Sunteți pe pagina 1din 36

The PEEC-Method and its Application by

Calculating Impedances of Complex


Power-Ground-Supply Systems
Dipl.-Phys. Matthias Trscher, Dipl.-Ing Armin Englmaier
SimLab Software GmbH
Krnerstrae 51
81373 Mnchen

The Partial Element Equivalent Circuit (PEEC) -method has proven to be useful
by modeling many different electromagnetic problems.
This paper presents the underlying theory of PEEC and applications of the
method in the field of EMC analysis on Printed Circuit Boards (PCB).
The goal here is to work out special methods by using PEEC to raise the
efficiency of calculations and to come up with the expectations of customers.
The paper will show that in that way it is possible also to treat very complex
PCBs, even in high frequency ranges.

Contents
Topic 1: Motivation using PEEC
Topic 2: Classification of the method
Topic 3: From theory to practice (first results)
Topic 4: Raising of efficiency
Topic 5: Use by handling complex structures
Topic 6: Conclusion

Topic 1 gives some reasons and motivations for using PEEC instead of other
well-known methods.
A detailed description is given in Topic 2 by classifying the method and by showing
relationships to other methods like Finite Differences and Finite Elements-method.
The PEEC-method can be derived by applying the Method of Moments (MoM)
to the Electric Field Integral Equation (EFIE).
Topic 3 leads from theory to first applications of PEEC and explains the great
advantage by using screening effects to raise calculation efficiency.
In Topic 4 preparations are made to make possible the calculation also of
non-academic complex structures on PCBs. This is done by introducing a
so-called Hierarchical Multipole Grid.
Topic 5 presents simulation results of a 400 mm x 200 mm Power-Ground-Supply
system.
Finally in Topic 6 a conclusion is made by working out again the advantages of
PEEC.

Demands on methods for EMC simulation on PCBs


1. Modelling of complex 3D structures
2. Simulation must substitute measurements 1:1
3. Insertion of additional models (e.g. bypass capacitors)
4. Exactness of modelling must be adjustable to save time

What does a customer expect from EMC simulation tools?


He wishes to model three-dimensional structures.
He wishes to work with a simulation tool as he is used with measurement tools.
He wants to take into consideration active and passive components in the
simulation such as capacitors and voltage sources.
And of course he wants to get exact results by at the same time saving running
time.

Description of electromagnetic effects by Maxwell's Equations


rot H = g + jD
rot E = - jB
div B = 0
div D =

Introduction of Potentials to simplify the description


B = rot A
E = - grad U

Actually, all we need are only these four equations: The Maxwell's Equations.
And after introducing the vector potential A and the scalar potential U all
problems have removed?!
But life isn't as easy as it seems.
Although these six equations can describe all effects we are interested in the
problems have just begun.

Possibilities of describing the problem through Potentials

Electromagnetic Problem

U(r) = 0

U(r) =

1
4

I |r (-rr')' | dV'
V

Coulomb Law

In general you have two possibilities to describe an electromagnetic problem.


The first possibility is using differential equations. There are many known
analytic methods to achieve common solutions for differential equations:
- Separation of the solution
- Spectral transformation
- Expansion of the solution in Fourier-series or series of polynoms.
To get a solution, special boundary conditions must be fulfilled. For complex
structures this can only be done by numerical methods and using computers.
The second possibility is using integral equations. They present the solution
in free space if the sources of fields (charges and currents) are known. If there
are additional conductors the charges and currents on these conductors must
be calculated with integral equations regarding boundary conditions. This also
can only be done by numerical methods and using computers.
Differential equations require consideration of the entire space in the calculation.
In intergal equations only the boundaries of the system must be analyzed.

