Register for free to join our community of investors and share your ideas. You will also get access to streaming quotes, interactive charts, trades, portfolio, live options flow and more tools.
Register for free to join our community of investors and share your ideas. You will also get access to streaming quotes, interactive charts, trades, portfolio, live options flow and more tools.
kpf -
This is a link to the company that markets a cost modeler for estimating product cost. They are the same company that gives away that yield modeler. This page is a sample of the $1000 product and it uses the original NorthWood 146mm2 die as an example. They are using the yield modeler and using .25d/cm2 as their estimation. I can't say if this DD number is correct but they seem to have enough confidence in both the DD number and the yield model to sell it as a commercial product. I think it shows the free yield model is pretty close to reality, assuming you have the correct DD number, whatever that is.
http://www.icknowledge.com/our_products/costmodelpages03.pdf
http://www.icknowledge.com/our_products/costmodelpages03.pdf
wbmw -
This link is to a sample page for their commercial cost modeler. This is a $1000 product. This page uses the original 146mm2 NorthWood die before the shrink and they are using .25d/cm2 as defect density. I can't say if this is correct or not but that's their estimate.
http://www.icknowledge.com/our_products/costmodelpages03.pdf
Reseller Mike -
The yield calculator does subtract the kerf from the Y dimension but I do not see a reference in the spreadsheet where it removes the X kerf. It also appears to be missing information on the flat section of the wafer. The model includes flat exclusions (subtractions from the available die) for 100mm, 125mm and 150mm wafers but uses 0 for 200mm and 300mm wafers. Maybe the model is a work in progress? By the way, the calculator also excludes 3 test sites that are equivalent to the die size in its model in the die count.
Damn! I'm going to demand my money back!
Reseller Mike -
Any comments?
Very good! I would point out a couple of mistakes though. You have the Madison6M and 9M as the same die size. Plus you are assuming Deerfield is a smaller die. The NorthWood die size should be 131mm2 and I don't believe Intel has announced Prescott's die size yet.
All should note that you are making an assumption about defect density that is not backed up by any data. The true number could be significantly different. Wafer cost may be different too.
Kudos to you for the work!
BTW - how about posting that on the AMD thread?
Thanks Duke!
Elpida ??
Anyone know exactly what's up with Elpida?
Here's something I found with a Google search. Anyone know any more?
Toshiba, Elpida to make fast chips
Correspondents in Tokyo
July 14, 2003
ELECTRONICS manufacturer Toshiba and chip maker Elpida Memory said they will make memory chips that are about eight times quicker than today's fastest chips for gadgets in future network-linked homes.
The two Japanese companies will separately make the new chip, called XDR DRAM, using technology by Rambus.
A sample of the chip will be produced next year and head into full-scale production in 2005, Toshiba spokesman Makoto Yasuda said.
The new chip, running at 3.2 gigahertz, will be able to handle large amounts of video data and may be used in home network servers, mobile systems and next-generation video game machines, including the successor to Sony PlayStation2.
"Keeping up with ever-increasing bandwidth requirements is a critical element of memory system design," said Hidemori Inakai, chief marketing officer for Elpida Memory, a joint venture between Japanese electronics giants Hitachi and NEC.
[edit:] Here's another link with more info:
http://techzone.pcvsconsole.com/news.php?tzd=1966
chipguy -
Where was this article published?
Sorry, BusinessWeek.
http://www.businessweek.com/technology/content/sep2003/tc2003092_6592_tc083.htm
Intel to Invest Extra $23 Mln in Elpida
Tuesday September 2, 2:31 AM EDT
TOKYO (Reuters) - Elpida Memory Inc said on Tuesday that Intel Corp had agreed to invest an extra $23 million in it, raising the semiconductor giant's total investment in Japan's last big maker of standard computer memory chips to $123 million. The latest investment is aimed at helping Elpida boost production of high-end DRAM (dynamic random-access memory) chips.
In June, Intel, the world's largest chip maker, agreed to invest $100 million in Elpida, a joint venture between Japanese electronics conglomerates NEC Corp and Hitachi Ltd.
