On how the NSA can be generations ahead

[Eric Murray]

On Sun, Aug 01, 2004 at 10:20:38AM -0500, J.A. Terranson wrote:
> On Sat, 31 Jul 2004, Major Variola (ret) wrote:
> 
> > Tyler D asked about how the NSA could be so far ahead.
> > Besides their ability to make 2" sq. chips at 10% yield (not
> > something a commercial entity could get away with)
> 
> What, exactly, would be the point of doing this?

More gates == more processing.

> > they can also *thin and glue* those chips into say stacks
> > of 5 thinned die.
> 
> As easily as you could do this to high efficiency chips.

It's possible, using technologies like flip-chip.  But its not
as good as having everything on one die.  The interconnects
are limited in number and large in size, so they take up a lot of
room.

Stacked die are also more difficult to keep cool.

> > 2" sq = 4 x performance
> 
> How do you figure 4x performance on a 2" chip?  Most of the chip
> performance is tied to the total distance that signals must traverse
> across the chip surface.

4x the gates (roughly) means 4x performance.
Chip performance, especially for highly parellizable things like
key cracking, is determined by the number of gates.


Eric