<}

}<

Computation of the Capacitance Matrix of

Manhattan Geometry Planar Conductors
Embedded in Multilayered Substrates
Enrique Drake, Francisco Medina, Manuel Horno
Microwaves Group, Department of Electronics and Electromagnetism, Facultad de Fisica,
Avda. Reina Mercedes s / n, 41012 Sevilla, Spain; e-mail: [email protected]
Recei ed 26 September 1997; re ised 28 April 1998

ABSTRACT: In this paper we propose an efficient technique for computation of the

capacitance matrix of a set of infinitely thin conductor patches embedded in a multilayered
medium. The patches present a manhattan-type shape, i.e., they can be subdivided into a
finite number of rectangular regions. The generalized biconjugate gradient method (GBGM)
in conjunction with FFT algorithms, is adapted to solve the convolution integral equation
governing the free-charge density distribution on the conductors. Important computational
improvements are achieved by including asymptotic extraction techniques in the determination of the space domain periodic Greens functions. The analysis is also applied to the
quasistatic modelling of some microstrip discontinuities. Q 1998 John Wiley & Sons, Inc. Int J
RF and Microwave CAE 8: 386]397, 1998.

Keywords: capacitance; planar; manhattan; multilayer; GBGM]FFT

1. INTRODUCTION
Planar multiconductor microstrip structures consisting of transmission lines, discontinuities, and
lumped elements are nowadays usually found in
microwave integrated circuits and devices. In
many situations of practical interest, particularly
at low frequencies, these structures can be accurately represented by lumped circuit models derived on the basis of static analysis. For this
reason, many papers that deal with computation
of the capacitance matrix of microstrip patch
capacitors have been published. The most widely
used approach is to formulate an integral equation for the surface free-charge density on the
conductors and to solve it by using the method of
moments w1]3x or variational techniques w4]6x.
Correspondence to: Manuel Horno.
Contract grant sponsor: DGICYT, Spain, Project TIC950447.
Q 1998 John Wiley & Sons, Inc.

386

Concerning the geometry of the patches, many

authors have focused their attention on the analysis of rectangular, circular, and elliptical patches
w6]9x. For these geometries, the method of moments is particularly suited because the freecharge density distributions on the conductors
can be expanded in terms of appropriate entire
domain basis functions. This may not be feasible
for more general shaped patches of practical interest. The analysis of arbitrary geometries implies either the adoption of more computationally
intensive techniques w10]12x or the exploitation
of subdomain basis functions in the method of
moments w1, 3, 13x. The main drawback of using
subdomain functions in the method of moments
is that a large set of these basis functions is
usually needed to approximate the free-charge
distributions, thus leading to very large systems of
linear algebraic equations. In an attempt to overcome the aforementioned drawback of the conCCC 1096-4290r98r050386-12

Capacitance Matrix of Manhattan Conductors

ventional method of moments, conjugate gradient-type iterative techniques offer an alternative

tool for solving linear operator equations w14, 15x.
However, integral operators have to be discretized in order to become amenable to digital
computer computation, and therefore it has been
argued w16x that almost all implementations of
these iterative techniques are more or less equivalent to moment method solutions using subdomain basis functions. Anyway, the large-size systems of linear algebraic equations can be more
efficiently solved by an iterative method than by a
direct inversion method e.g., Gaussian elimination. because the former circumvents the excessive storage problems inherent in the latter. In
addition, a recurrent algorithm can be stopped
once a prespecified degree of accuracy in the
solution is reached, thus resulting in CPU time
savings.
In the present paper, we are interested in the
static characterization of microstrip patch structures such as that shown in Figure 1 i.e., we hope
to compute its capacitance matrix.. This generic
structure is composed of a set of infinitely thin
metallizations located at a given plane in a multilayered medium. The geometry of each conductor
is manhattan type w2x, i.e., it can be subdivided
into a finite number of rectangular subregions.
This manhattan geometry includes a wide variety
of flat metallization geometries of practical importance as particular cases. The problem is conveniently formulated in terms of the solution of
the convolution integral equation for the surface
free-charge density on the conductors. The choice
of a method of moments with entire domain basis
functions is not readily available to analyze the

Figure 1. Manhattan-type multiconductor structure

embedded in a multilayered medium.

387

manhattan geometry, and this prompts us to use

subdomain basis functions. However, the discretization of the integral equation is performed
in such a way that the entries of the resultant
system of linear equations retain the convolution
nature of the integral kernel. At this point, the
large size of the linear equations system obtained
has led us to use an efficient iterative technique
for its solution: the generalized biconjugate gradient method GBGM. w17x in conjunction with
FFT algorithms for exploiting the convolutional
properties. The GBGM-FFT was successfully applied by the authors to the quasi-TEM characterization of planar transmission lines in w18x, and
now it has been adapted to treat the present
problem. There have been several previous attempts to apply iterative methods to the solution
of this type of electrostatic problem w19]22x.
However, the original contribution of the present
work resides in the following features:
v

The assumption of manhattan-type modelling for the conductor geometries, which

enlarges the scope of the analysis.
The adaptation of the GBGM-FFT to the
solution of the system of convolutional linear equations.
The rigorous treatment of the real or fictitious compelled by the use of FFT algorithms in the computation of the spatialdomain Greens function. periodicity of the
structures.
The special effort devoted to find a good
compromise between accuracy and computational cost in the determination of the
spatial-domain periodic Greens function
via the inclusion of asymptotic extraction
techniques.