Intel: Back in the Chips
Improved sales, new products, and new plants seem certain to see the outfit set the pace for an industry on the rebound
All through the economic downturn, Intel fared better than most rivals, persevering while its competitors struggled. However, the world's largest semiconductor manufacturer didn't endure the slowdown unscathed. When Intel's revenues slid 21%, from $33.7 billion in 2000 to $26.7 billion in 2002, its income fell from $10.5 billion to $3.1 billion and the company had to let go thousands of workers.
Advertisment
Now the tide appears to be turning. On Aug. 22, Intel (INTC ) served up a pleasant surprise: The chipmaker increased guidance for the current quarter, ending Sept. 27, to between $7.3 billion and $7.8 billion in sales, vs. the previous range of $6.9 billion to $7.5 billion, thanks to stronger than expected sales of its processors for desktops and notebooks. To put the improved numbers in context, listen to CIBC World Markets analyst James Jungjohann: "This is one of the strongest growth quarters for the company in the past decade."
Investors were pleased, pushing Intel shares up 6%, to $28, soon after the announcement. Other semiconductor stocks also moved higher, on the belief that Intel's news signals an industry recovery. While Intel CEO Craig Barrett urged caution, emphasizing that the market remains spotty and uncertain, industry observers see more reason to be upbeat about Intel's future.
STRONG MARGINS. Like other chipmakers, Intel is benefiting from recovering demand in Asia, where sales could soar 20% over the next couple of quarters as the SARS panic recedes, according to semiconductor consultancy IC Insights. Intel also is gaining from the gradual improvement of corporate spending on information-technology spending, which should become more pronounced by the first quarter of 2004, says analyst Greg Sheppard of chip consultancy iSuppli.
Company-specific factors are what set Intel's improved revenue-and-earnings outlook apart from the pack. Intel's margins are rocketing, for example, thanks to product introductions and technological investments it made over the past three years. In the recent preannouncement, guidance was for gross margins of about 56%, vs. the 54% company-specific factors previously company-specific factors, which Fulcrum Global Partners analyst Clark Fuhs believes should rise into the low 60s within the next year to 18 months. Given year-over-year sales growth of between 11% and 14%, Fuhs sees Intel's earnings per share rising by between 25% and 40%.
Such numbers would justify a higher stock price, adds Fuhs, who recently upped his 12-month target for Intel to $37 a share -- a 32% premium on the current $28.59, which is just off the 52-week high it hit Aug. 22. Others think that outlook is too rosy. For example, Michael Mahoney, a portfolio manager with EGM Capital in San Francisco, thinks the stock could move to the low 30s and recommends jumping in on any seasonal market dips. (Mahoney doesn't own the stock.)
The higher margins are partly due to greater economies of scale. Most of the chipmaker's costs are fixed, so as it ramps up production, margins should rise. Intel will benefit greatly from its new pair of 300-mm high-volume plants, which stamp out twice as many chips as the older, industry-standard 200-mm facilities, lowering per-chip costs. As Intel makes its chips ever tinier, unit costs should be down by 50%, as of 18 months to 24 months from now, according to Morgan Stanley.
WiFi's MARKET. Margins also stand to benefit from the growing demand for notebooks. The dominant supplier of laptop processors with 88.1% of the market, Intel has gained 1.5% share from main rival Advanced Micro Devices (AMD ) over the past year, according to tech consultancy IDC. The market is expanding at about twice the rate for desktop processors and offers higher margins, says Dean McCarron, founder of PC components consultancy Mercury Research in Cave Creek, Ariz. Better still, most of the growth is still ahead: Laptops accounted for 26.5% of total PC shipments in the second quarter -- a figure that will grow to 35% by 2007, estimates Alan Promisel, an analyst with IDC.
