saved a vid for the technical information on the parameters of layer structure. Needed this recently. You can google your head off finding out this kind of stuff. Thank you Robert, really helpful.
Robert, I love the content. This is one of the few really great RUclips channels that perfectly helps young engineers such as myself improve our designs. I've learnt so much about signal integrity on PCBs from you. I was wondering if you have any videos planned for signal integrity for signals leaving the PCB through different types of connectors and wires.
Thanks you for this video; literally eye-opening information. I'm sure this will also be useful for boards using the USB 3.2 Gen 2x2 standard, at 20 GBps. GDDR6x SDRAM Memory (as used on the RTX 3090) uses PAM4 signalling at 21Gbps, which is also a significant challenge for PCB designers.
Great video, thank you. PAM4 is widely used in optical transceivers for ethernet (QSFP, SFP, OSFP...). The host side would be PAM4 and the line side would be an optical (fiber) signal. And then when PAM4 is not enough there is QPSK and QAM16, these are being used for Coherent optical networks for 400Gbps ethernet. I am not an expert, but I have some experience working for a high speed test and measurement company (TDR, DSO, BERT ...)
Is the 270mil stub length shown at 6:00 split across multiple vias? Even accounting for the reflection going down and back up the via, it seems unlikely that you'd have a single via that long.
I was thinking the same, but, some PCBs really are very thick. I googled more info about backplanes and for example here it says " typically 4.0-6.0mm, sometimes even up to 10.0mm." www.pcbway.com/blog/Engineering_Technical/PCB_backplane.html
Yeah, it's funny how you can work with something like PCBs for decades, and get accustomed to say 0.6 to 2.0mm thicknesses, and then suddenly run into 6mm board thickness and it beggars belief. But apparently it's just more of the same, with the exception of vias now becoming an appreciable length to worry about.
You are absolutely right! Thank you. I have been reading and pronouncing that word wrong for years, and nobody has pointed it out. I have never heard it pronounced before and somehow I saw the letters in the wrong order. Now the problem will be to unlearn that pronunciation in the future. lol!
@@bertsimonovich7297 And you've probably realized it by now, but just in case not: the device measures a profile, hence it's a profile-o-meter.... hopefully that makes it easy to remember!
Nice...!! I was looking for the time to watch this video. I still have to watch a couple more videos. maybe if you separate them into parts it will be easier to find the time to see them. It's just a tip and you may have more views per video. Thank you very much for the content you make. It is simply excellent.
Can someone explain why Frequency = Baud / 2 @ 10:30? Does this have to do with Nyquist? I dont not see the correlation between nyquist (which has to do with sampling) and bit rates. :)
What simulation methods are there to simulate Fiber weave effect? Is there a standard skew in ps per length that we can use or something? Do a Monte Carlo or something like that by adding/subtracting skew? Or other thoughts?
In the skew discussion he mentioned that the Nyquist was at 14GHz, but that would be roughly the nyquist of an NRZ signal. With PAM4, the nyquist is actually 26Gbits/4bits=~8GHz?
Hmmm, if it's 26Gbits/sec (which is the same as 26Gbaud), and it's a 4- level code (2-bits per time slot), that means 13G time-slots/sec (or state-changes/sec). So the wave frequency is half that: (6.5 GHz). And the Nyquist frequency required to sample 6.5GHz, would be twice that... bringing is to 13GHz.
@@bertsimonovich7297 thank you, Bert. Your work is phenomenal. I always have problems with modeling of surface roughness of copper foil. This staff in this video and your works are great.
Actually I just found out ADS starting with version 2022 includes the full Huray model so you can find the necessary parameters using my Cannonball model
Hi Robert, small tip. Please dont interrupt the Speaker. Ask your question only after the other speaker has finished. Wearing headphones really helps, then u r conscious of the other speaker and wont speak over and interrupt them
Of the various technical youtube channels I watch that do interviews, I would say that Robert inserts questions relatively unobtrusively. In fact, sometimes I get the impression that Robert has found a point unclear, and waits quite a long time to bring it up. And sometimes he even waits to the editing stage to insert a pause and clarification. It's also difficult to wait for the speaker to "finish" because, unlike in an in-person interview, the speaker on zoom or skype may not pick up a non-verbal cue from the inteviewer to pause for a question. As a result, a certain amount of "interrupting" is inevitable, and actually welcome, because Robert is interceding on behalf of the audience. No doubt all interviewers can learn things to improve their technique, but I think Robert does well, and a simplistic suggestion like "wear headphones" is not helpful -- he's obviously listening to the speaker already.