A Fortran-77 code implementing the iterative

algorithm has been written on a Convex-240 computer. The reliability of this computer code has
been conveniently tested with data available from
the literature. A good benchmark for the method
is the comparison with data for the capacitance
parameters of rectangular microstrip patches and
some microstrip discontinuities. For these particular situations, accurate and reliable methods
w4, 6x are available from the open literature. The
method proposed in this paper is reasonably competitive, even for particular problems for which
specific techniques exist.

388

Drake, Medina, and Horno

2. STATEMENT OF THE PROBLEM:

THE INTEGRAL EQUATIONS
Figure 1 shows the generic planar multiconductor
structure under analysis. The substrate is a stratified medium composed of Nl layers of lossless
dielectric materials. Cylindrical symmetry in the
electrical properties of the substrate is guaranteed by assuming either isotropic materials or
uniaxial anisotropic ones with the optical axis
parallel to the Y-axis. According to this, each
layer of the substrate is characterized by a permittivity tensor given by the expression

we ix s

e xi x

e yi y

e xi x

i s 1, . . . , Nl . 1 .

The lower boundary of the configuration interface 0. is a grounded plate and the upper boundary interface Nl . is any one of three possibilities:
grounded plate, magnetic wall, or open boundary.
An arbitrary number, Nm , of manhattan-type
metallizations are printed on the Mth interface
of the multilayered substrate. These metallizations are assumed to be infinitely thin and perfect
conductors. A top view of the metallized interface
is shown in Figure 2a.
In order to apply FFT algorithms discrete and
circular in nature . to the computation of the
Greens function of the structure, we will consider the configuration enclosed between four
lateral walls of periodic repetition dashed lines
in Fig. 2a.. These walls are actually present if we
deal with periodic structures, but aperiodic structures can also be conveniently handled if the
lateral walls are removed far away from the metallized region so that the adjacent cells do not
have a significant effect on the value of the potential around the actual metallizations. Therefore, in any case, we analyze a periodic array
structure whose unit cell of repetition has the
dimensions T0 x = T0 z .
For simplicity, we define the following regions
at the Mth interface see Fig. 2a.: Sc , the whole
region bounded by the unit cell of repetition; Sk ,
the region occupied by the kth manhattan-type
metallization; S, the whole metallized region deNm
.
fined by S s D ks
1 S k ; S, the complementary
region of S; and Sr , the minimum rectangular
region containing all the Sk regions.
It is obvious that each capacitance coefficient,
Ck m , of the multiconductor structure can be iden-

Figure 2. Top view of the metallized interface. a.

Unit cell of periodic repetition. b. Rectangular discretization.

tified as the free charge on the kth conductor

when the voltage of the mth conductor is set to
unity and the remaining ones to zero mth canonical voltage excitation.. Taking into account also
that the free charge on each conductor can be
computed by integrating the free-charge density
on its surface, the computation of the capacitance
matrix, w C x, is straightforward from the solution
of the Nm systems of integral equations m s
1, . . . , Nm .

HHS G x y x9, z y z9. s x9, z9. dx9 dz9 s d

km ,

x, z . g Sk ,

s x, z . s 0,

k s 1, . . . , Nm ;

x, z . g S,

2.

where s x, z . is the surface free-charge density

on the metallized interface, d k m is the Kronecker
delta 1 if k s m; 0 if k / m., and the integral
kernel, G x y x9, z y z9., is a specialization Mth
interface . of the spatial-domain Greens function
of the periodic structure.

