Интересная информация из области полупроводников: Nikon vs. Canon

Нашел на ДПРевью. Из разряда Analyze this....

[quot]
michaeljberman написал:

The white paper was a good marketing paper from C (это он о документе про ФФ от Кэнон). I have been making computer chips for over 25 years. The price for a processed 8 inch CMOS wafer from fabs in Asia is less than $1,000 USD this is a fully completed wafer not the starting wafer talked about in the whit paper.

Steppers:

Yes, C makes steppers, but Nikon and ASML have >75% market share for the last 5 years. C was a big player in the market 10 years ago but fell way behind.Here is a story about this:

EE Times: Semi News
(05/27/2005 2:09 PM EDT)

Canon behind in lithography development, says exec

SAN JOSE, Calif. — For years, Japan's Canon Inc. has made various claims that it will one day become the world's largest lithography-tool vendor, surpassing ASML Holding NV and Nikon Corp.

But Canon (Tokyo) continues to lag behind its rivals and remain the third largest vendor in the arena amid lackluster results and product delays.

The company has experienced an assortment of delays in developing and shipping advanced lithography tools for semiconductor production, according to reports.

Canon is reportedly developing a dual-stage lithography tool, but the machine is not expected to hit the market until 2006. And it has not made any significant announcements about its immersion lithography tools, while rivals ASML and Nikon are aggressively pursuing the technology.

"To be honest, we haven't been able to keep up with the competition. There is a simple explanation — we were late to develop the most advanced machines," said Fujio Mitarai, president of Canon, in a report from Reuters on Thursday (May 26). "It will take one or two years for us to really regain our strength."

The comments by Mitarai come after Canon cut its 2005 target for shipments of steppers, from 143 units to 125, according to the report. Canon shipped 152 steppers last year.

Canon's share of the stepper market has hovered at about 20 percent in unit terms for the past three years, down from 35 percent in 2001, according to the report. ASML has 41 percent of the market, while Nikon held 37 percent, according to the report.

Mitarai said overall revenues in its lithography unit "were not falling," due in part to strong demand for LCD steppers, according to the report.

As for stepper field size ASML is at 26 mm x 33 mm this is 2 to 3 mm "short" but would work very well as it is. Look here for the ASML statement http://www.asml.com/asmldotcom/show.do?ctx=7077&rid=7095

Also yes stitching is done here is the ASML short paper on that subject: http://www.asml.com/asmldotcom/show.do?ctx=6874&rid=7241

I have worked in fabs that have used both Nikon and ASML steppers, they are both great. The cost increase in doing a stitch to get to 36 mm would be very small in the cost increase for the chip. With only 20 die on a 8 inch wafer that will be the main cost driver, that will mean 10 to 15 good units per wafer vs well over 150 on APS size so the die will cost about 10X as much. but that would be maybe $250 to $500 more for the D3XL, not a real big deal for the over all cost.

[/quot]

[quot]
michaeljberman написал:

>How much does the cost vary according to the number of layers or the feature size? (ignoring yield issues)

Feature size is a very big driver, I left wafer fab about 2 years ago, (I now am at a major University), 2 years ago .18 micron was the sweet spot in the off shore fabs, so in the last years there may have been a drop down to .13 micron, but that is a big jump due to copper interconnect (long story but major bucks to change a line to copper interconnect) If a 180 nm (1000 nm is one micron, most of the sizes below .13 micron are now being called by nm) wafer is $1,000 then a 130 nm wafer could be $2500 and the 90 nm would be $5000 each, that is also based on supply and demand but not a bad rule. Also keep in mind the design costs and tooling costs (the masks) also “about” double each time you drop from 180 to 130 to 90 so the other costs also go way up. The costs of more layers, going from 4 layer interconnect to 5 layer would only add about $100 per wafer so feature size is the key driver here, unless the added layer is a hard film like going to very low k.

On steppers, it is clear that Nikon steppers are for chip making in the wafer fab world (I do not know a lot about LCD’s but I know it is the same players for the most part) I do not know of Nikon making there own chips they may design them, but fab them they do not. Even if they did, it would not be the market driving there steppers. C on the other hand may very well use there steppers to help drive the cameras, it would make sense. (in th old days IBM used wafer fab to give them an edge on computers, the die they made costed a lot more than others and would not as a die make money, but it fit into a $50-100,000 computer and made it a much better deal)

C has been way behind both Nikon and ASML in fact a major stepper run off I know of in 1997 did not every look at C. but we were running C in one of the older fabs in Japan, the 3rd company was SVGL, it was in many ways the first out of the gate, but they are no more, ASML bought them up a few years go. C is not an advance player in steppers, they do very nice press releases, but I thank a lot of there sales are on older tools not state of the art (This is steppers) But they could be using there steppers to give them a heads up if FF was one of there early design goals for there steppers. The newest steppers are printing down to 65 nm in production (Intel, AMD in a few weeks, TSMC) and they are running 45 nm in R & D. This is way below want is needed for the cameras.


