Partial Element its Application Equivalent Circuit in the F’requericy (PEEC) Method and and Time Domain Albert Ruehli IBM Research Division, T. J. Watson Research Center Yorktown Heights, NY 10598 Abstract The PEE’C approach is a full wave electromagnetic electrical modeling technique foT conductors embedded in arbitrary dielectrics in terms of equivalent circuits. The models can be used in both the time as well as the frequency domain. It facilitates the solution of problems which have both an electromagnetic part as well as a circuit part. Also, it leads to an intuitive understanding of electromagnetic problems. 1 Introduction The solution of electromagnetic problems is formulated in terms of circuits in the Partial Element Equivalent Circuit (PEEC) approach [l] instead of the usual field solution and field variables. The variables are quantities like the potential ‘P, and some other circuit variables like i, v, q, where i is the current, v the voltage, and q some charges etc.. The basic formulation is an Electric Field Integral Equation (EFIE) full wave solution to Maxwell’s equations. The models evolved first in terms of partial inductance calculations for VLSI inductance calculations [2]. The full PEEC models were developed from the study of partial inductances and the method of moments (MOM) techniques [3]. Recently, the PEEC model has been extended to accurately include models for dielectrics [4] and also scattering or incident fields [5]. PEEC models can be employed in both the frequency as well as the time domain. So far, we have been able to extend all the new developments of the method to both domains without any restrictions. In the time domain a large system of delay differential equations (DDE) result. Many realistic problems, may also include a nonlinear part which isbest solved in the time domain. The time domain solution of EFIE has traditionally been plagued by instabilities [6] and the O-7803-3207-5/96/$5.00 0 1996 IEEE PEEC models are no exception. Much progress has been made in this area and we believe that the problem has been solved [7]. The solution of the PEEC models can be obtained in different ways. Many techniques like simplified PEEC models, macromodels and special circuit formulations can be adapted for the efficient solution of PEEC model problems. Macromodeling is a key technique for the solution of very large problems. These techniques are used to model complex and challenging problems as is evident from work done in this area e.g. [S]. Attention has been given recently to the solution of large circuit models which result from the PEEC approach e.g. [9]. In this paper we cover several aspects of PEEC circuit models. Section 2 describes the basic circuit models and the derivation of the PEEC model from Maxwell’s equations. Section 3 shows circuit models and aspects of PEEC model. Stability and time domain issues are given in Section 4. Finally, Section 5 considers model simplification and some example results are given. 2 Circuit Based EM Models Circuit type models have been in use in EM for many years. The most widely used circuit model is the lumped element Differential Equation (DE) model for transmission lines [lo]. An interesting aspect is the fact that the longitudinal direction is represented by a DE circuit model while the transverse direction for the coupling is represented in terms of an integral equation(IE) model which is in fact related to a PEEC model. The main difference is that the PEEC model is a full wave model while the transmission line model supports TEM modes only. We give an abbreviated derivation of the PEEC model which includes finite size dielectric regions besides the conductors or wires [4]. The starting point is 128 Figure 1: Two cell example the PEEC model for conducting objects [l] and some of the work on dielectrics for the MOM e.g. [ll]. We specifically concentrate on approximations of the currents, charges and potentials (voltages) which are appropriate for an equivalent circuit representation. The key idea is to treat the displacement current due to the bound charges for to the dielectrics with er > 1 separately from the displacement currents due to the free charges. We start from the sum of all the sources of electric,,field [lo] at any point in space Figure 2: Complete PEEC model for two cell example The scalar potential is similarly cl,)dvr (7) JVI,F')q(rr, G(i=, where & is a potential applied electric field, 9 is the current density in the conductors, d and 9 are vector and scalar potentials, respectively. As indicated above, the dielectric areas are taken into account as a current rather than a capacitance through the scalar potential. T-his is accomplished by adding and subtracting ~0% in the Maxwell equation for B, or vxIT=JfeO(er-l)~+Eg~ Finally, using the above we can formulate an integral equation for the electric field at a point F which is to be located either inside a