e
TE
n
__
S
i
e
TE
e;n
.
Eds
135
U
inc
i
^ nn
i
rE
inc
i

m
k
2
0
jk
m
e
TM
e;m
__
S
i
e
TM
e;m
.
E
inc
i
ds

n
jk
n
e
TE
n
__
S
i
e
TE
e;n
.
E
inc
i
ds
136
In the above, e
TM
t;m
denotes the transverse part of the
electric eld of the mth transverse magnetic (TM) mode,
and e
TE
t;n
denotes the transverse part of the electric eld of
the nth transverse magnetic (TE) mode. E
inc
i
denotes the
incident elds at port i. In accordance with the general
variational theory, the functional for the boundary value
problem dened above given by
FE; E
a

1
2
___
V
frE
a

.
rE e
r
k
2
0
E
a
.
EdV

1
2

N
i 1
__
S
i
E
a
.
P
i
E E
a
.
U
inc
i
ds
137
where N denotes the total number of ports and E
a
denotes
the solution to a properly dened adjoint problem. The
FEM discretization of (137) using vector basis functions
yields the resulting matrix equation:
AfEg fbg 138
The solution of (138) can be done using an iterative
solver such as conjugate-gradient method (CGM) and gen-
eralized minimum residuals method (GMRES), and
preconditioned techniques can be used to improve the con-
dition number of the FEM system and accelerate the
convergence rate of iterative solvers. As a result, this
will yield the electric elds everywhere, including
those over the ports from which the S parameters can be
determined. A three-port circulator loaded with a
full-height ferrite post is simulated with the FEM algo-
rithm given above. The circulation characteristics
are shown in Fig. 46 and are compared with the experi-
mental data of Ref. 31. Another example considers an 8-
mm waveband H-plane Y-junction waveguide circulator
with a partial-height ferrite post [30]. The computed and
experimental results are given in Fig. 47 and show good
agreement.
20
16
12
8
4
0
32 33 34 35 36 37 38 39
Frequency (GHz)
d
B
Reflection
Isolation
Insertion loss
Insertion loss(measured)
Isolation (measured)
Figure 45. Comparison of reection coefcients, insertion losses,
and isolations for a waveguide T-junction circulator with a ferrite
sphere with measurements.
20
15
10
5
0
8 9 10 11 12
Frequency (GHz)
d
B
Insertion
Isolation
Reflection
Figure 46. Performance of the H-plane Y-junction circulator
with a full-height TT1-109 circular ferrite post (values comput-
ed from this theory; values given by Ref. 31).
1470 FERRITE CIRCULATORS
6.4. Mode-Matching Method
Discrete-domain techniques such as the nite-difference
and nite-element methods are slowly convergent,
even for those with simple congurations. Straight appli-
cation of the aforementioned FDTD scheme to analyze a
waveguide Y-junction circulator with a ferrite specimen
of arbitrary shape is possible but is used only as a last
resort. The culprit is its complex boundary condition,
which involves rectangular walls in the feeding
waveguides, cylindrical truncation border of the wave-
guide junction, and the spherical surface of the ferrite
sphere. It is further complicated by the anisotropic nature
of ferrite. As illustrated earlier in this section, the
computation time can be significantly reduced if a three-
dimensional problem can be approximated by a two-di-
mensional one. Of course, this is not a realistic assumption
for most circulators, but sometimes this scenario can be
created in a piecewise sense by partitioning the congu-
ration of interest appropriately. For example, the partial-
height ferrite post shown in Figs. 23 and 24 can be divided
horizontally into three parts such that each part contains
a cylindrical resonator of uniform content. In contrast
to the previous example, the eld quantities in a resonator
of nite height is not independent of z, yet the subsequent
computation can be significantly cut by assuming that
its dependence takes up the e
jb
z
z
form. As a result,
the relevant characteristic equation is similar to that of
a two-dimensional case given in (100) with b
e
replaced by
b
n
, given by
b
n

b
2
e
b
2
z
_
139
The electric and magnetic elds in these resonators are
obtained by summing up all possible solutions or modes of
the characteristics equation. Finally, the overall solution
is deducted by stipulating that the tangential electric and
magnetic elds be continuous at both ends of the cylin-
drical resonators in the adjacent parts.
The mode-matching scheme can be extended to cover a
ferrite body of revolution such as a sphere. For illustra-
tion, if the circle shown in Fig. 48 is segmented vertically
into parallel parts, a cascade of tubes is formed by
revolving the partitioned circle around the z axis. Note
that each tube consists of three components of heights
h
1
, h
2
, and h
3
. Success of the present scheme relies on
our ability to nd the eld quantities in a waveguide tube
of uniform content analytically. Again, details are skipped
and readers are referred to another source [27] for an in-
depth analysis of a waveguide lube of an annular base.
The nal solution is obtained by requiring that the tan-
gential electric and magnetic elds be continuous on the
cylindrical surfaces as well as the annular bases of the
waveguide tubes.
Although the mathematical derivation is very time-
consuming and the equations obtained are very lengthy,
the required computation time is very short. Characteris-
tics of a waveguide Y-junction circulator with a ferrite
sphere can easily be determined. Results similar to those
given in Figs. 4951 can be computed in a personal com-
puter in milliseconds, and excellent agreement between
the analytical with experimental results is observed. Due
to the computational efciency, we can afford to nd the
general eld pattern inside a waveguide junction circula-
tor. Shown in Fig. 49 are the strength and direction of the
electric eld vector in a magnetized ferrite sphere. The
eld pattern in the waveguide junction exterior to the
magnetized ferrite sphere is shown in Fig. 50. In both g-
ures, the nature of wave circulation is clearly demonstrat-
ed. For comparison, the symmetric eld pattern in the
waveguide with a demagnetized ferrite sphere is sketched
in Fig. 51.
30
25
20
15
10
5
0
28 29 30 31 32 33 34 35
Frequency (GHz)
d
B
Isolation
Insertion
Figure 47. Performance of the H-plane Y-junction circulator
with a partial-height circular ferrite post (values computed
from this theory; experimental results given by Ref. 30).
waveguide wall
Z
h
3
h
2
h
1
o j
1
j
2
j
N
j
h
Figure 48. Segmentation of a half circle into a cascade of wave-
guide tubes with annulus bases.
FERRITE CIRCULATORS 1471
7. CONCLUDING REMARKS
Study of circulators is a multidisciplinary one that covers
material science, ferrimagnetism, wave propagation, and
circuit design; therefore, it nds a special position in many
engineering curricula. Design of circulators marks a sharp
deviation from the current trend in problem solving as
engineers are inclined to solve their assignments by brute
force, such as using a faster computer, a better database, a
larger memory, and a wider bandwidth at a higher fre-
quency. Here, an engineers ingenuity can be fully shown
as significant improvement can be obtained by using dif-
ferent material, changing the size and shape of ferrite,
ne-tuning its location, and other insignificant altera-
tions. Circulators may be replaced in the future, but the
underlying principles will prevail forever as it is an excel-
lent manifestation of our ability in harnessing the nature
for our benet.
BIBLIOGRAPHY
1. D. M. Pozar, Microwave Engineering, Addison-Wesley, Read-
ing, MA, 1990.
2. R. E. Collin, Field Theory of Guided Waves, 2nd ed., IEEE
Press, New York, 1991.
3. H. Bosma, Performance of lossy H-plane Y-circulators, IEEE
Trans. Magn., Mag-2:273277 (1966).
4. J. B. Davis, An analysis of the m-port symmetrical H-plane
waveguide junction with central ferrite post, IEEE Trans.
Microwave Theory Tech. 596604 (Nov. 1962).
5. J. Helszajn, Microwave Junction Circulators: Theory and
Practice, Wiley, New York, 1998.
6. D. K. Linkhart, Microwave Circulator Design, Artech House,
Norwood, MA, 1989.
7. R. A. Waldon, Ferrites: An Introduction for Microwave Engi-
neers, Van Nostrand, London, 1962.
8. F. Keffer, The Magnetic Properties of Materials, Freeman, San
Francisco, 1967.
9. P. J. B. Clarricoats, Microwave Ferrites, Wiley, New York,
1961.
10. R. B. Leighton, Principles of Modern Physics, McGraw-Hill,
New York, 1959.
11. R. F. Soohoo, Microwave Magnetics, Harper &Row, New York,
1970.
12. J. Helszajn, Ferrite Phase Shifters and Control Devices,
McGraw-Hill, New York, 1989.
13. B. A. Auld, The synthesis of symmetrical waveguide circula-
tors, IRE Trans. Microwave Theory Tech. 238246 (April
1959).
14. C. E. Fay and R. L. Comstock, Operation of the ferrite junc-
tion circulator, IEEE Trans. Microwave Theory Tech. 1527
(Jan. 1965).
Figure 49. Strength and direction of the electric vector inside a
magnetized ferrite sphere. (This gure is available in full color at
http://www.mrw.interscience.wiley.com/erfme.)
Figure 50. Strength and direction of the electric vector in a
waveguide Y junction with a magnetized ferrite sphere. (This
gure is available in full color at http://www.mrw.interscience.
wiley.com/erfme.)
Figure 51. Strength and direction of the electric vector in a
waveguide Y junction with a demagnetized ferrite sphere. (This
gure is available in full color at http://www.mrw.interscience.
wiley.com/erfme.)
1472 FERRITE CIRCULATORS
15. J. Helszajn, Design data for radial waveguide circulators us-
ing partial height ferrite resonators, IEEE Trans. Microwave
Theory Tech. MTT-23(3):288298 (March 1975).
16. J. Helszajn, Planar triangular resonators with magnetic
walls, IEEE Trans. Microwave Theory Tech. MTT-26(2):95
100 (Feb. 1978).
17. E. K. N. Yung, D. G. Zhang, and Roger S. K. Wong, Novel
waveguide Y-junction circular with a ferrite sphere for milli-
meter waves, IEEE Trans. Microwave Theory Tech. 44(3):454
457 (March 1996).
18. H. Bosma, On stripline Y-circulators at UHF, IEEE Trans.
Microwave Theory Tech. 6171 (Jan. 1964).
19. E. K. N. Yung, W. B. Dou, D. G. Zhang, and R. S. Chen,
Microstrip circulator made of a magnetized ferrite sphere,
Microwave Opt. Technol. Lett. 28(2):130131 (Jan. 20,
2001).
20. Y. Konishi, Lumped element Y circulator, IEEE Trans. Mi-
crowave Theory Tech. MTT-13(6):852864 (Nov. 1965).
21. E. K. N. Yung, H. Y. Ding, R. S. Chen, and Z. M. Xie, Finite
difference analysis of H-plane waveguide Y-junction circula-
tors, Microwave Opt. Technol. Lett. 20(6):414422 (March 20
1999).
22. J. B. Castillo Jr. and L. E. Davis, Computer-aided design of
three-port waveguide junction circulator, IEEE Trans. Micro-
wave Theory Tech. MTT-18:2534 (1970).
23. K. L. Kunz and R. Luebbers, Finite-Difference Time Domain
in Electromagnetics, CRC Press, Boca Raton, FL, 1993.
24. J. Schneider and S. Hudson. The nite-difference time-
domain method applied to anisotropic material, IEEE Trans.
Anten. Propag. 41(7):994999 (July 1993).
25. R. S. Chen. E. K. N. Yung, Z. M. Xie, and Y. Wang, FDTD
analysis of EM wave circulating a magnetized ferrite body in
free space, Int. J. Infrared Millimeter Wave 19(2):329341
(Feb. 1998).
26. S. A. Ivanov, Applications of the planar model to the analysis
and design of the Y-junction stripline circulator, IEEE Trans.
Microwave Theory Tech. 43(6):12531263 (June 1995).
27. E. K. N. Yung, R. S. Chen, K. Wu, and D. X. Wang, Analysis
and development of millimeter wave waveguide junction cir-
culator with a ferrite sphere, IEEE Trans. Microwave Theory
Tech. 46(11):17211734 (Nov. 1998).
28. J. Liu, J. M. Jin, Edward K. N. Yung, and R. S. Chen, A fast,
higher order three-dimensional nte-element analysis of mi-
crowave waveguide devices, Microwave Opt. Technol. Lett.
32(5):342352 (March 5, 2002).
29. R. S. Chen, E. K. N. Yung, C. H. Chan, D. X. Wang, and D. G.
Fang, Application of the SSOR preconditioned CG algorithm
to the vector FEM for 3-D full-wave analysis of electromag-
netic-eld boundary-value problems, IEEE Trans. Microwave
Theory Tech. 50(4):11651172 (April 2002).
30. R. S. Chen. D. X. Wang, and Edward K. N. Yung, Analysis of
ferrite waveguide devices by use of the edge FEM combined
with SOC technique. Int. J. Electron. 89(10):771790 (2002).
31. El-Shandwily, et al., General eld theory treatment of H-
plane waveguide junction circulators, IEEETrans. Microwave
Theory Tech. 21(6):392403 (June 1973).
32. N. Zeina, H. How, and C. Vittoria, Self-biasing circulators op-
erating at Ka-band utilizing M-type hexagonal ferrites, IEEE
Trans. Magn. 28(5):32193221 (Sept. 1992).
33. M. A. Tsankov, S. I. Ganchev, and L. G. Milenova,
Higher-order mode waveguide circulators for millimeter
wavelengths, IEEE Trans. Magn. 28(5):32283230 (Sept.
1992).
FERRITE ISOLATORS
B. BAYARD
B. SAUVIAC
D. VINCENT
Jean Monnet University
St. Etienne, France
1. ISOLATORS AND THEIR APPLICATIONS
The transmission of a radiofrequency or microwave signal
(above 1 GHz) along a line differs with that of a low-fre-
quency signal essentially in the fact that one of the di-
mensions of the line is greater than the wavelength. This
means that at a given time the instant value of the signal
varies along this dimension. Furthermore, the signal may
propagate along the line according to the opposite direc-
tions (forward and backward propagations).
The behavior of the wave is characterized by the map-
ping of the electric and magnetic elds. This mapping
depends on the geometry of the line as well as the con-
ductivity, the dielectric, and the magnetic constant of the
materials used in the structure. The characteristic imped-
ance is calculated from the mapping of the electric eld
and the magnetic eld; its value is often set at 50 O for
practical use.
If several kinds of lines are used to transmit a signal
from a source to a load, the electromagnetic eld map will
be different along each line. As a result, the signal will be
perturbed at each connection. The impedance mismatch
between two lines causes a return loss, that is, a reection
of a part of the signal toward the source and a weaker
wave transmitted to the load. Since this reection is un-
desirable and may be disruptive or even destructive, it has
to be eliminated by the use of an isolator.
Isolators are two-port circuits that allow the microwave
energy to propagate along a direction and stop it along the
opposite one (Fig. 1a). The ow is restricted to one direc-
tion; hence any reected energy at the load is trapped or
dissipated. Because the behavior of the isolator differs for
direct and reverse propagations, it is referred to as a non-
reciprocal device.
Operating frequencies above 10 GHz are currently the
higher limit for a correct functioning of active isolators or
circulators, not only because of the limited bandwidth of
semiconductors but also because of noise generation and
power dissipation. On the other hand, microwave ferrite
material can be regarded as a mature technology, and
their interesting anisotropic properties have been widely
exploited for passive nonreciprocal applications [1,2].
1.1. Applications
Isolators are intensively used in microwave communica-
tion systems. Their applications include the decoupling
between a generator and its load, the decoupling of several
ampliers, and the combining of two or more transmitters.
1.1.1. Decoupling between Generator and Load. Gener-
ators are usually affected by any power coming back to
FERRITE ISOLATORS 1473
them, which may result in nonlinear effects such as fre-
quency shift and instability. To overcome these problems,
an isolator is used to eliminate the coupling between the
generator and the load as shown in Fig. 1b, where the
isolator attenuates only the backward wave and protects
the generator.
1.1.2. Decoupling of Amplier Stages. In applications
where ampliers are connected in series, they can affect
each other in relation to their input impedance. The in-
uence increases if their working frequency band is nar-
row because their input impedance varies sharply. If
different amplier stages are decoupled by isolators as
shown in Fig. 1c, they can be tuned and adjusted without
affecting the others, and if one stage is unbalanced, the
others will not be overloaded.
1.1.3. Combination of Several Transmitters. A typical
emitter stage of a mobile phone base station includes
several transmitters working at different frequencies
(Fig. 1d). The signal of each transmitter passes through
an isolator, which has a low insertion loss, and a lter
centered on the transmitters frequency. It travels then to
the combiner. This lets the signal of each transmitter
travel to the antenna without disturbing the others.
1.2. Characteristics
A perfect isolator should transmit all the energy from port
1 to port 2, cut all the backward energy from port 2 to
port 1, and provide no reection at port 1 and at port 2
(Fig. 2a).
The corresponding S parameters can be expressed in
the following matrix:
S
11
S
12
S
21
S
22
_ _

0 0
1 0
_ _
1
Actual isolators response is not so ideal (Fig. 2b), and
their specication must take into account several funda-
mental parameters such as insertion loss, isolation,
and standing-wave ratio, as well as frequency band-
width, intermodulation, and thermal and power consider-
ations.
1.2.1. Insertion Loss. It corresponds to the neutrality
of the component in the forward direction and its magni-
tude must be close to the unity. When a signal is applied
to port 1, the insertion loss (IL) will be the ratio
of the input power (port 1) to the output power (port 2).
Typical values, expressed in decibels (dB), are between
0.1 and 1 dB and are calculated by the following
expression:
IL dB 20 log
10
a
1
b
2

20 log
10
jS
21
j\0 2
1.2.2. Isolation. This corresponds to the ability of cut-
ting the backward wave. Its magnitude must reach high
values. When a signal is applied to port 2, the isolation (IS)
will be the ratio of the input power (port 2) to the output
power (port 1). Typical values, expressed in dB, are
between 20 and 30 dB and are calculated by the following
Port 1 Port 2
(a) (b)
generator isolator load
mismatch
amplifier isolator amplifier
(c)
(d)
circulator
to receiver
antenna
amplifiers
c
o
m
b
i
n
e
r
isolators
narrow-band
filters
broad-band
filter
Figure 1. Some applications of isolators: (a) cir-
cuit symbol; (b) decoupling of generator and load;
(c) decoupling of ampliers; (d) combination of
transmitters.
1474 FERRITE ISOLATORS
expression:
IS dB 20 log
10
a
2
b
1

20 log
10
jS
12
jb0 3
1.2.3. SWR. The standing-wave ratio (SWR) species
how the input signal will be reected back toward the
source. It depends directly on the magnitude of return loss
(R) at port 1, that is, the ratio of the incident power to the
reected power. The SWR is always greater than one and
typical values are between 1.05 and 1.2, to maintain a
good impedance matching:
jRj
b
1
a
1

jS
11
j 0 4
SWR
1 jRj
1 jRj
\1 5
1.2.4. Bandwidth. The operating bandwidth can be ex-
pressed as the difference between the high and low work-
ing frequencies divided by the center frequency multiplied
by 100 (percentage bandwidth). Various bandwidth values
are available according to the technology of the isolator,
but 10% is often presented.
1.2.5. Intermodulation. When nonlinear elements such
as ferrite materials are used, harmonics (2f
1
, 3f
1
, etc.) ap-
pear but can be easily eliminated thanks to their high
frequency value. However, in the case of several but close
frequencies (f
1
and f
2
), intermodulation rays appear
(2f
1
f
2
and 2f
2
f
1
) and their inuence is qualied by a
third-order intermodulation product dened as the ratio of
the intermodulation magnitude to the fundamental mag-
nitude.
1.2.6. Temperature Range. The operating temperature
range of an isolator is limited by the materials used in the
device, especially by ferrites. Magnetic materials indeed
exhibit nonlinear behaviors and frequency shift due to
temperature variation and become nonmagnetic above
their Curie temperature. Operating temperatures from
20 to 701C are common.
1.2.7. Power Dissipation. For low-power isolators, ex-
ceeding the power limit causes nonlinearity effects in the
ferrite material and provides a frequency shift and an in-
crease in the insertion loss. For high-power isolators, av-
erage power induces an increase of temperature and
cooling of the device is sometimes needed. The peak pow-
er should cause breakdown or arcking, which generally
results in permanent degradation of performance. Power
depends on the technology and can be limited from a few
watts (planar dropin isolators) to several hundred watts
(waveguide isolators). Acceptable average power decreas-
es with the working frequency. Peak power can reach sev-
eral kilowatts.
2. DEVICES
The rst experimental microwave ferrite device was dem-
onstrated in 1949. The development of this type of device
was strongly linked to ferrite materials preparation and to
the knowledge of spin interaction in ferrimagnetic mate-
rials [35].
The behavior of microwave ferrite devices is based on
the gyromagnetic property of ferrimagnetic materials.
Several effects can be exploited to provide nonreciprocal
propagation of the signal, such as Faraday rotation, fer-
romagnetic resonance, and eld displacement.
Many microwave components such as circulators, iso-
lators, phase shifters, or gyromagnetic lters are using
ferrite materials and are always essential because there is
no alternative semiconductor-based device that satises
similar requirements.
Furthermore, microwave technologies are moving to
higher frequencies, up to 100GHz, where high-resistivity
materials are needed so that ferrites remain the rst
choice and semiconductor materials are currently not
competitive.
2.1. Waveguide Isolators
The microwave applications of ferrites had their founda-
tions in the Faraday effects, and the rst devices were cir-
cular waveguide Faraday rotators. Then, rectangular
waveguide components were developed using resonance
absorption or eld displacement [6].
2.1.1. Faraday Rotation Isolator. The linear polarization
of a wave rotates if it propagates in a longitudinally
magnetized ferrite. This phenomenon, called Faraday
S
12
S
22
S
21
S
11
Port 1
Insertion loss
Isolation
Return loss
and SWR
Port 2
b
1
b
2
a
1
a
2
(a)
50
45
40
35
30
25
20
15
10
5
0
7 8 9 10 11 12
Return loss
(S11)
Isolation
(S12)
Insertion loss
(S21)
Bandwidth
Frequency (GHz)
T
r
a
n
s
m
i
s
s
i
o
n

(
d
B
)
(b)
Figure 2. Characteristics of an isolator: (a) S parameters;
(b) actual frequency response.
FERRITE ISOLATORS 1475
rotation, is nonreciprocal so that the polarization always
rotates in the same direction whatever the propagation
direction. The Faraday rotation can be used in an isolator
if waves are absorbed or not, according to the rotation of
their polarization. The device is made from a circular
waveguide containing a 451 rotator between two 451 shift-
ed rectangular waveguides. Absorbing sheets are placed in
rectangular-to-round waveguide transitions (Fig. 3).
The polarization of the electric eld of a wave entering
port 1 is parallel to the smallest side of the rectangular
waveguide so that it is normal to the rst resistive sheet
and is not affected by it. The polarization is rotated 451
clockwise by the rotator and becomes normal to the second
resistive sheet. The wave can therefore emerge without
attenuation.
In the reverse direction, a wave entering port 2 is ro-
tated 451 in the same direction (clockwise) so that the
electric polarization is parallel to the port 1 resistive
sheet. The wave is then absorbed and cannot propagate.
The average power that can be absorbed is limited by
resistive sheets. On the other hand, insertion loss and iso-
lation are improved if the magnitude of circular polariza-
tions is not affected by the rotator so that no perpendicular
polarization appears before resistive sheets.
2.1.2. Resonance Isolator. A circularly polarized wave
that penetrates into a magnetized ferrite may be absorbed
or not, according to its rotation direction. The effect is
maximum at the gyromagnetic resonance frequency of the
ferrite. This phenomenon can be used in a rectangular
waveguide where the polarization is circular at two dif-
ferent positions (Fig. 4). If a ferrite is placed at one of these
positions, the nonreciprocal absorption of the microwave
energy is used to make an isolator [7]. A transverse DC
eld is required to magnetize the ferrite.
The optimum position of the ferrite is chosen for a cir-
cular polarization of the internal microwave magnetic
eld. This position may however differ from that of the
empty guide because of the ferrite presence. Several fer-
rite shapes can be used (E-plane or H-plane resonance
isolator), but H-plane ferrite geometry, consisting of two
thin at ferrite rods, shows better performance. It has the
particular advantage of setting the ferrite rods against the
waveguide with a maximum contact area so that the ma-
terial can be cooled through the metal waveguide.
2.1.3. Field Displacement Isolator. In a partially ferrite-
lled waveguide, the microwave energy may be rejected
from the material according to its direction of propagation.
As a result, the microwave electric eld vanishes near the
ferrite for direct propagation direction and is maximum
for the reverse direction (Fig. 5).
If a resistive sheet is placed against the ferrite slab
where the difference of energy is maximum, it will have
little effect on the direct propagation, but there will be
strong absorption in the backward propagation.
As for resonance isolator, a transverse DC magnetic
eld is applied to the ferrite, but a lower intensity is re-
quired. This kind of isolator gives low insertion loss about
0.1 dB. It is well suited for low-power operations because
the resistive sheet cannot be cooled as it is not in contact
with the waveguide.
2.2. Planar Isolators
The development of planar circuits, fabricated by conven-
tional printed-circuit techniques, has liberated the micro-
wave designer from high costs and many constraints
encountered with waveguides and coaxial lines [8,9]. Fer-
rite devices were rapidly developed for stripline and mi-
crostrip circuits such as edge-guided-mode isolators or Y-
junction circulators.
2.2.1. Coplanar Resonance Isolator. The coplanar wave-
guide is a surface strip transmission line and consists of a
central conductor strip separated from ground planes by
two slots on a dielectric substrate [10]. At high frequen-
cies, the propagation in this structure is no longer TEM
(transverse electromagnetic) because a longitudinal
F
o
r
w
a
r
d

