Jonh,
Thank you for digging into Rambus design requirements:)
<Are these the sort of things I should be reading?>
Yes, although there must be the other version,
RIMMDsgnGuide090.pdf
<As for illiteracy, the above document has some very clear diagrams showing what I describe.>
Not exactly. Let's see:
<I think that this document shows quite clearly that Rambus modulate the trace width to achieve a constant impedance and hence propagation velocity ..>
It has nothing to do with propagation velocity
which depends solely on effective dielectric
permittivity around the trace. The correct reason
RAMBUS is doing this is to reduce signal reflections
that would occur at trace impedance mismatches.
<I think that it also shows that if I am completely ignorant of the meaning of trace capacitance, then
so are Rambus.>
You seems to be not "completely", but "semi-".
So are they. Why? First, is was proven by
industry-wide "experiment". Second, let me suggest
at least one area that has escaped their attention.
In their RIMM design, half of the traces are external
microstrips, the other half are internal striplines.
The propagation delay along the microstrip has
lower effective electric permittivity because
about half of it is in surrounding air. Now watch:
the total RIMM propagation delay supposed to be
1.5ns, or 1500ps. On the top, all signal delays
must be matched up to 5ps, which is 0.3%. Now
what if there is misfortune of wet weather, with
100% humidity? The dielectric constant of water steam
is 1.2% higher than for the dry air, so it could
reduce the propagtion delay of external traces
by up to 0.6%, which will violate the RAMBUS
specifications! So, do not use your computer
in rainy weather, it may hang and corrupt your
data :)
Of course, the above example is somewhat of
exaggeration, but it shows how ridiculous
their tolerances are in terms of mass
manufacturability. More realistic scenario
is that air permittivity does not depend on
temperature (it is 1.0000xx) while the resin
inside the RIMM has a variation of about 20%
over the 0-70% temperature range. Again, when
the RIMM stick gets hot, internal microstrips
get slower by few %, and all specs blow up,
at least formally. And maybe not only:)
<Perhaps you should read it yourself.>
Already done :)
P.S. Dont be sorry for poor html. I can read it:)
|
