AMD Breaks Ground on New Chip Facility

philthedrill writes "AMD announced that they have broken ground on Fab 36, which again will be located in Dresden, Germany. The 300 mm fab is expected to start volume production in 2006. There's more information at CBS MarketWatch." AMD will be moving from its current 200 mm wafer process, and looking to save money through the higher efficiency of the new process, as well as keep up with expected demand for their next generation processors. The MarketWatch article also contains some speculation about probable partners for AMD.

Comments

size

[Anonymous Coward]

Why don't they skip to 10 foot wafers?

Re:size

[aardwolf204]

The better question is, why dont they switch to SQUARE WAFERS! Take a look at any of Tom's articles on new AMD cores and you'll see like 20% wasted space.

oh, and 10ft? Pfft, look at laserdisc, people want small ;)

Re:size

[philthedrill]

Why don't they skip to 10 foot wafers?

I know you're all joking, but I've actually been asked this sort of question a few times during my days when I did work in a fab. I started of on 6" wafers until we converted to 8" (200 mm) wafers. One of the challenges was film uniformity for deposition processes. Slight variations in film thickness can result in varying yields.

The other problem with 8" wafers was that they were a lot more sensitive to stress. Without going into too much detail, we'd have spare/scratch wafers that would go through some processes repeatedly (kind of like the seat filler at the Academy Awards). Certain processes/films such as thick silicon nitride would cause a lot of stress on the wafer, and they'd fracture, although we never saw that on 6" wafers (I put in a suggestion to mitigate the problem, it got accepted, and I got two free movie tickets out of it for saving the company millions). So this sort of problem could get worse as wafer sizes increase.

The better question is, why dont they switch to SQUARE WAFERS!

I guess the best way would be to make you watch a video on the wafer manufacturing process.

As for the article post, I apologize for making it vague. I could have clarified it and saved everyone some confusion... but then it'd be less interesting :)

Cost of labor

[ziggyboy]

Wouldn't it be cheaper for them to put facilities that mass produce chips in countries where labor is cheap? Most Intel chips I've seen are marked "Made in Malaysia" or "Made in the Philippines."

Re:Cost of labor

[stetsds]

In an interview one of the AMD managers said that worker skill was more important than cost in this case. AMD already has a chip fab in Dresden, employing about 2000.

Re:Cost of labor

[switched4OSX]

It might be cheaper, but poor quality control tends to go hand in hand with cheap labor.

Re:Cost of labor

[Barbarian]

I for one am happy to see AMD expanding in a first world country, and employing first world workers, against the trend of sending everything overseas, or of shipping in cut-rate employees from the third world.

Re:Cost of labor

[LarsWestergren]

From what I have read, for AMD the choice was between a fab in the US, and in Dresden. I think they picked former East Germany because the previous factory doing well, of the high number of skilled workers who are willing to work for relatively low wages, and the Euro starting to make Europe a more attractive place to do business. So again tech jobs are moving to eastwards, but not quite as far this time. Why they didn't go the Malaysia or Philippines as you mentioned I don't know. Perhaps the current unrest in the world?

Too bad for US and Asian workers, congratulations to the Germans.

Re:Cost of labor

[romanm]

Cost of labor is not the only factor here. By making the company in the Germany the chip becomes an European Union product and is less taxed than the Asian imported chip.

Re:Cost of labor

[WIAKywbfatw]

I doubt they built a fab plant in Germany to keep down labour costs.

As an EU nation, Germany has employment rights that are a lot more stringent than in the US or South East Asia - we're talking about a higher minimum wage, a cap on weekly working hours, sick pay, maternity and paternity leave, pension contributions by the employer, favourable redundancy payments, etc.

More likely is that other economic factors - tax breaks, being inside the Euro zone, etc - were the deciding factors.

Re:Cost of labor

[10Ghz]

AMD chips are packaged in Malaysia as well (that's what the "made in Malaysia" refers to). But the actual chips are made in Dresden. Intel's chips are also made elsewhere (for example, Israel).

Re:Cost of labor

[Anonymous Coward]

Devices are marked with the location of the packaging plant, not the location of the fab that made the actual chip inside. Malaysia, Thailand, the Philipinnes and other cheap labour areas are popular locations for packaging plants, as these used to be very labour intensive (they are highly automated now, but the plants haven't moved). Most fabs are located in more developed countries - US, Japan, Taiwan, etc.

Re:Cost of labor

[Anonymous Coward]

greetings from saxony,

thanks for the congratulations.
according to hector ruiz the main selling points of dresden wasn't wages (cost of labour is one of the highest in germany due to all the taxes) *but* high density of skilled workers, universities (laugh at it - but dresden was the center of microporcessor design in former eastern germany since the 60s or so :) and other chip makers. for instance infineon raised their first 300 mm fab there. of course many tool companies like applied materials are allready present there. (this was one of the drawbacks of SE-asia btw. nobody has done 300 mm there yet)

but the biggest selling points were guarantees for the debts they had to take and a huge chunk of money from the state (saxony) and the federation (frg) - several hundred millions.

also - opposite to the traditional prejudice about german bureaucracy - all the paper stuff was done real quick and without any hassle for amd. but the state of saxony has somewhat of a track record in this regard. you can see all that prejudices justified 100 miles to the north where intel and some shejkh from dubai want to raise a fab too - for several years now. google for communicant and frankfurt.

one more thing - amd *might* have build their factory in china too. but present us law prohobits exporting certain technologies there.

have a good day over there

andreas

Re:Cost of labor

[Barbarian]

Who said anything about race? I'd just like to see companies keeping jobs in the Western hemisphere, as opposed to the recent trends we've seen lately.