p
r
o
p
o
g
a
t
io
n
B
a
c
k
w
a
r
d

p
r
o
p
o
g
a
t
io
n
4
5
F
a
r
a
d
a
y
r
o
t
a
t
o
r
R
e
s
is
tiv
e
s
h
e
e
t
R
e
s
is
tiv
e
s
h
e
e
t
Port 1
Port 2
Figure 3. Waveguide Faraday rotation isolator.
H-plane
E-plane
H
DC
Figure 4. Waveguide resonance isolator.
Forward
propogation
Reverse propogation
H
DC
Resistance sheet
Figure 5. Waveguide eld displacement isolator.
1476 FERRITE ISOLATORS
component of the microwave magnetic eld appears. Thus
the resulting elliptically polarized magnetic eld in the
slot can be used for nonreciprocal operations if it interacts
with a ferrite material. A resonant isolator can be fabri-
cated by attaching ferrite rods at the airdielectric inter-
face between the conductors (Fig. 6). A horizontal
magnetic polarization is applied to the ferrite, perpendic-
ular to the microwave elliptic polarization plane.
This device will show better performances if the micro-
wave polarization is circular. High insertion loss and mod-
erate isolation values are the consequences of an
ellipticity different from unity. However, since the domi-
nant propagation mode is TEM, the microwave polariza-
tion will never be circular but the ellipticity could be
improved under one of these conditions: a higher permit-
tivity substrate, wider strip and slots dimensions, or a
higher working frequency [11].
2.2.2. Stripline Resonance Isolator. TEM transmission
lines are not suitable for nonreciprocal devices because
there is no longitudinal component of the microwave mag-
netic eld. However, the line may be antisymmetrically
loaded with a dielectric material so that a longitudinal
eld appears according to Maxwells equations and bound-
ary conditions [6]. Ferrite rods are placed near the air
dielectric interface, where the longitudinal eld is higher.
A longitudinal magnetic eld is required to obtain an el-
liptic polarization such as that in the device described
above as it will interact with the magnetic rods to provide
nonreciprocal attenuation of the signal. Figure 7 shows
ferrite isolators based on a coaxial line and a stripline.
They are both half-lled with a dielectric material and
contain vertically magnetized ferrite rods. The nonrecip-
rocal effect is maximum at the ferrite resonance as in the
case of the coplanar isolator.
2.2.3. Coupled Microstrip Line Isolator. The coupling be-
tween two microstrip lines on a ferrite substrate is non-
reciprocal if the gyrotropic material is magnetized [12].
The nonreciprocal effect is a differential phase shift be-
tween S
12
and S
21
that can be used to create an isolator.
The coupling length is adjusted so that the direct signal at
one port is coupled to the other line and the reverse one is
not delivered to the rst line. External matching loads are
placed at the end of each line to absorb reected power
(Fig. 8).
2.2.4. Microstrip Edge-Guided-Mode Isolator. A micro-
strip or a stripline using a ferrite substrate, with a mag-
netization normal to the ground plane, with a wide strip
conductor, shows an exponential eld variation of the
dominant mode along the device width (Fig. 9). This eld
displacement effect is nonreciprocal as it is reversed for
each direction of wave propagation and for opposite mag-
netic polarizations [13]. The energy is concentrated near
opposite edges of the component and may be dissipated in
the reverse propagation direction by placing an absorbing
lm on one edge of the strip.
The eld displacement is caused by the gyromagnetic
properties of the ferrite, which is magnetized below the
resonance in the low-loss region. The device therefore
shows better performance with a low eld bias. On the
other hand, the use of a strip much wider than the sub-
strate thickness produces a low characteristic impedance
so that suitable impedance transformers are required for
approximate matching to 50-O-impedance lines.
Because the presence of the absorber should provide
insertion loss, several improvements were proposed. The
absorbing lm can be replaced by a short circuit of one
edge of the strip with the ground plane. The addition of an
iron plate near the short circuit provides a strong inter-
action with the magnetic eld that improves the nonre-
ciprocal effect [14].
2.2.5. Sawtooth Edge-Mode Isolator. Other authors sug-
gested printing the wide strip with a slots on its edge
without absorber [15]. The direct wave is not affected by
the slot discontinuity, while the reverse one is reected
by this effective open circuit and dissipates in the ferrite
substrate. This device is strictly planar and requires no
absorber.
c
r
H
DC
Ferrite rods
Dielectric substrate
Figure 6. Coplanar resonance isolator.
H
DC
H
DC
Ferrite
rods
c
r
c
r
Figure 7. Coaxial and stripline resonance isolator. (From [10]; r
1969 IEEE, reproduced with permission.)
H
DC
Ferrite substrate
Port 2
Port 1
Matching loads
Figure 8. Coupled microstrip-line isolator.
H
DC
H
DC
Absorber
Reverse
Port 2 Port 1
Forward
Ferrite
substrate
Impedance
Transformers
Figure 9. Microstrip edge-guided-mode isolator. (From [13]; r
1971 IEEE, reproduced with permission.)
FERRITE ISOLATORS 1477
To improve the bandwidth, a multislot (sawtooth) con-
guration is used and a lossy material covers only the saw-
tooth to reduce insertion loss (Fig. 10). The behavior of the
device can be tuned by varying the size of those slots; iso-
lation is increased with a large sawtooth design, while low
insertion loss is obtained with a small sawtooth design [16].
2.2.6. Slotline Field Displacement Isolator. It is also pos-
sible to obtain a nonreciprocal eld displacement in a fer-
rite-loaded slotline (Fig. 11). If a horizontal static
magnetic eld is applied, the energy travels along each
side of the ferrite according to the propagation direction
[17]. Thus the presence of an absorbing material at the
bottom of the ferrite provides the isolation. An alumina
substrate may be placed between the ferrite and the lines
to reduce the insertion losses of the device. The main
drawback of such a structure is the transition between the
slotline and a coaxial line (or other lines) at each port of
the device that provides mismatch losses. The same kind
of device can be built using a nline structure [18].
2.2.7. Circulator-Based Isolator. One of the most popu-
lar planar isolators proposed by manufacturers for
low-power applications is based on a three-port circulator
by adding a matching termination to one port. A usual
form consists of a stripline symmetric Y junction with a
circular center conductor surrounded with two ferrite
disks (Fig. 12). The circular junction behaves like a reso-
nant cavity with two contrarotating modes resulting in a
standing wave. If a static magnetic eld is applied per-
pendicularly to the disks, the resonant frequencies of the
two modes are different. The pattern of the standing wave
is therefore rotated so that the third port is isolated while
the others are coupled [19].
The easy fabrication and the good performance of the Y-
junction circulator make this miniature nonreciprocal de-
vice one of the most widely used, as a circulator as well as
an isolator.
3. NONRECIPROCAL EFFECTS
The functioning of passive microwave isolators is based on
ferromagnetic resonance properties of ferrite materials.
Magnetized ferrites are indeed anisotropic and character-
ized by an antisymmetric permeability tensor, which is
necessary to obtain nonreciprocal effects such as isolation.
Most passive magnetic devices need a saturated state of
the ferrite or at least a magnetic polarization that is per-
formed by permanent magnets or magnetic coils.
3.1. Ferromagnetic Resonance
To understand the origin of the anisotropy, the interaction
between magnetic moments of the ferrite and a magnetic
eld needs to be studied.
3.1.1. Magnetic Units. The current ferrite literature
still uses cgs units (centimeter, gram, second) for histor-
ical reasons. Therefore, conversions between this system
and SI units (meter, kilogram, second, ampere) frequently
have to be performed.
In the SI system, the following relations correspond to
the response of a material (M and B) immersed in an ap-
plied eld (H)
wHM
dm
dV
6
Bm
0
HM m
0
H1 w m
0
m
r
H 7
The ratio of the magnetization to the applied eld is
the susceptibility (6), and the ratio of the induction to
the eld is the permeability (7). The susceptibility van-
ishes and the relative permeability equals one for
nonmagnetic materials. In the case of anisotropic materi-
als, induction, magnetization, and applied eld are not
parallel and the susceptibility as well as the permeability
become tensorial.
H
DC
Port 2 Port 1
Optional
absorber
Slots
Figure 10. Sawtooth edge-mode isolator. (From [16]; r 2001
IEEE, reproduced with permission.)
Forward
Reverse
Energy
distribution
H
DC
Air
Alumina
Ferrite
Absorber
Figure 11. Slotline eld displacement isolator. (From [17]; r
1975 IEEE, reproduced with permission.)
Ferrite
H
DC
1
2
Ground
Ground
Conductor
Ferrite
Port 1 Port 2
Transmission
Isolation
Port 3
Matching termination
Figure 12. Circulator-based isolator. (From [24]; r 1965 IEEE,
reproduced with permission.)
1478 FERRITE ISOLATORS
In the cgs system, vacuum permeability is set at the
unity and (7) must be written as follows
BH4pM 8
and the conversion between cgs and SI units are [20]
1 A=m4p10
3
Oe magnetic field H 9
1 A=m10
3
emu=cm
3
magnetization M 10
1 T10
4
G induction B 11
3.1.2. Static Magnetic Field. When a static magnetic
eld (H
0
) is applied to a magnetic material, its magnetic
moments (m) and the corresponding magnetization (M)
for a given volume precess around the eld axis with an
angular velocity proportional to the magnitude of the eld
(Fig. 13a). This phenomenon, called Larmors precession,
is governed by the following relations, which give the gy-
romagnetic equation of motion, the angular velocity, and
the gyromagnetic ratio, respectively:
dm
dt
gm
0
mH
0
12
o
0
gm
0
H
0
13
g 176 10
9
rad s
1
T
1
,28 GHz=T 14
3.1.3. Microwave Field. If, in addition, the material is
submitted to a microwave eld (h) perpendicular to the
static one, the magnetic moments tend to precess around a
total eld that oscillates (Fig. 13b). When the angular ve-
locity of the moments reaches the frequency of microwave
oscillations, the microwave energy is transmitted to the
material and absorbed.
3.1.4. Polders Model. After taking into account the mi-
crowave eld, the spectral expression of the equation of
motion leads to a relative permeability tensor:
" mm
r

m
r
0 jk
0 1 0
jk 0 m
r
_
_
_
_
_
_
_
_
15
This tensor is used for a y-axis applied eld (xy, yx, yz, and
zy elements vanish) and for a saturated material (yy ele-
ment is equal to 1). The presence of the imaginary number
j corresponds to the rotation of the moments; that is, there
is a 901 phase difference between m
x
and m
z
.
The elements of the permeability tensor depend on the
frequency
m
r
1
o
0
o
M
o
2
0
o
2
16
k
oo
M
o
2
0
o
2
17
o
M
gm
0
M
S
18
and they both depend on the magnitude of the applied
eld (13) and on the saturation magnetization of the ma-
terial (18).
3.1.5. Damping Factor. If losses are taken into account,
the angle between m and H
0
will decrease until these el-
ements are aligned (Fig. 13c). Tensor elements become
complex quantities, and a damping parameter is intro-
duced in (16) and (17):
m
r
m
0
r
jm
00
r
1
o
0
jao o
M
o
0
jao
2
o
2
19
k k
0
jk
00

oo
M
o
0
jao
2
o
2
20
The real and imaginary parts of the elements of the
permeability tensor are shown in Fig. 14 for a 180mT sat-
uration magnetization and 0.1 damping factor. Figure 14a
shows a frequency sweep with a xed applied eld, which
is currently the most common representation, thanks to
vector analyzer measurements facilities. For historical
reasons, Fig. 14b is the most widely known representa-
tion because it was previously easier to work with a single-
frequency microwave source and to vary the applied eld
by adjusting a continuous current.
According to Fig. 14b, the resonance losses are charac-
terized by the width of the imaginary part, which increas-
es for lossy materials. The width at midheight of the curve
is called linewidth and is frequently used by manufactur-
ers (often expressed in oersteds). It depends on the dam-
ping factor and on the working frequency:
gm
0
DH2ao 21
Figure 14b also shows three different zones according to
the magnitude of the applied eld. The expressions below,
above, or at resonance refer to a eld sweep and not to a
frequency sweep. The confusion could be reinforced if the
static eld were expressed as a frequency as in (13).
Polders model is dened for saturated material so that
the magnetization of the ferrite reaches the saturation
magnetization. Under particular operation conditions
such as low bias elds, several magnetic domains with
different magnetization magnitudes and orientations are
m
(a) (b) (c)
H
0
m
h
H
0
H
0
+h
m
H
0
o
o
0
o
0
o
0
Figure 13. Precession of ferrite moments under an applied eld:
(a) static applied eld; (b) static and microwave applied elds; (c)
with losses.
FERRITE ISOLATORS 1479
present. Therefore new models must be used to take into
account the resulting effects on the gyromagnetic phenom-
enon [21].
3.1.6. Demagnetizing Factors. All the previous deni-
tions concern materials inuenced by an internal magnet-
ic eld, that is, the actual eld experienced by the spin
dipoles. In a innite medium, the internal eld equals the
applied eld, but in the case of a nite medium, there is a
discontinuity of the magnetic eld at the boundary be-
tween materials with different permeability values. De-
termining the eld inside an arbitrarily shaped sample is
not always possible, but the problem is tractable for an
approximate ellipsoidal shape. Thus, the internal eld can
be expressed from the applied eld as follows
H
i
H
0
NM 22
N
N
x
0 0
0 N
y
0
0 0 N
z
_
_
_
_
_
_
_
_
23
where the demagnetizing factors, N
x
, N
y
, and N
z
, depend
on the shape of the sample. The sum of these terms are
equal to one. Each term equals
1
3
in the case of a sphere;
the term tends to zero if the corresponding size increases
and to one if it becomes thinner. Several approximations
have been calculated and are helpful in estimating inter-
nal elds for arbitrary shapes [22].
Since the demagnetizing factors concern static and mi-
crowave elds, the tensor elements are dramatically
changed and the resonant frequency is given by Kittels
equation for a z-direction applied eld:
o
r
gm
0

H
0
N
x
N
z
MH
0
N
y
N
z
M
_
24
3.2. Nonreciprocal Wave Propagation
Because waveguide or planar congurations are difcult
to solve, nonreciprocal effects in ferrites can be more ele-
mentarily understood by considering the microwave prop-
agation in innite media. Maxwells equations can be
combined to include the permeability tensor and, consid-
ering a propagation along the z axis, it gives
rrHo
2
e
0
m
0
e
r
" mm
r
H0 25
Hh expjot expgz 26
where
g a jbj
o
c

e
e
m
e
p
27
is the propagation coefcient that represents both the at-
tenuation and the phase delay of the signal. In an innite
medium, they depend on the effective permittivity and
permeability of the material.
According to the direction of the static magnetic eld,
which lls the permeability tensor (15) in different ways,
several phenomena on the effective permeability and on
the propagation occur.
3.2.1. Longitudinal Field. If a static magnetic eld is
applied parallel to the propagation direction (z axis), then
(25) leads to an eigenproblem where the eigenvalues are
m

m
r
k 28
m

m
r
k 29
These associated eigenvectors are respectively a right-
hand circular polarized magnetic eld for (28) and a left-
hand polarization for (29). This means that the microwave
must be decomposed according to these two polarizations
and each part propagates with the corresponding effective
permeability.
The effective permeability values, obtained from Fig. 14b,
are plotted versus the applied eld in Fig. 15a. In an in-
nite medium, the propagation coefcients of the corre-
sponding plane wave are given by
g

j
o
c

e
r
m

p
j
o
c

e
r
m
r
k
_
30
3
2
1
0
1
2
3
4
5
2 4 6 8 10
Frequency (GHz)
(a)

(b)
3
2
1
0
1
2
3
4
5
5 0 100 150 200
H0(kA/m)
250 300
H

Below
Above
resonance
resonance
Figure 14. Ferromagnetic resonance; real and imaginary parts
of the elements of the permeability tensor: (a) frequency sweep
and 150kA/m applied eld; (b) eld magnitude sweep and 5 GHz
frequency.
1480 FERRITE ISOLATORS
and are plotted in Fig. 15b. Since real parts and imaginary
parts are all different, nonreciprocal effects can be ob-
tained not only for absorption but also for phase.
3.2.2. Transverse Field. If the permeability tensor is
lled according to a perpendicular eld (x or y axis), prop-
agation will occur with two effective permeability values
m


m
2
r
k
2
m
r
31
m

1 32
The eigen vectors are now linear polarizations. Perme-
ability values and propagation coefcients are plotted in
Fig. 16. Nonreciprocal effects are also possible.
Although the explanations above are dened for plane
waves (without longitudinal eld components), they can
be generalized for cases of propagation in waveguides or
planar lines where longitudinal elds (non-TEM modes)
appear.
3.2.3. Faraday Rotation. Figure 15b shows different
values of the phase delay. If a linearly polarized incident
microwave eld is decomposed in both right-hand and left-
hand circular polarizations, each part will be differently
phase-shifted for a given distance (Fig. 17). Then the com-
bination of the two emerging circular polarizations (with
the same magnitude) gives a linear polarization with a
rotation equal to the half phase difference:
y
Df
2

b

2
L 33
The rotation is nonreciprocal. The polarizations of back-
ward and forward waves rotate in opposite directions (re-
spectively to the left or to the right) according to their own
coordinate system because the applied eld is inverted. As
a result, rotations are in the same direction according to
the ferrite axes.
Faraday rotation should be performed for low eld
values where losses are weak to ensure that the circular
magnitudes are not attenuated and the emerging
6
4
2
0
2
4
6
8
10
12
50 100 150 200 250 300
H0(kA/m)
(a)

j+

j+

H0(kA/m)
(b)
0
1
2
3
4
5
6
7
8
9
10
0 50 100 150 200 250 300
[
[+
:
:+
Figure 15. Microwave propagation parallel to the applied eld:
(a) effective permeability values; (b) nonreciprocal propagation
coefcients.
4
2
0
2
4
6
8
50 100 150 200 250
H0(kA/m)
(a)
300
j+

j+

H0(kA/m)
(b)
0
1
2
3
4
5
6
7
8
9
0 50 100 150 200 250 300


[
[+
:
:+
Figure 16. Microwave propagation perpendicular to the applied
eld: (a) effective permeability values; (b) nonreciprocal propaga-
tion coefcients.
incident wave
emerging
L
wave
H
0
a
p
p
lie
d
f
ie
ld
f
e
r
r
it
e

L
+
=
+
L
= 0

= 0
Figure 17. Faraday rotation of a linearly polarized wave.
FERRITE ISOLATORS 1481
polarization remains linear. The nonreciprocal rotation
effect, similar to the behavior of Faraday rotation in opti-
cal media, was the rst phenomenon exploited in micro-
wave ferrite devices and is widely used to build circulators
or isolators as shown in Fig. 3.
3.2.4. Resonance Absorption. Figures 15b and 16b both
show nonreciprocal absorption according to the polariza-
tion of the wave. Losses are added in (28) and vanish in
(29), where the imaginary part of the effective permeabil-
ity is nearly zero regardless of the applied eld. The max-
imum nonreciprocal attenuation occurs at the resonance
where phases are equal.
In the rst conguration, a left-hand circular polariza-
tion is absorbed while a right-hand polarization can prop-
agate. If the polarization of the incident wave is elliptic,
the attenuation depends on whether the magnitude of
the right-hand circular polarization is maximum or not
(Fig. 18).
In the second conguration, a linearly polarized micro-
wave magnetic eld parallel to the static eld does not
interact with the ferrite and the wave propagates in the
same way as in a dielectric material. On the other hand, a
polarization perpendicular to the applied eld is strongly
absorbed.
The resonance absorption is widely used in devices
where a longitudinal magnetic eld exists. If longitudinal
and transverse elds are 901 phase-shifted, then they form
an elliptic polarization that can interact with the rotation
of the moments in a ferrite. The material must be mag-
netized perpendicular to the plane where the microwave
polarization and the magnetic moments rotate. If longitu-
dinal and transverse magnitudes are equal, the polariza-
tion becomes circular and effects are maximum. This
conguration is used in rectangular waveguide resonance
isolators (Fig. 4) and in coplanar and stripline isolators
(Figs. 6 and 7).
3.2.5. Transverse Field Displacement. The negative
value of the effective permeability, in Figs. 15b and 16b,
implies that the microwave eld is rejected from the ma-
terial, while the other permeability allows the wave to
penetrate. If a waveguide is partially lled with a ferrite
material and if the case of a negative permeability occurs,
the energy therefore propagates in the empty region. This
phenomenon is a form of eld displacement and is used in
the rectangular waveguide isolator described in Fig. 5.
Another conguration where eld displacement can be
observed is the case of two wide strips separated by a fer-
rite substrate. In this type of structure, only TE modes are
considered because TM modes cannot exist between par-
allel strips when the spacing is less than a half-wave-
length [13].
The eld components can be expressed as follows
E
x
; H
y
; H
z
A
x;y;z
expa
y
y expjb
z
z 34
where the propagation coefcients depend on the elements
of the permeability tensor:
b
z
o

m
0
e
0
e
r
m
r
p
35
a
y
o
k
m
r

m
0
e
0
e
r
m
r
p
36
The corresponding wave can therefore propagate along
the z axis (purely imaginary coefcient in the propagation
direction) while it is attenuated along the y axis (purely
real coefcient along the width), so that the eld magni-
tudes decrease exponentially along the width. This is the
edge-guided mode (Fig. 19). The phenomenon is more ef-
cient if the material is magnetized below the resonance
with a weak internal eld to avoid losses, just enough for
the saturation.
If the applied eld is inverted, the sign of the attenu-
ation is changed and the exponential decay is inverted.
a=10
b = 5
A
left
= 0
A
right
= 7.5
A
right
= 7.5
A
right
= 7.5
A
left
= 2.5
A
left
= 2.5
A
left
= 2.5
A
right
= 0
strong
absorption
weak
absorption
imperfect transmission
insertion loss
a = 10
b = 5
strong
absorption
weak
absorption
imperfect isolation
Figure 18. Resonance absorption of left-hand and right-hand polarizations.
1482 FERRITE ISOLATORS
The same changes occur for a reverse propagation.
The eld displacement is therefore nonreciprocal and
can be used to build a stripline or a microstrip isolator
by adding an absorber on one edge as shown in Fig. 9. For
high frequencies, other modes of propagation with a sinu-
soidal pattern can exist. As a result, the eld displace-
ment, corresponding to the edge-guided mode, is
frequency-limited.
3.2.6. Ferrite Junction Circulation. A ferrite disk placed
between two conducting planes is a geometry of planar
junction circulator. Electric elds are supposedly mainly
perpendicular to the conductors and the boundary condi-
tions lead to two different congurations of electromag-
netic waves: right-handed and left-handed rotating waves.
A standing-wave pattern is established where elds in-
tensities are equal and oppositely directed on either side of
the conductor (Fig. 20).
For an isotropic substrate (or unmagnetized ferrite),
the corresponding effective indices where the rotating
waves propagate are equal. The standing wave intensity
is maximum at the input port (port 1), and vanishes at 901
from the input port. As a consequence, a small amount of
energy is coupled to ports 2 and 3.
For a gyrotropic substrate (magnetized ferrite), the two
rotating modes are oppositely phase-shifted so that a ro-
tation of the electromagnetic pattern occurs. If the rota-
tion reaches 301, port 2 is then coupled (transmission) and
port 3 is isolated [23].
In this case, the effective permeability is given by
m
e

m
2
r
k
2
m
r
37
and the solutions of the electromagnetic equations give
magnetic microwave elds [24] proportional to
A

J
n1
kr
J
n
kr
kr
1
k
m
_ _
38
A

J
n1
kr
J
n
kr
kr
1
k
m
_ _
39
where J
n
are nth-order Bessel functions, and where
k
2
o
2
e
0
m
0
e
r
m
e
40
is the wavenumber associated with the effective perme-
ability and r is the distance to the center of the ferrite disk.
The rst modes (n1) show two opposite areas of maxi-
mum energy as in Fig. 20. The positive and negative ro-
tating modes are strongly linked to the ratio of the terms
of the ferrite permeability tensor and differ in sign.
Magnetic elds vanish at the boundaries (magnetic
wall if r equals the radius). Magnitudes (38) and (39) are
then equal to zero, and their solutions lead to the radius of
the ferrite disk for given working frequency and perme-
ability tensor.
The input impedance of each port can be adjusted with
the thickness of the disk, and the width of the access strip
and can be improved by an impedance transformer.
The best performance is obtained above the resonance
(according to the applied eld, that is, at low frequencies),
but the circulator can also operate below resonance [25].
The rotation is then inverted because the sign of the ratio
is changed.
One of the most popular structure is the stripline Y-
junction circulator with two ferrite disks (Fig. 12).
3.3. Ferrite Materials
Microwave ferrites are high-resistivity ferrimagnetic ma-
terials used at frequencies above 0.1GHz. They are ferri-
magnetic oxides made of oxygen anions around metal
(such as iron) divalent or trivalent cations [26]. The pos-
sibility of mixing the composition to adjust their behavior
is one very attractive property of ferrites. If a divalent iron
cation is replaced by any divalent composition, it gives a
mixed ferrite with sometimes completely different proper-
ties. In addition, the substitution of a trivalent iron cation
H
DC
Ferrite
Edge-guided mode Higher mode
Figure 19. Field displacement and edge-guided mode. (From
[13]; r 1971 IEEE, reproduced with permission.)
Port 1
Input Input
Port 3
Port 2
H
DC
= 0
Port 1
Port 3
Isolation
30
Transmission
Magnetized ferrite Unmagnetized ferrite
Port 2
H
DC
Figure 20. Ferrite junction circulation. (From
[24]; r 1965 IEEE, reproduced with permis-
sion.)
FERRITE ISOLATORS 1483
gives a substituted ferrite whose behavior can be thinly
adjusted.
3.3.1. Properties. Although ferrites usually show weak-
er performance than do magnetic metals, they are highly
resistive, so that an electromagnetic wave can penetrate
and interact with the material. Their resistivity can in-
deed vary between 10 and 10
8
O cm, a value that must be
compared to the 10
4
O cm resistivity of ferromagnetic
metals. The relative dielectric permittivity of ferrites is
almost constant at microwave frequencies and close to 15.
Dielectric losses result mainly from the existence of both
trivalent and divalent iron cations. There is an excess of
electrons that may jump from one iron cation to another
and therefore cause some conduction and losses.
A fundamental property such as permeability has to be
considered because it governs the interaction between the
wave and the material. Low coercivity is a condition for
low losses. High Curie temperature and low magneto-
striction constant are the main factors in improving the
stability of ferrites. High saturation magnetization implies
high gyromagnetic anisotropy but also intense magnetic
eld to saturate the material.
3.3.2. Spinels. Historically, spinel ferrites were the rst
type of ferrite used in microwave devices, in the 330 GHz
range. The typical composition of a spinel ferrite is as
follows:
Fe
2
O
2

.
Fe
3
2
O
2
3
Fe
3
O
4
41
Nickel mixed ferrite has a Curie temperature of about
5701C and is therefore rather stable. Its saturation mag-
netization is about 320mT. Because it contains Ni ion,
which is a relaxing ion, the spinel ferrite is a lossy mate-
rial reserved for power applications. Lower magnetiza-
tions (down to 140mT) are obtained by aluminium
substitution for iron, while partial substitution of zinc
for nickel increases the saturation magnetization
(480 mT) at a lower temperature (4001C).
Lithium ferrite has a magnetization of about 360mT
with a high Curie temperature near 6451C. The lack of
divalent iron ions results in a narrow linewidth that ren-
ders this ferrite suitable for low power levels with low
losses. Magnetization can be decreased (down to 230mT)
by Ti substitution and increased by Zn substitution
(500 mT).
3.3.3. Garnets. Yttrium iron Garnet (YIG) is a very-
narrow-linewidth material (near 3kA/m, compared to 15
35 kA/m of spinels) that remains the best microwave ma-
terial in the 110-GHz band. It can easily be saturated
thanks to its weak saturation magnetization (175 mT). As
the Curie temperature of YIG is 2861C, the stability with
temperature is low. The chemical formula is
3
2
Y
3
2
O
2
3

.
5
2
Fe
3
2
O
2
3
Y
3
Fe
5
O
12
42
Aluminum substitution of iron ions decreases the magne-
tization as well as the Curie temperature, while Gd sub-
stitution of Y ions reduces the magnetization without
changing this temperature, which results in a better sta-
bility. Narrow linewidth as low as 0.8kA/m can be ob-
tained with In or Zn substitutions (associated with Ca
ions).
3.3.4. Hexaferrites. These materials have an hexagonal
structure with a c axis of symmetry. Their crystalline
structure is essentially uniaxial of the M-type with the
following formula
Ba
2
O
2

.
6Fe
3
2
O
2
3
BaFe
12
O
19
43
for barium (BaM) or strontium (SrM) hexaferrites. They
have a high saturation magnetization (480 mT) and a Cu-
rie temperature as high as 7001C.
They also have a high magnetocrystalline anisotropy,
which results in an anisotropy eld H
a
(mainly near
1500 kA/m and up to 2700 kA/m) that can be added to
the internal eld to reach resonance frequencies higher
than (24)
o
r
gm
0

