Issue link: https://iconnect007.uberflip.com/i/750006
36 The PCB Design Magazine • November 2016 ground plane pairs; as a result, the return cur- rent on a badly designed single-ended via finds itself trying to radiate out through each plane pair and radiates until it meets a boundary; of- ten, this is the open circuit at the edge of the board. Needless to say, this is a very complex 3D scenario to model. However, all is not lost. Recall my note at the start that modelling can't fix a poor design. So, how do you mitigate the effects of vias? There are several approaches, depending on how demanding the application is, and only once you have thought of an ap- propriate approach does it make sense to start modeling to confirm your design. A first step would be to keep the via as short as possible by using microvias or blind or bur- ied vias. Second, you should consider the chal- lenge of via stubs: if a trace goes from L1 to L3 but leaves a stub down to L16, the signal will traverse though all the layers to L16 before re- flecting to L3, giving an effective electrical stub length almost double the mechanical one. In the case of very thick boards with through-hole vias, back-drilling is an option to remove the unnecessary part of the via barrel. However, an- other approach is maybe a little counter intui- tive: It may be better to run the signal down the full depth of the via and back up an adjacent via to reach the desired layer, as this could present a shorter stub than just going down one layer and leaving a long stub. Going back a moment to the explanation of the challenging return path seen by the re- turn current from a single via, you can make life easier for the return current if you design a return path rather than leaving it to chance. So, by adding a return via adjacent to the signal via you give the return current a favourable path back to the source rather than allowing the high frequency to bounce around in the cavity between the planes. This also leads to one of the reasons that dif- ferential transmission is popular for high-speed signals: Aside from its inherent noise immunity, the balanced differential pair means you provide an outbound AND return path in the signal via pairs without having to think about it. In other words, the signal integrity of a differential pair is inherently less effected by transition through vias than for the single ended case. As always in The Pulse columns, I have avoid- ed too much in-depth analysis. I've rather given some guidance to help you get a better feel for the behavior of vias. I hope that you consider these points, so that when you do come to de- sign high-speed vias, you'll have a head start. When modelling is required, now you'll know that you should not expect a good model to fix a bad design. Instead, you can use modelling to confirm your design, which has started from a good perspective. In many cases, you can attempt to keep the geometries small enough that the via is invisible. (Shameless plug: The via stub length indicator in the Polar Si9000e PCB transmission line field solver can help.) One closing thought. Wouldn't it be great if an HDI wizard could create a coaxial via so the high-speed signal didn't see all those nasty slots between planes as it traverses down through the board? Well it has been done, and patented; an online search will show you some of the ideas. I imagine it is not as easy as it seems, or it would be a mainstream approach by now. I would love to hear back from fabricators on this topic. PCBDESIGN Martyn Gaudion is CEO of Polar Instruments. To contact him or view past columns, click here. " There are several approaches, depending on how demanding the application is, and only once you have thought of an appropriate approach does it make sense to start modeling to confirm your design. " VIAS, MODELLING, AND SIGNAL INTEGRITY