@@Graham_Wideman Thank you very much Graham. Yes, I was not intentionally commenting on ravi's comment as .... in many situations I have to interrupt. I know, it is not polite and I know, sometimes I cut out an interesting sentence, but otherwise the videos would be too long or not about the topic. Especially I interrupt when I see that my question was already answered or if the talk is going away from what I would like to focus on. If I would not interrupt, the talk would go sideways and videos and recordings would be too long or very hard to edit (recordings are often between 2-3 hours long, so even now it is a lot of work to record and then edit). Also, I interrupt, when I have a question in mind and I would like to ask it before we start talking about something else. PS: I often tell people before our call, that I will interrupt them, so they are aware of that (and they are ok with it, actually, they are happy about it, many are aware, that if I do not stop them, they would keep talking about the topics they love).
The specialist guest is showing graphs that go up into tens of GHz. How much relevance does this have to a person doing PCB designs that uses DDR3/4, PCIe gen 4 or Ethernet PCBs?
Do you have experience with differential pairs and trace length matching? Which effect could have a greater impact in my signal integrity, bumps on one line which influence my impedance or the asymmetry of the line lengths? Thanks in advance
Like everything with signal integrity, it depends on your noise or skew budget. Skew budget will be a function of your data rate. Matching trace lengths helps to mitigate skew, and matching impedance mitigates reflections and noise. We usually budget about 0.2UI for skew from all sources in the channel before it starts to degrade the eye, where UI is unit interval of the baud rate. So for example, PCIE Gen5 (32 GT/s) the UI is only 31.25 ps and total skew budget from all sources is 6.25 ps. With a propagation delay of ~165ps/inch for the dielectric, that is 38mils (~1mm). But as I explained in video, there is fiber weave effect skew, that is harder to predict and other sources of skew to consider in total budget. Skew also translates into mode conversion which could exacerbate EMI compliance. Because PAM-4 SNR is -9.5dB worse than NRZ, everything matters. Only good modeling and simulation can quantify the different effects.
maybe someone could talk about wiring between/inside equipment (non-differential), should there be multiple grounds in ribbon? plain twisted pairs for each signal? won't they create ground loops? what is worse ground loop or coupling between signals if ground is shared by multiple signals? always wanted to know answer...
You should have a few videos on the basics of RF engineering. I fact all high speed pcb designs should. Concepts like short circuited stubs are introduced in the first week. Gotta ask too, a 6 mm long via? How many layers was that board?
That’s a lot of layers. Definitely have to include those parasitics. In our millimeter wave designs we must create EM models for our different vias to design effectively.
Thank you Bert and Robert for taking the time and effort to explain how these things can affect high speed board design.
Thank you MrMike
@@RobertFeranec how much do all these industry expert speakers charge you for the presentations?
Thank you for putting together this video. I enjoyed working with you on this project!
Thank you Bert for finding time to help me with this video.
Thanks Bert for your time and effort!
@@Graham_Wideman You are welcome. Glad you found it helpful.
saved a vid for the technical information on the parameters of layer structure. Needed this recently. You can google your head off finding out this kind of stuff. Thank you Robert, really helpful.
I see the first 1:40 of the video and I know I NEED to give it a like. Super interesting stuff, Robert, thank you!!!
Thank you very much guillep2k
I really enjoy these high speed and power integrity application discussions! Keep them coming!
Thank you very much Skyler
Another spectacularly useful and informative video Robert!
Robert, I love the content. This is one of the few really great RUclips channels that perfectly helps young engineers such as myself improve our designs.
I've learnt so much about signal integrity on PCBs from you. I was wondering if you have any videos planned for signal integrity for signals leaving the PCB through different types of connectors and wires.
Thank you very much Rohit
I wish there would be a whole series on Hyperlynx created by someone.
Thank you very much Robert and Bert! Excellent information for the ones doing high-speed designs.
Thanks you for this video; literally eye-opening information.
I'm sure this will also be useful for boards using the USB 3.2 Gen 2x2 standard, at 20 GBps.
