Fast electromagnetic modeling of high speed interconnects in advanced electronic packaging.

摘要

This dissertation is concerned with the use of fast full-wave electromagnetic methods for modeling high speed interconnects and passive components in advanced electronic packaging. The main contents of this dissertation are twofold; to model the multiple vias of various structures and materials within the parallel waveguide, to evaluate the layered medium Green's functions for multilayers and lossy substrate in frequency domain as well as in time domain. Both projects demand high efficiency and accuracy in computation and simulation. The interior problem of massively-coupled vias is modeled using extended Foldy-Lax multiple scattering approaches. The kernel algorithm is the vector cylindrical wave expansions with Bessel's functions and addition theorem, based upon the magnetic waveguide modal solutions of dyadic Green's functions. Numerical simulations of practical problems are illustrated, where the signal integrity issues can be observed like the insertion loss, the return loss and the crosstalk coupling. The results are verified for the S-parameters of various via configurations and via array sizes in layered dielectrics with those generated from Ansoft's HFSS. It is shown that the Foldy-Lax approach is accurate within 5% difference from results of HFSS and is several thousand times faster than running HFSS. For the second project, the exterior via-line transition problem solved by the method of moments (MoM) requests for fast evaluation of layered media Green's functions. Fast all modes method (FAM) and the numerical modified steepest-descent path method (NMSP) are recently proposed to calculate the Green's functions for multilayers and lossy media over an imperfect or perfect conductor. The FAM locates all modes accurately including surface wave modes, leaky wave modes, and improper modes. For a typical 6-layer case over a ground plane, the FAM requires only 2.265 CPU seconds of pre-processing that includes computing 200 mode locations. The NMSP is then used to evaluate the steepest descent path integral. Accuracy within 0.2% is achieved in comparison with the benchmark calculations. Within this context of accuracy, the total CPU per distance point is less than 7.6 milliseconds for distances larger than 0.02 free-space wavelengths and is less than 2.7 milliseconds for distances larger than 2 free-space wavelengths.