If you look at Intel's design rules I think you'll see that historically Intel has never strived to have the most aggresive physical dimension scaling at any given process node.
The high density corner of process tech has usually been the domain of the foundry guys where production cost is paramount.
Intel has always been the king of Idrive and IDsat, which has always helped them with their clockspeeds.
At 45nm this difference in historical approach between Intel and the foundries is amplied by the fact that Intel's gate-last implementation of HK/MG comes with a penalty to areal scaling (density is inherently less on gate-last versus gate-first integration) but an added bonus for drive currents.
So the comparison between Intel's 45nm Atom and a TSMC produced Bobcat all the more underscores this difference.
Now TSMC's HKMG is gate-last as well, so the density of TSMC's 28nm HKMG versus Intel's 32nm HKMG isn't going to be as stark as the differences you see now at 45nm vs. 40nm.
Agree. If we look at Hand DV old post acedhardware here:
http://aceshardware.freeforums.org/...ng-momentum-is-this-the-new-wave-t966-45.html
"
45nm Nehalem: 729 M transistors on a 262mm^2 die
40nm ATI RV740: 826 M transistors on a 133mm^2 die
These numbers are still skewed since Nehalem has most transistors in
its large caches which allow a much denser transistor packaging as the
logic circuits which dominate the RV740, so the transistor density of
TSMC's process is well over two times higher.
The Lithography equipment Intel will be using for it's 32nm single
patterning Immersion process won't be able to reach the transistor
densities of TSMC's 40 nm
This is Intel's own choice which it can afford because it operates,
as we know, in a "less competitive market" as TSMC.
http://www.asml.com/asml/show.do?ctx=6717
http://nikon.com/products/precision/.../nsr/index.htm
As far as I know Intel uses the Nikon NSR-310F for 45nm production
which is a "dry" machine with an NA of 0.92 and an overlay accuracy
of 7nm.
and Intel will use the Nikon NSR-610C for 32nm production (without
double patterning) This is an immersion lithography tool with an NA
of 1.30 and an overlay accuracy of 6.5nm
TSMC uses an ASML immersion tool for 40nm, almost certainly the
XT-1950Hi with an NA of 1.35 and an overlay accuracy of 4nm.
TSMC will use the newer XT-1950i for it's 32nm and 28nm processes.
This tool has the same NA of 1.35 but has an improved overlay accuracy
of 2 to 2.5nm to allow reliable volume production using double pattering."
------
This is a post from 2009. And he took a beating from the likes of Anon and Charlie
,
He have later added at SA that:
"
Nikon's NSR-620D for 32nm production is currently being delivered to Intel
after a long delay."
And the 620 have the folowing specs:
http://www.nikonprecision.com/products/nsr_s620.html
2nm overlay accuracy
1.35NA lens
What can be interesting is that Intel as hinted otherwise by Hans, is that Intel chooses suppliers for other reasons that just spec/price, but also to keep future competition on the market. Read: They did get less dense 45nm to secure a future solild lithography suplier market.
It can sound crazy but i think its very good strategic policy, it happens where i stand to. Its far more important to build future deliveries for the tools, than just think on the short term.
Consequenses
The consequenses for the future development is just that i think we can expect better power management from the gpus with higher clocks and far better idle power. But it doesnt ofset the fight for wafer area.
If we could look at TSMC balance sheets over the years, i think the gpu takes less and less production, and that development will just continue.
It all comes down to what we can expect from the gpu in the future. I dont think perf. / price will continue the way it has. Dont hope for cheap high perf 28nm parts. They will be more lean, but i wouldnt hope for the current perf/price trend to continue.
What is needed
But do we need all those big gpu anyway, with all those console ports?
This is written on a sandy bridge mobile with Intel gfx, with excellent battery life, performance and temperature. It just shows the importance of granularity of power management. That will hopefully hit the gpus too, so we can have cooler gpus.
Perhaps that will cool the debate too. And Idontcare will have less to do
Facebook
Twitter