Techniques for Including Dielectrics when Extracting Passive Low-Order Models of High Speed Interconnect

Similar documents
Combined Electric and Magnetic Dipoles for Mesoband Radiation, Part 2

Modes are solutions, of Maxwell s equation applied to a specific device.

A Fast Block Structure Preserving Model Order Reduction for Inverse Inductance Circuits

Time Domain Method of Moments

n n=1 (air) n 1 sin 2 r =

Panel Session on Data for Modeling System Transients Insulated Cables

Advances in Radio Science

Part G-4: Sample Exams

Directional Coupler. 4-port Network

Evaluation of effect of blade internal modes on sensitivity of Advanced LIGO

DIGITAL DISTANCE RELAYING SCHEME FOR PARALLEL TRANSMISSION LINES DURING INTER-CIRCUIT FAULTS

1 sin 2 r = 1 n 2 sin 2 i

Control Theory association of mathematics and engineering

Hankel Optimal Model Order Reduction 1

HIGH speed interconnects both on-board and on-chip that

Ayan Kumar Bandyopadhyay

Developing Excel Macros for Solving Heat Diffusion Problems

Advanced Computational Fluid Dynamics AA215A Lecture 4

The simulation analysis of the bridge rectifier continuous operation in AC circuit

Inhomogeneous structure: Due to the fields within two guided-wave media, the microstrip does not support a pure TEM wave.

Measuring & Inducing Neural Activity Using Extracellular Fields I: Inverse systems approach

+Ze. n = N/V = 6.02 x x (Z Z c ) m /A, (1.1) Avogadro s number

Wave Propagation through Random Media

A Spatiotemporal Approach to Passive Sound Source Localization

A model for measurement of the states in a coupled-dot qubit

Experiment 3: Basic Electronic Circuits II (tbc 1/7/2007)

ELECTROMAGNETIC NORMAL MODES AND DISPERSION FORCES.

Models for the simulation of electronic circuits with hysteretic inductors

MultiPhysics Analysis of Trapped Field in Multi-Layer YBCO Plates

Wavetech, LLC. Ultrafast Pulses and GVD. John O Hara Created: Dec. 6, 2013

Green s function for the wave equation

Aharonov-Bohm effect. Dan Solomon.

Growing Evanescent Envelopes and Anomalous Tunneling in Cascaded Sets of Frequency-Selective Surfaces in Their Stop Bands

Automatic Generation of Geometrically Parameterized Reduced Order Models for Integrated Spiral RF-Inductors

Dynamics of the Electromagnetic Fields

Where as discussed previously we interpret solutions to this partial differential equation in the weak sense: b

Class XII - Physics Electromagnetic Waves Chapter-wise Problems

Nonreversibility of Multiple Unicast Networks

Lightning electromagnetic environment in the presence of a tall grounded strike object

Scalable Positivity Preserving Model Reduction Using Linear Energy Functions

Four-dimensional equation of motion for viscous compressible substance with regard to the acceleration field, pressure field and dissipation field

SURFACE WAVES OF NON-RAYLEIGH TYPE

The homopolar generator: an analytical example

ENERGY AND MOMENTUM IN ELECTROMAGNETIC WAVES

MOLECULAR ORBITAL THEORY- PART I

A Theorem of Mass Being Derived From Electrical Standing Waves (Adapted for a test by Jerry E. Bayles)

A Combination of BIM and BEM for efficiently analyzing optical. elements

General Closed-form Analytical Expressions of Air-gap Inductances for Surfacemounted Permanent Magnet and Induction Machines

Frequency Domain Analysis of Concrete Gravity Dam-Reservoir Systems by Wavenumber Approach

Contact Block Reduction Method for Ballistic Quantum Transport with Semi-empirical sp3d5s* Tight Binding band models

