Fun fact, in 2007, you could re-load the video and give as many thumbs up as you wanted. A small group of us were manipulating the system to bring obscure videos to the top. That ability ended shortly after. They tried to convince us to stop then they fixed the bug. Still gives me the chuckles.
Hey Robert, Thanks of this video. I love seeing your interviews with these experts and find it help full that you sometimes pause the video to explains stuff. Thanks again.
Funny thing is, I started in electronics by making high frequency FM radio transmitters on breadboards (something thought of as impossible by RF engineers). I'd be making 400MHz+ transmitters on normal solderless breadboards before I realised the issues of parasitics. After learning a bit more about it the fundamentals, I could never figure out how to make a transmitter work above about 60MHz on a breadboard. The joys of learning how to control things properly in electronics, and never being able to make crazy transmitters again! 😆 Looking at the examples in this video really takes me back to my early days. I'm so glad I learned about some RF stuff first, because it helped me understand a lot about AC theory. It's clarifying it all. Your videos are filling in MANY knowledge gaps for me! Thanks so much to you and Eric. Really appreciate it.
Thank you for this video Robert, maybe this is a level above my knowledge - I also tend to use my 'instincts' and experience when assuming the better approach and usually it comes out well but It's always good to know WHY and HOW exactly physics of these phenomenons work.
Ah, yes! This is exactly what I was wondering, different harmonic effects of the crosstalk depending on the method used! And time domain width is inverse to f fomain width, so we see the short "trombone" peak has spread its frequency peak, and the low, wide precursor for the"river" has the narrow GHz peak!
Microstrip Accordion routing at around 1:00:00 Bloch effect punishes the design at 9 GHz. What if, instead of equal spacing of the serpentine traces, one used a series of slightly different widths? For instance different spacing between vertical legs of 1.0L, 1.05L, 1.10L, 1.15L, 1.2L, 1.25L, 1.3L, and so on.
The plots discussed around 53:55 (microstrip) and 1:05:25 (stripline) are very interesting. The trombone curve is especially striking. If we see the trombone as having three legs (Input --> A --> B --> C --> output), then each leg would be 5cm, with propagation delay about 330ns. We can see that for both microstrip and stripline, the "precursor" bump starts about 0.33ns delayed from the input step. Why? Presumably because the crosstalk from the input end of A to the output end of B starts immediately. Any forward signal from B into C starts immediately too, as does crosstalk from output end of B across to input end of C. So it only requires conduction from begin to end of C until the output starts to rise, though only up to a small plateau. Interestingly, the plateau for stripline is higher than for microstrip. I think this is because for stripline, the A-to-B crosstalk is all antiparallel in the forward direction of B, which outputs from B into C, and the crosstalk from B to C is also antiparallel, so in the forward direction of C. So the two contributions to a signal in C add. However, in the microstrip case, the crosstalks from A to B, and B to C, have parallel and anti-parallel components, partially cancelling out the effect, and/or simply sending energy in the reverse direction, hence a lower precusor plateau, but with the same timing. Very neat!
9:20 regarding Far-end and near-end crosstalk. First, for purposes of discussing meandering traces, it would be helpful to call these crosstalks something like "parallel" and "antiparallel", or "same direction" and "opposite direction". Robert mentions that for stripline far-end crosstalk (ie: parallel) "they cancel". Evidently, "they" refers to the two crosstalk components attributable to electric and magrnetic fields, which happen to be equal for stripline and opposite for the "same direction" direction, and presumably when dimensions and materials are selected for 50-ohm impedance. I guess in microstrip they don't cancel.
Just one request Robert you have to let him finish before asking some questions otherwise write it down bcz he is explaining something and you interrupt him then sometimes he does not continue with that same point apart from that thanks for this uploads learning a lot from guys like you and Eric.
Great video. I was wondering what would be better in the case of differential pair routing. I would think trombone would be better in the case where you need to compensate for a large length mismatch as the distance that the differential pair travels uncoupled would be shorter than in the case of river length matching. Also, if I look just at the transient analysis of the last simulation with 40ps rise time, it seems as though trombone would be superior to river in the case of digital signals as long as it doesn't cause the receiving device to register a 1 at that low initial voltage. Although this is just a single edge. Maybe it would be different if there was an actual high frequency block signal that was simulated as that more closely resembles an actual digital signal. I myself use trombone for very large length mismatch, river for medium length mismatch and triangle for small length mismatch. I do of course keep a reasonable distance between the segments.