conductor or inside a dielectric region. Starting from Eq. 1 with the externally applied electric field set to zero, and substituting for A and 0 from Eq. 4 and Eq. 7 respectively qe, t) u J(F, t) = &(F, t) + E& - l)z ai. = /.A G(F, Ff)f(i;‘, td)dti! J VI where the retardation which simply points F and derived here, functions are + EO(G - l)p %&,, G(F, Fl)q(Ti, (3) where jc~(F,t) is the conductor current and the remainder of the equations is the equivalent polarization current due the the dielectrics. The vector potential d is for a single conductor at the field point ? given by &=,t) G(F, i;r) “(; p (2) Here, the current in Eq. 2 is written as a total current JVI ,-,(F ;,) d2E(er, td) 1 d-t2 JVI =0 (8) JVI t&h/ + (4 Note that Eq. 3 is used for the total current in both the conductors and the dielectrics. To solve the integral equation Eq. 8 numerically, we need appropriate approximations to the current, potential, and charge variables. We do not use the usual high frequency skineffect approximation but we use a more accurate approximation to the internal current flow which is detailed in [12]. This implies that we cannot utilize the continuity equation time is given by is the free space travel time between the or. It is noted that in the formulation both the retardation and the Green’s free space quantities where (6) to replace all the charge variables directly with the current variables as is done in the moment type solutions [3]. The charges are, in all the solution approaches, assumed to be located on the conductor surfaces. In our approach, we implement the continuity equation in the form of Kirchoff’s current law in the circuit equation solution. 129 Three coupled integral equations (I.E.) result if we represent the vector quantities in terms of the Cartesian coordinates. For this case, the vector quantities are f = J,O + Jy$ + J,2 and l? = E,S + Eyfi + E,&. Using this, the three I.E. are identical in form for all three space directions z, y, z. A very simple example geometry is given in Fig. 1 which is a metal strip. It should be noted that this problem cannot be solved with the usual transmission line circuit models. The aim is to find an equivalent circuit where the unknowns are the potentials as well as the partial inductance currents. We must use pulse functions matching for a Galerkin solution. The inner product is formed with respect to a cell of conductor cy -1 au . J f(r)czv = vp ; J i a, m da dl 1, i4”, 11 11 m 3 Tm F coupling to cell Q for a current 1~ in the cell /?. The dielectrics are represented with additional circuit elements. We define the excess capacitance of a dielectric cell as (10) For the second and third terms in Eq. 8 we make use of the laminar, uniform current flow through the cell to take $-ail/at integral. With some help from the second term represents the of the conductor cell CYIwhere between two parallel cells a! and JJ hap v.vB Lp,p = -k- II T 22 j; Figure 3: Voltage source capacitance model where er is the dielectric constant of the dielectric cell. The equivalent circuit for the dielectric bar or cell y is given by a partial inductance Lprr in series to a capacitor 6’: with the current i = C+ w. where aa and <Ps are the potential at thi two ends of the dielectric cell. This model is a very important addition to the PEEC approach [4]. 3 outside the ognire that inductance inductance by m I ; oL where v, is the volume of the conductor cell Q, a, is the cross-section of the cell (z - y direction) and I, is the length (z-direction) and f(F) is the integrand. This corresponds to averaging the field point f over the cell volume vcr. From the result, we can assert what the appropriate equivalent circuit is for each term of the IE, corresponding to Eq. 8. The first term of the IE can easily be shown to be the resistance of the cell R, with the dc current 1,” = v since the current through the cell is Iz = Jza,. The resistance is as usual a& -= ai iji; G(fa, $)dvadvp (12) [2], we recpartial self the partial p are given where the first term is the partial self inductance of the cell cz and.the second term represents the inductive Circuit Models Provided that a solver exists, equivalent circuit solutions represent a very flexible approach for the solution of practical problems. We give PEEC models in this section for both ,the time domain as well as the frequency domain. The most efficient model for the two cases differ in the circuit used to represent the capacitive effects. A complete model for the example metal conductors shown in Fig. 1 is given in Fig. 2 [5]. The model for the partial inductances is rather straightforward following the basic implementation of PEEC models [l]. We show in more detail how the capacitive effects are modeled. The general form of the capacitive term is Cp= PQ where P is the coefficient of potential matrix. The case where the retardation cannot be ignored is very interesting in terms of the