A NETWORK SIMPLEX ALGORITHM FOR THE MINIMUM COST-BENEFIT NETWORK FLOW PROBLEM

Beams on Elastic Foundation

ELECTROMAGNETIC WAVES

A hybrid-mode characterization of microstrip transmission lines

Analysis of discretization in the direct simulation Monte Carlo

UPPER-TRUNCATED POWER LAW DISTRIBUTIONS

Active Magnetic Bearings for Frictionless Rotating Machineries

EE 321 Project Spring 2018

INFLUENCE OF OPERATING AND CONSTRUCTION PARAMETERS ON THE BEHAVIOR OF HYDRAULIC CYLINDER SUBJECTED TO JERKY MOTION

Millennium Relativity Acceleration Composition. The Relativistic Relationship between Acceleration and Uniform Motion

Waveguide Introduction & Analysis Setup

Acoustic Waves in a Duct

SEISMIC ANALYSIS OF SPHERICAL TANKS INCLUDING FLUID-STRUCTURE-SOIL INTERACTION

Stabilization of the Precision Positioning Stage Working in the Vacuum Environment by Using the Disturbance Observer

Bäcklund Transformations: Some Old and New Perspectives

Complexity of Regularization RBF Networks

An Integrated Architecture of Adaptive Neural Network Control for Dynamic Systems

Simplified Modeling, Analysis and Simulation of Permanent Magnet Brushless Direct Current Motors for Sensorless Operation

Long time stability of regularized PML wave equations

20 Doppler shift and Doppler radars

Relativity in Classical Physics

Robust Flight Control Design for a Turn Coordination System with Parameter Uncertainties

Geometry of Transformations of Random Variables

An Adaptive Optimization Approach to Active Cancellation of Repeated Transient Vibration Disturbances

Development of an efficient finite element model for the dynamic analysis of the train-bridge interaction

Chapter 9. The excitation process

Non-Markovian study of the relativistic magnetic-dipole spontaneous emission process of hydrogen-like atoms

On refinement of certain laws of classical electrodynamics

A study of microstrip transmission line parameters utilizing image theory

IN-PLANE VIBRATIONS OF CURVED BEAMS WITH VARIABLE CROSS-SECTIONS CARRYING ADDITIONAL MASS

The universal model of error of active power measuring channel

Speed Regulation of a Small BLDC Motor using Genetic-Based Proportional Control

ELECTROMAGNETIC WAVES WITH NONLINEAR DISPERSION LAW. P. М. Меdnis

STUDY OF INHERENT FREQUENCY OF HELMHOLTZ RESONATOR

9 Geophysics and Radio-Astronomy: VLBI VeryLongBaseInterferometry

7 Max-Flow Problems. Business Computing and Operations Research 608

Subject: Introduction to Component Matching and Off-Design Operation % % ( (1) R T % (

Effect of Rotation, Magnetic Field and Initial Stresses on Propagation of Plane Waves in Transversely Isotropic Dissipative Half Space

Accelerator Physics Particle Acceleration. G. A. Krafft Old Dominion University Jefferson Lab Lecture 4

Automatic Generation of Geometrically Parameterized Reduced Order Models for Integrated Spiral RF-Inductors

SIMULATION OF A NONLINEAR GAAS MESFET MODEL FOR USE IN THE DESIGN OF NONLINEAR MICROWAVE CIRCUITS

Application of negative group delay active circuits to reduce the 50% propagation Delay of RC-line model

Duct Acoustics. Chap.4 Duct Acoustics. Plane wave

NUMERICAL SIMULATION OF ATOMIZATION WITH ADAPTIVE JET REFINEMENT

Effect of conductor thickness on the mode capacitances of shielded strip transmission lines

Likelihood-confidence intervals for quantiles in Extreme Value Distributions

Aircraft CAS Design with Input Saturation Using Dynamic Model Inversion

10.2 The Occurrence of Critical Flow; Controls

A NEW FLEXIBLE BODY DYNAMIC FORMULATION FOR BEAM STRUCTURES UNDERGOING LARGE OVERALL MOTION IIE THREE-DIMENSIONAL CASE. W. J.