Re:Cost of labor

[broeman]

Actually, if he said the opposite, I would call it for some kind of racism (not that I would). Employing poor people in poor countries would "force" people in the rich countries to get more "educated" work (and the rich/poor mismatch would be greater) ... but luckily it is not true, the poor would get more money in the hands by even a sh1tty job, increasing their life-conditions (much better than laying in the streets without food).

Re:Cost of labor

[sbryant]

It's not so much the cost - it's more the cost effectiveness.

I remember hearing a company say that even with the higher employee costs in Germany, it ends up costing the same as places like South America, because the workers are much more effective.

Germans are certainly good engineers - just look at their cars. Germany also has a strong infrastructure, so things like telecommunications and transport are very reliable, and I'm sure that the German government gave AMD

Of course, the other reason for choosing Dresden must be the good protection it offers from natural disasters, such as flooding... :-)

It's nice - for those living in Germany - that AMD chose to build there. A lot of companies are wanting to get out - due to higher taxation and overbureaucratisation (is that a word?).

-- Steve

Re:Cost of labor

[sbryant]

and I'm sure that the German government gave AMD

... an incentive.

Note to self: don't get involved in long discussions in the middle of writing a post.

-- Steve

Re:Cost of labor

[Paul Jakma]

Most likely they received lots of grants from Saxony government (Saxony, or whichever german state it is :) ). They're allowed to do this because Dresden is in the former DDR, and hence needs development, so have a concession from the EU to give subsidies to attract investment. Same reason why Intel have 2 FABs in Ireland and are building a third - because of grants and subsidies from the Irish govt.

Re:Cost of labor

[Paul Jakma]

Intel has quite a few FABs in various parts of the world, including Israel, Malaysia and Ireland as well as USA (cant remember where their FABs are in US. But they did buy DEC's old FAB in Cambridge, MA.)

Re:Cost of labor

[10Ghz]

I'm well aware of that. I just didn't feel like listing them all, so I just used their fab in Israel as an example.

Re:Cost of labor

[Jeff DeMaagd]

Packaging does not mean fabbing. Fabbing is building up and etching the silicon with several layers of different chemicals and processes. Packaging is mounting the die and connecting the pins to a PGA or some other type of package that you see when you buy a chip. A lot of Intel's chips are fabbed in the US but packaged in the countries you named. There was a guest article at anandtech that makes this clearer.

Re:Cost of labor

[4of12]

the biggest market is still the us

Is it?

Where do microprocessors go? I would have guess the EU to be a larger market than the US, and East Asia to be comparable, too.

Re:Cost of labor

[Plammox]

AMD also has factories in Malaysia. It's not the wafer fabs, it's where they assemble the CPUs. Same thing goes for Intel.

Re:Cost of labor

[hughk]

There is still a large wage differential between the former DDR lander (states) and the rest. Dresden has not only been in the chip business (Infineon is also there), it has also been the home to some high precision instrument manufacture and research centres.

There is also an international school there which is a prerequisite to bringing over non-german execs with families. I don't work there myself but a friend has moved there and quite enjoys it. It may cost more than SE Asia, but there are other advantages.

You are right about tax breaks. The former DDR states still do quite well.

Re:Cost of labor

[Daengbo]

Yeah, but the AMD factory next door to me in Thailand (the wall is within 50m of my window) just changed hands / names / was sold / who knows. Just noticed a sign announcing the change about two weeks ago. Maybe that'll stop some of the black soot covering everything in my house. Oh, wait, this is BKK, so someone else will step up and help out the pollution level.

Re:Cost of labor

[ifwm]

You are a moron

Re:Cost of labor

[Hoser McMoose]

As others have mentioned, Intel just has their chips packaged in Malaysia, the Philippines, etc. Similarly AMD has almost all of their chips packaged in Malaysia. FWIW most of Intel's actually chip fabs are in the US. Oregon to be exact. They also have a couple plants in Ireland and in Isreal. To the best of my knowledge they do not have any plants in south-east Asia.

In any case, here's a few numbers for you.. A typical plant these days costs about $2.5 billion dollars to build. Equipment in that plant is about another billion dollars and has a useful life of about 5 years if you can stretch it. The plant as a whole has a useful life of about 10 years before it needs a major overhaul.

So, simple bit of math tells us that you're looking at a fixed capital cost of somewhere around $4.5 billion. This gives you a rate of depresiation of roughly $1.2 million dollars a day.

Now, if you pay an average of $100,000/year to the ~1000 employees of the plant, you're looking at roughly $275,000 a day. As you can see, this isn't all that large of a number when you just compare it to depreciation, let along the (rather high) cost of electricity to keep the plant up and running and the huge cost of all the raw materials for producing and cleaning the chips. In short, your cost of labor is VERY low as a proportion of the whole business, but the quality of your employees has a VERY direct relation to the quality of the product (most importantly yields and speed bins) that is coming out of the fab.

Long story short, you REALLY want to place your fab where you can get skilled workers, not where labor is cheap.

Of course, the determining factor on where the fab is built is always which government gives the best grants and loans to build the fab.

Re:Cost of labor

[VAXman]

Intel's main fabs are in Oregon, New Mexico, California, Massachusetts, Israel, and Ireland (there may be a few others in the US which I'm forgetting).

Intel doesn't have a fab in Malaysia. There is chip packaging in Malaysia, as well as Costa Rica and Philipines.

Re:Cost of labor

[ProtonMotiveForce]

Uhh, buh? Unless you're from Germany, what possible logical sense does that make?

If you're from Germany, then sure - it's great. But if you're from the US then it makes no difference if the job's in Germany or Timbuktu, it's not in the US.

Re:Cost of labor

[Uma Thurman]

Thanks for clarifying your position, you HEMISPHERIST! hehe

Re:Cost of labor

[Dastardly]

Wouldn't it be cheaper for them to put facilities that mass produce chips in countries where labor is cheap? Most Intel chips I've seen are marked "Made in Malaysia" or "Made in the Philippines."

