Electromagnetic modeling of anisotropic

ferrites—Application to microstrip Y-junction

circulator design
Cite as: J. Appl. Phys. 123, 234503 (2018); https://doi.org/10.1063/1.5026482
Submitted: 20 February 2018 . Accepted: 29 May 2018 . Published Online: 20 June 2018

V. V K Thalakkatukalathil, A. Chevalier, V. Laur, G. Verissimo, P. Queffelec, L. Qassym, and R. Lebourgeois

ARTICLES YOU MAY BE INTERESTED IN

Bandwidth broadening for stripline circulator

Review of Scientific Instruments 88, 024706 (2017); https://doi.org/10.1063/1.4975097

Modeling non-saturated ferrite-based devices: Application to twin toroid ferrite phase shifters
Journal of Applied Physics 120, 073902 (2016); https://doi.org/10.1063/1.4960640

Permeability spectra of yttrium iron garnet and its granular composite materials under dc
magnetic field
Journal of Applied Physics 110, 053909 (2011); https://doi.org/10.1063/1.3626057

J. Appl. Phys. 123, 234503 (2018); https://doi.org/10.1063/1.5026482 123, 234503

© 2018 Author(s).
 JOURNAL OF APPLIED PHYSICS 123, 234503 (2018)

Electromagnetic modeling of anisotropic ferrites—Application to microstrip

Y-junction circulator design
V. V K Thalakkatukalathil,1 A. Chevalier,1 V. Laur,1 G. Verissimo,1 P. Queffelec,1
L. Qassym,2 and R. Lebourgeois2
1
Lab-STICC, Universit
e de Brest, CNRS, Universit
e Bretagne Loire, 29238 Brest Cedex 3, France
2
Thales Research & Technology France, 1 av. Augustin Fresnel, 91767 Palaiseau Cedex, France
(Received 20 February 2018; accepted 29 May 2018; published online 20 June 2018)
This paper presents a theoretical tool, which combines a generalized permeability tensor model, a
homemade 3D magneto-static solver, and a commercial electromagnetic simulation software
TM
Ansys HFSS for accurate modeling of ferrite-based devices regardless of their state of magnetiza-
tion. A magneto-static analysis is carried out to find the internal biasing fields in the ferrite sample.
Permeability tensor components are computed using a generalized permeability tensor model. Real
and imaginary parts of permeability tensor components are then integrated into the commercial 3D
electromagnetic simulation software HFSS retaining the spatial variations of internal biasing fields
in the ferrite sample. Frequency domain simulations are done using HFSS. This theoretical model-
ing approach is validated by comparing the theoretical simulation results with experimental ones in
the case of a coaxial line loaded by a magnetized ferrite. Proposed approach is then applied to the
design of a microstrip Y-junction circulator. Finally, a demonstration prototype is manufactured in
the Low Temperature Co-fired Ceramics technology. Published by AIP Publishing.
https://doi.org/10.1063/1.5026482

I. INTRODUCTION commercial simulation software is able to describe the

dynamic EM behavior of ferrite-based devices accurately at
Ferrite materials are widely used in microwave electron-
all magnetization states. Designers often tend to use a “cut
ics for many telecommunication applications, especially in
and try” approach to optimize the performance of ferrite
circulators/isolators because of their high electrical resisti-
devices, which is expensive in terms of time and money.
vity and high spontaneous magnetization. The working of
To meet the requirements for the rapid growth of micro-
ferrite circulators/isolators is based on the nonreciprocal
wave communication systems, the design of circulators has
behavior of electromagnetic (EM) wave propagation in mag-
to be based on new materials (low sintering temperature fer-
netized ferrites. Circulators are generally used for full duplex
rites, hexaferrites),4 new technologies8 (Low Temperature
communications, whereas isolators are used to protect the
Co-fired Ceramics (LTCC), 3D printing, etc.), and the use of
transmission components from parasitic radiations or impe-
predictive simulation tools. In this paper, we present a theo-
dance mismatch. Precise control over the performance of
retical modeling tool, which combines magneto-static analy-
these devices requires prior knowledge of the dynamic
sis, a general model for permeability, and a commercial EM
behavior of ferrite materials, namely, tensor permeability
simulation software for accurately describing the dynamic
ðl
^ Þ, each component of which has a double dependence
behavior of ferrite-based devices in all magnetization states.
with respect to the wave frequency and the static magnetic
This theoretical approach is validated by modeling and real-
field ðHdc Þ.
izing an X-band microstrip Y-junction circulator.
Conventional Y-junction circulators require strong DC
magnetic fields to bias the ferrite integrated in the device.1
II. DESCRIPTION OF THE DEVELOPED MODELING
These DC biasing fields are provided by permanent magnets
TOOL
which increases the size and cost of these devices. For minia-
turization of circulators, designers are encouraged to use pla- Due to its complex nature (nonreciprocal behavior, aniso-
nar technologies (e.g., microstrip circulator) or pre-oriented tropic properties of magnetic materials, heterogeneous config-
materials to avoid the use of permanent magnets.2–5 One of uration, etc.), the design of ferrite devices meeting the
the characteristics of the magnetic materials is the depen- miniaturization conditions with good EM performance
dence of their permeability to the shape of the sample. For requires better understanding of the physical processes
non-ellipsoidal shaped ferrite sample, which is the case most involved. A more realistic description of EM behavior of fer-
commonly encountered in ferrite devices, the internal DC rite devices requires accurate computation of permeability ten-
biasing fields are not uniform. sor components.
This DC field non-homogeneity within the ferrite mate- Commercial simulation softwares use different theoreti-
rial leads to the reduction of transmission band in Y-junction cal models to describe the permeability tensor components.
circulators.6 The effect of DC field non-uniformity on the Ansys HFSSTM uses Polder model9 to describe the perme-
performances of Y-junction circulator has been the subject ability tensor components of the anisotropic ferrites and
of previous studies in our laboratory.7 None of the existing always uses saturation magnetization (4pMs ) to describe the

0021-8979/2018/123(23)/234503/11/$30.00 123, 234503-1 Published by AIP Publishing.