Intel is positioned to take advantage of that growth, thanks to its Centrino package of chips, which integrates the Wi-Fi wireless broadband for use in homes, airports, schools, and other public places. About 42% of notebooks shipped in 2003 will be Wi-Fi enabled, estimates Promisel -- up from 20% last year. The Centrino chips, released in March, are specifically designed for laptops, and sales should rise from 1 million units in the second quarter to roughly 5 million units by yearend, estimates Pacific Crest Securities analyst Michael McConnel, who notes that Intel will be mounting a major marketing campaign in the fall.
What's more, Intel has benefited from the price war that has been raging since the beginning of the year between PC giants Dell Computer (DELL ) and Hewlett-Packard (HPQ ). Dell recently lowered prices by an average of about 22% -- more than usual during the back-to-school season, says Megan Graham-Hackett, an analyst with rating service Standard & Poor's. Intel's dominance in this market gives PC makers little leverage to demand similar discounts from their suppliers. Indeed, the average price of processors should grow 15% in 2003, estimates Brian Matas, an analyst with IC Insights.
NOTEBOOK UPSTART. Not everything is rosy, however. Intel will see more competition in the notebook market from tiny competitor Transmeta (TMTA ), which unveiled its Efficion energy-efficient chip on Aug. 12. "Our customers think our product is competitive to Intel's [current products] in clock speed [a key measure of a chip's efficiency]," says Mike DeNeffe, Transmeta's director of marketing, who adds, "and we'll be less than Intel in price." That could undercut Intel's margins on Centrino, which have been high and stable since the product's introduction. However, Intel will counter Transmeta's move later this year with a new iteration of the Centrino processor, says a company spokesperson. That should help keep chip prices up.
A bigger problem for Intel could be AMD's new PC chip, Athlon 64, due later this month. The chip could help AMD grab the fastest-chip crown from Intel -- at least temporarily -- and even take away some market share in 2004. AMD believes the chip will "begin a new evolution in computing technology," says John Morris, a senior brand manager at the company. Still, the victory could be short-lived. Intel plans by yearend to come out with its own new chip, Prescott, which will offer at least as good a performance as the upcoming Athlon, says Peter Glaskowsky, editor-in-chief at industry research newsletter Microprocessor Report. "Intel has a long history of underpromising and overdelivering," he says. Intel, says it's "very confident that the performance of Prescott will be outstanding," says a company spokesperson.
The stars seem to be in alignment for Intel. If a recovery kicks into full swing next year, as expected, it appears poised to ride the wave -- with what many investors see as plenty of room to run.
chipguy -
Come on, don't help put the paranoid in 'droid. Saying
something like this in Dan's presence is like rolling the
dessert cart up to Rosanne Barr at a WW meeting.
I don't see his posts but as I remember he won't need any help from me to go over the edge. Can't say anything more or it will be deleted for sure. In fact this will probably get deleted. Who can ever predict?
Edgar -
Using Intel's Fab12, which is scheduled to undergo 300mm conversion, as an example, Intel is shutting the fab down for almost 1.5 years and spending $2 billion. Obviously AMD can't shut down F30 or spend $2 billion so I think they have a problem.
wbmw -
This would make good Dan3 anti-FUD. Just throw it in his face each time he comes up with FUD for one of Intel's products.
Hmmmm.... I wonder if Intel's compilers might generate code that creates that condition?
Wouldn't that be unfortunate...
Thanks Semi! This one is bookmarked.
chipguy -
Has anyone ever told him that Coppermine PIIIs use aluminum
interconnect? ;^)
And the i486DX4 was 3x clocked.
kpf -
A two digit number. I know what comes now.. links or data.
Have no links and only dimensional data of intels capacities, output and inventories. By far not enough to predict anything accurate within the 10% range or so. But enough to determine the ballpark. K.
Interesting!
I don't expect any links but may I ask if your estimate is based in any way from experience in the field?
kpf -
That is why I am the one here who says that tool is not useful for complex devices. As simple as that.
Neither of us can settle this without referencing non public information so let's just agree to disagree.
