Answering Length Tuning FAQs | Mismatches, Impedance, Vias, and More

[@robertswilson3483]

Great stuff, explained well. As a designer without a degree I really appreciate this stuff. I will never claim to be an engineer, but at least I have a better picture of what is happening and this'll make me a better designer. Thanks!

[@Zachariah-Peterson]

You're very welcome, we'll keep posting these FAQ videos for different topics.

[@ricardomatias5160]

Excellent explanation. Thank you.

[@AltiumAcademy]

You are welcome!

[@asdfzxcv6268]

When it matters? Are there any estimations to distinguish when length matching is doing something and when it's only fancy looking over-engineering? A few milimeters of mismatch seems to be considerable only at some many GHz range. What are the most real risks: emmision, immunity, distortion of signal, something else? How fast the clocks and rise time must be to get any benefits from that careful routing of differential pairs?

[@Zachariah-Peterson]

1) Length tuning is always doing something. 2) The point of length tuning is to ensure that the two signals in a differential pair are switching states at the same time. This is because differential receivers are crossing detectors, meaning they detect when the edge rates cross each other. 3) The related point of length tuning is to ensure that any common-mode noise received closer to the driver on the interconnect is also received as common-mode noise at the receiver, this way the receiver can cancel that noise at all possible frequencies. This is not achievable in practice unless the two signals are perfectly routed in parallel with zero fiber weave effect. However, you can get pretty close up to the Nyquist frequency, which is the lower limit where the receiver can no longer cancel common-mode noise. 4) The clock frequency doesn't matter, it's actually the Nyquist frequency and the rise time that matter (see #2 above for the reason why). 5) The real risk is input impedance mismatch and mode conversion if length matching is not applied properly and if the differential pairs are too close together. The length tuning section will always create mode conversion (that means conversion of common-mode noise into differential-mode noise). Your goal should be to ensure minimum possible mode conversion up to the channel's Nyquist frequency. You can read more about this here: resources[dot]altium[dot]com/p/what-impedance-length-tuning-structures

[@TheLemon22]

Hi Zach - great content as always. Does Altium ever plan to add delay tuning columns (in picoseconds) to the xSignals tab of the PCB panel? At my company we length match in delay units and unfortunately the PCB panel only lists routed length in millimetres.

[@Zachariah-Peterson]

It's true that you do need to have the length tuning tolerance performed in time, you can actually set the time requirement in the design rules. Then you are routing, it will show a design rule violation on-screen if there is a timing mismatch. If you set the rule in ps, then the tuning tool will also show the target tolerance in terms of ps.

[@TheLemon22]

@@Zachariah-Peterson hi Zach, appreciate the reply! What I'm referring to specifically is the PCB panel in Altium where you can view the nets, xSignals, and their related routed and unrouted lengths. The units for these routed and unrouted lengths are ONLY listed in length (mm) and not delay (ps) which makes seeing which of your nets in a given length-matched group are the longest in terms of actual delay very difficult to see at a glance. Sometimes due to stackup, your net with the longest routed length in mm will not always be your longest in terms of delay (say if it were routed mostly on an outer layer).

[@Zachariah-Peterson]

@@TheLemon22 If they add it I'll definitely make a video about it, I need to do something on the xSignals tool anyways!

[@ajingolk7716]

What about board to cable to board mismatch impedance differential pair configuration: Cable < board Cable > board Thank you

[@Zachariah-Peterson]

Cables that are built to interface with a standard connector involving a differential interface (like USB) are tested for their differential impedance and skew across the length of the cable to ensure it is within spec. So in practical cases, you should not expect excessive length mismatch in the cable. You only need to worry about the PCB to ensure the design has the right length matching tolerance.

[@ajingolk7716]

@@Zachariah-Peterson hi Zach thanks for your reply, so his the issue am not an electronician but am interest in making my own cable MCIO in the 5.0 gen impedance required is 85 ohm so the problem is its hard to find(balanced) , thats why am asking what would happen if i use let say 70 ohm with thos board of 85 ohm and recto versa thank you.

[@Zachariah-Peterson]

​ @ajingolk7716 Well think about it like this. The inteface requires 85 Ohms, so the traces leading into the components at each end of the interconnect should be 85 Ohms. To interface into the cable then you would need to impedance match into the cable input and output, but if this is digital signals then you need to use wideband matching. That can be difficult, simplest thing you can try is a 15 Ohms series resistor right at the input and output of the cable. A better option might be to try an RC or RLC low pass filter but it can't have an underdamped response in its transfer function. Also you have to set the cutoff frequency to very high value that exceeds the required channel bandwidth. It will require some SPICE simulation to verify the functionaliy.

[@ajingolk7716]

@@Zachariah-Peterson still you didnt answered my quetion 😔. What happen if cable > board ; board > cable

[@Zachariah-Peterson]

@@ajingolk7716 Well I didn't say what would happen but this is well known, if you do not have the board impedance matched to the connector and cable impedance you will get reflections. This is due to the input impedance mismatch looking into the cable.