Natural Convection Experiment Measurements from a Vertical Surface

Transcription:

Tehniques for Inluding Dieletris when Extrating Passive Low-Order Models of High Speed Interonnet Lua Daniel University of California, Berkeley dlua@ees.berkeley.edu Alberto Sangiovanni-Vinentelli Univ. of California, Berkeley alberto@ees.berkeley.edu Jaob White Massahusetts Institute of Tehnology white@mit.edu ABSTRACT Interonnet strutures inluding dieletris an be modeled by an integral equation method using volume urrents and surfae harges for the ondutors, and volume polarization urrents and surfae harges for the dieletris. In this paper we desribe a mesh analysis approah for omputing the disretized urrents in both the ondutors and the dieletris. We then show that this fully meshbased formulation an be ast into a form using provably positive semidefinite matries, making for easy appliation of Krylovsubspae based model-redution shemes to generate aurate guaranteed passive redued-order models. Several printed iruit board examples are given to demonstrate the effetiveness of the strategy. 1. INTRODUCTION Dieletri materials are present in almost all modern eletroni iruits: from Printed Ciruit Boards PCBs), to pakages, Multi- Chips Modules MCMs), and Integrated Ciruits. Dieletris an signifiantly affet both the performane and the funtionality of eletroni iruits. For instane, they an hange interonnet delays, as well as the positions of frequeny response resonanes. Ignoring dieletris an therefore potentially lead to very inaurate results both in timing analysis tools and in signal integrity tools. Integral equation methods have proved to be very effetive tools for analyzing on-hip and off-hip interonnet strutures, and there are several approahes for inluding dieletri interfaes in integral formulations. For problems whih an be viewed as flat interfaes of infinite extent, suh as multilayer printed iruit boards, the dieletri interfae onditions an be satisfied by an appropriate hoie of Green s funtion [1, 2, 3, 4, 5]. For general shape or finite-size dieletri bodies, it is possible to replae the dieletris with equivalent fititious eletri and/or magneti surfae urrents [6, 7]. General dieletri shapes an also be handled by a volume integral equation VIE) approah, in whih ase the polarization urrents are introdued in the volume of the dieletris, and This work was supported by the MARCO Interonnet Fous Center, and by the Semiondutor Researh Corporation. The authors would also like to thank Matt Kamon for useful feedbak and disussion. Permission to make digital or hard opies of all or part of this work for personal or lassroom use is granted without fee provided that opies are not made or distributed for profit or ommerial advantage and that opies bear this notie and the full itation on the first page. To opy otherwise, to republish, to post on servers or to redistribute to lists, requires prior speifi permission and/or a fee. ICCAD 2001, November 4-8, 2001, San Jose, California, USA. Copyright 2001 ACM...$5.00. harges are introdued on their surfaes [8, 9]. As the last deade has made lear, detailed eletromagneti analysis is a vastly more effetive tool if it an be used to automatially generate aurate iruit-level models of the interonnet. In this paper we show that areful disretization of the VIE method leads to, at least in the low frequeny regime, a linear dynamial system with positive semi-definite matries. This positive-definite result is important beause it makes possible the straight-forward appliation of the Krylov-subspae based guaranteed passive model-order redution MOR) tehniques [10, 11, 12, 13, 14, 15, 16, 17]. The paper is organized as follows: in Setion 2 we summarize Maxwell equations in VIE form. In Setion 3, we desribe the mesh analysis formulation for dieletris. In Setion 4 we show that the matries resulting from the VIE full mesh analysis approah an be ast into a positive semidefinite form, hene allowing guaranteed passive low order models extration. Finally, in Setions 6 and 7 we show the results of our implementation on simple examples of interonnet strutures inluding dieletris. 2. BACKGROUND Maxwell equations in Mixed Potential Integral Equation MPIE) form are as shown in 1)-4), where V and are the union of the ondutor and dieletri volumes respetively, r and r d are vetors indiating points in V and respetively. µ is the magneti permeability, ε 0 is the permittivity, ε r is the dieletri relative permittivity, σ is the ondutivity of the metal, and ω is the angular frequeny of the ondutor exitation. J is the urrent density in the ondutors. J d jω ε ε 0 E is the polarization urrent density in the interior of the dieletris, and E is the eletri field. The kernel K for a full-wave formulation is a frequeny dependent funtion K r 1 r 2 e jω ε 0µ r 1 r 2 5) r 1 r 2 When the relevant length sales are muh smaller than a wavelength, the above kernel an be approximated with the frequeny independent funtion K r 1 r 2 1 r 1 r 2 6) The salar potentials φ and φ d, an be related to the surfae harge ρ and ρ d, on both the ondutor and dieletri surfaes as shown in 3)-4), where S is the union of the ondutor surfaes, S d is the union of the dieletri surfaes, r s is a vetor indiating a point in S, and r s is a vetor indiating a point in S d. Within eah ondutor, and within eah homogeneous blok of dieletri, J r 0 7) 0 8) J d r d