234503-2 V K Thalakkatukalathil et al. J. Appl. Phys. 123, 234503 (2018)

magnetization inside the sample. This model is based on the demagnetization coefficient in the direction of the dynamic
small signal approximation of the Landau-Lifshitz equa- magnetization. This grain is surrounded by an effective
tions,10 and it is valid only in the case of a completely satu- medium characterized with a mean RF magnetization h~ mi
rated ferrimagnetic medium where all the magnetic moments  
are aligned in applied DC field direction. In addition, these ~1 þ m
m ~2 M
h~g ¼ ng  h~
mi : (2)
simulation tools consider that the internal DC magnetic field 2 Ms
(biasing field) is uniform, which is not always true in
practice. To solve the system of equations [Eq. (1)], it is neces-
In most common microwave applications, unsaturated sary to compute static quantities, i.e., the equilibrium direc-
regions exist in ferrimagnetic materials even when strong tion of the magnetization Ms and the local static magnetic
DC fields are applied.11 This is due to the inhomogeneous field in each area of grain of the ferrite sample.
nature of the internal magnetic fields within the non- For a domain “i,” once all the static parameters are
ellipsoidal samples. The intensity of the internal magnetic determined, solution to the system of coupled equations
gives the relation m ~ where vi is the local susceptibil-
~i ¼ vi h,
field can be too low in some regions of the material and
totally saturated medium assumed by the Polder model is not l ) of
ity tensor. In order to calculate the permeability tensor (^
valid in those cases. This limitation makes the design of a bulk ferrite sample, a statistical summation of the local
devices laborious when the ferrite is neither in the saturated dynamic magnetizations is performed. This summation takes
nor in the demagnetized state. into account the possible directions of the vector m ~i
There are several other theoretical models available in the described by the angles ð#; uÞ with respect to the easy axis
literature to describe the complex permeability tensor compo- and the possible grain and domain shapes. Empirical distri-
nents12–15 of anisotropic ferrites. However, these models bution functions are used to represent the possible domain
remain limited to certain states of magnetization, due to the and grain shapes (demagnetization coefficient).
simplified assumptions on which they are based upon. Gelin From a self-consistent theoretical approach, GPT model
and Queffelec have developed a general model to compute the provides the permeability tensor, whatever the frequency
permeability tensor components at all states of magnetization: range and the state of magnetization are, from the demagne-
generalized permeability tensor (GPT) model.16,17 This model tized state to saturation. This model takes into account the
predictively describes the dynamics of the permeability tensor demagnetizing field effects, and the dynamic interactions
components of polycrystalline ferrites using a more realistic between domains and grains, and ensures causality of the
approach than previously proposed models. tensor components. For more detailed description of the
GPT model is based on the classical Landau-Lifshitz- GPT model, interested readers may refer to Gelin and
Gilbert (LLG) equations and takes into account the inherent Queffelec articles.16,17
physical phenomena in polycrystalline ferrites such as the In this study, the GPT model is chosen to describe the
Polder-Smit effect,18 the hysteresis phenomenon, and statis- permeability tensor components of the ferrite material. The
tical distribution on the shape of grains and domains. To take input parameters of the GPT model are saturation magnetiza-
into account the dynamic interactions between two adjacent tion (4pMs ), anisotropic field (Ha ), demagnetization coeffi-
domains, 1 and 2, we propose to solve a set of two coupled cient (Nz ) of the sample, damping factor (a), and the applied
equations resolved from LLG equations. Under the action of DC magnetic field (Hdc ). With these input parameters, GPT
RF microwave field h, ~ the evolution of magnetic moments model can simultaneously determine the frequency depen-
considering the inter-domain coupling is described by fol- dent, complex components of the permeability tensor (l, j,
lowing set of coupled equations: and lz ) at any state of magnetization.
In order to describe the dynamic behavior of ferrites
~1 ðtÞ   ~ more accurately, we decided to associate a magneto-static
@M ~1  H
¼ cM ~1 þ h~  h~d  h~g þ a M~ 1  @ M 1 ðt Þ ; analysis with the GPT model.19 A magneto-static analysis is
@t Ms @t
carried out to find the internal biasing fields and resulting
(1a)
magnetization in the ferrite sample. Permeability tensor
~2 ðtÞ   ~
~ 2  @ M 2 ðt Þ ;
components are computed with the GPT model considering
@M ~2  H
¼ cM ~2 þ h~  h~d  h~g þ a M
@t Ms @t the spatial variation of internal DC magnetic fields. These
(1b) complex permeability tensor components are then integrated
with the commercial simulation software Ansys HFSSTM.
where M1 and M2 are the magnetization vectors of the adja- Finally, scattering parameters (S-parameters) will be calcu-
cent domains with internal field H1 and H2 , respectively. lated using the frequency domain solver in HFSS.
The field h~d ¼ nd ð~ m1  m~2 Þ corresponds to the A theoretical modeling tool combining a magneto-static
dynamic demagnetizing field bounded to the shape of the solver,19 a general permeability tensor model (GPT),16 and a
domain and nd is the demagnetization coefficient in the direc- commercial simulation software - Ansys HFSS would enable
tion of the dynamic magnetization. This term allows us to take us to understand the dynamic response of the ferrite-based
into account Polder-Smit effect, that is to say the dynamic cou- devices more accurately at all magnetization states. In Secs.
pling between magnetic moments of the adjacent domains. II A–II B, we will discuss in detail about the magneto-static
Similarly, the field h~g represents the dynamic demagnet- solver and the integration of the permeability models with
izing field linked to the shape of the grain and ng is the HFSS.
234503-3 V K Thalakkatukalathil et al. J. Appl. Phys. 123, 234503 (2018)