Numerical solution by means of the Method of Moments

Basic idea: Reduction of a functional equation to a matrix equation


Solution of the matrix equation by known techniques

Lf =g

1. Expand f in a series of basis functions


f=

Known:

L,g

n
j
n

fn

2. Define a series of weighting functions


Unknown: f

Iw

n
j
n

L fn dV = Iwm g dV
V

The numerical method to solve differential and integral equations is called the
Method of Weighted Residuals also called Method of Moments. The basic idea
of this method is to reduce the differential or integral equations to an algebraic
matrix equation system and to solve this equation system by known techniques.
A first step to achieve this is to expand the unknown function in a series of
so-called basis functions fn. These functions mostly are very primitive and are
only defined in small discrete areas.
To determine the parameters n of the basis functions the expanded solution
function is inserted into the equation and the function is tested in discrete regions
to get n equations for the n unknown parameters. This testing is done by
integrating the whole equation with so-called weighting functions.
If this method is applied to differential equations the resulting matrices are sparse
and can be solved by iterative matrix solution techniques such as the conjugate
gradient algorithm.
If the method is applied to integral equations the resulting matrices are full and
must be solved with direct matrix solution techniques such as the Gaussian
elimination algorithm.

Method of Moments: Underlying principle for different methods


to solve Differential Equations

FDM
Finite Difference Method

FEM
Finite Elements Method

BEM

No special treatment
Basis: Quadratic polynoms
Weights: Dirac functions

Treatment: 1. Green's theorem


Basis: Linear polynoms
Weights: Linear polynoms

Treatment: 2. Green's theorem

Boundary Elements Method

Basis: Constant polynoms


Weights: Green's functions

Many known numerical methods for the solution of differential equations can be
derived from the Method of Moments.
The Laplace equation for example can be solved with the Finite Differences
Method (FDM), the Finite Elements Method (FEM) or the Boundary Finite
Elements Method (BEM).
In the FDM the second order differential operator is substituted by an appropriate
finite differences formula. The difference scheme can be achieved by applying
quadratic polynoms as basis functions and dirac impulses as weighting functions.
The FEM is a so-called Galerkin-method. That means that basis and weighting
functions are identical. Usually linear polynoms are applied as basis and weighting
functions. To be able to use linear functions in differential equations with second
order differential operators the equation is often transformed by applying Green's
first theorem.
In the BEM the differential equation is transformed to an integral equation. This
can be done by using Green's second theorem and Green's functions as
weighting functions. Often constant polynoms are used as basis functions.
In FDM and FEM the entire space must be discretized. Therefore the resulting
matrix equation system is large but also sparse.
In BEM only the boundaries have to be discretized. The resulting matrix equation
system is smaller but - and this is the great disadvantage - full.

Characteristics of Differential and Integral Equations

Advantages:

Advantages:

- Sparse matrices
- Inhomogenous materials

Disadvantages:
- Volume must be discretized
- Problems with unbounded regions

- Only discretization of boundaries


- No problems with unbounded regions

Disadvantages:
- Full matrices
- No inhomogenous materials

Integral equations are most suitable to handle


electromagnetic problems in EMC domain
Only full matrices are a serious problem

Which is the best method now?


Well, this depends on the problem you have in detail. If you choose differential
equations you get a great advantage because of sparse matrices and the
possibility of analysing inhomogenous materials.
On the other side you must discretize the entire space and must approximate
limits of unbounded regions.
In the area of EMC analysis most problems deal with homogenous materials
and with unbounded regions.
Generally spoken: By analyzing EMC problems we prefer integral equations, but
we must focus our effort in avoiding full matrices.

Direct solution of Integral Equations by the Method of Moments


MoM

Treatment belongs to problem

Method of Moments

Basis: Belongs to problem


Weights: Belongs to problem

The Integral Equations must


be prepared in such a manner
to harmonize with the
electromagnetic problem

Most important integral equation for EMC analysis


Retarded Electric Field Integral Equation (EFIE)

g(r)
E(r) = + jA(r) + U(r)

U(r) = 1
4

A(r) =
4

I
V

I
V

jk|r-r'|

| r - r' | (r') dV'


jk|r-r'|

e
g(r') dV'
| r - r' |

Many electromagnetic problems can be expressed directly in form of integral


equations.
Here the Method of Moments can be applied without any additional
transformations and without certain constraints in using basis and weighting
functions.
An important integral equation that can be treated by MoM is the
Electric Field Integral Equation (EFIE) with scalar potential U and vector
potential A.
The numerators in the integrands describe retardation effects while the
denominators include informations about the distance between cause and effect.
The electric field now is described as a sum consisting of current density g,
vector potential A and scalar potential U.