Great video, Robert! As always. Potentially, a hand drawn meander, with recognizable variances in per-wave coupling and variances in segment length could perform better with flatter frequency response. Further, since the typical meander is generated by computer, AKA Altium et al, all of the dimensions of the meander could easily be made to vary randomly within preset tolerances to maximally harvest this flattening effect. Just a thought 🤷♂
50:51 14dB loss @ 20G for stripline vs 12dB for microstrip. Microstrip does dip down to 14dB at one point though. The more consistent stripline behavior comes at a cost of attenuation.
I found this video very well worth watching. An interesting follow-up would be if you could produce a video which compared some of the results of the simulations with results measured from actual PCBs produced using the three type of transmission lines.
🤔 around 59:00 - if the river would have been a "broadband"/"spread frequency" layout (don't remember seeing option for that in any software so far)? So, like the period of the river starts a bit smaller/narrower, increasing with each period and ending in a bit bigger/wider period. That should spread the resonance frequency peaks approximately by the same "a bit" in width and reducing in strength/height, no?
Filling in some more gaps: With the total length of the interconnect being 15 cm, nominally 1 ns delay, the round trip is 2 ns, so the resonant frequency would be 0.5 GHz (assuming imperfect impedance match at the terminations) , and that is in fact the spacing of the peaks in the reflection plot for the straight trace, and also the closely-spaced peaks in the River plot. Nice! Then.. ... for the Trombome trace, the lengths are 5 cm per leg, nominally 0.33 ns, or 0.66ns round trip, or 1.5 GHz. That is indeed the peak spacing on the reflection plot. Finally, for the River trace, the individual segments are 1/16 of 15 cm, so we'd expect a resonance at 16 times the frequency of the total trace, or about 8 GHz. That's a bit off from the simulation which shows 9 GHz, but maybe that's accounted for by the factor that each leg is made from a coupled section and a non-coupled section, or just be the coupling of one leg to another. All these ways in which you can see physical dimensions correlate to time and frequency really demystify the subject in a very satisfying way. Thanks Robert and Eric!
Thanks for the very informative video. 13:50 In order to have not much coupling spacing(width between serpentines) should be roughly 3 or 4 times track width @50 ohm line impedance. What about height of the serpentines should be? 1:17:00 Rough general rule under 10gig frequencies many short coupled segments(serpentines or meanders) are better. What ratio of the width and height of the serpentine is short according to track width? Or according to reference plane?
Eric very briefly touched on Bloch frequency being related to semiconductor band gaps - I would absolutely *LOVE* to see him talk about that in more detail. I tried digging up more details about it but the Wikipedia articles on Bloch's theorem, Bloch oscillation, and FSS / Bloch wave - MoM methods were completely unreadable and seemed to be targeted at solid state physicists.
That was a great video. It's really interesting to see the underlying reasons as to why the rules for serpentine track layout are the way they are. Normally, only those few engineers lucky enough to have access to super-expensive field solver software ever get this insight. I do have one question though: Why does Eric say (@9:00) that Microstrip suffer from far-end crosstalk (FEXT), but Stripline does not? I don't understand why the crosstalk gets cancelled out simply by using stripline. What difference does that make?
Thank you Nicholas. PS: I don't remember the exact minute, but the crosstalk cancellation is explained in this video: ruclips.net/video/EF7SxgcDfCo/видео.html
@@RobertFeranec Thank you, it all makes perfect sense now! It's all down to the symmetry or asymmetry of the dielectric constant of the material above and below the trace, and the fact that the electric field interacts with the dielectric material, but the magnetic field doesn't.
You are correct - stripline does NOT eliminate FEXT entirely. Odd and even mode phase velocities are equal in homogeneous stripline and unequal in microstrip, which reduces the directivity of coupling at higher frequencies in the microstrip case. A quick ADS or HFSS simulation will show you that FEXT does exist in coupled striplines. An example of a little knowledge being dangerous, I think.
I was directed to this video after asking some questions about ST Micro's AN5122 appnote, where they recommend trombone meanders (aka switchback) instead of accordion meanders (aka river or serpentine) for LPDDR3 traces, claiming that accordion meanders "provoke orthogonal propagation which compromises signal integrity". This was surprising to me, because I had always heard that trombone routing was more suitable for lower frequency stuff. After watching this video I feel I understand the behaviours of the meanders a lot better, but I'm still left wondering what they meant. By "orthogonal propagation" I can only guess that they're perhaps referring to the propensity for the mutual crosstalk to cause the precursor to advance faster over accordion meanders, by "jumping" the legs, but it wasn't clear to me that it's any better for trombone meanders. Maybe they're worried that the higher Q of the resonance in accordion routing coupled is more likely to create a large resonance at the fundamental frequency of the rising edges for LPDDR3 on typical 50Ω impedance stackups? Still seems a bit odd to me - maybe it's just a bad recommendation? (they do also mandate a 4L SGPS stackup, which made me roll my eyes a bit)
I wanna see group delay too, i bet the group delay on the river will be shorter as freq rises And trombone gets alot shorter in even lower freq then river (You can tell ADS to calculate group delay Via the SP contoller settings)
What would happen if the meanders were less regular? e.g, _--___-__---_ so that the crosstalk from one hill would be at a different resonance than the crosstalk enabled by the next hill, instead of all 15 hitting the same frequency.
