Kevin,
Re: "You've told us that we can
expect to see AMD's chips generally to run at slower clock speeds ..."
I have focused on performance, but I have also discussed the power of the
AMD processes. Here is a post that I sent to Petz to "bring him up to speed"
on process performance and power.
BTW Kevin, I'm very consistent in my views and posts ... I'm not concerned
with my credibility. Hopefully after reading my post below ... you will
correct your statement. It's a matter of your credibility.
exchange2000.com
Re: "Idsat will not remain the same, it will go up with voltage."
You have asked a very good question ... let me explain. A process
generation is typically defined by the printed poly gate length. For
a fixed gate length, your are correct ... Idsat does go up with voltage.
Idsat also goes up as gate length gets smaller. The other constraint
is that as gate length gets smaller, the voltage that can be applied
across the device (and gate) also has to go down. The two equations
that we are trying to optimize are:
Power ~ CV^2f , C = Capacitance, V = Operating Voltage, f = Frequency
Performance ~ I/(CV) , I = Idsat, C = Capacitance, V = Operating Voltage
It has been interesting that from process generation to generation,
Idsat and capacitance have stayed relatively constant ... for example
.35 generation - Idsat = ~625ua/um @ 2.5V, Lgate = .28um
.25 generation - Idsat = ~600ua/um @ 1.8V, Lgate = .22um
So in developing a process for optimum performance, one needs to optimize
the device drive current at low operating voltages. The processes
to optimize are gate thickness, Source/Drain implants, channel implants,
gate length ... The company that achieves the highest drive currents
at the lowest operating voltages win in both performance AND power.
I believe that AMD's process probably gives the following:
.25 - Idsat = ~600ua/um @ 2.1V, Lgate = .25um (my estimate of gate length)
Therefore, their devices (CPU's) will dissipate more power and will have
lower performance. I think the data on AMD's power/performance prove
my statements.
Hopefully this helps."
Make It So,
Yousef |