H
0
H
a
N
x
N
z
MH
0
H
a
N
y
N
z
M
_
44
They are also characterized by a squared hysteresis loop
with high remanence that allows a self-polarization of the
material.
4. CURRENT DEVELOPMENTS
The manufacturing of ferrite materials is difcult, espe-
cially for small dimensions. These constraints do not make
this kind of components easy to integrate or to miniatur-
ize. Several ways of improving the performance and com-
pactness of ferrite devices are proposed [27,28].
In order to reduce the size of the devices, some authors
propose using hard magnetic materials with high reman-
ence so that no external eld is needed. Hexagonal ferrites
such as SrM (315 mT saturation magnetization and
1500 kA/m anisotropy eld) were successfully used in a
33-GHz rectangular waveguide circulator [29]. A 30-GHz
self-biased microstrip circulator using a 130-mm-thick
SrM substrate were realized [30]. Another interesting
property of this material is its high crystalline anisotro-
py, which can be added to the internal eld so that the
resonance occurs at higher frequencies, above 10 GHz and
up to 90 GHz.
In the case of thin ferrite plates, ohmic losses caused by
metal resistivity may become the main contribution to to-
tal loss. To overcome this effect, the integration of super-
conductive materials in ferrite microwave devices has
been investigated and its feasibility has been demonstrat-
ed in the case of a 10-GHz ring network circulator made of
niobium or YBaCuO superconductors cooled to 77 and 4K
[31]. The same criteria are considered for the choice of
ferrite material at either cryogenic temperature or at
room temperature with, however, higher saturation mag-
netization values.
1484 FERRITE ISOLATORS
The main issue of current research is the integration of
ferrite devices with semiconductor technologies. Several
attempts at depositing ferrite lms on silicon substrates
were performed using pulsed laser deposition, evapora-
tion, or sputtering [32]. Since the deposited lm is amor-
phous and nonmagnetic, a thermal process is needed to
recover ferrimagnetic properties. High temperature val-
ues above 6001C are required, either during the deposition
process or during a postdeposition annealing, in order to
crystalize the material.
It may be possible to use a semiconductor to make a
millimeter-wave isolator or circulator instead of a ferrite
because its permittivity becomes tensorial if a static mag-
netic eld is applied. This phenomenon is the dual of per-
meability tensor of ferrites. A 60-GHz circulator was
demonstrated experimentally using an n-type indium
antimonide (InSb) disk at an operating temperature of
77 K [33].
Isolator structures can also be used as a microwave
measurement technique for determination of the complex
permeability tensor components of magnetized materials.
A microstrip measurement cell based on Fig. 19 was real-
ized and provided measurements up to 6 GHz [34]. The
central zone between the strip and lower ground plane is
lled with the ferrite, and two different dielectric materi-
als are placed on each side to provide a nonreciprocal effect
necessary to extract all the tensor components.
BIBLIOGRAPHY
1. K. Chang, ed., Handbook of Microwave and Optical Compo-
nents, Vol. 1, Wiley, New York, 1989.
2. J. D. Adam, E. L. Davis, G. F. Dionne, E. F. Schloemann, and
S. N. Stitzer, Ferrite devices and materials, IEEE Trans. Mi-
crowave Theory Tech. 50(3):721737 (March 2002).
3. K. J. Button, Microwave ferrite devices: The rst ten years,
IEEE Trans. Microwave Theory Tech. 32(9):10881096 (Sept.
1984).
4. R. H. Knerr, An annotated bibliography of microwave circu-
lators and isolators: 19681975, IEEE Trans. Microwave The-
ory Tech. 23(10):818825 (Oct. 1975).
5. G. P. Rodrigue, A generation of microwave ferrite devices,
Proc. IEEE 76(2):121137 (Feb. 1988).
6. B. Lax and K. J. Button, Microwave Ferrites and Ferromag-
netics, McGraw-Hill, New York, 1962.
7. F. E. Gardiol, Anisotropic slabs in rectangular waveguides,
IEEE Trans. Microwave Theory Tech. 8(18):461467 (1970).
8. K. C. Gupta, R. Garg, I. J. Bahl, and P. Bhartia, Microstrip
Lines and Slotlines, 2nd ed., Artech House, Boston, 1996.
9. R. M. Barrett, Microwave printed circuitsthe early years,
IEEE Trans. Microwave Theory Tech. 32(9):983990 (Sept.
1984).
10. C. P. Wen, Coplanar waveguide: A surface strip transmission
line suitable for non-reciprocal gyromagnetic device applica-
tion, IEEE Trans. Microwave Theory Tech. 17(2):10871090
(Dec. 1969).
11. B. Bayard, D. Vincent, C. R. Simovski, and G. Noyel, Electro-
magnetic study of a ferrite coplanar isolator suitable for in-
tegration, IEEE Trans. Microwave Theory Tech. 51(7):1809
1814 (July 2003).
12. Y. C. Moon, J. R. Lee, S. W. Yun, and I. S. Chang, A broadband
planar isolator using coupled microstrip lines on a magne-
tized gyrotropic substrate, Microwave J. 44(11):90103 (Nov.
2001).
13. M. E. Hines, Reciprocal and nonreciprocal modes of propaga-
tion in ferrite stripline and microstrip devices, IEEE Trans.
Microwave Theory Tech. 19(5):442451 (May 1971).
14. T. Nuguchi, New edge-guided mode isolator using ferromag-
netic resonance absorption, IEEE Trans. Microwave Theory
Tech. 25(2):100106 (Feb. 1977).
15. R. C. Kane and T. Wong, An edge-guided mode microstrip
isolator with transverse slot discontinuity, IEEE MTT-S Di-
gest, 1990, pp. 10071010.
16. A. H. Aly and E. B. El-Sharawy, Performance and modeling of
saw tooth edge mode isolators, IEEE MTT-S Digest, 2001, pp.
475477.
17. L. Courtois and M. de Vecchis, A new class of nonreciprocal
components using slotlines, IEEE Trans. Microwave Theory
Tech. 23:511516 (June 1975).
18. A. Beyer and K. Solbach, A new n-line isolator for integrated
millimeter-wave circuits, IEEE Trans. Microwave Theory
Tech. 29(12):13441348 (Dec. 1981).
19. D. K. Linkhart, Microwave Circulator Design, Artech House,
1989.
20. J. D. Huba, NRL Plasma Formulary, rev. ed., Naval Research
Laboratory, Washington, DC, 2002.
21. Ph. Gelin and K. Berthou-Pichavant, New consistent model
for ferrite permeability tensor with arbitrary magnetization
state, IEEE Trans. Microwave Theory Tech. 45(8):11851192
(Aug. 1997).
22. M. Sato and Y. Ishii, Simple and approximate expressions of
demagnetizing factors of uniformly magnetized rectangular
rod and cylinder, J. Appl. Phys. 66(2):983985 (July 1989).
23. H. Bosma, On the principle of stripline circulation, Proc. IEE
109:137146 (Jan. 1962).
24. C. E. Fay and R. L. Comstock, Operation of the ferrite junc-
tion circulator, IEEE Trans. Microwave Theory Tech. 13:1527
(Jan. 1965).
25. E. Schloemann and R. E. Blight, Broad-band stripline circu-
lators based on yig and li-ferrite single crystals, IEEE Trans.
Microwave Theory Tech. 34(12):13941400 (Dec. 1986).
26. J. Nicolas, Ferromagnetic Materials, North-Holland, Amster-
dam, 1980, Vol. 2, Chap. 4, pp. 243296.
27. E. Schloemann, Advances in ferrite microwave materials and
devices, J. Magn. Magn. Mater. 209:1520 (2000).
28. M. Pardavi-Horvath, Microwave applications of soft ferrites,
J. Magn. Magn. Mater. 215216:171183 (2000).
29. M. A. Tsankov and L. G. Milenova, Design of self-biased
hexaferrite waveguide circulators, J. Appl. Phys. 73(10):
70187020 (May 1993).
30. S. A. Oliver, P. Shi, W. Hu, H. How, S. W. McKnight,
N. E. McGruer, P. M. Zavracky, and C. Vittoria, Integrated
self-biased hexaferrite microstrip circulators for millimeter-
wavelength applications, IEEE Trans. Microwave Theory
Tech. 49(2):385387 (Feb. 2001).
31. G. F. Dionne, D. E. Oates, D. H. Temme, and J. A. Weiss, Fer-
rite-superconductor devices for advanced microwave applica-
tions, IEEE Trans. Microwave Theory Tech. 44(7):13611368
(July 1996).
32. B. Bayard, J. P. Chatelon, M. Le Berre, H. Joisten, J. J. Rous-
seau, and D. Barbier, The effects of deposition and annealing
conditions on crystallographic properties of sputtered barium
ferrite thick lms, Sensors Actuators A 99:207212 (2002).
FERRITE ISOLATORS 1485
33. Z. M. Ng, L. E. Davis, and R. Sloan, Coplanar waveguide gy-
roelectric circulator, Int. J. RF Microwave Comput. Aided
Eng. 12(4):367374 (July 2002).
34. P. Queffelec, S. Mallegol, and M. Le Floch, Automatic mea-
surement of complex tensorial permeability of magnetized
materials in a wide microwave frequency range, IEEE Trans.
Microwave Theory Tech. 50(9):21282134 (Sept. 2002).
FERRITE-LOADED WAVEGUIDES
RICARDO MARQUE

S
University of Seville
Seville, Spain
Ferrites and their technological applications have been
known for a long time. Magnetite, the rst known mag-
netic material, is actually a ferrous ferrite. In 1269 Peter
Peregrinus gave a detailed description of a compass made
with a oating magnetite needle, probably a Chinese in-
vention. The Spanish word for the compass, brujula,
which literally means small witch, clearly shows former
navigators amazement with the mysterious magnetic
properties of the lodestone. The rst experimental ferrite
device in microwave technology was demonstrated in 1949
[1]. Since then, applications of articial ferrite materials
in microwave technology have grown rapidly and have
become a mature technology, which has been discussed in
many classical textbooks [210]. A good historical survey
of the beginnings of the microwave ferrite technology can
be found in Button [1]. A complete bibliography containing
the most relevant contributions in this eld during the
years that followed can be found in Refs. 11 and 12.
Finally, an authorized survey of the most recent advanc-
es in microwave ferrite technology in Europe, the United
States, and Japan is found in Refs. 13, 14, and 15, respec-
tively.
This article describes the main physical effects due to
the propagation and guidance of electromagnetic waves in
ferrite-loaded waveguides useful in microwave technology.
The linear approach, in which the high-frequency mag-
netic susceptibility of the ferrite is a function of the inter-
nal static magnetic eld, will be considered valid. This
approach includes the analysis of exchange free electro-
magnetic waves, as well as magnetostatic waves and other
approximations, but not the analysis of spin waves and
nonlinear effects, magnetoelastic waves, and other com-
plex interactions.
The choice of units in the analysis of microwave appli-
cations of ferrites presents some particularities. SI (Inter-
national System) units are preferred for most of the
electronic and electrical engineers. Nevertheless, cgs (cen-
timetergramsecond) units are used mainly by research-
ers in the area of the constitutive electromagnetic
properties of ferrites. The usage in this text is a compro-
mise between both alternatives. Since in the linear theory
the equations for the static biasing magnetic eld are
decoupled from the radiofrequency eld equations, we will
use cgs units in the derivation of the internal magneto-
static eld, as well as in the expression of the magnetic
permeability in terms of the static bias eld H
0
, measured
in oersteds, and the saturation magnetization 4pM
0
, with
M
0
measured in gauss (1Oe 4pG). For the electromag-
netic RF equations we will use SI units.
1. FUNDAMENTALS OF ELECTRODYNAMICS OF FERRITE
MATERIALS
As long as the linear approach remains valid, the problem
of nding the radiofrequency electromagnetic eld inside a
magnetized ferrite can be divided into three steps. First
we must nd the internal static eld H
0
as a function of
the external applied static eld H
ext
. Then we must obtain
the RF magnetic permeability tensor that will be a func-
tion of the internal static eld. Finally, we must solve the
Maxwell equations for the RF eld with the appropriate
boundary conditions. Note that, in the linear approach,
the equations for the static eld H
0
remains independent
from the equations for the RF eld. The coupling between
these two elds occurs only by means of the dependence of
the RF magnetic permeability tensor on the static mag-
netic eld.
1.1. The Static Field
As a general statement, the internal static eld H
0
is the
solution of the static equations inside the ferrite with the
appropriate constitutive relations and boundary condi-
tions. In the simplest case of a saturated isotropic ferrite,
the static constitutive relations reduces to B
0
H
0
4pM
s
(remember that we will use cgs units in this part of the
analysis), where M
s
is the magnetization of the ferrite at
saturation, which, in isotropic ferrites, will be parallel to
both B
0
and H
0
. In many cases, the ferrite is placed in a
known external static and uniform eld H
ext
provided by a
magnet. In this case the internal static eld is the sum of
the external eld and a demagnetization eld H
d
created
by the ferrite internal magnetization. For ellipsoidal fer-
rite samples, rods and plates this problem is a classical
one and is solved analytically, expressing the demagneti-
zation eld as the dot product of the saturation magneti-
zation by a known demagnetization tensor, which depends
on the shape and the orientation of the ferrite sample. In
particular, for ferrite plates and rods placed in an external
magnetic eld parallel to the rod axis or the plane of the
plate, it is easy to show that H
0
H
ext
. For ferrite plates
placed in an external magnetic eld perpendicular to the
plate, it is H
0
H
ext
4pM
s
.
1.2. The RF Magnetic Permeability Tensor
In this section we will state the more usual RF cons-
titutive relationships for magnetized ferrites. First we will
consider the simplest case of an intrinsically isotropic fer-
rite. In this context intrinsic isotropy means that the in-
ternal static magnetic eld is the only source of anisotropy
(induced anisotropy). Thus, after magnetization, the fer-
rite becomes an uniaxial medium with an RF permeability
tensor given by (the z axis is chosen as the direction of
1486 FERRITE-LOADED WAVEGUIDES
internal magnetization)
m m
0
m
t
0
0 m
z
_ _
m
0
m jk 0
jk m 0
0 0 m
z
_
_
_
_
_
_
_
_
1
where m and k are in general complex quantities that
depend on the internal magnetic eld and ferrite magne-
tization. This particular form of both [m] and [m]
t
ensures
the invariance of the permeability tensor after rotations
around the z axis, which is the unique symmetry require-
ment. Notice that, if ka0, the tensor is not symmetric.
Thus, a magnetized ferrite is a nonreciprocal medium.
For lossless ferrites the permeability tensor must be
Hermitian, and therefore m, k, and m
z
are real numbers.
The theory leading to appropriate expressions for m, k,
and m
z
, valid for intrinsically isotropic saturated ferrites
was rst developed by Polder in 1949 from an analysis of
the precession of molecular magnetic dipoles in the static
internal eld. The derivation of such expressions may be
found in many textbooks [27]. The nal expressions are
given here for completeness
m 1
o
M
o
H
o
2
H
o
2
2
k
oo
M
o
2
H
o
2
3
and m
z
1, where o
H
is the resonance frequency given by
o
H
gH
0
4
and
o
M
4pgM
s
5
(g is the gyromagnetic ratio g ge/2m
e
c, where g is the
Lande factor; e and m
e
, are the electron charge and mass,
respectively; and c the speed of light. In most ferrites the
magnetization is due to the electron spin alone; therefore
g 2 and g 1.76 10
7
rad/s.Oe).
There is also a wide class of useful microwave devices
that use ferrites at the remanent magnetization.
These are termed latch ferrite devices and use ferrite
materials with a square hysteresis loop, so that the re-
manence magnetization M
r
is very close to the saturation
magnetization. At remanence H
0
0 and (2) and (3)
simplify to m 1 and k 4pM
r
/o. Nevertheless, the use
of these expressions is subject to some restrictions related
to the magnetic losses that may appear at low values
of H
0
[6].
The lossless ferrite is an approximation. Actually, there
are many mechanism of losses in ferrites. Some of them,
such as the ohmic conductivity, can be introduced in the
RF constitutive relationships adding an imaginary part to
the dielectric permittivity. Moreover, they are magnetic
losses as a consequence of the damping of the magnetic
oscillations described by the Polder tensor. The presence of
the resonance frequency o
0
in that tensor clearly suggest
the presence of losses in real ferrites, with a maximum at
this frequency. Magnetic losses may be included in the
Polder tensor after the transformation [10]
o
H
!o
H
jao 6
where a is a new parameter that accounts for the losses.
The a parameter is often substituted by the resonance
linewidth DH, which is the width of the resonance curves
for the real part of k and m1 plotted against H
0
. The
resonance linewidth is related to a by DH2ao
H
/g. The
magnitude of magnetic losses varies widely in ferrites
used for microwave applications. The resonance linewidth
ranges from about 0.1Oe for single YIG crystals to several
hundreds for polycrystalline ferrite materials. In the rst
case we can obtain meaningful results neglecting magnet-
ic losses, but this approximation may lead to significant
misleadings in other cases.
Until now we have considered only ferrites with intrin-
sic isotropy. This assumption is not realistic in all cases
because ferrites are crystalline materials with complex
internal structure. Magnetocrystalline intrinsic anisotro-
py usually induces in the crystal easy and hard directions
of magnetization (i.e., directions on which the ferrite
canor cannotbe magnetized to saturation by a small
applied magnetic eld). The modications of the Polder
tensor induced by the magnetocrystalline anisotropy are
complex and will not be analyzed here. The reader inter-
ested in this topic is referred to any of the textbooks that
develop such expressions [e.g., 8].
The most important magnetocrystalline effect occurs in
uniaxial hexagonal ferrites. In these crystals, the reso-
nance is pushed to a extremely high frequency, even when
the applied static eld is small. Thus, hexagonal ferrites
found most of its practical applications at millimeter wave
frequencies.
1.3. The Radiofrequency Field
The RF eld in a ferrite, characterized by the tensor per-
meability (1), is obtained by solving the Maxwell equa-
tions with the appropriate boundary conditions. Assuming
a time-harmonic dependence of the kind exp jot, as well as
the vanishing of current sources, these equations will read
rE jom H 7
rHjoeE 8
where the dielectric permittivity e will be in general a
complex quantity, in order to incorporate dielectric and
ohmic losses.
1.3.1. Uniform Plane Waves with Longitudinal Magneti-
zation. Although the aim of this article is to analyze the
propagation along waveguides, the analysis of the propa-
gation in an unbounded ferrite medium will provide a
useful introduction to some relevant aspects of the prop-
agation in waveguides. We will rst suppose a uniform
plane wave with a spacetime dependence of the kind exp
j( kz ot) (this factor will be suppressed in the following
FERRITE-LOADED WAVEGUIDES 1487
text) and an internal static magnetic eld directed along
the z axis: H
0
H
0
a
z
.
The analysis of (7) and (8) with these restrictions leads
to two TEM-wave solutions with right-handed circular po-
larization (RCP) and left-handed circular polarization
(LCP) referred to the static eld H
0
orientation. These
two waves have different phase constants given by
k

o

em
eff
p
9
where the plus sign stands for the RCP polarization and
the minus sign, for the LCP polarization and m
eff
is an ef-
fective magnetic permeability given by the two eigenvalues
of [m]
t
:
m

eff
m
0
m k 10
Note that the hand of polarization has been referred to the
internal static magnetic eld orientation, regardless of the
direction of propagation. Therefore, if one of these waves is
fully reected (e.g., by a perfect conducting plate perpen-
dicular to propagation), the hand of the circular polariza-
tion, as well as the effective magnetic permeability, will
remain unchanged, giving rise to a stationary wave.
The values of k
7
for an isotropic lossless ferrite with
RCP and LCP polarizations are shown in Fig. 1 for a fer-
rite magnetized under the usual technological condition of
H
0
o4pM
0
. A forbidden frequency range for RCP waves, in
which k

becomes imaginary, is given by


o
H
oooo
H
o
M
11
If magnetic losses are considered, the transformation (6)
must be introduced in the expressions for m

eff
. This leads
to two complex propagation constants, that of the RCP
wave having a typical resonant behavior with high reso-
nance losses (see Fig. 1).
The most important effect related to plane-wave prop-
agation in a longitudinally magnetized ferrite is the non-
reciprocal Faraday rotation of the plane of polarization of
a linearly polarized wave. A linearly polarized wave is not
a solution of (7) and (8), but it can be obtained by adding
two contrarotating RCP and LCP waves of equal ampli-
tude. Since the phase constants of these two waves are not
equal, the result is a rotation of the plane of polarization of
the linearly polarized wave. The rotation angle after the
wave has advanced a length Dz is given by
y
1
2
k

Dz 12
When a linearly polarized wave is reected backward, the
hand of rotation of the polarization plane remains un-
changed. Thus, the planes of polarization of the incident
and the reected waves will be different at a given dis-
tance from the plane of reection. Therefore, the Faraday
rotation in ferrites is nonreciprocal. If losses are consid-
ered, both the RCP and the LCP waves in which the lin-
early polarized wave splits have different attenuation
constants. This leads to an unequal change in the ampli-
tudes of the RCP and LCP waves, which causes Faraday
elipticity of the original linearly polarized wave. Detailed
treatments of Faraday rotation and elipticity may be
found in the literature cited in the introduction [115].
1.3.2. Transverse Magnetization. We will now suppose
an uniform plane wave with a spacetime dependence of
the kind exp j( kx ot) (this factor will be suppressed in
the following text) and an internal static magnetic eld
directed along the z axis: H
0
H
0
a
z
. The solution to (7)
and (8) with these restrictions leads two independent uni-
form plane waves with the E eld linearly polarized. One
of them is a TEM wave with the magnetic eld parallel to
H
0
. Thus, there is no interaction between the RF eld and
the electronic spins, and the effective magnetic permeabil-
ity is m
eff
m
0
m
z
. This solution is called the ordinary wave,
with phase constant k
2
o

em
0
m
z
p
. There is also an
extraordinary wave, whose propagation constant is still
given by (9), but with m
eff
given by
m
eff
m
0
m
2
k
2
m
13
The extraordinary wave is a TE wave with the electric
eld polarized parallel to H
0
and the magnetic eld ellip-
tically polarized in the plane perpendicular to H
0
. The
values of the propagation constants of the extraordinary
waves for the lossless and the lossy ferrites of Fig. 1 are
shown in Fig. 2. For the lossless extraordinary wave there
is a frequency-forbidden range, in which k becomes imag-
inary, dened by

o
H
o
H
o
M

_
oooo
H
o
M
14
The presence of an ordinary wave and an extraordinary
wave with orthogonal polarization, recalls the birefrin-
gence of uniaxial crystals. This birefringence can be used
Im(k)
RCP Re(k)
RCP
RCP
1
LCP
o/o
H
Figure 1. Normalized complex propagation constant for the RCP
wave (10) in a lossy ferrite with o
M
1.5o
H
and gDH0.1o
H
(solid lines). The normalized phase constants for the two RCP and
LCP waves in a lossless ferrite with o
M
1.5 o
H
and gDH0 are
also shown (dashed lines).
1488 FERRITE-LOADED WAVEGUIDES
in the design of microwave devices such as half- and quar-
ter-wave plates and polarizers.
1.3.3. Magnetization at Any Angle. In this case, the
phase constant can still be written as in (9), with an ef-
fective magnetic permeability that depends on the angle
y
k
between the static magnetization and the wave phase
velocity. The nal expression for this effective permeabil-
ity is [10]
m
eff
y
k

m
0
2
m
t
m
z
1
_ _
sin
2
y
k

m
t
m
z
1
_ _
2
sin
4
y
k
4
k
2
m
2
cos
2
y
k

2
sin
2
y
k
m
z

cos
2
y
k
m
_ _
15
with m
t
(m
2
k
2
)/m. For a xed frequency, a plot of k, y
k
in
polar coordinates, with k o

em
eff
p
, gives the phase con-
stants corresponding to the two solutions of (15). These
curves can also be interpreted as isofrequency curves of
the dispersion equation oo(k, y
k
) in the k, y
k
plane.
Thus, the group velocity v
g
r
k
o will be perpendicular to
these curves at each angle y
k
. For an arbitrary y
k
, the di-
rection of the optical ray will not be parallel to the direc-
tion of propagation of the wavefronts.
1.3.4. Nonreciprocity. One of the basic theorems of
electromagnetism is the Lorentz reciprocity theorem. It
applies to any linear and causal medium whose cons-
titutive relationships can be described by symmetric fre-
quency-dependent dielectric permittivity and/or magnetic
susceptibility tensors. As was already mentioned, since
the tensor magnetic susceptibility (1) is not symmetric,
this is not the case for ferrites. In fact, many of the prac-
tical applications of ferrites in microwave technology, such
as circulators or isolators, arise from this nonreciprocal
behavior. However, it is still possible to reformulate the
reciprocity theorem in a form that is applicable to ferrite
media. Following Harrington [16] and McIsaac [17], we
will start from the Onsager relation, which states that any
tensor macroscopic susceptibility of a causal and linear
medium must be equal to the mediums transpose after
reversal in time of all the physical relevant quantities. For
an externally magnetized ferrite, taking into account that
the static bias eld changes of sign after reversal in time,
we conclude that the tensor magnetic susceptibility of a
ferrite (1) must equal its transpose after the change of sign
of the static biasing eld. From this conclusion, we can
directly state the reciprocity theorem for ferrite media
__
EH
0
E
0
H dS
___
E
0
J EJ
0
dV 16
where the physical quantities must be reinterpreted as
follows: E, H and E
0
, H
0
are two independent electromag-
netic eld congurations produced by source current den-
sities J and J
0
respectively, at frequency o in the same
ferrite medium, except that the medium in which the
prime quantities are dened has a reverse static magne-
tization: H
0
0
H
0
. The surface integrals on the left side
of (16) are over any surface containing the source current
densities included in the volume integral on the right side.
This generalized reciprocity theorem is useful in the anal-
ysis of mode orthogonality in ferrite-loaded waveguides, as
well as in the analysis of ferrite-loaded waveguide junc-
tions.
2. MICROWAVE PROPAGATION IN FERRITE-LOADED
WAVEGUIDES
In the preceding section microwave propagation in un-
bounded ferrite media was analyzed. Many of the studied
effects, such as Faraday rotation and nonreciprocity, also
appear when the RF eld propagates along waveguides.
Moreover, the microwave propagation along ferrite-loaded
waveguides presents new useful effects, such as unidirec-
tionality and eld displacement, complex and backward
modes, slow magnetostatic waves, and circulation.
In the following paragraphs we will choose the z axis as
the waveguide axis, and a spacetime dependence of the
kind exp j(ot kz) will be supposed. The mode phase con-
stant k will be in general a complex number kb ja.
Both b and a will be chosen real without loss of generality,
and the factor exp j(ot kz) will be suppressed.
2.1. Unidirectional and Bidirectional Modes
A mode with phase constant k that does not have a sym-
metric pair with the opposite phase constant k is called
unidirectional. All lossless and reciprocal waveguides are
bidirectional. This is not the case for ferrite loaded wave-
guides, because ferrites are nonreciprocal media that are
not invariant after time reversal. The presence of unidi-
rectional modes of propagation in ferrite-loaded wave-
guides is useful in many microwave devices, such as
isolators and nonreciprocal phase shifters. However,
many ferrite-loaded waveguides, which remain invariant
o/o
H
Re(k)
Im(k)
1
Figure 2. Normalized phase and complex propagation constants
for the extraordinary waves in the innite lossless (dashed lines)
and lossy (solid lines) ferrites of Fig. 1 with transverse magneti-
zation.
FERRITE-LOADED WAVEGUIDES 1489
after some symmetry transformations, are bidirectional;
specifically, unidirectional modes cannot propagate along
these waveguides. McIsaac [18] and, more recently, Dmi-
triyev [19] have investigated these symmetries. McIsaac
[18] concludes that bidirectionality is ensured if the wave-
guide remains the same after one or more of the following
transformations:
*
Reection in a plane perpendicular to z axis
*
Rotation by 1801 about an axis perpendicular to the z
axis
*
Inversion at any point
In performing these transformations, the pseudovectorial
nature of the static bias eld H
0
has to be taken into ac-
count (i.e., H
0
remains the same after spatial inversion
and after reection in a perpendicular plane, but changes
sign after a reection in a parallel plane). In particular,
any ferrite-loaded waveguide with longitudinal magneti-
zation must be bidirectional, because this waveguide re-
mains unchanged after reection in a plane perpendicular
to the z axis.
Bidirectionality does not imply that all the character-
istics of the modes remain unchanged when the direction
of propagation is reversed. For instance, the energy dis-
tribution and/or the polarization of a pair of bidirectional
modes having the same but opposite phase constants may
be different. Moreover, although all modes in bidirectional
ferrite-loaded waveguides must be bidirectional, not all
the modes in nonbidirectional ferrite-loaded waveguides
are unidirectional; some of them, having the appropriate
polarization, may be bidirectional.
2.2. Complex and Backward Modes
Complex modes in inhomogeneously lled lossless wave-
guides were rst reported by Tai in 1960 and Carricoats in
1965 [20]. Complex modes in lossless isotropic waveguides
are characterized by a complex propagation constant k
7b7ja and appear in groups of four solutions, in which all
the possible combination of signs are allowed. Carricoats
also shown that, for a single complex mode, power ows in
opposite directions along the different media lling the
waveguide, giving a zero net power ux. Therefore, com-
plex modes in lossless waveguides are reactive modes.
Complex modes have proved to be a very important part of
the spectra of ferrite-loaded waveguides [20] (in fact, they
were rst reported for ferrite-loaded waveguides by Tai).
In particular, all the unidirectional and reactive modes in
ferrite-loaded waveguides must be complex [21,22].
Complex modes are closely related to backward modes
(i.e., modes with negative group velocity). In fact, a pair of
complex modes in lossless waveguides usually changes to
a pair of propagating forward and backward modes when
frequency varies [20]. Backward modes in the spectra of
ferrite loaded waveguides, mainly in the magnetostatic
wave region, has been widely analyzed (see, e.g., Ref. 8).
2.3. Mode Orthogonality
Mode orthogonality in ferrite-loaded waveguides was an-
alyzed in Ref. 23. Applying the generalized reciprocity
theorem (16) to two modes bfe
m
, h
m
(x, y) exp j(ot k
m
z)
and bfe
0
n
; h
0
n
x; y expjot k
0
n
z of the actual waveguide
and the complementary waveguide (the complementary
waveguide is dened as the original one with the static
magnetic eld reversed), the following relation is obtained
k
0
n
k
m