J r jω µ K r r σ 4π V J r dr J d r d jω jω ε µ K r d r ε 0 4π V J r dr 1 K r s 4πε 0 S rs ρ rs drs 1 K r ds S rs ρ rs drs 4πε 0 K r r d J d r d dr d φ 1) K r d r d J d r d dr d φ d 2) K r s S d rds ρ d rds drds φ r s 3) K r ds S d rds ρ d rds drds φ d r ds 4) for all points r and r d in the interior of V and respetively. In addition, the urrent normal to the ondutor and dieletri surfaes is responsible for the aumulation of surfae harge, ˆn J ˆn J d r s r ds jωρ jωρ d r s 9) r ds 10) where ˆn is the unit normal to S and S d at the points r s and r s respetively. The main unknowns, J, J d, ρ, and ρ d an be approximated by a weighted sum of a finite set of basis funtions. One lassial hoie for the basis funtions is to over the surfae of eah ondutor and of eah dieletri with panels, eah of whih hold a onstant harge density. To model urrent flow, the interiors of all ondutors and dieletris are divided into a 3-D grid of filaments. Fig. 3 shows an example of the 3D volume disretization of a dieletri parallelepiped. Eah filament arries a onstant urrent. Other basis funtions hoies are possible for the interior of the ondutors [18]. A Galerkin method [19] an be used to transform the Mixed Potentials Integral Equations 1)-4) into an algebrai form R sl 1 s 0 Pol P I I d q q d V φ φ d 11) where I, I d, q and q d are vetors of basis funtion weights for the ondutor urrents, dieletri polarization urrents, ondutor harges and dieletri harges respetively. V,, φ C and φ d are the vetors generated by inner produts of the basis funtions with the potential gradient and with the potential itself. The resistane matrix R, the indutane matrix L and the oeffiients of potential matrix P are all derived diretly from the Galerkin ondition [19], R L P R 0 L L d L d L dd P P d 12) 13) P d P dd 14) L and P are frequeny dependent when using a full-wave kernel as in 5), and frequeny independent when using a quasi-stati kernel as in 6). Matrix Pol in 11) is a diagonal matrix arrying the polarization oeffiients Pol i i A i l i ε ε 0 15) where l i and A i are the length and the ross-setional area of dieletri filament i respetively. 3. THE MESH FORMULATION Imposing harge onservation 9)-10) and urrent onservation 7)-8) on surfae and on interior of both ondutors and dieletris makes it possible to use a mesh analysis approah. As a summary, a omplete mesh formulation for strutures inluding both ondutors and dieletris, after the Galerkin transformation an be written simply as: M Z d M T I m V ms 16) where I m are the unknown mesh urrents, V ms is the vetor of known mesh voltage soures, non zero only on the rows assoiated with the external iruit terminals. Z d is the Galerkin impedane matrix Z d R sl 1 s M is a very sparse mesh analysis matrix, 0 Pol 0 0 1 s P M M f M d M p M pd! 17) 18) where submatries M f and M p are the KVL s mesh matries for the ondutors filaments and panels as desribed in [20]. In a very similar way to [20], we an onstrut also M f d and M pd, the KVL s mesh matries for the dieletri filaments and panels. In fat, as for the ondutors, dieletri panel harges an be treated as displaement urrents flowing on iruit branhes to the node at infinity. A set of independent meshes for the three dimensional disretization of the blok of dieletri an be found using a minimum spanning tree. 4. PASSIVE MODEL ORDER REDUCTION In this Setion, we will limit ourself to the usage of the quasistati kernel in 6) whih produes frequeny independent L and P matries in 13) and 14). The tehnique to handle dieletris in [21] uses a similar quasi-stati assumption, and seems more advantageous requiring fewer unknowns. However, not only magneti oupling between ondutive and polarization urrents are negleted by that formulation, but also the matries used in that formulation are not in the form required for Krylov-subspae based passive model-redution shemes [10]. In this Setion, we show instead an easy way to ast our mesh analysis approah into a form suitable for passive redued order modeling. Let us hoose as state vetor for a linear system representation: x #" I m Q s Q ds Q dv $ T 20) In view of this hoie, we an rewrite 16) as shown in 19) on next page, where Q s Q ds Q dv! T M p M pd M f d! T I m s 21)

