Length Tuning and Differential Pairs

[@btoiscool]

I'm a designer a part of a research team at University. I'm actually in the process of training new people, I'm going to have them watch these videos when they are ready to start designing things!

[@AltiumAcademy]

Awesome! Glad you like them :)

[@richardhead8264]

Hi Zach! Thank you for the detailed explanation! The green marker is on it's last leg. Could you please obtain a new one? Thanks!

[@Zachariah-Peterson]

I guess I'm drawing too many PCBs! I'm ordering a pack on Amazon now

[@richardhead8264]

@@Zachariah-Peterson _So grateful for your response! Thanks again, Zach!_

[@thanosprionas6919]

Very good explanation! Thumbs up!

[@mitchelllague5499]

This is awesome info, thanks!!

[@electronichome1153]

Thank you for the great tutorials, sir!

[@AltiumAcademy]

Glad you like them!

[@luisabueno2360]

Is it necessary for an analog circuit with differential signals as well?

[@Zachariah-Peterson]

Yes if you have an analog circuit with a differential interface, then it will be necessary. However, it is important to note that the important factor is the phase difference between the two opposite polarity signals. Some analog circuit may be sensitive to phase differences between the two polarity signals, and the allowed distance mismatch will depend on the allowed phase difference and the frequency of the opposite polarity signals.

[@aitorsierra]

Altium Academy: How do you calculate the exact value of the length ?. Is it possible to make a video showing it in Altium ?. Thanks

[@Zachariah-Peterson]

Hey Aitor, thanks for watching, I'm going to film this today, stay tuned for the new video!

[@beamray]

wait a second, there is no automatic or semi-automatic way to even phases in diff pair in AD in correct place, tuning will add all length you need in some (any) place, but not right one. to make phase stable ypu will need to do it manually. Am I wrong?

[@Zachariah-Peterson]

Hi Egor, There is no button you can click that will automatically equalize everything and automatically put the perfect length tuning structure into the PCB layout. What I mean is, you can use a tool to click and drag across the trace you need to equalize, and it will automatically create the length matching segment for you, so you don't need to manually layout each section of a serpentine route (or any other length tuning structure for that matter). It will also automatically calculate the time mismatch as you do this, you won't have to check lengths and propagation delays and then manually determine the time mismatch as you add in length matching sections, you'll be able to see when you have minimized the time mismatch visually thanks to this automatic calculation. You do need to specifically select the place to put the length matching segment because the best spot to place it could be slightly different in every layout. It's about as "automatic" as you can get!

[@pigglewiggle175]

So does this mean that you can never cancel out all common mode noise? You will never perfectly align the traces length, nor is the temperature of the board uniform and hence one trace will have electrons that travel slightly faster. Maybe for practical purposes it is inconsequential but is true separation of common mode noise something that can only be achieved in theory?

[@Zachariah-Peterson]

I think that's a fair statement. Even if you did have perfect differential pairs with aboslutely perfect phase matching up to infinite frequency, differential receivers do not have 100% common mode noise rejection. It's pretty high but not infinite dB common mode noise reduction. I think the key here is you try to get it as perfect as possible without going too crazy on length tuning, but do it with the knowledge that there will be something else in the system that creates additional common mode noise and/or extra skew.

[@waleedarshad8160]

How important is coupling between P-N if you have let's say a loosely coupled differential pair? What if I have to uncouple these in the BGA area over a long distance due to routing constraints? Would that work?

[@Zachariah-Peterson]

This is a great question! We have an advanced differential pair concepts video coming up, I will make sure to address this question as we will be talking more about coupling.

[@Zachariah-Peterson]

I'll give you the short answer here: The coupling requirement depends on the differential impedance target. You can have some slightly uncoupled impedance over a short length, such as when you're routing under a BGA. Those components tend to put pins for the high speed differential interfaces around the outside of the BGA so that you don't have to worry about this problem so much, but yes sometimes you do have to route between pins and uncouple them for a short distance. It's okay though, this would create a differential impedance discontinuity, but if it is short enough then the input impedance will be affected very little.