The retarded Partial Element Equivalent Circuit (rPEEC) Method


Application of the Method of Moments to the retarded EFIE with local
constant functions as base and weighting functions (Galerkin-Method)
Transformation of the free space retarded Green's functions in standard
elements like capacitances, inductances and controlled voltage sources.

N2

N1

Advantage:

N2

N4

N3

N1

N4

N3

Possibility of full dynamic wave simulation

Disadvantage: Full matrices restrict the solution to small problems


Not applicable for industrial designs (like other methods)

If the MoM with constant functions for basis and weights is applied to the
retarded EFIE, the resulting equation system can be interpreted as an electric
network consisting of self-capacitances, self-inductances and controlled
voltage sources.
For that the surfaces of the conductors must be discretized into small areas
assigned to one distinct nodal point of the mesh. These areas have to be
connected by small segments. Now the areas can be modeled by
self-capacitances that are related to the charges in the expression of the
electric potential in the EFIE.
The segments can be modeled by self-inductances that are related to the
currents in the expression of the magnetic vector potential in the EFIE.
The retarded couplings between the areas and the segments are described by
the Green's functions and can be expressed by controlled voltage sources.
The advantage of this method is the transformation of dynamic field problems
into network principles with known solution algorithms.
The disadvantage is the full network description matrix because of the controlled
voltage sources taking into account the retarded couplings.

Retardation and Quasi-Stationarity

Retardation:
-> Field needs time to get from one point to another point in space
-> Considering retardation enables to simulate wave propagation through space.

-> Field needs no time to get from one point to another point in space
-> Time dependent fields can be described by models derived from
static field calculations (e.g. capacitances, inductances)

If retardation effects are taken into account the wave propagation of


electromagnetic fields can be explained. Unfortunately the equations regarding
redardation are difficult to solve.
On the other side, static electric and magnetic fields can be calculated without
time-dependent terms in the Maxwell's equations.
It can be shown that also time-dependent fields can be described by static
solutions if the fields are only observed in an area which is much smaller than
the smallest wave length of any signal propagating through the structure.
In this case the fields can be treated as so-called quasi-stationary fields. The
important condition for quasi-stationarity is given above. is the smallest wave
length of any signal and d means the size of the structure to be analysed.

The Partial Element Equivalent Circuit (PEEC) Method


Application of the Method of Moments to the non-retarded EFIE with local
constant functions as base and weighting functions (Galerkin-Method)
Usage of Free space static Greens's functions to calculate the
inductance and capacitance matrices of the dicretized system

N2

N1

Advantage:

N2

N4

N3

N1

N4

N3

Creation of sparse network matrices by using screening effects on PCBs.

Disadvantage: Prior condition is the quasi-stationarity of coupled elements

With this approximation we obtain the Partial Element Equivalent Circuit-method.


If retardation is neglected the whole system consisting of small areas and small
segments can be expressed in terms of one capacitance matrix and one
inductance matrix.
These matrices are calculated in separate steps.
The capacitance matrix must be determined using a static electric field
calculation.
The inductance matrix must be determined using a static magnetic field
calculation.
The description of electromagnetic effects with capacitance and inductance
matrices is only correct if the quasi-stationary condition between
coupled elements is fulfilled.

