TSMC skips 22nm in favor of 20nm....

[Dark4ng3l]

http://www.fudzilla.com/content/view/18450/1/

If this is accurate it will have a significant impact on video card offerings in the next year and a half at least...

At this point I would expect ATI to ditch TSMC to go straight to GF. As to what Nvidia does after 28nm remains to be seen but they might have to switch too to keep pace. This looks to me like TSMC is desperate to keep a competitive edge.

 

[BD231]

I guess we'll all be skippin right along then won't we ...

Makes sense though, GPU's are evolving fast so why not jump ahead within the same time frame it would take to get any other new process out the door. As for ATI and GF, that's a tough one to call; AMD dose a lot with TSMC.

 

[sandorski]

Within a Year they'll have skipped to 0nm, then what will they do?

 

[Voo]

So 20nm is the half node between 22 and 16nm? Haven't heart anything about that before..

 

[ViRGE]

This can't possibly go wrong...

 

[marsbound2024]

They are having problems with 40nm process and they already want to talk about 20nm? I hope they don't push it too fast.

 

[crazylegs]

In other news... TSMC renames 22nm -> 20nm!

 

[Lonyo]

Maybe they will be able to improve their track record when they do their 28nm process, which is them skipping 32nm.
Sure, 40nm is crappy, but that's pretty much a one off, and 20nm is still 2 process generations away, so maybe they will sort themselves out and do 28nm without major setbacks and pull off 20nm as well.

 

[Lonbjerg]

TMSC's 40nm is what Intel calls 45nm...fluff-words are cheap these days...PR FTL

 

[Lonyo]

TMSC's 40nm is what Intel calls 45nm...fluff-words are cheap these days...PR FTL

Pretty much.
Once they used to do actual half nodes, now they are just doing regular one step changes and claiming they are half nodes of a smaller size (how much smaller exactly is debatable).

Once we got 180 -> 150 -> 130 -> 110 -> 90 -> 80 -> 65 -> 55 then it switched to -> 40 (claimed) -> 28 -> 20
And for CPU it was 180 -> 130 -> 90 -> 65 -> 45 -> 32 -> 22.

Looks like TSMC are giving up on doing all the half nodes (cost reasons?) and just sticking with the regular nodes and PR-ing them.

 

[Hacp]

I thought TSMC's 40nm was actually 50nm on average.

 

[OCGuy]

Maybe they will be able to improve their track record when they do their 28nm process, which is them skipping 32nm.
Sure, 40nm is crappy, but that's pretty much a one off, and 20nm is still 2 process generations away, so maybe they will sort themselves out and do 28nm without major setbacks and pull off 20nm as well.

I thought 28nm was already having problems, which is why AMD is not going to get to release "Northern Islands" like they wanted to.

 

[RussianSensation]

We can't get 40nm with good yields (creating shortages of ATI and NV cards), let's do 32nm...no scrap that....let's do 28nm (end up being behind by 1 year but no one cares because there is little choice in the industry), suggest to move to 22nm.....no scrap that, make it 20nm. So essentially going from unreliable, poor yielded 40nm to 20nm (with no certainty of being successful at any node in between). Sounds like a realistic plan!

 

[Genx87]

Are these guys going to become the Bitboys of the fab world?

 

[Stoneburner]

Whatever happened to UMC?

 

[Fox5]

Whatever happened to UMC?

Bought by Global Foundries I think.

If TSMC doesn't get their game together, GF is going to eat their lunch in the performance sensitive market. They're just lucky GF doesn't have the fab capacity for the rest of their products.

 

[Lonyo]

Bought by Global Foundries I think.

If TSMC doesn't get their game together, GF is going to eat their lunch in the performance sensitive market. They're just lucky GF doesn't have the fab capacity for the rest of their products.

GF bought Chartered, UMC is still #2 for contract manufacturing.

 

[SunnyD]

Haven't they "skipped" the last 3 or so nodes already?

 

[nitromullet]

If TSMC doesn't get their game together, GF is going to eat their lunch in the performance sensitive market. They're just lucky GF doesn't have the fab capacity for the rest of their products.

IIRC. right now TSMC is the only game in town for building something like Fermi on 40nm.

 

[Genx87]

IIRC. right now TSMC is the only game in town for building something like Fermi on 40nm.

My understanding is TMSC will do bulk silicon where GF does SOI. GPU design from what I understand lends itself better to bulk silicon.

 

[Fox5]

My understanding is TMSC will do bulk silicon where GF does SOI. GPU design from what I understand lends itself better to bulk silicon.

Maybe, but nvidia is a member of the SOI consortium.

I think SOI is just a cost issue. It's more expensive to implement, and requires more design work to get working well. Products with short design cycles don't want to do SOI, but if nvidia is going to be happy with Fermi for a few years, SOI might be a good idea for them.

Plus, GF will have their own competing bulk processes, even if they're not on the same nodes, they're close enough.

 

[SHAQ]

They lose money when they skip. You miss out on the opportunity to charge customers more times. I guess they are really worried about falling behind.