Generation of rational model based SPICE circuits for transient simulations Nobby Stevens * and Tom Dhaene ** * Agilent Technologies Belgium, EEsof EDA, Kortrijksesteenweg 1093b, 9051 Gent, Belgium ** Ghent University – IBBT, Gaston Crommenlaan 8 bus 201, 9050 Gent, Belgium nobby_stevens@agilent.com Tom.Dhaene@UGent.be Abstract In this paper, we present an efficient and compact circuit netlist formulation for rational models. It is shown how to implement the method for several commercially available time domain solvers. Finally, an example is given describing the flow and procedures to follow in order to obtain a reliable and stable transient simulation. Introduction Designing a circuit with analog, digital and/or active components requires a rigorous analysis in the time domain. Especially with regard to signal integrity issues such as cross- talk and signal degradation, it is crucial to have a good quantitative knowledge of these effects during the design phase. As of today, a number of SPICE simulators are commonly used for transient simulations. Note however that most electromagnetic simulators and measurement tools perform in the frequency domain. The conversion from frequency domain to time domain is, from a numerical perspective, a delicate procedure. Here, rational macromodeling comes into play. Once a passive mathematical rational model of the frequency data is constructed, a last and important step to take is the realization of a compact SPICE-like circuit using lumped components, that can be efficiently processed by a time-domain circuit simulator (such as e.g., Hspice and Spectre). Equivalent SPICE realization Consider the scattering matrix S of a Linear Time-Invariant system (LTI), with b and a respectively the column vectors of the reflected and incident waves: a S b = (1) Suppose that for this system, a broadband passive rational model has been constructed by use of the Vector Fitting technique [1-3] or the direct interpolation scheme [4]. For each element S kl of the scattering matrix, one can write: = + = kl M 1 m kl m kl m kl kl p s c d S (2) where, the kl m c are the residues and kl m p are the corresponding poles. They both are real, or occur in complex conjugate pairs. Note that the order kl M may depend on the dynamic behavior of each individual element. Also a set of common poles can be used, or each element can have its own set. We obtain the following relationship between the reflected and incident wave of an N-port system: = = + = N 1 n n M 1 m kn m kn m kn k a p s c d b kn (3) Realization of vi/ab transformation SPICE circuit simulators work with port voltages and currents. The relationship between the reflected wave b k and incident wave a k and the corresponding port voltage v k and port current i k is given by (Z 0 is the reference impedance): 0 k 0 k k Z 2 i Z v a + = (4) 0 k 0 k k Z 2 i Z v b = (5) Using these relations, we can see the incident wave as a (scaled) virtual voltage and the reflected wave as a scaled virtual current: k 0 k 0 k b Z 2 i Z v + = (6) k 0 0 k k i 2 Z Z 2 v a + = (7) From a network point of view, this can be represented as in Figure 1. (CCVS stands for current controlled voltage source and VCVS for voltage controlled voltage source.) This way, no negative resistors are introduced as in the work by Lamecki [5] and Neumayer [6]. Some SPICE simulators may have difficulties with negative resistor values. Realization of rational transfer function At this point, equation (3), which describes the relation between incident and reflected waves, must be reformulated as a circuit equivalent network. The form of the rational representation (2) is particularly interesting for transient simulations since the inverse Laplace transform of each term is known. Recent versions of commercially available SPICE-like circuit simulators make use of this fact and actually support the rational presentation as defined in equation (2). 978-1-4244-2318-7/08/$25.00 ©2008 IEEE SPI 2008