No far end cross talk in stripline?! When did that come in? Whilst it's true to say that the odd and even mode phase velocities are equal in stripline, which is a true TEM structure and unequal in miscrostrip, which is quasi-TEM - it is NOT TRUE to say that there is no far end cross talk in stripline. A quick ADS or HFSS simulation will show you that.
Eric always amazes me how he rationally explains such complex, non-obvious phenomena... Robert's questions also help understanding. Bravo to both!
These two make an outstanding team!
I wish I could thumbs-up this video more than once. Well done, Eric & Robert. Your partnership is making the world better.
Thank you very much Bart
Fun fact, in 2007, you could re-load the video and give as many thumbs up as you wanted. A small group of us were manipulating the system to bring obscure videos to the top. That ability ended shortly after. They tried to convince us to stop then they fixed the bug. Still gives me the chuckles.
1:17:41 for the 10 seconds rule-of-thumb for those who don't care about the juicy details 😋
Hey Robert, Thanks of this video. I love seeing your interviews with these experts and find it help full that you sometimes pause the video to explains stuff. Thanks again.
Thanks a lot Robert ❤
Funny thing is, I started in electronics by making high frequency FM radio transmitters on breadboards (something thought of as impossible by RF engineers). I'd be making 400MHz+ transmitters on normal solderless breadboards before I realised the issues of parasitics. After learning a bit more about it the fundamentals, I could never figure out how to make a transmitter work above about 60MHz on a breadboard. The joys of learning how to control things properly in electronics, and never being able to make crazy transmitters again! 😆
Looking at the examples in this video really takes me back to my early days. I'm so glad I learned about some RF stuff first, because it helped me understand a lot about AC theory. It's clarifying it all. Your videos are filling in MANY knowledge gaps for me! Thanks so much to you and Eric. Really appreciate it.
Thank you for this video Robert, maybe this is a level above my knowledge - I also tend to use my 'instincts' and experience when assuming the better approach and usually it comes out well but It's always good to know WHY and HOW exactly physics of these phenomenons work.
Thank you again Eric & Robert ..also excited like the same when sir is explaining...
Ah, yes! This is exactly what I was wondering, different harmonic effects of the crosstalk depending on the method used! And time domain width is inverse to f fomain width, so we see the short "trombone" peak has spread its frequency peak, and the low, wide precursor for the"river" has the narrow GHz peak!
Hi Robert, I like all your videos but those with Eric are some of my favorites!!
Thank you very much. This was a super cool video.
Great explanation by the champion of Signal integrity simulations
thank you for this kind of good contents you provide. Its very helpful.
Microstrip Accordion routing at around 1:00:00
Bloch effect punishes the design at 9 GHz.
What if, instead of equal spacing of the serpentine traces, one used a series of slightly different widths?
For instance different spacing between vertical legs of 1.0L, 1.05L, 1.10L, 1.15L, 1.2L, 1.25L, 1.3L, and so on.
The plots discussed around 53:55 (microstrip) and 1:05:25 (stripline) are very interesting. The trombone curve is especially striking. If we see the trombone as having three legs (Input --> A --> B --> C --> output), then each leg would be 5cm, with propagation delay about 330ns. We can see that for both microstrip and stripline, the "precursor" bump starts about 0.33ns delayed from the input step. Why? Presumably because the crosstalk from the input end of A to the output end of B starts immediately. Any forward signal from B into C starts immediately too, as does crosstalk from output end of B across to input end of C. So it only requires conduction from begin to end of C until the output starts to rise, though only up to a small plateau.
Interestingly, the plateau for stripline is higher than for microstrip. I think this is because for stripline, the A-to-B crosstalk is all antiparallel in the forward direction of B, which outputs from B into C, and the crosstalk from B to C is also antiparallel, so in the forward direction of C. So the two contributions to a signal in C add.
However, in the microstrip case, the crosstalks from A to B, and B to C, have parallel and anti-parallel components, partially cancelling out the effect, and/or simply sending energy in the reverse direction, hence a lower precusor plateau, but with the same timing. Very neat!
