Issue link: https://iconnect007.uberflip.com/i/1511130
18 DESIGN007 MAGAZINE I NOVEMBER 2023 true. With insertion losses being directly pro- portional to the line length, then halving your line lengths (where possible) will halve your losses: simple and free. I sometimes see ques- tions about line impedance on very short lines, where the frequency of interest is relatively low, and the interconnect at each end of the line is a significant portion of the length. It pays to check whether impedance or loss is an issue in these circumstances. ere is a simple Micro- so Excel tool called a critical trace length cal- culator, which helps check the critical length. Whilst on the subject of small or short, mov- ing from the X-Y to the Z-axis, the same is true for vias: Keep them short. If you can't keep them short, keep the stubs short. Let's drill down (sorry) on all the possibilities for mini- mising the signal integrity challenges of vias. Remember that, in most cases, the vias are short compared with the trace length unless it is a really thick board. inner boards (where possible) have inher- ently shorter vias. Where the PCB is necessar- ily thick because of high layer count, there are several approaches: • Minimise the stub length • Use microvias • Back-drill the offending stubs Why are via stubs a problem? Simply put, it is because a signal arriving as a stub sees it electrically twice as long as it is mechanically. A signal arriving from the top layer of the PCB traversing down the via to a trace on an inner layer can't tell the difference between propa- gating further down the via (the stub) or con- tinuing along the trace, so it does both—the part propagating along the intended trace con- tinues unimpeded. But the part of the signal that chances to propagate down the stub finds an open circuit at the end of the stub and 100% of that diverted signal will reflect and rejoin the main trace, at which point part of the stub reflection will return toward the source and the rest will rejoin the signal headed toward its intended destination. Using buried vias, back- drilling the stub, or using microvias are all technical solutions to this, but an alternative is to (where possible) route so that the stubs are as short as possible—say, on a 10-layer board route a signal from L3 to L8 rather than L1 to L5, which would leave a far longer stub. Some- times a rapid check to see whether a feature, such as a stub, is a cause for concern can rap- idly put you at ease or flag if more attention is needed. Figures 1, 2, and 3 depict differing stub lengths and their relative impact, from "safe to ignore" (green 28-mil stub) to "do something about it" (red 200-mil stub) when the stubs are in a 4Gbit/s data channel. Figure 1: Via length of 28 mils causes no concern at 4 Gbit/s data rate. Figure 2: Via length of 149 mils causes some concern at 4 Gbit/s data rate. Figure 3: Via length of 200 mils requires attention at 4 Gbit/s data rate. (Find the math behind this in "Signal and Power Integrity Simplified" by Dr. Eric Bogatin)