SPICE Model Tuning using a Parallel Computational Optimisation Strategy

Signal integrity is of high importance to highspeed digital interfaces and is typically analysed in the time domain using eye-diagrams. The available data of devices under investigation are often given as S-Parameters in the frequency domain and hence have to be transformed into the time domain. Here, problems like passivity and causality might occur. Using a SPICE circuit model with only passive components avoids these problems. In this work, such a model is designed by an expert based on knowledge about the physics. Subsequently, component values of this SPICE model have to be tuned to match the given data in the frequency domain. Even for small networks the design space is too large for exhaustive search. Here, a novel computational optimisation algorithm is proposed and used for the automated tuning. An advantage of this new method is that no control parameters have to be tuned for the optimisation itself. Another advantage is that the algorithm can be parallelised and hence benefits from multi-core architectures of modern workstations. The method was applied to the modelling of a planar coil. It was shown that the algorithm is capable of finding near-optimum solutions within reasonable time, satisfying requirements for practical applications.