Capacitance Matrix of Manhattan Conductors

3. APPLICATION OF THE
GBGM-FFT ALGORITHM
The solution of 2. by means of a digital computer requires the previous discretization of these
systems of integral equations. The first step has
been to discretize the spatial domain of integration in the way illustrated in Figure 2b. The
region Sc has been subdivided into Nc x = Nc z
identical rectangular subdomains with dimensions
Tx = Tz this implies Nc x Tx s To x and Nc z Tz s
To z .. Note that each subdomain must be entirely
either inside or outside the metallized region S.
In particular, Nr x = Nr z subdomains are inside
the rectangular region Sr . We can define a pair of
discrete variables, n x , n z , related to the continuous variables x, z, in such a way that the coordinates of the center of each subdomain are x s
n x q 1r2.Tx and z s n z q 1r2.Tz . This discretization of the Mth interface makes it possible
to associate a new region S cD , SkD , . . . . defined on
the discrete variables with each region Sc ,
Sk , . . . , . previously defined on the continuous
variables.
As a second step, both the surface free-charge
density and the periodic spatial-domain Greens
function have been assumed to be constant in
each rectangular subdomain and equal to their
values at the center of that subdomain, s n x 9, n z 9
and G n xyn x 9, n zyn z 9 , respectively. The singular
sample G 0, 0 is replaced by a suitable integral
average.
Taking into account these approximations and
evaluating 2. only at the centers of the rectangular subregions inside Sc , each system of integral
equations is converted into the following system
of Nc x = Nc z linear algebraic equations m s
1, . . . , Nm .:
Nc xy1 Nc zy1

G n xyn x 9 , n zyn z 9s n x 9 , n z 9 s

n xXs0 n z9s0

n x , n z . g SkD ,

s n x , n z s 0,

dk m
Tx Tz

k s 1, . . . , Nm ;
n x ,n z . g S D .

3.
4.

From the point of view of a method of moments

practitioner, this discretization technique may be
considered as a particular case of delta function
expansion and weighting. Observe that the uniform size of the rectangular subdomains has made

389

possible the propagation of the convolutional nature from the integral equation system of 2. to
the linear equation system of 3.. In addition, the
condition stated in 4. allows the reduction of 3.
to the smaller system
G n xyn x 9, n zyn z 9s n x 9 , n z 9

n x9, n z9 .gS rD

dk m
Tx Tz

n x , n z . g S kD ,
k s 1, . . . , Nm .

5.

In fact, according to 4., it is sufficient to sweep

the region S D by the discrete variables n x 9 , n z 9 ..
However, it is preferable to sweep the rectangular
region SrD in order to explicitly identify the first
member of 5. as a finite bilinear convolutional
sum.
The large size of the system in 5. makes the
choice of an iterative method advisable for its
solution. In our experience, the generalized biconjugate gradient method GBGM. often converges with a rate many times faster than the
ordinary conjugate gradient method CGM. w18x.
In addition, a computationally compact form of
the GBGM involving less storage requirements
than the general form has been proposed w17, eqs.
88. ] 94.x for the solution of symmetric operator
equations. We have chosen this compact GBGM
algorithm for solving 5. since the symmetry of its
coefficient matrix can be proved the next section
deals with the computation and properties of this
coefficient matrix.. Observe that the iterative solution of 5. by the GBGM requires the recurrent
computation of one finite bilinear convolutional
sum for each iteration. Important CPU time savings are obtained if this bilinear convolution is
converted into a bicyclic convolution, thus taking
advantage of the use of bidimensional FFT algorithms. Furthermore, the use of FFTs allows us to
exploit the redundant character of the coefficient
matrix of the system in order to minimize the
storage requirements.
Once 5. has been solved by the GBGM-FFT
algorithm, the free charge, Q k , existing on the
kth conductor k s 1, . . . , Nm . can be approximately computed by integrating the discretized
version of the free-charge density on the surface
of that conductor:
Q k f Tx Tz

n x , n z .gS kD

s nx , nz.

6.

Drake, Medina, and Horno

390

Finally, the capacitance matrix of the multiconductor structure is determined by repeating the
process for the Nm canonical voltage excitations.

4. TREATMENT OF THE
GREENS FUNCTION
Prior to formulating the integral equation system
of 2., we need to derive the spatial Greens
function, G x y x9, z y z9., for the periodic geometry under consideration. In practice, according to the discretization process, we only have to
compute the coefficients of the linear equation
system in 5., which are defined by

n u , n

s~

Txr2

n u s n s 0,

G n T , n T . ,

u x

sG

n x9 , nz9. g

S rD ,

q`

1
To x To z

7.

'a

q b 2 . 10.

q`

Go , n

a , nb

nasy` nb sy`

sin na prNc x . sin nb prNc z .

elsewhere,

n xyn x 9 , n zyn z 9

with g s

From Fourier transform theory, it can be proved

that the coefficients G n u , n defined in 7. are
o a , b . by
related with the SDGF G

where the singularity of G0,0. has been replaced

by the integral average of G x y x9, z y z9. in
the central rectangle wyTxr2, Txr2x = wyTzr2,
Tzr2x.
Note that the maximum number of coefficients
G n u , n involved in the solution of 5. is Nr x =
Nr z . 2 :
n u , n

o a , b . s Go g .
G

Tzr2

H H G u, . du d ,
T T yT r2 yT r2
x

tition walls from the original structure. The twodimensional spectral-domain Greens function
o a , b ., of this open structure is availSDGF., G
able in closed form, for instance, by using the
recurrent algorithm reported in w6x. In the present
case, this algorithm becomes especially simple
because the cylindrical symmetry of the substrates imposes on the SDGF the following symmetric dependence of the spectral variables:

G n u , n s

na prNc x .
q`

q`

Go , n

a , nb

nasy` nb sy`

=exp 2p

na n u
Nc x

nb n

11.
S rD .

8.