Check out an AMD chip someday they also say "Made in Malaysia". The reason is that, is the location where the chips are packaged. Intel chips are packaged in the Phillipines and Malaysia as you have noted. AMD's microprocessor Fab is in Dresden, but packaging is in Malaysia, and Final Test is in Singapore. AMD has previously made microprocessors in Austin Texas, and before that in San Jose, California. Intel has Fabs all over the world including the US, Europe and Asia, and packaging in at least the two places you have described. Intel isn't as forthcoming on their website about what their sites do.

Re:Cost of labor

[ncc74656]

Wouldn't it be cheaper for them to put facilities that mass produce chips in countries where labor is cheap? Most Intel chips I've seen are marked "Made in Malaysia" or "Made in the Philippines."

Intel doesn't have any fabs in those countries either...most of theirs are in the US, with a few others scattered here and there around the world. The only work that Intel has done in places like Malaysia or the Philippines is splitting dice off of wafers and attaching them to packaging to make a completed microprocessor (or whatever). The wafers are produced elsewhere. AMD does the same thing...wafers get sent from the US or Germany to Malaysia.

I suspect that production of wafers is mostly automated, so the cheap-labor argument for moving fabs to 3rd-world countries is out the door. On top of that, you want skilled people nearby who can fix the machinery if it goes on the fritz.

As for packaging...you would think that would be fairly automated as well. (Could you eyeball the correct positioning of an Opteron die such that all 940 contacts, all of which are shoehorned into an area on the order of 100 mm^2, would line up properly?) Why they send out for packaging is beyond me, especially since you'd think that shipping costs would eat up a fair chunk of whatever they're saving.

Re:Cost of labor

[Dastardly]

Oregon to be exact. They also have a couple plants in Ireland and in Isreal.

Don't forget Arizona and New Mexico.

Also, wafer fab is a very automated process. I have seen some Fab lines, and there are maybe 10 people inside the clean room where the lines are running. Then, there are another 10-20 people in wafer test. Then, maybe a few 10s of people in other locations doing other manufacturing or test processes. But, what you end up with is that you are probably talking about less than 100 people per shift that actually handle or in the case fo the Fab watch devices being made. It is possible I am over guessing, and maybe it is closer to 50 people per shift. The most of the other 800-900 people in a Fab are planners, repair techs, and process engineers. The process engineers though are the ones that make the company money. They are the ones who have to get the yields as high as possible. And, 1% of yield is worth a lot more than whatever salary you pay the process engineers.

Re:Cost of labor

[Dastardly]

As for packaging...you would think that would be fairly automated as well. (Could you eyeball the correct positioning of an Opteron die such that all 940 contacts, all of which are shoehorned into an area on the order of 100 mm^2, would line up properly?) Why they send out for packaging is beyond me, especially since you'd think that shipping costs would eat up a fair chunk of whatever they're saving.

Actually, shipping wafers probably doesn't cost much per die because you get around 100-200 chips per wafer which meand you can ship a thousand or so chips in a 1 foot by 1 foot box. Given bulk shipping costs of hundreds of wafers you are probably talking about pennies per die.

The reason packaging and final test is typically in Asia. Is because packaging and final test is much mor elabor intensive. Instead of a thousand or so chips in a 1' x 1' cube you end up with 100-200 in a 1' x 1' cube. So, every 100 or so chips have to be taken off the packaging or test equipment, and boxed or wrapped, so you could say that packaing and test is about 10x more labor intensive than wafer fab and test.

Re:Cost of labor

[Pharmboy]

There is still a large wage differential between the former DDR lander (states) and the rest. Dresden has not only been in the chip business (Infineon is also there), it has also been the home to some high precision instrument manufacture and research centres.

One of the reasons we build some units in the US is the gap, Germans make MORE than US citizens for our tasks, lower skill, highly repetitive. The German companies we hire to manufacture are farming it out to Muldavia or such, where labor is cheap (again, low tech, highly repetitive).

My experience is that on the mid/lower skill level, Germans make more than the US worker, although it is not likely they get to take it home after taxes.

Re:Cost of labor

[nelsonal]

Intel generates about 40% of their sales from Asia (ex Japan), I think that is not end use, but adds revenue for manufacturers in Asia that export to other areas, about a third from the US, about 20 percent from Europe (including the Middle East and Africa) and the last 10% from Japan. The EU has a similar GDP to the US (the two are within 5% of each other both at about 10 trillion), but buys fewer computers per economic unit. Japan also spends considerably less of their income on PCs (that use Intel chips), with an economy of about 5 trillion.

Re:Cost of labor

[nelsonal]

I think they have a big one in or around Arizona, they have quite a few. A recent 10-K should have a list of them.

Re:Cost of labor

[nelsonal]

To reiterate your point, AMD gave out a plethora of details regardig this fab. In AMD's case the building costs about $300 million, the equipment cost an additional $2.4 billion (this is what mnost of the loan promises are for) and the company plans to employ about 1000 people. Figure a trained EE with microprocessor desing experince costs about what you have, and one with a lot of classwork (but little experince) might go for 20,000-30,000 in India or China, it just doesn't make a lot of sense to spend 2.5 billion on a fab and then try to save the little bit on wages while your brand new shiny equipment is wasted on training purposes while you improve yields. Not that the Indian or Chinese workers couldn't do a great job, but I'll let you risk your company in a place that doesn't already have a fab or two. I know Fishkill, NY and locations in the UK were under consideration. The new fab will be a nearly exact copy of the new IBM fab in NY.

Re:300 mm? I hope that's wrong.

[ComaVN]

300 mm refers to the size of the wafer. One wafer contains lots of chips.

misread the article

[dido]

The article seems to be misleading. Apparently it's wafer size, not feature size, on reading a bit more carefully. ;)

Re:300 mm? I hope that's wrong.

