Re: gEDA-user: Eliminate separate Vcc planes?

[Andy Fierman <andyfierman@xxxxxxxxxxxxxxx>]

There are some nice little tools to show the behaviour of multilayer
chip ceramics (and some of Murata's inductors)


Although they don't include their single layer microwave caps in the
database, the Murata Chip Capacitor and Inductor S-Parameter &
Impedance Library program from here:

http://www.murata.com/products/design_support/mcsil/exe/mcsil-setup.exe

(This one runs OK on Linux too using WINE:

http://appdb.winehq.org/objectManager.php?sClass=version&iId=18139)


And as similar tool from Cap Cal from here:

http://www.dilabs.com/UserFiles/File/CapCadv302.exe


Fun to play with and give a good feel for how the caps and their
parasitics behave.


Single layer uwave caps may have a lower series L than MLCC parts but
it's not clear from many of the manufacturers data. By their nature
they are only available in relatively low capacitances (up to a couple
of nF).

As Neil Hendin explained, using supply traces rather than planes can
lead to resonance problems which may then need extra decouplers to, in
effect, tune these resonances out of you working frequency range. For
RF and microwave stuff that is fine but for the more broadband and
digital circuitry this is a real challenge. At high enough frequencies
you may be able to use the supply trace skin effect resistance to help
damp any resonances.

A further complication is that when you put several decouplers of
different values in parallel, you get unwanted resonances between the
series inductance of one cap and capacitance of another. This can
introduce some very wild parallel resonances where the supply suddenly
becomes high impedance to ground. Other series resonances from each
cap can form between the parallel ones to make a highly complex (!)
supply impedance.

These resonances aren't well behaved because the parasitics they are
caused by are not specified in the device datasheets so they should
not be relied on to do things like null out clock noise on Vcc rails.

As Gene's commented, very careful attention must be given to how any
decoupling cap is physically connected: the series L of MLCC's is so
small it can easily be swamped by the external PCB trace used to
connect it to the device and to ground.

If you have the component side signal layer over a ground plane then
using a small supply plane (via'd to the internal supply plane or
tracked to the nearest Vcc) directly under a chip with the decouplers
connected off that can create a very effective decoupling scheme. The
plane reduces the Vcc supply inductance at the device pins and
provides a very small but reasonably high quality decoupling C
directly at the device pins. The larger lower frequency decoupling can
then be located a little further away. There's a very nice apps note
showing that here:

http://www94.web.cern.ch/HSI/s-link/devices/g-ldc/decouple.pdf

It was originally from Cypress but seems to have disappeared from their website.

         Andy.

www.signality.co.uk



2009/10/20 Gabriel Paubert <paubert@xxxxxxx>:
> On Mon, Oct 19, 2009 at 06:43:42PM -0500, Darrell Harmon wrote:
>> On Mon, Oct 19, 2009 at 5:27 PM, Dan McMahill <dan@xxxxxxxxxxxx> wrote:
>> > my recent experiences are more in line with Larry's.  Most C for a given
>> > package and voltage seems to be the best meaning that above resonance it
>> > is no worse than smaller capacitance value devices and below resonance
>> > it is better.  And yes, this seems to fly in the face of what has been
>> > recommended in the past.  I've not done any careful measurements of
>> > older technology bypass caps but I wonder if this is one of those
>> > "rules" which became obsolete 15-20 years ago and no one noticed...
>> >
>> > -Dan
>>
>> I have done some testing of various passives (mostly 0402) and came to
>> the conclusion. I tested both shunt and series capacitors on a 50 ohm
>> transmission line with a VNA. What was most interesting to me was that
>> the large value capacitors performed better as series coupling caps
>> than the small ones. The single layer caps in the 1 to 100 pF range
>> frequently had parallel resonances in the 10 to 30 GHz range. Most of
>> the multilayer caps (10 nF to 1 uF) performed well to at least 30 GHz.
>
> And for the shunt case?
>
> Some manufacturers (ATC, Dielectric Labs) also offer specific broadband
> coupling capacitors, did you try them?
>
>        Gabriel
>
>
> _______________________________________________
> geda-user mailing list
> geda-user@xxxxxxxxxxxxxx
> http://www.seul.org/cgi-bin/mailman/listinfo/geda-user
>


_______________________________________________
geda-user mailing list
geda-user@xxxxxxxxxxxxxx
http://www.seul.org/cgi-bin/mailman/listinfo/geda-user