Re: MISC-d Digest V97 #28

["Larrie Carr" <larrie_carr@xxxxxxxxxxxxxxxx>]

Mail*Link(r) SMTP               RE>MISC-d Digest V97 #28

I am a designer at PMC which designs/manufactures mixed-signal integrated
circuits.  We have always moved agressively down the technology curve (now
looking at 0.25um).

It is true that the speed increase of 0.5um is not that significant over
0.6um.  While the transistor length was scale to 0.5um, most of the other
dimensions (like wire width) did not.  So at 0.5um 3Volts, the transistors
see about the same capacitance, but don't turn on as hard, making it slower
than
0.6um!  And at 0.5um 5Volts, you are getting about the same performance as 
0.6um (but with a slight density improvement).

However, at 0.35um 3.3V, the story changes.  Most of the other dimensions did 
scale and the speed increase is impressive.  In this technology, I stopped 
caring about gate delays and worry about the wire delay.  Typical logic 
gate delays are well into half a ns.  I didn't believe the numbers when I
first
saw them.

Of course, the above depends on your fab, libraries and design methodology. As
well, the above are my opinions and do not reflect those of my employer.

Larrie Carr

----------------------

There have been some comments in MISC recently about future chip
speeds. I thought I should pass on what Chuck has been saying about
this since last fall.  Chuck has been saying that although he has
seen an increase in transistor speed with each reduction in geometry
the transition from .8u to .5u micron may not yield any increase in
speed.  The reason for this is that .8 may be run at 3 or 5 volts
while .5u may only be run at 3v.  The reduction in voltage causes
a reduction in speed similar to the increase in speed from the smaller
geometry.  This step will result in a smaller and lower power chip
however.  This means .5u will not let Chuck break the 1G barrier.
I have heard elsewhere that .5u can be 3v or 5v, but the indication I 
got from Chuck was that he is not expecting to see .5u running twice
as fast as .8u because it was also going to be running at 3v
instead of 5v.  Also the characterization of these processes as
.8u or .6u or .5u is somewhat misleading.  The .8u process from HP
has some parameters about the same as some other's .6u processes.