Quote Originally Posted by informal View Post
I think AMD went for DFM with Shanghai/Deneb(and will continue to do so in future). The GloFo slides show better than expected clock targets(N+2) but yields are not where they want them(yet).
Maybe to an amateur. To professionals, that is a process out of control.

Quote Originally Posted by Paul Demone
32nm HKMG / Fmax distribution ?

They seem to be suggesting that they're doing better than the previously expected median of whatever that is...

max frequency of something or other?
That is max frequency.

It is one of those things that will impress the more technically
inclined AMD supporters who will inform the mouth breathers
and all will cheer and rejoice. It takes an experienced industry
type to read the real message there.

So fmax yield skewing high like that is good right?

Well fmax doesn't happen in isolation. It is highly correlated
to leakage and electric field strength related reliability and
aging effects. You get a lot of fast parts but those parts run
very hot, have high sleep mode power, and high failure rates
in the flat part of the bathtub curve.

That graph is not of a process 1) in control, or 2) meeting its
desired target by a long shot. You *want* a nice tight bell
curve approximately centered around your target median.