Electronic Thesis and Dissertation Repository


Doctor of Philosophy


Electrical and Computer Engineering


Dr. Anestis Dounavis


With the rapid growth in density, operating speeds and complexity of modern very-large-scale integration (VLSI) circuits, there is a growing demand on efficient and accurate modeling and simulation of high speed interconnects and packages in order to ensure the signal integrity, reliability and performance of electronic systems. Such models can be derived from the knowledge of the physical characteristics of the structure or based on the measured port-to-port response.In the first part of this thesis, a passive macromodeling technique based on Method of Characteristics (referred as Passive Method of Characteristics or PMoC) is described which is applicable for modeling of electrically long high-speed interconnect networks. This algorithm is based on extracting the propagation delay of the interconnect followed by a low order rational approximation to capture the attenuation effects. The key advantage of the algorithm is that the curve fitting to realize the macromodel depends only on per-unit-length (p.u.l.) parameters and not on the length of the transmission line. In this work, the PMoC is developed to model multiconductor transmission lines.Next, an efficient approach for time domain sensitivity analysis of lossy high speed interconnects in the presence of nonlinear terminations is presented based on PMoC. An important feature of the proposed method is that the sensitivities are obtained from the solution of the original network, leading to significant computational advantages. The sensitivity analysis is also used to optimize the physical parameters of the network to satisfy the required design constraints. A time-domain macromodel for lossy multiconductor transmission lines exposed to electromag¬netic interference is also described in this thesis based on PMoC. The algorithm provides an efficient mechanism to ensure the passivity of the macromodel for different line lengths. Numerical examples illustrate that when compared to other passive incident field coupling algorithms, the proposed method is efficient in modeling electrically long interconnects since delay extraction without segmentation is used to capture the frequency response.In addition, this thesis discusses macromodeling techniques for complex packaging structures based on the frequency-domain behavior of the system obtained from measurements or electromagnetic simulators. Such techniques approximate the transfer function of the interconnect network as a rational function which can be embedded with modern circuit simulators with integrated circuit emphasis (SPICE). One of the most popular tools for rational approximations of measured or simulated data is based on vector fitting (VF) algorithms. Nonetheless, the vector fitting algorithms usually suffer convergence issues and lack of accuracy when dealing with noisy measured data. As a part of this thesis, a methodology is presented to improve the convergence and accuracy issues of vector fitting algorithm based on instrumental variable technique. This methodology is based on obtaining the “instruments” in an iterative manner and do not increase the complexity of vector fitting to capture the frequency response and minimize the biasing.