__
e
0
n
h
m
e
m
h
0
n
u
z
dx dy 0 17
where the integral is over the cross section of the wave-
guide. This equation may be considered as a general or-
thogonality relation since the integral must be zero unless
k
m
k
0
n
. This relation simplies for most practical situa-
tions in which the static magnetic eld is either parallel or
perpendicular to the waveguide axis. The explicit orthog-
onality relationships for these particular but important
magnetizations can be found in Ref. 23.
2.4. Field Displacement Effects
It was pointed out earlier that electromagnetic wave prop-
agation in ferrites at an arbitrary angle with respect to the
magnetizing eld usually implies that the optical rays
(and the power ux) are not parallel to the phase velocity.
However, in a nonradiating waveguide, both power ux
and wave propagation are forced to be parallel to the
waveguide axis. Thus, this effect cannot be present in
nonradiating waveguides. Instead, this tendency of power
to ow in a direction different from the wave propagation
may cause strongly unsymmetric accumulations of elec-
tromagnetic energy across the waveguide section. This
nonreciprocal eld displacement effect is used in the
design of microwave isolators and phase shifters.
2.5. The Magnetostatic Approximation
Near the resonances m
eff
-Nand the effect of the Maxwell
displacement current may be neglected with regard to the
Faraday induction effects. This leads to the magnetostatic
approximation. Taking into account that r B0, a mag-
netostatic potential H rc is dened, which must
satisfy
r m rc 0 18
(in ferrite-loaded waveguides, the nabla operator is re-
placed by r !r
t
jku
z
). The solutions to (18), with the
appropriate boundary conditions, are the magnetostatic
modes of the waveguide. Magnetostatic waves in ferrite-
loaded waveguides, have been analyzed extensively [8].
The main applications of magnetostatic waves in micro-
wave technology arise from its small wavelength. This re-
sult in broad applications in miniature controllable
devices, such as delay lines, lters, power limiters, and
signal-to-noise enhancers [24].
2.6. Basic Properties of Ferrite-Loaded Waveguide Junctions
A waveguide junction is characterized by its scattering
matrix S
i,j
. It is a well-known fact that if the materials
lling the junction are reciprocal, the scattering matrix
must be symmetric. If the junction is nonreciprocal, this
1490 FERRITE-LOADED WAVEGUIDES
statement must be modied as a consequence of the re-
formulation of the reciprocity theorem (16). This modi-
cation leads to the following relations between the
scattering matrix elements of a ferrite-loaded junction
and its complementary (i.e., the junction with the biasing
static magnetic eld reversed):
S
i;j
o; H
0
S
j;i
o; H
0
19
If the junction is also lossless, the scattering matrix must
be unitary (S
i;j
S

k;j
d
i;k
, where the rule of summation over
all the repeated subindex has been used). Other symme-
tries of the scattering matrix may be deduced from the
spatial symmetries of the junction (including the bias
eld) [19].
The use of the scattering matrix symmetry properties is
useful in the design of many microwave devices, such as
isolators, phase shifters, and circulators. The Y circulator
is perhaps the most useful and best known nonreciprocal
junction. An Y circulator is a symmetric three-port junc-
tion with some specific properties. A symmetric three-port
junction must have S
1,1
S
2,2
S
3,3
, S
1,2
S
2,3
S
3,1
, and
S
2,1
S
1,3
S
3,2
. These relations are fullled by any junc-
tion having a rotation symmetry axis of third order and
magnetized along this axis. The circuit theory of three-
and N-port circulators may be found in Ref. 6 and other
textbooks. It can be shown that if a lossless, nonreciprocal,
and symmetric three-port waveguide junction is matched
(i.e., S
1,1
0), it is also an ideal Y circulator [i.e., S
1,2
1 or
0 and S
1,3
0 or 1]. If the magnetization of a nonreciprocal
three-port Y circulator is reversed, the direction of circu-
lation is also reversed, as a consequence of (19).
3. FERRITE-LOADED WAVEGUIDES FOR
PRACTICAL APPLICATIONS
In this section we will describe the most widely used fer-
rite-loaded waveguides. There are many classical text-
books and papers [2,3,57], and more recent ones [9,25]
that describe these waveguides, as well as the most useful
microwave devices that may be designed using them. The
reader may use these and other texts to broad the infor-
mation contained in this section.
3.1. Circular Waveguides with Longitudinal Magnetization
It is well known from the theory of hollow waveguides that
the fundamental mode of the empty circular waveguide is
the TE
1,1
mode, which is a double degenerate mode with
two perpendicular polarizations in the waveguide cross
section. This mode has a eld distribution that is almost
TEM in the vicinity of the waveguide axis. Thus, if a fer-
rite rod with longitudinal magnetization is placed at the
center of the waveguide (see Fig. 3), the two orthogonal
and degenerate TE
1,1
fundamental modes will interact as
a consequence of the Faraday rotation effect, giving rise to
two nondegenerate circularly polarized RCP and LCP
modes (with the hand of polarization dened with respect
to the static bias eld orientation). These modes can be
approximated by the same RCP and LCP modes of a cir-
cular waveguide with an inner isotropic rod with an scalar
magnetic susceptibility given by (10). In the same way, if a
linearly polarized wave with linear polarization enters the
ferrite-loaded waveguide, this wave will experience a
Faraday rotation by an angle given approximately by
(12), where k

and k

are now the phase constants of


the RCP and LCP ferrite-loaded waveguide modes.
The phase constants, as well as the mode elds of the
two nondegenerate RCP and LCP ferrite-loaded wave-
guide modes, were obtained analytically by Waldron in
1958. Analytical solutions, not only for the cylindrical
waveguide with a ferrite rod but also for many other re-
lated structures, such as cylindrical waveguides loaded
with ferrite and dielectric tubes, may also be found in the
literature [4]. Modes in this kind of ferrite-loaded
waveguide are not TE nor TM, but becomes TE and TM
at cutoff [4], these modes are termed HE and EH depend-
ing on whether the magnetic H
z
or the electric E
z
eld
dominates. At cutoff, HE modes become TE and EH modes
become TM.
Ferrite-loaded circular waveguides with longitudinal
magnetization are extensively used in Faraday rotation
devices, based on the aforementioned rotation of the po-
larization plane of a linearly polarized wave. The most
known Faraday rotation device is the four-port circulator,
described in many textbooks. Faraday rotation may be
also used in the design of magnetically tuned variable at-
tenuators, isolators and phase shifters (see, e.g., Ref. 9 and
references cited therein). In the early years of microwave
ferrite technology, much effort was devoted to develop
Faraday rotation circulators and other microwave devices
with cylindrical geometry. In the following years, however,
the Y-junction circulators, as well as phase shifters and
attenuators in rectangular and/or planar technology, were
found to be smaller, simpler, and more appropriate for
Ferrite
(j)
H
0
a
Figure 3. Cylindrical waveguide with an inner centered ferrite
rod of longitudinal magnetization.
FERRITE-LOADED WAVEGUIDES 1491
most applications; the research effort then began to focus
on these devices.
3.2. E-Plane Transversely Magnetized Ferrite-Loaded
Rectangular Waveguides
Figure 4a shows the variation of the magnetic eld com-
ponents H
x
and H
z
of the fundamental TE
1,0
mode in a
hollow rectangular waveguide. The magnetic eld is cir-
cularly polarized around the y axis where |H
z
||H
x
|.
This condition occurs at two symmetric positions, at a dis-
tance d of the rectangular sidewalls, given by
d
a
p
cot
1

4a
2
l
2
0
1
_ _
20
If an E-plane ferrite slab biased with a static magnetic
eld directed along the y axis is placed at a distance d of
one of the lateral sidewalls (see Fig. 4b), the wave prop-
agating in the positive (negative) direction along the z axis
is right (left)-handed-polarized with respect to the static
bias eld. Thus we can expect that the forward (backward)
wave will see the effective magnetic permeability of the
slab m

eff
m

eff
, given by (10). Therefore, wave propagation
will be unidirectional, with different phase constants for
opposite directions of propagation. Moreover, since the de-
pendence of m

eff
with H
0
is much stronger than those of
m

eff
, the forward wave will be much more affected by vari-
ations in the intensity of H
0
than the backward one. If the
bias eld is chosen in a region in which m

eff
is positive and
presents a strong dependence on the bias eld intensity,
the waveguide may be used as a nonreciprocal phase shift-
er. If the bias eld is chosen at the resonance condition
(o
H
o), the forward wave will see a resonant magnetic
permeability and will experience strong attenuation due
to the resonance losses. Then, the waveguide may be used
as a resonance isolator. If the bias eld is chosen at the
antiresonance condition o
H
oo
M
, the forward wave
will see a perfect diamagnet with m

eff
0, which imposes
perfect diamagnetic boundary conditions at both slab
sides and, therefore, zero tangential electric RF eld at
these boundaries. If an absorber is located at the inner
boundary of the slab, it is expected that the forward wave
will not be attenuated whereas the backward wave will be
strongly attenuated. This waveguide may thus be used as
a eld displacement isolator.
The wave propagation characteristics along this wave-
guide may also be found analytically. The rst published
results on this subject are due to of Kales (1953). Gardiol
[26] gave a general method for computing the propagation
characteristics of rectangular waveguides lled with an
arbitrary number of anisotropic slabs, including ferrite
slabs, making use of the transverse transmission matrix
method. Referred to the geometry of Fig. 4b, the trans-
verse transmission matrix of the i region [T
i
] is dened as
the matrix relating the tangential elds, E
y
and H
z
, at
both sides of the ith region of the waveguide. In a notation
that becomes apparent, we can write
E
y
H
z
_
_
_
_
x a
T
3
T
2
T
1

E
y
H
z
_
_
_
_
x 0

t
1;1
t
1;2
t
2;1
t
2;2
_
_
_
_

E
y
H
z
_
_
_
_
x 0
21
Since the lateral sidewalls are assumed to be perfect con-
ducting walls, the tangential electric eld must vanish at
these boundaries. This imply that t
1,2
0, which may be
considered an implicit equation for the phase constant.
This method may be applied to rectangular waveguides
with any number of E-plane transversely magnetized fer-
rite slabs and/or lossy dielectric slabs, thus providing a
general method for the analysis of isolators and phase
shifters.
The nonreciprocal isolation and phase variation effects
of the E-plane ferrite-loaded waveguide of Fig. 4b may be
increased by placing a symmetric ferrite slab with reverse
magnetization at the remaining plane of circular polar-
ization of the TE
1,0
mode. Since at this plane the circular
polarization of the TE
1,0
wave has opposite handedness,
the effect of the new ferrite slab adds to the effect of the
former one. A variation of this two-slab ferrite-loaded
waveguide is the latch ferrite toroid in rectangular wave-
guide (Fig. 5a) proposed by Treuhaft (1958) [9] for phase-
shifting applications. The main advantage of this cong-
uration is that the permanent magnet is substituted by a
ferrite toroid magnetized at remanence by an electric cur-
rent pulse, driven by a single wire at the center of the
waveguide (this wire is perpendicular to the RF electric
eld and has a negligible effect on microwave propaga-
tion). This structure is also suitable for rapid switching
between the two opposite nonreciprocal states of the wave-
guide. In the analysis of this structure, the upper
and lower branches of the ferrite toroid, which do not
have substantial effect on phase change, may be neglected
H
z

H
z

H
x

x=a
a
3
H
0
1
F
e
r
r
i
t
e
2
(a) (b)
Figure 4. (a) Plot of the intensities of the mag-
netic eld components |H
x
| and |H
z
| of the fun-
damental TE
1,0
mode in a rectangular hollow
waveguide (frequency 9 GHz, a23mm; dashed
linesplanes of circular polarization); (b) rectan-
gular waveguide loaded with a transversely
magnetized ferrite slab at a plane of circular
polarization of the RF magnetic eld.
1492 FERRITE-LOADED WAVEGUIDES
leading thus to the simpler structure of Fig. 5b. Gardiol
[27] gives expressions that transform the geometry of Fig.
5a into the geometry of Fig. 5b with a gain in accuracy. An
alternative for reducing the unwanted effects of the upper
and lower branches of the ferrite toroid is to replace this
into a rectangular grooved waveguide, as proposed in Ref.
28. If nonreciprocity is not desired, a reciprocal phase
shifter may still be obtained magnetizing both slabs of Fig.
5b with parallel and equal static magnetic elds. This
structure is symmetric after inversion at a point in the
waveguide axis and, therefore, is bidirectional.
3.3. Other Useful Cylindrical and Rectangular
Ferrite-Loaded Waveguides
Although the circular and rectangular geometries seem to
be the natural geometries for longitudinal and transverse
magnetization, respectively, there are also some useful
devices that use transversely magnetized circular wave-
guides and longitudinally magnetized rectangular wave-
guides. Field displacement effects similar to those
reported earlier for rectangular waveguides may be
achieved in cylindrical waveguides loaded with latch fer-
rite tubes magnetized in the azimutal direction. The dual-
mode ferrite phase shifters include latch and transversely
magnetized circular waveguide sections [29]. A widely
used ferrite-loaded rectangular waveguide with longitu-
dinal magnetization is the ReggiaSpencer phase shifter
[30], which consists of a ferrite rod with longitudinal mag-
netization placed at the center of a rectangular wave-
guide. If the dimensions of the hollow waveguide allows
for the propagation of only the rst TE
1,0
mode, Faraday
rotation cannot take place. Instead, a strong variation of
the wave phase constant with the applied static magnetic
eld occurs. Like all waveguides having longitudinal mag-
netization, the ReggiaSpencer phase shifter is bidirec-
tional; thus the phase change is reciprocal.
4. FERRITE-LOADED MICROSTRIPS, SLOTLINES, AND
FINLINES
After the mid-1960s, when planar microwave integrated
circuits became a viable technology, ferrite-loaded micro-
strips and slotlines began to be investigated as an alter-
native to traditional ferrite-loaded waveguides for the
design of reciprocal and nonreciprocal phase shifters
[31], isolators [32], and other useful devices, which have
been summarized in some classical review papers [25] and
textbooks [9]. Later, when nlines emerged as a useful
alternative for planar technology in millimeter-wave
circuits, ferrite-loaded nlines [33,34] also began to be
investigated.
4.1. The Ferrite-Loaded Parallel-Plate Waveguide
Before considering more complicated structures, it will be
useful to analyze the simpler parallel-plate waveguide
loaded with ferrite slabs (Fig. 6). It has been shown [25]
that for magnetization parallel to propagation, these
waveguides supports a quasi-TEM mode and that the fer-
rite layers may be characterized by an effective perme-
ability given by (13). For magnetization perpendicular to
both the direction of propagation and the plane of the
waveguide, the ferrite layers may be again characterized
by the scalar effective permeability (13), but the eld is no
longer TEM because of the birefringence effects. Finally,
for magnetization perpendicular to propagation and par-
allel to the waveguide plane, there is almost no interaction
between the static magnetic eld and the RF eld and the
ferrite layers are characterized by the scalar permeability
m
z
. Near the forbidden frequency range, the analysis be-
comes more involved, due to the apparition of magneto-
static modes. Magnetostatic modes in parallel-plate
ferrite-loaded waveguides have been extensively analyzed
(see, e.g., Ref. 8 and references cited therein).
4.2. Ferrite-Loaded Microstrip Lines
Microstrip line (see Fig. 7a) is the most widely used wave-
guide in planar technology. Although exact methods of
analysis are now available, considerable insight into the
physical behavior of ferrite-loaded microstrip lines may be
obtained from the well-known parallel-plate microstrip
y
x
c=c(y), [j]=[j](y;H
0
)
Figure 6. Parallel-plate waveguide lled by a multilayer medi-
um including one or more ferrite slabs [e(y) and [m](y) are piece-
wise constant functions of y].
Ferrite
H
0
Ferrite
H
0
H
0
c
0
c
0
c I
(a) (b)
Figure 5. (a) Rectangular waveguide loaded with a latch
ferrite toroid; (b) rectangular waveguide loaded with two
oppositely magnetized ferrite slabs at the planes of circular
polarization of the RF magnetic eld.
FERRITE-LOADED WAVEGUIDES 1493
model (see Fig. 7b). In this model, the microstrip line is
substituted by a section of parallel-plate waveguide be-
tween two magnetic walls. This section is slightly wider
than the microstrip, in order to incorporate the effects of
the fringing elds. Nevertheless, if, at the operating fre-
quency, magnetostatic modes could be excited in the hous-
ing surface waveguide at both sides of the microstrip (see
Fig. 7a), they could actually be excited by the microstrip,
which would then become a magnetostatic wave trans-
ducer [8,24].
4.2.1. Microstrip with Longitudinal Magnetization. Us-
ing the model of Fig. 7b and the results reported earlier,
the qualitative behavior of these kinds of microstrips may
be investigated by means of the analysis of a section of
parallel-plate waveguide loaded with one or more slabs of
effective magnetic permeability given by (13), bounded by
two perfect magnetic walls. This model provides a quasi-
TEM and bidirectional fundamental mode with a phase
constant that is a function of the biasing magnetic eld.
The same qualitative results are provided by more accu-
rate quasi-TEM analyses of the actual microstrip line,
using either the effective permeability (13) [25], the tensor
magnetic permeability (1) [35,36], or by a full-wave
analysis, discussed later in this article. The main applica-
tion of microstrip lines with longitudinal magnetization
is in phase shifting by meanderlines, a design that min-
imize the size of the device [31]. The phase variation
with the applied magnetic eld may be increased if strong-
ly coupled quarter-wave meanderline sections are used.
These structures provides strong nonreciprocal phase
shifting [31] were nonreciprocity is due to the coupling
effects.
4.2.2. Transversely Magnetized Microstrip Lines. We will
consider the two orthogonal magnetizations, perpendicu-
lar and parallel to the ground plane. For the second one
there is no expected interaction between the static bias
eld and the RF eld in the parallel plate waveguide mod-
el of Fig. 7b because of the RF magnetic eld parallel to
the static bias eld. In fact, only a slightly nonreciprocal
phase shift may be observed, due to the fringing eld near
the microstrip edges.
Of much more interest is the microstrip line with mag-
netization perpendicular to the ground plane. Considering
anew the parallel plate model of Fig. 7b, the ferrite slab
can be characterized by a layer of effective magnetic per-
meability (13). However it must be realized that, for this
magnetization, the RF eld is not quasi-TEM, due to the
birefringence effects in the ferrite layer. Hines [32] showed
that in a semiinnite ferrite-lled parallel-plate wave-
guide, with a perpendicular magnetic wall at its end,
and magnetized along the direction perpendicular to the
ground planes, there is a TEM mode propagating along
the waveguide and attenuating in the direction perpen-
dicular to the magnetic wall. The phase constant of this
mode is given by kk
0
j

e
r
m
p
a
z
k=m

e
r
m
p
a
x
[note that
k kk
2
0
e
r
m
eff
=m
0
; that is, k is the same as for a nonuni-
form plane wave in a slab of effective permeability m
eff
].
Hines also showed that this mode is unidirectional. The
TEM mode described by Hines fullls all the requirements
imposed by the lateral magnetic walls of Fig. 7b for suf-
ciently wide strips. Moreover, since the mode in the semi-
innite parallel-plate waveguide was unidirectional, the
microstrip mode will present a strong eld displacement
from one to another strip edge when the direction of prop-
agation is reversed.
Although the analysis above has focused on ferrite-
lled parallel-plate waveguides, the same qualitative re-
sults also apply to multilayered ferrite-loaded parallel-
plate waveguides and microstrips. Hines modes are usu-
ally termed edge modes in microstrips because the RF eld
is concentrated mainly in the vicinity of a microstrip edge.
Edge modes in microstrip are useful in the design of wide-
band edge-mode isolators and nonreciprocal edge-mode
phase shifters [32,38,39], which use the same principle as
do eld displacement isolators and phase shifters. Analy-
sis of such structures is usually performed using approx-
imate models. More recently, spectral-domain analysis has
been successfully applied to analyze edge-mode isolators
without approximations [40].
4.3. Ferrite-Loaded Slotlines and Finlines
Slotlines (Fig. 8a) and coplanar waveguides are an useful
alternative to microstrip in the design of microwave inte-
grated circuits. In millimeter-wave technology, nlines in
rectangular waveguides (Fig. 8b) also represent a good al-
ternative for integration without radiation losses. By add-
ing ferrite layers to these waveguides, many of the effects
described for ferrite-loaded microstrips may be achieved.
Since the RF magnetic eld in slotlines and nlines is
concentrated mainly in the slot and directed perpendicu-
lar to the air interface, it is expected that the strongest
effects for transversely magnetized slotlines and nlines
will occur for magnetization parallel to the ns. These
structures have been extensively analyzed for the design
of eld displacement isolators and phase shifters [25]. A
c=c(y)
[j]=[j](y;H
0
)
w
y
x
w+w
w+w
Perfect
magnetic
wall
Perfect
magnetic
wall
c=c(y),[j]=[j];(y;H
0
)
(a) (b)
Figure 7. (a) Microstrip line on a multilayer
ferrite-loaded substrate [as in Fig. 6, e(y) and
[m](y) are piecewise constant functions of y];
(b) parallel-plate waveguide model for the mi-
crostrip line of (a).
1494 FERRITE-LOADED WAVEGUIDES
millimeter-wave eld displacement nline isolator has
been proposed [33]. Transversely magnetized slotlines
and nlines for nonreciprocal phase-shifting applications
have also been analyzed [37,4244]. Applications of n-
lines with longitudinal magnetization have been also in-
vestigated [34].
4.4. Methods of Analysis of Ferrite-Loaded Quasiplanar
Layered Structures
With a few exceptions, quasi-TEM analysis usually pro-
vide sufciently accurate results for the analysis of con-
ventional microstrip and coplanar or slotlines. However,
this analysis is seldom suitable for these structures when
they are ferrite-loaded. In fact, quasi-TEM modes are by
definition bidirectional, and therefore the quasi-TEM
analysis cannot account for many of the most relevant
physical effects in ferrite-loaded transmission lines. Quasi-
TEM analysis in its standard form is restricted to
longitudinally magnetized lines [35,36]. More recently,
however, some attempts have been made in order to gen-
eralize this analysis to transversely magnetized struc-
tures [41]. Nevertheless, in general, planar and
quasiplanar ferrite-loaded transmission lines require
full-wave analysis.
With respect to numerical techniques, spectral-domain
analysis (SDA) is by far the most widely used technique
for the analysis of ferrite-loaded strip and/or slot struc-
tures [3537,40,4345]. The basic fundamentals of SDA
may be found in many textbooks, such as that of Mi-
rshekar-Syahkal [46]. SDA is specially well suited for the
analysis of microstrips and/or slot- and nlines on planar
single- or multilayer substrates, because of the translat-
ional symmetry of these substrates. Since the SDA applied
to microstrip or microslot structures is adequately de-
scribed in Ref. 46 and other textbooks, we only briefly de-
scribe here the main specific characteristics of the SDA
when it is applied to ferrite-loaded microstrip and/or mi-
croslot waveguides. The main difculty in the application
of the SDA to ferrite-loaded microstrip or microslot on in-
nite planar substrates is in the determination of the
spectral-domain Greens function dyad
"
GGk
x
; k
z
, which re-
lates a surface current source J
s
J
s,0
expjk
x
x expjk
z
z
in the plane of the structure, with the RF tangential elec-
tric eld E
t
E
t,0
expjk
x
x expjk
z
z (E
t
(E
x
, E
z
)
t
) over
the same or other parallel plane:
J
0

