AMD to Build 3rd FAB in Dresden

It's in german, but I'll sum it up for all of you:

I do hope that it would be a 450mm fab.

http://www.heise.de/newsticker/meldung/73529

Mabey it will bring down AMD prices so there will be even MORE competition **prays.**

I wouldn't count on significantly lower prices from it. After all, one of these fabs has a price tag equal to about 2X AMD total revenue for last year.

Good point.


Cheers!

Just goes to show that you should check the numbers before you just go and pull something out of your arse like I just did a couple of posts above. Way to catch my goof Jack.

I actually had in mind a number of about $3 billion for a fab these days. Just didn't bother to check the earnings data. Must have been thinking quarterly instead of yearly

Its a joke from my intel buys amd thread. Lighten up.

Good point.


Cheers!

Just to be clear, that should be net profit, and last year they made a net profit of 18 million dollars in 2005. The original investment commitment is 2.5 billion, I believe the price now is 1.9 Billion roughly, of which 1.3 Billion came from government loans and private contributions. So let's take 1.9 billion. That is 1900 million or more like 100X

If you want to take a look at AMD's last 11 year track record for what they have returned to their investors:

Year....Net Sales....Operating Income (Loss) in Millions....Net Income (Loss)
1994....2155.........469................................271
1995....2468.........222................................216
1996....1953.........(253).............................(69)
1997....2356.........(91)..............................(21)
1998....2542.........(163).............................(104)
1999....2857.........(321).............................(89)
2000....4644.........888...............................983
2001....3891.........(58 )..............................(60)
2002....2697.........(1225)............................(1303)
2003....3519.........(233).............................(110)
2004....5000.........222................................91

2005
2005..q1..1272.......(45)..............................(17)
2005..q2..1260.......(7)...............................(30)
2005..q3..1522........79...............................(31)
2005..q4..1838........205...............................96

Cumulative 11 year net income (-178 million) through Q4 2005.

Now that Q1 was roughly 200 million, they are finally in the black by 20-30 million after 11 years and 1 quarter. (Source: AMD's web site, Annual revenue reports back to 1994 -- they have not published any farther back in time so this is an 11 year analysis).

There you have it. Just believed gr8Mickey's numbers without checking.
I guess he already said it all, so...


Cheers!

More info coming from the Inq:
http://www.theinquirer.net/?article=31974

No, 450mm wafers would not make 8-core chips a possibility. The limiting factor of creating large dies is that, as the area of the die increases, the chance of a defect increases. If a die is twice as large as before, then it will have twice the chance of defects. Many times, all it takes is a single defect on the die area to make that die useless.

Having larger wafers just means that you waste less space on border losses (since rectangles don't really fit all that well in circular waffers) and it also tends to be cheaper to do things in larger batches.

In other words, 450mm waffers would just make it a bit cheaper to produce.. however, AMD doesn't have the money to go on developing a completely new waffer size. I don't think it's used anywhere else, and you would have to completely retool your fabs to deal with waffers that large. It's a waste of money.

Still.....Going from 1 operating 200mm fab to having 3 high-capacity 300mm fabs would be quite the improvement for AMD.

Agree.

Just playing with numbers... supposing that the die area scales linearly with node transition (which it doesn't), and taking a dual-core 200mm^2 chip @ 90nm as the basic reference (a square, for simplicity sake), one might get an idea of the die yield-per-node-per-wafer size, given the # of cores per die.

- Area of a 200mm wafer (A= Pi*r^2): 31 415,90mm^2 (just for reference);
a. Area of a 300mm wafer: 70 685,80mm^2;
b. Area of a 450mm wafer: 159 043,13mm^2;
The area scaling constant is the same, going from 200 to 300 to 450mm: 2,25.

Now, if one dual-core die is 200mm^2 @ 90nm, the next node's die area is given by the Reference Area (200mm^2) * x node / 90nm * # dual-cores:

-> @ 65nm: 200 * 65 / 90 * 1= 144.40mm^2 (dual-core)

-> @ 65nm: 200 * 65 / 90 * 2= 288.80mm^2 (quad-core)

-> @ 45nm: 200 * 45 / 90 * 1= 100,00mm^2 (dual-core)

-> @ 45nm: 200 * 45 / 90 * 2= 200.00mm^2 (quad-core)

-> @ 45nm: 200 * 45 / 90 * 4= 400.00mm^2 (octo-core)

_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _


Given the basic reference, one would achieve the following yield-per-wafer, @ 90nm: 200 * 90 / 90 (reference node) * 1 / wafer Area (wA, but referring to the diameter, in mm, for simplicity; 300mm)= 353 dies-per-wafer;

-> @ 65nm: wA (300mm) / 144.4mm^2= 410 (dual-core)

-> @ 65nm: wA (450mm) / 288.8mm^2= 550 (quad-core)

-> @ 45nm: wA (450mm) / 200,0mm^2= 795 (quad-core)

-> @ 45nm: wA (450mm) / 400,0mm^2= 397 (octo-core)

Just playing around, like I said... but, all things being equall (which aren't), this simple arithmetics gives some perspective on node scaling/wafer diametre yields, given the number of cores-per-die.