A. Magneto-static analysis on the Stoner-Wohlfarth energy model.20 This approach ena-

bles us to model the magnetization states of polycrystalline
Several commercial software solutions are available to
ferrites containing domains by considering the magnetic
compute the internal magnetization as a function of the
domains as uniaxial anisotropic particles.
applied DC magnetic field. However, a valid range of appli-
An iterative Broyden algorithm21 (Quasi-Newton
cations of these software products are limited due to the sim-
method) is used to solve a system of equations, which com-
plified assumptions on which their theoretical models are  approximation. When
prises the use of a Jacobian matrix (J)
based upon. A magneto-static study with the Ansys Maxwell
TM the algorithm is converged, in each mesh cell we obtain the
3D can be used to find the non-uniform internal fields of
microscopic quantities such as local domain magnetization
the magnetized ferrite material. In Maxwell 3D magneto-
(computed using the Stoner and Wohlfarth model20) along
static simulations, the ferrite sample is defined as a nonlinear ~ Each mesh cell can
~m , M).
with macroscopic quantities (H
(from a static point of view) anisotropic material.
follow a different hysteresis cycle (major or minor) and this
In order to describe the evolution of ferrite magnetiza-
analysis is applicable for all magnetization states. This
tion with magnitude and direction of the applied DC mag-
magneto-static analysis presents a more realistic description
netic field, created by a permanent magnet or a current,
of internal DC fields especially in the partially magnetized
Maxwell 3D uses a single B-H curve (major hysteresis curve,
state.
where B is the magnetic flux density and H the magnetic
In order to validate the results obtained with our
field). Most of the time this B-H curve is extracted from the
magneto-static solver, we consider the case of a 3 mm thick
measured major hysteresis cycle and it is applied to the
hollow cylindrical shaped Yttrium Iron Garnet (YIG) ferrite
whole volume of the sample under study. Finally, the perme-
material uniformly magnetized along the longitudinal direc-
ability in each mesh cell defined in HFSS is calculated from
tion (z-axis, see Fig. 1). Ferrite sample is of APC7 standard
the Polder formulations.
(outer diameter 7 mm and inner diameter 3.04 mm) and has a
However, for an applied DC magnetic field, there can be
saturation magnetization (4pMs ) 810 G, an anisotropic field
several states of magnetization in different regions of the
(Ha ) 54 Oe, and a damping factor (a) 0.0068.
magnetized ferrite depending on the magnetization history
Internal DC biasing fields are calculated using the
(variation of the strength of the applied DC biasing field). In
magneto-static analysis and the results are compared with
certain regions of the ferrite, a minor hysteresis curve should
a commercial software solution-Ansys Maxwell 3D.
be taken into account due to the weak value of the local DC
Calculated internal DC biasing fields of a 3 mm thick, hol-
biasing field. But Maxwell 3D cannot take this phenomenon
low cylindrical shaped, YIG ferrite sample when the mate-
into account, since a single B-H curve (major hysteresis
rial is magnetized along the z axis with an applied DC
curve in general) is used to represent the magneto-static
magnetic field 1600 Oe are shown in Fig. 1. The field plot
behavior of whole sample.
calculated using our magneto-static solver shows very
good agreement with the results obtained with Maxwell
1. Magneto-static solver developed in our laboratory
3D simulations. The fact that the results obtained from the
In order to study the inhomogeneity of the internal DC two theoretical approaches are in good agreement is not a
magnetic fields in ferrite materials more accurately, a surprise since the ferrite sample is set in a saturated state
magneto-static solver was developed in our laboratory.19 due to the strength of the DC field. For this particular case
This magneto-static solver is based on the discretization of (saturated ferrites), magneto-static properties of most of
space and resolution of Poisson equation in each mesh cell the regions in the ferrite sample can be described by a
using the finite difference method. major hysteresis curve. However, for a general description
The algorithm used is a differential method, so an airbox of magnetization states of ferrites (major or minor rema-
is defined around the ferrite sample. The computational nence, partially magnetized states, etc.), the magneto-static
nodes are defined on the edges of the mesh. For each mesh analysis of the material from a single B-H hysteresis curve
cell, this solver takes into account a hysteresis cycle based remains insufficient.

FIG. 1. Magnitude of the internal DC

magnetic fields calculated inside a
3 mm thick, hollow cylindrical shaped
YIG ferrite sample using a magneto-
static analysis, for an external DC bias-
ing field Hdc ¼ 1600 Oe directed along
the z-axis. (a) Magneto-static solver
developed in our laboratory; (b)
Maxwell 3D.
234503-4 V K Thalakkatukalathil et al. J. Appl. Phys. 123, 234503 (2018)

FIG. 2. Magnetization of the ferrite

slab at remanent state (Maxwell-3D
simulation). (a) Intensity variation. (b)
Variation in direction.