[heliocentric]

Yeah, I was impressed when we were making 6" wafers a few years ago and darn proud of it. I remember back when they were the size of a US quarter dollar.

I worked in final visual inspection on a 6" line and that was very dicy. You'd get someone failing too many parts, whole wafers in some cases due to what their eye saw as too much FLUC. Tricky balancing act since FLUC identification is more of an art than a science (metering thing is a science with acceptable ranges and such) in that we didn't want to ship things that would fail in the wild, but we didn't want to fail too many things and incrase costs (a wafer at the final end has had a lot of effort put into it).

Oh wait, I just re-read your post. Where you meaning sarcasim or did you not understand what they were talking about?

Re:300 mm? I hope that's wrong.

[affenmann]

It's the size of the wafer, the round silicon thing with *many* chips on it.
See this link (to Intel, inappropriately) for more info.

Re:300mm?

[Anonymous Coward]

> Or 300mm^2? I'm guessing the latter, which equates to about 1.73cmx1.73cm

Nope. that is 300mm in diameter. And that's the diameter of the wafer, not the chip.

300mm

[Anonymous Coward]

This is what 300 mm refers to.

Re:300 mm? I hope that's wrong.

[dido]

I misread the article, it would seem. I thought they were referring to feature size when it read 'the 300 mm fab will begin operation...' Sorry about that.

Re:300mm?

[mattjb0010]

I was thinking the wafer itself was larger, but 300mm is what Intel are using (or were using when that article was written).

The Foundation of the Saxony Valley

[xcomm]

Thank you AMD for laying the foundation of the Saxony (Silicon) Valley together with Infineon. Thank you for recognizing the talent, education, pracmatism and working power of the patient and friendly Saxony people. Your payback is visible as you are now nearly your break even. Thank for enjoying our great land and cultural as well as industrial heritage.

May also come the great R&D Transmeta, Big Blue, Samsung and Motorola here. You will get our working power and you will fall love too.

Re:The Foundation of the Saxony Valley

[matticus]

As someone who has recently assimilated himself into the Dresden culture by moving here, I have to agree. I love Saxony, and Dresden is one of the jewels of the world.

Re:The Foundation of the Saxony Valley

[benzapp]

This one factory is going to play just a small part in the recovery from 50 years of soviet occupation and the razing of the city by allied bombing.

Dresden is still a mere shadow of what it once was.

I wish the people of Saxony well with the hard work ahead of them.

Re:The Foundation of the Saxony Valley

[Hoser McMoose]

Good thing too since the entire valley flooded not too long ago!

Re:The Foundation of the Saxony Valley

[gmhowell]

Dresden is one of the jewels of the world

Well, take a lump of coal, subject it to enough heat and pressure, and you get a diamond. You should be thanking the Americans profusely.

(Horrible joke. Why did I have to read Slaughterhouse Five to find out just how obscene what we did to that city was? Seems like the folks in Dresden could have a serious discussion with those of Hiroshima and Nagasaki regarding who got the worst of WWII)

Re:The Foundation of the Saxony Valley

[nelsonal]

Tokyo and Nanking were also brutilized during WWII as well. Tokyo was firebombed with the intent of overwhelming the fire departments fighting all the wood and paper buildings. Just to keep it balanced the Japanese killed between 200,000 and 300,000 Chinese in Nanking near the begining of the war.

Re:The Foundation of the Saxony Valley

[gmhowell]

I'm the first to bring up Nanking (particularly when told how insensitive it is for the US to honor the crew of the Enola Gay), but I think what happened in Dresden (and Hiroshima and Nagasaki) is a bit different. Hell, it's even different in Hiroshima, Nagasaki, and Tokyo, owing to the fact that the stuff would light off and burn for much less time than the buildings in Dresden.

Anyway, yeah, don't want to forget Nanking.

Re:Just so people know ...

[MasterDirk]

Do you people actually know this, or is it just kibbitzing about a possible typo? I thought mm seemed large for chips, as I was under the impression nm was a more relevant unit. I may just be mistaken, yet again :)

Building size...

[MosesJones]

So we've all agreed that 300mm isn't the chip size, so it must be the size of the fab itself. Yes AMD are building the worlds smallest fab, and employing little goblins to make, by hand, the worlds smallest chips.

Next generation AMD processors will require no power supply as they will have tiny elves inside on treadmills which act both as the power and the clock cycle. Elves have been tested at IBMs labs up to 4.2THz which appears to be their physical limit for peak speed. The advantage of using Elves is that they can intelligently act as a variable clock speed, slowing down when nothing is happening, potentially to a stop, then giving a quick blast of 4.2THz power when required. It should be noted that each chip will have accomodation for around 100 elves, but only 50 will be delivered with the chip, the remaining 50 berths are for expansion and also for much more effective chip to chip communication in "ElfBUS"(tm) multi-processor systems.

Yes Elves are the future alright, and goblins are taking our jobs.

Or am I reading a bit too much into this ?

Re:Building size...

[FireFury03]

300mm is the diameter of the silicon wafer.
A large number of dies are constructed on the circular wafer, tested and the wafer is then cut up with a diamond saw. At this point the dies that failed the tests are binned (there certainly used to be a very high failure rate - not sure how high it is these days though).
AMD makes their CPUs as "flip-chips" these days, thich means that the die is bonded directly onto a PCB, instead of embedding it in ceramic or plastic.

People know.

[Hittite Creosote]

Do you people actually know this, or is it just kibbitzing about a possible typo?

Yes, they do. Or have googled it. For example, this article in the Industrial Physicist mentions 300mm wafer sizes in the sixth paragraph.

I was under the impression nm was a more relevant unit.

Relevant for the features within devices, not the wafers the devices are fabricated on. Many, many devices are made on a single wafer.

Re:300 mm? I hope that's wrong.