If the accountability is correct , yields would reach a peak @ 45nm node/450mm wafers, for quad-cores; assuming some truth in this speculation, octo-cores should downscale to the 32nm node, in order to be profitable using the same 450mm wafers.
Of course, this would obviously reflect profitability...

As a sort of disclaimer, many other factors influence, decisively, all the parametres I referred; hence, this is a biased speculation.


Cheers!

If you go 450mm, then you need new FOUPs, new OHV's, new loadports, handling robots, process chamber fittings, Quartz tanks tubes boats washers and on and on and on and on.

That stuff is NOT cheap. . . and the fact that you have to talk AMAT into making prototypical hardware, qualing it, wait over a year to find out if they have a ROI or not. . . then if youre AMD you have to convince AMAT that you're going to buy enough of the POS6400 that it warrants the training of FSE's and a new plant for themselves dedicated to manufacturing this one tool - for you - because you THINK it will work. Then do that with 15 other vendors at the same time.

It takes alot of cash to say the least.

Hehe!

Yeah, I had to google AMAT & FOUP (the non-acronyms, are somewhat easier to grasp), although I was already familiarized with Applied Materials.
As for POS6400... if you say so!

Thanks for the help.


Cheers!

Well, I'm 100% positive I didn't understand most of the acronyms you referred to; I'll search 'em, after.
Anyway, I get your point and, aside the - apparently huge - investment, ROI would be a near term big issue for such a "small" company. As for AMAT, your two last paragraphs seem pretty enlightening.
Intel will certainly have an order of magnitude advantage over AMD/ (IBM?) but, the 300mm process has plenty of room for the next node, yet.
What I fail to grasp is this condition: «3-4 billion dollars for a new 300mm fab. if they go 450mm before intel does, they will be paying the R&D costs of it out to the tool manufacturers. and that's something they can't afford to do, private investors or not.»; If AMD goes into 450mm wafers AFTER Intel... means that, Intel would assume all the R&D & tooling costs (because they can?), allowing AMAT to warrant its investment & making it easier for AMD/(IBM?) to purchase the necessary tooling & training assets at lower costs. Is this correct?

Edited: «before» -> «AFTER».


Cheers!

exactly so. when intel moves, the industry moves with it.

D1C was a huge leap into the unknown. . . but everyone was willing to jump in with Intel because they knew we were going to pull it off. Now any fabrication company can afford 300mm.

I know Intel's powerful: It wouldn't make sense for AMD to take the 450mm step and be the first at it, claiming that achievement & go bankrupt the next day; AMD hasn't even got the "full steam ahead" on 300mm tooling, manufacturing & all; hence, that's an obvious "not-yet" by Intel, clamping AMD's (& others) eventual intentions from getting there first. But, hasn't IBM a [strong] word to say about it?


Cheers!

What's a WIPT, this is a new one....

Jack

Had to search that one too: "WIPT" -> "Working Integrated Process Team", where 'P' can also stand for "Product", "Project".
I wonder if it's correct, though. As far as I can infer, that'll be an "automated process" team(s)...

I wonder if Jlflipflop98 would be kind enough to post the most common acronyms, since he works THERE!


Cheers!

Once you go over your engineered fab threshold, you build inventory at your various constraints in the process line. If you're smart, you stop your wafer starts at the front until the line empties out a little.

If you're AMD, you just keep shoveling them in and hoping for the best.

No, below is what investors had as return:

March, 83: AMD Stock Price $9.00
May, 06 AMD Stock Price $31.63

A modest 250% return on investment.

(Source Yahoo, prices adjusted for dividents and splits)

In that same time period, the Stock Market itself (going by the S&P 500) provided a 1300% return on investment. So that is a huge opportunity loss. Don't try to be clever with facts, they bite back.

More info on AMD's upcoming Fab39:
TheStreet.com

No, I am not confusing anything. Below is what you said:

Investors' return is what they get from increased stock value and/ordividents which don't have much to do with present profitability of a company.

A company's stock value goes up without one single penny profit in the income statement if investors have faith in its future (like bio-tech or superconductor companies)

Sure.
May 06, $18.22, Jul-86 $0.36 (divident and split adjusted). Much higher than AMD. Just for the record, I never said AMD was more profitable than Intel. I own Intel shares by the way.

Jeez man. if I'd read this before, I'd have changed my tone in my previous message.

Now, listen up. I bought Intel shares when Dell double-crossed them and their stock was bashed. When Conroe comes up, Intel will recover and AMD will fall. I will sell my Intel shares and buy AMD. Intel, having the dumbest marketing department will screw up again, their stock value will go down once again and AMD shares will go up. I will sell AMD shares and buy Intel. Intel will eventually die but I will be grateful to them.

Get it?

Your Intel buddy was talking about being clever. I guess it starts with not having an Intel icon below your name.