I'm curious, take NorthWood with a 131mm2 die. What is your estimate of good yield for a complex device such as this? The yield model I posted says 143 GDPW with world class yields. Do you think this is unrealistic? What would you predict?
kpf -
May I offer a suggestion? Use the {i} and {/i} (but use [ & ] instead of { & }). This will allow us to see what is your post and what you are quoting.
kpf -
China has the $$$, look at their trade balance
Well I guess Countries could conceivably fund it but I'm not holding my breath. Additionally, I'll believe schedules when they deliver.
In the meantime FAB30 plus East-Fishkill would be sufficient for AMD to grow into the 30% MSS range in the 90nm node.
IBM has yield problems at 130nm. I'm not convinced they can make Opteron in volume on 130nm much less 90nm. Somebody needs to step up and demonstrate they can make this thing before I get all excited.
UnD -
you had a post a few months ago where you gave figures that, when plugged into the yield calculator, would have given Intel's yields, when interpreted according to accompanying statements you made. Was that a test to see whether we were paying attention? :)
That was an attempt to be helpful. I gave people here a tool to help them make informed predictions about yields. There is someone here who says that tool is not useful for complex devices but I strongly disagree. It does a very good job predicting yields on the most complex devices I've tracked and if you believe me that <.25/cm2 is the threshold for world class yields then you can do the math and see what a fab should be putting out if they're world class. Keep in mind that I've made no claims about Intel's yields and they may not be in this range. AMD on the other hand has made statements that they have the "best yields in the world" prior to moving to .13u SOI. If one ran the numbers at the time that claim was made then it would be very hard to explain the low fab output. I am surprised that there hasn't been another shareholders lawsuit over this, in addition to the pending ones, because it would be pretty easy to prove.
UnD
And my speculation is that that particular battle, with AMD64 added to the x86 side, is too uphill for any other alternative architecture to succeed.
I never argue religion but I will say that A64 has yet to demonstrate the level of superiority you seem to think. This will take time to play out. I've put a few bucks on AMD but I'm not betting the farm.
chipguy -
then why didn't you invest in AMD instead?
I invest in both!
I agree that SGI looks compelling as a speculative
investment. A basically good company down on its luck,
but holding some great technology and a well known
brand.
That's the way I see it too. With Madison they now have the top position in their nitch. If they can just stay afloat until they gather some mommentum they have a good chance. I think they got caught between technologies but now they are on track.
kdf -
Did the Sandia-project help the stock lil bit already?
I think the layoffs last week were the cause of the slight rise.
Semi -
I said in the semiconductor manufacturing world nobody calls them that [dice].
Where I work people would look at you funny if you used the word "dice". It may be grammatically correct but it just isn't used.
BTW: Easy big fella. The thread nannies have really itchy trigger fingers.
aixman -
Ok, time to start another "discussion".
I don't think there is enough 130nm - 90nm capacity in the world to displace Intel. If x86-64 suddenly became the CPU of choice it would be many years before capacity could be put in place to overtake Intel and nobody has the $$$ and process/manufacturing technology to do it. Intel is #1 for years to come if for no other reason than without Intel's fabs and manufacturing infrastructure the world can't produce enough silicon to replace them.
No, I don't have all the data but that's the way it seems to me.
kpf -
I bought some SGI shares last week. They're so cheap that I can't lose much and who knows, maybe they can turn it around? They certainly have compelling technology but such poor financials. Kind of reminds me of AMD.
kpf -
Is Altix 3000 available for sale already?
http://www.sgi.com/servers/altix/
DDB -
Nobody has to believe my numbers if he/she doesn't want ;)
I certainly don't.
Spokeshave -
I think it is clear that either Intel is experiencing very poor yields, or they have a huge amount of capacity sitting idle.
I almost agree with you here. Either Intel:
#1 Has poor yields and is running at or near capacity.
#2 There are significant volumes of products you haven't accounted for.
#3 Has less capacity than you've estimated.
#4 Has excess capacity.
#5 A combination of any or all of the above.
Now, as you know Intel has been pumping a lot of money into Communications. What will happen if they hit the big one and score a major design victory in a recovering market? If they don't have excess capacity then why did they spend all that money developing new comms products? What if Madison takes off? What about the expanding sales of large die Xeons? What is Opteron fails and AMD goes under?