[WebMasterP]

He speaks the truth. Think of a wafer as a slice out of a log of beef (except the beef is silicon).

I assume this means we'll get smaller chips from AMD now...

Re:Just so people know ...

[berkut1337]

It's the diameter of the wafer, at 30 cm, it looks about right.

Re:Just so people know ...

[imsabbel]

its off by an factor of around 100.
with you forumula, the silicon of an opteron would cost 2000$ alone....

Re:Finally - a computer chip with visible features

[bobdotorg]

So the entire chip must be about 10 square meters? ... and running hot at 10Hz?

... and illegal to operate in California, 'cos it draws too much current!

Dude - don't diss the Pentium V.

Re:Just so people know ...

[TheOnlyCoolTim]

Alright - I know my Athlon is bigger than a millimeter squared, but I paid way less than $10,000 for it...

Tim

Re:Just so people know ...

[digitalunity]

Please, for god sakes, do not talk out of your ass. How does this shit get modded up? Just baffles the mind.

I work in the industry. Your numbers and your assumptions are way off. Fist off, a 200mm polished non-epitaxial silicon wafer is going to cost about $60-75 depending on the specific processing you want done to it. A 300mm wafer isn't a whole lot more. A big piece of the cost of a wafer is the processing, labor, and subsidization of investment capital for the huge plants required to manufacture wafers. The silicon itself isn't terribly expensive. 300mm wafers do not cost an arm and a leg. They are the most cost effective way to produce chips right now and 300mm is the market standard. AMD has been using 200mm wafers in the past but with the larger die size of their newer chips, 200mm is biting into their profits. The problem is silicon wafers are round and CPU dies are square. All of the silicon around the edges is wasted where a whole core won't fit. 300mm makes this wasted silicon a much smaller percentage of the total wafer's surface area.

Wrong assumption two: 300mm will be here for a while. There are still a lot of companies using 200mm wafers. I know this because I personally make 200mm wafers, and market forecasts has us producing a shit load of 8" wafers for THE NEXT TWO YEARS. 300mm wafer demand is growing, and will continue to grow for quite some time as companies make the transition. I would expect 300mm to be standard until at least '07 or '08. I heard someone talking a while ago about 350mm wafers, but I have a strong suspicion this person was, like you, also talking out their ass. To my knowledge, 300mm is the largest wafer being produced now or in the near future.

But the real question....

[Hexydes]

When will Cyrix be coming out with their next big thing?!

Re:But the real question....

[Hoser McMoose]

Considering Cyrix no longer exists, probably never.

Re:300 mm? I hope that's wrong.

[WalksOnDirt]

300 mm is 11.8 inches, so backwards Americans usually refer to them as 12 inch wafers. But not forwards Americans, of course.

Re:Just so people know ...

[popeyethesailor]

Why are learned people incredibly prone to verbally abuse somebody who is wrong?

Re:Just so people know ...

[sharkey]

Please, for god sakes, do not talk out of your ass. How does this shit get modded up?

Welcome to Slashdot. It's always nice to see new faces.

Re:Just so people know ...

[digitalunity]

I don't like verbally abusing anyone. I'm actually a super nice guy. It is just frustrating to see one person make a few assumptions, a few other people make the assumption he is right and mod him up, and then the rest of the people thusly assume that this highly rated post is accurate. It's the snowball effect of assumptions. I have no issue with people being wrong about anything. I just don't think people should speak as though they know something when they don't.

Then again, this is /.
Everyone's the expert.

What feature size?

[p3d0]

I couldn't find any reference anywhere to the feature size to be used on these 300mm wafers. Is it 90nm?

Re:What feature size?

[glassesmonkey]

The feature size is independant of the fab tooling. AMD will implement the smallest affordable feature size at the time the fab comes on line and most likely will be running two feature sizes. Depending on who is making their chips in two years (probably IBM) they will most likely use the same masks and try to get matching silicon up to production levels.

The fab is mostly just the facility (shake proof bldg, class 1, 10, 100, etc, wafer handling). What goes into the fab is the latest equiptment. That equipment will support multiple generations of lithography.

Re:What feature size?

[Hoser McMoose]

65nm initially, with plans to switch to the next generation (45nm?) a few years down the road.

Re:What feature size?

[Dastardly]

The feature size is independant of the fab tooling. AMD will implement the smallest affordable feature size at the time the fab comes on line and most likely will be running two feature sizes. Depending on who is making their chips in two years (probably IBM) they will most likely use the same masks and try to get matching silicon up to production levels.

This does not sem to be entirely true. Because there appears to be a point at which it is less costly to just build a new plant, than to refit an old plant with new equipment for smaller feature sizes. I am more familiar with back end equipment than front-end equipment, so I can't say for sure.
But, I would think the switch from Austin to Dresden where the wafer sizes are identical, but it does not appear that they have put 130nm equipment in Austin, is an example of this choice.

Re:What feature size?

[p3d0]

Got a reference?

Re:What feature size?

[Hoser McMoose]

Meant to provide one with my original post, but got distracted. Here it's mentioned in a CNet story about it. I think AMD actually stated the feature size would be 65nm in their conference call, though I didn't listen in.

Shouldn't really be a big surprise though, 2003 was when everyone was supposed to be switching to 90nm production (though it looks like it's actually going to happen mid to late 2004), and normally new feature sizes come out about every two to three years. So, with the new fab opening up in 2006, a 65nm feature size is right on track with expectations. Actually I think Intel is hoping to get to 65nm before the end of 2005, but then again, they also wanted to start shipping chips with 90nm features months ago.

Re:Just so people know ...

[Flaming Death]

Thanks for the accurate info. Its a pity Slashdot has degenerated so much. I used to really enjoy the industry info responses by people like yourself. Now its like trying to filter spam out of my mailbox..

I wish that Slashdot had a 13 year old kid filter... would make it alot less like fatbabies.com...

