Thanks to those who responded. I think some may have misunderstood. Skew here is the difference between the propagation delay in differential mode (hence received bit times) between PHY lanes (differential pairs).
I am asking because I am working on the laning architecture for 50GBASE-T2 and 100GBASE-T4, and the time variability of the delay difference between each pair of 25GBASE-T1s link matters when you bond 2 or 4 links into a higher speed link.
So, in this context since skew is the difference between differential mode propagation delay, and it would be worst-cased on an 11m link. (so about 60nsec total). A 25G bit time is 0.04 nsec, or 0.07% of the total prop delay, so any change
in the relative delay more than that would be something we start to pay attention to – obviously, this is a small number, one that many might think is negligible. If that is too small, I’ll look at the 25GMII clocking, which is at 390.625 MHz, or 2.56nsec,
about 4% of the total delay, but it would be nice to have time/temperature variation in the relative delay of 2 links to be substantially less than that.
As an aside, I have not heard a PHY person refer to the difference in delay between the + and – conductors of a differential pair as ‘skew’. That is all encompassed in the (complex) response of the insertion loss for the differential mode.
I’m not really looking for proprietary data or theoretical treatises on this, but rather impressions formed from physical testing as to whether tracking the relative delay at these levels is a necessary feature. It adds complexity to the
laning (basically determines whether you do alignment once at the startup, or track it continuously with inserted markers).
If you wish to respond privately, that’s OK too.
-george