Thank you very much for this video! Very informative and listening to prof Bogatin is a pure pleasure!
9:20 regarding Far-end and near-end crosstalk. First, for purposes of discussing meandering traces, it would be helpful to call these crosstalks something like "parallel" and "antiparallel", or "same direction" and "opposite direction". Robert mentions that for stripline far-end crosstalk (ie: parallel) "they cancel". Evidently, "they" refers to the two crosstalk components attributable to electric and magrnetic fields, which happen to be equal for stripline and opposite for the "same direction" direction, and presumably when dimensions and materials are selected for 50-ohm impedance. I guess in microstrip they don't cancel.
Great content. Definitely, you make us better engineers at work, and now we can take more informed decisions in design. Thank Eric and robert
Right questions, clear answers. Thank you for your work!
Thank you for your comment and feedback.
Just one request Robert you have to let him finish before asking some questions otherwise write it down bcz he is explaining something and you interrupt him then sometimes he does not continue with that same point apart from that thanks for this uploads learning a lot from guys like you and Eric.
Was waiting for this one! Thanks!
Great video. I was wondering what would be better in the case of differential pair routing. I would think trombone would be better in the case where you need to compensate for a large length mismatch as the distance that the differential pair travels uncoupled would be shorter than in the case of river length matching.
Also, if I look just at the transient analysis of the last simulation with 40ps rise time, it seems as though trombone would be superior to river in the case of digital signals as long as it doesn't cause the receiving device to register a 1 at that low initial voltage. Although this is just a single edge. Maybe it would be different if there was an actual high frequency block signal that was simulated as that more closely resembles an actual digital signal.
I myself use trombone for very large length mismatch, river for medium length mismatch and triangle for small length mismatch. I do of course keep a reasonable distance between the segments.
Thanks for putting this out.
What a great informative video!
Great video, Robert! As always.
Potentially, a hand drawn meander, with recognizable variances in per-wave coupling and variances in segment length could perform better with flatter frequency response. Further, since the typical meander is generated by computer, AKA Altium et al, all of the dimensions of the meander could easily be made to vary randomly within preset tolerances to maximally harvest this flattening effect. Just a thought 🤷♂
This is awesome! I remember working at…. This stuff was new and top secret lol. Love your information…
Thank you sir for giving us these kind of knowledge by making videos. Keep it up your good work. It would be very helpful to us. 👍
50:51 14dB loss @ 20G for stripline vs 12dB for microstrip. Microstrip does dip down to 14dB at one point though. The more consistent stripline behavior comes at a cost of attenuation.
Also on the transmitted signal graph -12dB for microstrip and -14dB for stripline @20GHz looks like there are more losses.
I found this video very well worth watching. An interesting follow-up would be if you could produce a video which compared some of the results of the simulations with results measured from actual PCBs produced using the three type of transmission lines.
🤔 around 59:00 - if the river would have been a "broadband"/"spread frequency" layout (don't remember seeing option for that in any software so far)?
So, like the period of the river starts a bit smaller/narrower, increasing with each period and ending in a bit bigger/wider period.
That should spread the resonance frequency peaks approximately by the same "a bit" in width and reducing in strength/height, no?
Filling in some more gaps: With the total length of the interconnect being 15 cm, nominally 1 ns delay, the round trip is 2 ns, so the resonant frequency would be 0.5 GHz (assuming imperfect impedance match at the terminations) , and that is in fact the spacing of the peaks in the reflection plot for the straight trace, and also the closely-spaced peaks in the River plot. Nice! Then..
... for the Trombome trace, the lengths are 5 cm per leg, nominally 0.33 ns, or 0.66ns round trip, or 1.5 GHz. That is indeed the peak spacing on the reflection plot.
Finally, for the River trace, the individual segments are 1/16 of 15 cm, so we'd expect a resonance at 16 times the frequency of the total trace, or about 8 GHz. That's a bit off from the simulation which shows 9 GHz, but maybe that's accounted for by the factor that each leg is made from a coupled section and a non-coupled section, or just be the coupling of one leg to another.
All these ways in which you can see physical dimensions correlate to time and frequency really demystify the subject in a very satisfying way. Thanks Robert and Eric!
Great content as always! Thanks for sharing with us! 😄
Really enlightening. Thank you
Thanks for the very informative video.
13:50 In order to have not much coupling spacing(width between serpentines) should be roughly 3 or 4 times track width @50 ohm line impedance. What about height of the serpentines should be?
1:17:00 Rough general rule under 10gig frequencies many short coupled segments(serpentines or meanders) are better. What ratio of the width and height of the serpentine is short according to track width? Or according to reference plane?