However, the convolutional nature of the linear

equation system induces high redundancy in the
coefficients 8.. Indeed, the convolution variables,
n u , n , sweep the region SGD defined by,
SGD s  ; n u , n . < y Nr x y 1 . F n u F Nr x y 1 . ;
y Nr z y 1 . F n F Nr z y 1 .4 ,

Nc z

elsewhere,

nx , nz . g

n u s n s 0,
1
To x To z

nb prNc z .

9.

which involves only 2 Nr x y 1. = 2 Nr z y 1. different coefficients. The reduction of storage requirements arising from these redundancies is the
main advantage of the knowledge of the Greens
function in the spatial domain. Furthermore, the
cylindrical symmetry of the substrates introduces
even more redundancies G n u , n s G n u , yn s
Gyn u , n s Gyn u , yn ., yielding a maximum of
Nr x = Nr z different coefficients.
Our starting point to compute G n u , n has been
the consideration of the laterally open structure,
which is obtained by removing the periodic repe-

where

o , na , nb s Go na Ta , nb Tb . ,
G
Ta s

2p
To x

Tbs

2p
To z

12.

has been used as the notation.

As a first possibility, we might truncate the
series of 11., retaining only Nc x = Nc z terms, and
compute their sums by two-dimensional inverse
FFTs. However, the slow convergence of these
double series would force us to use very large
values for Nc x and Nc z this is equivalent to make
the computation domain very large andror the
discretization very fine. in order to reduce the
inherent error associated with the spectral truncation. Owing to this, it is preferable in the previous application of an asymptotic extraction technique on the SDGF to minimize the storage requirements and the CPU time of the inverse
FFTs. This is our next aim.

Capacitance Matrix of Manhattan Conductors

aso a , b . be the spectral Greens function

Let G
of an isotropic and homogeneous medium, with
dielectric permittivity eas , bounded by a grounded
plate at a distance of h as from the source-field
plane. Its analytical expression is
y1

aso a , b . s Gaso g . s w eas g 1 q coth g h as ..x .

G
13.

a , nb

To x To z
=
G n u , n s

eas s

si s

e xi x

/
e yi y

na prNc x .
1

Txr2

nb prNc z .
Tzr2

H H G
T T yT r2 yT r2

as

u, . du d ,

n u s n s 0,
1

1
e yMq
sMq1
y

na nb

aso , na , nb
yG

sin na prNc x . sin nb prNc z .

Let us define the parameters eas and h as as

e yMy sM

Go , n

391

Tx Tz

o , na , nb y Gaso , na , nb
FFTy1 G

qGas n uTx , n Tz . ,

elsewhere,
15.

o 0 . ,
h ass eas G

1r2

i s M, M q 1 . .

14.

Then, the asymptotic extraction technique consists of carrying out the computation of the coefficients G n u , n by

Gas x y x9, z y z9 . s

q`

q`

4peas

k xsy` k zsy`

where both the double sum and the two-dimensional inverse FFT extend over Nc x = Nc z pairs
of values of the indexes na and nb , and where
Gas x y x9, z y z9. is the spatial Greens function
corresponding to the periodic counterpart with
period To x = To z . of the structure whose spectral
aos a , b ., i.e.,
Greens function is G

'

x y x9 y k x To x . q z y z9 y k z To z . 2

' x y x9 y k T

x ox

With the choice of eas made in 14., the asymptotic spectral behavior g `. of the functions
aso g . and Go g . is the same. Therefore, the
G
o y Gaso has
residual series of 15. associated to G
a narrower range of terms significantly different
from zero, thus allowing a drastic reduction in the
size of the two-dimensional inverse FFT. In spite
of this computational advantage, the reduction of
Nc x = Nc z has the drawback of causing a decrease
in the discretization level which may damage the
accuracy of the results. However, this drawback
can be overcome by a posterior bicubic interpolation process aimed at increasing the discretization
level. On the other hand, with the choice of h as
o y Gaso is not very
made in 14., the function G
different from zero in the surroundings of g s 0,
thus avoiding numerical problems.

. q z y z9 y k z To z . 2 q 4 h 2as

16 .

In addition, the computational feasibility of an

asymptotic extraction technique requires the analytical or quasianalytical availability of the asymptotic tail wlast term of 15.x. In this sense, the
efficient computation of the function Gas xyx9,
z y z9. has been the main difficulty of the present
technique, whereas in w18x the homologous function of a single variable could be analytically
obtained in closed form. However, we have computed Gas x y x9, z y z9. by a quasianalytical approximation based on the physical meaning of
this function. Observe that the expression of
Gas x y x9, z y z9. given in 16. may be identified
as the electrostatic potential at the upper charge
plane of the double periodic array of unit charges
shown in Figure 3a. We have replaced this distribution of sources by the hybrid model shown in

392

Drake, Medina, and Horno

Figure 3. a. Double periodic array of unit charges

whose electrostatic potential is Gas x y x9, z y z9..
b. Hybrid model of discrete and continuous charges to
approximate Gas x y x9, z y z9..