In the partially magnetized state, we consider the case of not uniform in direction. These differences between the
a ferrite slab of dimensions 40  40  60 mm3 magnetized results of the two approaches are explained by the fact that,
in the direction Oz. Ferrite has a saturation magnetization in our case, the magnetization process is not described only
Ms ¼ 12.25 kG, and anisotropy field Hk ¼ 1250 Oe. The mag- by a major hysteresis loop. Minor cycles can be taken into
netization of the ferrite slab at the remanence calculated account according to the intensity of the DC biasing field
using Maxwell 3D is shown in Fig. 2, where we find that it is that prevails locally. Thus, the evolution of magnetization,
relatively homogeneous in modulus and direction. This is described by the model of Stoner and Wohlfarth in each
obviously related to the fact that ferrite was considered mesh cell, can be different from one cell to another. This is
completely saturated before it returned to the state of not the case with Maxwell 3D which imposes an identical
remanence. relation B (H) for all the mesh cells, and only the value of
We used our magneto-static solver to simulate the same the field can vary from one cell to another. The magnetiza-
magnetic structure as the one previously studied via tion state strongly influences its dynamic EM response of
Maxwell-3D software. Results from our approach differ sig- ferrites, so the precision of the field calculation and magneti-
nificantly from those given by Maxwell 3D. Indeed, in Fig. zation is essential to predict the permeability tensor
3, we find that the magnetization is relatively uniform in accurately.
modulus and direction in the center of the ferrite and on its For the next step, we compare the permeability spectra
transverse sides. However, in the upper and lower regions, obtained using the GPT model and those given by the Polder
the magnetization module is much weaker in magnitude and analytical formulations for the ferrite previously studied in

FIG. 3. Magnetization of the ferrite

slab at remanent state (homemade
magneto-static solver). (a) Intensity
variation. (b) Variation in direction.
234503-5 V K Thalakkatukalathil et al. J. Appl. Phys. 123, 234503 (2018)

terms of its magneto-static properties (ferrite set in a rema-

nent state). In the center of the ferrite, we find an internal DC
biasing field of 425 Oe, which corresponds to a magnetiza-
tion of 1860 G (calculated from Stoner and Wohlfarth energy
model) oriented along the z axis. Using these magneto-static
quantities and a damping factor of 0.1 as the dynamic param-
eter representing the magnetic losses, the diagonal compo-
nent of the permeability tensor (l11 ) spectra computed using
Polder formulations is shown in Fig. 4. In Polder formula-
tions, saturation magnetization (Ms) is replaced by remanent
magnetization (Mr). Polder model predicts a gyromagnetic
resonance around 1.19 GHz and narrow absorption peak, the
characteristics of a saturated ferrite which is not the case in
point.
Permeability spectra computed using GPT model for the FIG. 5. Spectra of the real and imaginary parts of the diagonal term of the
same input parameters are shown in Fig. 5. In this case, the permeability tensor computed using the GPT model.
gyromagnetic resonance frequency is found to be around
4.77 GHz, and importantly, spectra of the real and imaginary definition of all the nine elements of permeability and/or per-
parts of permeability differ completely from those computed mittivity tensor. This opens up an opportunity to integrate
using Polder model. GPT model predicts a widening of the theoretical models other than Polder model with HFSS.
absorption region according to the frequency as qualitatively In our approach, permeability tensor components are
predicted by the Polder-Smit effect. This is realized by tak- calculated using a generalized permeability tensor (GPT)
ing into account a statistical distribution on the shape of the model. Frequency domain solver in HFSS needs to calculate
grains and domains in the GPT model. the value of the permeability tensor components at any fre-
For a simple case scenario (properties, ferrite geometry), quency value. Since GPT model does not give analytical for-
this comparative study demonstrates the limitation of the mulations, each component of the permeability tensor is
commercial EM simulation software which uses Polder computed using GPT model and then interpolated into fre-
model, whether the non-uniformity of the internal DC bias- quency dependent polynomial functions.
ing fields is taken into account or not. Permeability of the material is defined by the tensor
function, and each component of the permeability tensor is
B. Integration of permeability tensor models defined by frequency dependent functions. Real and imagi-
with HFSS nary part of the permeability tensor components are imported
In HFSS, anisotropic nature of ferrite material can be as separate data sets in HFSS. Each data set is then interpo-
modeled by defining permeability tensor components. lated into frequency dependent functions using the inbuilt
Permeability tensor components can be defined either by piecewise linear interpolation function of HFSS software.
measured data or by entering frequency dependent polyno- These functions are then used to define the complex compo-
mial functions. HFSS commonly uses the Polder model to nents of the permeability tensor in HFSS.
compute the permeability tensor components by taking mea- A set of MATLAB functions are created for the defini-
tion of new materials in HFSS software. Using these func-
sured material properties (saturation magnetization, anisot-
tions, we can create or edit materials in HFSS, and it is
ropy field, resonance linewidths) as input parameters. Ansys
TM possible to define the complex elements of the permeability
Electronic Desktop (version 2016.2 and above) allows the
or permittivity tensors from the MATLAB environment.
Scripting with MATLAB allows us the easy control over the
material definition in HFSS simulations. Precisely we can
thus define the real and imaginary part of permeability tensor
components using theoretical models more generalized than
the Polder model (only valid for saturated media), general-
ized permeability (GPT) model in HFSS.