Principle of hierachical simplifications


rPEEC
1. Simplification:
Retardation is neglected

PEEC

Static matrix calculation


of R, L, C, G

2. Simplification:
Conductors with boundary
conditions (U = 0) can cause
sparse network matrix

Network

Two simplifications can reduce the complexity of the problem in orders of


magnitude.
The first simplification is to treat the electromagnetic fields on PCBs in the quasistationary case. This allows the modeling of PCBs in terms of frequency
dependent resistance (skin effect), inductance, capacitance and frequency
dependent conductance (dielectric losses). These values must be calculated
exactly once and are valid for all frequencies that fulfill the condition of
quasi-stationarity.
But what does quasi-stationarity mean in practice?
Quasi-stationarity is satisfied if the distance between coupled elements is
remarkable smaller than the smallest wavelength of any signal propagating
through the structure. This condition is simply fulfilled by considering screening
effects on PCBs, the second simplification.
Screening effects make the capacitance and inductance matrices sparse if there
are conductors on the PCB that can be handled as ground conductors (U = 0).

Screening effects
System without boundary condition

System with boundary condition

Lumped Capacitance Matrix [F]

Lumped Capacitance Matrix [F]

S1

S2

S1

S2

S1

4.41e-14

2.02e-15

S1

7.73e-14

3.23e-16

S2

2.02e-15

4.41e-14

S2

3.23e-16

7.73e-14

Here a brief description is given what boundary condition (U = 0) means.


In the figure above you can see two conducting planes above a dielectric with
permittivity 4.5. On the left side no groundplane is contained while on the right
side a ground plane is placed below the dielectric.
What happens if we load the left plane with 1 Volt and the other one with 0 Volt?
The charges on the left conductor influence charges on the right conductor.
Moreover, currents in the left conductor induce currents in the other conductor.
These effects are expressed in terms of non-diagonal elements of capacitance
and inductance matrices.
Without the ground plane (left picture) the influence of one conductor to its
neighbour conductor is about 4.5 % of the self-capacitance.
In presence of the groundplane (U = 0) in the right picture there is a remarkable
less influence on the right conductor. This is due to the additional charge influence
on the ground conductor. The smaller influence is expressed in the smaller
non-diagonal values of the capacitance matrix. Now the coupling capacitance
has decreased to about 0.42 % of the self-capacitance.
This effect we call screening effect.

Requirements for conductors to observe boundary conditions


Distinction between signal and ground conductors
The boundary condition on ground conductors is

U=0
Ground conductors are assumed to be ideal zero
Valid model

Non-valid model

A ground conductor with potential U = 0 everywhere on its surface is a kind of


theoretical assumption. In reality you always have voltage drops on conductors
when currents are flowing through them. This voltage-"noise" is often called
ground bounce.
But when can a conductor handled as an ideal ground conductor? We get an
answer to this question if we ask:
When can a voltage drop be neglected?
This depends on the relation between the voltage drop on a signal conductor and
a so-called potential ground conductor.
If the ground bounce on a conductor is below a defined limit of the normal
signal-voltage behaviour in an electric circuit the conductor can be modeled
as a ground conductor.
In principal the following can be stated:
The more plain and homogenous a conductor is the more likely this conductor
can be modeled as a ground conductor.

Example: Conductor above non-ideal groundplane

15 mm

Loading:

Contact

Source port: I [A], Ri = 1M


Sink port:

To ideal GND

Contact:

R = 50
Source Port
Sink Port

It is time to show some results and applications of the PEEC-method.


This example should demonstrate the ability of PEEC to calculate the return
path of current in a plane at different frequency points. The distance between
the conductor loop and the ground plane below is 0.7 mm.
The loop is loaded with a current source and terminated with a resistance of
50 Ohm to the groundplane.

Result: Current distribution in lower plane


f = 1kHz
f = 100 kHz
f = 1 MHz

These three pictures present the current density distribution in the plane.
At any frequency point the system is endeavoured to produce the lowest
impedance in combination with the signal track or, generally spoken, to save
energy.
At low frequencies (f = 1 kHz) the path with the lowest impedance is the path
with the lowest Ohmic resistance. Therefore it is identical with the shortest
way between the contact and the sink port.
The higher the frequency the greater the inductive effect. Therefore at high
frequencies (f = 1 MHz) the path with the lowest impedance is the path with the
lowest inductance. The current takes the way almost directly below the signal
trace to minimize the current loop.

Example: Two planes in different layers

A = 5 cm x 5 cm
= 40
d = 1 mm
f = 100 MHz - 2.1 GHz
Source Port: I [A],

Another very nice example are two planes in different layers.