Figure 3b, which retains the discrete nature of

the charges neighboring the observation point but
converts the rest of the charges into a continuous
distribution. The contribution of the discrete
charges is a finite double summation involving
few terms. The contribution to the potential of
the continuous region of charge is analytical. The
main nontrivial result necessary to obtain such
analytical expression is the value of the potential
created by a rectangular uniform surface charge
distribution, which can be derived easily from the
results reported in w23, pp. 245]246x. The computational key of this hybrid model lies in the number 2 Nd x q 1. = 2 Nd z q 1. of charges whose
discrete nature must be preserved according to
the accuracy desired. Our empirical observations
drive us to assert the prime influence of the
geometrical ratio To x = To z .rh2as on the number
of discrete charges required for the model.

5. NUMERICAL RESULTS
The technique presented in this work has been
implemented in Fortran language on a Convex240 computer. The performance of the code has
been exhaustively checked, and results have been
compared with those available in the literature to
ensure its reliability and accuracy. First, several

aspects related to the convergence behavior of

the method have been investigated. To restrict
the size of the present paper, we collect here only
the most outstanding conclusions arising from
these convergence analyses. So, in the analysis of
aperiodic geometries, we have observed a strong
relation between the adequate size of the unit
cell of periodic simulation and the geometrical
features of the real structure analyzed. It has
been noticed that any circumstance forcing a
closer confinement of the electrostatic field to the
region near the metallizations, e.g., a tight proximity between conductors, allows us to select
smaller dimensions for the periodic unit cell. On
the other hand, the algorithm GBGM-FFT has
been compared with other iterative techniques
especially, algorithms derived from the ordinary
CGM. for the solution of the system of linear
algebraic equations, and the results of these comparisons confirm that the GBGM-FFT provides a
very good compromise between convergence rate
and convergence stability.
We have paid particular attention to the check
of the advisability of the asymptotic extraction
technique in the computation of the periodic
Greens function. The following numerical example illustrates the advantages of this asymptotic
extraction. The relative capacitance of a square
patch width w . on alumina substrate e r s 9.6.
supported on a grounded plate wrh s 0.1. has
been found to be Crw s 220.8 pFrm. These four
significant digits for the capacitance are obtained
if the patch is discretized by 20 = 20 square
subdomains and the ratio periodic cellrpatch is
60 = 60. Without asymptotic extraction, we need
a two-dimensional inverse FFT of 1200 = 1200
samples involving 34 s on the Convex-240. With
asymptotic extraction, the residual series only requires a 240 = 240 inverse FFT, which only spends
1.8 s on the same machine a bicubic interpolation factor of 5 = 5 allows us to keep the same
accuracy level in the result.. However, this reduction of CPU time and storage requirements is not
the only advantage of the asymptotic extraction
technique. The accuracy of the results is also
increased because the central region of the periodic spatial Greens function, involved in the convolution process, is almost analytically built up.
Note that this region of the spatial Greens function is affected mainly by the spectral asymptotic
tail, which has been quasianalytically computed.
As stated before, rectangular patches constitute a good benchmark to check our method

Capacitance Matrix of Manhattan Conductors

since this geometry is amenable to analysis with

accurate techniques different from those discussed in this paper. Table I shows our results for
two coupled rectangular patches using three discretization levels. Note that the increase in the
number of samples is obtained just by increasing
the interpolation factor for the computation of
the Greens function so as to minimize the size
of the inverse Fourier transform used to build up
the spatial Greens function. Thus, we start with a
1.
.
low discretization level Nr1.
x s 10, Nr z s 11 and
more samples are obtained simply by applying the
interpolation procedure mentioned in the previ n.
1.
ous section Nrxn. s n = Nr1.
x ; Nr z s n = Nr z ; n
being the interpolation factor.. Note that the
values of the capacitances computed for n s
1, 2, 3 can be used in a simple extrapolation
scheme to meaningfully increase the accuracy of
the results. In terms of CPU time consumption, it
is better to use low levels of discretization combined with extrapolation than to use directly a
large amount of samples for a given accuracy..
The extrapolated results in Table I are in very
good agreement with converged data computed
with the method reported by some of us in w24x
the method in w24x is especially suitable for rectangular conductors and makes use of entire do-

393

Figure 4. Electrode capacitance matrix for a GaAs

FET as a function of width offset gate.. Comparison
with the results of w13x. Data: e r s 12.5, h s 200 m m,
w 2 s L2 s 25 m m, L1 s L3 s 100 m m. Asterisks *.
denote data computed with the method in w24x. Capacitances in fF s 10y15 F..

