3

I am working on a device that runs 100 Base-T. When routing the differential pairs I was wondering if there is a maximum differential skew specified somewhere and if so, what is it? Is there a formula, does it depend on transfer speeds etc.? I haven't been able to find any information on this topic, so thanks in advance!

VicTic
  • 59
  • 5
  • You can probably find this information in TIA-EIA-568, Cabling. – SteveSh Aug 17 '22 at 13:22
  • So just to be clear, you are asking how much difference there can be between two wires of a differential pair, i.e. intra-pair skew? And you are not asking how much difference is allowed between differential pairs, i.e. inter-pair skew? – Justme Aug 17 '22 at 17:24
  • @Justme I'm sorry for being unclear there. I am talking about the skew between two wires of one pair (intra-pair skew) – VicTic Aug 18 '22 at 06:38

2 Answers2

1

You can probably work backwards from the cabling specification, which is 50 ns for Cat5 and 6, 30 ns for Cat7/7a. This is for inter-pair skew, that is, the difference in path length between for example the Tx+/- pair and Rx+/- pair. For intra-pair skew, the difference between Tx+ and Tx- in the same pair, TI's MDI guidelines specify matching within 50 mils.

winny
  • 13,064
  • 6
  • 46
  • 63
vir
  • 14,718
  • 13
  • 28
  • I have no idea where these numbers come from, but they are certainly wrong. Cat5 has a bandwidth of 100 Mhz, for a symbol rate in the same ballpark. That would allow for not more than 2-3 ns margin. – asdfex Aug 17 '22 at 17:23
  • Edited to clarify inter/intra-pair skew. – vir Aug 17 '22 at 20:39
  • @asdfex They come from standards and requirements and are correct, 50 ns time difference is allowed between pairs. When data is transmitted over a single pair, e.g 100Base-TX, time difference between pairs is irrelevant. But 1000Base-T and others that use multiple pairs of course must handle up to 50 ns time difference between pairs when receiving data, as per the specifications. – Justme Aug 18 '22 at 08:10
  • 50 mils = 0.05” = 1.27 mm? – winny Aug 21 '22 at 06:34
1

While 100BaseT uses differential pairs, it doesn't run at that fast of speed, at only 66 MHz (full duplex) and at that speed skew matching would be really long.

1000BaseT runs at 125MHz and has tighter tolerances. (why doesn't it run 10x faster? because of encoding)

Even microchip (AN2054) only recommends 50 mil for gigibit, although I have seen some other gigibit phys call out 20mil (intra-pair) matching.

Keep the intra-pair and inter-pair skew between the device and RJ45 to less than 50 and 600 mils, respectively.

Source: http://ww1.microchip.com/downloads/en/Appnotes/00002054A.pdf

If you used the ratio of the speed between 1000BaseT and 100BaseT, which is a around a factor of two then you could say that 100BaseT needs 100mil of skew matching (which is not correct but it could give you an idea of the skew).

I think there is not much information because most designers are going to match better than 100mil for 100BaseT at 66MHz.

Voltage Spike
  • 75,799
  • 36
  • 80
  • 208
  • Of all the different 100Base-T variants, this answer seems to refer specifically to the -TX which is the most common, but incorrectly. 100Base-TX has a symbol rate of 125 Mbaud due to 4B5B coding and 31.25 MHz bandwidth requirement due to MLT-3 coding. – Justme Aug 21 '22 at 06:56