Professional Documents
Culture Documents
net/publication/3125935
CITATIONS READS
902 175
1 author:
Albert E Ruehli
Missouri University of Science and Technology
275 PUBLICATIONS 10,591 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Albert E Ruehli on 01 June 2020.
Multiconductor Systems
~lxtract—Multiconductor or multiwire arrangements find many here and elsewhere [13] since they encompass a wide
applications in electronic systems. Examples are interconneciioris frequency spectrum.
between digital circuits or integrated microwave circuits. Equivident
The formulation presented here affects three seemingly
circuit models are derived here from an integral equation to establish
an electrical description of the physical georheiry. The models, which
different subject areas. First, it provides a more compre-
are appropriately called partial elefient equivalent circuits (PEEC), hensive interpretation of the relations between circuit
are general iri thdt they iriclude losses. Models of different complexity theory and field theory [14]. Second, it maybe considered
can be crmstructed, to suit the application at htid. as an extension of an integral equation solution for induct-
ance computations [15] in that it includes capacitance.
I. INTRODUCTION Third, it relates to multiwire antennas [16], which corre-
sponds to the formulation below if retardation tirhes are
ULTICONDUCTOR or rnultiwire arrangements included.
M find many applications in electronic systems, e.g., A basic integral equation approach is discussed in Sec-
the intercommunications between digital logic circuits tion II, while an interpretation in terms of inductances is
[1] or microwave integrated circuits [2]. Several methods considered in Secticm III. The portion that relates to
are available today for the analysis of a set of long parallel capacitance is the subject of Section IV, and in Section V
wires close to a common conducting plane [3]–[5]. Two- the relation to retardation times is discussed. In Section
dimensitmal approximations are, however, not applicable VI circuit interpretations are considered, while in Section
to an important class of three-dimensional conductor VII an example is given and further approximations are
arrangements. These geometries consist of conductors discussed.
that may be highly intercoupled because of the tibsence
or remoteness of a comon ground plane. Examples of II. INTEGRAL EQUATION FORMULATION
three-dimensional geometries are integrated circuit pack- The unknowns in a multiconductor or multiwire system
ages and wires or conductors located on dielectric layers are the charges on the surfaces and the. current densities
with remote ground planes. These geometries may, be within the conductors. An integral equation solution is
viewed as i ‘consisting of discontinuities” from a trans- appropriate for problems with large free-space regions
mission line point of view afid are not easily amenable to since a differential equation formulation would require
analysis. However, additiorial electrical design flexibiM y that all regions including the free space be described by
is obtained above two-dimensional striplines for many riodes.
applications due to the multitude of relative conducto~ The approximate integral equation solution pursued
locations possible. here is based on the smnmation of all sources of electric
The ptirpose of this paper is to present a theory for an
fields within a conductor [14], or~lu = ~, – d~/Ot – V+.
approximate computer modeling approrich for, three- Here, ~ is the current density in the conductor of con-
dimensional geometries that is appropriately called partial ductivity u, and ~ and @ are the vector and scalar poten-
element equivalent circuit (PEEC) analysis. The PEEC tials, respectively.
method is based on an integral equation description of the
geometry that is interpreted in terms of circuit elements. J(F,t) ad (?,t)
Eo(?,t) = —+7 + v@(F,t) (1)
The circuit elements, tiiz., the partial inductances and i?
partial capacitances, can be found from computer solu-
where P is the vector from the origin. The vector potential
tions, as has been shown previously [G]–[9]. A gene:al-
~ for K conductors is
purpose network analysis pr~gram [lCi3-~12] is then used
to obtain voltages and currents in both the time domain ‘P
~ (F,t) = ~~, ~ / l?(~,?’)j(r’,t’) dv’ (2)
and the frequent y domain. Time domain results are gi+en .k
where the retarded time is given by
Manuscript received June 25, 1973; revised September 28, 1973.
The author is with the Circuit and Computer Design Automation
t,=t_l~–~’l
Group, IBM Thomas J. Watson Research Center, Yorktown (%.%) ‘l’ (3)
Heights, N. Y. 10598. c
RUEHLI: MODELS E’OR THREE-DIMENSIONAL MULTICONDUCTOR SYSTEMS 217
for uniform regions with properties e,, W, where c repre- quantity. Hence, the expansion function for the charge
sentsthe speed of light. Tbefunction l?is defined by density is
Mh
R(T7”)Q ,~ : # ,“ (4)
qk(tn) = x p?nkqmk (k) (lo)
m=l
(8)
t’ t’
D
~1 I
II
[1 CAPACITIVE
PARTITIONS
II I
NM
Ii ~1
for y = x,y,.z.
!_t II
Furthermore, the charge density is written as another
Iln
expansion function of the same type. Since the free charge x
is restricted to the outside surfaces of all conductors,
Fig. 2. Surface cells for capacitive partitions for rectangular
q(F) is a surface layer charge density rather than a volume conductors.
