• Embed Doc
  • Readcast
  • Collections
  • CommentGo Back
Download
 
A Novel Deembedding Technique Applied to theS–Parameter Computation of Multichip-Modules via TLM
Wolfgang Dressel, Tobias Mangold, Larissa Vietzorreck and Peter RusserInstitut f ¨ur Hochfrequenztechnik, Technische Universit¨at M¨unchenArcisstr. 21, D-80333 Munich, Germany, email: dressel@hft.ei.tum.de
 Abstract 
—TheTLM-method isusedforcomputingthescattering parameters of multichip-modules via time do-main simulation. A novel deembedding method is ap-pliedwhichallowstoreducetheerrorintroducedbynon-ideal port termination.
 Index Terms
—MCM, numerical modeling, TLM
I. I
NTRODUCTION
In this contribution the TLM method is usedfor the numerical characterization of 
m
ulti
c
hip
m
odule (MCM) structures. In MCMs several ICchips are mounted in a silicon or ceramic sub-strate. As a group of highly functional electronicdevices MCMs reduce cost and size providing op-portunities to integrate unique functions of chipsfromvariousprocessingtechnologiesintoonesys-tem [1], [2], [3]. MCMs exhibit general three-dimensional geometry. For the modeling of MCMstructurestherefore afield solverisrequired whichimposes no restrictions on the geometry of theelectromagnetic structures. Considering this re-quirementtheTLMmethodiswellsuitedformod-elingofMCMstructuresincludinglossesinthedi-electrics as well as in theconductors. In modellingan MCM structure the task usually is to com-pute its multiport parameters in a wide frequencyrange. The TLM method already has been appliedto multichip modules and to extract lumped ele-ment equivalent circuits of MCM structures [4].The frequency characteristics is obtained from thetime domain simulation of the MCM multiportstructures by Fourier or Laplace transform [4] orby parameter estimation methods [5]. Full–wavesimulation of the multiport characteristics may bedone under open circuit conditions, short circuitconditions and matched conditions. Computa-tion under matched conditions directly yields thescattering parameters afterFourier transformation.The main advantage of simulation under matchedconditions is that the decay of the scattered pulsesis considerably faster than in the case of open–circuit or short–circuit termination. This may re-duce the computation time considerably. How-ever, this method suffers from the difficulty of acorrect modeling of the termination. In this work we therefore propose a new deembedding methodwhich allows direct time domain computation of the scattering parameters without exact termina-tion of the multiport under consideration.II. TLM S
IMULATION AND
D
EEMBEDDING
The S-parameters are calculated from the com-puted generalized port currents or voltages respec-tively. We obtain for example
  ¢¡£¡
from the inci-dent generalized voltage wave amplitude
¤¦¥¨§©
andthe resulting generalized input port voltage
¤¡
via
  ¡£¡! ¤¡¢"¤#¥$§© ¤#¥$§©
(1)The application of this formula, however is nota trivial task. The generalized voltages are spec-ified via a path integral or an area integral overthe transverse electric field component. Since themodes are non–TEM and furthermore the trans-verse field distribution also is frequency depen-dent, an exact modelling of a matched waveguidetermination will not be feasible.In the following we therefore propose a deem-bedding procedure which can be applied in caseof modelling the scattering behaviour of the mul-tiport without matching the ports. Let us considera multiport with the scattering matrix
%
. The vec-tor of the incident waves
&
and the vector of the
 
scattered waves
  
are related via
  
%&
(2)Ifwewantto characterize amultiportwith
¡
portswe have to perform
¡
simulations and to computethe scattering from the input ports to the outputports. Usually we will excite only one port at eachsimulation, and each time we will excite anotherport. However, due to non-perfect termination of all ports each time all ports will be excited by in-cident waves. Let
&¡
,
&
£¢
, ...
&
¥¤
be the vectorsdescribing the excitation in the first, second and
¡§¦©¨
simulation, and let
  
¡
,
  ¢
, ...
  ¤
denominatethe vectors summarizing the scattered waves. Wethen may introduce the
¡¡
matrices
&¡
&
£¢
&
£¤"!
(3)
#
  
¡
$  ¢   ¤"!
(4)A number of 
¡
simulations is carried out, wherein the
%
–th simulation,(i = 1 ...N)a primarypulsewave
&('0)
is applied to the
%
–th port. The portsof the multiport are terminated with the reflectionfactors
1
¡
12¢ 1(¤
. The waves flowing into themultiport are given by
436587 #
(5)with
7
@9%0&BA1
¡
C12¢ 1(¤D!
. If 
FE
¡
exists, thescattering matrix of the multiport can be obtainedfrom
%
#E
¡
(6)For this evaluation both,
and
#
have to be eval-uated. This procedure yields a deembedding of non–perfectly terminated multiports.III. E
XAMPLE
In this section, S-Parameters calculated by theconventional method according to (1) and thedembedding method after (6) are shown for a in-terdigital capacitor mapped in fig. 1. The inter-digital capacitor has got a size of 880
G
m
250
G
m and is manufactured by a thin film method ona ceramic substrate. The computed scattering pa-rameterresultshavebeen comparedwithmeasure-ments made by a HP-8510C network analyzer anda Cascade on-wafer prober.
500/100 nodes190/19 nodesplane of the calculated currents and voltages
12
measurement plane
190/19 nodes
measurement plane
        1        0
220/22 nodes
        1        0
85/17 nodes
Fig. 1. Interdigital capacitor with dimensions (in
H
m) andnumber of nodes
The simulation was performed using 5
G
m assmallestmesh sizeand 10
G
mas biggest. Thesim-ulation box has got a size of 880
G
m
305
G
m
335
G
m, whereas 240
G
m from the 335
G
m rep-resent the free space above the structure. A mag-netic wall can be used to half the simulation box.So computation resources and time can be saved.The path integrals for the voltages and currents arein the same cross-section defined as the measure-ment plane at every port of the interdigital capaci-tor. With this cell sizes the computation time took 5 hours on a HP-C360 workstation with 500MBmemory. The time includes the two device sim-ulations needed for all wave amplitudes, and thereference simulation needed for the characteristicimpedance of the microstrip-lines.Fig. 4 till Fig. 2 show the S-parameters
¢¡£¡
,
  
¢
¡
and
  
¢I¢
derived by the conventional method ac-cording to (1), the dembedding method after (6)and the measurement.In order to proof the deembedding technique,one simulation was computed with non-ideal
Fig. 2. Simulated and measured
PRQSQ
of the interdigital ca-pacitor
of 00

Leave a Comment

You must be to leave a comment.
Submit
Characters: ...
You must be to leave a comment.
Submit
Characters: ...