main basis functions with edge singularity.. Similar agreement with the results computed with the
method in w24x has been confirmed for a wide
variety of geometries. Our CPU times are also
quite similar to the ones required by the technique in w24x. Another example involving coupled
rectangular patches is shown in Figure 4. In this
figure we plot our results for the two-terminal
capacitance matrix of a configuration which represents the source and drain electrodes of a GaAs
FET with a gate bonding pad offset from them.
Comparison is made with the results obtain in
w13x by using the method of moments. An excellent agreement is found between our diagonal
capacitance coefficients and those reported in
w13x. On the contrary, some discrepancies are
observed in the off-diagonal capacitance coefficients, which typically have a higher relative numerical error. However, once again, the offdiagonal coefficients are very close those computed by the method in w24x. Since that method
and the method in this paper are completely
different, it seems that our results for the mutual
capacitances are also accurate enough in this
example.
Another interesting problem that can be used
as benchmark for our method is the computation
of the capacitances of the equivalent circuits for
microstrip open-end and gap discontinuities.
These parameters can be very accurately computed by using the excess charge model in
conjunction with the spectral domain method w25,

394

Drake, Medina, and Horno

26x. It is also possible to compute these capacitances from the capacitance matrix of adequately
chosen systems of rectangular patches by extracting the contribution to the overall capacitances of
the portion due to straight line sections, such as
was done, for instance, in w4]6x. Since our method
deals only with the computation of the matrix
capacitance of a system of printed patches i.e., it
does not incorporate the excess charge model.,
we are forced to use the latter technique, which is
intrinsically poorer than the former for such purpose the reason for this is that even very small
errors in the computation of the capacitance matrix elements of the patch system might result in
large error for some of the capacitances of the
equivalent circuit of the discontinuity.. Therefore,
reasonably good agreement of our computed results with those reported in other papers using
the excess charge model would be indicative of
high performance of the method proposed in this
paper when applied to the computation of the
capacitance matrix of coupled rectangular conducting patches. Thus, as a first example, we have
computed the open-end capacitance of a microstrip in vacuo that also was analyzed in w25x.
For wrh s 10 we obtain C o crw s 11.77 pFrm
by using four samples along the strip width direction, interpolation factors n s 3, 4, 5, and subsequent extrapolation 16-s CPU time.. The value
reported in w25x was C o crw s 11.7 pFrm. This
degree of agreement requires one to compute the
capacitance of the auxiliary rectangular patch with
accuracy better than 0.01%. As a second example,
we consider a more complicated situation in Fig-

Figure 5. Series and parallel capacitances in femtofarads. of the p network of an asymmetric gap microstrip in suspended configuration. Data: e s 10.4e 0 ,
h1 s h 2 s 635 m m, h 3 s 10 h1, w 1rh1 s 1.

ure 5. We plot in this figure the values of the

capacitances of the p-circuit model for an asymmetric gap in the microstrip. Our results and the
results reported in w26x excess chargemodel.
agree quite satisfactorily. However, it should be
recognized that the method we employed in this
paper is not accurate for the small parallel gap
capacitances arising in tightly coupled configurations. Figure 6 illustrates this fact for a symmetrical microstrip gap configuration in a multilayeredranisotropic medium. Although good agreement with the results reported in w25x can be
noticed for a wide range of values of sr h1 q h 2 .,
important discrepancies appear in the C prw parameters when sr h1 q h 2 . becomes small. The
accurate computation of C prw in such cases
would require extremely high precision in the
determination of the capacitance matrix of the
auxiliary patches. Anyway, from a practical point
of view, these discrepancies are not too meaningful since the circuital behavior of the discontinuity is dominated in such cases by C srw, which is
accurately computed by using any of the mentioned methods.
Let us finish this section by analyzing a nonrectangular manhattan geometry. Table II shows
the two-terminal capacitance matrix for a configuration of two manhattan patches that were previously analyzed by different versions of the
method of moments in w1x and w10x. We present
our results for different levels of discretization

Figure 6. Series and parallel capacitances normalized

to w . of the equivalent circuit of a symmetric gap
microstrip on sapphire e xsx s e zsz s 9.4e o , e ysy s
11.6 e o . in inverted solid line, e 2 s e o , e 3 s w e s x., suspended dashed line, e 2 s w e s x, e 3 s e o ., and sandwiched dot]dashed line; e 2 s e 3 s w e s x. configuration. In all cases h1 s h 2 s h 3 and wr h1 q h 2 . s 1.

