On Oct 8, 2007, at 12:45 AM, Alan B. Pearce wrote:

>> so that the processor's
>> ground pad is the only place that this cluster touches any other
>> ground.
>
> Personally I would have the two inner ends of the caps as the ground
> connection, rather than the outer ends as illustrated, making it  
> closer to a
> point ground.

Ah, I think I'm starting to understand.  While a ground plane provides
nice shielding and power distribution, "bypass caps" should be wired
directly across the pins of the IC they are bypassing, rather than  
between
a power pin and the ground plane (even if the ground pin of the chip is
relatively close by.)  This helps prevent the transients that are  
produced
by the chip from being propagated onto the ground plane via the  
"infinite
resistor array" effect that Dave mentioned.
In other words, the ground plane is not your ideal power distribution  
plane.
Hmmph.  That's rather inconvenient.  Are there any tricks to  
implementing
this using common CAD packages?  Even if you directly wire caps to pins,
EAGLE will happily fill in the ground plane as well.  I guess it  
falls out
nicely on 4-layer boards, where the inner layers are power planes but  
the
outer layers (and the caps and their wiring) are on the outside.
How does the tradeoff of wanting separate ground planes and power  
connections
go?  It's counter-intuitive that I might be able to reduce EMI by  
deleting a
component side ground plane, for instance, but taking the concept to  
the limit,
that seems to be what is implied...

Are there amateur-level tests that can give you relative EMI info?   
Not exact
values, but "it's better if I do THIS than if I did THAT" ?  Scope  
with a
dipole or something ?

BillW
-- 
http://www.piclist.com PIC/SX FAQ & list archive
View/change your membership options at
http://mailman.mit.edu/mailman/listinfo/piclist