Global Foundries' secret is APM
THE QUESTION Global Foundries has been asking itself over and over again since it became an independent fab firm is "how do you outgun a competitor with 10 times more resources and engineers?" How will the firm overtake UMC and take on giant Taiwanese chip shop TSMC? The answer, according to senior vice president and general manager of the AMD spinoff's Dresden fabs, Jim Doran is "you get smarter."
Getting smarter, however, entails rather more than wooing journos with fancy Powerpoint presentations at media summits, and Global Foundries knows it has to be nimble on its feet if it is even going to come close to competing with Intel and TSMC.
Much of the secret sauce that will enable the company to compete, says Tom Sonderman, GloFo's veep of manufacturing systems and technology, is in Automated Precision Manufacturing (APM) and yield management systems.
"GF needs to be agile" says Sonderman, emphasizing APM's alleged speed and accuracy in responding to manufacturing issues and its ability to help customers fine tune the process to their specific needs.
It's "like putting an EKG on your process tool," Sonderman told us, adding that all of Global Foundries' critical tool sets are hooked up to the APM framework, which constantly monitors their health as well as the health of all chips in production.
After gathering up information about how things are ticking along, APM can then purportedly use that knowledge to make automatic adjustments to the manufacturing process, making sometimes small but often critical changes to yield a better batch of chips.
Finding the best yield to performance ratio is the challenge, according to Sonderman, whose ideal scenario is one he describes as "living on the edge", balancing optimal yield with optimal performance.
Sonderman reckons Global Foundries already has a "rich and powerful set" of its own yield and engineering data analysis tools, and that the Dresden 300mm fab is already considered "world-class" in performance. In fact, added Doran, Dresden is "as good as it gets in the semiconductor industry at the moment."
"Yield management systems are where a lot of the secret sauce is", explained Sonderman, adding that such systems are capable of quickly identifying any yield problems which may have cropped up and fixing them. This, said Sonderman, was greatly helped along by something called dynamic wafer level sampling, which uses wafer processing history to decide which wafers will be most useful in providing the relevant info.
"Customers are are going to feel they can get their hands on and customise the factories", added Doran, claiming "it's really all leading edge technology, no trailing edge stuff".
Sonderman says TSMC doesn't have APM, and this is a huge competitive advantage for Global Foundries. Intel, admits Sonderman, probably has aspects of the technology "but I dont think they've integrated it the way we have," he said.
He added, alluding to Intel's well-known technological arrogance, "if you asked them, they'd probably say they had all of it". µ
Share this:
ShareBusiness Intelligence Software
Customer Relationship Management (CRM)
Enterprise Accounting Software
Enterprise Resource Planning (ERP)
Enterprise Systems Management
If you are smarter aren't you supposed to keep that a secret instead of telling everyone why.
Looks Fancy.How About Secret LunchWagon in N.Y. FAB, Ummmm, kind o' Grab ibm Team As wonder About FishKill. Ah, Haw. Screams Uniformed ALL Meat Beef Lingua Burito Server: "You've Got IBM Hands". (Straight From Prof. HILL) That'll Do IT. GRAB IBM & Phoney Excuse- Charttered Knows Better & WaLLA, Global Supreme HQ. Signed:RoadRunner Beep,Beep.
Why should APM work any better for GloFo than it did for AMD?
Asset smart.. Anyone?...
Globo is getting good at PPT presentations though! Take that UMC, TSMC, DAWG
Three questions:
1. What SPECIFIC, QUANTIFIABLE benefit does APM give? $XMil saved through better yield, Y% better bin splits? You will never get anything like this, because it doesn't exist and AMD, like IBM, likes to tout 'general' improvements without actually quantifying them. Remember how much better the 65nm process was over 90nm... how'd that work out in real life?
2. What, specifically, is unique about this system? EVERY ADVANCED FAB uses some ofr of feed forward control, tool level sensor sampling, dynamic wafer monitoring.
3. Doesn't APM go against what you want from manufacturing? The goal of any manufacturing process is stability - the goal is to develop as wide a process window on your tools, so you don't need to 'tweak' the process from wafer to wafer and lot to lot. It makes the ability to fight excursions (problems) in the fab very difficult if every wafer can have a slightly different process - it makes troubleshooting very difficult as you now have a multitiude of variables to explore.
Without any actual specifics on real life benefot, other than these vague generalities - APM is just a PR smokescreen that sites like the INQ publish every year or so when AMD decides to roll out the press again.
as a general rule, the one trying to catch the process technology leader is the one struggling with yield.