Re:Just so people know ...

[spectral]

why don't they make square wafers then? I'm seriously not being a troll, I assume that the wafer isn't being rolled around or spun, at least after being manufactured.. so why not use a square wafer, or recycle the silicon that's wasted?

I'll admit I know nothing of the production of silicon wafers, but it doesn't seem like it'd be harder to get square ones than round ones, or that it'd be impossible to make round ones, chop off the edges to make it square, ship that to AMD, and then melt the edges down in to another wafer.

And hence, I ask.

Re:Just so people know ...

[popeyethesailor]

Your point is well-taken. I have seen a number of +5 info. posts which are out-right wrong. Moderators do take action when pointed out.

I'd just like to think unless the OP was deliberately trying to mis-inform, there is no reason for abuse.
Thanks for responding:)

Re:Just so people know ...

[Short Circuit]

As another poster pointed out, the move to 300mm wafers is intended to reduce wasted space.

AMD's die size has gotten large lately, so they need to increase the overall die size so that more dies will fit as opposed to areas where a die wouldn't fit. You can't run on half a CPU.

Re:300 mm? I hope that's wrong.

[Bushcat]

On the other hand, it would herald a welcome return to traditional QA techniques. Wheeltappers from the rail industry could be retrained to walk around chips, tapping suspect components with their little hammers.

There's more to it than that (IIC)

[Ethelred Unraed]

Check out IIC's website. The IIC is a quasi-governmental group attracting businesses to eastern Germany, and they also had a hand in all of this.

This is great news for eastern Germany, in particular the Dresden area, which has really been on hard times since reunification. Hopefully this will also help fight the nascent neo-Nazism that was budding in Saxony for a while...that seems to have quieted down in recent years.

Cheers,

Ethelred

Re:Just so people know ...

[MrPink2U]

Due to the nature in which silicon crystals are grown, they will always come out round. A seed of perfectly aligned silicon is dipped into a crucible of molten silicon. Both are counter-rotated and the seed is slowly pulled from the melt, thus producing a round crystal.

I have a feeling that making a square wafer out of a round crystal is possible, but it probably isn't cost effective compared to the current ID saw method.

I worked in the grower industry for a few years but not in a fab. Someone closer to the industry please correct me if I am wrong, but doesn't the waster silicon go back to the supplier for re-purification?

Re:Just so people know ...

[Sique]

Wafers are produced as slices from long, round silicium blocks. They have this form because that's the way they come out of the zone heating ovens used to purify the silicium and melt out all the atoms you don't want on the chip. Non silicium atoms cause defects in the crystalline structure because of different size and number of electrons in the shell.

That's one of the issues with 300mm wafers (where 300mm refers to about one foot in diameter): The silicium blocks are wider in diameter, thus needing larger zone heating ovens, thus more energy evenly distributed over the whole zone. If the energy isn't exact the right one in the melting zone, the wrong atoms get out.

The purified silicium blocks then get sawed into thin slices. Larger slices are more prone to breaking during the handling, another issue to overcome while going to larger wafer sizes. They have larger surfaces which could be scratched or damaged.

On the other hand: Once you have the whole process running, you get more chips from the same area ;)

Re:Finally - a computer chip with visible features

[Garak]

Wouldn't it draw less current? Less IR^2 losses and greater heat dissipation(no need for a heat sink, the full chip is one), the speed would be limited more by the time it takes for the signal to travel from one end to the other?

Re:But WHY?

[stevesliva]

.why a company named American Micro Devices is locating their plant in Germany and not here in America?

That's Advanced Micro Devices. I could harp on how idiotic that makes the parent post look, but continuing on...

RTFA-- Germany offered them $1.5 billion in incentives, and they're building right next to an existing fab with experienced employees. $1.5 billion would be tough for a US state to put together.

Oh, come on.

[raygundan]

Nobody opposes freedom for Iraqis. Some people, however, think that maybe we should have kept on with the diplomacy a bit longer before sending in troops. Who is right? I don't know. At the very least, it's fairly expensive international-relations suicide to go invading stuff on your own without the help of the UN. But suggesting that because they do not advocate war, Germany opposes freedom for Iraqis is ridiculous.

Wafers round, chips square.... Why?

[Dawn Keyhotie]

I can see why they want bigger wafers, so that proportionately you get less wastage from the non-square edge areas left after inscribing all those square processors. But for the love of God, why are the processors square? They should be hexagonal. This would increase yields by filling in the edges of wafers much more efficiently than rectangular designs.

Let's see, the current Opterons are 193mm2 using 130nm process, as you can see here, so AMD is getting at most 148 dies from one wafer.

If we assume a regular hexagon of 193mm2, using the formulae for regular hexagons found here (Google to the rescue, Insta-Math!) each hexagonal die would be 14.93mm wide and 8.619mm to a side. That'll give you 13.39 dies across and 11.6 dies verically on a single wafer. SO, ok, all you Slashdot-lurking mathemeticians, how many hexagons of the given size can be completely inscribed by both 200mm and 300mm diameter circles?

And, as an additional exercise, what are the maximun number of hexagonal dies for 200mm and 300mm wafers when circuit dimensions are halved, i.e. 65nm process as planned for Fab-36?

Cheers!

Re:Wafers round, chips square.... Why?

[sirsex]

They use a very fine diamond tipped band-saw to cut the wafer up to seperate all the dies (dice?). This only works with rectangular patterns. A hexagonal pattern would require a laser cutter or some such, and that kind of heat would likely damage the lattice. And thoughput would be a low.

Re:Wafers round, chips square.... Why?

[Dawn Keyhotie]

Thanks, sirsex! That makes sense, if a laser or abrasive jet couldn't be used to cut the wafers into dies. That still leaves triangles as an option, but I guess that would just be too different. After all, everyone likes right angles, they're just so natural and easy to work with.

Cheers!