capacitive model. Two different forms of the capacitive equivalent circuit are most appropriate depending on whether the PEEC model is applied in the time or frequency domain. The most efficient model for the MNA time domain formulation is the one shown in Fig. 2. The controlled sources are derived from the (13) For the Cartesian coordinates considered here, all perpendicular cells have zero mutual inductances. Hence, the inductive terms are simply interpreted as Equivalent 130 PEEC - Model + “’ Source ve+ w Figure 4: Simple PEEC model for one cell above coefficient of potential equation with the relationship between current and charge i = $. The capacitive current is for K capacitive cells where i,, is the total capacitive current for cell m and from Eq. 5 the retardation time is (17) where &,, is the distance between conductor cells m and n and c is the speed of light. We measure the distance K&, between two points on two different cells m and n. The second circuit model for the retarded coefficients of potential shown in Fig. 3 is more suitable for a loop oriented formulation like the MLA in [12]. This is the preferred formulation for the frequency domain since it leads to the minimum number of unknowns. This formulation without additional circuit elements included corresponds to a Galerkin MOM method [3]. The PEEC model for this case is the same as the one given in Fig. 2 where the capacitive part is replaced with the one in Fig. 3. It can easily be proven that the ThCvenin equivalent circuit has voltage controlled voltage sources which are given by (18) where s@):,is the potential across the self capacitance n, shown in Fig. 3. Both formulations use the same P matrix and the same delays. Figure 5: PEEC model for dipole antenna 4 Model Iution stability and time domain so- Stability is a key problem for time domain solution, for both the MOM and PEEC formulations e.g. [S] and PEEC formulation. In the time domain, the model instability will be observed as an unstable solution which will totally mask the real solution starting at some point in time. The confusing aspect of the stability issue is that there may be two different contributing factors which may cause instabilities. First, the discretized model may be unstable. However, this can be corrected by adjusting the delay times Eq. 17 in the model shown below and by using other techniques which improve the model stability [7]. The second issue for the stability is the time integration methods used for the integration of the resultant Delay Differential Equation (DDE) [13]. We cannot determine from an unstable solution which one of the sources caused the problem. However, a carefully implemented method leads to a stable time domain solution. The key issue for an efficient time domain formulation for the PEEC model is to introduce all the state variables which are required without introducing additional unknowns. A condensed modified nodal analysis (MNA) matrix can be found by utilizing Kirchoff’s 131 DIPOLE (45cm x 9.0cm) 1 DescriDtion * t Full Model All delays small Resistance small High Currents Chip power,Gnd Low Currents Chip, Signals PEEC Model (W,,W Ignore all 7 (W,,P) or (R,L,,C) (Lp,P,~) or (Lp,C) WP) or (LPI (R,P) or (R,C) 4.08 _........______...... i ........._..__..... $ _..___............... w .............+.................... -009 ........._........... j ................+... ..................p.....- ......t ................. . -8dOOOO 4O;OOO 's.OdOOO 12.0bOOO Frtlquency (GHr) 16.0bOOO 5 Figure 6: Impedance of patch dipole current law at each node to eliminate unnecessary currents. This is possible due to the fact that the internal nodes of the PEEC model have a very specific known structure [5]. The values of the partial inductances Lp for the model are found using the formulas in [2]. We give a simple example for the complete time domain formulation which corresponds to the one cell model in Fig. 8. The corresponding PEEC model is shown in Fig. 9 and the time domain equations are where I means that its variable is evaluated at t - r. Equation (19) can be rewritten as a Delay Differential Equation (DDE) in an MNA form similar to [9] which is G(t) = Py(t) + Qy(t - T) + ~(4 t - T) (20) where P,Q and u are evident. The model for larger problems with multiple delays leads to the equation for a PEEC model in the general form Qiy(t - 7i) + c jr = Py + c i zci(t - pi) Table 1: PEEC Model Reduction 20.00000 (21) i For the time domain solution of the DDE, a specially stable integration method needs to be used. The result of work in this area shows that explicit integration methods are only marginally stable. Implicit integra tion methods which have excellent stability properties like the backward Euler or the Lobatto III-C have to be used. It is clear that the solution of the implicit system is much slower than the explicit ones. Simplified PEEC models and results Macromodels obtained by pruning