III. VALIDATION OF THE THEORETICAL APPROACH

Consider the case of a 7 mm APC7 standard (outer
diameter 7 mm and inner diameter 3.04 mm) coaxial line
with a toroidal shaped ferrite sample (square cross section)
inserted in between the inner and outer conductors. The sam-
ples used are the bulk ferrites (Yttrium Iron Garnets, YIG-
39) from Exxelia TEMEX with following properties: relative
FIG. 4. Spectra of the real and imaginary parts of the diagonal term of the permittivity 14.3, saturation magnetization 4pMs ¼ 810 G,
permeability tensor computed using the Polder model. anisotropic field Ha ¼ 54 Oe, and damping factor a ¼ 0.0068.
234503-6 V K Thalakkatukalathil et al. J. Appl. Phys. 123, 234503 (2018)

Under the action of DC magnetic field, the material is mag-

netized and becomes anisotropic in nature.
Scattering parameters (S-parameters: reflection S11, and
transmission S21 coefficients) are measured using a vector
network analyzer in the frequency band 130 MHz–20 GHz,
for DC magnetic field strengths ranging from 0 to 2200 Oe.
Power absorption is calculated from the measured S-
parameters (1  jS11 j2  jS21 j2 ). Resonant frequency has a
linear relationship with the internal field present in the sam-
ple (Fr / Hin ). The internal static field seen by the magnetic FIG. 7. Schematic representation of different computational regions exhibit-
moments is decreased by the demagnetizing fields present in ing uniform internal field.
the sample. These demagnetizing fields strongly depend on
the shape of the sample. Moreover, for the case of finite non- formulations for the computation of the permeability tensor)
ellipsoidal samples, internal DC magnetic fields are not uni- with Maxwell-3D software (major hysteresis cycle assump-
form. This internal field non-homogeneity causes secondary tion for the determination of the local magnetization know-
resonance peaks in the power absorption spectrum. Presence ing the DC biasing field). In order to consider the variation
of these secondary resonance peaks results in the broadening of the demagnetizing fields, ferrite sample is divided into dif-
of the power absorption curve near the resonance. Figure 6 ferent concentric annulus regions (Fig. 7) in which the inter-
presents the power absorption spectra of YIG39 samples nal DC field can be considered uniform. Permeability tensor
with different thicknesses leading to different macroscopic is calculated for each computational region using GPT model
demagnetization factors. In each case, secondary resonances taking into account the internal DC biasing field (strength
are present and this phenomenon results in the broadening of and direction).
power absorption spectrum. In HFSS design, these zones are then defined as different
Classical frequency domain simulation using uniform materials or simply as computational regions in which the
biasing in HFSS does not predict the secondary resonances material is the same but with different internal DC biasing
and fails to account for the broadness of the power absorption fields and thus different permeability tensors. For each
near the resonant frequency. Indeed, these simulations con- region, real and imaginary parts of the permeability tensor
sider that the internal fields are constant whereas in reality, components are defined as frequency dependent functions by
the internal DC fields are inhomogeneous in nature. using the inbuilt piecewise linear interpolation function of
A magneto-static analysis is carried out using the home- HFSS. S-parameters are then calculated using the frequency
domain solver in HFSS.
made magneto-static solver to find out the internal DC bias-
Measured and simulated (non-uniform bias) power
ing fields in the ferrite sample. The magneto-static study of
absorption spectra of coaxial line loaded with a 3 mm-thick
sample shows that demagnetizing fields are space dependent
YIG sample for different applied DC magnetic field values
and internal DC fields are inhomogeneous in nature. The
are shown in Fig. 8. The power absorption results are also
internal fields do not show any variation along the “h”
compared with results obtained using the Maxwell 3D-HFSS
direction.
simulations.
HFSS does not allow users to accurately define material
A very good agreement is found between the measured
properties according to each mesh cell, with the exception of
results and those simulated from the theoretical tools we
completely saturated media when HFSS is coupled (Polder have developed. From Fig. 8, it is clear that results obtained

FIG. 6. Power absorption spectra of YIG 39 samples (Exxelia TEMEX) of FIG. 8. Measured and simulated power absorption spectra of 3 mm thick
different thicknesses, 4pMs ¼ 810 G, e ¼ 14.2, a ¼ 0.0068, Hdc ¼ 2000 Oe. hollow cylindrical YIG ferrite sample.
234503-7 V K Thalakkatukalathil et al. J. Appl. Phys. 123, 234503 (2018)