"
GGk
x
; k
z
E
t;0
22
General methods for the computation of the spectral
Green dyad in multilayered ferrite-loaded substrates (in
fact, in general layered bianisotropic substrates) have
been reported [47,48]. The Fourier-transformed matrix
method reported in Ref. 47 is, in fact, an application of
the transverse transmission matrix procedure described
earlier to the determination of the spectral Green dyad
(22) in general bianisotropic media with arbitrary surface
current sheets. The equivalent boundary method, de-
scribed in Ref. 48, proposes a procedure for computing
the spectral Green dyad of an n-layered medium starting
from 4n (in the worst case) single-layer Green dyads de-
ned for each layer of the structure with appropriate
boundary conditions. This results in a recurrence algo-
rithm that is found to have a high numerical stability
[45,48].
The SDA may also be applied to boxed stripline and
nline structures. In this case the integral Fourier trans-
form of the eld and currents must be substituted by a
Fourier series transform in an equivalent periodic struc-
ture. For magnetized ferrite-layered media in rectangular
metallic boxes, this imposes an important restriction
strictly speaking, the SDA can be applied only to sub-
strates with static magnetization perpendicular to the lat-
eral sidewalls. In any other case, due to the properties of
the magnetic eld after spatial reection, it is not possible
to nd an equivalent periodic structure with translational
symmetry suitable for the application of the series Fourier
transform. Therefore, the application of the SDA to boxed
structures magnetized in any direction different from the
aforementioned one, must be considered only as an
approximation.
The SDA, as described previously, applies only to struc-
tures with strips or ns of negligible thickness. Structures
with nonnegligible n or strip thickness may be analyzed
using a mode-matching technique in the transverse direc-
tion, which also implies an SDA [49]. A similar method
may be applied to boxed structures with asymmetric rect-
angular piecewise boxes [50]. Finally, the SDA also applies
to structures having fully or partly lossy strips or ns,
provided these lossy strips or ns can be described by a
suitable surface impedance, dened over the strip or the
n region [40].
BIBLIOGRAPHY
1. K. J. Button, Microwave ferrite devices: The rst ten years,
IEEE Trans. Microwave Theory Tech. 32:10881096 (1984).
2. A. G. Gurevich, ferrites at Microwave Frequencies, English
transl. Heywood, 1963.
3. P. J. B. Carricoats, Microwave ferrites, Chapman & Hall,
1961.
4. R. A. Waldron, ferrites, Van Nostrand, 1961.
c=c(y), [j]=[j](y;H
0
)
c(y)
H
0
[j](y)
(a) (b)
Figure 8. (a) Slotline on a multilayer ferrite-loaded sub-
strate; (b) nline with a multilayer ferrite-loaded substrate.
As in Figs. 6 and 7, e(y) and m(y) are piecewise constant func-
tions of y.
FERRITE-LOADED WAVEGUIDES 1495
5. B. Lax and K. J. Button, Microwave ferrites and Ferrimag-
netics, McGraw-Hill, 1962.
6. J. Helszajn, Principles of Microwave ferrite Engineering,
Wiley, 1969.
7. C. E. Fay and W. Von Aulock, Linear ferrite Devices, Academic
Press, 1969.
8. M. S. Sodha and N. C. Shrivastava, Microwave Propagation in
Ferrimagnetics, Plenum Press, 1981.
9. A. J. Baden Fuller, ferrites at Microwave Frequencies, Peter
Peregrinus, 1987.
10. A. G. Gurevich and G. A. Melkov, Magnetization Oscilations
and Waves, CRC Press, 1996.
11. L. E. Whicker and D. E. Bolle, Annotated literature survey of
microwave ferrite control components and materials for 1968
1974, IEEE Trans. Microwave Theory Tech. 23:908919
(1975).
12. R. H. Knerr, An annotated bibliography of microwave circu-
lators and isolators: 19681975, IEEE Trans. Microwave The-
ory Tech. 23:818825 (1975).
13. L. E. Davis, Status of microwave ferrite technology in Europe,
1993 IEEE Microwave Theory Tech. Symp. Digest, 1993,
pp. 199202.
14. D. C. Webb, Status of ferrite technology in the United States,
1993 IEEE Microwave Theory Tech. Symp. Digest, 1993,
pp. 203206.
15. Y. Murakami, Microwave ferrite technology in Japan: Cur-
rent status and future expectations, 1993 IEEE Microwave
Theory Tech. Symp. Digest, 1993, pp. 207210.
16. R. F. Harrington and A. T. Villeneuve, Reciprocity relation-
ships for gyrotropic media, IRE Trans. Microwave Theory
Tech. 6:308310 (1958).
17. P. R. McIsaac, A general reciprocity theorem, IEEE Trans.
Microwave Theory Tech. 27:340342 (1979).
18. P. R. McIsaac, Bidirectionality in gyrotropic waveguides,
IEEE Trans. Microwave Theory Tech. 24:223226 (1976).
19. V. A. Dmitriyev, Symmetry of microwave devices with gyro-
tropic mediacomplete solution and applications, IEEE
Trans. Microwave Theory Tech. 45:394401 (1997).
20. M. Mrozowski, Guided Electromagnetic Waves. Properties
and Analysis, Wiley, 1997.
21. R. Marques, F. Mesa, and M. Horno, On the complex nature of
higher order modes in lossless nonreciprocal transversely
magnetized waveguides, IEEE Microwave Guided Wave
Lett. 2:278280 (1992).
22. R. Marques, F. Mesa, and M. Horno, Nonreciprocal and re-
ciprocal complex and backward waves in parallel plate wave-
guides with a ferrite slab arbitrarily magnetized, IEEE Trans.
Microwave Theory Tech. 41:14091418 (1993).
23. P. R. McIsaac, Mode orthogonality in reciprocal and nonre-
ciprocal waveguides, IEEE Trans. Microwave Theory Tech.
39:18081816 (1991).
24. P. Kabos and V. S. Stalmachov, Magnetostatic Waves and
Their Applications, Chapman & Hall, 1994.
25. F. J. Rosenbaum, Integrated ferrimagnetic devices, in
L. Young and H. Sobol, eds., Advances in Microwaves, Aca-
demic Press, 1974, Vol. 8, pp. 203294.
26. F. E. Gardiol, Anisotropic slabs in rectangular waveguides,
IEEE Trans. Microwave Theory Tech. 18:461467 (1970).
27. F. E. Gardiol, Computer analysis of latching phase shifters in
rectangular waveguide, IEEE Trans. Microwave Theory Tech.
21:5761 (1973).
28. A. Mizobuchi and H. Kurebayashi, Nonreciprocal remanence
ferrite phase shifters using grooved waveguide, IEEE Trans.
Microwave Theory Tech. 26:10121016 (1978).
29. W. E. Hord, C. R. Boyd, and D. Diaz, A new type fast switch-
ing dual mode ferrite phase shifter, IEEE Trans. Microwave
Theory Tech. 35:12191225 (1987).
30. F. Reggia and E. G. Spencer, A new technique in ferrite phase
shifting for beam scanning of microwave antennas, Proc. IRE
45:15171570 (1957).
31. G. T. Roome and H. A. Hair, Thin ferrite devices for micro-
wave integrated circuits, IEEE Trans. Microwave Theory
Tech. 16:411420 (1968).
32. M. E. Hines, Reciprocal and non-reciprocal modes of propa-
gation in ferrite stripline and microstrip devices, IEEE Trans.
Microwave Theory Tech. 19:442451 (1971).
33. A. Beyer and K. Solbach, A new n-line ferrite isolator for
integrated millimeter-wave circuits, IEEE Trans. Microwave
Theory Tech. 29:13441348 (1981).
34. L. E. Davis and D. B. Sillars, Millimetric nonreciprocal cou-
pled-slot nline components, IEEE Trans. Microwave Theory
Tech. 34:804808 (1986); see also comments and reply in E.
Jemsen and C. Schieblich, IEEE Trans. Microwave Theory
Tech. 35:470471 (1987).
35. T. Kitazawa, Variational method for planar transmission
lines with anisotropic magnetic media, IEEE Trans. Micro-
wave Theory Tech. 37:17491754 (1989).
36. M. Horno, F. Mesa, F. Medina, and R. Marqus, Quasi-TEM
analysis of multilayered, multi-conductor coplanar structures
with dielectric and magnetic anisotropy including substrate
losses, IEEE Trans. Microwave Theory Tech. 38:10591068
(1990).
37. E. B. El-Sharawy and R. W. Jackson, Full wave analysis of
innitely long magnetic surface wave transducer, IEEE
Trans. Microwave Theory Tech. 38:730737 (1990).
38. S. H. Talisa and D. M. Bolle, On the modeling of the edge-
guided mode stripline isolators, IEEE Trans. Microwave The-
ory Tech. 27:584591 (1979).
39. D. M. Bolle and S. H. Talisa, The edge-guided mode nonre-
ciprocal phase shifter, IEEE Trans. Microwave Theory Tech.
27:878882 (1979).
40. T. M. F. Elshaey, J. T. Aberle, and E. B. El-Sharawy, Full
wave analysis of edge-guided mode microstrip isolator, IEEE
Trans. Microwave Theory Tech. 44:26612668 (1996).
41. R. Marques, F. Mesa, and F. Medina, Theory of magnetoelec-
tric multiconductor transmission lines with application to
chiral and gyrotropic lines, Microwave Opt. Technol. Lett.
38:39 (2003).
42. G. Bock, New multilayered slot-line structures with high
nonreciprocity, Electron. Lett. 19:966968 (1983).
43. E. B. El-Sharawy and R. W. Jackson, Coplanar waveguide
and slot-line on magnetic substrates: Analysis and experi-
ment, IEEE Trans. Microwave Theory Tech. 36:10711078
(1988).
44. E. B. El-Sharawy and C. J. Koza, Dual-ferrite slot line for
broadband, high-nonreciprocity phase shifters, IEEE Trans.
Microwave Theory Tech. 39:22042210 (1991).
45. F. Mesa, R. Marqus, and M. Horno, An efcient numerical
spectral domain method to analyze a large class of nonrecip-
rocal planar transmission lines, IEEE Trans. Microwave The-
ory Tech. 40:16301641 (1992).
46. D. Mirshekar-Syahkal, Spectral Domain Method for Micro-
wave Integrated Circuits, Wiley, 1990.
1496 FERRITE-LOADED WAVEGUIDES
47. C. M. Krowne, Fourier transform matrix method of nding
propagation characteristics of complex anisotropic layered
media, IEEE Trans. Microwave Theory Tech. 32:16171625
(1984).
48. F. Mesa, R. Marques, and M. Horno, A general algorithm for
computing the bidimensional spectral Greens dyad in multi-
layered complex bianisotropic media: The equivalent bound-
ary method, IEEE Trans. Microwave Theory Tech. 39:1640
1649 (1991).
49. T. Kitazawa, Analysis of shielded striplines and nlines with
nite metallization thickness containing magnetized ferrites,
IEEE Trans. Microwave Theory Tech. 39:7074 (1991).
50. Z. Fan and R. Pennock, Analysis of asymmetrical multilayer
ferrite-loaded nlines by the extended spectral domain ap-
proach, IEEE Trans. Microwave Theory Tech. 44:497504
(1996).
FERRITE PHASE SHIFTERS
WILLIAM E. HORD
Microwave Applications Group
Santa Maria, CA
Phase shifters are used extensively in the microwave and
millimeter-wave region primarily as array elements in
phased-array antennas. The rst application of ferrite
phase shifters in antenna arrays was during the decade
of the 1960s. Since then growth has been dramatic with
military applications being the motivating force. The air
defense systems of the former Soviet Union are designed
and built around ferrite phase shifting devices as are sev-
eral of the ground-based, naval, and airborne systems of
the United States. The article describes the evolution of
ferrite phase shifters that have made the transition from
the research laboratory to the production oor. Wherever
possible, the author has identied the systems that use a
specific type device.
Ferrite phase shifters are two-port devices operating in
the microwave and millimeter-frequency range between
1.4 and 100GHz. A variable insertion phase between in-
put port and output port is accomplished by varying the
bias magnetic eld of the ferrite material. The insertion
phase of a phase shifter is the phase delay experienced by
a radiofrequency (RF) signal propagating between port 1
and port 2 and is the angle of S
21
, the transmission coef-
cient from port 1 to port 2. If the angle of S
21
equals the
angle of S
12
, the transmission coefcient from port 2 to
port 1, the phase shifter is reciprocal, while if these two
angles are not equal the device is nonreciprocal. The
phase shifter consists of (1) a microwave circuit contain-
ing magnetized ferrite whose purpose is to provide a vari-
able insertion phase to the RF signal and (2) an electrical
circuit containing electronic components whose purpose is
to vary the magnetic bias of the ferrite and control the
amount of variable insertion phase. Because the state-of-
the-art of electronic control circuits changes rapidly de-
pending on device availability, the focus of the following
discussion will be on the microwave portion of the ferrite
phase shifter and the electronic control techniques will be
limited to basic principles.
The use of magnetized ferrite to provide the variable
phase shift was recognized as early as 1953 [1]. Phase
shifter applications were stimulated by the discovery of
the reciprocal, ferrite phase shifter in 1957 [2] and the
latching ferrite phase shifter in 1958 [3]. During the
1960s, major efforts were undertaken on phase shifter de-
velopment, and the toroid phase shifter [4] and dual-mode
phase shifter [5] evolved into their present congurations
during this decade. The rotary-eld phase shifter was
reported in the early 1970s [6].
Although they nd application in many areas, the ma-
jor use of ferrite phase shifters are as phase shifting ele-
ments in electronic scanning antennas where the data
rate is high enough to preclude the use of a mechanical
scanning antenna or where the aperture must be shared
by several functions requiring the antenna have an agile
beam shape. Antennas used for systems tracking large
numbers of targets such as the AWACS (airborne warning
and control system) require data rates not attainable with
mechanical scanning antennas. A ground-based air de-
fense system such as the Patriot must track a particular
target while continuing to search for other threats neces-
sitating an electronic scanning system. Ferrite phase
shifters are also used as feed elements for reector anten-
nas where the pattern of the reector may be changed by
changing the feed pattern providing different coverages.
Electronic control is desirable since the system may be lo-
cated in space and the reliability of mechanical switches is
not adequate. Other antenna feeds use four phase shifters
to provide a variable phase to each quadrant of the an-
tenna resulting in a conical scanning beam for the anten-
na. A third use is as the variable element in microwave
circuits used for high-power switches and variable-power
dividers and combiners. Because of the low insertion loss
and excellent phase accuracy attainable with ferrite phase
shifters, high-power electronic switches with insertion
loss as low as 0.5 dB and with isolations approaching
40 dB may be realized. A circuit providing variable
phase shift in each leg of a bridge circuit has been used
to combine the azimuth and elevation difference signals
from a monopulse antenna and by properly phasing the
signals compensate electronically for aircraft roll. Finally,
ferrite phase shifters have been used as Doppler simula-
tors, frequency translators, and so on.
1. FERRITE MATERIALS AND PROPERTIES
A material is called magnetic if it exhibits a magnetic mo-
ment in the absence of an applied magnetic eld. The
magnetic moment is due to the presence in the material of
at least two different electronic spin systems. If these spin
systems are equal and parallel, the material is ferromag-
netic; if the spin systems are equal and antiparallel, the
material is antiferromagnetic; and if the spin systems are
unequal and antiparallel, the material is ferrimagnetic
and is generically referred to as ferrite. Ferrite materials
are ionic crystals with no free electrons, resulting in
high resistivity and making them potentially useful for
FERRITE PHASE SHIFTERS 1497
application in the microwave and millimeter frequency
ranges. Two types of ferrites both with cubic crystalline
structure but one (spinels) having structure similar to
spinel and the other (garnets) having the garnet structure
have been used for phase shifter fabrication. A sample of
ferrite material of a size required for microwave compo-
nents usually does not exhibit a net magnetic moment in
the absence of an external bias eld. The material is com-
posed of magnetic domains; each of these exhibits a net
magnetic moment but is randomly aligned, resulting in
zero net magnetic moment when summed over the sample.
Application of an external magnetic bias eld rotates the
domains that align with the bias eld and produce a net
magnetization. When all domains in a sample are aligned,
the material is saturated and the magnetization is called
the saturation magnetization 4pM
S
. A virgin sample of
material exhibits a magnetization curve similar to that of
iron.
When the magnetizing current is removed, some mag-
netic ux may remain in the sample, and a current in the
opposite direction must be applied to reduce this ux to
zero. This phenomenon is called hysteresis. A ferrite ma-
terial formed in a closed loop exhibits a hysteresis loop
similar to that shown in Fig. 1, where the squareness of
the loop is a function of the chemical composition of the
material. The magnetic eld intensity required to reduce
the magnetic ux density to zero is called the coercive force
H
c
, while that magnetic ux density remaining after the
magnetic eld intensity has been reduced to zero is called
the remanent ux density B
r
. Magnetic material proper-
ties are sensitive to temperature; and above a certain
temperature, called the Curie temperature, the magnetic
properties vanish.
Ferrite phase shifters require values of saturation mag-
netization ranging from approximately 200 to 5500 G
(gauss) (the maximum attainable value with commercial-
ly available materials). By substituting aluminum ions for
ferric ions in YIG, the saturation magnetization may be
reduced from 1780 gauss (the value for pure YIG) to about
175 G. For the lithium spinel family the substitution of
titanium ions for ferric ions is used, and both aluminum
and zinc separately or in combination have been used to
vary the saturation magnetization for the magnesium
manganese ferrite family and the nickel ferrite family. In
general, when substitution is made the Curie temperature
is lowered from that of the unadulterated material. Dop-
ing with rare-earth ions may be used to increase peak
power capacity, although the insertion loss may increase
slightly.
Phase shifters providing variable values of insertion
phase operate with the ferrite partially magnetized. The
ferrite exhibits a tensor permeability whose on-diagonal
elements m vary slightly as a function of the applied mag-
netization. The off-diagonal elements of the tensor 7jk are
equal but of opposite sign, leading to the nonreciprocal
behavior of ferrite devices. These off-diagonal elements
are in phase quadrature with the on-diagonal elements
and are proportional to the ratio of applied magnetization
to saturation magnetization. If the RF energy is circularly
polarized, the effective permeability of the ferrite medium
is equal to (mk) for one sense of circular polarization and
(m k) for the other sense of circular polarization, both of
which are dependent on the applied magnetization
through the off-diagonal element of the permeability ten-
sor. The permittivity of the ferrite is scalar with a dielec-
tric constant in the range 1018. The dielectric loss
tangent of ferrite is about 0.0002 for garnets, 0.0003 for
magnesium spinels, 0.0005 for lithium spinels, and 0.001
for nickel spinels.
Phase shifter characteristics determined by the micro-
wave circuit are the mode of operation, either reciprocal or
nonreciprocal; operating frequency; instantaneous and
tunable bandwidth; polarization of the input and the out-
put signals; peak and average RF power; insertion loss
and modulation of the insertion loss; and return loss. Pa-
rameters determined by the microwave circuit and the
electrical control circuit are phase quantization, phase ac-
curacy, switching time, switching rate, and control power.
Physical parameters of the phase shifter include size;
weight; cooling requirements; input interfaces for RF sig-
nal, data, and control power; and output interfaces for RF
signal and built-in test. The phase shifter must conform
with environmental requirements such as operating and
storage temperature range, operating and transportation
shock, operating and transportation vibration, and oper-
ating and storage humidity. Finally, requirements gener-
ally exist for the reliability, interchangeability, and
maintainability of the phase shifter.
2. TYPES OF FERRITE PHASE SHIFTERS
Many ferrite phase shifters are described in the literature,
but only three types have been produced in quantity and
deployed in the eld in various antenna systems. The one
characteristic shared by these three different devices is an
insertion loss less than 1 dB. System and antenna design-
ers are unwilling to use devices with higher loss because of
B
r
H
c
B
H
H
c
B
r
Figure 1. Ferrite hysteresis loop for a square-loop material de-
ning the remanent magnetization B
r
and the coercive force H
c
.
1498 FERRITE PHASE SHIFTERS
reduction in antenna gain and cooling problems associated
with dissipation of the transmitter power in the antenna.
The J-STARS airborne surveillance system, the Patriot
mobile, ground-based air defense system, and the Aegis
naval air defense system use variations of the nonrecip-
rocal toroidal phase shifter. This device operates in a non-
reciprocal mode, requiring that the phase shifter be
switched immediately after the transmitter pulse to pro-
vide the proper phase shift for propagation in the receive
direction. The phase shifter is switched again just prior to
the next transmitter pulse, resulting in a switching rate
twice the pulse repetition rate of the radar. Switching time
is a few microseconds depending on operating frequency.
This type of phase shifter may be designed to have ex-
tremely wide operating bandwidth approaching 2 octaves
in some cases.
The second type of phase shifter, also nonreciprocal,
consists of a longitudinally magnetized ferrite rod located
on the axis of either square or circular waveguide. Several
of the ground-based, mobile air defense systems of the
Confederation of Independent States use these phase
shifters in antennas, radiating one sense of circular
polarization and receiving the orthogonal sense of circu-
lar polarizationthe single bounce return. An adapta-
tion of the phase shifter that provides reciprocal phase
shift is referred to as the dual mode device and has
been successfully employed in several antenna designs
most notably the multimode offensive radar systems
of the B-1B. Switching times for these phase shifters are
in the tens of microseconds depending on the frequency
of operation. Operating bandwidth is more modest
than the toroid phase shifter, typically being 1015%,
although 40% bandwidth has been achieved for experi-
mental devices.
The phase shifters described above provide a variable
insertion phase by varying the magnitude of the bias mag-
netic eld, resulting in a variation in the equivalent in-
ductance of the waveguide and yielding a variable
propagation delay through the device. The nal phase
shifter described in this section does not use this phenom-
enon but rather makes use of the variation in the direction
of the bias magnetic eld to effect change in insertion
phase with no change in the propagation delay. Because of
the similarity of the phase shifter to the rotary-vane [7]
mechanical phase shifter, it has been called the rotary-
eld phase shifter. The rotary-vane device uses a dielectric
vane to realize a half-wave plate that can be rotated
about the axis of a circular waveguide housing. A circu-
larly polarized RF signal receives phase shift when pass-
ing through the half-wave plate equal to twice the
mechanical angle of rotation of the half-wave plate. Sub-
stitution of a transversely magnetized ferrite rod for the
dielectric half-wave plate results in an electrically vari-
able version of this phase shifter. This phase shifter has
been employed in a single-axis scanning conguration for
the antenna for the AWACS surveillance aircraft and for
several single-axis scanning, ground-based, mobile air de-
fense systems. The switching time for the device is of the
order of hundreds of microseconds, and the operating
bandwidth is about the same as that of the dual-mode
phase shifter.
2.1. Toroidal Phase Shifters
The toroidal phase shifter consists of one or more ferrite
toroids inserted into a rectangular waveguide as shown in
Fig. 2. The cross section shown in Fig. 3a is the original
version of the toroidal phase shifter reported by Truehaft
and Silber (3) in 1958. The toroid is fabricated from a ma-
terial with a square hysteresis loop. Current owing in the
switching wire induces a magnetic ux in the toroid that
remains after the current is removed. The phase shifter is
said to be latched; operation in this mode is desirable
since control energy is required only when the phase shift-
er is set to a new state. The magnetic eld intensity of the
TE
10
mode in a rectangular waveguide is circularly polar-
ized in a longitudinal plane parallel to the narrow wall of
the waveguide and located a distance from the waveguide
centerline, which makes the longitudinal magnetic eld
intensity equal in magnitude to the transverse magnetic
eld intensity. The opposite sense of circular polarization
exists in the longitudinal plane located the same distance
on the other side of the waveguide centerline. If a sample
of magnetized ferrite is placed in the region of circular
polarization, a strong interaction between the RF and the
ferrite will occur, provided the direction of the bias eld is
interchanged on either side of the waveguide centerline.
The geometry shown in Fig. 3b makes more efcient
use of the ferrite toroid and provides more total phase shift
than the geometry of Fig. 3a at a minimal increase in in-
sertion loss by loading the window in the toroid with a
dielectric. The dielectric loading effect of the ferrite toroid
and dielectric used to load the toroid window distorts the
behavior of the elds of the waveguide mode so that the
region of circular polarization no longer lies in the same
plane as that of the air-lled waveguide. Computer-aided
analysis programs have been evolved [8,9] to predict the
performance of the phase shifter as a function of toroid
placement and dielectric loading. A disadvantage of the
geometries shown in Figs. 3a and 3b is the excitation of
the TEM mode, which is easily established by the switch-
ing wire coupling to the rectangular waveguide. Two
toroids separated by a dielectric slab spacer as shown in
Switching wire
Ferrite toroid
Rectangular waveguide
Figure 2. Prototypical toroidal phase shifter. First described in
the 1950s, this device was the rst latching, ferrite phase shifter.
By pulsing the switching wire the ux may be latched to either
B
r
or B
r
. Sustaining drive current is not required.
FERRITE PHASE SHIFTERS 1499
Fig. 3c reduce this coupling by concentrating the RF en-
ergy in the region adjacent to the dielectric slab and lo-
cating the switching wires in regions of very low RF
energy. This geometry is referred to as the dual-toroid
and has come to be the preferred realization for this class
device.
The ferritedielectric composite is housed in a rectan-
gular waveguide that typically has a cross section smaller
than that of the connecting waveguide because of the di-
electric loading. Quarter-wave transformers are used to
match into and out of the ferritedielectric composite.
These transformers increase the length of the device but
do not contribute significantly to the insertion loss. When
the connecting transmission line is a TEM-type line such
as microstrip, a high dielectric constant (E100) material
may be used as the dielectric spacer to reduce the cross-
sectional dimensions of the phase shifter and lower the
characteristic impedance level to around 50 O. This has
the added benet of reducing the length of the device be-
cause of the increase in electrical length caused by the di-
electric loading. The ferrite toroids are bonded to the high-
dielectric-constant rib, and the composite is coated with a
conducting material to form the waveguide. Connection to
the TEM line may be made with a short length of wire
with a chip capacitor located at the point of connection of
the wire to the phase shifter in order to resonate the in-
ductance of the wire loop.
In order to establish a reference condition for the phase
shifter the toroid is reset; that is, a voltage pulse of mag-
nitude and duration sufcient to saturate the toroidal core
is applied to the control wire. The current in the control
wire remains roughly constant until the core saturates, at
which time the current increases sharply. Sensing the
current and removing the drive voltage when a predeter-
mined current has been attained allows the core to relax to
the remanent ux and establishes a stable reference point.
The magnitude of the voltage pulse is not critical for the
reset operation. The set operation establishes a ux level
in the core corresponding to a given value of phase shift.
Faradays law states that the change of ux is equal to the
time integral of the applied voltage; a variable ux level
may be set by using a variable amplitude voltage pulse for
a xed time duration or a xed amplitude voltage pulse for
a variable time duration. If either the voltage or the pulse-
width varies substantially from that used to calibrate the
phase shifter, the error in setting the ux may be exces-
sive and resorting to more complicated methods such as
integration of the voltage pulse may be necessary. Typical
switching waveforms are shown in Fig. 4.
During switching, the toroidal core presents a resistive
load to the source. Application of a voltage pulse to the
control wire results in a current pulse whose amplitude is
determined by Amperes law, NI
_
H dl. The magnetic
eld intensity is constant and equal to the coercive force,
the number of turns is unity, and the integral of dl is equal
to the mean length around the ferrite toroid. Thus I H
c
l.
The ux change from one remanent state to the other re-
manent state is 2B
r
A, where B
r
is the remanent ux den-
sity and A is the cross-sectional area of the toroid normal
to the direction of ux. For a constant voltage Vapplied to
the core for a time T, VT2B
r
A, yielding a switching time
T2B
r
A/V. This is the maximum value of the time to es-
tablish the reset condition; the total switching time would
Air Air
Air
Switching wire
Ferrite
Air Air
Switching wire
Switching wire
Ferrite
Dielectric
Dielectric
Ferrite
Ferrite
(a)
(b)
(c)
Air Air
Figure 3. Evolution of the toroidal phase shifter into the twin-
toroid device. The window inside the toroid is loaded with high
dielectric material in order to improve the phase shifter RF per-
formance in (b) while in (c) another toroid is added that greatly
decouples the switching wires from the RF eld.
Voltage
Current
Reset Set
t
t
Figure 4. Switching waveforms of the latching phase shifter
when driven from a constant voltage source. The voltage remains
approximately constant during switching until the ferrite satu-
rates. Saturation of the core to obtain a stable reset state is ex-
hibited by the abrupt rise in the current in the reset waveform.
The area under the voltage waveform for the set pulse determines
the amount of ux switched into the ferrite core and hence the
amount of phase shift.
1500 FERRITE PHASE SHIFTERS
be at least twice this value. Increasing the applied voltage
reduces the switching time and increases the dynamic
core resistance, which is given by R2B
r
A/(TH
c
l).
2.2. Dual-Mode Phase Shifters
Latching operation of the ferrite rod phase shifter is real-
ized by lling the waveguide completely with ferrite and
providing a magnetic return path for the bias ux through
the use of external ferrite yokes as shown in Fig. 5. The
control power is furnished by a coil that is wound around
the waveguide. Again the insertion phase of the device is
dependent on the value of magnetic ux existing in the
ferrite rod, and variable phase is realized by changing this
value. The RF energy propagating through the ferrite rod
must be circularly polarized, which may require the input
polarization be converted from linear polarization to cir-
cular polarization. Each sense of circular polarization re-
ceives a different value of insertion phase when
propagated through the device, and these values are in-
terchanged when either the direction of propagation or the
direction of magnetization is reversed. However, if an an-
tenna uses these phase shifters and receives the single
bounce return, transmitting right-hand circular and re-
ceiving left-hand circular or vice versa, switching between
transmit and receive is not required.
Adaptation of the ferrite rod phase shifter to the recip-
rocal dual-mode phase shifter is illustrated schematically
in Fig. 6, and the physical realization of the phase shifter
is shown in Fig. 7. Nonreciprocal circular polarizers
(NCPs) located on either end of the ferrite rod function
as the circulators shown in the schematic diagram con-
verting linearly polarized RF energy into circularly polar-
ized RF energy and vice versa. The NCP physically
consists of a section of the ferrite rod that is transverse-
ly magnetized with a four-pole bias eld by a permanent
magnet located exterior to the microwave circuit. This
four-pole bias eld provides a differential phase shift of 901
to cross-polarized signals. If the input to the NCP is
linearly polarized at an angle of 451 with respect to the
axis of the NCP, the output will be circularly polarized
with the sense of circular polarization depending on the
orientation of the input linear polarization. For a circu-
larly polarized input signal, the output of the NCP will be
linearly polarized with orientation depending on the sense
of the circularly polarized input signal. The NCP on the
left side of the device in Fig. 7 converts input linear po-
larization into right-hand circular polarization in the fer-
rite rod section when RF energy is propagated from left to
right. This circularly polarized energy receives a variable
value of insertion phase dependent on the magnitude of
the remanent bias ux density in the ferrite rod. The cir-
cular polarization is then restored to linear polarization by
the NCP on the right side of the gure. For propagation
from right to left, the RF energy in the ferrite rod is con-
verted to left-hand circular polarization by the NCP on the
right side of the gure, resulting in the signal receiving
the same value of variable insertion phase irrespective of
the direction of propagation through the device. The sig-
nal is restored to linear polarization by the NCP located on
the left side of the device.
The electronic control of the dual-mode device is similar
to that used for the toroid, but two major exceptions exist.
Figure 5. The Faraday rotation phase shifter. The ferrite-lled
circular waveguide provides the signal path for the RF energy.
Longitudinal magnetization is obtained using a drive coil
wrapped around the ferrite rod. Latching operation is realized
using latching yokes for a return path for the bias magnetic ux.
1
[
2
[
2
[
2
[
Figure 6. A reciprocal phase shifter realized using four nonre-
ciprocal components. The two circulators and two nonreciprocal
phase shifters provide port-to-port reciprocal behavior.
Nonreciprocal
circular polarizer;
2 places
Ferrite-filled
circular waveguide
Drive coil
Latching yoke
Dielectric transformer
Waveguide flange;
2 places
Latching yoke
Figure 7. Physical realization of the dual-mode reciprocal phase
shifter. This compact realization of the schematic shown in Fig. 6
has proved to be compatible with the packaging requirements for
electronic scanned phased array antennas with wide-scan-angle
requirements.
FERRITE PHASE SHIFTERS 1501
The drive coil is almost always a multiturn coil, which re-
sults in the apparent resistance and inductance of the fer-
rite core being increased by the square of the number of
turns of the drive coil. Second, the waveguide walls do not
enclose the magnetic circuit in its entirety. The magnetic
ux cuts through the waveguide walls as it passes from
the ferrite rod and is returned through the external yokes,
resulting in an induced voltage in the waveguide walls
whenever the ux is changed. Since the waveguide walls
are made of low-resistivity material, a low-resistance path
allows eddy currents to ow in the waveguide walls and
produces the phenomenon called shorted-turn damping.
This effect may be modeled by including a parallel resis-
tance due to this damping in the equivalent-circuit of the
phase shifter.
2.3. Rotary-Field Phase Shifters
The geometry of the rotary-eld phase shifter is shown in
Fig. 8. A composite ferritedielectric rod is metallized with
a thin metallic coating to form the microwave portion of
the circuit. This is inserted into a laminated steel yoke
that provides the variable magnetic bias eld. The rod
yoke assembly is housed in a two-piece metallic housing
that interfaces to standard rectangular waveguide. Two
interlaced windings wound on the multislot yoke generate
the four-pole bias eld. Dielectric quarter-wave plates on
either end of the ferrite rod convert linearly polarized RF
energy to circularly polarized energy, and vice versa, for
propagation through the ferrite half-wave section.
The linearly polarized RF input signal is converted to
circularly polarized energy by means of the dielectric
quarter-wave plate. This circularly polarized energy pass-
es through the ferrite half-wave plate and receives a
variable phase shift dependent on the orientation of the
ferrite half-wave plate. At the output of the ferrite half-
wave plate the sense of circular polarization is reversed,
allowing the output dielectric quarter-wave plate to
reconstitute the same sense of linear polarization as the
input polarization. The phase shift through the device is
the same for either direction of propagation so that it is
referred to as a reciprocal device. In the strictest sense it is
nonreciprocal since a xed 1801 phase shift exists between
signals propagating through it in opposite directions.
Latching operation of the rotary-eld device was re-
ported in 1995 [10] and units presently deployed operate
with continuous holding current, resulting in a substan-
tial DC power supply for array applications. This has lim-
ited the device to single-axis electronic scanning antenna
applications such as the surveillance antenna for the
AWACS aircraft. Typically the electronic control for these
devices are two parallel drivers to control the two inde-
pendent windings on the yoke, with the control current on
one winding set proportional to the cosine of the desired
phase angle and the control current on the remaining
winding set proportional to the sine of the desired phase
angle.
3. PHASE SHIFTER CHARACTERISTICS
Most phase shifter designs are custom designs having
been developed for specific purposes and programs. Spe-
cific operating parameters will not be provided, but rather
general electrical and physical characteristics will be dis-
cussed. Finally, typical numbers are provided for the
reliability of the devices.
3.1. Electrical Characteristics
Electrical characteristics of importance are the operating
mode (reciprocal/nonreciprocal), the operating frequency,
the instantaneous bandwidth, the tunable bandwidth, the
input polarization, the output polarization, the peak and
average RF power, the insertion loss and the insertion loss
modulation, the return loss, the quantization of the phase
shift, the phase accuracy, the switching time, the switch-
ing rate, and the control power. Although it would seem
that reciprocal operation would be preferred over nonre-
ciprocal operation, there are many more nonreciprocal
phase shifters deployed than reciprocal ones. The interac-
tion of ferrite with RF energy is nonreciprocal and histor-
ically the earlier successful phase shifters were the
nonreciprocal toroidal types. The choice of operating
mode is generally dictated by system requirements; and
in several cases such as communication satellites, nonre-
ciprocal operation is not a handicap.
The operating frequency is another choice of the system
designer with phase shifters having been developed from
1.4 to 94 GHz. The instantaneous bandwidth refers to the
frequency band over which the phase shift remains within
specied tolerances, while the tunable bandwidth refers to
the frequency band over which the phase shifter may be
adjusted to bring the phase shift within the specied tol-
erance. The nonreciprocal toroidal phase shifter has been
designed to yield two octaves of instantaneous bandwidth,
while the dual-mode phase shifter and the rotary-eld
phase shifter yield instantaneous bandwidths in the 24%
range with tunable bandwidth of the order of 15%.
Figure 8. The rotary-eld phase shifter. This device provides
excellent phase accuracy and is capable of relatively high values
of RF power.
1502 FERRITE PHASE SHIFTERS
The input and output polarization depends on the ap-
plication for which the phase shifter is intended and may
be either linear or circular or switchable between the var-
ious linear and circular polarizations. The nonreciprocal
toroidal phase shifter operates in a linearly polarized
waveguide mode so that the input and output polariza-
tions are linearly polarized. The dual-mode phase shifter
may use circularly polarized inputoutput polarization
operating in the non-reciprocal mode or linearly polarized
inputoutput polarizations when operating in the recipro-
cal mode. This phase shifter lends itself well to incorpo-
rating polarization switching so that various output
polarizations are available even when the phase shifter
is excited with a linearly polarized input. The rotary-eld
phase shifter usually uses linearly polarized input and
output signals, although this is not required.
The peak RF power capacity of a ferrite phase shifter is
determined by the choice of ferrite used to realize the
phase shifter. If the RF magnetic eld intensity exceeds a
threshold value, excitation of spin waves results and the
RF insertion loss increases substantially. Doping of garnet
material with rare-earth ions may be used to increase the
threshold value, but at the expense of increased low power
insertion loss. The average RF power capacity of a phase
shifter is determined by the mechanical design and may
be increased only by improving the heat ow path away
from the ferrite.
The insertion loss of the phase shifter, as mentioned
previously, should be below 1 dB in order to merit consid-
eration from antenna designers. Of the phase shifters dis-
cussed, the rotary-eld device has the lowest loss with
values as low as 0.3 dB obtained in production quantities
for a device operating in the 5 GHz frequency range. The
toroidal phase shifter and the dual-mode phase shifters
have insertion loss in the range from 0.6 to 1.0 dB. The
variation of the insertion loss as a function of the insertion
phase of the device is greatest for the reciprocal devices,
generally of the order 0.20.4 dB, while the toroidal device
has loss modulation less than 0.1dB.
The return loss of the ferrite phase shifters depends on
the RF input and output connections, and values cited will
be for linearly polarized inputoutput congurations. The
toroidal phase shifter generally has a maximum value of
return loss of 20 dB, while the reciprocal phase shifters
have maximum values of return loss of the order of 14 to
17 dB.
Ferrite phase shifters are generally designed to provide
3601 of electrical phase shift. Early ferrite phase shifters
were realized using discrete lengths of ferrite to provide
quantization of the phase shift in steps of 1801, 901, 451,
and 22.51. This method results in a simple electronic driv-
er design but a complicated microwave structure. Modern
ferrite phase shifters are realized using a continuous piece
of ferrite in order to minimize the cost of the microwave
structure, resulting in a continuous range of phase shift.
Quantization is provided by the electronic driver. Since
the driver commands are distributed over the total range
of phase shift, which is often greater than 3601, the nal
quantization is 1 bit less than that provided by the elec-
tronic driver; that is, an 8-bit driver command results in
7-bit phase shift quantization. Quantization levels greater
than this are found in the control electronics of variable
power dividers/combiners but rarely are used in other
phase shifter applications.
The phase accuracy of the phase shifter refers to the
precision with which the insertion phase of the device may
be set. For the toroid phase shifter, this parameter is a
function of the stability of the reset state, the operating
frequency, and the operating temperature, and accuracies
of the order of 231 RMS error can be achieved. Improve-
ments in accuracy can be achieved at the expense of added
calibration. For the dual-mode phase shifter the align-
ment of the nonreciprocal polarizer magnets is another
source of phase error, and accuracies of the order of 341
RMS error are common with this device. The rotary-eld
phase shifter may be set very accurately since the inser-
tion phase is proportional to the ratio of the two currents
that control the rotation angle of the magnetic bias eld.
Typical phase accuracies for this device are in the range
11.51 RMS and are not particularly sensitive to frequency
and/or temperature.
The switching time of the phase shifter is the time re-
quired to establish a new insertion phase state and in-
cludes the times for resetting and setting the toroid and
dual-mode phase shifter. The switching time is a function
of the operating frequency since the size of the phase
shifter is dependent on the frequency. Typical switching
times for a toroid phase shifter range from about 20 ms at
2 GHz operating frequency to about 5ms at 20 GHz. Typ-
ical switching times for a dual-mode phase shifter range
from about 200ms at 5 GHz to about 50 ms at 20 GMz. The
nonlatching rotary-eld phase shifter requires switching
times ranging from 200ms at 3 GHz to about 100ms at
10 GHz but requires a high-voltage boost circuit in order
to attain these speeds. The switching rate of the phase
shifter is generally determined by system requirements,
where the control power is directly proportional to the
switching rate.
3.2. Physical Characteristics
Important physical characteristics of the phase shifter are
the size, the weight, the cooling requirements, and the
physical location of the input and output interfaces. The
size is dictated by the operating frequency, average power
requirements, and type of phase shifter. In general the
toroid phase shifter has the smallest cross section, and the
rotary-eld phase shifter has the largest cross section with
the dual-mode somewhere between the two. The weight of
the phase shifter is proportional to size; the toroid phase
shifter is generally the lightest weight unit, the rotary
eld the heaviest, and the dual-mode in between the two.
The RF interfaces are the input and output RF ports,
which may be coaxial cable, microstrip transmission line,
rectangular waveguide, radiating elements into free
space, or any combination of these. Electrical interfaces
consist of the input data and any output data, such as
built-in-test, control, power, and ground.
The electrical and mechanical design must be such that
the phase shifter meets the specied values over operating
temperatures ranging from 40 to 951C. In many cases
a reduced temperature range for full performance with
FERRITE PHASE SHIFTERS 1503
degraded performance over the temperature extremes is
allowed. Nonoperating temperatures normally range from
55 to 1251C. Operating shock and vibration require-
ments are determined by the mechanical design of the
system and the phase shifter mounting.
The reliability of the phase shifter as measured by the
mean time before failure (MTBF) is an important consid-
eration. The microwave portion of the phase shifter has
high reliability since it is composed of a ferrite core and
associated windings. The overall reliability is generally
determined by the electronic driver with values of roughly
200,000 h for the latching phase shifters and values of
50,000h for the nonlatching phase shifters because of the
requirement for continuous drive current.
4. FUTURE DEVELOPMENTS
Two developments in waveguide devices offer promise.
The rst is an attempt to provide a reciprocal phase shift-
er using the toroid phase shifter in a geometry similar to
that of the dual-mode device but using microstrip circula-
tors. The difculty with this approach is that the micro-
strip circulator is realized naturally as a three-port device
and the schematic diagram of Fig. 6 requires a four-port
circulator to adequately isolate the input and output ports.
Work continues in this development. The second develop-
ment in the waveguide area is the latching rotary-eld
phase shifter, which has been reported in the literature [6]
but has not been deployed in the eld. Data taken on ex-
perimental units are very encouraging.
There is a continuing effort to develop a ferrite phase
shifter in a planar geometry suitable for integration with
microstrip transmission line. The literature contains many
references to these devices, but the insertion loss continues
to be a drawback to deployment. The textbook cited in the
Further Reading section contains several examples of pla-
nar phase shifters as well as many references.
BIBLIOGRAPHY
1. N. G. Sakiotis and H. N. Chait, Ferrites at microwaves, Proc.
IRE 41:8793 (1953).
2. F. Reggia and E. G. Spencer, A new technique in ferrite phase
shifting for beam scanning of microwave antennas, Proc. IRE
45:15101517 (1957).
3. M. A. Treuhaft and L. M. Silber, Use of microwave ferrite to-
roids to eliminate external magnets and reduce switching
power, Proc. IRE 46:1538 (1958).
4. W. J. Ince and E. Stern, Non-reciprocal remanence phase
shifters in rectangular waveguide, IEEE Trans. Microwave
Theory Tech. 15:8795 (1967).
5. C. R. Boyd, Jr., A dual-mode latching reciprocal ferrite phase
shifter, IEEE Trans. Microwave Theory Tech. 18:11191124
(1970).
6. C. R. Boyd, Jr. and G. Klein, A precision analog duplexing
phase shifter, IEEE Int. Microwave Symp. Digest 248250
(1972).
7. A. G. Fox, An adjustable waveguide phase changer, Proc. IRE
35:14891498 (l947).
8. E. Schloemann, Theoretical analysis of twin-slab phase shift-
ers in rectangular waveguide, IEEE Trans. Microwave Theory
Tech. 14:1523 (1966).
9. J. L. Allen, D. R. Taft, and F. K. Hurd, Computer-aided design
of ferrite devices using intrinsic material parameters, J. Appl.
Phys. 38:14071408 (1967).
10. C. R. Boyd, Jr., A latching ferrite rotary-eld phase shifter,
IEEE Int. Microwave Symp. Digest 103106 (1995).
FURTHER READING
S. K. Koul and B. Bhat, Microwave and Millimeter Wave Phase
Shifters, Vol. I: Dielectric and Ferrite Phase Shifters, Artech
House, Norwood, MA, 1991.
FERROELECTRIC MATERIALS
R. K. PANDEY
The University of Alabama
Tuscaloosa, Alabama
1. INTRODUCTION
Before discussing the fundamentals of ferroelectric mate-
rials and their applications, it is important to acquire
some background on how any material, especially insula-
tors such as ferroelectrics, behave under the equilibrium
conditions given by forces such as electrical, mechanical,
and thermal. Each of these forces gives rise to certain
properties depending on the basic physical nature of the
material. For example, an electric eld (E) causes dis-
placement (D) [or polarization (P)], stress (T
ij
) causes
strain (S
ij
), and entropy (S) is caused by temperature
(T). These effects are not isolated from each other. In
fact, when the system is in equilibrium, they interact
with each other and give rise to additional properties. It
is helpful to understand the relationship between these
effects in order to appreciate the nature and properties of
electronic and other materials.
In 1925 Heckmann [1] proposed an equilibrium dia-
gram between electrical eld, temperature, and stress,
which was later modied by Nye [2] in 1957. A simplied
version of this diagram is given in Fig. 1. It is assumed
that the system is in equilibrium and the properties can be
described with reference to changes that are thermody-
namically reversible. Interaction between stress (T
ij
) and
displacement (D) causes direct piezoelectric effect whereas
the converse piezoelectric effect is caused by the strain
(S
ij
) and electric eld (E). The relationship between dis-
placement and temperature is called the pyroelectric effect.
The piezoelectric and pyroelectric effects are inherently
important for discussing ferroelectricity.
The interactions between the three principal agents
(E, T
ij
, and T ) within themselves and between the sec-
ondary agents such as displacement (D), strain (S
ij
), and
entropy (S) are responsible for materials to acquire certain
properties. Many of them are identied in Fig. 1. However,
for our purpose we need to consider only two more effects:
1504 FERROELECTRIC MATERIALS
(1) the thermal expansion effect, which is caused by the
interaction between strain (S
ij
) and temperature (T); and
(2) the heat of deformation, which originates from the in-
teraction between the strain (S
ij
) and entropy (S). Both
these, especially the thermal expansion coefcient, is a
very important parameter for the selection of a suitable
substrate for the growth of piezoelectric or ferroelectric
lms to produce integrated structured devices for many
novel applications.
The three parametersentropy, displacement, and
strainundergo small changes corresponding to the small
changes experienced by the agents temperature, electric
eld, and stress, respectively. Mathematically they can be
represented by the following three simple relationships
ds
C
v
T
dT 1
where c
v
is the specific heat per unit volume and it is as-
sumed that the system is in equilibrium and fully revers-
ible; and
dD
i
k
ij
dE
j
2
where k
ij
is the permittivity tensor and
dS
ij
s
ijkl
dT
ij
3
where s
ijkl
is the elastic compliance.
In discussing ferroelectric materials and associated
topics, we will make use of these concepts in gaining a
good insight into the eld of ferroelectricity.
The phenomenon now universally known as ferroelec-
tricity got its name more because of its phenomenological
similarity with ferromagnetism than because of the un-
derlying physics describing these two phenomena found in
materials. The name is deceptive in some sense. Ferro-
electricity, as the word implies, leads us to assume that it
has something to do with iron (ferum in Latin). In reality
it has hardly anything to do with this magnetic metal. In
fact only a handful of materials having iron (or other
prominent members of the ferromagnetic family such as
nickel and cobalt) have been reported to exhibit ferroelec-
tricity. However, they are certainly not members of the
mainstreamferroelectrics and are rarely researched active-
ly. Some examples are cadmium iron niobate, Cd
2
FeNbO
6
,
some members of the barium uoride group [e.g., barium
iron uoride (BaFeF
4
), barium cobalt uoride (BaCoF
4
),
and barium nickel uoride (BaNiF
4
)], and antimony sul-
de-iodide-type compounds such as iron sulde (FeS) [3,4].
Apart from the naturally found mineral FeS, another min-
eral termed ilmenite (FeTiO
3
), containing copper has also
been reported to be ferroelectric with a Curie point of
approximately 580K [5]. No other group has made such
a claim. This authors research group [6] has searched
extensively for ferroelectricity in laboratory-processed
doped ilmenite without any success.
Almost all well-known ferroelectric materials are man-
made (synthetic), in contrast to leading ferromagnetic ma-
terials, which are found abundantly in nature. Classic
examples are iron, cobalt, and nickel and their alloys. So
far as ferroelectricity is concerned only FeS and FeTiO
3
are naturally found in which the possibility of ferroelec-
tricity might exist. Further research is warranted to
establish the fact.
Ferromagnetism was already a well-established eld of
science and technology long before the birth of so-called
ferroelectricity. In 1921 Valasek [7] discovered the hyster-
esis effect between polarization and electric eld in potas-
sium sodium tartarate tetrahydrate (KNaC
4
H
4
O
6
.6 H
2
O),
which is commonly known as Rochelle salt after its dis-
coverer who lived in France in the seventeenth century.
Since its discovery over 80 years ago, the eld of ferro-
electricity has established itself as an important branch of