[@waleedarshad8160]

@@Zachariah-Peterson thanks for the explanation, can't wait for the next video!!

[@asingh4152]

"Thanks for tuning in" I see what you did there.

[@Zachariah-Peterson]

🙃

[@ehsanbahrani8936]

Thank you ❤

[@Dr.Bigglesworth]

Hi Zach. Your first 4 minutes are what I have always called "Static Phase Tolerance", which as you say, is specific to the receiver (end of the diff pair), and has to do with the "eye" or amount of time the receiver "sees" a valid signal. In-between that will be noise, which decreases the eye. After that, you start to talk about what I have always called "Dynamic Phase Tolerance", which is maintaining the wave front along the two differential signals, and is specifically, as you say, for common-mode noise rejection. With dynamic phase tolerance, you need to adjust each trace to maintain a specified difference in length all along the entire length of the differential pair. Those minor length "adjustments" should be done at intervals along the diff pair routing to maintain the length matching is maintained along the entire length. I would strongly urge to not use the moose antler shaped routing you showed, as it does not maintain differential impedance. Plus, if that much is needed, and you've added it all in one place, that means the dynamic phase tolerance is WAY off all along the rest of the diff pair routing, which compromises the common mode noise rejection. I believe Altium only supports Static Phase Tolerance, as in, gives an overall distance comparison of the two legs of the diff pair. It does not give any readout along the diff pair or where along the diff pair it is out of Dynamic Phase Tolerance. If I'm wrong about this, I'd love to know, but it's a feature that IMO Altium needs to add. Other tools do have this. In Altium, I just look at the routing, and add the "adjustments" just after spots that cause the length mismatch (typically after a turn that doesn't have a corresponding turn in the opposite direction right after). All of this and how particular you are with your differential routing depends on signal level and frequency. Since I am typically in the many GHz, this can mean the difference between working and not. But, it should all be spec driven by the design requirements. Length matching across a group of 1 MHz signal to within 10 mils is typically a waste of effort. Understanding when design rules make sense is in my opinion very important.

[@Zachariah-Peterson]

You're not wrong but there is some context needed. The serpentine length tuning can create large deviations in differential impedance that you mention, they can also create lots of mode conversion. But this is only in the condition where the traces are already very close together (spacing

[@wilfredswinkels]

no there is no mis match. the induced noise cannot come later because it's from the same source. the noise is on the edge of the lower signal but should subtract in a similar way.

[@Zachariah-Peterson]

Noise can absolutely come later and does not have to originate from the differential driver. Most common is from crosstalk or ISI. Another possibility is mode conversion, although we left it out of this video as it is a more advanced topic. With crosstalk into a differential pair, there is both a common mode and differential component due to the different magnitudes of the noise signal on each trace in the differential pair. And noise only subtracts if the noise is common-mode, differential mode noise does not subtract. In general, any noise signal can be decomposed into common mode and differential components.

[@SeNoGERfreak]

Length matching a differential pair is not a critical thing. Watch this, what Rick Hartley says about the topic: kzbin.info/www/bejne/h3iTcqOloZKioJI So why should i do this at all?

[@Zachariah-Peterson]

It is critical, but not always for the reasons people think. There are three reasons for length tuning: noise suppression, timing skew, and mode conversion suppression. If you just compensate timing within the rising edge, the channel will work fine and you can be really loose with the length tuning tolerances (assuming long rise times), that is generally what Rick and Lee refer to. In fact, if you see a distance-based tuning tolerance in a datasheet, then it's most likely bogus. What people rarely bring up is noise suppression and mode conversion. Noise suppression is less important than mode conversion because differential receiver circuits have decent CMRR, and as long as you do everything else correctly then common-mode noise won't be major. But with mode conversion, it does not take much mode conversion within your signal bandwidth to cause channel compliance/EMC failure.