Yeap Swee,
This is a very general question - but I'll try my best to answer your
questions.
I will limit my response to signal integrity/propagation only and not try
to address such issues as switching noise (which is probably most
affected by 2nd level packaging) and power/ground plane noise isolation
on the pcb.
I would try to analyze these following areas:
1) pcb vias - Probably your major electrical discontinuity.
Through hole vias may not be acceptable so blind vias
may be required. Also, if the number of high speed lines
are limited then microstrip may be the way to go. However
microstrip lines will raise EMI emissions.
2) dielectric characterization, especially loss tangent
Standard FR4 materials used for PCBs have
one of the highest loss tangents (~.02) of the
commonly used materials in the industry.
3) metal loss due to skin effect - PCB metals have very
jagged surfaces - apparently to allow the metal to adhere
well to the dielectric materials. If skin effect losses are
a concern for smooth metals, then do the jagged surfaces
worsen the skin effect losses significantly?
4) internal inductance effects - Not normally a concern for
todays technology because I theorize a) internal inductance
would lead to dispersion which leads to errors in predicting
propagation delays - but these errors may be small when
compared to sub-Ghz timing requirements, and b) pcb
conductors are typically quite wide compared to the signal
thickness and such a structure would not have as much
internal inductance as a conductor that is more squarish.
5) impedance and propagation control - At > 1Ghz propagation
for a digital signal the likelihood of having more than one
signal on a wire between clock cycles is greater. Perhaps
impedance control becomes a bigger issue for this situation.
Propagation delay control certainly becomes a bigger issue
as the clock period decreases.
Hopefully I have answered your question regarding what the
main problems could be for >1Ghz on pcbs. With respect
to overcoming the problems, first you have to create models and verify
your models to determine which factors are most detrimental
in meeting your requirements. Then you have to design around
these factors - or invest in more advanced technologies.
Mike
-- _______________________________________________________________ Mike Degerstrom Email: [email protected] Mayo Clinic - Gugg. Bldg. RM 1011-B Phone: (507) 284-3292 Rochester, MN 55905 FAX: (507) 284-9171 WWW: http://www.mayo.edu/sppdg/sppdg_home_page.html _______________________________________________________________