Implementation issues and performance evaluation of table models in SPICE3

All of the existing semiconductor device models in SPICE3 are based on a set of analytical equations which describe the currents and charges resulting from the applied bias. These have familiar names such as Gummel-Poon and BSIM. Because modern devices exhibit more complex behaviour it is very difficult to represent their operation by such a set of analytical equations. One possibility to overcome this difficulty is the use of models based on look-up tables (the so called table model approach). This paper will discuss the implementation of table models into a circuit simulator such as SPICE3 and examine the merits of this approach with respect to the more familiar compact models. In particular, we discuss how to construct the table models so that they provide the information needed by the SPICE non-linear solvers for DC and transient analysis and by the linear solver for AC analysis. This will include discussion of the table size requirements for a given accuracy specification, the interpolation schemes required, calculation of derivatives and charge conservation. The methods of extrapolation used for device characteristics beyond the limits of the data tables will also be discussed. Quantitative and qualitative analysis of the linear interpolation method will be presented in terms of memory requirements, calculation speed and the accuracy of the interpolation. Additionally, the performance of the table model will be compared to that of a widely used analytical compact model, such as the BSIM3 model.