an existing PEEC model are used quite frequently. We denote a full model by the notation (R,L,,P,r,V,)PEEC where the terms mean that the quantities are included in the PEEC part of the model. The notation which indicates the presence of at least one element of a particular type is R for series resistances, Lp for partial inductances, P for the coefficients of potential, C for capacitances, T for delay and VP for potential incident field sources [5]. Table 1 gives most of the situations for reduced circuit PEEC models. A small delay 7 implies that for the maximum angular frequency of the spectrum w, (UT) << 1. This may be due to a small geometry or due to a sufficiently low frequency w. A substantial reduction in model size may be achieved in situations where these approximations apply. In the (L,,R)PFEC case we use the L, and the R matrices only and the circuit equations degenerate to V(s) = RI(s) + sL;l(s) The admittances (22) are easily found from I(i) = (R + sLp)-lV(s) (23) This model may be further reduced if the resistance is small compared to the partial inductances for the frequencies of interest. Then a frequency independent reduction results I(s) = f(Lp)-lV(s) (24 which leads to a simple solution. It is evident from the extensive use of inductance calculators in the EIP and EIP areas for noise calculations that many practical problems can be solved with (Lp) or (Lp, R)PEEC models. Such calculations have been used for years 132 three dimensional multiconductor systems. IEEE Trans. on Microw. Theory and Techn., MTT22(3):216-221, March 1974. PI A. E. Ruehli. Inductance calculations in a complex integrated circuit environment. IBM J. of Res. and DeveIop., 16(5):470-481, September 1972. [31 R. F. Harrington. Field Computation Methods. Macmillan, 1968. 0 0.2 0.4 0.6 0.6 PI A. Ruehli and H. Heeb. Circuit models for three-dimensional geometries including dielectrics. IEEE Trans. on i&Tow. Theory and Techn., 40(7):1507-1516, July 1992. 1 time (ns) Figure 7: Current response for simple PEEC model where the model complexity has been drastically reduced by, for example, including much simplified capacitances in the model. The reduction of models using systematic macromodeling techniques is an area of intensive research activity today e.g [9]. W e conclude the paper by giving a simple result in both the frequency and the time domain. Figure 5 gives the most simple PEEC model for a dipole antenna. Figure 6 shows the input impedance of a patch dipole antenna (4.5 cm wide and 9 cm long) with a comparison between the M O M solution in [ll] and a PEEC model. The dashes curves are the imaginary parts and the solid and dotted curves the real parts. As is evident, the results are very close. The second example is a time domain solution for a 0.5 cm by 2 cm patch excited by a pulse function with a PEEC model similar to Fig. 4. The result in Fig. 7 shows the convergence of the current waveform with the number of cells along the length. 6 Conclusions and Acknowledgements The PEEC method offers several additional features above the conventional field results due to the circuit oriented solution method. The author would like to acknowledge many discussions with B. Rubin as well as the frequency domain results obtained from the collaboration with J. Garrett and C. Paul and the time domain results obtained form the work with W . Pine110 and A. Cangellaris. [51 A. E. Ruehli, J. Garrett, C. R. Paul. Circuit models for 3d structures with incident fields. In Proc. IEEE Id. Symp. on Electrom. Cornpat., pages 28-31, Dallas, TX, August 1993. PI B. P. Rynne. Comments on a stable procedure in calculating the transient scattering by conducting surfaces of arbitary shape. IEEE Trans. on Ant. and Propag., APP-41(4):517-520, April 1993. [71 A. Ruehli, U. Miekkala, H. Heeb. Stability of discretized partial element equivalent efie circuit models. IEEE fians. on An-L. and Propag., APP43(6):553-559, June 1995. PI R. F. Milsom, P. A. Jamieson, K. J. Scott. EM1 prediction in system level simulation. In PTOC. ht. Symp. on EIMC, volume 10, Ziirich, Switzerland, March 1995. PI E. Chiprout, M.S. Nakhla. Asymp-totic waveform evaluation. KLUWER, 1994. [101S. Ramo, J. R. Whinnery and T. Van Duzer. Fields and Waves in Communication Electronics. John W iley and Sons, 1994. Pll Equivalent circuit B. J. Rubin, S. Daijavad. Radiation and scattering from structures involving finite-size dielectric regions. IEEE Trans. on Ant. and Propag., 38(11):1863-1873, November 1990. P21J. E. Garrett, A. E. Ruehli, C. R. Paul. Efficient frequency domain solut. for speec efie for modeling 3d geom. In PTOC.Int. Symp. on EMC, volume 10, Ziirich, Switzerland, March 1995. A. 1131 E. Ruehli, U. Miekkala, A. Bellen, H. Heeb. Stable time domain solutions for EMC problems using PEEC circuit models. In Proc. IEEE Id. Symp. on ElectTom. CornpaL, Chicago,Ill, August 1994. References [l] A. E. Ruehli. by Moment models for 133