You have been watching AMD revamping their strategy to plan for new capacity should their plans succeed. Why is it so hard to expect Intel to do the same? Of all companies, Intel has shown a history of planning for success, not failure. You keep analyzing Intel's capacity in terms of today's demand. If you were to think like Intel you'd plan for success too.
wbmw -
My estimate did say that 2,777 WSPW of 300mm wafers were required to meet demand, assuming ALL of the 200mm wafer space is used.
You are also making some assumptions about Intel's yields. They could differ from your assumptions, either way.
DDB -
I was being facetious.
kpf -
Sorting is done in the backend. AMD would always know exactly the yields. But IBM as well needs to know all data of it to be able to improve their process. Apart from the business-model, outsourcing needs a very close collaboration.
I'm sure you know IBM's procedures better than I do but I wouldn't think they need to monitor every product so closely. A alternative would be to pick the highest volume product and concentrate on it. Whatever process improvements or controls needed to improve yield on that product would naturally trickle down to the other products as well, assuming all products used the same design rules. Sort of the design version of "copy exactly".
As for keeping things confident, there are always people who know exactly. I would assume the (very few) ones at IBM who need to know are as trustworthy than the (very few) at AMD.
The word you want to use here is "confidential".
kpf -
Assuming IBM would only be used for front-end, and assuming their business model for the Fishkill-foundry opereration to charge only for useable dies
My experience with IBM is they charge for the wafer. I wouldn't want IBM sorting them anyway because I would want to control the test coverage and I would also want to track their yields. BTW, this is why it's impossible to keep yields a total secret if you contract out wafers.
TWY -
Thus, it will always yield an incorrect number and depending on the actual size of the kerf......possibly a very incorrect one. Of course, they don't tell you that. But ...perhaps sophisticated enough to fool you
Did it ever occur to you that they could just subtract out the scribe line?
DDB -
Does your source work at Intel? If so they should send him to some math refresher course. My math courses in university were some years ago.
My math is fine thanks and my yield number was every bit as likely as yours.
kpf -
Btw, the Yield calculator elmer recommended is an approach for fairly trivial one-layer-chips - the implied yield-models are not usable for complex CPUs like Athlon or Opteron consisting eight layers of metal interconnects.
Really! That tells me more about AMD then you probably realize. I've found it to be a very accurate yield model for extremely complex devices with 6 to 8 metal layers, but I can't say who that might be...
DDB -
My source told me that yield for Barton wafers is 96 percent. That doesn't say at which point - so I assume it means 96% working chips on a wafer.
My source told me Intel's yields are 102%.
New TPC-C score from HP running Windows
http://www.tpc.org/tpcc/results/tpcc_perf_results.asp
Keith -
BTW, didn´t you basically trade AMD on the premise that it´s very predictably staying around the 7 US$ mark? What´s your strategy now?
That's close. I base my AMD trading on the observation that it has always returned to the ~$7.50 area. I have short puts out there and CCs in this range. I will roll the CCs as long as I can continue to make money. I will continue to write puts but I don't think I will go above $9-$10. As long as I can make money I'll keep playing AMD but I don't want to chase this one too far until and unless they show real potential for sustained profitability.
Keith -
For me, it´s impossible to predict AMD´s earnings at this point. Only a trend, and that´s looking "up", enough for me to stay fully invested.
When you say "fully invested" do you mean your entire investment capitol or the percentage you have allocated to AMD? If you mean the latter then I guess I'm fully invested too.
Spokeshave -
Perhaps you should refine this statement a bit to say something like "...you can't conclusively determine Intel yield problems to be the cause of your discrepancies...". In the past, you have had no trouble attributing similar such discrepancies for AMD to yield problems <G>.
In this you are right! And on both accounts. <G>
AMD's discrepancies have far fewer unknowns and a conclusion can be drawn with much higher probability of being right. Why do you think AMD stopped announcing volumes?