D
S
ij
Converse
Piezoelectric
Effect
Direct
Piezoelectric
Effect
Pyroelectric
Effect
E
T
T
ij
Thermal Expansion
Elasticity

Heat Capacity
Piezocaloric
Effect
Permittivity
Piezoelectricity

Pyroelectricity
Thermoelastic Effects
Electrocaloric
Effect
S
Heat of deformation
Figure 1. Modied equilibrium diagram between electric eld (E), temperature (T), and stress
(T
ij
) and their associated effects.
FERROELECTRIC MATERIALS 1505
physics and engineering. During this span of time, hun-
dreds of new ferroelectric materials have been discovered
leading to the emergence of novel technologies that exploit
various properties commonly found in ferroelectrics. The
range of applications of this class of materials is vast, en-
compassing the spectrum from classical to ultramodern
applications. On one hand, because of their high dielectric
constant, ruggedness, and the reliability with which the
materials can be processed reproducibly in very large vol-
umes, ferroelectrics are the materials of choice for the
fabrication of capacitors varying in size from miniature,
used in microelectronics, to large, used in power circuits.
On the other extreme, they nd applications in such mod-
ern technologies as electrooptics, nonvolatile memory, un-
cooled focal plane arrays, microelectromechanical system
(MEMS), and wireless communication.
The physical mechanisms involved in these two elds
are entirely different and have practically nothing to do
with each other. While ferromagnetism is anchored in the
quantum-mechanical properties of an electron, especially
its spin, ferroelectricity is macroscopic in nature, originat-
ing from the long-range interactions of the electric dipoles
and the noncentrosymmetry of the unit cell of the crystal.
The phenomenological similarity between these two
mechanisms goes beyond the existence of a hysteresis
loop. Like ferromagnetism, ferroelectricity also shows
the existence of the Curie point, at which it ceases to be
ferroelectric and enters the nonpolar phase called the pa-
raelectric state, analogous to paramagnetism. Further-
more, as there are antiferromagnetic materials, there
are also antiferroelectric materials. Both ferromagnetic
and ferroelectric materials have domains, and their struc-
tures help in describing terms such as paramagnetism and
paraelectric as well as antiferromagnetism and antiferro-
electricity. Piezoelectricity and pyroelectricity are inher-
ently present in all ferroelectric materials and are
strongly coupled with each other. In other words, all fer-
roelectrics are piezoelectric as well as pyroelectric, where-
as not all piezoelectrics are ferroelectrics; nor are all
pyroelectrics also ferroelectrics. The simultaneous pres-
ence of piezoelectricity and pyroelectricity in a ferroelec-
tric material makes it truly a multifunctional material
and therefore very attractive for the emerging elds of
applications where multifunctionality is the most desired
property of a material. Ferromagnetic and ferroelectric
materials are not the only ones that display hysteresis
loops. There is another class of material in which a hys-
teresis loop exists between the stress and the strain; these
are called ferroelastic materials. Grouped together, ferro-
magnetic, ferroelectric, and ferroelastic are called ferroic
materials.
According to the IEEE standard definitions [8], ferro-
electrics, ferroelastics, and ferromagnetics dene the class
of primary ferroics. It is entirely possible to switch the di-
rection of spontaneous magnetization, spontaneous polar-
ization, or spontaneous strain by applying magnetic or
electric elds or stresses, respectively. Figure 2 shows the
idealized examples of the three types of hysteresis loops
for typical (a) ferromagnet, (b) ferroelectric, and (c) ferro-
elastic materials [8].
2. FERROELECTRICITY AND ASSOCIATED PHENOMENA
2.1. Background Information
Ferroelectric materials are important members of the fer-
roic group. Compared to magnetism, they represent a rel-
atively new eld and yet have significantly impacted the
development of electronic technology. As described in the
preceding section, ferroelectric materials are truly multi-
functional in nature because at least two other important
phenomena, namely, piezoelectricity and pyroelectricity,
coexist with them. Even a casual inspection of Fig. 1
makes it obvious that cause and effect between the elec-
tric eld (E), temperature (T), and mechanical stress (T
ij
)
respectively play a dominant role in inducing dielectric,
electric, and thermal properties in any material. Ferro-
electricity is a typical example of such a manifestation.
Because of the interplay between these forces, special ef-
fects take place, giving rise to distortion of the unit cell of a
crystal structure and rendering it noncentrosymmetric.
Table 1 and Fig. 3 will help in understanding the back-
ground of crystal symmetry, structure, and other classi-
cations.
As shown in Table 1, the seven crystal systems are di-
vided into three groups based on their optical classica-
tion. Isotropic means that the refractive index of the
cubic crystal remains unchanged irrespective of whether
it is measured in the a, b, or c direction of the crystal; thus,
cubic crystal has no birefringence. The terms tetragonal,
hexagonal, and trigonal are lumped together as uniaxial,
meaning that they have one value of birefringence be-
cause they can have two different values of refractive in-
dex when measured along the different crystallographic
T
31(0)
T
31
S
0
M
M
r
H
c
E
3
(a) (b) (c)
H
3
P
r
P
E
c
S
31
=S
13
Figure 2. Idealized hysteresis loops of typical
(a) ferromagnet, (b) ferroelectric, and (c) ferro-
elastic materials [4].
1506 FERROELECTRIC MATERIALS
directions a, b, or c. Similarly, because of three possible
values of refractive index, the biaxial crystals can have
maximum of two values of birefringence. Such a classi-
cation is very important for determining the electrooptic
properties of ferroelectrics and other electrooptic crystals.
The third column in Table 1 gives the number of charac-
teristic symmetry for each crystal system. As we can imag-
ine, the order of symmetry is the highest in the cubic
system and the lowest in triclinic. Progressively from the
lowest to the highest order, it goes from triclinic to cubic
system as indicated by the arrow in the last (rightmost)
column.
It is a well-established fact of crystal physics that out
of the seven fundamental crystal systems one can
generate 14 fundamental three-dimensional (3D) congu-
rations called the unit cells. It was the French physicist
Auguste Bravais in 1850 who rst demonstrated that,
according to the periodic arrangements of atoms found
in a crystal, there could be only 14 possible arrangements
of atoms in space. These are the famous 14 Bravais
lattices, which are also known as fundamental unit
cells. They can be in four types: primitive (P), body-cen-
tered (I), face-centered (F) and base-centered (C). For
example, a cubic crystal can exist as simple cubic, or
body-centered, or face-centered. Only in the orthorhombic
structure are all four classes of unit cells found. Out of
the possible 14 Bravais lattices, 32 point groups originate.
These are possible combinations of macroscopic symmetry
elements, and once again, on the basis of the atomic
periodicity of crystal, there can be only 32 point
groups.
In Fig. 3 we show how these 32 point groups can be
subdivided into groups of centrosymmetric and noncen-
trosymmetric unit cells. Out of these 32 classes, 11 are
centrosymmetric and 21 noncentrosymmetric. These 21
are theoretically capable of exhibiting either ferroelectric-
ity, piezoelectricity, or pyroelectricity, or a combination
thereof. In reality only 20 of these 21 noncentrosymmetric
crystals do show these properties. One of them has other
symmetry elements, making this a special class of point
groups.
Noncentrosymmetry of the unit cell of a crystal is a
necessary condition for the existence of ferroelectricity, pi-
ezoelectricity and pyroelectricity. Figures 4a and 4b show
a simple representation of a centrosymmetric crystal be-
fore and after being subjected to a mechanical force [9].
Obviously, in the presence of an external force, the inter-
planar distance becomes smaller in the direction of the
applied force. But the crystal still retains its center of
symmetry. From Figs. 5a and 5b we also see that the crys-
tal retains its noncentrosymmetry before and after appli-
cation of an external force while experiencing, as in the
Table 1. Crystal System and Symmetry
Crystal System Optical Classication Characteristic Symmetry Order of Symmetry
Cubic Isotropic (also called anaxial) Four 3-fold axes Highest
Tetragonal Uniaxial One 4-fold axis
Hexagonal Uniaxial One 6-fold axis
Trigonal (also called
rhombohedral)
Uniaxial One 3-fold axis
Orthorhombic Biaxial Three mutually perpendicular 2-fold axes;
no axes of higher order
Monoclinic Biaxial One 2-fold axis
Triclinic Biaxial A center of symmetry or no symmetry at all Lowest
7 Crystal Systems
14 Bravais lattices
32 point groups
21 noncentrosymmetric
(polar groups)
11 centrosymmetric
(nonpolar groups)

7 crystal systems
Other symmetry elements still
can be present in 1 out of the 21
polar groups
10 can be spontaneously polarized
and show pyroelectricity as well as
piezoelectricity ; the spontaneous
polarization is not reversible
20 show piezoelectric effect
(polarized under mechanical stress)
Ferroelectricity is found in the remaining
10, and the spontaneous polarization
is reversible
Figure 3. Point groups and their subgroups.
FERROELECTRIC MATERIALS 1507
previous case, contraction in the interplanar distance
along the direction of the force [9].
2.2. Parameters, Symbols, and Units
As are many other elds of science, ferroelectricity is also
described by inconsistent symbols spread over decades of
literature. This causes confusion, which might lead to
misinterpretation of phenomena and the applications orig-
inating from their properties. To avoid such confusion, we
follow here the definitions and symbols recommended by
the Committee of the Institute of Electrical and Electron-
ics Engineers, Inc. (IEEE). This group was charged with
developing standard definitions of terms used in ferroelec-
tricity and associated polar phenomena of piezoelectricity
as well as pyroelectricity [8]. Important parameters, their
symbols and units are reproduced in Table 2.
2.3. Ferroelectricity
We have already discussed some of the general features of
ferroelectricity in order to grasp an understanding of its
differences with ferromagnetism. A ferroelectric material
is dened as a class of material in which the spontaneous
polarization exists even when no electric eld is applied to
it. It is capable of existing in two equivalent states that are
totally reversible, making ferroelectrics suited for many
applications.
There is another class of materials known as electrets,
which are often confused with ferroelectricity. These are
also dielectric materials in which quasipermanent real
charges can reside on the surface. They can also be found
in the bulk of the material, or as frozen-in aligned dipoles
in the bulk. Electrets behave in the same way as a battery
or an electrical counterpart of a permanent magnet. How-
ever, they are not ferroelectrics. Ferroelectric materials
are distinctly different from electrets in two respects:
(1) its polarization is spontaneous, which is retained
even at zero electric eld; and (2) this polarization is con-
trolled by the crystals symmetry.
From Fig. 3 we nd that the ferroelectric effect can be
present in the 10 noncentrosymmetric polar groups in
which the spontaneous polarization is reversible. Absence
of the center of symmetry is the essential condition for the
existence of not only ferroelectricity but also of piezoelec-
tricity and pyroelectricity. As we will see later, a piezo-
electric material assumes its polarization when subjected
to a mechanical stress. Here the polarization is not spon-
taneous. It disappears when the mechanical force is re-
moved. However, pyroelectrics are also spontaneously
polarized in the same way as ferroelectrics but are not
switchable. This is an important distinction between these
two closely related phenomena. Ferroelectrics are also dis-
tinguished from pyroelectrics in one additional important
wayonly in ferroelectrics does the spontaneous polariza-
tion disappear at a well-dened temperature, called the
Curie point. There is no concept of the Curie point in ei-
ther piezoelectric materials, pyroelectric materials, or
electrets. This characteristic temperature enables ferro-
electrics to exist in two distinct phases: the polar (i.e., fer-
roelectric) phase and the nonpolar (the paraelectric)
phase. Furthermore, ferroelectrics are well known to be
materials showing very high dielectric constant, some-
times in excess of 50,000. The uniqueness of ferroelectric
materials reects itself in many ways, which make them
(a) (b)
Force
+ +
+ +
_
_
__
+
+
+
+
+
_
_
_
_
_
+
Figure 4. Schematic representation of a unit cell with center of
symmetry (a) before and (b) after application of a mechanical force
[9].
(a) (b)
Force
+
+
+ +
+