Re:Just so people know ...

[Betelgeuse on Ice]

Actually, 200mm is still the industry standard. After the intial build-out of 300mm fabs by the three I's (Intel, IBM, Infineon) its taking MUCH longer for other companies to financially justify building 300mm fabs.

In most cases, 200mm fabs cannot be retrofitted to 300mm, so entire new facilities must be built. Also, 300mm equipment is almost universally more expensive that its 300mm counterpart. (Our equipment runs anywhere from double to triple the cost.) A few 300mm fab runs about 4 billion US to build. In order to recieve the benefit of the larger wafers, chip makers have to run high volume chips with well defined processes. Few companies are as good as Intel or IBM at manufacturing chips. If you process is screwed, scrapping a 6 or 8 inch wafer costs you less lost (potential) revenue than a 12" wafer. I have heard of companies looking into running multiple devices of different sizes on wafers (alt-kerf) to better utilize 12" wafers, but I've never actually seen it work.

One interesting side-note about 300mm, it requires a huge amount of fab automation, due partially to the fact that a full cassette of 300mm wafers (25, generally) is too heavy for a fab worker to safely life unassisted. You can imagine the fallout of someone dropping a full cassette containing 25000 Pentium 4's. IBM's plant in New York state is one of the most advanced implementations of this. Quite amazing to see.

A question for the industry people:

[El Camino SS]

This is about the whole keep it in the States vs. Maylasia thing.

I know nothing about this industry so I'll just let someone take this:

Is there really that much of a economy of scale for moving a chip plant to Maylasia? I see that industry of creating and manufacturing processors and other computer whatnots to be EXTREMELY quality oriented work where the clock, quality, and price are an paramount. It looks like a tightrope act.

So my question becomes this: with all of that quality and issue at stake, does moving to Maylasia really make that much difference? It would seem to me that the capital investment is maximum and labor is not the largest factor (once again, I don't know). If you are looking for educated workers, I wouldn't think that Maylasia would be as good as America... but I may be wrong.

I just don't get all of this corporate moving. If you make shower flip-flops, I could see it. But if you make anything with a real standard that can't be found in a Wal-Mart, I just don't get it.

Either way, someone please explain.

Thank you AMD. By the way, I will continue buying AMD because I like it.

Re:A question for the industry people:

[Dastardly]

See my earlier post. US companies are not for the most part moving FABs to Asia. Mostly packaging and final test. I go through the reasons for it, but it basically boils down to packaged processors are about 10x as bulky as the die on a wafer. Therefore, it takes about 10x the labor just to move them to and from the package line or test equipment.

Re:But WHY?

[Dun Malg]

why a company named American Micro Devices is locating their plant in Germany and not here in America?

Fool. AMD is "Advanced Micro Devices". You're probably confusing them with the BIOS maker AMI, which is "American Megatrends Inc" (stupidest name I've ever seen).

The German guy

[Hoser McMoose]

Take a look at the photo in the top-left corner of the linked press release from AMD.

Could that guy possibly be any more German?! :>

Re:Just so people know ...

[wramsdel]

Damn, and just when I thought Gallium Arsenide had a cost advantage over silicon! Thanks for raining on my parade, DU...now I'll have to go back to work.

Re:Just so people know ...

[FatherOfONe]

Like the other poster I know very little about chip production, but you didn't seem to answer his question.

It appears that these ovens could produce square chips. That would appear to save some money and then it appears that you wouldn't need to keep making bigger and bigger wafers.

Again I know verry little about chip design, so I am not trying to be a troll or start any wars here.

German Subsidy

[jmichaelg]

Why they didn't go the Malaysia or Philippines as you mentioned I don't know. Perhaps the current unrest in the world?

Because the Germans guaranteed the necessary loans. AMD's balance sheet has been so bloody of years past, that there wasn't anyway that AMD would have been able to swing the loans without a consigner. Aren't a lot of parents out there who can co-sign a $2.4 Billion note. From the article...

Press reports quoted regional government sources in the state of Saxony as anticipating that AMD will receive a debt guarantee from the German governement by next week. Other sources report that the Saxonian government has already provided guarantees for 80 percent of the required loans.

If I was a German taxpayer, I'd be wondering how AMD can possibly pay off the note.

Bombing of Dresden

[kneel]

The 1,000 jobs created by this plant are a far cry from the 100,000 that were killed on Feb 13-15, 1945. The city of dresden was basically annihilated by American and British allied forces in 3 days by 1250 sorties. They used a firebomb technique to make sure that they could kill as many people as they could. These were refugees, not Nazi soldiers. Bombing of Dresden

Re:Bombing of Dresden

[Tailhook]

These were refugees

Indeed, they were refugees. After Germany invaded the Soviet Union and got pushed back by the Red Army, for some reason civilians felt the need to flee. Thus, they became "refugees." I'll bet the Russians had another name for them...

The 1,000 jobs created by this plant are a far cry from the 100,000 that were killed on Feb 13-15, 1945.

100,000 is on the high side and likely to be exaggeration. 1000 is on the low side. That's just the number of people directly employed by AMD.

The city of Dresden was basically annihilated by American and British allied forces in 3 days by 1250 sorties.

A few things could have been done to prevent the annihilation of Dresden (Hamburg suffered a similar fate.) Below I offer a few guidelines you may use:

1.) Don't invade France and other European countries.
2.) Don't invade the Soviet Union.
3.) Don't sink American supply ships.
4.) Don't bomb London.

If you follow my guidelines I can assure you that Dresden will go un-firebombed for the foreseeable future.

Thanks!

can't be recycled

[rebelcool]

The unused silicon chips left over aren't useful - they're not square shaped (the edges are round), of odd and varying sizes, and they often already printed on with partially etched chips, depending on how the equipment works.

