It will be great to bring this and other past/future videos together focusing on the complications of flex boards. In particular also considering the emi coatings that can be applied top/bottom of flex boards.
Isn’t it true in most layouts that the field of each trace is mostly contained in the space between the trace and the reference plane (not the other trace)? I think the magic of diff pairs in cables and uhdi was shown, but for most board designs it’s a different animal, right?
Many thanks for the video. Everything is clear, but I also would like to know the answer to the following questions: 1. For example, I use a 2-layer PCB without ground under a diff pair. Can I route other signals on the opposite side? 2. In the previous case, should I place ground traces/polygons around the diff trace? 3. For example, I have a ground plane under a diff trace. And then the plane ends, but the diff trace goes further, and from some point, there is no ground plane under it anymore. That's clear that this is not a good thing, but "how much" is it bad? For example, if I converted analog video signal to a diff line, there are no spectrum above 10 MHz, so how will it affect this kind of signal? Should I change the trace parameters in this case after leaving "grounded" area? Offtop: There is one more topic that would be interesting to me. There are a lot of devices that were designed decades ago and almost completely out-of-date now. For instance, the control system for trains. It must have high reliability, and they work great. However, their PCB design (70s -90s) doesn't comply with the current understanding of how to design PCBs. In this condition, managers often think to keep the past approach and avoid enhancements. What do you think about it? Thank you for your content!
This is super cool stuff. I haven't seen anyone else cover this type of detail in PCB design, and it's really useful for someone less experienced to get an intuition for behavior on a PCB. It sounds like this heavily applies not just to things like USB but also to PCIe. I have always wondered how the boards are routed in order to achieve the bus clock speeds for PCIe devices, and this is starting to help make sense of it. Seems like the traces would need to be very close together to make sure they are as self-referrential as possible, in order to make each serial pair as stable as possible. Or maybe I am misunderstanding.
Hi Zach, very insightful video on theoretical vs practical case on this topic, although it still begs the question as to how much degradation we see in signal integrity of differential pair when the tight coupling of that pair is broken due to choke/coupling cap/via in its path?? Since in this case, diff pair without any GND reference depends upon tighter coupling present between its traces, the further you pull them apart, higher the deviance from its characteristic impedance... Which might not be that bad of a case when there is somewhat loose coupling and GND reference plane is present below the pair.... Would love to hear your views on this...
Hi thank you for a great video. Maybe it’s is just semantics 😊 But at the example at 10:32, where you set the width to a 100mil, the spacing is 25.9mil (not 21mil) and then it actually decrease to 22mil. So the width is actually not increasing.
This is sorta unrelated to this video, but I had another question: Why do we panelize boards if the fabricator will be distributing our design on the panel anyway? Do they not configure the design to be as optimal as personally laying out a panelized version? I know a lot of pcb design speakers that say we shouldn't leave design factors up to the manufacturer (such as impedance control, stackup, etc.) Is Panelizing a design included in this idea of not leaving design factors up yo the manufacturer? If so what are your thoughts on why?
In some situations, the panel itself is the deliverable and must conform to specific requirements for further processing. If you're just buying individual pcbs, let the board house work out panelization for fab. If you're processing the panels as a whole, you should specify the exact panelization you need.
@@jonathonmcmillan9410exactly, sometimes the test fixture is made in such a way to test a panel so you have all the connections made on pcb outside your actual board.
The board orientation relative to the panel orientation also matters in cases where fiberglass weave direction makes a difference. Even if it doesn't matter electrically (thick traces), for elongated boards it could matter mechanically.
I have recently panelized a collection of three boards that were a part of one design. 1) seems like ordering one larger design is often cheaper than ordering a few smaller ones. 2) i needed the same stackup for all of them, so doing it all at once saves time and reduces the likelihood of mistakes during ordering. The mfg has noticed there are 3 board, and bumped up the price quite a bit unfortunately... I feel like it's quite unfair.
Ground is still important for all practical differential pairs, because of limitations in the common mode range of receivers. The absolute voltages on the members of the pair have to be within that common mode range or the receiver will not reliably translate the differential signal to a single ended logic signal.
indeed. I've recently been preparing myself to route some usb3 signals for the first time ever, and i've found a routing guide from Renesas. It says that having a track cross the diffpair through the groundplane layer is an absolute nono. It is certainly very bad to break a ground plane under a single-ended stripline, but i feel like it should actually be okay with a diffpair.