Capacitance Matrix of Manhattan Conductors

395

tor microstrip structures by offering an efficient

technique to compute the capacitance matrix of
a set of manhattan-type patches. The proposed
method combines the versatility of a method of
moments using subdomain basis functions with
the computational efficiency of an iterative algorithm using FFTs. Special attention has been paid
to obtaining an efficient means for computing the
infinite-array periodic spatial Greens function of
the structures analyzed. The validity of the method
has been checked by testing its numerical convergence behavior. The results obtained have been
compared with those appearing in the literature,
and good agreement has been found in most
cases. The method can be used with confidence in
all those situations where the capacitance matrix
of manhattan-type conducting patches embedded
in layered substrates is required.

REFERENCES

interpolation factor ranging from n s 1]5. We

can see that our results for n s 1 agree quite well
with the data reported in w1x. This is not surprising, since the authors of that paper used exactly
the same discretization. The slight discrepancy for
C12 with the result reported in w10x can be attributed to the different discretization scheme
used by those authors. However, if we systematically increase the discretization level by increasing n., the values of the capacitance matrix elements steadily converge to new values which are
slightly larger C11 and C22 . or meaningfully
higher C12 . than those reported in w1x and w10x.
We conclude that the discretization level used in
those papers was not detailed enough. All these
tests confirm that the method reported in this
paper is useful and efficient for the electrostatic
analysis of rectangular and nonrectangular manhattan-type conducting patches.

CONCLUSIONS
The current paper contributes to the computeraided analysis and design of planar multiconduc-

1. P. D. Patel, Calculation of Capacitance Coefficients for a System of Irregular Finite Conductors

on a Dielectric Sheet, IEEE Trans. Microwa e
Theory Tech., Vol. MTT-19, Nov. 1971, pp. 862]869.
2. P. Balaban, Calculation of the Capacitance Coefficients of Planar Conductors on a Dielectric
Surface, IEEE Trans. Circuit Theory, Vol. CT-20,
Nov. 1973, pp. 725]731.
3. A. E. Ruehli and P. A. Brennan, Efficient Capacitance Calculations for Three-Dimensional Multiconductor Systems, IEEE Trans. Microwa e Theory Tech., Vol. MTT-21, Feb. 1973, pp. 76]82.
4. B. Bhat and S. K. Koul, Lumped Capacitance,
Open-Circuit End Effects, and Edge-Capacitance
of Microstrip-Like Transmission Lines for Microwave and Millimeter-Wave Applications, IEEE
Trans. Microwa e Theory Tech., Vol. MTT-32, Apr.
1984, pp. 433]439.
5. S. K. Koul and B. Bhat, Evaluation of End Effects
and Lumped Capacitance of Microstrip with
Anisotropic Substrates, Proc. 14th European Microwa e Conf., 1984, pp. 148]153.
6. R. R. Boix and M. Horno, Lumped Capacitance
and Open End Effects of Striplike Structures in
Multilayered and Anisotropic Substrates, IEEE
Trans. Microwa e Theory Tech., Vol. 37, Oct. 1989,
pp. 1523]1528.
7. S. Coen and G. M. L. Gladwell, A Legendre
Approximation Method for the Circular Microstrip
Disk Problem, IEEE Trans. Microwa e Theory
Tech., Vol. MTT-25, Jan. 1977, pp. 1]6.

396

Drake, Medina, and Horno

8. R. R. Boix and M. Horno, Capacitance Computation of Elliptic Microstrip Disks in Biaxial

Anisotropic Multilayered Substrates, IEEE Trans.
Microwa e Theory Tech., Vol. 38, Jan. 1990, pp.
30]37.
9. A. K. Verma and Z. Rostamy, Static Capacitance
of Some Multilayered Microstrip Capacitors, IEEE
Trans. Microwa e Theory Tech., Vol. 43, May 1995,
pp. 1144]1152.
10. P. Benedek, Capacitances of a Planar Multiconductor Configuration on a Dielectric Substrate by a
Mixed Order Finite-Element Method, IEEE Trans.
Circuits Syst., Vol. CAS-23, May 1976, pp. 279]284.
11. M. Naghed and I. Wolff, Equivalent Capacitances
of Coplanar Waveguide Discontinuities and Interdigitated Capacitors Using a Three-Dimensional
Finite Difference Method, IEEE Trans. Microwa e Theory Tech., Vol. 38, Dec. 1990, pp.
1808]1815.
12. M. Davidovitz and Z. Wu, Semi-Discrete Finite
Element Method Analysis of Arbitrary Microstrip
Elements. Static Solution, IEEE Trans. Microwa e
Theory Tech., Vol. 41, Apr. 1993, pp. 680]686.
13. N. G. Alexopoulos, J. A. Maupin, and P. T. Greiling, Determination of the Electrode Capacitance
Matrix for GaAs FETs, IEEE Trans. Microwa e
Theory Tech., Vol. MTT-28, May 1980, pp. 459]466.
14. P. M. Van den Berg, Iterative Computational
Techniques in Scattering Based Upon the Integrated Square Error Criterion, IEEE Trans. Antennas Propagat., Vol. AP-32, Oct. 1984, pp.
1063]1071.
15. T. K. Sarkar and E. Arvas, On a Class of Finite
Step Iterative Methods Conjugate Directions. for
the Solution of an Operator Equation Arising in
Electromagnetics, IEEE Trans. Antennas Propagat., Vol. AP-33, Oct. 1985, pp. 1058]1066.
16. S. L. Ray and A. F. Peterson, Error and Convergence in Numerical Implementations of the Conjugate Gradient Method, IEEE Trans. Antennas
Propagat., Vol. AP-36, 1986, pp. 1824]1827.

