Reading time ( words)
What does it take to design predictable PCB or packaging interconnects operating at tens of Gbps? Properly identified dielectric and conductor roughness models, known manufacturer geometry adjustments, and properly validated simulation tools are necessary conditions. One of the sufficient conditions is the localization property; to be predictable, all elements of an interconnect link must be localized up to a target frequency. This article introduces and illustrates the localization concept, with the power-flow density computed using the unique Trefftz finite element solver available in Simbeor THz software.
Ideally, all interconnects should look like uniform transmission lines (or wave-guiding structures) with the specified characteristic impedance. In reality, an interconnect link is typically composed of transmission lines of different types (microstrip, strip, coplanar, coaxial, etc.) and transitions between them such as vias, connectors, breakouts and so on. Transmission lines may be coupled to each other that cause crosstalk. The transitions may reflect and radiate energy due to discontinuities in signal and reference conductors. The crosstalk, reflections and radiation cause unwanted and sometime unpredictable signal degradation. If analysis of traces or via hole transitions is possible in isolation from the rest of the board up to a target frequency, the structure is called localized. Structures with behavior that is dependent on other structures and board geometry are called not localized, and they should not be used in multi-gigabit interconnects in general.
Examples of non-localized structures are coupled traces, striplines with non-connected reference planes, traces crossing gaps in reference planes, vias with far, no or insufficient stitching vias (vias connecting reference planes of the connected traces). Analysis of non-localized structures is usually possible only at the post-layout stage with substantial model simplifications that degrade accuracy at higher frequencies. To design predictable interconnects, only localized structures must be used—this is one of the most important elements for design success. The localization is always bandwidth limited for striplines (two reference conductors) and for vias (two or more reference conductors). How do we estimate the localization property of a transition? One way is to run an electromagnetic analysis of the structure with different boundary conditions or simply change simulation area size without changing phase reference planes and evaluate the differences in the computed S-parameters. If the difference is small, the structure may be considered localized and suitable for final design.
To read this entire article, which appeared in the September 2018 issue of Design007 Magazine, click here.