- M f M f d! R sl! M f M f d! T I m M p M pd! P Q s Q ds! T M f d Pol! Q dv V ms 19) Or finally in linear system terms: ˆL dx dt y t where matries ˆL and ˆR are defined as ˆRx t Bu t 22) Cx t 23) M f M f d* L M f M f d* T ˆL % &') 0 P T 0 Pol* T +, M f M f d* R M f M f d* T M p M pd* P ˆR % &' - P T M p M pd* T Pol* T M T f d +, M f d Pol* 24) 25) V ms. Vetor u t ontains the exitation voltage soures, Bu t Vetor y t ontains the observed output urrents, derived through matrix C from the mesh urrents I m in the state vetor x t. THEOREM 1. Matries ˆL and ˆR in 24) and 25) are positive semidefinite. PROOF. The polarization matrix Pol! is diagonal with positive oeffiients, hene it is positive semidefinite. When using a Galerkin tehnique [19], the oeffiient of potential matrix P in 14) and the indutane matrix L in 13), are both positive semidefinite. The matrix M f M f d! L M f M f d! T is then also positive semidefinite. Sine all the three bloks of the blok-diagonal matrix ˆL in 24) are positive semidefinite, ˆL is positive semidefinite. This onludes the first part of the proof. To prove that ˆR in 25) is positive semidefinite let us alulate: 2 M f M f d! R M f M T f d! ˆR Rˆ T 0 0 26) The resistane matrix R is positive semidefinite, hene the submatrix 2 M f M f d! R M f M f d! T is positive semidefinite and so is ˆR Rˆ T. Sine x T ˆR Rˆ T x 2x T ˆRx, we an then finally onlude that ˆR is positive semidefinite. OBSERVATION 1. When modeling the input impedanes and the transfer funtions of a 3D struture, we apply input voltages at some ports, and we measure the resulting urrents on the same set of ports, hene we are hoosing C B T in eq 22) and 23). OBSERVATION 2. From Theorem 1 and from Observation 1, one an onlude that the formulation in 22)-25) satisfies to the onditions for guaranteed passive Krylov subspae based model redution in [10]. 5. SUMMARY OF OUR PROCEDURE We summarize here briefly for the onveniene of the reader our entire proposed proedure in its final form: 1. First, we disretize both the volumes and the surfaes of the ondutors and dieletris. An example is shown in Fig. 3. 2. We use a standard Galerkin tehnique [19] to onstrut matries R L P Pol in eq.12) to 15). 3. A mesh analysis approah is used to onstrut the sparse KVL s matries M f M f d M p and M pd in 18). More details on how to handle ondutors are in [20]. For the dieletris, we use a minimum spanning tree to find a set of independent meshes. 4. A Krylov subspae based model redution algorithm suh as [10] is then used to produe redued order linear system models. At eah step of the algorithm the quantity ˆR s 0 ˆL! 1 ˆLv, ould be omputed using fast matrix vetor produts and Krylov subspae iterative methods. 5. The redued order model is then used to obtain a plot of the frequeny response as in Fig. 1, and Fig. 4 or to produe an equivalent SPICE iruit for a time domain simulation inluding the non-linear iruitry. The overall omplexity of this proedure is O N m Nlog N, where N m is the total number of moments mathed by the model redution algorithm at all frequeny expansion points. N is the size of the original full linear system model in 22)-23), or about the number of basis funtions used in the volume and surfae disretization. 6. A TRANSMISSION LINE EXAMPLE Two PCB traes are onsidered in this example in a transmission line onfiguration. Traes are loated on opposite sides of a dieletri substrate, and shorted at one end. Fig 1 shows the frequeny response of suh transmission line struture. In Fig 1 we also show the response of the alulated redued order model. Admittane [1/Ohm] 1 10 1 10 2 10 3 Shorted T Line with dieletri in between. epsr=4. lenght=30m. mesh analysis, redued order model mesh analysis, full model 10 4 0 1 2 3 frequeny [Hz] 4 5 6 x 10 8 Figure 1: Redue order modeling of a shorted PCB transmission line. Traes dimensions are 250µm x 35µm x 30m. A 100µm thik dieletri layer ε r. 4) is present between the two traes. The ontinuous line is the admittane vs. frequeny of the alulated redued model. The irles are the response of the original system. At the frequenies where the frequeny independent kernel in 6) yields aurate results, it may also be reasonable to neglet magneti oupling between ondutors and dieletri polarization urrents. However there are ases where even with a non-fullwave