+
+
+ +
+

Figure 5. Schematic representation of a unit cell with no center


of symmetry (a) before and (b) after application of a mechanical
force [9].
Table 2. Physical Properties and Their Symbols and Units
Terms Symbol
Units
(SI Units)
Curie constant C K (Kelvin)
Capacitance C
x
,C
p
,C
i
F (Farad)
Elastic stiffness constant c
ijkl
,c
pq
N/m
2
(newtons/meter
2
)
Electric displacement
(vector)
D
i
C/m
2
(coulombs/meter
2
)
Piezoelectric charge
(or strain) coefcient
d
ijk
,d
ij
C/N (or m/V)
Electric eld (vector) E
i
V/m (volts/meter)
Coercive eld (or coercivity) E
c
V/m
Piezoelectric stress
coefcient
e
ijk
,e
ij
C/m
2
Pyroelectric current i
p
A (ampere)
Dielectric polarization
(vector)
P
i
C/m
2
Remanent polarization P
r
C/m
2
Spontaneous polarization P
s
C/m
2
Maximum polarization P
max
C/m
2
Pyroelectric coefcient p
i
C/m
2
K
Strain (second-rank tensor) S
ij
Dimensionless
Elastic compliance constant s
ijkl
,s
pq
Dimensionless
Curie point T
c
K or 1C
Stress (second-rank tensor) T
ij
N/m
2
Permittivity of free space e
0
F/m
Permittivity (second-rank
tensor)
e
ij
F/m
Relative dielectric constant K
ij
Dimensionless
Source: Ref. [8].
1508 FERROELECTRIC MATERIALS
extremely attractive for diverse applications. We will dis-
cuss some of these later in this article. To sum up, the
characteristic features of a ferroelectric material are (1)
switchable spontaneous polarization, (2) high dielectric
constant, (3) the Curie point that clearly denes the tran-
sition between the polar and nonpolar states, (4) hyster-
esis loop between polarization and electric eld, and (5)
crystal structures that lack center of symmetry.
The relationship between barium titanate (BaTiO
3
)
and ferroelectricity is similar to that between silicon and
a semiconductor. It is the most prominent member of the
family of ferroelectricity and well established in technol-
ogy. It is isostructural with the mineral calcium titanate
(CaTiO
3
) and synthesizes in perovskite structure, com-
monly represented by the formula ABO
3
. The typical
structure of BaTiO
3
in its ferroelectric phase (below the
Curie point of 1201C) is shown in Fig. 6.
Here the Ba
2
ions occupy the eight corners of the unit
cell, which is based on close-packed face-centered cubic
crystal, the Ti
4
ions are located in the body-centered po-
sition, and the O
2
ions are at the face-centered positions.
The barium ions are coordinated with 12 oxygen ions and
the Ti ions in the octahedral interstices. At and above the
Curie point, the unit cell of BaTiO
3
undergoes a phase
transition from nonpolar to polar state. That is, it is no
more ferroelectric above the transition point. The pres-
ence of the Curie point is an important characteristic of all
ferroelectrics. Obviously, this is a material constant and
can vary for different materials.
The majority of the well-known ferroelectric materials
synthesize in perovskite structure. Some examples of tech-
nologically important ferroelectric materials with this
structure are K niobate, K tantalate niobate (KTN), Pb
titanate, and Pb zirconate-titanate (PZT). It is interesting
to note that besides ferroelectrics, many other electronic
materials of great scientific and technological importance
can also crystallize in this structure. This was realized in
the late 1980s, when high-temperature superconductivity
was discovered. Leading examples were prominent mem-
bers of the high-temperature superconductor family,
namely, 123 YBCO (YBaCu oxide) and the colossal mag-
netoresistive materials of the type LaCaCu oxide and
LaSrCu oxide. These discoveries have made the perovs-
kite group of materials a subject of intensive research in
attempts to discover some novel phenomenon.
Another prominent ferroelectric group is found in the
tungsten bronze structure shown in Fig. 7. They are rep-
resented by the generic formula of AB
2
O
6
. Leading mem-
bers of the family are Pb (meta)niobate (PbNb
2
O
6
), Pb
(meta)tantalate (PbTa
2
O
6
), PbK niobate (PKN) having the
formula (Pb
2
KNb
5
O
15
), and BaSr niobate (SBN) with the
formula of Ba
2
Sr
3
Nb
10
O
30
[4]. PKN has been identied to
have the largest piezoelectric coupling coefcient and was
researched heavily in the mid-1980s. However, it was al-
most impossible to grow good crystals without producing
multiple cracks. With the advancement of lm technology
and integration with silicon, PKN may once again receive
renewed attention. Among all the electrooptic materials,
SBN crystals show the most interesting properties and
was the choice material for the fabrication of delay lines.
Ferroelectric crystals, like their magnetic counterparts,
consist of a large number of domains in which the polar-
ization is oriented in one unique direction. The neighbor-
ing domains are separated from each other by domain
walls. All the polarization vectors point in one single
direction and are parallel to each other, resulting in the
maximum value of the spontaneous polarization for the
bulk of the material. This picture is attainable under two
conditions: (1) close to the absolute zero temperature and
(2) when subjected to a strong external electric
eld. Between absolute zero and the Curie point the do-
main structure goes from fully ordered state (maximum
polarization) to fully disordered state (zero polarization).
Figure 8 shows schematically the parallel and antiparallel
congurations of ferroelectric domains. Even at room
a
A1 site B1 site B2 site C site A2 site
b
Figure 7. The framework of the tungsten bronze structure look-
ing down the tetragonal c axis. The interstitial sites labeled A1,
A2, and C can accommodate A-type cations. The B-type cations
occupy the octahedron centers labeled B1 and B2 [10].
Ba
Ti O
Figure 6. Perovskite structure of BaTiO
3
below T
c
.
FERROELECTRIC MATERIALS 1509
temperature the energy associated with lattice vibration
is sufcient to destroy the ordered structure and cause the
domains to orient randomly, resulting in zero polarization
in the absence of an external eld.
The hysteresis loop of ferroelectrics is obtained by using
the simple circuit originally given by Sawyer and Tower
[11] in 1930 and later modied by Sinha [12]. The switch-
ing time between the two states of remanent polarization
is determined by the method proposed in 1966 by Fatuzo
and Fatuzo [13]. A typical hysteresis loop of a ferroelectric
material is shown in Fig. 9.
The material at room temperature has randomly ori-
ented domains, as discussed earlier, and therefore zero
value of the polarization. Initially when the eld is applied
and its magnitude is small, the P-versus-E curve is linear
because the eld induces the polarization here. This cor-
responds to the linear portion of the curve OA in Fig. 9
when the applied eld is very small. Once its magnitude
increases, an increasing number of domains orient in its
direction. Eventually the polarization enters the nonlin-
ear region and continues to increase with the eld until it
reaches its maximum value P
m
in the rst quadrant
labeled I. This path is followed as shown by the dashed
curve OA and called the virgin curve. After this the effect
of increasing eld is negligible. At point P
m
all the do-
mains are fully oriented in one unique direction as shown
by the upward-pointing arrows. The interpolation of the
uppermost curve in quadrant I cuts the polarization axis
at zero eld at point P
s
; P
s
is termed the spontaneous po-
larization. Strictly speaking, this definition is not exactly
correct because theoretically the value of the spontaneous
polarization is dened as the polarization at absolute zero.
However, the value derived from the hysteresis curve is
the value of the spontaneous polarization for all practical
purposes. In quadrant I, when the eld is reduced from its
maximum value at P
m
, it does not retrace its original path.
Instead, it shows hysteric effect and meets the polariza-
tion axis at P
r
. This is the value of the polarization re-
maining from its original value of P
m
at zero eld; it is
called the remenant polarization. Usually the ratio of P
r
to
P
m
is less than 1. Ferroelectric materials with this ratio
close to 1 are of great importance to memory applications.
On further reduction of the eld, in quadrant II, we see
that the polarization disappears at point E
c
; E
c
is called
the coercivity or coercive eld. In the second quadrant this
is the eld that is needed to bring the remenant polariza-
tion to zero. Here the hysteresis curve has completed its
half-cycle. Further reduction of the eld, in the third
quadrant (III) brings us to P
m
. Thus, here the sponta-
neous polarization has switched by 1801 as indicated by
the downward arrows representing the orientation of the
domains at this point. Once this point is reached, there is
no sense in increasing the eld in the negative direction
any further. The eld direction now is reversed and the
P-versus-E curve cuts the polarization axis at P
r
. From
the symmetry of the curve we observe that once again it is
the remanent polarization.
(a) (b)
Figure 8. Schematic of domain congurations of a ferroelectric
crystal: (a) parallel orientation (P
s
maximum) and (b) antipar-
allel (P
s
zero) orientaiton.
Pm
E (V/m)
Polarization
Electric field
Ec +Ec
III
II
I
IV
0
A
Pr
+Pr
Ps
+Pm
P(C/m
2
)
Figure 9. Polarization (P) versus electric eld (E) hysteresis loop of a ferroelectric material.
1510 FERROELECTRIC MATERIALS
The presence of two equivalent states of P
r
lends fer-
roelectrics bistable state similar to that found in ferro-
magnetism. This particular property is of great technical
significance and is the basis of nonvolatile ferroelectric
memory. This property is also exploited for many other
applications where memory effects are important. In the
fourth quadrant (IV) the increasing eld brings us back to
the coercive eld, E
c
. Further increase of the electric
eld allows the polarization to traverse the rst quadrant
nonlinearly. It eventually reaches the P
m
point. This
completes the full cycle of the loop.
No matter how many times the loops are generated,
they retrace their original paths. This reproducibility is
needed for designing devices based on the nonlinear prop-
erty of ferroelectrics. The shape of the loop is dependent on
the frequency of the AC electric eld applied. It is also
strongly dependent on temperature. At and very close to
the Curie temperature of the ferroelectric material, the
nonlinearity ceases to exist because of the disappearance
of the spontaneous polarization. For all practical purposes
it reduces to a straight line. The temperature at which the
hysteresis loop collapses is also a measure of the Curie
point. However, it gives only the approximate value. The
exact value is to be determined by careful measurement of
the spontaneous polarization as a function of temperature.
The temperature dependence of the spontaneous polariza-
tion is shown in Figs. 10 and 11 [8].
From Figs. 10 and 11 it becomes clear that at the Curie
point the material goes through a phase transition from a
polar to a nonpolar state. Phase transition can be of two
kinds: rst-order, in which the change is discontinuous;
and the second-order, in which the change is continuous.
For many ferroelectrics the rst-order transition occurs
when it goes through a change in crystal structure such as
from tetragonal to cubic. Such is the case for BaTiO
3
, PZT,
and KTN. Also, we observe that the Curie point is the di-
viding line between the polar and nonpolar states. Below
this temperature ferroelectricity and therefore the non-
linear behavior dominates whereas above it the material
is linear, as is any other dielectric. The polar and nonpolar
phases in a ferroelectric are totally reversible, and this can
be explained on the basis of domain theory; specifically,
above the Curie point the domain orientations are ran-
dom, and as the temperature is lowered below the Curie
point, the spontaneous polarization in these domains be-
gins to renucleate and the number of these domains con-
tinues to increase as the temperature is lowered. At
absolute zero they attain the highest order, allowing the
spontaneous polarization to reach its maximum possible
value.
The nature of the phase transition is also reected in
the temperature dependence of the relative dielectric con-
stant (K) as shown in Fig. 12. In each case, the value of K
goes to innity at the Curie point. Ideally above the Curie
point K
1
increases linearly with increase in tempera-
ture. This is predicted by the CurieWeiss law
K
C
T T
c
4
In Fig. 12(a), y is the asymptotic Curie point at which K
reaches its maximum value and in almost all cases it is
close to the value of the Curie point determined by other
P
s
T
c
T
Cooling
(Ferro)
(Ferro)
(Para)
(Para)
Heating
Figure 10. Spontaneous polarization versus temperature for
rst-order phase transition [8].
(Ferro)
P
s
T
c
T
(Para)
Figure 11. Spontaneous polarization versus temperature for
second-order phase transition [8].
T
c
T
c
1 1
0
(a) (b)
T T
0

Figure 12. Temperature dependence of inverse of relative dielec-
tric constant of a ferroelectric material undergoing phase trans-
formations at the Curie point of the (a) rst order and (b) second
order [8].
FERROELECTRIC MATERIALS 1511
experiments. It should also be emphasized that not all
ferroelectrics follow the CurieWeiss law strictly.
The capacitancevoltage plot, shown in Fig. 13, is also a
good indication of the materials quality of a ferroelectric.
Here too we observe that the nature of the curve is hys-
teric with respect to the decreasing and increasing bias
voltage. The width of the peak of the curves, as shown be
dashed lines, is a measure of the coercive force. Selected
examples of ferroelectric materials and their properties
are presented in Table 3.
2.4. Piezoelectricity
Like many other terms of science, this, too, is derived from
a Greek word meaning to press. As we have seen in Fig.
1, the relationship between the mechanical stress (T
ij
) and
the electric eld (E) gives rise to piezoelectricity. Also, we
nd from this gure that there are two types of piezoelec-
tric effects: direct and converse. Direct piezoelectric effect
occurs because of the interaction between the mechanical
stress (T
ij
) and the displacement (D). Similarly the inter-
action between the electric eld (E) and the strain (S
ij
)
gives rise to the converse effect. Both these piezoelectric
effects have significant importance in technology.
If a stress is applied to a piezoelectric crystal, it devel-
ops an electric moment per unit volume (or charge per unit
area). The magnitude of polarization is proportional to the
stress applied. This is called direct piezoelectric effect.
Mathematically this is represented by the following
equation
PdT 5
where d is the piezoelectric coefcient, P the polarization,
and T the stress; or, more precisely as
DP
i
d
ijk
DT
jk
6
800
750
700
650
600
20 15 10 5 0 5 10 15 20
C
a
p
a
c
i
t
a
n
c
e

(
P
F
)
Voltage (V)
E
c
Figure 13. Capacitance versus bias voltage plot of PNZT solgel
lm [14].
Table 3. Selected Ferroelectric Materials and Their Properties
Name (Abbreviation) Chemical Formula
Curie Temperature
(1C)
Spontaneous
Polarization P
s
(mC/m
2
) at (1C)
Crystal Structure
4T
C
oT
C
Barium titanate BaTiO
3
120 26.0 (23) Cubic Tetragonal
Lead titanate PbTiO
3
490 50.0 (23) Cubic Tetragonal
Potassium niobate KNbO
3
435 30.0 (250) Cubic Tetragonal
Potassium dihydrogen
phosphate (KDP)
KH
2
PO
4
150 4.8 ( 177) Tetragonal Orthorhombic
Triglycine sulfate
(TGS)
(NH
2
CH
2
COOH)
3
H
2
SO
4
49 2.8 (20) Monoclinic
(centrosymmetric)
Monoclinic
(noncentrosymmetric)
Potassium sodium
tartrate tetrahy-
drate (Rochelle salt)
KNaC
4
H
4
O
6
. 4H
2
O 24 0.25 (5) Orthorhombic
(centrosymmetric)
Monoclinic
(noncentrosymmetric)
Lead zirconium
titanate (PZT) [21]
Pb(Zr,Ti)O
3
200480 Composition-depen-
dent
Cubic Rhombohedral or
Tetragonal (depends
on composition)
Lead niobium
zirconium titanate
(lm), PNZT
a
See below 400 58 (20) Cubic Tetragonal
Antimony sulfoiodide
(crystal)
SbSI 22 25.0 (5) Orthorhombic
(centrosymmetric)
Orthorhombic
(noncentrosymmetric)
Antimony sulfoiodide
b
SbSI 19 0.03 (19) Orthorhombic
(centrosymmetric)
Orthorhombic
(noncentrosymmetric)
Potassium tantalate
niobate
c
(KTN)
See below 108 to 19 Composition-depen-
dent
Cubic
(centrosymmetric)
Tetragonal
(noncentrosymmetric)
a
Pb
1.1
Nb
0.04
Zr
02
Ti
0.8
O
3
[17,18].
b
PLD-grown lm [19].
c
KTN of vialbe compostions,KTa
(1. x)
Nb
x
O
3
, [20].
Source: Ref. 15.
1512 FERROELECTRIC MATERIALS
The converse piezoelectric effect takes place on the appli-
cation of an electric eld to a piezoelectric material that
induces the shape to deform (or, change) slightly. This is
also a linear effect and can be represented by the following
simple relationship
DS
jk
d
ijk
DE
i
7
where S is the strain and E the electric eld.
We have seen earlier that the elastic coefcient, the pi-
ezoelectric coefcient, and the dielectric constant are ten-
sors and there can be as many as 45 independent
coefcients. For example, the elastic constant can have
21 values; the dielectric constant, 6; and the piezoelectric
coefcient, 18. But in practice one need not worry about all
these different coefcients. One would usually apply the
electric eld, for example, in only one direction and mea-
sure the mechanical response in some other direction. For
all practical purposes, the following two simple equations
will sufce to describe piezoelectricity [9]
TcS dE 8
DeEdS 9
where e is the dielectric constant. The validity of these
equations is given by the following arguments.
If d0, that is there is no piezoelectric effects present
in the material, then we have TcS, which is simply the
famous Hookes law; and DeE, which is the most famil-
iar equation of electromagnetism. Alternatively, by setting
E0 in Eqs. (8) and (9) TcS, we once again get the fa-
mous Hookes law and DdS. The latter relation indi-
cates that even in the absence of an electric eld in a
piezoelectric crystal a nite amount of polarization can
develop on the application of strain. Similarly, an electric
eld can induce strain in a piezoelectric crystal even when
there is no mechanical stress applied.
In short, because of the piezoelectric properties, a
crystal develops strain when subjected to an electric
eld. Alterenatively, polarization develops under mechan-
ical stress. What follows from these arguments is that
piezoelectric materials are ideally suited for applications
related to electromechanical transducers. They are
the materials of choice for many MEMS devices. Some
examples of leading piezoelectric materials are shown in
Table 4.
2.5. Pyroelectricity Effect
We have already encountered the term pyroelectricity in
Figs. 1 and 3. The origin of this effect lies in the following
two facts: (1) interaction between the displacement and
temperature and (2) noncentrosymmetry of the unit cell.
Thus, pyroelectricity is caused when a crystal having a
noncentrosymmetric unit cell and spontaneous polariza-
tion undergoes a temperature change. The change in tem-
perature can induce a change in surface charge, which in
turn can cause a change in electrical polarization of the
material. The end result is the emergence of a thermal
current, which can be detected, in an external circuit.
From electromagnetic theory we know that when an
electric eld is applied to a polar material, the resultant
displacement is given by the following simple equation
De
0
EP
net
e
0
EP
s
P
ind

10
where D is the displacement, E the electric eld, e
0
the
permittivity of vacuum, and P
net
the net polarization.
Since the total polarization consists of both spontane-
ous component (P
s
) and induced part (P
ind
), we rewrite Eq.
(10) considering that P
s
cP
ind
as follows
D % eEP
s
11
where e is the dielectric constant of the material. Differ-
entiating this equation with respect to dT, we get
dD
dT
%
dP
s
dT
E
de
dT
12
assuming that the E remains constant. We obtain
p
eqv
% p
i
E
de
dT
13
where p
eqv
represents the equivalent pyroelectric coef-
cient and p
i
, the true pyroelectric coefcient.
It is obvious from Eq. (13) that the true pyroelectric
coefcient p
i
can be evaluated by differentiating the
P
s
temperature curves as shown in Figs. 10 and 11. It is
also obvious from Eq. (12) that the change in temperature
will induce change in the polarization vector. Therefore,
Table 4. Representative Piezoelectric Materials
Material Type Piezoelectric Constant C/N10
9
Relative Permittivity
Quartz,SiO
2
Crystal d
33
2.33 4.04.5
Polyvinylidene uoride (PVDF) Polymer d
31
23, d
33
1.59 12
Barium titanate (BaTiO
3
) Ceramic d
31
78 1700
crystal d
33
190 4100
Lead zirconate (PZ) Ceramic d
31
110, d
33
370 12003000
Zinc oxide (ZnO) Ceramic d
33
246 1400
Nb-doped PZT
a
(PNZT) Textured lm d
31
54 2200
a
For further details, see Refs. 16 and 18.
Source: Ref. 15.
FERROELECTRIC MATERIALS 1513
the pyroelectric coefcient can now be dened as
DP
i
p
i
DT 14
where i 1,2,3,y.
The basic definition of the true pyroelectric coefcient
is given by Eq. (14), whereas Eq. (13) gives the actual
pyroelectric coefcient measured. The contribution from
the second term representing the temperature coefcient
of the dielectric constant (de/dT) cannot be neglected if
the pyroelectric measurements are done on a ferroelectric
material for which, as we know, the temperature coef-
cient of the dielectric constant can assume large values.
We have already seen that the large values of the di-
electric constant are also one of the most important fea-
tures of a ferroelectric material and reaches its maximum
value at the Curie point. This would mean that the equiv-
alent pyroelectric coefcient (p
eqv
) would be large for a
ferroelectric with the Curie point near room temperature.
This is an important consideration while designing a pyro-
electric detector. Another important consideration in the
device design is the contribution made by the piezoelectric
coupling on the performance of the detector. Most of the
practical pyroelectrics are also piezoelectrics. Therefore
they will experience strain because of the thermal expan-
sion leading to the development of surface charges.
The thermal current (I
p
) that can be generated by heat-
ing uniformly a pyroelectric material is given by
i
p

dQ
dt
A
dP
s
dT
.
dT
dt
15
or
i
p
Ap
i
dT
dt
16
where A is the area of the electrode and p
i
the pyroelectric
coefcient. Here it is important to note that the heating of
the sample must take place at a uniform rate. The current
is generally determined when the experiment is conducted
in a dynamic mode. Alternatively, the pyroelectric coef-
cient can be evaluated directly by measuring the charge
(using an electrometer) that is generated at different tem-
peratures. The following equation can be used for this
purpose:
p
i

Q
2
Q
1
AT
2
T
1

17
Some examples of technologically important pyroelectric
materials are given in Table 5. These materials are used
mostly for infrared imaging.
2.6. Pyrooptic Effect
A lesser known property called the pyrooptic effect of di-
electric materials can also be potentially exploited for de-
tection of infrared. The temperature dependence of the
refractive index gives rise to this parameter. It is dened as
y
p

