.15 micron versus .13 micron process

[ricercar]

What are the advantages of NVIDIA going with .13 micron process for NV30? Does the .13 process produce a faster device? less hot? More die per wafer? Able to leap taller buildings in a single bound?

If this .15 -> .13 migration allowed ATI to "catch up to" or "pass" NVIDIA's leadership with a .15 Radeon 9700, won't this be ephemeral glory? Eventually ATI will have to go to .13 and NVIDIA will be a jump ahead again.

[thcdru2k]

lamen terms. smaller = faster 0.13mm is smaller than 0.15mm. you make more use of the space given.

[The Baron]

smaller = more transistors in same space = faster. and it's usually cooler, too--look at Tualatin...

[thcdru2k]

yup its always cooler

[StealthHawk]

Originally posted by ricercar
What are the advantages of NVIDIA going with .13 micron process for NV30? Does the .13 process produce a faster device? less hot? More die per wafer? Able to leap taller buildings in a single bound?

yes, yes, yes, and yes(hopefully) :D

number 1 may not always be true, but it should be true in most cases because of #2.

also, #3 is true, but since .13 is a new process and the fab is reportedly having trouble with it, the yields are a lot lower than they would be on .15

[PreservedSwine]

Cooler, and less power consumption.

In turn, the cooler temps should allow for a faster core frequency, in theory:)

[SurfMonkey]

As the process gets smaller you use less power, the die runs cooler and you have more silicon real estate to play with, so you can add lots of extra features. You can run the die at a faster speed because you are running cooler. You also have a better part for mobile computing because of this.

Check out the EETimes web site for more stuff.

[jbirney]

smaller die != faster all of the time. In most of the cases you could but it really depends on the design and were the bottle necks are and what is the critical path (as the critical path is the slowest and sets the speed).

But I have had several die shrinks were we keep the clock at the same rate in order to maximize yields. Yes going to a smaller die gave us more parts per wafer. Keeping the clock at the same speed gave us more usable parts at the end of probe (or yield).

Keep in mind part of the reason why ATI hit the 325 Mhz with the R9700 is that they went in and hand tweaked the critical paths to optimize the speed. There is no way of knowing how much that helped them.

Your main enemy is heat. The hotter it gets, the more thermal noise generated inside the part. Once this noise is "high" enough to effect bits then you get errors. Heat is caused by the transistors switching on/off. The faster you go the more heat you generate. However the amount of heat is also related to the size of the deceives and their power consumption. The smaller the devices the less power it consumes, which impels that it produces less heat.

So for us lower size is better. But as the die sizes shrink other factors make it more difficult to build....

[DadGT]

You have mentioned all the primary advantages to the .13um process when done correctly. I think your second paragraph is more interesting. When chip desingers start on a new chip, one of the first choices they make is what die size to use. Since NV30 has been 2 years in the making, that means nVidia looked into their crystal ball two years ago and guessed .13um would be ready to go. Looks like that was a little off, as evidenced by the delay people have supposed.

ATI with the R300 choose stay with the mature .15um technology and it allowed them to bring a product out quick. However, the large die can make cost high and there was some speculation at Beyond3D that the R300 chip alone costs >$100. (xbox chips cost ~$40)

ATI will eventually move to a lower die size. Everyone always does. And everyone (Intel, AMD, TMSC, nVidia) has had unexpected problems with getting .13um processes chips up to "speed." How ATI deals with the transition is yet to be seen. I personally think ATI's first .13um chip will either be very close to the R300, in which case it won't be radically faster, or a new design that is delayed to market. Either way, if NV30 is a massive jump, ATI could be playing catchup again soon. But that is just my opinion.

[Lezmaka]

According to the Anandtech article, deciding which process to use is one of the first things ATI does. I would assume nvidia would do the same thing.

http://www.anandtech.com/video/showdoc.html?i=1711&p=2
The cost limitations also help determine figures like transistor counts; at this stage the target manufacturing process is also chosen depending on a number of factors. As you can expect the manufacturing process (e.g. 0.15-micron, 0.13-micron, 90nm, etc…) contributes to the cost structure of the chip and imposes die sizes/transistor count limitations as well. What's important to note is that the target manufacturing process is decided upon at the very start of the design cycle based on estimates of where the foundry (the people that actually manufacture the chip - e.g. TSMC) will be at by the completion date. If this estimate is off, which was the case with NVIDIA's NV30 design, then the GPU will be inevitably delayed. Once a manufacturing process is decided upon, it is extremely difficult to, months later, go back and attempt to revise the design for a different process.

[Lezmaka]

I was wrong until I read that article too

[Mod]

Wouldn't it be the smaller the transister the hotter it is ?

[thcdru2k]

no

[DadGT]

Originally posted by Mod
Wouldn't it be the smaller the transister the hotter it is ?

I know it's the opposite way.

From a simple physics view (and this may not be the real reason) but:

Power = i^2 * R
where i = current and R = resistance.

(or another equal way of expressing it)

Power = i * V
where i = current and V is voltage drop across any one transistor.

Reducing the die size means that at least one (and possibly all) of the terms for power can be reduced and hence, less power consumped and lower temperatures.