GDDR6x SDRAM Memory (as used on the RTX 3090) uses PAM4 signalling at 21Gbps, which is also a significant challenge for PCB designers.
Thank you Nicholas. PS: Eye opening :D
... eye-opening... hahaha
that is wild that even the Fiberglass used impacts the Signals, and impacts it in multiple ways at once.
one of your best videos. thank you
Great video, thank you.
PAM4 is widely used in optical transceivers for ethernet (QSFP, SFP, OSFP...). The host side would be PAM4 and the line side would be an optical (fiber) signal.
And then when PAM4 is not enough there is QPSK and QAM16, these are being used for Coherent optical networks for 400Gbps ethernet.
I am not an expert, but I have some experience working for a high speed test and measurement company (TDR, DSO, BERT ...)
Thank you Hisham
Is the 270mil stub length shown at 6:00 split across multiple vias? Even accounting for the reflection going down and back up the via, it seems unlikely that you'd have a single via that long.
I was thinking the same, but, some PCBs really are very thick. I googled more info about backplanes and for example here it says " typically 4.0-6.0mm, sometimes even up to 10.0mm." www.pcbway.com/blog/Engineering_Technical/PCB_backplane.html
@@RobertFeranec Ah, yeah, on a re-watch I realised he's talking about backplanes with huge layer counts. That makes more sense.
Yeah, it's funny how you can work with something like PCBs for decades, and get accustomed to say 0.6 to 2.0mm thicknesses, and then suddenly run into 6mm board thickness and it beggars belief. But apparently it's just more of the same, with the exception of vias now becoming an appreciable length to worry about.
A profilometer (pro-file-ometer) is the tool he mentioned; I think he pronounced it as "prolifometer" by accident.
Thanks for commenting on that -- I was pretty sure that profil-ometer was what he was trying to say, but it wasn't what I was hearing!
You are absolutely right! Thank you. I have been reading and pronouncing that word wrong for years, and nobody has pointed it out. I have never heard it pronounced before and somehow I saw the letters in the wrong order. Now the problem will be to unlearn that pronunciation in the future. lol!
@@bertsimonovich7297 And you've probably realized it by now, but just in case not: the device measures a profile, hence it's a profile-o-meter.... hopefully that makes it easy to remember!
Nice...!! I was looking for the time to watch this video. I still have to watch a couple more videos. maybe if you separate them into parts it will be easier to find the time to see them. It's just a tip and you may have more views per video. Thank you very much for the content you make. It is simply excellent.
Can someone explain why Frequency = Baud / 2 @ 10:30? Does this have to do with Nyquist? I dont not see the correlation between nyquist (which has to do with sampling) and bit rates. :)
What simulation methods are there to simulate Fiber weave effect? Is there a standard skew in ps per length that we can use or something? Do a Monte Carlo or something like that by adding/subtracting skew? Or other thoughts?
Suggest to disable your own camera window when showing PPT information as it block the infor.
In the skew discussion he mentioned that the Nyquist was at 14GHz, but that would be roughly the nyquist of an NRZ signal. With PAM4, the nyquist is actually 26Gbits/4bits=~8GHz?
It doesn't say Gbits/s but 26.56 GBaud, so 26 G signal state changes per second, which is 13.28 GHz
Hmmm, if it's 26Gbits/sec (which is the same as 26Gbaud), and it's a 4- level code (2-bits per time slot), that means 13G time-slots/sec (or state-changes/sec). So the wave frequency is half that: (6.5 GHz). And the Nyquist frequency required to sample 6.5GHz, would be twice that... bringing is to 13GHz.
Baud is state changes per second. With PAM4 you transmit 2 bits with each state, so 50 Gbit/s
@@bene5431 Hmm, on looking this up, it does appear you're correct.
Thank you for so usefull content. Hope to see one time something more deep about s-parametrs and analyzing the channels:)
Compared to these guys I've always worked at DC, worrying about a few measly MBauds.
thank you for this video really helpful
nice video! very useful, however, modern transceiver do equalization, and therefore the eye diagram will look as the good ones
Another great video. Thank Robert and Bert. For surface roughness modeling which numbers use in Advanced Design System?
Advanced Design Systems currently does not include Huray model, so my Cannonball model does not apply.