The upper plane has an area of 5 cm x 5 cm with a distance to the lower plane
of 1 mm. The system is embedded in a dielectric with permittivity 40.
Note that there is no termination resistance but only a current source on the
left front corner.

Result: Impedance curves

First of all the input impedance was calculated between 100 MHz and 2.1 GHz.
Since there are 12 resonance points in this frequency interval we were interested
in what happens at these resonances.

Current distribution
between
100 MHZ and 454 MHz
(1st Resonance)

To get a first answer we looked on the current density distribution up to the


first resonance point.
As can be seen the current density distribution is blown up like a lullaby by
increasing the frequency.
At 454 MHz the first resonance point is reached.
Remember the size of the structure: 5 cm x 5 cm.
At 454 MHz and a permittivity of 40 we obtain a wave length of exactly 10 cm!
This explains the high impedance in the diagram before.

Resonances between 454 MHz and 2.03 GHz

If we sweep the frequency up to 2.1 GHz we observe other resonances and


the corresponding current density distributions as shown above.
To follow the resonances from low to high frequencies the succession is from
left to right and from top to bottom.

Contrast between current and voltage distribution


1.82 GHz

2.03 GHz

Current distribution

Voltage distribution

Of course it is not only possible to view the current density distribution but also
the voltage distribution on the surfaces of the conductors.
The voltage distribution can have maxima at those points where the current
density distribution is minimum and vice versa.

Example: Two planes in different layers

A = 5 cm x 5 cm
= 40
d = 1 mm
f = 100 MHz - 2.1 GHz
Source Port: I [A],

If we change the loading of the current structure as above we are able to calculate
the transmission behaviour by performing a scatter-parameter analysis.
The S-parameters are calulcated between the two ports P1 and P2.

Result: S-Parameter Analysis

If we analyse the transmission S21 from Port P1 to Port P2 we get maxima at those
points where we already found maxima of impedance.
This could be expected from theory, too.
But what happens at 827 MHz or 1.96 GHz, respectively? There are very deep
minima in transmission!

Current and voltage distribution at f = 1.96 GHz


(No Transmission)

If the look at the graphic we get an answer: No current density and no voltage at
Port P2.
Therefore we obtain zero power in transmission.

Elimination of resonance effects


1. Contact

2. Slits

C = 10 pF

Current distibution

C = 100 pF

Voltage distribution

f = 645 MHz

One very interesting point is, how resonance effects at a distinct frequency point
can be eliminated, e.g. at f = 645 MHz as illustrated in the left bottom graphic?
One possibility is to place a discret capacitance between the two planes. The
resonance effect removes the more the higher the capacitance is. This is shown
in the middle column of the grphic above.
Another possibility is to cut slits into the structure. The resonance effect
immediately breaks down, at least for this frequency. In the right column you can
see the current density and voltage distribution of this case.

Advantages of the PEEC-Method


- Consideration of skin effect and dielectric losses
- Creation of sparse networks by introducing
boundary conditions for conductors
- Combination of field simulations and circuit simulations.
Possibility of inserting additional loadings
- Simulation of wave propagation if distance between
coupled discret elements fulfil the quasi-stationary condition
- Simulation of DC- and low frequency currents

Disadvantage of the PEEC-Method


- Shifting the problem only from dynamic field determination to the
calculation of static capacitances and inductances
- Quasi-stationary condition must be fulfilled

Obtaining these results the PEEC-method can be considered as a very powerful


method
- to consider skin effect and dielectric losses
- to combine field simulation and circuit simulation
- to simulate DC- and low frequency currents.
Although the quasi-stationary condition can be fulfilled very often one serious
problem still remains: The determination of capacitance and inductance matrices
using the MoM.
It was possible only to reduce the complexity of the network matrix to a sparse
matrix by introducing screening effects. But at the same time the problem of full
matrices was shifted from dynamic field calculation to the calculation of static
capacitances and inductances.

