Search

This paper introduces physics-based stripline ports for a segmented optimization of component footprints on printed circuit boards (PCBs). A segmentation is used for dividing a digital communication system into well-defined parts that are considered to be mostly independent. This technique is widely used in conjunction with generic 3-D electromagnetic field solvers, but it is new to physics-based models that until now only allow for ports at the upper and lower via ends, i.e., on PCB surfaces. Evidence of feasibility is provided by means of correlation with measurements and full-wave simulations up to 40 GHz. Since physics-based models can efficiently capture and simulate even systems of thousands of elements within acceptable computation time, the results of a segmented simulation can easily be compared to the full-board environment. The corresponding results are correlated to investigate limitations. [ABSTRACT FROM AUTHOR]

This paper presents, for the first time, the comprehensive modeling of complete via constellations consisting of several thousands of vias in multilayer printed circuit boards using the physics-based approach. For each computational step of the physics-based approach, several alternatives are analyzed with regard to their computational efficiency, and calculation times are discussed as a function of the number of simulated vias. The results of this analysis are used in combination with previous studies to determine an efficient yet accurate algorithm for the simulation of large numbers of vias. The impact of the stackup configuration on the computational effort of the algorithm is analyzed, and the most computationally expensive parts of the calculation process are identified. A parallelization of the algorithms is carried out to accelerate the critical calculation tasks. As an evaluation example, simulation results for a via array consisting of 10 000 vias and eight cavities are shown. With the proposed simulation methods, the computation time for this via array is about 6.5 h per frequency point on a single CPU and about 40 min per frequency point with the parallel version running on 16 CPUs. [ABSTRACT FROM PUBLISHER]