@@bertsimonovich7297 thank you, Bert. Your work is phenomenal. I always have problems with modeling of surface roughness of copper foil. This staff in this video and your works are great.
@@Hadi-zw9mb Thank you. My web site has more detailed papers on modeling copper roughness.
Actually I just found out ADS starting with version 2022 includes the full Huray model so you can find the necessary parameters using my Cannonball model
Hi Robert, small tip. Please dont interrupt the Speaker. Ask your question only after the other speaker has finished. Wearing headphones really helps, then u r conscious of the other speaker and wont speak over and interrupt them
Of the various technical youtube channels I watch that do interviews, I would say that Robert inserts questions relatively unobtrusively. In fact, sometimes I get the impression that Robert has found a point unclear, and waits quite a long time to bring it up. And sometimes he even waits to the editing stage to insert a pause and clarification. It's also difficult to wait for the speaker to "finish" because, unlike in an in-person interview, the speaker on zoom or skype may not pick up a non-verbal cue from the inteviewer to pause for a question. As a result, a certain amount of "interrupting" is inevitable, and actually welcome, because Robert is interceding on behalf of the audience. No doubt all interviewers can learn things to improve their technique, but I think Robert does well, and a simplistic suggestion like "wear headphones" is not helpful -- he's obviously listening to the speaker already.
@@Graham_Wideman Thank you very much Graham. Yes, I was not intentionally commenting on ravi's comment as .... in many situations I have to interrupt. I know, it is not polite and I know, sometimes I cut out an interesting sentence, but otherwise the videos would be too long or not about the topic. Especially I interrupt when I see that my question was already answered or if the talk is going away from what I would like to focus on. If I would not interrupt, the talk would go sideways and videos and recordings would be too long or very hard to edit (recordings are often between 2-3 hours long, so even now it is a lot of work to record and then edit). Also, I interrupt, when I have a question in mind and I would like to ask it before we start talking about something else. PS: I often tell people before our call, that I will interrupt them, so they are aware of that (and they are ok with it, actually, they are happy about it, many are aware, that if I do not stop them, they would keep talking about the topics they love).
The specialist guest is showing graphs that go up into tens of GHz. How much relevance does this have to a person doing PCB designs that uses DDR3/4, PCIe gen 4 or Ethernet PCBs?
Do you have experience with differential pairs and trace length matching? Which effect could have a greater impact in my signal integrity, bumps on one line which influence my impedance or the asymmetry of the line lengths?
Thanks in advance
This is on my todo list of future videos
Like everything with signal integrity, it depends on your noise or skew budget. Skew budget will be a function of your data rate. Matching trace lengths helps to mitigate skew, and matching impedance mitigates reflections and noise. We usually budget about 0.2UI for skew from all sources in the channel before it starts to degrade the eye, where UI is unit interval of the baud rate. So for example, PCIE Gen5 (32 GT/s) the UI is only 31.25 ps and total skew budget from all sources is 6.25 ps. With a propagation delay of ~165ps/inch for the dielectric, that is 38mils (~1mm). But as I explained in video, there is fiber weave effect skew, that is harder to predict and other sources of skew to consider in total budget. Skew also translates into mode conversion which could exacerbate EMI compliance. Because PAM-4 SNR is -9.5dB worse than NRZ, everything matters. Only good modeling and simulation can quantify the different effects.
maybe someone could talk about wiring between/inside equipment (non-differential), should there be multiple grounds in ribbon? plain twisted pairs for each signal? won't they create ground loops? what is worse ground loop or coupling between signals if ground is shared by multiple signals? always wanted to know answer...
Excellent.
You should have a few videos on the basics of RF engineering. I fact all high speed pcb designs should. Concepts like short circuited stubs are introduced in the first week. Gotta ask too, a 6 mm long via? How many layers was that board?
I have some RF videos on my todo list. PS: I found this about the backplanes: www.raypcb.com/backplane-pcbs-fabrication-tips/
It was a 26 layer backplane 280 mils (7.1mm) thick
That’s a lot of layers. Definitely have to include those parasitics. In our millimeter wave designs we must create EM models for our different vias to design effectively.
exclusive content
Nice Joss
I am trying to run DDR4 batch simulation on Hyperlynx, but couldn't find a very detailed tutorial. Sad.