Large space RFIC spiral inductor parametric modeling technique

In this paper, we present a novel method to develop wideband scalable inductor models that are accurate over a large dimensional space. We use a modified double-π topology for the inductor equivalent circuit model. The circuit element values are computed using empirical functions which are formulated using inductor geometries and numerical coefficients. Values of the coefficients are extracted from electromagnetic simulation data over a large inductor geometrical space. A neural network model is trained to learn the relationship between the geometrical dimensions and coefficients. The model has been developed and tested in integrated circuit process technology. Result shows that the model is very accurate over a large geometrical space and frequency range. The technique is general and is applicable to other passive components model development.