dZ
dT
_ _
18
where y
p
is the pyrooptic coefcient and Z the refractive
index.
Performance of a pyrooptic detector will be significantly
superior to pyroelectric or photon detector because no me-
tallic contacts to pixel elements are needed to operate this
device. Because it is an optical system, it allows for non-
contact readout. This leads to an ideal thermal detecting
structure as it eliminates a major source of signal-to-noise
ratio in the system. In addition, for a pyrooptic detector,
the sample thickness need not be greater than the opti-
mum thickness required for supporting the optical waves
on reection. Thus, the pixel volume (or, mass) can be very
small.
The concept of a pyrooptic detector was proposed in the
early 1990s [24,25]. The device can be built on a single-
crystal base or on lms of a pyrooptic material. The lm
can be freestanding or supported on a transparent sub-
strate having poor thermal conductivity.
Leading pyrooptic materials with their pyrooptic coef-
cients are presented in Table 6. As we can see from this
table, once again ferroelectrics appear to be the material
for the pyrooptic technology.
It is to be noted that in spite of many attractive features
and simplicity, no practical device has become commer-
cially available that is based on this effect. However, with
the phenomenal progress made since the mid-1990s in
processing of materials, especially lms, and in producing
Table 5. Representative Pyroelectric Materials
Materials Type Pyroelectric Coefcient/p
i
(mC m
2
K
1
) at 1C
Triglycine sulfate (TGS) Single crystal 280 (35)
Deuterated TGS Single crystal 550 (40)
Lithium tantalite (LiTaO
3
) Single crystal 230 (25)
SBT, (SrBa)Nb
2
O
6
Single crystal 550 (23)
Modied lead zirconate (PZ) Ceramic 400 (23)
PVDF Polymer lm 27 (25)
BST (Ba
0.65
Sr
0.35
)TiO
3
(bias eld0.6 V/mm) Ceramic 7000 (20)
PST,Pb(Sc
0.5
Ta
0.5
)O
3
(bias eld10V/mm) Sputtered lm 850 (20)
Antimony sulfoiodide, SbSI
a
Film, laser ablated on Pt/Si 12.4 (19.2)
PNZT, Nb-doped PZT
b
Film, solgel grown on Pt/Si 1080 (23)
a
Thick lm grown on Pt/Si substrate by PLD [22].
b
Solgel lm on Pt/Si ofPb(Nb
0.02
Zr
0.2
Ti
0.8
)
3
[14].
Source: Ref. 23.
1514 FERROELECTRIC MATERIALS
integrated structures with silicon [26], it is reasonable to
assume that the pyrooptic effect might attain its rightful
importance in the infra-red technology.
2.7. Antiferroelectricity
Once again we have copied this term from the vocabulary
of magnetism. In antiferromagnets electronic spins are in
antiparallel conguration. Similarly, the dipoles are anti-
parallel to each other in an antiferroelectric material. In
these crystals the polarization is distributed throughout
in the bulk of the material in an antiparallel arrangement
such that the net polarization is zero. Figure 8b may be
considered to schematically represent such a system. Note
that this is a highly ordered conguration and must not be
confused with the randomness of the dipoles caused by
lattice vibrations. According to the classic work by Jona
and Shirane [21], an antiferroelectric material is to be de-
ned as an antipolar crystal whose free energy is compa-
rable to that of a polar crystal. In an antipolar crystal the
dipole interactions are such that they cause the antipar-
allel domains to dominate. Obviously these materials do
not exhibit hysteresis loops between the polarization and
the electric eld. But a large anomaly is experimentally
detectable when the structure changes from a completely
unpolarized state to the antiferroelectric state at the tran-
sition point. Antiferroelectricity is a fundamental part of
ferroelectricity and plays an important role in under-
standing the physics behind dielectric materials. Lead zir-
conate (PbZrO
3
) is one of the most famous examples of
antiferroelectricity. It shows a large anomaly in the tem-
perature dependence of the dielectric constant at about
2501C and no dielectric loop at all [21]. This is a very im-
portant constituent of PLZT, PNZT, PZT, and other com-
pounds that have enormous technical importance.
2.8. Multiferroic Magnetoelectrics
No discussion of ferroelectricity can be complete without
considering multiferroic materials. These are the materi-
als in which ferroelectricity and ferromagnetism can co-
exist in the same phase. They are also known by different
terms, such as magnetoelectrics and ferromagnetoelectrics.
Ferromagnetism and ferroelectricity have been exten-
sively researched and used in developing many applica-
tions and technologies. They both remain to a great extent
independent elds of science with little or no overlap.
However, the scientific and technological importance has
long been realized for materials showing both ferromag-
netism and ferroelectricity in the same phase and prefer-
ably above room temperature. Unfortunately such
materials are not found in nature and have until relative-
ly recently frustrated the efforts to produce them with a
high degree of reproducibility in laboratories. This scarci-
ty might be anchored in the fact that the transition metal
d electrons, which are essential for the origin of ferromag-
netism, tend to reduce the lattice distortion that causes
ferroelectricity [27]. This results in a weak coupling be-
tween the spontaneous magnetization and spontaneous
polarization. Because of the advancement in lm technol-
ogy and successful fabrication of integrated structures and
superlattices, the search for new multiferroic materials
has become very active. This covers a broad spectrum of
materials ranging from composites to epitaxial lms. It
has been reported that magnetoelectric effects can be in-
troduced in the thin lm consisting of nanometric level of
CoFe
2
O
4
[cobalt ferrite (CFO)], which is a ferromagnetic
material, and ferroelectric lead titanate (PbTiO
3
) [28].
Bismuth manganate (BiMnO
3
) is another material in
which magnetoelectricity has been predicted [29]. The ex-
perimental search for this effect in BiMnO
3
was presented
at a workshop conducted under the sponsorship of the US
Ofce of Naval Research [30].
Of special importance are the two new discoveries. The
presence of ferroelectricity in magnetic bismuth ferrite
(BiFeO
3
) has been conclusively shown by Palkar et al. [31].
They grew pure phase of bismuth ferrite on platinized sil-
icon substrates by the pulsed-laser ablation method (PLD)
under multiple partial pressures of oxygen. They report
the existence of saturated ferroelectric loop for their sam-
ples, and the typical anomalous behavior of the tempera-
ture dependence of dielectric constant, which is the
signature of a ferroelectric material. This anomaly occurs
at the antiferromagnetic transition temperature (T
N
) of
B3801C. The ferroelectric transition occurs at B8101C.
This is a remarkable result and is certain to contribute to
the eld of multiferroic materials and their applications.
Another equally important discovery has been reported
by Hur et al. [32]. Electric polarization reversal and mem-
ory effects were induced by external magnetic eld rang-
ing from 0 to 2 T (tesla) in terbium manganate (TbMn
2
O
3
)
multiferroic material. This is the rst time anyone has
been able to demonstrate strong coupling between mag-
netic and ferroelectric states. Besides polarization rever-
sal induced by a magnetic eld, the investigators were also
successful in attaining memory effects under the combined
inuence of magnetic and electric elds. Like the previous
work, this outstanding discovery is certain to impact the
science and information technology.
Table 6. Examples of Some Pyrooptic Materials
Materials Formula Temperature Range (1C) Pyrooptic Coefcient y
p
( 10
4
K
1
)
Antimony sulfoiodide SbSI 015 75
Bismuth vanadate BiVO
4
20100 2.8
Molybdenum disulde MoS
2
100 to 18 1.63
Lead titanate PbTiO
3
60 to 40 1.5
Barium titanate BaTiO
3
29120 3.1
Triglycine sulfate TGS 40300 5.0
Source: Ref. 24.
FERROELECTRIC MATERIALS 1515
3. MATERIALS PROCESSING
The ferroelectric materials are processed in three princi-
pal ways: ceramic processing, lm deposition, and single-
crystal growth. Each of these methods has its merits and
demerits. Their selection depends primarily on the factors
such as cost, volume of materials needed, and intended
use. In the next few paragraphs we will briefly survey the
three processing techniques applicable to ferroelectric ma-
terials.
3.1. Ceramic Processing
For large-scale production and applications this is the
most widely preferred processing technique applied to
produce dielectric materials including ferroelectrics. It is
a relatively simple technique, is moderately expensive,
and gives high returns on investment. Additionally, it
lends itself easily to implementation of modications and
parameter adjustments for producing materials with pre-
dictable properties in large volumes. On a smaller scale, it
provides an excellent research tool for designing and dis-
covering new dielectric materials.
This consists of seven major steps as shown in Fig. 14.
The rst and foremost step is to begin with highest possi-
ble purity grades of rawmaterials in proper weight or mole
ratio. Then they should be mixed thoroughly such as by
ballmilling to break down the particle size and obtain a
homogeneous mixture. The subsequent steps of heat treat-
ment and forming desired shapes and sizes followthis step.
Then the most critical step is undertaken. The compaction
takes place by applying uniformly high pressures to the
dies containing the desired shapes and forms. Hydrostatic
pressing at elevated temperatures is the most desirable
approach to obtain a high-density and high-quality mate-
rial. However, if such a system is not available, cold hy-
drostatic pressing, or simply cold or hot isostatic pressing
are available alternatives. Once the pressing is done, the
green ceramic needs to be sintered at high temperatures,
which are mostly 90% of the melting temperature of the
materials, for hours in a specialized atmosphere. Air an-
nealing is usually sufcient. However, depending on the
chemistry of the material, it is necessary to do annealing in
reducing, oxidizing, or neutral atmospheres. The logic be-
hind sintering at high temperatures for extended time pe-
riods is to allow the diffusion process to produce single-
phase, homogenous materials with large grains (crystal-
lites) and with practically no voids. Each heat treatment
aids in attaining this goal. It is important that the steps
shown in boxes (second, third, and fourth seen from the
top) in Fig. 14 should be repeated at least 3 times before
high-pressure operation takes place. This laborious and
time-consuming processing method ensures the quality of
the material to be produced. Once the ceramic samples be-
come available, they are ready for use either for scientific
experiments or for product development.
3.2. Film Deposition
Because of their multifunctional nature, ferroelectrics
lend themselves to a large number of applications and de-
velopment of novel devices for which high-quality epitax-
ial or textured lms are required. They are grown by a
variety of techniques; the most common ones are sputter-
ing, electron-beam evaporation, solgel, metallorganic de-
position (MOD), physical vapor transport (PVT), and
pulsed-laser deposition (PLD). As in the case of ceramic
processing, the choice of the method for lm growth de-
pends on many factors; again, the foremost are the cost
and quality of lms needed. As can be expected, each of
these methods has its merits and limitations. The litera-
ture is full of excellent books and publications dealing
with each of these techniques. Unfortunately, we are not
in a position to cover any of these methods in depth in this
article. However, we are giving here some basic informa-
tion related to the methods mostly used by the ferroelec-
tric community for the benet of the reader.
For large-scale production, sputtering and solgel are
the leading methods used on industrial scale. Having sim-
ilarity with solgel MOD is also widely used but it is not as
effective a method as solgel. The PVT method is used only
when other methods are not applicable such as if the ma-
terial has high vapor pressure. These are mostly iodides,
uorides, suldes, and similar compounds. The PVT meth-
od is capable of producing high-quality textured or even
single crystalline lms on a variety of substrates. Howev-
er, it has its own limitations and is not easily modiable
High - purity - grade raw materials mixed in desired
proportion by weight or mole percent
Ballmilled to achieve thorough mixing
Sintered at high temperatures in appropriate
atmospheres to allow diffusion process to produce
homogenous mixture
Fine - ground powder mixed with organic binders are
packed tightly in stainless - steel dies of appropriate
geometries .
Well - formed green ceramics are released from dies
and then sintered at high temperatures in an
appropriate atmosphere for several hours to obtain
dense and single - phase polycrystalline samples
The sintered ceramic samples can be used as target for
laser ablation or for fundamental studies or device
fabrication
Pressure is applied uniaxially or isostatically and at
room temperature or at higher temperatures depending
on facilities available
Figure 14. Flow diagram for ceramic pressing.
1516 FERROELECTRIC MATERIALS
for large-scale applications. Figure 15 shows a simple ex-
perimental arrangement developed by us [22] for the
growth of antimony sulphoiodide (SbSI) lms for pyroelec-
tric and pyrooptic research [26]. High-quality textured or
even single-crystal lms could be grown of this high vapor
pressure material by the PVT method.
The PLD method is versatile and has been used for the
growth of a large number of oxides including ferroelectrics,
high-temperature superconductors, and colossal magneto-
resistive (CMR) materials. Since its discovery in late 1980s
when it was rst introduced very successfully for
the growth of high-temperature superconductor, especial-
ly 123 YBCO, PLD has become the choice method for re-
search on oxide growth. It is a versatile method and can
be easily adopted to implement necessary changes in
growth parameters. Its greatest strength lies in the fact
that it retains the chemical homogeneity of the lm with
respect to its source (called targets in the language of
PLD) and can produce polycrystalline, textured, or single-
crystalline lms; integrated structures; and superlattices.
It is also used for the growth of complex device congura-
tions. An excellent treatment is given of PLD technology in
Ref. 33.
3.3. Single-Crystal Growth
Single-crystal growth has been the most fascinating and
challenging area of materials processing. Ceramics, lms,
integrated structures, and superlattices can meet many
scientific and technical needs, yet there are still some very
specialized needs for high-quality bulk single crystals.
Substrates for epitaxial growth of lms are exclusively
bulk single crystals. This has been a very well-established
eld for more than a century. Its biggest impact in tech-
nology came with the invention of transistors in the late
1950s when it was realized that device-quality single crys-
tals of silicon were needed for superior performance and
reproducibility of the transistors. The crystal growth
method, which is today widely known as the Czochralski
technique, lived up to this enormous technical challenge.
It is the most widely used method for crystal growth.
Other growth methods are high-temperature solution
growth (HTSG), aqueous solution growth (ASG), Bridg-
man growth, top-seeded solution growth (TSSG), and tem-
perature-gradient transport growth (TGTG), to name only
a few. The literature is full of crystal growth methods ap-
plicable to a variety of materials. Obviously it is not within
the scope of this article to survey the literature for the
crystal growth of even ferroelectric materials. Interested
readers are referred to some excellent work such as given
in Refs. 3436.
Here we plan to limit ourselves to a very introductory
presentation of this subject. The selection of an appropri-
ate growth method depends on all the factors outlined in
the previous two sections. But the most important criteri-
on is to determine whether a material melts congruently
or incongruently. When, for example, the melt and the
solid of a material have the same chemical composition,
then the materials melting point is congruent. Otherwise,
it is incongruent. Very few ferroelectrics melt congruently.
As a result, the famous Czochralski technique cannot be
employed for the growth of these materials. This is unfor-
tunate in the sense that other methods are not as versatile
as the Czochralski technique in producing large single
crystals in a reasonable amount of time. The most suc-
cessful methods for the growth of ferroelectric crystals
have been HTSG, TGTG, TSSG, and ASG. The Czochral-
ski technique has been widely used for the congruently
melting members of the family of barium strontium titan-
ate (BST), strontium barium niobate, and lithium niobate
(LiNbO
3
). All these are highly attractive materials for
electrooptic applications and fabrication of optical wave-
guides. Lithium niobate, an excellent ferroelectric and
piezoelectric material, is also widely used as substrates
for building integrated structures.
The famous pyroelectric material, TGS, is grown by
AGS method. This is an excellent material for IR imaging.
But its hygroscopic nature is a drawback. A large number
of ferroelectric crystals have been grown by the HTSG
method, which is also known as the ux growth method.
Here a suitable solvent is used for dissolving the raw ma-
terials (called charge in the vocabulary of crystal
growth) at high temperatures and then cooled very slow-
ly (usually 121/h) over many weeks. Relatively large (a
few millimeters to a few centimeters) crystals grow that
are harvested by dissolving the frozen ux in a suitable
solvent. Barium titanate (BaTiO3) has been grown in
device quality by the TSSG method.
A variation of this method is TGTG. This is a very in-
novative technique and has been applied for the growth of
some otherwise difcult-to-grow ferroelectrics with a min-
imum of compositional gradient within the bulk of the
sample. Potassium tantalate niobate (KTN), (Fig. 16),
which has excellent electrooptic, ferroelectric, and pyro-
optic properties, has been grown as a large single crystal
by this method [20].
4. APPLICATIONS
In the preceding sections we have established the multi-
functional nature of ferroelectrics. Ferroelectrics, as we
have already seen, are high dielectric constant materials,
which also possess reversible spontaneous polarization
Film
Source
Figure 15. SbSI lms grown by physical vapor transport
method.
FERROELECTRIC MATERIALS 1517
and piezoelectric coupling as well as pyroelectricity
simultaneously. These properties are the driving forces
behind the importance of ferroelectrics in a wide range
of technologies. In fact, they play a very vital role in the
advancement of technologies ranging from classical to ul-
tramodern. Classically, ferroelectrics are ceramics. Owing
to their unique electrical properties they are also impor-
tant members of the family of electroceramics.
It is not within the scope of this article to discuss in
depth even the most important applications based on the
multifunctional nature of ferroelectric materials. Interest-
ed readers will nd the two publications [15,37] useful in
understanding of applications based on ferroelectrics and
associated phenomena and their impact on technology, es-
pecially electronic, electrooptic, communication, and in-
formation technology.
By carefully evaluating the multiple properties and
phenomena associated with these classes of polar materi-
als, we can get an idea of the enormity of its importance to
technology and its potential for the development of novel
devices. We can summarize them as follows:
*
Because ferroelectrics are piezoelectric as well as
pyroelectric, they are also multifunctional.
*
Piezoelectricity can exist alone in some polar materi-
als, or it can be accompanied by pyroelectricity,
or in some other materials all three effectsferro-
electricity, piezoelectricity, and pyroelectricitycan
coexist.
*
All pyroelectrics are piezoelectrics but not necessarily
ferroelectrics as well.
*
The discovery of existence of both ferroelectricity and
ferromagnetism in the same phase (previously re-
ferred to as mutltiferroic or magnetoelectric) opens
avenues for world-class research in materials science,
physics, and microelectronics.
*
Because piezoelectricity is an excellent class of elect-
ro-optic materials, its impact on the optical industry
is conspicuous.
In the beginning of this article we briefly examined the
equilibrium diagram (Fig. 1) between three forces, electric
eld (E), temperature (T), and mechanical stress (T
ij
),
and their respective consequences; namely, displacement
(D), entropy (S), and stress (S
ij
). These six agents couple
together in noncentrosymmetric crystals to produce
the multifunctional phenomenon of ferroelectricity. Fig-
ure 17 presents some of these interactions and their con-
sequences.
We have already seen that the domains in ferroelectrics
even at room temperature can be randomly oriented.
Therefore, the materials need to be poled before they
can be utilized for device fabrication and integration. Pol-
ing is a relatively simple operation. First, metallic elec-
trodes are deposited by thermal evaporation or sputtering
or by other means on the surfaces of the samples so
that an electric eld can be applied to them. Then they
are subjected to a DC electric eld, which is much greater
than the samples coercivity, while slowly raising the
temperature to a point above the Curie point of the ma-
terial, where it is held for some time to ensure that
the entire material is in equilibrium in its paraelectric
state. Subsequently, the temperature is slowly lowered
to room temperature in the presence of the eld. The pro-
cess involves the nucleation of domains as soon as
the temperature is lowered slightly below the Curie point.
Once the domains form they orient themselves in the
Figure 16. Potassium tantalate niobate (KTN) single crystals
[20]; the length scale shown here corresponds to 1cm.
Agents
Electric field Mechanical stress Temperature
Direct
piezoelectric
effect
Piezocaloric
effect
Piezooptic
effect
Pyroelectric
effect
Thermoelectric
effect
Heat capacity
effect
Converse
piezoelectric effect
Electrocaloric
effect
Electrooptic effect
(a)
(b)
(c)
(a)
(b)
(c)
Memory effect (d)
(a)
(b)
(c)
Figure 17. Agents and effects responsible for fer-
roelectric applications [15].
1518 FERROELECTRIC MATERIALS
direction of the eld that is present in the sample. Their
number keeps on increasing, as the sample is nally
reaches room temperature. Now all the domains in the
material are should be fully aligned and will remain so
even after the eld is removed. At this stage, the poling
has been completed and the eld is removed. The sample
will retain its poled status as long as it is not reheated
above the Curie point. In that event, one will have to re-
pole the specimen.
On one end of the application spectrum, ferroelectrics
are the dominant material for design and fabrication of
ceramics. They are produced in large volumes and have
the largest share of capacitor market throughout the
world. Because of their unusually high dielectric constant,
a large quantity of electrical energy can be stored in a
relatively small volume and capacitor size. Therefore, fer-
roelectric-based capacitors can be made in large to minia-
turized shapes and fulll vital functions in electronics and
integrated circuits.
On the other end of the spectrum, ferroelectrics
are conspicuously present in an array of modern applica-
tions covering as varied elds as MEMS technology,
radiofrequency and microwave communication, informa-
tion technology, electrooptics, optical communications,
infrared imaging, uncooled focal plane arrays, and
nonvolatile memory, to name just a few applications and
technologies to which ferroelectric materials have contrib-
uted.
Ferroelectrics dominate the scene of actuators and
transducers because of their piezoelectric properties.
Among all commercially available piezoelectric materials,
Pb zirconate titanate (PZT) plays a very prominent role.
Table 4 lists other piezoelectric materials. PZT is the most
widely used piezoelectric material and has established it-
self as a leading transducer material in microelectrome-
chanical (MEMS) technology.
When it comes to infrared imagining applications, fer-
roelectric materials once again lead the pack of materials
suitable for fabrication of detectors that can operate above
room temperature. These are the so-called uncooled focal
plane arrays. Around 1990 Texas Instruments in Dallas,
Texas championed ferroelectric-based uncooled focal plane
arrays. The original material used was highly dense ce-
ramic of barium strontium titanate (BST) having a tran-
sition temperature of around room temperature. In this
type of device, the unique pyroelectric properties of ferro-
electrics are exploited. In Table 5 we list the leading pyro-
electric materials.
Memory devices based on ferroelectric materials
are nonvolatile, meaning that the data stored are retained
in the memory even when the power fails or is switched
off. This is comparable to the well-established magnetic
memories. Another class of widely used memory is DRAM
(or dynamic random access memory), which consists
of volatile memories. These memories are based on silicon
integrated circuit technology, in which the stored data are
lost once the electrical power of the device is switched off.
In spite of this serious drawback and inconvenience,
DRAMs are widely used because of their high integration
capability. Nonvolatility of a ferroelectric memory is the
result of the stable bipolar states of its hysteresis loop
discussed previously in this article. Constant improve-
ments are being made in the design and integration of
FeRAMs.
Ferroelectrics are also an established group of materi-
als for microwave communications. Barium strontium ti-
tanate (BST) appears to be the leading material for such
applications. The literature is full of reports on how BST
can be used in developing integrated structured devices to
meet various requirements of the microwave communica-
tion. Ferroelectrics are attractive materials for microwave
devices because (1) they have low microwave losses in the
paraelectric phase, (2) dielectric permittivity is DC-eld-
dependent, (3) tunability is high, and (4) extremely high
values of permittivity make it possible to miniaturize the
devices. The ferroelectric integrated structures (with sil-
icon and superconductors) are attractive for many micro-
wave applications:
*
Varactors and varactor-based devices such as mixers
and harmonic generators
*
Tunable resonators, lters, and antennas
*
Miniature tunable delay lines and phase shifters
*
High-density capacitors and small-size low-imped-
ance transmission lines
*
Thin-lm bulk acoustic wave resonators and lters
*
Surface acoustic wave (SAW devices) lters, convolve-
rs, converters, and similar devices
*
MEMS switches for actuation and transduction.
A good survey of applications of ferroelectric materials in
microwave communications is covered by Patel et al. [38].
Interested readers will benet greatly from consulting
this resource. Integrated structured devices using ferro-
electric materials as one of the components are becoming
increasingly important for development of new microwave
[39] and other applications [4046].
The following books and other publications are recom-
mended for serious readers who would like to gain an in-
sight in the eld of ferroelectricity and related
phenomena.
BIBLIOGRAPHY
1. G. Heckmann, Ergebnisse der exacten Naturwissenschaften
4:140 (1925).
2. J. F. Nye, Physical Properties of Crystals, Oxford Science
Publications, 1986.
3. M. E. Lines and A. M. Glass, Principles and Applications
of Ferroelectrics and Related Material, Clarendon Press,
1977.
4. E. C. Subbarao, Ferroelectrics 5:267 (1973).
5. R. P. Viswanath and A. T. Seshadri, Solid State Commun.
92:831 (1993).
6. R. K. Pandey, P. Padmini, and P. Kale, unpublished results,
Laboratory for Electronic Materials and Device Technology
(www.emdtech.eng.ua.edu), Univ. Alabama, Tuscaloosa, AL.
7. J. Valasek, Phys. Rev. 17:475 (1921).
8. Draft 16 of a Working Document for a Proposed Standard
to be entitled IEEE Standard Definitions of Terms Associated
with Ferroelectric and Related Materials; see IEEE
FERROELECTRIC MATERIALS 1519
Trans. Ultrason., Ferroelectrics Freq. Control 50:1613
(2003).
9. L. Solymar and D. Walsh, Electrical Properties of Materials,
6th ed., Oxford Science Publications, 1999.
10. P. B. Jamieson, S. C. Abrahams, and J. L. Bernstein, J. Chem.
Phys. 48:5048 (1968).
11. C. B. Sawyer and C. H. Tower, Phys. Rev. 35(3):269 (1930).
12. J. K. Sinha, J. Sci. Instrum. 42:696 (1965).
13. E. Fatuzo and W. J. Merz, Ferroelectricity, North-Holland,
Amsterdam, 1966.
14. H. Han, X. Song, J. Zhong, P. Padmini, S. Kotru, and R. K.
Pandey, Appl. Phys. Lett. (Nov. 2004).
15. K. C. Kao, Dielectric Phenomena in Solids, Elsevier Academic
Press, 2004.
16. G. T. A. Kovacs, Micromachined Transducers Sourcebook,
McGraw-Hill, 1988.
17. C. Nistrorica, J. Zhang, P. Padmini, S. Kotru, and R. K. Pan-
dey, Integr. Ferroelectrics 62 (2004).
18. C. Nistorica, Integrated PNZT Film Structures for MEMS
Gyroscope, dissertation, Univ. Alabama, Tuscaloosa, AL,
2003.
19. S. Kotru, W. Liu, and R. K. Pandey, Proc. 12th IEEE
Int. Symp. Applications of Ferroelectrics (ISAF 2000),
2001.
20. K. W. Goeking, R. K. Pandey, P. J. Squattrito, A. Cleareld,
and H. R. Beratan, Ferroelectrics 92:89 (1989).
21. F. Jona and G. Shirane, Ferroelectric Crystals, Dover Publi-
cations, 1993 (originally published in 1962).
22. S. Kotru and R. K. Pandey, unpublished result, Laboratory for
Electronic Materials and Device Technology (www.emd-
tech.eng.ua.edu), Univ. Alabama, Tuscaloosa, AL.
23. A. J. Moulson and J. M. Herbert, Electroceramics, 2nd ed.,
Wiley, 2003.
24. J.-F. Li, D. Viehland, A. S. Bhalla, and L. E. Cross, J. Appl.
Phys. 71(5):2106 (1992).
25. L. E. Cross, A. S. Bhalla, F. Ainger, and D. Demjanovic,
Pyro-Optic Detector and Imager, US Patent 4,994,672
(1991).
26. R. K. Pandey, S. Kotru, X. Song and D. Donnelly, Bulletin of
the American Physical Society, paper P. 21.6, presented at
March 2003 Meeting of American Physical Society, Montreal,
Canada, 2004.
27. N. Spaldin, Why Are So Few Magnetic Ferroelectrics? gradu-
ate seminar abstract, Univ. California, Santa Barbara,
2003.
28. I. Takeushi, K.-S. Chang, M. Murakami, M. Aronova,
C. I. Lin, and J. Hattrick-Simpers, Bulletin of the American
Physical Society, paper P. 21.3, presented at March 2003
Meeting of American Physical Society, Montreal, Canada,
2004.
29. N. A. Hill and K. M. Rabe, Phys. Rev. B 59:8759 (1999).
30. D. Schlom, V. Gopalan, X. Pan, A. K. Cheetham, and R.
Ramesh, ONR Workshop on Frontiers of Epitaxial Engineer-
ing, Moab, UT, 2004.
31. V. R. Palkar and R. Pinto, PRAMANA-J. Phys. (Indian Acad-
emy of Sciences) 58(5):1003 (2002).
32. H. Hur, S. Park, P. A. Sharma, J. S. Ahn, S. Guha, and S.-W.
Cheng, Nature 2572 (2004).
33. D. B. Chrisey and G. K. Hubler, Pulsed Laser Deposition of
Thin Films, Wiley, 1994.
34. D. Elwell and H. J. Scheel, Crystal Growth from High-
Temperature Solutions, Academic Press, 1975.
35. J. J. Gilman, ed., The Art and Science of Growing Crystals,
Wiley, 1963.
36. B. R. Pamplin, ed., Crystal Growth, 2nd ed., Pergamon Press,
1980.
37. K. Uchino, Ferroelectric Devices, Marcel Dekker, 2000.
38. D. M. Patel, J. M. Pond, and J. B. I. Rao, Microwave ferro-
electric devices, in J. G. Webster, ed., Wiley Encyclopedia of
Electrical and Electronics Engineering, Vol. 13, Wiley,
New York, 1999, p. 109.
39. S. Hontsu, H. Nishikawa, H. Nakai, J. Ishii, M. Nakamori,
A. Fujimaki, Y. Noguchi, H. Tabata, and T. Kawai, Supercon-
duct. Sci. Technol. 12:836 (1999).
40. Y. Watanabe, Appl. Phys. Lett. 66:1770 (1995).
41. S. Mathews, R. Ramesh, T. Venkatesan, and J. Benedetto,
Science 276:238 (1997).
42. W. Wu, K. H. Wong, C. L. Mak, C. L. Choy, and Y. H. Zhang,
J. Appl. Phys. 88:2068 (2000).
43. A. M. Grishin, S. I. Khartsev, and P. Johnsson, Appl. Phys.
Lett. 74:1015 (1999).
44. T. Wu, S. B. Ogale, J. E. Garrison, B. Nagaraj, A. Biswas,
Z. Chen, R. L. Greene, R. Ramesh, T. Venkatesan, and A. J.
Millis, Phys. Rev. Lett. 86:5998 (2001).
45. S. R. Surthi, S. Kotru, and R. K. Pandey, Integr. Ferroelectrics
48:263 (2002).
46. S. Surthi, Integration of Colossal Magnetoresistive Materials
with Ferroelectrics, dissertation, Univ. Alabama, Tuscaloosa,
AL, 2001.
FURTHER READING
F. Jona and G. Shirane, Ferroelectric Crystals, Dover Publications,
1993 (a classic piece of literature on ferroelectricity).
J. F. Nye, Physical Properties of Crystals, Oxford, Univ. Press,
1985 (a fundamental work essential for all students of ceram-
ics and dielectrics).
J. C. Burfoot and G. W. Taylor, Polar Dielectrics and Their Appli-
cations, Univ. California Press, 1979 (a good introductory book
on polar materials).
M. E. Lines and A. M. Glass, Principles and Applications of Fer-
roelectrics and Related Materials, Clarendon Press, 1977 (this
book deals comprehensively with physics of ferroelectricity and
its applications; intended for advanced study and research in
the eld).
K. C. Kao, Dielectric Phenomena in Solids, Elsevier, 2004 (an ex-
cellent treatment of dielectric materials, including ferroelec-
trics and associated phenomena; a must-read book for students
at the senior undergraduate and graduate levels).
J. Moulson and J. M. Herbert, Electroceramics, 2nd ed., Wiley,
2003 (offers an in-depth treatment of electroceramics).
K. Uchino, Ferroelectric Devices, Marcel Dekker, 2000 (an intro-
ductory book of its kind on device aspects of ferroelectricity and
related materials).
IEEE Standard Definitions of Terms Associated with Ferroelectric
and Related Materials, Draft 16, reprinted in IEEE Trans.
Ultrason. Ferroelectrics Freq. Control 50:1613 (2003)
(this standard should be consulted for basic definitions
and terms in the eld of ferroelectricity and related
materials; it is an essential work to consult for ferroelectric
research).
1520 FERROELECTRIC MATERIALS
Next Page

You might also like