Behavioral Modeling of Nonlinear Circuit Elements: Application to Signal Integrity and Electromagnetic Compatibility

Nowadays the availability of computational models of nonlinear dynamic components is becoming a key requirement for the analysis and design of complex systems. The complexity of many components, however, as well as the lack of information on their internal structure often prevent from the development of traditional physical models. This scenario raise the interest for behavioral models, which are models obtained from the observation of the external behavior of components. In this thesis we focus on the development of behavioral models for the assessment of Signal Integrity (SI) and ElecroMagnetic Compatibility (EMC) effects on fast digital circuits. Such an assessment, that is mainly achieved by simulating the evolution of signals sent on interconnects by digital integrated circuits (ICs), requires e±cient and accurate models of IC ports driving and loading the interconnects themselves. The required models must allow the simulation of large realistic problems and must performs at an accuracy level useful to the prediction of sensitive effects, like crosstalk and radiation. Behavioral models meet such requirements and are establishing as the best tools for the description of IC ports. We analyze possible behavioral modeling methods for IC ports and concentrate on behavioral modeling via black-box identification. It amounts to the selection of a suitable parametric model and to the estimation of its parameters from measured transient responses. The selection of a suitable class of parametric models leads to Radial Basis Function (RBF) representations, that offer many advantages in the modeling of systems with strong nonlinear nature and multiple inputs. We developed a simplified RBF model that can be obtained from measured port voltage and current. The estimation of such model is simple and relies on a robust algorithm. In order to test the effectiveness of the proposed approach and its feasibility, we apply it to the modeling of several virtual devices and to an actualdevice of interest. The obtained models perform at a fairly good accuracy and effciency levels and turn out to be weakly sensitive to driven loads and measurement setup. Besides, since the model structure is selected by the estimation process itself, all the relevant physical effects relating input and output signals (e.g., substrate or packaging effects) are automatically taken into account.