Slashdot Mirror
Next-Generation Chip Fabs
PaulBu writes "As reported in EE Times, a new IBM $2.5B fab will be the first one to 'produce chips using all three of the sophisticated technologies on the industry's bleeding edge: low-k dielectrics, copper interconnect and silicon-on-insulator based transistors' on 300mm wafers. And it runs entirely on Linux! Quote from the article: 'The state of automation in Building 323 is such that 20,000 sensors are used to track wafer lots in front-opening unified pods that are transported from one tool to the next on rails using linear induction motors. The setup resembles an intricate monorail system tuned to millimeter-precision specs. A central control system monitors all stations and tracks wafer lots via 802.11 wireless communications.'"
For they will wreck havoc with your 802.11 control infrastructure.
"An internally developed master software system called SiView controls all manufacturing operations. An IBM spokesperson said the manufacturing execution system is being licensed to others for fab control.
As for the intended output of Building 323, Bijan Davari, vice president for technology and emerging products, said the company has "spent $500 million on process development alone in order to maintain our technology leadership, and we are experiencing a significant recovery via intellectual-property licensing and alliances. Our value proposition is that we are one to two years ahead of the best of the best."
They spent 2.5 BILLION bucks on this fab and the only thing they could think of naming it was "Building 323". That's so weak. How about SupaFab? Fab:TNG? Absolutely Fab-ulous? MegaFab2k2? It's not like this is a super secret government base like Area 51. Come on IBM, have some flair.
-B
"IBM" and "Flair" are two words that just don't go together.
Oops. I was wrong Google says there are 12,100 hits.
"Live Free or Die." Don't like it? Then keep out of the USA
Uh... do you have any idea how much fabs cost? Six years ago a state-of-the-art fab, which was designed to manufacture nothing smaller than 0.15 micron transistors (and 0.25 was top notch at the time) cost nearly $1.5B.
Once in full production the fab paid for itself in under 9 months. Amazing what happens when fabbing lots (a lot is 12 or 24 wafers, at least where I worked) that have a street value of $250,000.
Chip costs won't rise. They'll continue to fall, just as they always have. Building a fab is indeed a large investment, but if you have the money to invest then it's one that'll pay for itself in a very short amount of time.
Frankly, $2.5B for a 65 nm (aka 0.065 micron) fab is a good value. Sure, if they're starting off with 150 nm or 130 nm equipment they'll have to replace nearly everything to go down to 90 or 65 nm, but that's probably less than a billion per cycle. Equipment is no big deal -- the building itself is a huge deal. Getting all the tolerances tight enough for 65 nm work costs a LOT of money.
According to rumours, IBM will unveil a PPC-based desktop processor - something like a Power4 Lite - on October 15th. Some people speculate that Apple will ditch Motorola in favour of IBM and get the new breed of processors from them, since Motorola is lagging behind and doesn't seem to like having Apple as customer (apparently they got burnt when Jobs killed the clone market).
:)
So perhaps they will fab the next-generation (G5?) processor for Apple there. I at least hope so
-- The plural of 'anecdote' is not 'data'.
Let's start off by saying that I like Linux and I think that it is great. It sounds like IBM did some fantastic things at this plant and I applaud the innovation.
/.). The reboots are due more to external factors than the box needing it. Reliability is not an issue in the Windows based systems that I build.
The Windows system fails after 6 or 7 day? I work with Industrial controls all the time. As I write this, I am working on an NT based server that monitors chemical production. It has only been rebooted 4 times in the last year (I'm waiting for a backup to complete so I can change tapes hence the time to cruise by
If the Windows based system failed after 6 or 7 days then they f'ed something up. There are a lot of things that you can blame on Bill Gates but I don't think that is one of them.
I think that it is great that they are using Linux. I would like to see a lot more of this type of thing. I'd love to take a look at what they have done, but the crap about the Windows system failing is FUD. It smells just as bad coming from the Linux crowd as it does coming from MS.
As others have pointed out, the system is for moving wafers, not loading them into the machines. This is nothing new -- I worked at Texas Instruments several years ago and they had a rail system moving lots around the fabs, keyed to barcode scanners and a Unix backend (we used Solaris on oodles of Sparc 5's).
Honestly, it's not clear from the article if the rail system does end-to-end transport, or if it's just a lot shuttle. At TI it was just a shuttle - you'd ask for the next lot to be processed for a particular machine and the system would retrieve the lot and move the tray to you. A technician would pick the basket up off the rail and then use vacuum wands to move the wafers into the loading mechanism for the machine. Once processing was done, vacuum wand the wafers back into the basket and place it back on the track.
This process is error prone -- TI would only hire technicians with at least a high school diploma, but it's still human intensive and distractions can (and did) cause problems. Grab the wafer by the wrong side? Toast. Vacuum seal break while moving the wafer? Shatter. Drop the basket? Many shatters. Accidentilly forget which wafers have been processed already (many of the machines could only load 5 or 10 wafers, and a lot was 24 wafers)? Bad things happen when you double-dope or double-etch wafers.
If IBM's new automation system is end-to-end, meaning that the rail system somehow automatically loads and unloads the wafers to/from machines then that's a real advancement. It would allow you to eliminate 80% of the humans from inside the fab, and humans are one of the primary causes of particles. When you start talking about 65 nm processes, you have to seriously consider eliminating humans as much as possible from the environment. Or at least having them wear self-contained suits -- hair, skin, and clothing all shed humongous particles at a frightening rate (to a silicon wafer that is). And don't even think about being a smoker.
A technician would pick the basket up off the rail and then use vacuum wands to move the wafers into the loading mechanism for the machine. Once processing was done, vacuum wand the wafers back into the basket and place it back on the track.
You must have been in one of the older fabs. There are two industry standard automated wafer carrier pods used these days: SMIF and FOUP. SMIF (Standard Mechanical InterFace) is used for 200mm wafers, and FOUP (Front Opening Unified Pod) is used for 300mm wafers. The pods are sealed from their environment and are not opened by fab technicians under normal circumstances. The overhead tracks run directly to each machine in the fab, and each fab tool loads the wafers directly from the pod without human intervention.
A major benefit of all this is that the wafers never enter the cleanroom air - they only encounter the air in the pod, and the air in whatever tools they enter. As a result, the air in the cleanroom doesn't have to meet such a high spec, which leads to big savings on air scrubbers.
Accidentilly forget which wafers have been processed already (many of the machines could only load 5 or 10 wafers, and a lot was 24 wafers)? Bad things happen when you double-dope or double-etch wafers.
This is the reason behind the wireless control system. Old fabs use paper-based flow logging, meaning that each wafer lot has a paper attached to show where it has been and where it has to go. Did I mention that this is special (read: expensive) cleanroom paper, because regular paper flakes off lots of particles that are a no-no in the cleanroom environment? In modern fabs, the SMIF and FOUP pods have electronic tags that carry all the information needed to process the wafer lot - the recipe for which machines it has to go to, what to do when it gets to the machine, notes by technicians, etc etc.