[rwolf]

When ATI goes to .13 microns with the R350 they will save big because nVidia and TSMC will have resolved the process issues at nVidia's expense. Just like ATI had to resolve the AGP 8X issues with the motherboard manufactures because the spec was new and much more tempermental than AGP 4X was.

[Lezmaka]

Originally posted by rwolf
When ATI goes to .13 microns with the R350 they will save big because nVidia and TSMC will have resolved the process issues at nVidia's expense. Just like ATI had to resolve the AGP 8X issues with the motherboard manufactures because the spec was new and much more tempermental than AGP 4X was.

Just because nvidia got the problems worked out for its chip doesn't mean ATI won't have problems as well. VIA's had it's C3 chip on .13u for a while but it sure doesn't seem like it helped nvidia any.

[Dr_Colossus]

Just because the process will be more refined doesn't mean they won't have problems. They'll have to make changes on an architechtural level not simply start using the .13u process. You say the R350 will be .13u but ATI says otherwise (or maybe they were talking about a R300 refresh) . By the time ATI does release a .13u card Nvidia will have had two flag ship products (NV30 and the NV35(just a guess cosidering their history)) using the .13u process.

[Richthofen]

"
When ATI goes to .13 microns with the R350 they will save big because nVidia and TSMC will have resolved the process issues at nVidia's expense. Just like ATI had to resolve the AGP 8X issues with the motherboard manufactures because the spec was new and much more tempermental than AGP 4X was.
"

This does not help ATI in any way.
They have to develop their chip layout for 0.13.
They also have to optimize the 0.13 process for their chip design.

Don't make the mistake and think, well Nvidia works all out and ATI will profit from it.
Forget it guys. There is no company in the world which is that fool.
It won't help ATI in any way because of the reasons described above. If it would help on the engineering side i promise you Nvidia would have contracts with TSMC which will make sure that ATI will not profit from Nvidia's work. Every company would do that.
If R350 is still in 0.15 than it is pretty damn clear that the change to 0.13 is not that easy for other companies like you might or would like to think.
All companies i know have had problems with that 0.13 process. I think even Intel had problems but they started a lot earlier than all other companies and invested billions. That's why it was pretty quit on their front.

[DadGT]

Originally posted by rwolf
When ATI goes to .13 microns with the R350 they will save big because nVidia and TSMC will have resolved the process issues at nVidia's expense. Just like ATI had to resolve the AGP 8X issues with the motherboard manufactures because the spec was new and much more tempermental than AGP 4X was.

Just because TMSC seems to have worked out the process doesn't mean ATI will have any easier time tranisitioning.

The biggest argument for problems other than just process is AMD. AMD recently went .13um with the Athlon with the Thoroughbred core. If you recall, at inital launch of the the 1700+ to 2200+ models, people expected to be able to overclock to the moon. But they couldn't. Now AMD has "released" the 2400+ and 2600+ and those appear to be more overclockable.

What happened? The core design of the Thoroughbred A (1700+) and B (2200+) are different. It was found after manufacture that the chips just would not run fast when they had the A design. They shifted things around, added another layer and made the B design which moves it into the proper realm.

The simple fact is this, every chip designer/maker who has transistioned to .13um has had problems. Don't fool yourself that just because TMSC has a better handle on process issues that ATI will have no problems moving. ATI sure isn't assuming it will be easy. Why else didn't they already make the switch?

[jbirney]

No,

but as time goes on TMSC learns how to make .13 large scale (100million+ transistor) parts better. Its a simple fact that as time passes the more mature the process gets. Of course ATI will have some issues. But they wont have to trouble shoot some of the process issues that NV found by going with the .13 so early.

Also they felt for the time being the .15 process was a safer bet. They were right.

[Spiritwalker]

Everybody seems to be forgetting the fact that ATi has a cross licensing agreement with Intel in some of these areas.

Begs the question, how much has the big I helped in tweaking the R300 and what help might they put in to aide in the transition to .13u?

[PreservedSwine]

Some of the rumours I heard regarding TSMC, NVIDIA, and the .13 delays, have had more to do w/ TSMC needing to re-tool many parts of their machinery. (Mind you, these are just rumours) I would think this will be of some benifit to ATI, as this would suggest that TSMC will be a bit more prepared to hande complications w/ new tools.

[jbirney]

yea all that adds up. I am not say ATI will have a cake walk with the R300/350/400 when they take it to .13u. They will have their own set of issuse to deal with. However they will most likely not have as many PROCESS related issues.

[Dr_Colossus]

If TSMC had to retool some of their machinery they're the ones flipping the bill not Nvidia. While this will help ATI make the transition easier, Nvidia will still have more experience with the process and once a new process comes around (.09u most likely) Nvidia wont have the same growing pains.

[StealthHawk]

Originally posted by Dr_Colossus
If TSMC had to retool some of their machinery they're the ones flipping the bill not Nvidia. While this will help ATI make the transition easier, Nvidia will still have more experience with the process and once a new process comes around (.09u most likely) Nvidia wont have the same growing pains.

err...why would experience with .13 qualify nvidia to not have growing pains with .09? the two have nothing to do with each other.

even if nvidia is not paying financially for jumping to .13 early on, they certainly are suffering time-wise. and since yields are supposedly bad, doesn't that also mean that they have to purchase more wafers than normal?