kernel one might observe some effets of the magneti oupling between / dieletri polarization urrents and ondutors. One of suh ases is illustrated in Fig. 2. A via is loated in proximity of the shorted PCB transmission line. The line is then exited at a frequeny lose to the first quarter-wavelength resonane. In this situation most of the urrent loses its path through the dieletri layer in the form of polarization urrents. If a nearby via orresponds to a quiet vitim line, some oupling an be observed between the vertial polarization urrents and the via. Figure 3: Two traes part of an MCM interonnet system figure above). A dieletri layer εr 4 is present underneath the traes and the hips. The figure below shows the volume polarization urrents inside the dieletri layer at the 3 GHz resonane. For visualization purposes, the axes in this piture are not to-sale. Traes are 2m long, 4mm far apart, 250µm wide and 40µm thik. Figure 2: Via loated near a PCB transmission line. In this piture we do not show the dieletri layer whih is loated between the two dark PCB transmission line traes. Shadings orrespond to urrent density amplitudes. On top we show the urrent densities orresponding to the ase where magneti oupling between polarization urrents and ondutors is aounted for. On the bottom we show the same example but setting Ld 0, Ld 0 and Ldd 0 in 13) whih orresponds to negleting magneti oupling between polarization urrents and ondutors. 7. MCM INTERCONNECT EXAMPLE In a seond example, we have applied our tehnique to analyzing two wires of an interonnet bus on an Multi-Chip Module MCM), as shown in Fig. 3. A dieletri layer εr 4) is present underneath the traes and the hips. In Fig. 4 we show the frequeny response of the two interonnets when shorted on one side and driven on the other. We show the frequeny response with and without the dieletri substrate. A signifiant differene in the resonane position an be observed. Fig. 3 shows the polarization volume urrents at the first resonane f 3 GHz. In Fig. 4 we ompare the redued order model to the full model for the ase when the dieletri substrate is present. The redued order model has been built mathing four moments around eah of the following expansion points: s1 j2π 100 KHz, s2 j2π 3 GHz, and s3 j2π 6 GHz. In order to inlude also the point s0 0 among the other expansion points, a non-singular R in 25) would be neessary. It an been shown [22] that a matrix of the form suh as in 25) is non-singular under the ondition that there are no ut-sets of only apaitors. Unfortunately, eah node in our dieletri disretization is suh a ut-set when dieletri losses are negligible. Therefore, for lossless dieletris the point s0 0 annot be inluded in the multipoint expansion algorithm, and a non-zero low frequeny expansion point is used instead. From our experiments, we have observed that this expansion point restrition is not interfering with auray. For instane in this partiular example, the zero frequeny behavior of the struture has been aurately aptured as shown in Fig. 5, whih is a magnified view of the low frequeny part of the plot in Fig. 4, 8. CONCLUSIONS In this paper we desribed applying the mesh analysis approah to solving for the disretized urrents and harges in a VIE formulation. We showed that the approah leads to a system with provably positive semidefinite matries, making for easy appliation of Krylov-subspae based model-redution shemes to generate aurate guaranteed passive redued-order models. Several printed iruit board examples demonstrated the effetiveness of the strategy. Arguably, it is tempting to assume that the VIE approah is a step bakward, as it involves disretizing volumes instead of surfaes. However, volume integral equation methods are used for magneti analysis of ondutor problems, beause ondutors oupy a vanishingly small region of the problem domain. The same vanishingly small oupany argument an be made for dieletris as well. In addition, sine polarization urrents are not outputs, it

