AMD to Turn to TSMC for ''Bulldozer'' Manufacturing

[btarunr]

AMD is rumored to be seeking ties with TSMC, Taiwan's premier semiconductor manufacturing foundry, for future manufacturing of its "Bulldozer" architecture processors, according to a report by DonanimHaber. This has two very distinct implications: first, AMD could be facing issues with GlobalFoundries 32 nm HKMG node, its de facto foundry for CPU manufacturing, and second, this could just be an obvious development of future low-power APUs based on the new x86 architecture being manufactured at TSMC, much like how current E-series and C-series APUs are.

Then again, AMD doesn't exactly have any APUs in works that use "Bulldozer" architecture for the x86 cores, rather, its successor codenamed "Piledriver". Another couple of important things to note here are that TSMC does not have a 32 nm bulk node (it was scrapped with the transition to 28 nm bulk), and its HKMG (high-K metal gate transistor) manufacturing technology is deployed rather recently. It would be interesting to follow this development.



View at TechPowerUp Main Site

 

[caleb]

Didn't AMD move their manufacturing 2 years ago or so?

That donanim hebel knows a lot of AMD rumors...

 

[Fourstaff]

Didn't AMD move their manufacturing 2 years ago or so?

That donanim hebel knows a lot of AMD rumors...

No, they are still largely using GF for their processors, only their graphics processors are done by TSMC.

 

[twilyth]

tsm is always a couple of nodes behind. Since a lot of fab expertise probably still resides at AMD, this could be like striking gold for tsm. They get the knowledge to upgrade their fabs with a lot less pain and a much shorter learning curve.

GF has to be able to survive on their own and maybe they see the best chance of doing this is inconsistent with chasing the bleeding edge in fabrication.

 

[Completely Bonkers]

AMD is clutching at straws... if they can't deliver a significant realignment in market perception of CPU performance (and they can't in the short term... there is nothing in the pipeline except PileDriver and that will only win about 10% performance gain) then they need to go for POWER PER WATT, ie take the existing architecture and get it onto a smaller chip, make it cheaper, use low-leak gates, and therefore run much lower power... or RANK UP those GHz.

It's the right and only defensive move they can take right now.

Good luck to 'em

 

[Sihastru]

tsm is always a couple of nodes behind. Since a lot of fab expertise probably still resides at AMD, this could be like striking gold for tsm. They get the knowledge to upgrade their fabs with a lot less pain and a much shorter learning curve.

GloFo actually only looks ok on paper, in reality they are in over their heads. TSMC on the other hand already build some of AMD's earlier APU's (E350/E450 and C series). The Llano APU's (A series) are in short supply now because AMD tried GloFo once again and they can't really deliver the market required amount.

It's also very clear that AMD wanted higher clockspeeds out of Bulldozer at launch (4.2 - 4.6) to actually be more competitive in the market, but I guess GloFo messed up again, and we can all see the sky-rocketing power consumption of a Bulldozer FX chip when overclocked around 4.6 GHz.

So AMD promisses that soon-ish they will improve Bulldozer's performace numbers by up to 50%, and I really think this claim has EVERYTHING to do with migrating production to TSMC.

 

[Bundy]

AMD is clutching at straws... if they can't deliver a significant realignment in market perception of CPU performance (and they can't in the short term... there is nothing in the pipeline except PileDriver and that will only win about 10% performance gain) then they need to go for POWER PER WATT, ie take the existing architecture and get it onto a smaller chip, make it cheaper, use low-leak gates, and therefore run much lower power... or RANK UP those GHz.

It's the right and only defensive move they can take right now.

Good luck to 'em

Yes I think this is marketing to try and distract away from bulldozer

 

[HalfAHertz]

It's in AMD's best interest that Glo Fo is successful, after all they won a huge percentage of the shares...

 

[phanbuey]

It's in AMD's best interest that Glo Fo is successful, after all they won a huge percentage of the shares...

Thats such a bad thing for AMD - means they are more or less stuck with Glo Fo for better or worse.

 

[bear jesus]

I would be interested to see what TSMC could do although i would not expect a lot.

It's in AMD's best interest that Glo Fo is successful, after all they won a huge percentage of the shares...

The last i read AMD only has 14% of the shares, i would not really consider that a huge amount.

 

[Fourstaff]

10% to 14% is a good improvement if you ask me

 

[nt300]

This is interesting can it be that AMD is quickly moving it's high end line to 28nm bulk HKMG and leaving 32nm for low to mid range?

 

[xenocide]

This is interesting can it be that AMD is quickly moving it's high end line to 28nm bulk HKMG and leaving 32nm for low to mid range?

Unlikely. You can't just jump like that, it takes a lot of investment to move to a smaller process. As it is, these lower processes are still kind of works in progress, so trying to get a whole line of Consumer CPU's out the door for your customers on a still semi-experimental process is a gigantic risk. That's basically what GF did with 32nm Bulldozer.

 

[WarraWarra]

So BD is the reason why the last AMD guy in charge was fired, they knew back then already it is a "BD HKMG APU" and not a BD desktop CPU.