218 IEEE TRANSACTIONS ON MICROWAVE THEORY AND TEcHNIQUES, MARCH 1974
K Mb
1
v. = x ~ qmk(tmk) — 2? (7z+,?’) ds’
k=l m=l [ 47K \ /gmk
The vector associated with the positive end of the lth cell
is designated with a + sign while the negative lead is
indicated with a — sign. The retarded time refers to the
where y = x,y,z as previously, where the charge density is center of the volume cell, in the same way as for the partial
included as’ a surface quantity in (11). Next, (11) is inductances.. The kernel for the integrals in (15) is re-
divided by the cell cross section al perpendicular to the written as a Green’s function G = K/47rc. Then (15) is
direction of current flow. The local approximations used K Mk r,
here to find network elements &e the same as are used in
finite difference techniques. The first term on the left-
hand side of (11) is then interpreted as the resistive
voltage
written
drop along the cell v, = RTZIYJ and (11) is re-
?s v, + VL + V. = 0.
—
J G(?z-,?’)
S71A
ds’
1
. (16)
where in general
(12) 1
ppij = – G (7,,7’) ds’. (17b)
aj / Si
If the kernel (4) is inserted into (12), the term in
brackets is recognized to be in the form of partial in- Equation (17) is clearly in the form of partial coefficients
ductances [6], with the sums representing the coupling of potential (7), which is essential] y the same as the
among them: coefficients of potential except that partial surfaces of a
K NW conductor are involved. Equation (17a) represents the
VL = ~ x LPLYn~ $ [Irk(k) 1. (13) voltage differences Vij = @i — @j across the elements. TO
/+=1 n=l further extend the definition, retarded partial coefficients
Again, the inductances of straight conductor segments of potential are defined that include the specification of
corresponding to the cells are called partial inductances the retardation time. Since the charges reside on the con-
Lpi, to avoid confusion with the loop inductances Lij. ductor surfaces, the potentials are only due to nodes
The purrents I,~k are for most practical systems instan- external to the conductors. Therefore, generalized partial
taneous in that k = t in (8), due to physical smallness. coefficients of potential can be written in matrix form as
For widely spaced segments in uniform regions, retarded *P’
[1[1
PP’
partial inductances can be defined since the retardation ---- = (18)
---- Q,
depends on the physical distance. *pi
PPi
IV. CAPACITIVE TERM where @ and ppi correspond to the internal nodes, while
It is next shown that the last term in (11) corresponds @J.’,pps, and Q. refer to the surface cells.
to capacitive coupling. The inside integral is defined as
V. RETARDATION TIME
F(7) A f,., l?q cts’ over the local surface cells defined in
(10). Then the integral in the ~ coordinate is approxi- The formulation given above is general in that it in-
mated as cludes retardation times for widely spaced objects in a
uniform dielectric region. As indicated in Sections III and
VI, retarded partial inductances and retarded coefficients
of potential can be defined for a network interpretation of
the formulation. A basic difference between the wire
(14)
antenna problem [16] and the interconnection systems
This sh~ws tha$ the capacitive cells in Fig. 2 are shifted considered here is that retardation times are large for the
from the cells in Fig. 1 by half the size of a cell. If the former, while for the latter they may be ignorably small.
above results are applied to (11) and use is made of (10), Furthermore, the coupling of minute signals from’ the
a circuit-type equation results, or antenna to the receiver is of interest. In contrast, for
RUEHLI: MODELS FOR THREE-DIMENSIONAL MULTICONDUCTOR SYSTEMS 219
circuit-type geometries only relatively strong intercoupling inductance, and network analysis. General formulations
is of interest. This immediately suggests that sparsity can exist for all these tasks such that responses for a wide
be introduced into the interactions by ignoring small range of geometries can be computed.
coupling terms. For most situations, the small coupling
elements are also the ones having larger retardation time. VII. APPROXIMAIrIONS AND RESULTS
–vL+
fi H >INDUCTIVE PART,T,ON
& . — ——
I
i “
LP14 14
‘JII 8
( )
: :R14 : R15
LP12 12
1b
s %2 CP6 12 ‘P6 II
I 91 u — + ;
61 mm
1b n
%99
%9
LPIO 10 \
V56 ( ) \
—cp@
CAPACITIVE
: F9 RIO \ PARTITIONS
L P77
! /’
~’ q
5
R7
6 /
—
+,
\
I
Fig. 3. Equivalent circuit for corner of a conductor. Fig. 4. Conductors on insulating substrate.
220 3EEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, MARCH 1974
. v~ TABLE I
LIST OFNETWORKELEMENTS
R~ = 50 L = 21 c = .61
p55 p12
~1 = 300 L = 59 c = .525
p66 p13
RS
%2 .50 = 20 c = .225
JJp77 p14
=8 = 14.7 c = .725
CL1 Lp13 p15
= 27 L = -10.6 c = .285
c p23
L2 P16
= 57.2 L =2 c = .275
L p27 p25
LP66 pll
L = 11.6 L = -1 c – .225
p44 p57 p45 –
Lpm
R in ohms; L m nSI; C in PF
—---+&------
resistance can usual] y be found from simple calculations.
