There's no reason for that rule. Most of the time a reasonable, like
0.33 uF, 0603 maybe, surface-mount cap has a very low RF impedance,
essentially the same ESL as most any other 0603 cap. SRF doesn't
matter, impedance does.
A 0.33 uF 0603 cap is a very low z well into the GHz. Ideally, match
the cap body width and the trace width.
John
I'll second that, with a bit more explanation: smt caps have
inductance that's determined by their physical size. A piece of wire
has inductance, after all; a length even .06 inches or .08 inches of
microstrip has inductance. What it also has, though, is capacitance
to a ground plane. The distributed inductance and the distributed
capacitance cause it to have a particular characteristic impedance,
sqrt(L/C) (neglecting the reactive part of the impedance caused by
resistance, which is generally quite small in the GHz region). If you
model the MLCC smt part as a conductor the length, width and thickness
of the part, you'll be really close to the way it actually performs in
the circuit. That is, make its inductance part of the transmission
line you mount it in. If the capacitor is slightly narrower than the
microstrip trace, the impedance should remain very close to constant
through the capacitor. If you want to get fancy about it, use the
freeware ATLC program to calculate impedance for that cross-section;
or use Agilent's ADS to accurately predict performance to well beyond
1GHz, if you put in an accurate model of your system. -- The reason
I say that ideally the capacitor will have a width slightly less than
the microstrip trace is that the added height of the part adds
capacitance to the ground plane and slightly decreases the series
inductance compared with a thin trace the same width. But...the
propagation velocity of the line is such that .06 inches of line
represents only about 3 electrical degrees at 1GHz using an FR4
substrate, less for lower dielectric constant substrates. One way to
think about what that means is that a reflection off the "leading"
edge is very nearly canceled by the corresponding reflection off the
"trailing" edge, where you have an impedance discontinuity only that
long. If you try to think about this stuff in terms of self-resonant
frequencies, you'll get yourself all needlessly worried. The model I
suggest above is far closer to what's really going on. (This assumes
the series reactance of your capacitor is very small compared with the
transmission line impedance, of course; otherwise, just include that
as a series impedance at that point along the line.)
Cheers,
Tom