from our theoretical tool are closer to the measurement

results than those predicted by Maxwell 3D-HFSS simula-
tions. Our theoretical tool efficiently predicts the secondary
resonance peaks and the broadening of the power absorption
spectrum.
Increasing the number of computational regions (finer
mesh) typically results in a more accurate solution.
However, finer mesh means more computational nodes and
significant increase in the computational time. In some cases,
computational time can be doubled for an improvement of
1% in accuracy which is not reasonable. Definition of com-
putational regions is therefore done based on a criterion,
which balances accuracy and computational resources. As
long as the variation of magnetization in a region is under
this limit, those regions are assumed to be biased with uni-
form magnetic field. This criterion is taken as the variation
of magnetization in the material and fixed as a compromise
between accuracy and computational time.
Keeping in mind the fact that GPT model is valid at all
magnetization states unlike Polder model which is only valid
in the case of saturated ferrite medium, proposed theoretical
approach shows good advantage over conventional EM sim-
FIG. 9. Y-junction microstrip circulator—Design parameters: Quarter wave-
ulation methods. Integration of magneto-static analysis
length line: l1 ¼ 1.8 mm, w1 ¼ 0.93 mm, 50 X Access line: lin ¼ 2 mm, w2
enhances the predictive nature of GPT model. By taking into ¼ 0.4 mm. Ferrite sample: Rf ¼ 1.82 mm, thickness ¼ 1 mm. Thickness of
account the inhomogeneity of the internal DC biasing fields, substrate, hf ¼ 1 mm. Metallization: Silver. (a) Top view and (b) side view.
proposed theoretical tool can predict the dynamic behavior
of anisotropic ferrites more accurately, at all magnetization These material properties (see Table I) are used to
states. This association of theoretical tools with EM simula- model the microstrip Y-junction circulator with Ansys HFSS
tion software opens up new perspectives for the study and in the classical way (Polder model). The frequency domain
design of microwave ferrite devices. simulations of the microstrip Y-junction circulator are done
for a frequency range from 6 GHz to 14 GHz. In the classical
IV. APPLICATION TO THE DESIGN OF A MICROSTRIP
EM simulations, the ferrite disk is assumed to be magnetized
Y-JUNCTION CIRCULATOR with uniform internal field of 1710 Oe. This static DC mag-
netic field is sufficient to saturate the ferrite material.
In order to validate the proposed theoretical modeling Simulated S-parameters of the Y-junction circulator are
approach in the limit case saturation, we modeled and real- shown in Fig. 10. The simulation results show a broadband
ized a microstrip Y-junction circulator using LTCC technol- behavior over the frequency range around X-band (see Fig.
ogy for an extended frequency band around X-band 10). The circulator shows minimum insertion losses of
frequencies. The determination of the initial geometric 0.44 dB at 9.8 GHz. Isolation and return losses remain below
parameters of microstrip Y-junction circulator is based on 15 dB between 8.3 GHz and 11.7 GHz, leading to a relative
Bosma principles.22 Starting from these initial geometric bandwidth of 34%. The insertion losses remain lower than
characteristics, the circulator design is optimized using 0.77 dB in this frequency band.
HFSS simulations (Fig. 9).
The quarter wavelength lines are designed on the LTCC A. Experimental results
compatible substrate T950 (made from MRA powder VLF-
The Y-junction microstrip circulator is fabricated with
220Aq3) and connected to the 50 X access lines. The width an additive multilayer process using LTCC technology. An
of the quarter-wavelength line is tuned using HFSS for maxi- YIG based ferrite material (chemical composition
mum isolation. The design does not show ternary symmetry Y2:28 Cu0:05 Bi0:67 Fe5 O12 ) is chosen for the core of the circu-
as the branches of the junction are not identical. lator.23 With the substitution of bismuth and copper cations,
The operation frequency is chosen above the FMR reso- the sintering temperature of the yttrium garnet is consider-
nance region. Yttrium Iron Garnet (YIG) ferrite substituted ably decreased from 1450  C to down to 950  C. This
with bismuth, and copper cations, is used for the design and
realization of microstrip Y-junction circulator. The ferrite mate- TABLE I. HFSS design—Material properties.
rial has the following properties: relative permittivity ef ¼ 21,
tan d ¼ 0.003, saturation magnetization 4pMs ¼ 1710 G, anisot- Material e tand 103 4pMs ðGÞ Ha (Oe) DHðOeÞ
ropy field Ha ¼ 50 Oe, and resonance linewidth (magnetic Ferrite 21 3 1710 50 10
losses) DH ¼ 10 Oe. The demagnetization coefficient of the fer- T950 21 3 … … …
rite disk is calculated to be 0.636.
234503-8 V K Thalakkatukalathil et al. J. Appl. Phys. 123, 234503 (2018)

the ferrite material. Compared to conventional stripline tech-

nology, this asymmetrical topology increases the non-
homogeneity of the internal DC biasing fields in the ferrite
disk.
The S-parameter measurements are done with Rohde &
Schwarz ZVA24 (4 port) vector network analyzer in the fre-
quency range from 6 GHz to 14 GHz. Figure 12 shows the
measured insertion loss, isolation, and return loss of the
microstrip circulator. Minimum insertion losses of 0.38 dB
are measured at 8.8 GHz, where the isolation level is 26 dB
and return losses are near 25 dB level. Isolation and return
losses remain below 15 dB from 7.46 GHz to 12.06 GHz (rel-
ative BW 47%). The insertion losses remain lower than 1 dB
in this frequency band. A smaller bandwidth can be defined
FIG. 10. Simulated (HFSS-Polder model) S-parameters of a microstrip Y-
junction circulator.
between 8 GHz and 11.37 GHz where the insertion losses are
below the 0.6 dB level. Measured results show a pronounced
decrease in sintering temperature allows the co-firing of bi-band behavior. This dual-band response helps the broad-
compatible dielectric materials with silver or gold. The addi- band behavior of the circulator.
The measured results show a shift in center frequency to
tion of bismuth ions increases the permittivity of the material
and subsequently reduces the core size of the circulator. lower frequency value compared to the HFSS simulations
Circulator ground plane, conductive strips, and junction using a Polder model. The measured isolation bandwidth is
are realized by screen printing layers of silver. The junction found to be higher than the simulated one. This difference is
radius is 1.82 mm. The quarter wavelength lines are depos- due to the fact that internal field in ferrite material is not
ited on the ferrite substrate and connected to the 50 X access homogeneous as assumed by classical HFSS simulations.
lines. The length and width of microstrip line are 1.8 mm and Different parts of the ferrite material may have different
0.93 mm, respectively. The ferrite disk has a radius of static biasing fields and different magnetization states.
1.82 mm and a thickness of 1 mm. The substrate material
used for access lines is LTCC compatible dielectric T950 B. Theoretical modeling tool: Non-uniform bias fields
with a relative permittivity 21 and a dielectric loss tangent
tan d ¼ 0.003. In order to understand the performance of the circulator
The experimental setup of the microstrip Y-junction cir- more accurately, an EM analysis is carried out considering
culator is shown in Fig. 11. The DC bias magnetic field is the non-homogeneity of internal static fields and a general-
applied normal to the ferrite disk by placing a Samarium ized permeability model. The inhomogeneity of the internal
R
Cobalt permanent magnet (SmCo magnet)—RecomaV 20 static magnetic fields in the ferrite is studied using the
above the junction which has a residual induction (Br) 9 kG magneto-static solver we have developed.19 This magneto-
and a coercivity of 8.8 kOe. The static field at the center of static analysis presents a more realistic description of inter-
the ferrite is controlled by placing a dielectric spacer in nal DC fields. For each mesh cell, this solver takes into
between the junction conductor and the magnet. The DC account a hysteresis cycle based on the Stoner-Wohlfarth
magnetic field provided by the magnet is enough to saturate energy model.