@@victortitov1740I think you are spreading the little return current and that makes a very big field that induces unnecessary noise into the diff pair itself. Why are you asking for trouble?
@@victortitov1740 Check Rick Hartleys talk on "what your differential pairs wish you knew", he describes why this is not a good idea. I personally trust his explanation more, coupled with Lee Ritchey's text on Differential Pairs (nothing against you Zach! 😅)
hi zach... a european viewer here... why does altium keep pushing the imperial measurement system, ie: 'mil' when the whole world (your audience) uses metric..? when was the last time you saw a component built according to imperial measurements..? even the ipc and other american engineering organisations strongly recommend using the metric system... other than that i'm a big fan of your content 🙂
No one is "pushing" people to only use imperial measurements, and Altium certainly isn't. I'm American so I'm more used to using imperial when it comes to building things. I'm also a physicist so I'm also used to SI units and you'll notice that I always do the on-the-fly SI calculations with SI (MKS) units. I never took the time to learn all the conversions to CGS intuitively, otherwise you would see me doing everything in cm/mm.
0402, 0603, PCB copper thickness oz per square foot, all of these (understandably bad) measurements came with US first adoption. As such, currently the PCB world seems to be stuck with these units.
Hi there. I am from Poland. I use ONLY imperial (mils) measurements because it is more convinient into a design. For instance, I use 10 mils trace, 20 mils, 150 mils etc. In metrics it would be 0,254mm, 0,508mm, 3,81mm. It's then to complicated. And mils are just as it name says, the inch divided by a thousand, so it is very accurate and you don't have to use a comma in numbers. Even simplyfying measurements to 0,25mm, 0,5mm, 3,8mm, they are still numbers with comma and they are not so easy to operate.
This is the explanation I have been missing. Thanks, Zach!
It will be great to bring this and other past/future videos together focusing on the complications of flex boards. In particular also considering the emi coatings that can be applied top/bottom of flex boards.
Isn’t it true in most layouts that the field of each trace is mostly contained in the space between the trace and the reference plane (not the other trace)? I think the magic of diff pairs in cables and uhdi was shown, but for most board designs it’s a different animal, right?
What would happen if we have a thick trace(GND) between a differential pair on the same plane (on a two-sided board)?
Many thanks for the video. Everything is clear, but I also would like to know the answer to the following questions:
1. For example, I use a 2-layer PCB without ground under a diff pair. Can I route other signals on the opposite side?
2. In the previous case, should I place ground traces/polygons around the diff trace?
3. For example, I have a ground plane under a diff trace. And then the plane ends, but the diff trace goes further, and from some point, there is no ground plane under it anymore. That's clear that this is not a good thing, but "how much" is it bad? For example, if I converted analog video signal to a diff line, there are no spectrum above 10 MHz, so how will it affect this kind of signal? Should I change the trace parameters in this case after leaving "grounded" area?
Offtop: There is one more topic that would be interesting to me. There are a lot of devices that were designed decades ago and almost completely out-of-date now. For instance, the control system for trains. It must have high reliability, and they work great. However, their PCB design (70s -90s) doesn't comply with the current understanding of how to design PCBs. In this condition, managers often think to keep the past approach and avoid enhancements. What do you think about it?
Thank you for your content!
When do we choose single coplanar and differential coplanar in 9:28? Can make another video to discuss about this?
This is super cool stuff. I haven't seen anyone else cover this type of detail in PCB design, and it's really useful for someone less experienced to get an intuition for behavior on a PCB. It sounds like this heavily applies not just to things like USB but also to PCIe. I have always wondered how the boards are routed in order to achieve the bus clock speeds for PCIe devices, and this is starting to help make sense of it. Seems like the traces would need to be very close together to make sure they are as self-referrential as possible, in order to make each serial pair as stable as possible. Or maybe I am misunderstanding.
Hi Zach, very insightful video on theoretical vs practical case on this topic, although it still begs the question as to how much degradation we see in signal integrity of differential pair when the tight coupling of that pair is broken due to choke/coupling cap/via in its path?? Since in this case, diff pair without any GND reference depends upon tighter coupling present between its traces, the further you pull them apart, higher the deviance from its characteristic impedance... Which might not be that bad of a case when there is somewhat loose coupling and GND reference plane is present below the pair.... Would love to hear your views on this...