Changes in the geometry are easily accommodated since
Fig. 5. Partial element equivalent circuit. Partial capacitances
labeled according to node numbers. general computer solutions are involved in the task.
It is noted that the PEEC models are rather inefficient
for two-dimensional geometries, especially if a high degree
0.2
\v= of resolution is required. For three-dimensional geometries,
f 7
very good results are obtained with a few elements only,
as is exemplified in Section VII.
01 -
—VL MEASUREO, CL2 = 27PF
/ ACKNOWLEDGMENT
––– VL CALCULATE, CI_2 = 27PF
/’
—.–VL MEASURED, CL2 = o The author wishes to thank M. Handelsman of the
I
—-—vL CALCULATED, CL2 = o University of Vermont for many helpful discussions on the
0
5 10 15 20 subject.
TIME IN ns
and T. R. Scott, “Networka nalysisu singasparset ableauwith [14] E. C. Jordan and K. G. Balmain, Electro-Magrzetic Waves and
tree selection to increase sparseness,” in Proc. IEEE .7nt. Radiating Systems. Englewood Cliffs, N. J.: Prentice-Hall,
Syrnp. Circuit Theory (Toronto, Ont., Canada, Apr. 1973), 1968, ch. 14.
pp. 165-168. [15] A.. Gopinath and P. Silvester, “Calculation of inductance of
[12] IBM Advanced Mati.stical Analysis Program, ASTAP manual fimte-length strips and its variation with frequency,” IEEE
GH20-1271-O. Trans. Microwave Theory Tech., vol. MTT-21, pp. 380-386,
[13] A. E. Ruehlij ``Electrical analysis ofinterconnections inasolid- June 1973.
state circuit environment, ” in Dig. IEEE Int. Solid-State [16] R. F. Harringtoqj Field Computation b~ Moment Me,thods.
Circuit Corzf. (New York, 1972), pp.6+65andp.216. New York: Macmdlan, 1968, ch. 1 and 4.
in F ie Id Problems
W. KINSNER, MEMBER, IEEE, AND EDWARD DELLA TORRE, SENIOR MEMBER, IEEE
Abstract—An iterative approach to the finite-element method is vector (s) associated with the particular eigenvalue (s).
presented. Several finite-element formulations are presented for the
It is noted that for the H modes A k only semidefinite.
Laplace, Poisson, and Hehnholtz equations. These formulations
The finite-difference formulation of these problems also
permit iterative solutions. The convergence of the vector sequences
generated by the iterative method is accelerated using successive leads to (1), however, the matrix B is the identity matrix,
extrapolation and other methods. Accuracy and convergence of the and A k not necessarily symmetric.
solutions are discussed. A variety of methods for solving (1) have been pre-
sented (e.g., [9], [10], and [12]). The finite-difference
I. INTRODUCTION
solutions most frequently employ iterative techniques
HE THEORETICAL background of the finite- and the finite-element solutions almost exclusively use
T element method has been given by Aubin [1]. dwect methods.
Other authors [2], [3] have introduced some practical Iterative methods for solving eigenvalue problems may
aspects of the method as applied to structural mechanics. be divided into two categories. The first methods use the
Sivester [4] and others [5], [16], [20]-[22] discussed fact that eigenvectors of a system form a linearly inde-
the method as applied to the electromagnetic field prob- pendent set which spans an n dimensional space. The
lems. Convergence of the method, as a function of the methods in the second category use the property that the
relative size of the discretizing elements and the order of generalized Rayleigh quotient
the approximating polynomials, is discussed in many
XTA X
recent mathematical and technical journals [6]. In Q=— (q
XTBX
particular, explicit discretization errors are given in [1],
[2], [6], and [7], and some experimental results are given is equal to an eigenvalue and is stationary when x is the
in [5] and [8]. corresponding eigenvector. A method using this property
The variational formulation of waveguide problems, with Fletcher–Powell iteration has been described [13].
using complete polynomials, leads to the general eigen- These methods combined with the deflation or ortho-
value problem gonalization yield, however, only partial eigensolutions,
AX = xl?% (1) i.e., the dominant and several closest eigenvectors, An
iterative method for the complete eigensolution shall be
where A and B are symmetric positive-definite n x n
presented.
band matrices, A is the eigenvalue (s) and x the eigen-
II. FINITE-ELEMENT FORMULATION
Manuscript received June 26, 1973; revised November 7, 1973. Let R be either a simply or multiply-connected bounded
This work was supported by the National Research Council of
Canada. region in an n dimensional space V“ with boundar,y r.
The authors are with the Group on Simulation! Optimization, The boundary 1? consists of a finite number of closed,
and Control, Department of Electrical Engineering, McMaster
University, Hamilton, Ont., Canada. nonintersecting hypersurfaces rl, (h = O, ..”, ~) such