FIG. 11. Microstrip Y-junction circulator—Experimental setup.

234503-9 V K Thalakkatukalathil et al. J. Appl. Phys. 123, 234503 (2018)

tensor components are then imported into HFSS as frequency

dependent dataset points. Each of the annulus regions is
defined as a new material in Ansys HFSS by using the fre-
quency dependent functions obtained using an inbuilt piece-
wise linear interpolation function.
The frequency domain simulations are done for a fre-
quency range from 6 GHz to 14 GHz. The simulated S-
parameters are then compared with the experimental results
and with HFSS-Maxwell 3D combined simulations.
Measured and simulated S-parameters of microstrip Y-
junction circulator considering the non-uniform DC biasing
fields are shown in Fig. 14. Theoretical results show good
agreement with the experimental results and HFSS-Maxwell
FIG. 12. Measured and simulated (HFSS-Polder model) S-parameters of a simulations. These results validate the use of proposed theo-
microstrip Y-junction circulator. retical modeling approach in the limit case saturation. Our
approach shows good agreement with the results obtained
The ferrite disk shows a variation in internal field along with HFSS-Maxwell 3D combined simulations although
radial direction as well as along the thickness of the ferrite these simulations use the Polder model. This is due to the
material. The maximum field is observed at the sample face fact that GPT model gives similar results with the Polder
closest to the magnet. The ferrite regions farther from the model in the saturated state. From Fig. 14, it is clear that the
magnet show low internal biasing fields and this can possibly results obtained from our theoretical tool are closer to the
result in unsaturated areas in the ferrite disk. The anisotropic measurement results than those predicted by HFSS-Maxwell
dynamic properties (permeability tensor which appears in 3D combined simulations.
the Maxwell equations) of the material are dependent on the Proposed approach shows very good advantage over
magnitude and direction of internal DC biasing fields. conventional EM simulations considering the fact that GPT
In order to take the non-homogeneity of the internal model is valid at all magnetization states unlike Polder
static fields into account, the ferrite disk is divided into dif- model, which is only valid in the saturated state. Moreover,
ferent computational regions where the internal fields are for a given static field value, there may be several possible
considered to be uniform. The sample disk is divided into states of magnetization depending on the magnetization his-
ten concentric annuli regions around a central cylinder (see tory. HFSS-Maxwell 3D simulations cannot take this phe-
Fig. 13). In each region, the internal field is considered to be nomenon into account because a single major B-H hysteresis
uniform in magnitude and direction. EM properties of these curve is used to represent the static magnetic behavior of the
computational regions are characterized by their respective whole ferrite sample. In the proposed theoretical approach,
permeability tensors. Each region is defined as a separate each mesh cell can follow a different hysteresis loop (major
material with its own permeability tensor in HFSS simula- or minor) based on the Stoner-Wohlfarth energy model,20
tions. In each region, the permeability tensor components are which gives a more realistic description of internal DC
calculated using the GPT model considering the internal fields.
biasing field and hysteresis phenomenon. Integration of generalized permeability tensor model
For each region, input parameters of the GPT model are and magneto-static analysis with HFSS made the EM simu-
the same except for the internal field value. The predictive lations as a function of applied biasing field intensity possi-
nature of the GPT model makes it possible to find permeabil- ble in all states of magnetization. This approach will be more
ity tensor components for all the regions in a single computa- advantageous for the modeling and design of devices which
tion procedure. Real and imaginary parts of permeability used ferrite set in the completely demagnetized (miniaturized

FIG. 13. 3D modeling of ferrite disk—

Graphical representation of annulus
sectors exhibiting uniform internal
field. (a) Top view and (b) side view.
234503-10 V K Thalakkatukalathil et al. J. Appl. Phys. 123, 234503 (2018)

FIG. 14. Measured and simulated S-parameters of Y-junction microstrip circulator: Non-uniform biasing.