Thank you Eric!
Eric very briefly touched on Bloch frequency being related to semiconductor band gaps - I would absolutely *LOVE* to see him talk about that in more detail. I tried digging up more details about it but the Wikipedia articles on Bloch's theorem, Bloch oscillation, and FSS / Bloch wave - MoM methods were completely unreadable and seemed to be targeted at solid state physicists.
Thanks for the video. Great content.
That was a great video. It's really interesting to see the underlying reasons as to why the rules for serpentine track layout are the way they are. Normally, only those few engineers lucky enough to have access to super-expensive field solver software ever get this insight.
I do have one question though: Why does Eric say (@9:00) that Microstrip suffer from far-end crosstalk (FEXT), but Stripline does not? I don't understand why the crosstalk gets cancelled out simply by using stripline. What difference does that make?
Thank you Nicholas. PS: I don't remember the exact minute, but the crosstalk cancellation is explained in this video: ruclips.net/video/EF7SxgcDfCo/видео.html
@@RobertFeranec Thank you, it all makes perfect sense now! It's all down to the symmetry or asymmetry of the dielectric constant of the material above and below the trace, and the fact that the electric field interacts with the dielectric material, but the magnetic field doesn't.
You are correct - stripline does NOT eliminate FEXT entirely. Odd and even mode phase velocities are equal in homogeneous stripline and unequal in microstrip, which reduces the directivity of coupling at higher frequencies in the microstrip case. A quick ADS or HFSS simulation will show you that FEXT does exist in coupled striplines. An example of a little knowledge being dangerous, I think.
Very nice video, thank you !!!!!!!
great content. i also guessed that the short leg version performs better.
This content is great!
This is the really cool stuff!!!!!
thanks, great talk!
This video is great.
Thank you for the video :)
Thank you!
A very informative video, though as always! Thank you, Robert for your work!
Very interesting, great analysis, Have you just discovered a PCB filter?
I was directed to this video after asking some questions about ST Micro's AN5122 appnote, where they recommend trombone meanders (aka switchback) instead of accordion meanders (aka river or serpentine) for LPDDR3 traces, claiming that accordion meanders "provoke orthogonal propagation which compromises signal integrity". This was surprising to me, because I had always heard that trombone routing was more suitable for lower frequency stuff. After watching this video I feel I understand the behaviours of the meanders a lot better, but I'm still left wondering what they meant. By "orthogonal propagation" I can only guess that they're perhaps referring to the propensity for the mutual crosstalk to cause the precursor to advance faster over accordion meanders, by "jumping" the legs, but it wasn't clear to me that it's any better for trombone meanders. Maybe they're worried that the higher Q of the resonance in accordion routing coupled is more likely to create a large resonance at the fundamental frequency of the rising edges for LPDDR3 on typical 50Ω impedance stackups? Still seems a bit odd to me - maybe it's just a bad recommendation? (they do also mandate a 4L SGPS stackup, which made me roll my eyes a bit)
“All app notes should be assumed wrong until proven otherwise”
- Rick Hartley
Really good video. While what Fringe Field effect makes the effective dielectric constant smaller so leads to trombone and meander delays? Thanks. Br.
Do the shorter rivers legs means little spacing between the legs(horizontal direction) or just smaller legs(vertical direction)?
I wanna see group delay too, i bet the group delay on the river will be shorter as freq rises
And trombone gets alot shorter in even lower freq then river
(You can tell ADS to calculate group delay Via the SP contoller settings)
What would happen if the meanders were less regular? e.g, _--___-__---_ so that the crosstalk from one hill would be at a different resonance than the crosstalk enabled by the next hill, instead of all 15 hitting the same frequency.
What we heard from SI engineers that Trombones are better than accordians in Skew point of view. Is that true?
What’s the URL link to the masters degree?
I asked Eric, here it is: www.colorado.edu/ecee/academics/graduate-programs/professional-masters/high-speed-digital-engineering-coming-soon
As summary what is the answer
No far end cross talk in stripline?! When did that come in? Whilst it's true to say that the odd and even mode phase velocities are equal in stripline, which is a true TEM structure and unequal in miscrostrip, which is quasi-TEM - it is NOT TRUE to say that there is no far end cross talk in stripline. A quick ADS or HFSS simulation will show you that.
Hyperlynx is awesome... PADS as a layout tool is soooo archaic :(
Схемотехника выше 10ГГц - является магией.
Hi
Spoiler alert to viewers! If you watch this video, but can't afford newfangled field solver software, you're gonna feel like a "Meanderthal".
Someone please Tell me in comment !!!🙄I not have enough data