GlobalFoundries Charters a New Course, Delays 32nm - HotHardware
GlobalFoundries Charters a New Course, Delays 32nm

GlobalFoundries Charters a New Course, Delays 32nm

September was a busy month for GlobalFoundries, the pure-play foundry AMD spun off earlier this year. For now, AMD remains the company's only customer; GF won't begin shipping STMicro products until the fall of 2010 at the earliest. Its current lack of customers hasn't deterred the company from thinking big, even as its slowed the pace on its 32nm process ramp. Earlier this month, ATIC (Advanced Technology Investment Company) announced its purchase of Chartered Semiconductor Manufacturing (Chartered for short). Once the transaction is complete, ATIC will completely own Chartered while maintaining its own major stake in GlobalFoundries. The idea is to play to the strengths of both foundries. According to the PR: "The transaction will allow ATIC to build on the complementary platforms of Chartered and GlobalFoundries, with Chartered’s customer relationships and capabilities in both 8-inch and 12-inch fabrication, and GlobalFoundries’ advanced technology expertise, capacity profile and global footprint."


Chartered's main production facilities. More information on which fab is which is available here.

The two foundries currently have distinctly different foundry assets. GlobalFoundries' two fabs are referred to as Fab 1, Module 1 (aka Fab 36) and Fab 1 Module 2 (aka Fab 30/38). Both of these facilities produce 300mm wafers; both are scheduled for (or have completed) the migration from 65nm to 45nm technology. Until GF's Fab 2 comes online in 2011 or 2012, GF is somewhat limited in its ability to grow its business.

Enter Chartered. At present, Chartered is the 4th largest independent dedicated foundry, has six fabs open and supports a wide range of process technologies. It's not clear which of these ATIC may (or may not) choose to close or streamline, but the company's Fab 7 is capable of shipping up to 45,000 300mm wafers a month and supports a range of nodes from 130nm to 40nm. Even as ATIC moves to purchase Chartered, GlobalFoundries has been forced to push back its roadmap for 32nm introduction. Originally, GF was planning to tape out AMD's next-generation 32nm parts in the first quarter of 2010. That's since been pushed back by six months--32nm production is now set to begin in the third quarter of next year, or a little under a year from now.While it'll take some time to integrate the two companies, each should be able to benefit from the specialties of the other.

As for GF's 45/40nm bulk silicon, both the low-power and standard flavors have been pushed back into Q3 2010 as well, from an original Q2 early production prediction. Oddly, GlobalFoundries is still claiming that it'll begin working on 28nm bulk silicon production in Q4 of 2010, just three months after its repositioned risk production of 40/45nm bulk silicon is scheduled to begin. If you buy the optimism, AMD will then begin early work on a 28nm SOI flavor in Q1 of 2011.

GF, meanwhile, claims everything is business as usual. "Our roadmap for 32nm SOI has not slipped," according to a spokesman for the company. "Yes, the timeframe for introduction has been altered slightly to the roadmap we showed you in July, but that is not because of any issues with the technology. The roadmap has simply been adjusted to align with AMD's product needs...we have high confidence in our ability to demonstrate the same robust yields and manufacturing capability on 32nm that we have historically had."

If GF actually delayed 32nm introduction to "align" with AMD's product needs, it implies that AMD is either waiting to transition until it's 45nm technology is fully developed, or that Sunnyvale might be concerned about footing the bill. AMD still claims it will return to profitability in Q4--CEO Dirk Meyer may feel the company needs a few quarters in the black before it jumps for a new process.
0
+ -

"CEO Dirk Meyer may feel the company needs a few quarters in the black before it jumps for a new process."

What about falling behind Intel?

0
+ -

Gibbersome,

Right now, AMD can't afford to aggressively compete with Intel; it may not be in a position to do so at all until its next-generation Bulldozer/Bobcat architecture is ready.

If you can't aggressively compete, you trim costs, stay in the ballpark, and do your best to compete on the margins--backwards compatibility, support for a wider array of RAM standards, performance-per-watt, etc.

0
+ -

Yeah...