antennas24), partially magnetized (tunable filters,25 tunable accurately, at all magnetization states. This new theoretical
antennas26) or remanent (self-biased circulators,4 phase- tool will be advantageous for the design of devices working
shifters27) states. in the partially magnetized states: self-biased circulators/
insulators, miniature antennas, phase shifters, etc.
V. CONCLUSION
ACKNOWLEDGMENTS
In this paper, we presented a predictive theoretical
modeling approach to study the EM properties of ferrites in a This work was funded under the framework of
predictive way, regardless of the magnetization state of the EURIPIDES type project LOCCIMIM (Low Cost
medium. The proposed theoretical tool combines a magneto- Circulators for Microwave Modules). We would like to
static solver, a generalized permeability tensor model, and a thank Cobham Microwaves for the measurement of S-
commercial EM simulation software Ansys HFSS. Proposed parameters of the circulator. All calculations in this work
theoretical tool is validated by comparing calculated and were done on the high performance computing cluster
measured power absorption spectra in APC7 coaxial line WINCHYPS1. This cluster is managed by the Tec’Hyp
loaded with a YIG ferrite placed in the saturation state. Our platform of the Lab-STICC.
approach is then applied to the design of a microstrip Y-
junction circulator for an extended frequency range around 1
V. G. Harris, IEEE Trans. Magn. 48, 1075 (2012).
X-band, and a demonstrator is realized in LTCC technology. 2
M. A. Tsankov and L. G. Milenova, J. Appl. Phys. 73, 7018 (1993).
A good agreement is observed between the S-parameters of 3
S. A. Oliver, P. Shi, N. E. McGruer, C. Vittoria, W. Hu, H. How, S. W.
the circulator calculated from our theoretical approach and McKnight, and P. M. Zavracky, IEEE Trans. Microwave Theory Tech. 49,
385 (2001).
those measured. 4
V. Laur, G. Verissimo, P. Queffelec, L. A. Farhat, H. Alaaeddine, E.
Thanks to the generalized character of the GPT model, Laroche, G. Martin, R. Lebourgeois, and J. P. Ganne, IEEE Trans.
this theoretical tool will be applicable for the design of Microwave Theory Tech. 63, 4376 (2015).
5
ferrite-based devices regardless of the state of magnetization B. K. O’Neil and J. L. Young, IEEE Trans. Microwave Theory Tech. 57,
1669 (2009).
of the magnetic medium. By taking into account the inhomo- 6
H. How, S. A. Oliver, S. W. McKnight, P. M. Zavracky, N. E. McGruer,
geneity of the internal biasing fields, the proposed modeling C. Vittoria, and R. Schmidt, IEEE Trans. Microwave Theory Tech. 47,
tool can predict the dynamic behavior of ferrite devices more 1982 (1999).
234503-11 V K Thalakkatukalathil et al. J. Appl. Phys. 123, 234503 (2018)

7 19
A. Guennou, P. Queffelec, P. Gelin, and J. L. Mattei, IEEE Trans. Magn. A. Le Gouellec, G. Verissimo, V. Laur, P. Queffelec, I. Albert, and T.
43, 3645 (2007). Girard, J. Appl. Phys. 120, 073902 (2016).
8 20
V. Laur, J.-L. Mattei, G. Verissimo, P. Queffelec, R. Lebourgeois, and J.- E. C. Stoner and E. P. Wohlfarth, Philos. Trans. R. Soc., A 240, 599
P. Ganne, J. Magn. Magn. Mater. 404, 126 (2016). (1948).
9 21
D. Polder, Physics 15, 253 (1949). C. G. Broyden, Math. Comput. 19, 577 (1965).
10 22
L. D. Landau and E. Lifshitz, Phys. Z. Sowjetunion 8(153), 101–114 H. Bosma, Proc. IEE Part B: Electron. Commun. Eng. 109, 137 (1962).
23
(1935). L. Qassym, G. Cibien, R. Lebourgeois, G. Martin, and D. Colson, Int.
11
A. Guennou, B. Della, P. Queffelec, P. Gelin, and J. L. Mattei, IEEE Symp. Microelectron. 2016, 000586.
24
Trans. Magn. 43, 2642 (2007). F. Canneva, F. Ferrero, A. Chevalier, J.-M. Ribero, J.-L. Mattei, P.
12
E. Schloemann, J. Appl. Phys. 41, 204 (1970). Queffelec, and R. Staraj, Microwave Opt. Technol. Lett. 55, 2007 (2013).
13 25
J. J. Green and F. Sandy, IEEE Trans. Microwave Theory Tech. 22, 641 X. Yang, J. Wu, S. Beguhn, T. Nan, Y. Gao, Z. Zhou, and N. X. Sun,
(1974). IEEE Microwave Wireless Compon. Lett. 23, 184 (2013).
14 26
M. Igarashi and Y. Naito, IEEE Trans. Magn. 13, 1664 (1977). L. Huitema, T. Reveyrand, J. L. Mattei, E. Arnaud, C. Decroze, and T.
15
G. T. Rado, Phys. Rev. 89, 529 (1953). Monediere, IEEE Trans. Antennas Propag. 61, 4456 (2013).
16 27
P. Gelin and P. Queffelec, IEEE Trans. Magn. 44, 24 (2008). A. Abuelma’atti, I. Khairuddin, A. Gibson, I. Morgan, and A. Haigh, in
17
P. Gelin, P. Queffelec, and F. Le Pennec, J. Appl. Phys. 98, 053906 (2005). 2007 IEEEMTT-International Microwave Symposium (2007), pp.
18
D. Polder and J. Smit, Rev. Mod. Phys. 25, 89 (1953). 2067–2070.