Designing high speed boards without microvias

[matrixofdynamism]

So I am trying to understand the topic of microvias in detail and have created this thread.

So first thing I want to clear is, is it possible to design high speed boards without microvias i.e just rely on plated through hole vias without back drilling? I believe its not since plated through hole vias will have a short stub if we are not connecting between top and bottom layers and this shall cause reflections that can cause problems with signal integrity. I believe that knee frequency of the signal in this case has to have some mathematical relation with the size of the stub otherwise problems will happen. Is this correct?

[tszaboo]

Yes, you can design high speed boards without those. I expect this to only become an issue well above 4-5 GHz. Basically look at a PC motherboard, or a video card, all those are possible without these microvias or buried vias.

[T3sl4co1l]

How high are we talking?

People often vastly overestimate the requirements.  You see plenty of articles for bleeding edge (10s Gbps say) interfaces, and sure, those extend to lower frequencies as best-practices, but how much of that you really need depends on the exact edge rate, and noise tolerance, in use.  PCIe for example (PLL clock recovery, numerous signal conditioning and line coding tricks) is a hell of a lot more tolerant of poor signal quality than a bespoke low-jitter timing interface (no error correction, sensitive to exact analog levels/crossings).

People most especially overestimate Ethernet; I regularly see questions asking whether some layout or schematic is adequate at 100Mbps, i.e. where the critical mismatch/stub length is on the order of 10s of cm.  Accordion traces and via stubs are just utterly undetectable at such bandwidths.  Just do a quick reality check, and relax, right?

Tim

[twospoons]

Rule of Thumb #18 : Stub length(cm) < 0.75/ Bit Rate(Gbps)
from https://www.edn.com/how-long-a-stub-is-too-long-rule-of-thumb-18/

I was concerned about testpoint stubs on USB lines - until I read the above and realised I could have 10mm stubs on 480Mbps USB without issue.

[alligatorblues]

Vias are very useful on the analog side, for DIYs. With analog, you have to preserve the signal. With digital the sampling becomes numerical. If you lose some of the signal, you just refer back too the transmitting side, and tell it to send it again.

There's many algorithmic tricks to make the digital side robust. Unless you do what my friend, Marylin, asked me to do with her computer: she asked, "What are all these wires doing everywhere? (You know, monitor, printer, scanner, USB, AC power, speakers, network) Could you just cut those off, and get rid of them?"

I said, "I don't recommend that. It may be detrimental to perforrmance." So, I went and got some flex cable tubing, stuffed the wires in that, and she was so pleased she pulled all the cash out of her purse--and she's married to a doctor--and gave it to me. It was about $500. That was for 1/2 hour of labor and $16 of flex tubing.

And, she invited me for dinner. And there were six people for dinner. She prepared six different entrees, because everyone wanted something different. I  had a steak. The doctor had Vietnamese soup. I can't remember the rest. But  I learned there are things really important to others that I couldn't care less about.

And if you get them what they want, it can be worth a lot.

[matrixofdynamism]

Lets define what is meant by high speed so this question can be answered.

External Memory Interface: for DDR3, DDR4, DDR5.
Multi Giga Bit Transceiver: lets take an example Arria 10 at 17.4 Gbps

And as someone said, it is about the edge rate (rise time and fall time) and not the frequency of the signal.

[tszaboo]

Rule of Thumb #18 : Stub length(cm) < 0.75/ Bit Rate(Gbps)
from https://www.edn.com/how-long-a-stub-is-too-long-rule-of-thumb-18/

I was concerned about testpoint stubs on USB lines - until I read the above and realised I could have 10mm stubs on 480Mbps USB without issue.

Not surprisingly, if you plop in my 4-5GHz estimate into this, at 5 GHz you get 1.5mm, which is the thickness of regular PCB.
And if you have a stub, it's not the end of the world. It depends where, and how it shows up in the eye diagram. I would just route it with regular PCB techniques and only upgrade if necessary. Not to mention, a 6 layer board with with blind vias is probably cheaper than back drilling, you really need to go deep comparing the prices of your board supplier.

[T3sl4co1l]

Ah, that makes sense; cross-link to avoid duplication: https://electronics.stackexchange.com/questions/687181/how-to-know-what-length-of-via-stub-is-acceptable-for-a-given-signal

Tim

[matrixofdynamism]

So can one really do DDR3 and PCIe layout using all plated through hole vias, no blind or buried or backdrilled vias?

[tszaboo]

So can one really do DDR3 and PCIe layout using all plated through hole vias, no blind or buried or backdrilled vias?

Sure, just take apart a computer from the last 10 years, and you will see it.

[Infraviolet]

From what I've heard having even one blind or buried via on a PCB massively adds to the per board production cost. Quite possibly by more than a board without such features would cost in the first place.

I would guess such via types are only used as a last resort where mechanical constraints (forcing a small board) and other constraints meant board designers have "painted themselves in to a corner" from which only those special via types let them escape.

I'd guess then that blind and buried vias are only to be found on boards where after a huge amount of work nobody on the design team could find a way to avoid them.

[tszaboo]

From what I've heard having even one blind or buried via on a PCB massively adds to the per board production cost. Quite possibly by more than a board without such features would cost in the first place.

I would guess such via types are only used as a last resort where mechanical constraints (forcing a small board) and other constraints meant board designers have "painted themselves in to a corner" from which only those special via types let them escape.

I'd guess then that blind and buried vias are only to be found on boards where after a huge amount of work nobody on the design team could find a way to avoid them.

I had a board that had a narrow pitch BGA package, and mmwave radar, and other design contains on it. So I sat down with a PCB supplier for a meeting, we went through the options of blind or buried vias multiple layers and so on. Basically deviating from a standard board that I would do. The PCB cost in their quotation ended up something like 9 times the cost I would need to pay for a standard board, and thats for thousands quantities, it was the single most expensive part of the BOM. There was a way to route the board in the end without them.
So I guess this technology is fine for mobile phones or FPGA boards. In phones it's a postage stamp sized PCB anyway, that you can put hundreds on a panel.

[MathWizard]

If parasitic's of typical through hole parts start to become important around 10MHz (or maybe it's less).....how hard is it to make FM radio circuit ? How did they do all this in the 1930's in big bulky radio's, and then in RADAR, etc ? Do they just have much lower power factor's, and low gain's ?

I was playing with a 5MHz TX circuit on a protoboard lately, and some results surprised me, not that I understand it all. But it makes me want to try and measure some parasitic's, at least of the BJT's.

[T3sl4co1l]

FM BCB is a couple meters wavelength, so some cm lead length is okay as long as the impedance is near Zo.  It's when the impedance is very high or low in relation to the geometry, that the frequency limit is much lower.

Hence why you can do an RF amp with Zo = 50 ohm say, and do it at 1GHz with <1cm parts, but you can't do an SMPS at a couple amperes (Z ~ ohms) with 10ns edges (~100MHz harmonics) and the same dimensions.

Put another way: the impedance of say average 1/4W axial resistors is flattest up to the highest frequency, for values say 100-330 ohms, and progressively worse outside of that range.

Tim