What a great and telling analogy. How often is an EKG used when you have a healthy and stable heart? You use an EKG when you are having problems and you don't use it to monitor your heart CONTINUOUSLY unless you have some serious problems! What a great analogy, I'm sure the VP had no idea what he was saying. I'll be looking out for all those healthy people who have EKG's hooked up to them continuously, I must have missed it - where I'm from generally you see these in hospitals hooked up to sick people.
By the way - there is NO WAY you can monitor all chips at all times. I assume this is a typical INQ, re/mis-interpretation of something slightly different. With hundreds of chips per wafer, you simply can't do this. Instead specific locations on a wafer are chosen at the various process steps - typically areas that are most variable for a given process, and as mentioned in the article not even EVERY WAFER, yet alone every chip is monitored.
And I don't think Intel would 'arrogantly' say they've already implemented all of this as the VP whines about. Instead I'd think they'd point out, the idea is to have a process not designed to be on a cliff where all this automated twaeking is needed to get yields up. The goal of manufacturing is not to tweak, it is to repeatably make the same part over and over.
The more you have to tweak - whether by hand or dressed up as an advanced and automated process like APM, the more unstable and variable your process is.
This whole notion of APM is a bad reshuffling of the Toyota Method that prof Christensen and others have been touting around.
I concede that APM minimizes inventories, and reduces the "failure rate", however its have never been applied in the context of a fabricator for hire. Nothing indicates that this secret sauce will have any impact on the cost structure of GF which is utterly out of whack due to the sub utilization of its production capacity in Germany.
Toyota benefited from APM because their overall costs are dominated NOT by technology but by aborted products, distribution, recalls and after sales warranty costs? Clearly better yields result in a cascade of cost savings, this is why the Toyota APM method works! GF cannot benefit from APM since all these costs are not present in their model, only capital costs and utilization factor are relevant.
Why would any cost conscious fabless company switch to a leading edge startup fabricator that does not understand its new business model?
Clearly Fab 2 becomes by the day a greater collective delusion likely to become a failed reality as long as the losses in Saxony are not reigned on very, very soon.
This APM mumbo jumbo is simply an extension of Toyota, In-line Closed Loop Adjustments, Power Point Drama and huge lack of sense of reality of a desperate management team.
Keep in mind that since Fab 2 is purely imaginary, no TSMC customer will desert a successful tangible reality to explore the benefits of a mirage of a promised land.
How many managers would want to risk their careers in pursue of dreams, subsidies, massive losses and public relations mumbo jumbo?
Is this "Adolfo Gutierrez " the same as this one who is so dedicated and blogging on this all over the place?
http://albany.bizjournals.com/albany/stories/2009/02/23/editorial1.html :
"..Adolfo Gutierrez makes no secret of his strong opinions about the deal to get a chip plant built in Malta.
Gutierrez adamantly opposes the Advanced Micro Devices/Foundry project.
Gutierrez is a prolific writer in the comment sections of newspaper and trade publication Web sites..."
"Gutierrez suggests the money would have been better spent on Intel. That would be a sure bet, he says."
...
"“What was a promise has become a delusion.”" <amd
From letter
"GF cannot benefit from APM since all these costs are not present in their model, only capital costs and utilization factor are relevant."
Not so. GF cannot charge its customers over market rates, its utilization rate is not its customer's problem. So gains in quality and faster ramp to yield do matter. It can only make itself financially viable with more customers. Do fab customers pay some NRE costs? If so then perhaps GF could absorb much of any risk by spreading most of it in the unit costs. It's customer's risks are apparently reduced somewhat by GF being in IBM's fab club. Malta is even physically close to IBM.
GF seems solidly financed at this point by motivated people. I can't know if GF will prove viable over time, especially through the fog of present economic troubles. Intel's attempts to get x86 license payments from a fab seem odd since they didn't do that when Chartered was set up to make x86's. It seems unlikely that yet more FUD, however, will shed much light on any questions.
Stop talking like a freakin' idiot. Or are you just trying to sound cool?
If you study AMD and their manufacturing history - you will see that APM has been a key factor in AMDs ability to keep high yields and production costs low.
It is probably only APM that's been keeping AMD afloat when they where using smaller wafers and was one tech generation behind Intel.
AMD has improved APM several times to keep up with new production methods/technology.
The automated process makes it possible to get as much bang for the bucks as possible. Now they need more customers to keep the production as high as possible - so the technology investment is divided by as many wafers as possible.