I believe - and I could be wrong here, please someone correct me - that the silicon ingots have to be grown in a fairly complex fashion involving centrifuges to spin impurities out and the like resulting in a cylinder shaped ingot, from which they cut the wafers from.

Re:Just so people know ...

[imsabbel]

they cant produce square wafers.
This kind of cleaning works that way:
you have a silicion rod and move is SLOWLY through a heating device that heats it up enough to let non-silicon atoms migrate.
The end that leaves the heated area slowly cools, and (if all goes right) silicon atoms create a monocrystal. The wrong atoms stay in the heated area and wander to the end which is cut off.
in reallity, you often need many passes...
now, if you would use a square rod, the corners are HIGHLY sensitive and very likely to cause defects in the crystal...

Re:Just so people know ...

[imsabbel]

you can forget GaAs as silicon replacement because the low hole mobility prohibits fast CMOS logic.
And ECL would really define "a new kind of hot"

Re:Just so people know ...

[p3d0]

Good God. Did you read the parent of my post? In fact, did you even read my post? I think I made it clear I know the difference between wafer size and feature size.

If this isn't the most frustrating thread I have ever been involved with, I don't know what it.

Re:Just so people know ...

[wramsdel]

I was mostly being facetious. GaAs hasn't been used seriously in digital for awhile now. The real battleground is in analog, specifically RF. RFCMOS is constantly threatening to knock GaAs out of the market, primarily because of downward cost pressures. There's also the attractiveness of integrating things like the power amplifier onto the radio chip (which is already RFCMOS.)

Re:MOD PARENT DOWN

[Aaron England]

The above is off-fcuking-topic!

Re:Just so people know ...

[dustinmarc]

Besides the fact that it is easier to make the wafers circular, I thought that I would let you know that the wafers are indeed spun. Usually this is to dry them off after they've been dipped in a chemical or water.

The 300mm is a real advantage for any company that has a Fab which supports it. You can produce more than twice as many chips on a 300mm wafer then on a 200mm wafer.

It's not really a breakthrough on AMD's part at all. Intel already has 300mm fabs up and running.

Re:Just so people know ...

[digitalunity]

Most of the machines we have here at our plant are capable of processing 6" and 8" wafers. Most of our cleanlines could accomodate 12" but that's about it.

You think dropping a cassette of 12" wafers is bad? We had a guy who pulled too much on an ingot and cracked the furnace, then dropped an ingot. That amounted to about $850K in damage, lost time, and lost revenue. That guy got fired faster than you can imagine.

Re:300mm fab?

[HiggsBison]

You can't get much work done in such a tiny building...

Well, duh! That's why they have 36 of them. Besides, it's so much better than those old 100mm and 200mm buildings.

Re:300mm?

[nelsonal]

300mm (12") wafers are by and large cutting edge technology. Most fabs still use 200mm (8") stuff, only the most expensive chips, PLD CPU some cell and memory applications justify the sort of investment that new tools require. There is still some 6" stuff running, and I think a few analog companies still use 4" wafers. The only 300 mm fabs I know of are owned by TI, Intel, AMD, IBM, Taiwan Semi, and Samsung. I'd assume that Micron, Infeneon, United Micro, and ST Micro have one or plans for one soon. AFAIK all the 300mm stuff runs a 130 nm process and Intel and IBM are already trying to get 90 nm going. Everyone kinda slowed their 300mm migratin when unit demad fell in 2002. The cost advantage comes along with a huge increaes in chips produced, and there were some problems getting yields up to a decent level.

Re:But WHY?

[nelsonal]

IBM got a pretty sweet incentive package from NY to build a simialr fab there. Not quite as nice, but this deal wouldn't have happened without tremendous funding from almost any potential location. It's fairly common for at a bare minimum property taxes to be rebated if you bring the hope of high paying jobs.

Re:Just so people know ...

[p3d0]

I don't understand the connection with my post.

Where did the parent go?

[raygundan]

Holy reparenting, batman! When I initially posted this, it made sense. But that parent post is gone. It's not the first time I've noticed this on /., but it's the first time it's happened to me. The parent post is gone, and my post has been "reparented" up to connect with the parent's parent. Anybody know what gives?

Found it in "preferences"

[raygundan]

There is a checkbox labelled "reparenting" which is enabled by default. When checked, this causes posts that are children of posts below your threshold to be reparented up and still displayed even though their parent post is no longer visible.

That explains a whole lot of the top-level "reply" posts you see on stories, as well as the complete lack of sense my post makes with its parent gone.

There ought to be some sort of "Parent Missing" indicator, and maybe a link to the low-modded parent, just for the sake of making sense of "adopted" posts like this one.

Re:Just so people know ...

[Short Circuit]

First, you were getting nanometers (nm) and millimeters (mm) mixed up. nm refers to the minimum feature size. (Or your resolution, if you will.) the 300mm slabs refer to the size of the entire wafer, on many dies are fit.

Have you ever seen a picture of a circular piece of silicon? That's a wafer. All of the rectangles you see on the wafer are the individual dies, which are placed in a plastic(If you're really cheap, and you're not worried about static in handling) or ceramic(what's normally used) casing. Once you place the die in the casing, and perform a little bit of wiring inside, you have your chip.

Re:Just so people know ...

[p3d0]

First, you were getting nanometers (nm) and millimeters (mm) mixed up.

No, I was not. I was responding to someone else who made that error. Check again.

Re:Just so people know ...

[Short Circuit]

My apologies. I was replying, long after I had last posted in the thread. I should have paid more attention the the context than just reading your post.

Mixing up nm and mm was rampant in this story. I was beginning to think nobody had a frigging clue what they were talking about. If a guy just reads the background articles on Ars Technica, they're a veritable authority compared to the average discussion participant.

Re:Just so people know ...

[p3d0]

Well, of all the people that accused me of not understanding the issues, you're the only one that appologized, so thank you. :-)