Problem: Handling complex strucutres

Uniform and gridded areas


6 layers, = 0.018 - 0.035 mm
A = 400 mm x 200 mm
= 3.9, d = 0.13 - 0.23 mm
120.000 discrete elements

One day a customer aked us to simulate the behaviour of the structure above
under different conditions and to calculate impedances.
The structure consists of uniform and gridded areas, it includes 6 conducting
layers with thicknesses between 0.018 mm and 0.035 mm and has an expansion
of 400 mm x 200 mm.
Shortly spoken, about 120.000 discrete elements would be necessary to simulate
this structure.

EMC analysis on PCBs

These pictures show details of the discretization of the structure.


The total view on the left side gives an overview of the 6-layer board. The green
triangular elements represent the dielectric boundaries and the blue and red
elements deal with capacitive and inductive elements.
In the middle of the board you can see a so-called gridded area. The right pictures
show a zoom view of the grid.
In this example the user wants to "measure" the impedance of the power-ground
system on several points in a wide frequency range. Furthermore, the user wants
to load the structure with different devices and wishes to observe the effects
during the "measurement".
The user is not interested in theoretical aspects and reasons why the discrete
elements must not exceed a defined maximum size. All he wants to know is:
How long does it take to run the simulation?

Introduction of a hierachical multipole grid with screening rules


rPEEC
1. Simplification:
Retardation is neglected

PEEC
2. Simplification:
Separation of the system into small
areas and calculation one by one

Static matrix calculation


of R, L, C, G
3. Simplification:

Network

Conductors with boundary


conditions (U = 0) can cause
sparse network matrix

What we are looking for is a possibility to reduce the complexity of the calculation
in an earlier stage, namely before the calculation of the capacitance and
inductance matrices.
Why not introducing the knowledge about screening effects also by calculating
the L- and C-matrices?
If there are ground conductors it is very useful to subdivide the whole PCB into
smaller regions using a so-called hierarchical multipole grid. Then the
capacitances and inductances can be calculated separately in these smaller
regions. So the problem of solving one large equation system can be reduced to
the problem of solving many small equation systems.

Principle of the hierachical multipole grid with screening rules

SQUARE 1

SQUARE i
SQUARE 2
Ri

SQUARE 3

R3

SQUARE j

Rj

Grid consisting of
SQUARES with
Search Radius R

No sooner said than done! A hierarchical grid is defined over the whole structure.
Then the distance d is determined for each square to the next conductor with
boundary condition U = 0.
This distance multiplied with an adjustable constant factor determines the search
radius r that gives information which neighboured squares must be considered
by the calculation.
Finally the self- and mutual capacitances and inductances of all partial elements
are calculated that are located inside the origin square and neighboured squares.
With the help of this algorithm the effort of calculation for the capacitance and
inductance matrices can be reduced by orders of magnitude. The procedure
is valid if there are ground conductors on the PCB.

Schematic description of the hierarchical multipole grid

In this schematic you can see the treatment of the problem.


With the help of the grid we zoom into an interesting region and calculate only
the matrix coefficients of the elements located in this square.
This, of course, requires the consideration of neighboured squares, too.

Result: Impedance curve

Now it is possible to calculate even very complex structures.


Here you can see the result of an impedance calulation in the frequency range
from 200 MHz to 1 GHz.

Current distribution without bypass capacitors

In this picture the current density distribution is shown with special loadings.
Please compare it with the current density distribution on the next page.

Current distribution with bypass capacitors

Here you can see the current density distribution with the same loadings but
additional bypass capacitors.
The current density and the current path have changed totally.

Conclusion
The PEEC-method is a network interpretation of the EFIE and can be
derived by a Galerkin - MoM - scheme
Main Advantage:
- Only boundaries have to be discretised
- Circuit-models of loadings can easily be integrated
Main Disadvantage:
- Network matrix can be full
But:
If known 0 V boundary conditions can be introduced, the
network matrix becomes sparse.
So the PEEC-Method can be used as a powerful tool for the
simulation of effects on conducting structures of PCB's..

S-ar putea să vă placă și