17. T. K. Sarkar, On the Application of the Generalized Biconjugate Gradient Method, J. Electromag.
Wa es Appl., Vol. 1, 1987, pp. 223]242.
18. E. Drake, F. Medina, and M. Horno, An Improved Iterative Technique for the Quasi-TEM
Analysis of Generalized Planar Lines, IEEE Trans.
Microwa e Theory Tech., Vol. 40, 1992, pp. 652]658.
19. T. K. Sarkar and S. M. Rao, An Iterative Method
for Solving Electrostatic Problems, IEEE Trans.
Antennas Propagat., Vol. AP-30, July 1982, pp.
611]616.
20. M. F. Catedra, Solution to Some Electromagnetic
Problems Using Fast Fourier Transform with Conjugate Gradient Method, Electron. Lett., Vol. 22,
Sept. 1986, pp. 1049]1051.
21. C. H. Chan and R. Mittra, Analysis of MMIC
Structures Using an Efficient Iterative Approach,
IEEE Trans. Microwa e Theory Tech., Vol. 36, Jan.
1988, pp. 96]105.
22. V. V. Sanjaynath, N. Balakrishnan, and G. R.
Nagabhushana, Application of Conjugate Gradient Method for Static Problems Involving Conductors of Arbitrary Shape, IEEE Trans. Antennas
Propagat., Vol. 42, July 1994, pp. 1028]1033.
23. E. Durand, Electrostatique, Tome I, Les Distributions. Masson et Cie, Paris, 1964.
24. G. Plaza, F. Mesa, and M. Horno, Computation of
Capacitance Coefficients of Rectangular Thick
Patches on Multilayered Substrates, Int. J. Microwa e Millimeter-Wa e CAE, Vol. 6, June 1996,
pp. 445]453.
25. J. Martel, R. R. Boix, and M. Horno, Static Analysis of Microstrip Discontinuities Using the Excess
Charge Density in the Spectral Domain, IEEE
Trans. Microwa e Theory Tech., Vol. 39, Sept. 1991,
pp. 1623]1631.
26. J. Martel, R. R. Boix, and M. Horno, Static Characterization of Offset Asymmetric Microstrip Gap
Discontinuities, Microwa e Opt. Tech. Lett., Vol.
6, 1993, pp. 277]280.

BIOGRAPHIES
Enrique Drake was born in Montilla,
Cordoba, Spain, in September, 1966. He
received the Licenciado degree in October 1990 and the Doctoral degree in May
1995, both in Physics, from the University
of Seville, Spain. He is currently an Associate Professor in the Department of Applied Physics at the University of Seville.
His research interest focus on the analysis of planar multiconductor structures embedded in anisotropic materials.

Francisco Medina was born in Puerto

Real, Cadiz, Spain, in November 1960.
He received the Licenciado and the Doctoral degrees, both in Physics, from the
University of Seville, Seville, Spain, in
1983 and 1987, respectively. From 1986
to 1987, he spent the academic year
at the Laboratoire de Microondes de
lENSEEIHT, Toulouse, France, on
scholarship from MEC-MRT. From 1985 to 1989, he was an
Assistant Professor in the Department of Electronics and

Capacitance Matrix of Manhattan Conductors

Electromagnetism, University of Seville, and since 1990, he
has been a Profesor Titular Associate Professor. of Electromagnetics. His research deals mainly with analytical and numerical methods for planar structures and circuit applications
of multiconductor lines.
Dr. Medina was a member of the Technical Programme
Committee of the 23rd European Microwave Conference,
Madrid, Spain, in 1993.

Manuel Horno was born in Torre del

Campo, Jaen, Spain. He received the Licenciado and the Doctoral degrees, both
in Physics, from the University of Seville,
Seville, Spain, in 1969 and 1972, respectively. Since 1969, he has been with the
Department of Electronics and Electromagnetism, University of Seville, where
he became an Assistant Professor in 1970,

397

Associate Professor in 1975, and Full Professor in 1986. His

main fields of interest include boundary value problems in
electromagnetic theory, wave propagation through anisotropic
media, and microwave integrated circuits. He is presently
engaged in the analysis of planar transmission lines embedded
in complex materials, multiconductor transmission lines, and
printed antennas.
Dr. Horno is a member of the Electromagnetism Academy,
Massachusetts Institute of Technology MIT., Cambridge, MA.