AMD retire this BD batch to APU's or midrange cpu's, adjust their power and speeds to match midrange cpu specs then cook a new batch for performance cpu's and name then BD2 / BD Gen2 for Feb/March 2012.
IE: BD goes to best power / performance capable and labeled as hard core mid / APU then do BD gen2 with fixes for desktop performance.

Same as Half baked Sandy-e's and full baked Sandy-e's for next year, we can forgive you AMD, just recover like this.

AMD we are still rooting for you, just hope you can quickly recover from your BDangover "hangover" .

LOL AMD likely did not catch the old saying, do not keep all your eggs in 1 basket or AMD translation, do not keep all your CPU's in HKMG.
Time to do your own foundry's in Canada, Brazil, Australia or EU.

Is this poor HKMG performance or is it another insider sabotage in HKMG ?
LOL who am I kidding it is not like AMD/Intel spends time reading decent website like this one.

 

[Casecutter]

But didn't we just get told TSMC was bulging at the seams and couldn't keep up with capacity/ demand... and they would be raising prices?

This really just appears to be FUD, why does it seem when this donanim... "herbal" speaks it's like some enlightened maharishi.

Move along nothing to see here!

 

[techtard]

Wow, so they ditched GloFo for the TSMC? Weren't these guys having all kinds of problems the last few years?

AMD looks like they are actually trying to hit rock bottom.

 

[zithe]

As much as we can speculate and gather from articles, we don't know what AMD knows.

 

[[H]@RD5TUFF]

Does this mean BD will not suck after this ? If not why does this matter ?

 

[mastrdrver]

There is so much Fud there it is unbelievable.

As far as I know, TSMC does not even have a 28nm SOI of which AMD currently designed BD on and is using GF 32nm SOI. TSMC only produces on Bulk which AMD uses for their GPUs and Bobcat APUs.

This would be like moving a Chevy Corvette from it manufacturing plant to the Ford Taurus plant. It just is not going to work without a lot of reworking of the car or assembly line.

This is one of the reasons AMD didn't move their GPU production to GF this last time because they were designed for Bulk. GF did not offer Bulk, so AMD would have had to redesign the chip to accommodate SOI production. No worth the time since it would take about a year just to get something down the line and back to find out if the initial silicone (A0) worked or not and get it out for production (given that there were no major bugs). By that time AMD will have moved on to 28nm and it would have been a waste of money and human resources that could have been better used for something else.

 

[Peter1986C]

Move along nothing to see here!

That made me think of this guy:

 

[Super XP]

GF used to be AMD's fab, there is no way they would move the 32nm to TSMC. They already have issues of there own. GF needs to get 32nm working 100% and fast.

 

[NC37]

Wow, so they ditched GloFo for the TSMC? Weren't these guys having all kinds of problems the last few years?

AMD looks like they are actually trying to hit rock bottom.

You mean the mess after mess NV had to deal with because of them?

Sure outsourcing to these other companies might be cheaper, but with as often as they are in the news for mistakes and screw ups, think they'd be better off doing it all themselves.

 

[Fourstaff]

You mean the mess after mess NV had to deal with because of them?

To be fair, AMD used the same node and their 5xxx turned out fine. Nvidia was expecting a smooth production, but was shafted. AMD accounted for problems, and came out with a stellar product.

 

[Horrux]

You mean the mess after mess NV had to deal with because of them?

Sure outsourcing to these other companies might be cheaper, but with as often as they are in the news for mistakes and screw ups, think they'd be better off doing it all themselves.

That might be true, and to quote Jerry Sanders, former Chairman and CEO of AMD: "Real men own fabs". However, the cost of the ever newer and better fabs increases rapidly for each new process node, to the point where the fabs were worth a huge chunk of AMD's share price. That was the reason they spun out their fabs a few years back. The fabs should be self-sustaining in and of themselves.

As of now, AMD could not possibly afford to build a next-gen fab or buy a current-gen one. They need market share for that, and to get that, they need a good product. Chicken and egg. I don't know what they did during the Athlon/Athlon 64 days but it seems they missed out on doing the right things back then and today they are paying for it.

 

[NirXY]

That might be true, and to quote Jerry Sanders, former Chairman and CEO of AMD: "Real men own fabs". However, the cost of the ever newer and better fabs increases rapidly for each new process node, to the point where the fabs were worth a huge chunk of AMD's share price. That was the reason they spun out their fabs a few years back. The fabs should be self-sustaining in and of themselves.

As of now, AMD could not possibly afford to build a next-gen fab or buy a current-gen one. They need market share for that, and to get that, they need a good product. Chicken and egg. I don't know what they did during the Athlon/Athlon 64 days but it seems they missed out on doing the right things back then and today they are paying for it.

yes separating the fabs from the design is a good thing if you plan to outsource those fabs, i'm not sure if GF has any other clients then AMD, but even then the capacity will probably not be enough.

though i'm sure many people will disagree, but I still thinks that purchasing ATI was a big mistake, at $6.4B (if memory serves correct) it was a huge risk, and it eventually caused the spin-off, and less investment in R&D further down the road.

yes Llano is nice, but is it all AMD is out there for ?