Admittane [1/Ohm] 10 1 1 10 1 10 2 10 3 10 4 10 5 MCM interonnet with dieletri, redued order model with dieletri, full model without dieletri 0 1 2 3 4 5 6 7 frequeny [Hz] x 10 9 Figure 4: Admittane vs. frequeny for the two traes in Fig. 3. Admittane [1/Ohm] 12.749 12.748 12.747 12.746 12.745 12.744 12.743 MCM interonnets. Zoom at DC. mesh analysis, redued order model mesh analysis, full model 12.742 0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000 frequeny [Hz] Figure 5: Magnified view of the low frequeny part of the plot in Fig. 4, to verify that the redued model ontinuous line) aptures orretly the DC behavior of the original system irles). might be possible to align them with a regular grid. Suh an alignment might improve the performane of fast solvers, suh as the Conjugate Gradient FFT CGFFT) [23] or Preorreted-FFT [24] methods. This is an important onsideration sine suh solvers are required when using any integral formulation on models with ompliated geometries. 9. REFERENCES [1] Y. L. Chow, J. J. Yang, D. G. Fang, and G. E. Howard. A losed-form spatial green s funtion for the thik mirostrip substrate. IEEE Trans. on Mirowave Theory and Tehniques, 393):588 592, Marh 1991. [2] M. I. Aksun and R. Mittra. Derivation of losed-form green s funtions for a general mirostrip geometry. IEEE Trans. on Mirowave Theory and Tehniques, 4011):2055 62, November 1992. [3] K. A. Mihalski and J. R. Mosig. Multilayered media green s funtions in integral equation formulations. IEEE Trans. on Antennas and Propagation, 453):508 519, Marh 1997. [4] J. Zhao, W. W. M. Dai, S. Kapur, and D. E. Long. Effiient three-dimensional extration based on stati and full-wave layered green s funtions. In Proeedings of the Design Automation Conferene, pages 224 229, 1998. [5] A. C. Cangellaris and V. Okhmatovsky. New losed-form eletromagneti green s funtions in layered media. In IEEE MTT-S International Mirowave Symposium, pages 1065 8, June 2000. [6] A. W. Glisson. An integral equation for eletromagneti sattering from homogeneous dieletri bodies. IEEE Trans. on Antennas and Propagation, AP-322):173 175, February 1984. [7] D. R. Swatek and Ioan R. Ciri. Single soure integral equation for wave sattering by multiply-onneted dieletri ylinders. IEEE Trans. on Magnetis, 323):878 881, May 1996. [8] A. E. Ruehli and H. Heeb. Ciruit models for three-dimensional geometries inluding dieletris. IEEE Trans. on Mirowave Theory and Tehniques, 40:1507 1516, July 1992. [9] T. K. Sarkar and E. Arvas. An integral equation approah to the analysis of finite mirostrip antennas: Volume/surfae formulation. IEEE Trans. on Antennas and Propagation, 383):305 312, Marh 1990. [10] A. Odabasioglu, M. Celik, and