>Glad to read about this. Does the stitching process decrease yields, all else equal?

Stitching has been around for some time, here is the best I know of to date: http://dalsasemi.com/news/news.asp?itemID=252 This is a 100 mm by 100 mm 100 meg chip that is stitched, but it will be very costly do to single chip per wafer.

I have not worked with stitching, most chips in the last 20 years were about 10 mm by 10 mm. On a long lived die, it may start out at 14mm by 14 mm on say a 250 nm process, but will end up in the 130 nm process as a 7mm x 7mm the same wafer will give 4 times the die, this is the big cost driver for pushing to smaller sizes (if device speed is not a driver).

Stitching will have costs at are added, the tooling (the masks used) will need to be 2X (or more) per layer, this cost for tooling was in the 250 nm days maybe $500,000 for a device but has doubled on each smaller size, so the cost for the tooling maybe $2,000,000 vs $1,000,000 OK so if there are 10,000 units made the NER cost will be $100 more per die for 100,000 unit the extra cost per die will be onlly $10. The wafers per hour on the stepper would also drop as each wafer will now have 2x the number of masking layers. There may also be some reduction in yield due to Stitching, but I thank that it would be small.

>Here I am confused. Doing the math, the area of an 8" wafer equals 37 + FF sensor dies. Of course, a "partial" sensor is no good, so perhaps the true number is only 20 or slightly more.

OK, you are right, the numbers I used I miss read, I come up with total die of 24 FF and 54 APS here is the math: Start with a dia. of 200 mm, take 6 mm off for edge (3 mm on each side) we are now down to 194 mm, next for the partial die remove the sq. root of the die size (FF = 24 x36 or 864 total are with a sq. root of 29.4 giving you a 164.6 mm disk to cut the die out of giving you 24.6 or 24 die (LESS YEILD) for APS (24 x 18) you have a 173.2 mm disk with 54.5 die or 54 die (LESS YEILD)so with 5 killer defects, we end up with 19 FF and 49 APS on a wafer, due to the process not being a standard TSMC type process, if we use the $2500 per wafer number, we have each good FF at $131 vs APS at $51 this does not include packaging of the die or testing. Packaging should be in the $10 range (+-75%) testing should be less than package.

This is the cost of wafers, not the design or tooling costs and sometimes they like to make profit.

OK now lets see the die at a store down the street, add the tooling cost and design costs of say $2,000,000 (maybe this is the reason Sony likes to sell the same die to many different camers makers?) lets say we are going to sell 100,000 so the cost per die is NRE $200, the die cost is $130 or $50 testing and package is $20. costs is then FF $350 or APS $270, now lets talk mark up, retail should be about 4 X cost or FF is now $1400 and APS is now $1080. (take 20% off if you mail order out of New York)

I started using a new F2 in 1972 when did you start?

Mike
[/quot]


[quot]
michaeljberman написал:

I made a math error in the last big section of the last posting I said:


OK now lets see the die at a store down the street, add the tooling cost and design costs of say $2,000,000 (maybe this is the reason Sony likes to sell the same die to many different camers makers?) lets say we are going to sell 100,000 so the cost per die is NRE $200, the die cost is $130 or $50 testing and package is $20. costs is then FF $350 or APS $270, now lets talk mark up, retail should be about 4 X cost or FF is now $1400 and APS is now $1080. (take 20% off if you mail order out of New York)

But when you $2M with 100,000 units the NRE is $20 not $200, so that drops the prices to : FF $170 APS $90 then the times 4 numbers would be FF $680 APS $360.

It must be a bad math day for me, I need to get a better computer

Mike

-------------------------------------------------------

The math was based on retail cost if you buy it at the store downtown, the camera companys buy in very large amounts and will pay a lot less, If you take the retail price of all the parts in a computer, it will be in many cases a lot more that the price you can buy the computer for. May companys look to make the profit on the finished goods not each part, call Nikon and ask them for the cost of a replace chip and the prices they quote will be a lot closer to the ones I was using for retail.

I was once told that if you go into a car dealer and buy all the parts to make the car yourself, it will cost you 4 to 6 times more than retail on the car, and you will have the fun of putting it together yourself.

mike

[/quot]

Dejstvitel'no interesnaya. Spasibo!

Все таки с калькуляциями, что-то у автора не то.
Или Кэнон в своей white paper уж очень сильно приврал...