Wow! That is contrary to popular belief and what we've been told by other
academics.
Maybe I've missed something, but I was under the impression that the
impedance should be controlled as impedance discontinuities will cause
signal reflection.
Correct. But the point being made in the article is that they can be
treated as two separate terminated lines, therefore any differential
impedance becomes irrelevant. Even for a seemingly "differential
termination" like that in the LVDS example.
As the impedance is normally set to 100 Ohm differential in all other parts
of a PCIe system, then I should maintain this on my PCIe card (not my first
PCIe design, but the most complex one I'm working on).
I think that linked 1st article is a tad misleading. It does conclude by
saying that you need two signals of equal length and of a specified
characteristic impedance. By far the easiest way to achieve this is to use
closely coupled signals.
Yes, I agree. But the focus on differential impedance is usually
unwarranted, and that's really the main point. Getting correct
differential impedance can often be quite touchy, so it can be easier
to simply ignore it and work with single trace impedance. Mine usually
end up within 10% for the differential impedance anyway.
And I'm not sure saying 12.5 mm for trace length differences is ideal as
this is 1/10 of a bit flight time.
Yes, 12.5mm is pretty big. I try to keep mine within 100thou, just
because I'm a bit fussy about those sorts of things :->
I thank you for the link(s) but will likely continue to place my traces in
the 'accepted' way.
Sure, no harm in that.
Dave.