L. T. Pileggi. Prima: Passive redued-order interonnet maromodeling algorithm. IEEE Trans. on Computer-Aided Design of Integrated Ciruits and Systems, 178):645 54, August 1998. [11] P. Feldmann and R. W. Freund. Redued-order modeling of large linear subiruits via a blok lanzos algorithm. In Proeedings of the Design Automation Conferene, pages 474 479, June 1995. [12] K. J. Kerns, I. L. Wemple, and A. T. Yang. Stable and effiient redution of substrate model networks using ongruene transforms. In IEEE/ACM Internat. Conf. on Computer-Aided Design, pages 207 214, November 1995. [13] E. J. Grimme. Krylov projetion methods for model redution. PhD Thesis, Univ. of Illinois at Urbana-Champaign, 1997. [14] I. M. Elfadel and D. D. Ling. A blok arnoldi algorithm for multipoint passive model-order redution of multiport rl networks. IEEE/ACM Internat. Conf. on Computer-Aided Design, November 1997. [15] L. M. Silveira, M. Kamon, I. Elfadel, and J. White. Coordinate-transformed arnoldi algorithm for generating guarantee stable redued-order models of rl. Computer Methods in Applied Mehanis and Engineering, 1693-4):377 89, February 1999. [16] J. E. Braken, D. K. Sun, and Z. Cendes. Charaterization of eletromagneti devies via redued-order models. Computer Methods in Applied Mehanis and Engineering, 1693-4):311 330, February 1999. [17] A. C. Cangellaris and L. Zhao. Passive redued-order modeling of eletromagneti systems. Computer Methods in Applied Mehanis and Engineering, 1693-4):345 358, February 1999. [18] L. Daniel, A. Sangiovanni-Vinentelli, and J. White. Condution modes basis funtions for effiient eletromagneti analysis of on-hip and off-hip interonnet. In Proeedings of the Design Automation Conferene, June 2001. [19] R. F. Harrington. Field Computation by Moment Methods. MaMillan, 1968. [20] M. Kamon, N. Marques, L. M. Silveira, and J. White. Automati generation of aurate iruit models. IEEE Trans. on Comp., Pakaging, and Manufat. Teh., August 1998. [21] K. Nabors and J. White. Fastap: a multipole aelerated 3-d apaitane extration program. IEEE Trans. on Computer-Aided Design of Integrated Ciruits and Systems, 1011):1447 59, November 1991. [22] M. Kamon. Fast parasiti extration and simulation of three-dimensional interonnet via quasistati analysis. PhD Thesis, Massahusetts Institute of Tehnology, 1998. [23] B. G. Salman and A. MCowen. A omparative study of the omputation of near-field sattering from resonant dieletri/pe satterers. IEEE Trans. on Magnetis, 323):866 9, May 1996. [24] J. R. Phillips and J. White. A preorreted-fft method for eletrostati analysis of ompliated 3-d strutures. IEEE Trans. on Computer-Aided Design of Integrated Ciruits and Systems, 1610):1059 1072, Otober 1997.