> I will agree with your approach if the package has rails, namely
> separated power planes for its use, however, i meant a plane-less package
> inwhich the cavity is used as the down-bond plane which means noise is injected
> to the whole core, doesn't it..?
Ah! Something like a PQFP, then.
Be careful here. The die attach isn't really a reliable
electrical connection to the bulk silicon, and if your
package is typical it doesn't have especially better feeds
than the signal connections. Since there are bulk connections
all through the core, you may as well tie the core VSS to
the plate. This is also true since the core is less sensitive
to supply symmetry than the I/O ring is.
The PLL supplies, though, should still be isolated.
> > > I am inveswtigating the various options of using the down bonds
> > > for our design. naturaly, one would like to connect all the VSS to the
> > > down-bond plate. However, the periphery supply, which makes the most
> > > noise, can then inject noise into the core logic, especialy to the PLL
> > > circuitry through the substrate (influencing input buffers' trip-point
> > > as well). So, my conclusion is to conect core and other non periphery
> > > VSS to the down-bond plate (paddle) and the periphery should be directly
> > > connected to the packages' pins.
> > >
> > > Do u agree with the analysis, can u point me to some areticles
> > > and/or literture on the subject..?
> > Our standard practice is just the opposite. Core current is
> > generally balanced so onchip bypassing takes care of the worst
> > of the high-frequency components, while most I/O types are
> > unbalanced and heavily dependent on supply inductance. We
> > run the I/O ring VSS and VDD supplies to the package rails
> > and supply core through either signal or secondary supply
> > balls. Either way, PLL supplies are as completely isolated
> > as possible, with no connection to either core or I/O supplies
> > short of the PWB planes.
-- D. C. Sessions firstname.lastname@example.org