Re: [SI-LIST] : A Question About Power Noise.

Ray Anderson (raymonda@radium.eng.sun.com)
Wed, 17 Mar 1999 15:09:34 -0800 (PST)

>
> OK Ray. Here is a loaded question for you. (See below)
>
... snip snip ...
>
>
> There has been considerable discussion about how to mount/route a bypass
> cap. One theory is to run traces from the cap to the device and then vias
> from the cap to the plane. The other theory is to run vias to the plane
> from both the device and the bypass cap and let the currents flow over the
> 'low inductance' plane.
>
> Now, if the highest inductance part of the path is the pad/via, would that
> favor running (wide) traces from the cap directly to the device, since that
> eliminates two vias between the cap and the device? A wide, short path only
> has a few nH inductance (I think).
>
> Doug Brooks

Just spoke with my colleague Larry smith about your question
and the concensus is that it is a BAD idea to run traces from
the capacitor to the chip and then connect the composite
mess to the planes via vias .


Why?? Loop area. The loop area of the current path that
describes the circuit from plane 1 up through a via, through
a trace, through a capacitor, back through another trace and
thence to the other plane is much larger than the loop area
from a plane, through a via, through a cap and then to the
other plane.

Why do we care about loop area? Because the bigger the loop,
the more lines of magnetic flux and hence the higher the inductance.
This translates to a lower SRF among other things. These other
things include making the Q of the bypass higher and hence of
lower bandwidth. This requires the use of more capacitors to
achieve the same bypass effectiveness since our goal is
to create a broadband low impedance for the power distribution
system.

To supplement all the buzzwords, Larry had empirical experience
in that he tried it wrong way once upon a time and swears
he will never do it again..

Run your chip power pins directly to the planes. Run your
decoupling caps directly to the planes (while minimizing the
current loop area and via length).

-Ray Anderson

**** To unsubscribe from si-list: send e-mail to majordomo@silab.eng.sun.com. In the BODY of message put: UNSUBSCRIBE si-list, for more help, put HELP. si-list archives are accessible at http://www.qsl.net/wb6tpu/si-list ****