Hi thank you for a great video. Maybe it’s is just semantics 😊
But at the example at 10:32, where you set the width to a 100mil, the spacing is 25.9mil (not 21mil) and then it actually decrease to 22mil. So the width is actually not increasing.
he also changed the tracks gap
This is sorta unrelated to this video, but I had another question:
Why do we panelize boards if the fabricator will be distributing our design on the panel anyway? Do they not configure the design to be as optimal as personally laying out a panelized version?
I know a lot of pcb design speakers that say we shouldn't leave design factors up to the manufacturer (such as impedance control, stackup, etc.) Is Panelizing a design included in this idea of not leaving design factors up yo the manufacturer? If so what are your thoughts on why?
In some situations, the panel itself is the deliverable and must conform to specific requirements for further processing. If you're just buying individual pcbs, let the board house work out panelization for fab. If you're processing the panels as a whole, you should specify the exact panelization you need.
@@jonathonmcmillan9410exactly, sometimes the test fixture is made in such a way to test a panel so you have all the connections made on pcb outside your actual board.
The board orientation relative to the panel orientation also matters in cases where fiberglass weave direction makes a difference. Even if it doesn't matter electrically (thick traces), for elongated boards it could matter mechanically.
I have recently panelized a collection of three boards that were a part of one design. 1) seems like ordering one larger design is often cheaper than ordering a few smaller ones. 2) i needed the same stackup for all of them, so doing it all at once saves time and reduces the likelihood of mistakes during ordering.
The mfg has noticed there are 3 board, and bumped up the price quite a bit unfortunately... I feel like it's quite unfair.
@victortitov1740 yeah they charge the designs not the gerbers
Ground is still important for all practical differential pairs, because of limitations in the common mode range of receivers. The absolute voltages on the members of the pair have to be within that common mode range or the receiver will not reliably translate the differential signal to a single ended logic signal.
This seems like a really controversial topic! I wish there was some more testing on design aspects like this.
indeed. I've recently been preparing myself to route some usb3 signals for the first time ever, and i've found a routing guide from Renesas. It says that having a track cross the diffpair through the groundplane layer is an absolute nono. It is certainly very bad to break a ground plane under a single-ended stripline, but i feel like it should actually be okay with a diffpair.
@@victortitov1740I think you are spreading the little return current and that makes a very big field that induces unnecessary noise into the diff pair itself. Why are you asking for trouble?
@@victortitov1740 Check Rick Hartleys talk on "what your differential pairs wish you knew", he describes why this is not a good idea. I personally trust his explanation more, coupled with Lee Ritchey's text on Differential Pairs (nothing against you Zach! 😅)
hi zach... a european viewer here... why does altium keep pushing the imperial measurement system, ie: 'mil' when the whole world (your audience) uses metric..? when was the last time you saw a component built according to imperial measurements..? even the ipc and other american engineering organisations strongly recommend using the metric system... other than that i'm a big fan of your content 🙂
No one is "pushing" people to only use imperial measurements, and Altium certainly isn't. I'm American so I'm more used to using imperial when it comes to building things. I'm also a physicist so I'm also used to SI units and you'll notice that I always do the on-the-fly SI calculations with SI (MKS) units. I never took the time to learn all the conversions to CGS intuitively, otherwise you would see me doing everything in cm/mm.
0402, 0603, PCB copper thickness oz per square foot, all of these (understandably bad) measurements came with US first adoption.
As such, currently the PCB world seems to be stuck with these units.
I use mils
@ - thanks for answering...
Hi there. I am from Poland. I use ONLY imperial (mils) measurements because it is more convinient into a design. For instance, I use 10 mils trace, 20 mils, 150 mils etc. In metrics it would be 0,254mm, 0,508mm, 3,81mm. It's then to complicated. And mils are just as it name says, the inch divided by a thousand, so it is very accurate and you don't have to use a comma in numbers. Even simplyfying measurements to 0,25mm, 0,5mm, 3,8mm, they are still numbers with comma and they are not so easy to operate.
@Raman from PUNJAB