28nm SOI will never exist. That's fairly common knowledge. CPUs are produced on full nodes, which on IBM and Intel processes are 45-32-22 etc. TSMC used to be the same, but changed because of the 40nm process (which is a full node but initially would've been a half node...they skipped 45nm and adjusted their roadmap accordingly). For them 32nm is a half node and 28nm will be a full node. Hence, it's safe to say Evergreen and Nvidia refreshes will use the 32nm half-node at TSMC. Optical shrink, no major redesign, yadda yadda.

Other than that, I guess it's safe to say ATi will stay on at TSMC for R9xx series (28nm) if indeed they won't be manufacturing until Q4. I except R9xx a year from now on 28nm. I know AMD prefers GF, but they're going to do it with TSMC if they have to.

I'm very glad AMD hasn't bitten the bullet, and hopefully will not in the future, to use their...erm...GlobalFoundries, fabs for ATi's Bulk production when their processes are lagging behind. Cost surely must be the reason everything is lagged, and bulk hasn't been introduced for customers yet...They want to at least be on even keel with TSMC when they do, but doing so is uber expensive, especially when you're not making money because you are currently behind in process tech; it becomes a catch-22. You need the latest process to fund the next process, which TSMC is doing, and AMD currently is still required to take the first plunge...so to speak. I suppose the they hope the 40/45nm bulk process will retroactively fund 32nm SOI and 28nm bulk as it will be cheaper and more commonly used, but know they need the most current process as well to propel them forward.

I wish they'd just eat (skip) one step and go from there. I guess they ate 32nm bulk, as that looks like it isn't happening, but it looks like they may need to eat 28nm too, as from that roadmap it will ship the same time TSMC does 22nm...and if so, unless their gate-first tech is marginally better than TSMC's tech, they will continue to fail. They need 28nm shipping BEFORE TSMC's 22nm is a realistic tape-out option or they're screwed. That could happen. It could also NOT happen.

They don't need to worry about catching up to Intel, they'll surely ship 32nm SOI before Intel does 22nm...They need to worry about the whole point of the GF venture; producing silicon for companies OTHER than AMD, whom use bulk and not SOI silicon, and that means arriving to market quicker or at even keel with TSMC. That is something this revised roadmap (if 28nm bulk in Q4 is true) still lacks, and in my estimation will cause GF to continue to falter sans leveraging the acquisition of Chartered's customers, and allowing them some extra fab space.

I sincerely think the reason they bought Chartered, other than for customers, is because they can now have one location with tech tied to the lagging SOI process tech tied to AMD, and rapidly shrink the bulk tech to keep pace with TSMC in a separate facility. They HAVE to do this to survive.

I truly hope IBM's 22nm process doesn't suck, is on time, AMD uses it's SOI derivative quickly in cpus and ATi for it's bulk. They really seem to be betting the whole farm on that tech (GF in NY state). I hope it pays off and is as good or better than what TSMC has to offer at that time...or it could end in catastrophe.

0
+ -

I don't understand most of this post and the comments too being a certified 'Old Fart'. But however it all ends up, I hope that AMD stays in business because the CPU world without them would be terribly expensive. The same goes for Video Card Maker ATI's health and well being.

These fields need to remain as competitive as possible to reign-in higher prices. Without them, we would already be bleeding out.

0
+ -

The whole reason why AMD sold it's Fabs to ATIC to make GF is exactly for that reason. The technology may face some setbacks, but as AMD is now a fabless designer, they don't have to worry about that. Their staying in business is based solely designing and selling CPU's and GPU's and not on making the CPU's themselves.

0
+ -

Turtle,

I think you make some good points, but "skipping" processes, as far as I'm aware, isn't that simple. For example: When 45nm deployments began, everyone *except* Intel moved to immersion lithography. Intel elected not to do this and pursued double patterning instead. Part of the reason it took AMD longer to launch 45nm was the need to install and vet the IL equipment.

So now GF is ahead, right? Wrong. Now it looks as though both double patterning and immersion lithography are required for 32nm.

Login or Register to Comment
Post a Comment
Username:   Password: