3.03.2009

"At heart, we're a reverse engineering design company" - AMD

"Advanced Micro Devices Inc closed a deal to spin off its manufacturing operations on Monday, and said it expects the new company to assume responsibility for paying off about $1.1 billion of debt.
The plants which make AMD's chips are now part of a $5 billion joint venture with Advanced Technology Investment Co, of Abu Dhabi, temporarily called The Foundry Co."
- Reuters

In a very creative way, AMD has ridden itself of its crippling debt and the massive burden of capital investment going forward. While AMD may sound as if this strategic move brings them closer to their core expertise, it is without a doubt that this back-to-the-corner decision was the only way for AMD to remain viable. This new lease allows AMD, maybe for a few more product lifecycles to continue and remain as the only challenger to Intel.

At the bleeding edge of semiconductor technology, it has yet to be seen whether a fabless company can challenge one with a foundry. In the not so bleeding edge such as memory products companies with their own foundry like Samsung are dominating over the rest of the industry but competition remains vibrant. But in the x86 space where process leadership creates cost and performance advantages, history isn't kind to fabless companies. Starting this week, AMD is effectively what Transmeta was back in 2000. The difference is Transmeta had a lot of hype going for them and probably with a more compelling product offering in the mobile space.

973 comments:

«Oldest   ‹Older   801 – 973 of 973
SPARKS said...

"Considering that he is using a source that is still pushing the "cost structure" thing, I think says it all. I shot that theory full of holes a while back."

ITK, Yes you did, quite well.


"Surely you can see the difference. As far as I have seen, Intel is not guilty of the latter and if they are, you had better bring compelling evidence."

KORGANO, That piece was excellent, coherent, and well thought out. KUDOS. However, he doesn't have the capacity to do either.

SPARKS

SPARKS

SPARKS said...

"Other partners were a bit more cautious about the new program. One prominent AMD partner on the West Coast confessed that he had not even heard about the Fusion Partner Program a week before its launch. And another Midwestern partner of both AMD and Intel said the legacy of AMD's deals with Tier 1 partners in recent years "still stings."

"The Dell (NSDQ:Dell) thing is still on a lot of people's minds," said the source, who asked not to be named. AMD's late 2006 deal to supply Dell with desktop processors was seen by many in AMD's whitebox channel as a slap in the face."

Remember when poor beleagered AMD thought they had it all? One bitten twice shy. Oh no wait, let me think, how can we twist this so we can blame INTC.

http://www.crn.com/white-box/220300557;
jsessionid=FFHVYNL2H2Y1BQE1GHOSKH4ATMY32JVN

SPARKS

SPARKS said...

There's an article at X-Bit Labs entitled:

"Globalfoundries on Track for 50% Natural Yield by Year End with 32nm Process Technology."

Ok, what is a "Natural Yield"? Is it like a "Operating Profit" in the process world?

SPARKS

SPARKS said...

Well, that didn't work.

"It says to us: "let's put handcuffs and shackles on you and duct tape your mouth"."

http://channel.hexus.net/content/
item.php?item=20509

SPARKS

InTheKnow said...

Sparks, I have no idea what a "natural" yield is. Since semiconductor devices don't spontaneously assemble themselves the term seems out of place to me.

But you missed the really juicy parts of the article.

The 28nm technology offers the smallest SRAM cell size (0.120┬Ám2) currently reported in the foundry industry, and an advantage in die size relative to 28nm “Gate Last” approaches. In addition, the Globalfoundries’ “Gate First” approach to HKMG simplifies 28nm design implementation and IP re-use for customers using conventional poly/SiON-based technology at the 45/40nm and 32nm nodes due to similar process flows and design rules.

and

IBM and its research partners first introduced the “Gate First” HKMG innovation in 2007 as the basis for a long-sought improvement to the transistor to deal with power leakage that emerged at the 45nm node.

First, I'd like to know where I can get one of these 28nm "Gate Last" devices. Since Intel isn't doing any half-nodes that I'm aware of, I have no idea what they are comparing to unless the IBM consortium evaluated a gate last device as an option.

Then there is the bit about the process flows being similar. I'm here to tell you the circuit designers that are going to purchase foundry designs couldn't care less about the process flow. The toughest customer we ever built PCBs for was EMC. They documented our process flows and validated all of our process control systems. But they never once tried to convince us that some other process flow might be better. All they really wanted to know was "can you hit our specs and can you assure us that you can do it reproducibly."

Now the bit about design rules does have some value, particularly if you want to do a dumb shrink from 32nm to 28nm. But if your design rules are too restrictive or don't fit my needs, I'll just take my design somewhere else. So a foundry's design rules need to be as broad as possible. I just don't think they are bringing anything different to the table here that what I would expect from any other foundry.

And that last bit is my favorite. They "introduced" Hk/Mg in 2007. Tomorrow starts the 4th quarter of 2009. So just where can I get a gate first Hk/Mg device 2 years after it's introduction? I can't you say? Surely you jest!

Oh, and lets not forget that gate leakage is something that "emerged" at the 45nm node. Didn't gate's leak prior to the development of the 45nm node. Did the IBM consortium discover gate leakage too?

In short, all I see here is a huge steaming pile of marketing manure.

Orthogonal said...

Anyone else excited about the prospects of Light Peak? I know it isn't quite as sexy as operational 22nm SRAM test chips or as jaw dropping as the prospects of Gulftown, but this thing sounds really cool. I'm still a little fuzzy on the exact implementation, but the idea of a single cable standard to run pretty much any and all I/O protocols is a great and promising idea.

http://techresearch.intel.com
/articles/None/1813.htm

Anonymous said...

Sparks you forgot your decoder ring again?

"Yield progress continues with 24Mb SRAMs in double-digit natural yields on path to 50 per cent natural yields by year-end."

- On path means - we hope we can get there (or at least that's what the engineers tell management).

- Not sure how strong the market is for 24MB SRAMs working at (conveniently undefined) speed, yet alone how that will translate to actual logic... SRAM is a "natural" :) first step but that's all it is. When you start pushing performance on logic parts you could run into some non-defect related yield issues which you may not see on simple SRAM structures.

- Double digit obviously could mean 10% and certainly means <50% as they are on a 'path' to get there.

- Yield on a wafer (or small subset of wafers) does not equal High volume yield. It is easy (or I should say easier) to get yield fixes working on a single wafer and then have them work not so great in volume. It's not like AMD is running a massive amount of volume baseline 32nm material... oh and again it's SRAM only!

One would think if they will be truly be at 50% natural yield (whatever the hell that means) by year end, they'd be queuing up risk prodiuction starts in Q4 and have 32nm production material out in Q1'10 (and either on the market late Q1 or early Q2.

You'd also think if the yield was thiis good you'd be seeing 32nm samples floating around... but again it's only SRAM!!!!

- Finally it's easy to put a yield roadmap together (or I should say a "path"?) to fix X, Y and Z yield issues and also demonstrate those fixes individually, but when you put them all together sometimes they don't add up or introduce new issues. The other problem is when your yield is really low, you often times have no idea what new failure modes you wil find when the first one is fixed... In other words there may be 2 or 3 causes to a specific yield issue and you may think it is only one until you implement it and realize there's more to the problem.

Given the complete subjectivity of the statement, the use of complete marketing/ and nebulous engineering terms (probably to introduce wiggle room), there's not much to glean from it.

Tonus said...

Khorgano: "Tonus, I understand the distinction you're trying to make, but in reality there is no difference in the scenario's you've made. Whether it's offering volume discounts or preference discounts it's all the same. The fact remains that it's not Intel that's being "anti-competitive", it is AMD that is being uncompetitive."
I'm thinking about it in the legal sense. I'm afraid I don't know enough about the specific laws in this instance (or even if they exist) but the only way the scenarios I describe would be the same is if there are only two competitors in a particular market, and if the market was much narrower than it is. And the law is rarely (if ever) written with such narrow scenarios in mind.

Setting terms in order to boost your own sales is not the same as targeting a specific competitor in order to cut his sales. It may have a similar effect in the CPU market today, but that is unlikely to be relevant in the legal sense, especially since the first example (offering quantity-based discounts on your own product) means that you do not have to use the latter in order to boost your own sales. This would also mean that Intel could offer broad-based cuts across their line of products while targetting a narrower range of AMD products (ie, we'll cut 5% off of our high end product if you block AMD's sales of low-end product). This could potentially harm both OEMs and customers. (Which would be a reason to believe that Intel didn't do this, as it would probably be illegal but it also pisses off your customers at both levels.)

Now, assuming that Intel did in fact use this tactic, I believe that that in itself would be an indicator of the respective market positions of the two companies. If all else were equal and Intel had offered OEMs a bonus for limiting their purchases of AMD products, they'd probably be told where to stick their offer. But things are not equal, and OEMs understand that Intel has the werewithal to provide them with CPUs if AMD was unable to, but the reverse is not true. Only a company with that level of domination in a market would be able to make such an odd offer to companies and get away with it.

--continued below--

Tonus said...

That said, I believe that most OEMs don't have any sense of allegiance to Intel or AMD. They probably don't even care about the competitive aspect from a general standpoint (Dell's competition isn't Intel, it is HP, IBM, Gateway, etc). They may not even worry about Intel's ability to dictate terms-- barring direct action against an OEM, which could easily bring government intervention, I doubt that Intel is interested in harming any OEM for its own sake. Having more than one supplier for any particular part is probably more of a hassle for an OEM than anything else. If OEMs have it in for Intel or AMD, it's probably a personal issue and not conducive to good business practices. (This also seems like a good reason for Intel not to offer incentives designed to harm AMD directly.)

So this is not, in my mind, an issue of who is at fault. AMD is, IMO, primarily responsible for its current situation. For years they had problems matching Intel in speed and performance, and more importantly they had supply issues for a time (this is aside from the fact that they cannot match Intel's overall output). When AMD seemed to finally clear both of those hurdles and Intel was struggling with the late P3 and early P4, OEMs were not sure that they could trust AMD. The biggest factor in AMD's inability to gain more marketshare than they did was that OEMs were worried that AMD could not produce enough CPUs for them and that they might not be able to follow-up on a great CPU design before Intel got its house back in order. If Intel really is the bully that they're said to be, then OEMs would have loved to stick a finger in its eye back then. But they had to be sure that they had an alternative that was good enough that they wouldn't find themselves crawling back to Intel later on. I don't think they trusted AMD to do that.

And the current situation, in which Intel has more say in AMD's prices than AMD does, confirmed those fears.

pointer said...

Orthogonal said...
Anyone else excited about the prospects of Light Peak? I know it isn't quite as sexy as operational 22nm SRAM test chips or as jaw dropping as the prospects of Gulftown, but this thing sounds really cool. I'm still a little fuzzy on the exact implementation, but the idea of a single cable standard to run pretty much any and all I/O protocols is a great and promising idea.



Apple will be likely showing us how to use it :)

Khorgano said...

Setting terms in order to boost your own sales is not the same as targeting a specific competitor in order to cut his sales.

Actually, there is no difference. This is why I said there is no change in aggregate demand. Sure, it could cause a short-term spike/dip over the course of a month or two, but over the long term of many quarters and years (which is the timeframe we're talking about) there is no effect on aggregate demand. The business tactics trend toward the same outcome.

When AMD seemed to finally clear both of those hurdles and Intel was struggling with the late P3 and early P4, OEMs were not sure that they could trust AMD. The biggest factor in AMD's inability to gain more marketshare than they did was that OEMs were worried that AMD could not produce enough CPUs for them... But they had to be sure that they had an alternative that was good enough that they wouldn't find themselves crawling back to Intel later on. I don't think they trusted AMD to do that.

AMD's perception among OEM's is their problem, and theirs alone. If AMD needed more capacity then they could have easily done a secondary stock offering/sale of bonds/sale of underperforming business units years ago. If OEM's don't trust AMD, that's their fault, not Intel's.


And the current situation, in which Intel has more say in AMD's prices than AMD does, confirmed those fears.

This gets back to my original point which you have yet to address. This is not an Intel so-called "Anti-competitive" problem, but an AMD competitiveness problem. Intel can dictate their pricing schedule because they have the superior product and cost model. If AMD was more efficient and had a superior cost model (better products certainly help), then regardless of what sale tactic Intel uses, they can position themselves in the market accordingly and the "anti-competitive" problem takes care of itself.

If AMD had the margins to dictate their own pricing, they would and Intel's main bargaining chip would be gone. It's really that simple (from an economic point of view, obviously execution is more difficult).

Look, over the past couple years, AMD's break even point was in the $1.9B - $2.0B which is rather high for them, this is why they had such huge and punishing losses in the last 3 years. With the recent reorg announcements, cuts in spending etc... they have now dropped the break even point to the $1.4B - $1.5B range. Had they done this years ago, not only would their losses be significantly reduced, but they would have had several profitable quarters. I hope you're getting the picture. AMD can do it, it's just a matter of if they can.

Tonus said...

"Actually, there is no difference. This is why I said there is no change in aggregate demand.
"

It's not demand that is the focus, it's supply, and attempts to artificially throttle someone else's directly. As I explained, there is already a mechanism for providing rebates that does not require a business to target a specific competitor. It's not logical to target a single competitor's potential sales as a way of increasing your own.

"This gets back to my original point which you have yet to address. This is not an Intel so-called "Anti-competitive" problem, but an AMD competitiveness problem."

I addressed it when I said "So this is not, in my mind, an issue of who is at fault. AMD is, IMO, primarily responsible for its current situation." I specifically placed the bulk of the blame for AMD's problem on AMD in my explanation.

But that remains aside from my comment about practices such as offering rebates to reduce purchases of someone else's items. That still strikes me as illegal and anti-competitive behavior.

Tonus said...

Having read back through the comments, I just want to clarify my point-- the extent of AMD's fault in its own problems is not what determines whether an action is deemed illegal according to the law. If you're caught stealing from someone, the fact that they happen to be indebted to you does not mitigate in your favor.

If Intel is accused of anti-competitive behavior, it is based on whether or not they did things that violate the law. It is not based on whether AMD did a poor job of running its company.

How much AMD is at fault for their problems becomes relevant if Intel is found guilty of anti-competitive behavior, as it may have a bearing on any penalties that are handed down (ie, if Intel's anti-competitive behavior had little impact on AMD because they ran their own business so badly, the penalties against Intel may be less than anticipated).

I've used the USFL/NFL example in the past. The NFL was found guilty in civil court of conspiring to harm the USFL. But the jury felt that the USFL's financial problems were of its own doing, and granted them only a $1 award. If AMD is guilty of its own bad performance, that by itself would not stop a court from finding that Intel broke the law.

Khorgano said...

"It's not demand that is the focus, it's supply, and attempts to artificially throttle someone else's directly. As I explained, there is already a mechanism for providing rebates that does not require a business to target a specific competitor. It's not logical to target a single competitor's potential sales as a way of increasing your own.


Maybe you can clarify or reword this for me since that doesn't make any sense. How does Intel rebates have anything to do with AMD's supply? AMD control's their own supply with their own manufacturing? Did you mean throttle their sales?

Further, and most importantly, why can't Intel put terms and conditions on use of rebates? If they are not fraudulent, then the OEM's are free to accept or reject those terms. Is Intel anti-competitive because they have a strategic partnership with Apple as a single source supplier who gets sweetheart deals? Does it even matter? Or is it ok because Apple has a relatively low marketshare? What if their marketshare was 10%, 15%, 30%, 50%. At what point is it wrong?

Ultimately, you need to justify why Intel can't put terms on the sale of their products. If an OEM freely agrees to it, why is that wrong? AMD just needs to offer a better deal and it would get made.

Finally, it IS logical to target another competitors sales directly to benefit your own. How do you think business works? You don't think Coke and Pepsi go after each others clients? Nearly everywhere you go, restaurants, theaters etc. they are exclusive purveyors of those products at the expense of the other.

"If Intel is accused of anti-competitive behavior, it is based on whether or not they did things that violate the law. It is not based on whether AMD did a poor job of running its company.

You're right, but for all the wrong reasons. If they broke the law, they deserve to pay for it. However, there is nothing wrong with the alleged activities of Intel. Just because countries have anti-trust laws, doesn't make them right. Government control and regulation of markets does nothing to foster competition. Can you explain how the EU decision to take a money grab of Intel for over $1B does anything to help the progress of Semiconductors? There is no benefit to consumers, no benefit to AMD and now OEM's have lost a rebate option from Intel to lower their costs.

SPARKS said...
This comment has been removed by the author.
SPARKS said...

ITK, G, sorry I didn't get back sooner. (Stripers out at Montauk last night!)

Thank you for the clarification. However, in my case, a decoder ring wasn't enoungh. An Enigma Machine or a week at Bletchley Park would have been the order of the day!

Yes of course, I (again) forgot the memory thing, as opposed to the logic thing.

SPARKS

Anonymous said...

The memory thing isn't that bad - SRAM is a good first step, but you do run into new issues with logic... memory is more or less just the same damn cell over and over.

The real issue is the "double digit" yield claim which could be as little as 10%... it's impossible to say something is healthy or on track based on "double digit yield"

Or to put it another way... Intel already has working 22nm SRAM chips - don't know if this is one or a handful per wafer,but I guess they could claim "single digit" or who knows, possibly "double digit" yield on 22nm at the same time AMD (ummmm... "global foundries") is making claims on 32nm. Though I don't think Intel is as far along on 22nm as AMD is on 32nm - that would mean a full generation lead.


It's just another one of those nebulous intermediate press statements to confuse folks... I guess all the other PR slants have grown old "ramping production" (remember how they were ramping 45nm in H1'08.... well they were but of course that means add 6-9months for actual product out), mixing and matching the various different 32nm processes, making claims and comparisons to some competitors 28nm gate last technology which doesn't exist, and then going into shipping product, risk production, accepting orders, and other nebulous timelines which insiders understand but the press assumes means product available.

It's all a shell game designed to confuse... all they have to say is this is when the first 32nm SOi product will be available comercially, this is when the first 32nm bare Si product will be available... but those timelines won't sound as impressive.

Tonus said...

"How does Intel rebates have anything to do with AMD's supply?"

You said that it didn't matter because aggregate demand remains the same. However, Intel isn't being accused of trying to change the level of demand, but the level of supply of AMD products available to consumers via OEMs.

"Further, and most importantly, why can't Intel put terms and conditions on use of rebates?"

I never said that they could not.

"Is Intel anti-competitive because they have a strategic partnership with Apple as a single source supplier who gets sweetheart deals?"

Exclusive contracts are, by definition, anti-competitive.

"At what point is it wrong?"

I have not discussed right or wrong. I gave my opinions on whether a specific action that Intel is accused of is legal or anti-competitive.

"Ultimately, you need to justify why Intel can't put terms on the sale of their products."

I never said that they could not.

"Finally, it IS logical to target another competitors sales directly to benefit your own."

Competing implies confidence in your product and services. Attempts to restrict supply via non-competitive practices implies lack of confidence in your product and services. Promoting a product that you lack confidence in is not logical.

"there is nothing wrong with the alleged activities of Intel. Just because countries have anti-trust laws, doesn't make them right."

The morality of market regulation is a different discussion. I was discussing what is, not what should be.

"Can you explain how the EU decision to take a money grab of Intel for over $1B does anything to help the progress of Semiconductors?"

Why are you asking me to explain something that I have not claimed, and which I do not support?

"There is no benefit to consumers, no benefit to AMD and now OEM's have lost a rebate option from Intel to lower their costs."

Even assuming that the accusation is true, Intel is still free to offer rebates. OEMs are still free to negotiate contracts. And consumers do not have their choices artificially limited. I don't see where there is a problem.

Khorgano said...

Competing implies confidence in your product and services. Attempts to restrict supply via non-competitive practices implies lack of confidence in your product and services. Promoting a product that you lack confidence in is not logical.

This is a non-sequitor, confidence in your product has nothing to do with competition. Also, exclusive contracts are not "by definition" anti-competitive, they are in fact competitive. If you beat out your opponent to win a clients business, that is due to competition.


""Ultimately, you need to justify why Intel can't put terms on the sale of their products."

I never said that they could not."


You most certainly did, if Intel is forbidden from offering exclusive contracts with OEM's, that is a restriction on the terms they can use to sale their products.

Look Tonus, you seem like a well meaning person, and I've enjoyed our discussion thus far, but I think it's time we take a step back from this current line of reasoning and look at the big picture.

I am not for or against Intel, and I am not for or against AMD, I am for freedom and against the use of force. My questioning boils down to one thing. Is it ok to use aggression or force against another party? If no, then lets take it to the next step. If 2 parties freely agree to a mutual exchange/transaction or contract, is it ok for a third party to use force against them and forbid it? If no, then take it to its logical conclusion and Intel's use of rebates or exclusive contracts with a willing OEM are perfectly fine. Otherwise you are advocating the use of force or aggression against a free exchange.

If you believe force should be used to ban this, then this is where we part ways and we agree to disagree. If so, I would ask that you re-examine your position. Granting someone the use of force (i.e. government) to ban free exchange is something that should not be taken lightly. Throughout history, this use of force has led to unfortunate conclusions.

Good day

Tonus said...

"This is a non-sequitor, confidence in your product has nothing to do with competition. Also, exclusive contracts are not "by definition" anti-competitive, they are in fact competitive."

What is your reasoning behind this? You're just responding "no, it's not." I am genuinely interested in why you hold those views or disagree with my statements.

"If you beat out your opponent to win a clients business, that is due to competition."

But you have also excluded them from that market, which means that you have no competition on the purest level-- the consumer. When you exclude a potential competitor, then they cannot compete. Ergo, exclusive contracts are by their nature anti-comptetitive.

"You most certainly did, if Intel is forbidden from offering exclusive contracts with OEM's, that is a restriction on the terms they can use to sale their products."

I have not, at any time, proposed that Intel be limited or restricted in any way in how they do business. As I explained in my previous post, I have been discussing existing laws and statutes, as those are what will affect AMD's lawsuit against Intel. I have not discussed right and wrong, and I have not discussed my own beliefs on how markets should work. I have tried to analyze the current situation dispassionately.

If you feel that I've been discussing anything other than that, despite my assurances to the contrary, please excuse any confusion on my part. But I have simply been, from the start, discussing what currently is (or may be, as I have admitted not being sure regarding the basis of some of my opinions) going on and why the case of AMD vs Intel may shake out a certain way.

Honest, read through my posts. If there are any statements that give the impression that I'm moralizing on the issue, I will explain them or take them back.

"If 2 parties freely agree to a mutual exchange/transaction or contract, is it ok for a third party to use force against them and forbid it?"

What if the agreement between those two people is harmful to me?

Sorry, couldn't resist. That one was too easy. :)

"but I think it's time we take a step back from this current line of reasoning and look at the big picture."

I think what you want is for me to clarify my personal stance on market economics. I was purposely avoiding it because it winds up distracting from the discussion, but I think we've wrung what we can out of it.

I am at heart an unabashed capitalist. No, I don't like government interfering in business aside from whatever minor necessary regulation may be required to provide an even playing field for all entrants. I am not in favor of government placing limitations on agreements that companies may reach (although I dislike those agreements and would prefer that companies allow the market to determine winners and losers on its own). I feel that the consumer defines the market and should be its ultimate focus.

Bear in mind that I do not trust business people any more than I trust politicians. But business people have, IMO, a more reliable motivation to play by the rules. I think this is easiest to fix by limiting the damage that government can do and remove any incentives it has to allow business to do any damage. However, the only way to do that is via regulations and restrictions (although I do like the thought that if we're going to restrict freedoms, it is best to restrict that of the people who run the government). There's no happy medium when you have markets (money) and government (power).

In any event, I don't take these discussions personally. If it's a discussion that I find of interest, I'll participate. I don't mind if anyone believes differently from me or thinks I am wrong. It's the internet and I'm a big boy, etc etc. Sometimes I prefer to debate an issue in isolation because I find that it's a good way to analyze a situation without letting personal bias creep into it.

SPARKS said...

"Anyone else excited about the prospects of Light Peak? I know it isn't quite as sexy as operational 22nm SRAM test chips or as jaw dropping as the prospects of Gulftown......"

Sure, go ahead Ortho, rub it in. Sure, leading edge tech can get sooo tiring at times. You work at the highest tech factory on the planet, you get to see all this stuff before it's released, (I know, I know, hush, hush and all that),and then "test" the juicy "engineering samples."

Peh, you eat all the Filet Mignon, while we all speculate on the flavor of the croutons.

I'm licking my chops on Gulftown, third gen Hi-K @ 32nM, hooked up to WIN7 64.

I can smell the chips on the grill (with the unlocked multipliers) from here, and that's ~2500 miles away.

Just another day at the office, eh?

SPARKS

SPARKS said...

"But business people have, IMO, a more reliable motivation to play by the rules."

Sure they do. They have to play hardball with the politicians.

SPARKS

InTheKnow said...

Well, I have found at least one interpretation for "natural yields". All I had to do was wander into the zone and look for our old friend Abinstein. Here is his take...

He is comparing GF's natural yield (one that is not helped by redundancy) with Intel's redundancy enhanced yield.

So what Abinstein is claiming is that what GF is calling "natural yields" are simply Non-redundant yields under another name. While I can buy that, I have to debunk his claim regarding Intel using redundant yields.

In order to use redundant yields, you have to have redundant circuitry in the transistor. While someone may have done such a thing I'm not aware of anyone that builds redundancy into SRAM structures. So when we are talking about SRAM, which is what all processes are developed and validated on, NO ONE is using redundant yield numbers.

I have no idea why Abinstein would think that Intel is fudging these numbers since Intel does not make their yield numbers public. It does Intel no good to lie to themselves.

As you might expect, though, he couldn't leave it there and had to dive off into some fantasy land where redundancy is some sort of miracle pixie dust that results in perfect yields. I offer his follow up comment for your entertainment without additional analysis. Trust me, this claim just isn't worth any real analysis.

A 20% natural yield can be increased to 60% to 70% with minor redundancy added. (50% natural yield plus redundancy would've pushed the overall yield to close to 100%)

pointer said...

ITK: ... ... As you might expect, though, he (Abinstein ) couldn't leave it there and had to dive off into some fantasy land

he has low intolerance towards any stuff showing bad on AMD .. even if it is GF ... but when someone said the GF delay is due to AMD ...

someone said:
(AMD is the gating factor on 32nm release, since fab guys are ready even in 1st half of 2010, it depends on AMD's execution now)"


He replied: We know this is basically what GlobalFoundries have claimed. But do you have evidence to back this up, or just a guess?


The funniest part of the reply is that he ask people to back the claim of GF or else just being a guess ... isn't what GF said can be a supporting statement (while not necessary true btw), what else an unrelated individual can prove beside depends on those statement ? at least AMD has not denied it.

I keep wondering what would he said if some one in the same thread state an absolute reverse statement .. AMD has to delay its 32nm design because GF is not ready ...:)

Anonymous said...


In order to use redundant yields, you have to have redundant circuitry in the transistor. While someone may have done such a thing I'm not aware of anyone that builds redundancy into SRAM structures. So when we are talking about SRAM, which is what all processes are developed and validated on, NO ONE is using redundant yield numbers.


ITK -- you are wrong here. SRAM is the thing that has a fair amount of redundancy built in. It's just a bunch of repeating features, after all. After processing, you "turn off/fuse" the extra sram you don't need/want.

Logic circuitry generally has -zero- (or very minimal) redundancy. There just isn't the room for it.

With respect to yields, I know for a fact that all the major players track "non-redundant" yields on their test vehicles. (which is almost, but not quite, exactly the same metric as defect density.) That will tell both the process guys and finance guys what to ultimately expect when they ramp to high-volume.

I actually think that "natural yield" might be *redundant" yields... ie, what the current defect density will give you in terms of working product. Without being too specific, being around 50% would put them pretty close to about a year out from selling product. If that 50% is "non-redundant" yield, then they are much further away than that.

InTheKnow said...

ITK -- you are wrong here. SRAM is the thing that has a fair amount of redundancy built in. It's just a bunch of repeating features, after all. After processing, you "turn off/fuse" the extra sram you don't need/want.

And therein lies the danger of assumption. :) I've never seen a redundant yield reported on SRAM, and so assumed that there was no fusing built into the structure. Thanks for correcting my error.

I actually think that "natural yield" might be *redundant" yields

I agree that Abinstein is probably right on this.

To be charitable, Abinstein could have been talking about SRAM when referring to the high yield numbers he was was tossing out there. The problem with that, as you clearly stated, is CPUs don't have much if any redundancy in the logic circuits. The cache does, and this certainly helps yields, but defects in the logic areas still have a high kill rate. And since no one has yet discovered the big market for SRAM, CPU yields are the only thing that counts in the long run.

being around 50% would put them pretty close to about a year out from selling product.

Since they expect to be at about 50% yield in 3 months time (around December), then they would be ready to introduce 32nm at the end of Q4'10. That is right on time for a late Q1 early Q2 ship date for first products. So it looks like GF and AMD's schedules are pretty much in synch at this point.

InTheKnow said...

ITK wrote you have to have redundant circuitry in the transistor.

That didn't come out the way I intended at all.

Obviously, there is no redundancy within the transistor. It would have been far more correct to say that you had to have redundant routing/functionality in the circuitry.

Khorgano said...

I apologize to those who think Tonus and I are distracting from the other conversations, but I think we've nearly reached an impasse, but I digress.

What is your reasoning behind this? You're just responding "no, it's not." I am genuinely interested in why you hold those views or disagree with my statements.

My point of it being a non-sequitor, (aka doesn't follow), is that your confidence in your own product has little to do with how you sale your product. Whether you believe it to be world class or entry level crap is irrelevant. You're still going to try and make the best deals possible. However, continue reading below since I think it better adresses your true concerns.

"If 2 parties freely agree to a mutual exchange/transaction or contract, is it ok for a third party to use force against them and forbid it?"

What if the agreement between those two people is harmful to me?

Sorry, couldn't resist. That one was too easy. :)


How quant, but I think my point went right over your head. Allow me to expound this. I agree with you, if your actions cause harm to someone, then you deserve to be prosecuted to the fullest extent of the law. In this point, I think we're in accord. Where the disconnect comes from is how each of us qualify "harm". Judging by your personal beliefs, we would both like to see a market that is free from aggresion and harm, but the root cause of our disagreement, is what qualifies as that harm.

To me, anti-competitive behavior is one in which force or aggression is used to influence market outcomes. As I've stated earlier, it would be things like: Fraud, defamation, libel, sabotage, corporate espionage, vandalism etc...

A competitive behavior is one in which the parties involved in a transaction do so freely without deception/fraud/force etc. If you lose a client's business because a competitor made a better offer or the client just chose to use them exclusively, that is not a form of aggression, that is just called business. If you are going to argue that this is aggression or harm (like in the case of Intel vs AMD), then using this logic, a reductio ad absurdem would conclude that ANY transaction Intel makes is harmful to AMD, and any transaction AMD makes is harmful to Intel. With this type of conclusion we are then well on our way to centrally planned economies and/or communism.

There are nothing wrong with exclusive contracts. Going back to my point about Apple. Maybe the correct question to ask is, Why is AMD entitled to Apples business? Should Apple be forced to purchase from AMD in the interest of "competition". What should be done to Apple if they refuse?

I am not in favor of government placing limitations on agreements that companies may reach (although I dislike those agreements and would prefer that companies allow the market to determine winners and losers on its own). I feel that the consumer defines the market and should be its ultimate focus.

You're partially right, but mistaken in assuming that only end user's are consumers. Consumers exist up and down the production chain. Intel is a consumer of AMAT, and Nikon. HP and Dell are consumer's of Intel/AMD. Company XYZ are consumer's of HP and Dell etc... It is the collective decisions of millions of people and different entities that comprise the market and no government or regulation can properly account for the voices of all those desires.

In any event, I don't take these discussions personally. If it's a discussion that I find of interest, I'll participate. I don't mind if anyone believes differently from me or thinks I am wrong.

Agreed, this has been fun and this is healthy debate. It's not often I find people who can intelligently discuss things without resorting to personal attacks and disrespect.

SPARKS said...

"Agreed, this has been fun and this is healthy debate. It's not often I find people who can intelligently discuss things without resorting to personal attacks and disrespect."

Well said, I agree. Well done, both.

There's a fine line that separates the point at which you both have reached the impasse. I seriously doubt the EU gave the debate the attention you both did.

The fact is, the EU has a vested interest (1.2B EU) in recouping their 'investment' with AMD in Dresden. Conversely, the United States has a vested interest in the survival of INTC, with billions in tax revenue and jobs, among other things.

In the end, it will be politics and self serving interests that will prevail, not a well debated intellectual argument.



SPARKS

InTheKnow said...

being around 50% would put them pretty close to about a year out from selling product.

Sorry, I missed my queue there.

That is the point where I'm supposed to bring up APM or some other secret sauce and tell you that GF will be able to get this puppy dialed in and churning out perfect wafers in about 6 weeks. :-P

And, yes, I do know that APM will be useless for defects. I'm not aware of any automated system that is capable of digging into a tool and finding out where the defects are coming from. For those that bothered to read my long rambling explanation of APM a while back you may remember I said that APM wasn't any good for defects back then.

Anonymous said...

APM may not be any good for random defects - true - but to account for known process variability - maybe.

APM runs a serious risk in responding to every little non-ideality which can end up making the problem worse. My stats class guy called it a word which I don't recall.. but it's bad.

IF they can truly identify process shifts real time (that isn't noise in the data itself) AND know that it's a global thing AND have a means of counteracting it (which I can't come up with a good example for), THEN you have a shot at something big.

Maybe something like an anneal temperature was bad, so the next step needs to up a voltage by 0.1V or something. (shrug) The places you can do this are few and far between and are certainly NOT the largest sources of defects in the process.

Just run your damn tools within their control limits and be done with it.

SPARKS said...

I swung over to MaximunPC today and I found some info on ThermalTake's beautiful Level 10 case. I can't wait to get my hands on this 47 pound behemoth. (Netbook my ass) I can imagine myself as one of the apes in '2001: Space Odyssey' feeling it's black monolitic shape with 'Also Sprach Zarathustra' playing in the backround. A perfect home for a new 6 core build-up! 700 bucks, YIKES!

Dum, Dum, TA-TA!

http://www.maximumpc.com/article/news/
exclusive_thermaltakes_jawdropping_level_10
_chassis_unboxed

SPARKS

Khorgano said...

I should probably leave well enough alone, but when I responded last night, I must have glossed over this paragraph or didn't completely register it. I think I was a little hasty and it was getting rather late.

I have not, at any time, proposed that Intel be limited or restricted in any way in how they do business. As I explained in my previous post, I have been discussing existing laws and statutes, as those are what will affect AMD's lawsuit against Intel. I have not discussed right and wrong, and I have not discussed my own beliefs on how markets should work. I have tried to analyze the current situation dispassionately.

Sure, if you're going to do business internationally, you have to play by their rules. If you break one of their laws or statutes, then yes, technically you are in the "wrong" and must suffer the consequences. However, my entire argument has not been based on whether Intel broke some arbitrary statute in some backwards country (and I use that term broadly), but whether or not their actions were anti-competitive in the purest sense. Not what some bureaucrat or regulator in the EU or Japan or anywhere else thinks.

Frankly, discussing how the DoJ or whatever government regulator decides in the Intel vs. AMD case is not very appealing to me. Sparks put it succinctly, regardless of outcomes, the politicians and special interests will rule the day. I apologize that I never fully understood your position and where you were coming from.

When you say that you want the market to decide outcomes, but what I've been trying to say all along is that the market did decide. Just because you don't like it doesn't mean the government should get involved and amend what you perceive as an injustice. It's fine if you believe that a company should do something differently, or you have a better idea on how it should be run, but the moment you allow government to intercede I have a problem with it. Because now you have introduced a force that changes market outcomes, ergo, an act of aggression. You see, this is why our economy is in so much trouble, we've allowed all the special interests to lobby congress and make policies that shape the market to function how they want it to work and are essentially creating entities with priviledged status. This is what I mean by unfair and anti-competitive in addition to whatever someone may personally do.

The reason why the government always wins is because they have something that we do not, and that is a monopoly on force. The government can intervene in the market at any time, at any place and whenever they want and you can't do anything to stop it. (see the recent housing bubble and subsequent collapse for example). When their interventionist policy inevitably collapses in on itself, the government then points fingers at everyone else to assign blame and then attempt to fix it by trying to tax/borrow/spend their way out of the mess without ever fixing anything or taking responsibility.

If we are to have a truly fair market, the government needs to be taken out of the equation all together. When there are no priviledged entities, then a free market is one where your desires and ends are weighed against the collective desires and ends of everyone else and when your wants align with someone elses and you both decide that a trade would be mutually beneficial, a sale is made.

Again, I apologize if this is not the discussion you meant to have and I'll understand if you choose to reply or not, but I've made my case and I think the discussion has run its course.

Thanks

InTheKnow said...

My stats class guy called it a word which I don't recall.. but it's bad.

I believe the word you are looking for is tampering. And the risk of tampering for those that don't know, is that you may actually introduce more variability than you remove.

SPARKS said...

"If we are to have a truly fair market, the government needs to be taken out of the equation all together."

Too late, the Genie is out of the bottle, it has been for decades, and there's no going back. Government is bigger and more pervasive than it's had ever been. The political authoritarian long term objective is to control the markets. In a way they already are. SEC

Let's start with power, exclusively controled by government sanctioned authorities. They tell you where and when you can build a nuclear power plant, then tell you when to take it down. NRC and NYS vs. Shorem

Agriculture, they tell you what to grow and pay what not to grow. Talk about economies of scale, the small farmer is a thing of the past, despite the vain efforts of USDA.

Housing-Fanny May, etc., government sponsored debacles causing last year tragic mortgage meltdown. Today interest rates are so low you are competing with Uncle Sam to earn any interest from the banks. Don't believe me? My buddy on Wall Street put $100K on a 6 month treasury note. His return? $6, yes SIX DOLLARS!

Need I mention tariffs?

For every type of business you can think of, there's a pervasive government regulation of one sort or another. Do we need side impact airbags? How much do they add to the cost of the finished car? DOT

Taxing cigarettes, liqueur, no smoking in bars, come as the lawmakers continue to add, refine, and amend new laws, while our personal freedoms erode further.
ATF

Government intervention of any sort strives to create equality without regard for the long term negative effects of such regulation. In most cases, for the most part, it's always detrimental in the long term, to someone else or some other group, usually some unexpected, unforeseen scenario. All things are not equal, they never will be.

Let's hope for this countries sake they stay the hell out of the semiconductor/software industry, and let the market run itself. It is too big and complicated for the most intelligent politician (yes I know a contradiction in terms) to grasp it on the most simplest terms. No matter what position you take, ANY action against INTC, or Microsoft, would have catastrophic effect on the entire industry as a whole. The long term trickle down effect for the entire industry would weaken the United States in the long term on the whole.

This is precisely what the Europeans and Asian's would love.

As long as the politicians can see a revenue gain in the short term, the long term deleterious effects are inconsequential, so long as said politicians' political tenure is completed. I give you AMD and the NYS debacle, a failure before the first hole was dug. MY TAXES ARE PAYING FOR IT!

1.2 BILLION FUCKING DOLLARS OF NYS TAXPAYER MONEY TO SUPPORT RICH (OIL) ARAB OWNED GLOBAL FOUNDRIES, WHILE THIS STATE FACES BANKRUPTCY!

As ITK put it:

"I believe the word you are looking for is tampering. And the risk of tampering for those that don't know, is that you may actually introduce more variability than you remove."

The same is true with a government/political influence on any "free" market sector.


SPARKS

SPARKS said...

Oh, let's not forget the intrusions of the FTC with this new tidbit.

It's never gonna stop, folks.

http://www.theinquirer.net/inquirer/
news/1557524/us-regulate-bloggers


SPARKS

Anonymous said...

EUV litho still looks sketchy...

http://www.eetimes.com/news/semi/showArticle.jhtml;jsessionid=LXJYQ1JB1NWX5QE1GHRSKH4ATMY32JVN?articleID=220301445

An analyst speculated Hynix, IMEC, Intel, Samsung, Toshiba and possibly TSMC are the initial customers for the NXE 3100. But one system will probably also go to the Albany research center that along with IMEC received the first prototype tool.

Wasn't IBM talking up 22nm and EUV not too long ago? :)

I'd expect with almost certainty that immersion is the solution at 22nm and who knows maybe even 15nm...

The article also talks about ASML (one of the big litho equipment suppliers) targetting a production type throughput in the 2015 range (although there would likely be "production-ready" tools at an early date with a lower throughput).

Figuring 32nm is late 2009, 22nm is late 2011, volume 15nm would be 2013/2014 for the technology leaders. This would mean ramping/installing tools in early-mid 2013 and having pretty close to ready production tools for development in early 2012. Not sure this is going to happen.

InTheKnow said...

Here is a sneak preview of Intel's 32nm from Chipworks.

The early speculation is that Intel has incorporated SiC stressors into the NMOS structures. That would be an industry first in a mass produced device. Once again Intel would have moved something out of the science lab and into production.

It looks like Intel has improved their low-k intermetal dielectric used in back end processing. I guess IBM isn't the only one that can do that. :)

There is also evidence that Intel has changed the process flow on their gate last processing to incorporate the high-k material after the dummy gate removal.

InTheKnow said...

Justin Ratner has given us a little bit of a preview of what to expect from Larrabee.

The article is largely about supercomputing, but for those that just want the Larrabee tidbits, they are in the paragraph next to the picture of Pat Gelsinger.

Ratner says that initial testing shows "that the Larrabee outperforms both the Nehalem and NVIDIA’s GT280 on volumetric rendering problems."

I figure that puts it about 18-24 months behind a state of the art GPU at the time it gets released next year. Not earth shattering by any means, but not a total dog either if the report is true. In fact, I'd call it downright respectable for a first attempt at a discrete graphics offering.

The bigger question in my mind is whether or not Intel will be able to close the gap with the top of the line offerings from NVIDIA and AMD/ATI.

tagskie said...

hi.. just dropping by here... have a nice day! http://kantahanan.blogspot.com/

Tonus said...

"Judging by your personal beliefs, we would both like to see a market that is free from aggresion and harm, but the root cause of our disagreement, is what qualifies as that harm."
That sounds correct.

"To me, anti-competitive behavior is one in which force or aggression is used to influence market outcomes."
I agree with that definition, but my own view is that it should be expanded. I see anti-competitive behavior as any that limits competition artificially.

The reason that I don't think exclusive contracts should be outlawed is that there are going to be circumstances where it may be the logical choice to make. Sure, I'd like to make my own choice of sodas when I buy a Big Mac. But it may not be efficient for McDonalds to try and carry all brands of soda and allow them to compete for the limited counter space (and inventory). They may be better served by doing something that reduces competition, but is probably more reasonable for all involved.

As a matter of principle, though, I'd rather that companies not enter into exclusive contracts and allow the consumer to determine who wins and loses.

And I agree that the buying public is not the consumer or "end point" for some markets. A corporation will be the end point for many products. I purchase Intel CPUs or AMD CPUs. I doubt that I'll ever purchase chip fabrication tools-- Intel or AMD or a foundry is the consumer, even though they're using that to make a product I buy.

I'd rather that the companies that provide tools and equipment to Intel and AMD and etc compete directly for their business, as opposed to signing agreements to limit choices beforehand. It's not realistic to expect the markets to work ideally, but that's what I'd like to see.

Tonus said...

Well I guess that this is timely: DoJ opens anti-trust probe... against IBM.

pointer said...

Just to throw something out and see what's people view on this.

Intel was DELL sole PC CPU supplier, is it anti-competitive.

Intel is Apple sole PC CPU supplier. Is it anti-competitive

AMD was Sun's sole PC CPU supplier, is it anti competitive

and there are so much sole supplier case here ... and of course one could judge each of them differently base on other reason but not on the single reason of being the sole supplier right?

p/s: i have to admit ... it didn't read through the long anti-competitive stuff that you all wrote ... but hey i have glanced thru :)

Khorgano said...

"I'd rather that the companies that provide tools and equipment to Intel and AMD and etc compete directly for their business, as opposed to signing agreements to limit choices beforehand. It's not realistic to expect the markets to work ideally, but that's what I'd like to see."

This is a great point that feeds directly into what I've been trying to say all along. What is an "ideal" market. This is completely subjective to all participants. Everyone has a different idea how they would like to see the market work. This is why you can't have regulations or government control because it influences and favors certain parties over others based on what some bureaucrat in office subjectively thinks the market should do. The only fair market is one in which all participants are free to pursue their ends and whatever happens, regardless of what you personally feel, is the right outcome for the market as a whole from a purely objective point of view.

An additional takeaway is that people do make mistakes, and companies do make mistakes, and if they do, then they will pay for it as the market corrects itself from malinvestment.

SPARKS said...

"EUV litho still looks sketchy..."

According to the article:

"Inside that vacuum, engineers need to steer the 13.5nm light through an array of focusing mirrors with the equivalent accuracy of a flashlight beam hitting a penny on the surface of the moon. To maintain that accuracy the mirrors need to be extremely flat, varying no more than the equivalent of a millimeter every 1,000 miles."

Just a couple of questions, please.

Can this be done? Wouldn't slight changes in temperature effect microscopic variations in the shape(s) of the mirrors and lenses?

EUV, I thought that this narrow frequency of light has a negative effect on the lens and the mirror material itself. How could this degradation of tool optics be circumvented, special coatings, special glass, etc.?

Say they found solutions to the two problems above, wouldn't it be very difficult for motorized steping/control mechanisms (for the lack of a better term and knowledge) have to built with extreme, if not impossible, tolerances?

What about the tools that build the tools? Wouldn't they need the same, if not more, precision? What grinds this stuff and how big are these mirrors and lenses?

Lastly, from the article, would these EUV tools and their relatively low wafer output be practical considering the output of more practical, proven, commercially successful, solutions?

Sounds like a red herring AND a white elephant to me.

SPARKS

Anonymous said...

Can this be done? Wouldn't slight changes in temperature effect microscopic variations in the shape(s) of the mirrors and lenses?

Yes. I suspect piezoelectronics would be used. Current lithography stages / etc routinely require accuracy down to the ~1nm level. Plus they move. While not trivial, it could be overcome.


EUV, I thought that this narrow frequency of light has a negative effect on the lens and the mirror material itself. How could this degradation of tool optics be circumvented, special coatings, special glass, etc.?


That's the big question -- EUV is nothing but a big fancy x-ray. you do need special coatings/materials that do not break down over time.

And to be specific, it's not "narrow" frequency spread which is the issue, it's the energy said photon that is.


Lastly, from the article, would these EUV tools and their relatively low wafer output be practical considering the output of more practical, proven, commercially successful, solutions?


There is one more factor: capability. Right now immersion lithography is NOT capable, ragardless of what any semiconductor company will tell you. That's why you've got double/triple/more patterning for some layers. Sure, EUV may be slow and expensive compared to an immersion tool, but what about compared to *three* immersion tools? Four? And then consider how long the processing is as each said immersion step..

What you have hit on is questioning the assumption of "smaller" is "cheaper". At some technology cycle, this will no longer be true.

SPARKS said...

"What you have hit on is questioning the assumption of "smaller" is "cheaper". At some technology cycle, this will no longer be true."




Destructive Photon energy levels, got it, thanks. (I can feel the Gamma rays aging me by the moment)

You get X number of wafers before (extremely expensive) tool refurbishsment. ("Actual milage may vary")

The cost of tooling rises exponentially, offsetting the financial rewards going to a smaller node. (Yes thats where I was going.)

Additional 30 million dollar tools, more process steps, adding varability, thereby lowering yield by default. (Murphy takes no prisoners)

It is a white elephant. (You've been say this for years.)

I'll believe in aliens when a ship lands in my backyard, and I'll believe in this process when INTC (and you) say it's possible.

Thanks, again.

SPARKS

SPARKS said...
This comment has been removed by the author.
Anonymous said...

I'll believe in aliens when a ship lands in my backyard, and I'll believe in this process when INTC (and you) say it's possible.

Oh, it's possible. No doubt about that. Hell, it's even *yield worthy* for current technologies (at some level) as we speak.

high-volume-manufacturing capable is a whole other story.

Anonymous said...

I think the limiter on the throughput right now is the power of the EUV source... the lower the power, the longer the exposure time you need (and thus lower throughput).

There still are probably issues with the mirrors, and I'm by no means an expert but those seem solvable.

I think it's going to come down to the power of the EUV source (in order to get manufacturable and cost effective throughputs) and I still think they're are problems with the resists - assuming you can get the light to the wafer you still have the whole patterning thing. :)

EUV vs immersion is also not simply a cost issue - at some point immersion will not be capable even with multi-patterning. You have alignment tolerances with each printing step so eventually stacking these errors will be an issue.

And of course there is the whole - you're still printing a feature smaller then the wavelength of the light...multi-patterning allows you to deal with pitch and density issues but you don't get smaller features with each successive printing - there is a fundamental limit to how small a feature you can get with a 193nm light source.

Anonymous said...

I think it's going to come down to the power of the EUV source (in order to get manufacturable and cost effective throughputs) and I still think they're are problems with the resists - assuming you can get the light to the wafer you still have the whole patterning thing. :)

You are correct here. The issue with patterning is, while the wavelength of EUV is small, the length scale of the resist interaction is LARGE. In "conventional" 193nm photoresists, light directly excites a molecule to a dissociative state, which can then create a chemical reaction. Stuff diffuses around, but overall the reactions remain local.

Contrast this to EUV, where the x-ray actually ejects and electron from the resist and this secondary electron creates the above excitations and reactions. This is a much longer length scale, and results in the ultimate resolution of EUV to be severely materials limited, not wavelength limited.

And of course there is the whole - you're still printing a feature smaller then the wavelength of the light...multi-patterning allows you to deal with pitch and density issues but you don't get smaller features with each successive printing - there is a fundamental limit to how small a feature you can get with a 193nm light source.

Actually, this isn't completely true. The wavelength defines the minimum pitch, not the minimum linewidth. Which is why double patterning works. You're more limited by resist effects here. However, given that you can play some games and "backfill" resist trenches with other materials, there are ways to make things really really really small.

It's just SLOW and not terribly cost effective.

Anonymous said...

Patterning have no fear those with bucks will find a way provided that scaling to the next node provides a return on investment.

If the goverment chooses instead to take the reward away than said companies won't find it too attractive to invest the billions in R&D and 10's of billions more in factories. Even goverment subsidies won't fianance a factory if their is no technology to run in it or a company capable of designing, validating and testing it.

Be careful liberal goverments, go to far barking up the anti trust tree and you could fuck up the one of the few manufacturing business still left where US is king.

By the way there are lots of little tricks one can play to pattern things much smaller than curent dimension. Sure double or triple pattern is the brute force way but there are others toos.
But again only those guranteed some reward will invest in it.

Soon even Arab oil money could dry up as soon as they realize that semiconductors isn't where the smart money should be spent.

Anonymous said...

Intel 32nm samples are out in the wild and we've seen the first reverse engineering.

Where oh where is that AMD/IBM sample of that innovative gatefirst approach.

Didn't IBM annouce their HighK metal gate the same weekend as INTEL did years ago. Last I checked INTEL has shipped more than 200 million CPUs from Atom to servers with HighK/Metal Gate. To date I still can't find one fukcing AMD or IBM chip. What the hell have those guys been doing for the last 3 years since the annoucment? Can they only publish papers, or can they actually make it?

SPARKS said...

"What the hell have those guys been doing for the last 3 years since the annoucment?"

Hey, that should be easy, especially for you process geniuses. My circuit traces are 3 inches in diameter, and I know that answer!

They haven't found a guy like Biswahoman Pani, yet.

Here's some nice pictures for your enjoyment.

http://www.netcrunch.org/tag/amd/

SPARKS

OXOX

Anonymous said...

AMD can't develop it, can't steal it. All they can do these days is to try and sue it from their competitor. They can't compete, don't got balls to take risks so what do they do. Go see a lawyer and hope to sue it out of their competitor.
Tick tock tick tock, the clock is running out you you Hector and his jolly gang of laywers. You got no 32nm got no HighK, got no metal gate.

Yeah you got a fistful of oil money but even that don't buy you no highK metal Gate.. nor can any lawyer get what you need which is technology that works.

SPARKS said...

Even though graphics architecture is far different from CPU architecture, I thought perhaps the difference in process tech may be of interest here. Further, as you all know I am no fan of the "Imitator." However, I have always been a fan of ATI. And there's the rub, at least for me, my money going to AMD and I despise NVDA. I needed a new Graphics card, however.

ATI/AMD or NVDA, NVDA or ATI/AMD, GTX 295 or HD 5870, I was absolutely MANIC about the purchase. Like a Ping-Pong ball bouncing around my empty scull for weeks, the decision was finally made. It was NVDA's CEO's 'One Hung Low' and his big mouth that made the decision for me. I went for the AMD HD 5870 for some of the reasons below.

http://www.hexus.net/content/
item.php?item=20568

http://www.theinquirer.net/inquirer/
news/1558036/nvidia-snubs-intel

http://www.hexus.net/content/
item.php?item=20633&page=1

The good news is NVDA is out of the chipset business, a kick in the crotch for sure and the new AMD card is a definite a kick in the teeth. New Egg HD 5780s ran out of stock twice (!) last week, alone. Forget the rest of the vendors. In short, these things are selling like Dunken Donuts at a PBA meeting. A good kick in the ass for Nvidia as well.

AMD's HD 5870 is the Godfather of graphics cards. Like Don Corleone, AMD made me an offer I couldn't refuse. At $379 it runs on par with the $500 GTX 295 and it absolutely trashes the GTX 280/285 which has now been rendered obsolete by comparison. Further, the AMD card is DX10/Win7 compliant for my 2Q 2010 "Gulftown" build up. A second card in Crossfire will work quite nicely in next years X58, Extreme, 6-Core behemoth.

Upon receiving the massive card, shoe horning in to my case, and comparing it to the QX9770/Zalman 9700 cooler combo, it occurred to me what a graphic representation of the difference in the INTC process. (Pun intended) If not for the manual fan override (60% and noisy) in the CCC utility, this card would run at temps the QX9770 has NEVER seen! It uses twice as much power as the CPU does at full tilt, and I'm talking 3.85 GHz for the QX, guys. The GPU runs at 850 MHz.

It made me wonder, what the hell are they doing wrong? (GURU, if you think I'm baiting you into a comparison of the two processes, you're damned straight I am!)

I haven't run everything on the new card, but I have run Crysis. The visuals are absolutely stunning, textures, trees, water, mist, blood splatter, butterfly's, all of it. All settings defaulted to HIGH.

"They can't compete, don't got balls to take risks so what do they do."

Sure, in the CPU space AMD is on the balls of their ass, no doubt, but they do have a winner in the graphics arena, top dog. That's a fact. I don't know if it will take them 200M out of the red, however. They can't hurt anyone but NVDA.

That's the real good news, for me anyway. AMD and INTC beating up bitchy NVDA. Oh, the joy.


SPARKS

Pat Gelsinger said...

ARM showcases 40% power savings when using SOI

...In case of ARM 1176 core, similar to the ARM core used in nVidia Tegra chip and Texas Instruments OMAP chips - power savings were as high as 40%. The goal of the test was comparison between 45nm SOI high-performance and 45nm bulk Low-Power process, clearly demonstrating that even a non-power optimized, high-performing SOI process can result in significant power savings and enable a large jump in performance inside the same thermal envelope. When ARM's engineers overclocked the 45nm SOI chip by 20%, the chip still saved 30% over the standard bulk silicon, meaning you could clock ARM core significantly higher and still achieve remarkable power savings.

Power saving was not the only benefit of switching to SOI, though - ARM reported a seven percent "circuit area reduction", meaning you can build even smaller chips. With Intel talking about Light Peak optical connectivity and having already demonstrated a SOI wafer for silicon laser demonstration back in 2006, we feel that you'll be starting to hear a lot of SOI-related news during 2010 and onwards. Quite the opposite of "SOI is dead" mantra we heard from Intel and other bulk silicon vendors since the beginning of 2001...


After all, AMD did made the right choice with SOI. It's only a matter of time for arrogant intel to use it in their processes just the same way they had to use AMD64 instructions in their crappy processors and later on implement the integrated memory controller after living almost for 4 years in complete denial about it.

Time will tell... ;)

Anonymous said...

So basically GF/AMD is saying they would be more than 40% behind were it not for SOI... wow their process technology is even worse than I thought.

SOI helps, but this "up to 40%" is what technical folks call HORSESHIT - provide some actual power #'s. If they've run the analysis why not just show X vs Y and stop with this "up to" qualifier crap?

For those with short term memory problems you may also recall Phenom I being "up to" 40% better than Core2 (pre-release of course)

Anonymous said...

Pat G, why oh why did you leave INTEL?

Seemed to be a stupid reason to leave just because you wouldn't be the next CEO. Better to be on the winning team and get a ring than join a losing team and QB it but have no ring. Rather shallow just to want to be top dog I think

Anonymous said...

Funny so many new armchair QBs believe ARM will give x68 a run around on netbooks.

Duh, do them said dumb fuck analysts know anything about scaling and process technology advantage. Duh do them said dumb fucks study and find out what happened to the first netbooks that were 50 bucks cheaper and ran linux. Did thme said dumb fuck analysts study and see MS track record for porting OS, let alone develop a new one. There won't be any magical new windows look feel OS for ARM. Without that ARM is broken business trying to get into netbooks. Don't matter dual core or quad core or 2 GHZ. Its the OS you dumb fucks.

No different that x86 can't compete in smart phones. Its the power and interface that matters. Us cellphone user don't give a fuck about x86 compatibility when it comes to our cell phones. iTouch is enough, damm good enough. No x86 required, deliver the power and performance and you'll be in every socket

InTheKnow said...

I'm surprised our resident troll hasn't shared this article with us. Since it accuses Intel of cheating on benchmarks, it seems right up his alley.

Though lately, he does seem determined to expose his complete and utter lack of process knowledge to one and all instead.

I believe the driver development in the report demonstrates an example of shoddy workmanship by Intel. I see no reason not to write the optimization so that it detects GPU loading and offloads some of the work onto the processor any time the load on the GPU reaches some high (pre defined) percentage. Singling out specific applications is just plain lazy in my opinion.

On the other hand, I also believe the author of the piece is putting the worst possible spin on this finding. He pushes Intel's statement regarding the state of this driver to the second page as part of a response from Intel. Specifically, Intel says "Our driver is currently in the certification process with Futuremark and we fully expect it will pass their certification as did our previous DX9 drivers."

So Intel has submitted the optimization for Futuremark's review. It isn't like they are trying to pull a fast one here. Futuremark has the optimization for evaluation, and will get to decide themselves if it is keeping with the intent of their policy or not. And who is in a better position to make that determination than the people who created the policy in the first place.

Again, what I find disappointing is that Intel does seem to be doing is a poor job of implementing the optimization.

Of course, there could just be a HUGE anti-Intel conspiracy brewing out there. ;-P

SPARKS said...

Well, if you owned INTC before the meltdown, bought INTC immediately after at, say, 13 bucks a share, and stayed with the company through the long haul, you would have done EXTREMELY well.

I particularly enjoy dividends clocking in at terrific 2.8% per quarter. Not bad, not bad at all, especially in this chickenshit market.

INTC is the blue in the blue chips.

9.4 B in sales, nearly 2B in profit, and Christmas is coming.

Well done boys, well done. God I love INTEL.

http://finance.yahoo.com/news/
Intel-Reports-Strong-bw-196357517
.html?x=0&.v=1

SPARKS

A Nonny Moose said...

^ Yep, another impressive showing by Intel, beating the street EPS by 6 cents. I just wonder if AMD will also manage to beat the analysts - consensus is -42 cents EPS, which works out to another $280M loss.

SPARKS said...

"For those with short term memory problems you may also recall Phenom I being "up to" 40% better than Core2 (pre-release of course)"

Dare I take issue with you? It was 40% better (pre-release) under "simulations", as I recall.

SPARKS

SPARKS said...

Moose, is that 10 consecutive quarterly loses? I've lost count.

SPARKS

Tonus said...

"Dare I take issue with you? It was 40% better (pre-release) under "simulations", as I recall."

I believe that it was 40% better 'over a wide range of workloads', or words to that effect.

That earnings report for Intel makes me think that "Pat Gelsinger" will be kicking himself if he cashed out his options early. :)

Anonymous said...

Best quarter on quarter in 30 years!

Gross margin up 7 points even as ASPs trend down.

Q4 margins expected above 60 points!

That my friends is the power of being first on manufacturing! Get the die shrink before everyone else enable new products that are smaller, faster, lower power. The performance/power is a big win but nothing beats gettting a 2x in density to increase your margins and drop your product costs! Intel invest and reaps the margins.

Who are the dumbfuck analysts talking about ARM taking over the netbooks with a dual core 1 GHZ+ ARM core runnig Linux. Next year INTEl will have 3rd generation ATom on 32nm, and another two years later 1/2 the cost one more time. What does an Atom cost today 50-70 bucks, in 4 years the same transistor count can be had for 15 bucks or maybe INTEL will charge 20-25 bucks for somethign with double the transistor count of todays Atom. Oh lets not forget the reduction in power too, what another 10-50x. ARM bend over and kiss your ass goodbye in netbook. I'd be pretty scared the smartphone will also be gone by then too! If intel executes on 32nm I figure they will have a pretty competitive x86 and fully pC compatible cell phone next year and I can only imagine what they will have two years hence.

Lets also not forget that ARM and its licensee's don't have the manufacturing prowess or the volumes to invest as required to get to the next node or leverage that node for return on investment for chips from high margin servers, medium margin high performance, and low margin netbooks processors.

I can't wait to see how stinky an egg AMD is going to lay when they annouce shortly. I really wonder how stupid them Arab investors are feeling, they should have invested in potato chip manufacturing instead of silcon chips.

Pat Gelsinger said...

InTheKnow Said:
Of course, there could just be a HUGE anti-Intel conspiracy brewing out there. ;-P


Does people always live in denial in this blog even though they have enough evidence in their noses?

I wonder if you guys are still being spoon-fed like babies?
If that's the case, that wouldn't definitely surprise me.

Anonymous said...

Earnings call transcript for those interested.

http://seekingalpha.com/article/166325-intel-q3-2009-earnings-call-transcript?source=yahoo&page=5

One interesting bit which I know I've mentioned many times here and on places like the INQ that say Intel will be complacent without AMD and innovation will halt:

Paul S. Otellini

I don’t think that we can grow much above the PC growth rate at the end of the day. I mean, there is -- there has been some share gain over the course of this year. We’ll see what this quarter maps out to but I would think that trend could continue this quarter. But at the end of the day, we need the volume. The PC market has to grow.


If Intel killed off AMD and absorbed the 20% market share, it would be a one time growth of ~25% (which is about 2.5years of general market growth). If Intel was to stop innovating, the market growth would slow to a halt and there would be no growth - Intel realizes this and innovation is as much about staying ahead of competitors as it is about continuing to grow the market. It is needed whether or not AMD exists and that will continue a key driver for innovation.

Oh and Atom chip (and assoictaed chip sales) crossed the $1Bil mark for the year...

InTheKnow said...

And the troll speaks on queue. I wonder if we can teach him to fetch and roll over. LOL

A Nonny Moose said...

Sparks: Moose, is that 10 consecutive quarterly loses? I've lost count.

12 in a row I believe. First loss was 4Q06 (after Core2 sales kicked K8's slow butt), although 3Q06 saw some decline from the previous year.

IIRC AMD was predicting a return to profitability, or at least break-even, in 4Q07, 2Q-3Q-4Q08, and 4Q09 (during the 1Q09 conference call).

The street sense is that AMD will continue its losing streak next quarter and all of next year, since they don't have any worthwhile mobile product (biggest gainer this past quarter and mentioned by Otellini as mainly responsible for the large gains), they have zip in the netbook market, desktop segment is declining, and when corp IT starts buying again, both workstation & server will probably lose yet more marketshare to Intel.

I hope Abu Dhabi has plenty of cash to bail out this clunker! :)

Tonus said...

"And the troll speaks on queue."

The correct term is "on cue." Unless you're making a clever turn on the phrase and implying that he is "out of order." :)

SPARKS said...

ITK, Tonus, there's a bit of psychology behind the "trolls" posts. He sees himself as a lone crusader, the single voice of opposition crying out in the wilderness, "we will not go silently into the night."

Despite the years we have watched the "Imitator" fall perilously close to bankruptcy, despite the fact they will never manufacture microprocessor again, and despite 12 consecutive quarterly losses, and without doubt more to follow, he remains in denial. Either this, or he just a fool, lingering on simply to disrupt and annoy.

Intel just reported 2B in profit for the quarter in a foundering economy, whereas AMD will report additional losses into the hundreds of millions. They have a broken business model, and product line that sells at a loss merely to be a presence in the market.

They have been relegated to a second tier company, which I suspect, is now being supported by the successes of a graphics card company.

I submit they should stay with the graphics. At least ATI has a fighting chance in that end of the market. Anything else from AMD is yesterdays technology and simply low end, hyped up garbage that falls further and further behind by each successive generation.

SPARKS

SPARKS said...
This comment has been removed by the author.
Anonymous said...

t is time again for the troll to speak right ON queue :-P

I wonder if the armchair technologists that post here can do anything anything but spew shit? Never a rebutal to my preposterous claims. People either to stupid or its the truth!

Lets look at the perfect storm that has come and gone and how Intel sailed thru to smooth rich fishing grounds.

Invest invest invest in new technology, deliever new products on new faster and smaller process node. How could they invest so much because the got marketshare, got the balls, not like some other silly companies without the backbone to do what they need. Now intel has higher performance, lower leakage and twice as much transistor density in manufacturing than ANY other manufacture. Let the margins continue to roll in. A few claim to be close but in reality none is. Today more than 200 million HighK/Metal gate CPus have been sold by INTEL, I think that amounts to something more than a thousand TRILLION transistors to some paper annoucments by others. Intel's second generation HighK/Metal Gate is now in production while all others still struggle to figure out how to do it reliably at all at the single transistor level. To make matters even more amazing they already have working 22nm chips, NOT transistors, so right on schedule we should see 22nm products roll out in 2011. No wonder ARM is running scared and trying to stop the attack at the netbooks. Better to try and divert attention of the battle front to a foreign soil than the home turf that will soon be lost.

Whats up in 2010? Everyone wants a mobile chip, guess who offers the best? Business upgrade to new generation of laptops, guess who gets the sale? Business updates to next generation server, guess who is best? Consumer wants more wifi connnected cheap netbooks, guess again who owns that. Everywhere you look where people buy guess who supplies the silicon? The after thought second source will be exactly that, pray they don't lose another 2-5% market share during 2010 rebound upgrade cycle. All they got is price lever but how low will they go and how much money will the choose to loose to fill their underutilized factory?

You think ARM got a chance in netbooks, forget it. The cellphone is going to be a fight, like I said before ARM will be on even footing in two years time, and behind in 4 years time. The advantage of 2x transistor density and 30-50% power by being a generation ahead on technology is too much for for any achictecture to make up. This isn't a Netbust versus AMD situation. Its about a 3rd generation optimized x86 core competing in stretch space. Just like x86 took over servers/risc. Game is over unless INtel fumbles the ball. Intel is like a pro football team driving toward the goal line down by 3 competing against a college defense. Unless they fumble the ball the endzone and a win is all but assured.

to be continued

Anonymous said...

How do they do that, let the troll remind everyone ( Scientia, Sharikou, Abinstein pay attention! ) Huge economies of scale, with a huge guranteed volume to come they can afford to make the bet and invest in the best, hire the best and try everything at any cost to deliver the next generation node. NODBODY else can afford this, not iBM, not TSMC, not Samsung, not a consortium. Maybe some dumbfuck stupid rich arabs might think they can throw a ton of money and get in the game. But they won't have the stomach to put 2 billion for another 3 years and than invest say another 10-20 billion for two generation of factories could. Lets see that is a total of 30 billion dollars required over the next 4 years. Even if you got the money to burn are you backing the right people? Does AMD/ATI have the expertise to do it? Big money bet, I'd think the Arabs should be rethinking what they are doing if they aren't already. The Charter deal already says they are going after TSMC market. That isn't a good sign for AMD as the foundries largest customer. The users of high end silicon are far and few. Trailing end silicon isn't a big money maker and the suppliers their are a dime a dozen. Very broken business to invest in dumb arabs!

Now I'll go to the back of the line and stay in the queu to re-remind everyone shortly on this all again!

Having_Fun said...

What a bunch load of crap. I wonder if your fingers "smells" after you've finished typing that load. :)

Only a fanboy like you would believe the flawed logic that Intel is bigger than IBM.

That's why I love coming to this crappy blog once in awhile for some comedy relief.

Keep on with the good job!!!!!!!!!!

:D

Pat Gelsinger said...

"What a bunch load of crap. I wonder if your fingers "smells" after you've finished typing that load. :)

Only a fanboy like you would believe the flawed logic that Intel is bigger than IBM."


Priceless. LOL

pointer said...
This comment has been removed by the author.
pointer said...

signed with 2 names and talk to himself? speechless :)

anyway, AMFudZoner, welcome to the blog! :)

SPARKS said...

AMD's 3rd quarter results are in, another loss. I'm still waiting for the big AMD turn around. It's been 3 years. I guess from the AMDzone perspective this is short term. When will they pay down that debt, and with what?

It any event, the numbers (again) show another "favorable impact" of some sort totaling $54M. Anyway you cut it they lost at least 128M. The table on the page below shows 244M.

'AMD Product Company non-GAAP net income (loss)" of 244M. I don't get how they turned a quarter of a billion loss into $128M loss.



http://www.streetinsider.com/Press+Releases/
AMD+Reports+Third+Quarter+Results/5020966.html

SPARKS

Khorgano said...

AMD's 3rd quarter results are in, another loss. I'm still waiting for the big AMD turn around. It's been 3 years. I guess from the AMDzone perspective this is short term. When will they pay down that debt, and with what?


Things are still down, but this is the first quarterly report in the last 3 years that actually looks promising. There is a very good chance AMD has an operating profit in Q4 '09 and they will be very close to break even on Net profit which is saying something considering the down economy.

They still have the problem of -$6.2B in retained earnings (their gross profit over the entire life of the company), but if the economy starts to pick up, they might be able to squeak by and make a profit and start paying down their debt. They aren't out of the woods yet, but I think there is a glimmer of light at the end of the tunnel. They have the looming senior note maturation problem in 3 years, but who knows where AMD or ATIC will be by then. Either way, I think they need the Arab's to make at least on more ~$1B "donation" to weather those storms.

SPARKS said...

"Things are still down, but this is the first quarterly report in the last 3 years that actually looks promising.....", et al.

I agree completely, points all.

In fact, the performance and current lead with their newly released graphics, points to a better than expected fourth quarter. I recently purchased an HD 5870 (XFX), the performance is outstanding. They are selling extremely well, competitively priced, and personally, I intend to get another.

NVDA was late to the party with their new WIN7 (DX 11) capable graphics cards giving AMD a good lead going forward, at least till Christmas.

They truly have a winner here.


SPARKS

Anonymous said...

Intel has great growth huge margin increase and what does AMD do?

Another quarter of losses. Damm maybe having lagging silicon technology is a problem. Figure late every time results in larger dies, more power, and slower products. what a fucking handicap, its like showing up to a fight with a pocketknife while Intel comes with a bat. Guess who wins Arabs.. INTEL. Keep throwing your good money into that black hole.

Let me understand how spinning off to GF will help AMD. GF will now need to make money. How will they do that? Charge for the silicon at a profit. Ouch.. that means AMD is going to have to PAY. I doubt GF other customers will want to be subsidizing AMD. AMD will have higher costs.. How again are they going to make money?

Of course right now its all hidden in the billions the dumb Arabs are pouring in, but in the end the silicon must cost AMD more. How can that be a good thing. They can't make money when times are good how will they continue to make money thru the down cycle. Plus GF must now spend money developing processed and dedicating capacity to other customers who don't want high performance technology. More processes means more money spent and less for AMD... losing proposition all around.

Good bye AMD

Anonymous said...

Hey Patty, why are you posting here. You should be concentrating on your new job at EMC. After all you do want to be CEO at that company don't you?

Pat Gelsinger said...

Hey Patty, why are you posting here. You should be concentrating on your new job at EMC. After all you do want to be CEO at that company don't you?

Things are going well at EMC. The larrabee fiasco (and other personal reasons) were the main factors why I quit from intel.

:)

SPARKS said...

ITK, G, I found a nice article (a bit beyond me, I might add) that may interest you both. (If you haven't read it already).

In any case, from the nice pictures, it seem that INTC has really tightened up it's act. Pitches have been reduced to 112 nM and drive currents have been increased 22%. The process, when compared to 45nM, is cleaner and the structures seem more refined and symmetrical. (Better lenses, masks, etc?)

I would like to know what is "increasing the porosity of the Hi=K stack". Does this mean electrons move more easily through the gate during the 'on state', therefore, the the 22% increase?

For your pleasure.



http://www.semiconductor.net/blog/
Chipworks_Inside_Angle/
23620-Intel_s_32_nm_Clarkdale_
Shows_Many_Changes.php

I'm really looking foward to my 2010 buildup, and reporting it here.


SPARKS

SPARKS said...

Intel's warm-up to the enthusiast community (bless them) is very obvious. There are i9's up and running in the wild. I'm sure this particular site is giving the new and improved Nehalem a thorough going over.

I think (hope) that INTC will pull a fast one, lift the NDA (open the flood gates), and release the XE chip before the year is out. They did it with QX9770, if you recall.

Merry Christmas, Sparks!

Naturally, they won't be hot sellers, and they will not be cheap. (Ask me if I G.A.F.) 22% over i975, tweaks galore, new instruction sets, and I want one.

Incidentally, I've read some F. U. D. that a "special X58" motherboard tweak or older boards will not run the new baddie, which is why I've resisted the purchase of a new hardware setup. (Sneaking in new hardware every few weeks or so keeps everything below the wifey radar.)

Good things do come in small packages, that's for sure.

I'm not so sure about the ThermalTake LEVEL 10, however. That will be about as easy as hiding a Nuclear Fast Attack Sub in the backyard.

SPARKS

http://www.pcgameshardware.com/aid,695432/
Core-i9-Intels-6-core-desktop-CPU-pictured/News/

SPARKS said...

And you think I'm nuts? Check this out!

http://www.million-dollar-pc.com/

HOO YA!

SPARKS

InTheKnow said...

Sparks, I had read the article and posted a link and comment earlier. In any event, you misread the article. It said:

If we look above the transistors, Intel has been steadily increasing the porosity of the low-k dielectric in the metallization stack; even though there’s been no announcement confirming 2nd-generation low-k, it looks similar to the 2nd-gen material we have seen in other chips

So it is the low-k material used to isolate the metal traces put down in the back end of the process that is affected. It isn't a direct effect on the transistor, though there is obviously an effect on the circuit.

SPARKS said...

"So it is the low-k material used to isolate the metal traces put down in the back end of the process that is affected."

Thanks ITK.

SPARKS

Anonymous said...

Intel has new lower K in production. I recalled IBM made lots of noise about airgaps. Where is that these days. Is that another great invention like Silk and bi axle strain from them silicon gods, LOL

Anonymous said...

Is that another great invention like Silk and bi axle strain from them silicon gods, LOL

"Bi axle" is that like what they have on semis?

It's SiLK and biaxial strain. If you are going to mock, at least do it intelligently.

SPARKS said...

I guessing here, but I think AMD is going back to it's old formula of holding on to market share at all cost. They are cutting prices on their entire lineup. I think this may be in response to INTC's introduction of the i5, i7 series.

http://www.hexus.net/content/item.php?item=20798

Additionally, they've come out with very inexpensive quads that are being called the an INTC killer bellow $90, assuming their not including ATOM in this analysis. I have no experience in calculating the cost per wafer/die count. However, logic tells me that the margins must be razor thin with die this size.

http://www.anandtech.com/
cpuchipsets/showdoc.aspx?i=3663


As an added bonus to their bottom line, their recently introduced Radeon 5xxx is selling like mad. According to Dirk Meyer, sales in the graphics division totaled 300M. I suspect this number will increase substantially in the forth quarter. Shortages are abound and they caught NVDA completely flat footed here.

http://www.xbitlabs.com

In any event, I think AMD is looking somewhat positive after the train wreak of the past 3 years.

Man, are they survivors, or what? Ya gotta give 'em credit, they are a tenacious lot.

Sorry LEX.

SPARKS

Anonymous said...

Sparks - what I think what has helped them on the CPU pricing front is the transition to 45nm. Despite the loons and 'technologists' at CANzone, the transition to 45nm was not as fast or as 'world record' as forecasted. If I recall correctly in Q2 AMD still had sold more 65nm chips than 45nm and much if not virtually all of the mobile area, (which is the fastest growing segment and more important than desktop from a P/E perspective) is still 65nm.

As things migrate more to 45nm AMD has 2 options - pass on the cost benefit as price cuts (and hope to get share) or beef up your earnings and try to turn a profit. With the CPU market expected to pick up in 2010, not sure why AMD isn't holding firm on price and looking to get into the black or maybe even make some money and invest it back in AMD or GF.

I think you are correct with the market share fixation... who knows maybe they go back old school and try to get a larger chunk of the Dell business - that worked well the first time!

I do find the business strategy curious - slash prices in a declining market segment and gain share of an increasingly smaller piece of the pie that they are already fairly competitive in....

If the CPU market grows, the price sensitivity (short term) will be less critical - if nothing else Intel will not be able to flood 1156 chips instantly - it will take some time... don't know why AMD would try to preempt it before things ramp up and they have true competition from it. It's not like slashing prices is going to crush the 1156 uptake... just seems like they are cutting prices before they have to from a business perspective (again) to try to get some sort of PR price/watt platformance argument...

SPARKS said...

"I do find the business strategy curious - slash prices in a declining market segment and gain share of an increasingly smaller piece of the pie that they are already fairly competitive in...."


Excellent, and I didn't factor in the fastest growing market segment as you did, mobile. Ottelini himself made this quite clear recently. Unfortunately for AMD, their presence in this area is very weak, especially with power consumption.

I thought the preemptive price cutting was a bit early, too. Although I didn't elaborate on the possibilities the way you did. Why they did it before LGA 1156 got serious traction in the market is anybody's guess. Maybe, AMD is thinking they may hold on to buyers with the older established AMD platform before buyers went forward to the newer more expensive Intel platform; fighting for dollars or holding on to existing AMD customers, perhaps? Who knows?

In the final analysis, it seems keeping up appearances with bottom line dollars, and market share, is paramount. Value and performance per dollar is one way of being competitive, even at a loss.

One thing for sure, they are consistent.

SPARKS

Anonymous said...

45nm helpings AMD a lot as it cuts die size, the power of Moore's law. The sad part is they are only starting and I'm sure yield is still got a ways to go. INTEL on the other hand has a year plus almost two years on 45nm and at mature high yields and low cost pumping atoms, i7, and i5s out. They got record margins this quarter and can slash prices big time if they want to engage AMD in a price war. Remember they are forecasting almost 60% GM next quarrter.

And they got 32nm coming with even smaller, faster and lower power products.

Tick Tock Tick Tock, a relentless march. Sorry even Arab oil money ain't going to help AMD.

INTEL is already focused on their next competitor to conquer ARM, AMD is a non issue.

SPARKS said...

"The introduction of Intel's LGA1156-based chips has further strangled AMD's attempts to charge a higher price for its AM3 parts. Indeed, now, AMD's fastest desktop CPU etails at the same price as Intel's slowest Core i5."

This is Hexus' (unabashed) opinion of the AMD preemptive price drop before the full ramp of i5. Our original speculations seem to be inline with the price cuts. INTC doesn't have to drop prices, they just release new, cheaper, more powerful chips.

I think LEX may be right this time. INTC just keeps tic-tocking away relentlessly keeping AMD on the back of their heels.

http://www.hexus.net/content/
item.php?item=20798


SPARKS

SPARKS said...

Now this is a damned shame for AMD/ATI. Although the source of this report is questionable, I'm inclined the believe it given the WEEK long unavailability of the high end HD 5780 graphics cards. I was lucky to get one 2 weeks back and am happy to say it's the finest card out there.

I speculated that the shortage was due to high demand. Fuddie is reporting this and AMD underestimating the cards potential by not ordering enough wafers to build the cards. This is absolutely incredible. They finally get a lead in SOMETHING and they blow a great opportunity by underestimating it's sales/market potential.

As listed on NEW EGG, all HD 5780 (7 companies) will not be available until 10/31/09. Since it's release, that's nearly 3 weeks of sales shot!


AMD spends 5.2B on a graphics company and they underestimated the value of their purchase. STUPID! I'm marking this one down on my list as another AMD blunder.

http://www.fudzilla.com/content/view/16095/1/

SPARKS

Anonymous said...

Sparks this is one of the issues with foundries and will be an issue going forward on the CPU side probably as well...

A lot of times there is some upside capacity (or "burst" capacity) where you can run full throttle a little above capacity plan to make up for shortfalls. When you are ordering from foundries and forecasting capacities in advance, a foundry may not have the upside to support short term bursts (or they may charge more for it).

I imagine given the current volume AMD is of GF business they probably have some flexibility but as time goes on and they become less of a player on a % basis it may be hard to lean on them during high demand.

It's just another one of those things that adds to complexity in having a foundry (as opposed to owning capacity). If you own the fab - you might go overtime, pull-in tools early from equipment suppliers, pay that slacker Sparks extra to wire the tool up ahead of schedule (pullin install/qual timelines for capacity coming online), etc... AMD now effectively has no real say in this (though I suspect they will have some say for a little while with GF)

Anonymous said...

Intel Clarksdale 670 @4.7GHz (on air by the way)

http://www.hexus.net/content/item.php?item=20842

Pretty impressive stuff when considering it's on air and the Voltage is 1.42V (high but not obscene). This is a dual core chip (with integrated graphics) that operates @3.2GHz stock.

Anonymous said...

Oops...

http://www.eetimes.com/news/semi/showArticle.jhtml;jsessionid=WKVNEG4GNIFRFQE1GHRSKH4ATMY32JVN?articleID=220900808

Ex-AMD CEO Ruiz reportedly linked to insider-trading case

I also love how everyone and their brother calls GF a spinoff, yet somehow it is considered an AMD subsdiary and not a violation of the x86 license (and managed to sail through US regulatory approvals without so much as a second glance)

Tonus said...

That is an interesting defense being offered in the article- that there was no wrongdoing on Ruiz' part because everyone knew that AMD was in a death spiral and had to spin off manufacturing in order to survive.

pointer said...

Blogger Tonus said...

That is an interesting defense being offered in the article- that there was no wrongdoing on Ruiz' part because everyone knew that AMD was in a death spiral and had to spin off manufacturing in order to survive.


May be you can tell the author about this:

Similarly, when country ABC and country XYZ are at war, there is nothing wrong for country ABC's president to tell someone in country XYZ that at what time, which target, that country ABC would launch its missiles at, ... since they are at war anyway and thus the missile launch should be expected ... :)

Anonymous said...

Biggest joke or should I say what a joker.

Hector Ruiz the big AMDer who whines about his competition cheating is now caught being a cheat and violating the law. Sure he'll have some sorry ass excuse. But come on is the guy a retard or what. You share the impending spinoff of your fucked up company to an outsider. This isn't some dumb or greedy middle line guy. Its supposidily a quality leader and CEO. There are two kinds of people on the top honest and cheats. Nobody gets to the top and is this stupid.

But maybe again he really is that stupid look what he did to AMD.

So whats the opinion, stupid dumb fuck or a cheat?

Anonymous said...

So whats the opinion, stupid dumb fuck or a cheat?

I'd say none of the above. He could buy AMD shares directly (of course this would be reported as he is a senior exec) if he wanted to profit off of it, so unless he tipped this info and then invested in the fund in question to profit indirectly (which should be fairly easy), my guess is he probably didn't profit from this. It's also not like AMD shares were 50%of this fund so even a large run up in the AMD shares has less of an impact on the overall fund performance.

I think this kind of info also gets passed among many upper execs. Usually it is done through runors or hypotheticals or what if's and more often than not it's about talking up your company or trying to exchange info to get some other info that may not be public (usually to gain an advantage on strategic and planning areas, not for indvidual profit).

Its wrong and the guy running the fund is completely screwed (he's done this with more than just Ruiz) - but I don't think much will happen with Ruiz. It's not right, but it's not a huge deal for him - my guess he gets a slap on the wrist if anything.

Anonymous said...

Sorry - I meant to say it should be fairly easy to figure out if he invested in the fund.

SPARKS said...

"So whats the opinion, stupid dumb fuck or a cheat?"

Going into 2Q 2006 he was a frigg'n genius and the darling of Wall Street.

Due to this new found brilliance with the Street and his board, they thought this guy was the new God of the semiconductor world.

He sold the banks, he sold Wall Street, he sold the board, he even sold the shareholders. He was flying high, and subsequently, he sold them the ATI deal that was going platform INTC into oblivion.

They bought the whole enchilada, especially the Barcelona promise. In fact, at the time, they applauded the deal.

He's no cheat or a dumb fuck. He bit off way more than AMD could chew and Barcelona fell on it's ass. The truly dumb fucks are the greedy bastards who thought this "vision" was going to work.

Today, Wrecktor is a corporate joke. After losing several billion dollars, AMD's manufacturing arm, and 63% of the company, he was lucky to get out before AMD went belly up.

Insider trading? Na, he gets a pass. There's no personal gain here, just a ray of sunshine he shine up his "partners" asses.

He now heads up Global Foundries, and he's selling his vision to the ARABS!

So who's the stupid fucks (again)?


The guy is BRILLIANT!

SPARKS

SPARKS said...

Oh, by the way, speaking of "vision" and GF:

"Customers of Taiwan Semiconductor Manufacturing Company (TSMC) continued to rather aggressively migrate to newer process technology in the third quarter of 2009, the world’s largest contract maker of semiconductors revealed. It is noteworthy that TSMC’s revenue from 40nm wafers quadrupled in the third quarter and is now around 4%."


The article is at X-BIT Labs.

SPARKS

Anonymous said...

4% on leading edge, wow what a technology company. Wonder when TSMC will get to 25% of revenue.

Anonymous said...

4% on leading edge, wow what a technology company. Wonder when TSMC will get to 25% of revenue.

SPARKS said...

OK, what are "chamber matching issues?"

Now we know why AMD has had problems building HD 5870s. Lesser cards are available presumably from less than optimal die.

INQ is reporting a 40% drop in yield on 40nM at TSMC. Ouch.

http://www.theinquirer.net/inquirer
/news/1560346/problems-40nm

SPARKS

Anonymous said...

Many process steps are processed via single wafer - a wafer goes into a chamber and has something deposited, etched, implanted, cleaned in some case, polished, etc... Since most process steps take some time the only way to make things cost effective (for some steps) is to group a bunch of chambers onto a single tool and do a bunch of single wafer processing in parallel.

This of course opens up another source of variation as chambers might have minor differences and lead to "chamber matching issues" - films out of one chamber may end up being different then the other chambers.

Basically this is similar to issues with tool to tool matching (something that Intel's copy exactly strives to eliminated), except it is occurring on a chamber level instead of a tool level.

As an example, if you have a 3 chamber tool with one chamber being "off" (temperature, maybe a physical dimension impacting gas flows... or any # of 100's of variables that can be off) suddenly 1 out of every 3 wafers looks or behaves differently and might behave differently in subsequent processing steps which could lead to all sorts of problems, including yield issues.

Anonymous said...

What a crock of shit, chamber matching is the cost for 40% yield? So one chamber has good yield say greater then 80% and the other is zero yielding so average is 40%. Sorry that don't fly. You got a bad chamber you shut it down and fix it. Take what a week or so. In the meantime you deliver good yields from the other chambers.

But given the low volume on TSMC likely they only got one tool for each step and a couple chambers. I don't buy the chamber matching. Better PR to say ramp is going slow as you don't have all your capacity versus lying about a chamber problem. They yield problem is much bigger then they are admiting and they need to say something.

GF won't be any better

Anonymous said...

...additonally

To fix this you can do a few things - you can change the process in that specific chamber to account for the differences (which can be a nightmare)
- you can adjust downstream processes to account for this (feed forward control/APC/whatever you want to call it)
- You can design the processes with a wide window so the differences don't matter (a lot of times you can't do this without sacrificing performance or something else)
- You can fix and identify the problem (new chamber, new parts in the chamber, etc)

While it might sound nice to fix it on the fly (change the recipe in the chamber to account for the difference, or tell the downstream module to account for everyt 3 or 4th wafer being different it is usually not practical... even if you have some fancy "APC" system with all sorts of bells and whistles. The problem is that in addition to chamber to chamber matching issues, you also have tool to tool matching issues, and a lot of chambers you have an aging effect where the process will drift (usually predictably) over time. While in tehory you can account for all of these huge variables it gets to be a tough juggling issue.

The goal should be to match the chambers perfectly - sometimes though you just can't find the problem....

Anonymous said...

Looks like TSMC yields are still struggling at the leading edge 40nm. Frankly I'm not suprised at all. The golden days of foundries matching IDM are over. The sweet spot was in the 240nm to 90nm nodes. Then any monkey with money could buy capable tools from the tool vendors and use their baked process and create a tasty cake. Below 65nm the game changed, complext litho, subtle process interactions with polish, etch and deposition make developing a process without close and early learning on a real product results in FAILURE. How do I define failure, low yields. Look at TSMC, they are a world class act. They got good tools, good engineers, lots of money and lots of designs to work. But at 65nm and lower customer volume and adoption fell to just a few products with low volumes. Without products and volumes you simply can NOT get the processing learning to adjust your process to get high yield. The problem is magnifying as you get to 45, 40 or smaller nodes as more complext patterning interact with etch, polish and deposition. Without partning with designs up front during the defintion of design rules is a gurantee for low yield issues. Test chips of SRAMs are not enough. You need a product running hundreds of wafers a day to shake out tool and process shifts. Then the fab and designers often need to make quick corrections. Tell me what foundry or deisng house can afford this kind of investment? Only two come to mind Samsung and Samsung. Sorry GF even with Arab oil cash can't afford it unless they are willing to subsidize their customers as well as AMD.

I think this clearly indicates the problem with the foundry model. its a chicken or egg issue. Leading edge technologies push tools and their process capability. They have complex layout and density interactions that can't often be predicted till you run volume. Where will GF get that additional business and who will pay? Where will TSMC get that business and who will pay. There are really only CPUs and GPUs that can generate the high ASP and levarge these advance nodes. Cheap Atom, ARMs must draft behind this learning. Intel has a huge learning as they make money on high end CPUs for Atom to draft. GPUs don't have enough volume to drive and are now segmented so TSMC/GF can't share between ATI and Nvidia. Summary, the foundry model is broken below 65nm.

TSMC is scrambling as they know they have a busted model for leading edge silicon. chamber matching my ass. The problem is far larger, you shut down the bad chambers or chamber and continue on the good ones with fewer wafers. No they got a bigger problem then a bad chamber.

Anonymous said...

What a crock of shit, chamber matching is the cost for 40% yield? So one chamber has good yield say greater then 80% and the other is zero yielding so average is 40%. Sorry that don't fly. You got a bad chamber you shut it down and fix it. Take what a week or so. In the meantime you deliver good yields from the other chambers.

A - you can't assume the overall yields are 100% or even 80%... if TSMC is saying 40% yield - maybe that's a drop from a 60% baseline (which would be a 1 chamber out of 3 issue on a critical module?)

B - While some problems may take a week to fix, some may not - maybe you need a whole new chamber and you are talking 1-2 quarters on some tools to do this. Not to mention the impact to the working chambers while you are deinstalling/installing a new chamber. This gets into the dicey situation of managers trying to decide whether they can figure out a way to live with it by making adjustments on the fly or whether they are willing to take the capacity hit and fix the root cause of the issue. WHile it's easy to say just replace the chamber, that not only takes that one chamber down but will also impact the other chambers that are good to some degree.

The problem is you don't always see these issues in development at low volumes - I suspect TSMC thought they had a fairly stable process and then when running it in volume may have found some subtle matching issues (just a guess).

SPARKS said...

"As an example, if you have a 3 chamber tool with one chamber being "off" (temperature, maybe a physical dimension impacting gas flows... or any # of 100's of variables that can be off) suddenly 1 out of every 3 wafers looks or behaves differently and might behave differently in subsequent processing steps which could lead to all sorts of problems, including yield issues."


......and/plus.......


"But given the low volume on TSMC likely they only got one tool for each step and a couple chambers. I don't buy the chamber matching. Better PR to say ramp is going slow as you don't have all your capacity versus lying about a chamber problem. They yield problem is much bigger then they are admiting and they need to say something."

.......equals.......


"I suspect TSMC thought they had a fairly stable process and then when running it in volume may have found some subtle matching issues (just a guess)."


First, I love you guys.

Second, I thought wafers were delivered to each tool in "Pods", where each wafer was processed individually, like my 300+ CD player. I didn't realize you could dump the whole lot, or a portion thereof, all at once, like my wife loading the dishwasher. (Are the small plates at the top getting really cleaned?) After all, some dishwashers are better than others, soap and loading is important, and water temperature/hardeness/mineral content could be factor. God forbid the glasses come out spotted!

Third, It just goes to show that INTC has got some serious chops when in come to HV production. Kickass chips for the masses, obviously, that's where the money is.

However, I'll take my very expensive, nearly perfect, XE chips, well done, thank you very much. Considering the number of variables at each tool, the number of steps, and the number of tools, I want my badboys CERTIFIED bad ass. There's more to this stuff than an unlocked multiplier!

And they tell me I'm wasting my money. You guys just keep enabling ole' Sparks to spend his money in the right places.

Thanks all.


SPARKS

InTheKnow said...

I thought wafers were delivered to each tool in "Pods", where each wafer was processed individually, like my 300+ CD player. I didn't realize you could dump the whole lot, or a portion thereof, all at once, like my wife loading the dishwasher.

Sparks, what you are talking about here is a "cluster" tool. Manufacturers will group several chambers together to reduce cost and the foot print of the tool (shared robots, vacuum system, etc.). Remember, floor space is money in the fab so anything done to reduce a tool's footprint gives the tool manufacturer a selling point vs the competition.

Some cluster tools have chambers dedicated to specific process steps and others have multiple chambers running the same process step in multiple chambers. In a general sense a cluster tool with multiple chambers running the same process step will process wafers like this.

The FOUP (Front Opening Unified Pod) docks to the tool. The door is opened and the wafers are moved into the tool individually. The tool will put one wafer into a chamber and get another wafer processing in a second chamber while the first is still processing. If you have a third chamber, then it will get the third one started, and so forth. Once the first wafer is finished, it is returned to the FOUP and another wafer is pulled out of the FOUP and moved into the first chamber.

Lather, rinse, repeat.

Tools will have multiple load ports so you can dock more than one FOUP and don't have to wait for the next FOUP to be delivered before you can start processing.

There are a handful of tools that do process multiple FOUPs of wafer at the same time. These are the diffusion furnaces and some of the wet benches. Though there are newer wet benches on the market now that have multiple "chambers" and run single wafer processing.

Batch processing (running multiple FOUPs at once) is a high risk activity and if you look at semiconductor processing history, you will find a steady move towards single wafer processing. If something goes wrong in a single wafer tool, you lose a wafer. If something goes wrong in a batch tool you can lose 100+ wafers. You can see why single wafer processing is considered desirable.

My personal opinion is that the diffusion tools will never be able to move away from batch processing. The process steps take too long (30 minutes is not unheard of). The slick new ALD (Atomic Layer Deposition) processes tend to have even longer times (on the order of an hour or more) than the traditional diffusion furnaces. So I think that batch processing to some degree is here to stay.

But there are a lot of smart people out there trying to find a way around batch processing, so they may just prove me wrong.

InTheKnow said...

While some problems may take a week to fix, some may not - maybe you need a whole new chamber and you are talking 1-2 quarters on some tools to do this.

I agree completely. Many problems take far more than a week to resolve.

I've been involved in, and seen others involved in, dealing with problems that required far more than a week to identify, far less fix the source of the problem. Just because it is obvious that there is a problem doesn't mean you know what is causing it, or how to fix it. Until you have identified the source of the problem, you don't know what to fix.

Once you understand what is causing the problem you have to propose a solution, get something new manufactured if it is a mechanical issue (more often than not in my experience) and test the new part to validate it works. Each of those steps can take far more than a week all by themselves.

There is also the issue of where the problem occurs vs where you can detect it. If it is very early in the process and you need electrical test data to validate the fix you will be waiting weeks to see the effects of any change you make. If you need end of line data, it will be even longer. Worst case would be a reliability issue. Then you need to finish the wafer, get through packaging and spend several weeks doing reliability testing to determine if the issue is fixed.

SPARKS said...

"The tool will put one wafer into a chamber and get another wafer processing in a second chamber while the first is still processing. If you have a third chamber, then it will get the third one started, and so forth."

"Batch processing (running multiple FOUPs at once) is a high risk activity"


F**ken a Bubba!

Got it. Multiple chambers on a single tool, sequentially performing separate tasks common to that tool. All things being equal, (although they never are) in theory, this should cut operation time by x number of chambers for that particular part of the process. It reminds me of the old axiom:

Beauty is only skin deep, but ugly goes to the bone.

The problem may not be with chambers, but with the constancy of the tools' mechanism common to each specific operation. If so, can that tool adjust perimeters on the fly for each FOUP? Can't they pick off a sample immediately after the operation to see if they screwed the pooch, without compromising the whole lot to failure?

Or, will it be this?:

http://www.fudzilla.com/content/view/16217/1/


SPARKS

Anonymous said...

Can't they pick off a sample immediately after the operation to see if they screwed the pooch, without compromising the whole lot to failure?

If you had something to measure, yes. Single wafers (from lots) generally do get some sort of metrology step afterwards -- but only if you can actually measure what you are looking for.

If you are doing a deposition, you *could* measure what you have deposited, however, how would you do that without disturbing/destroying the thing you just made? (there are ways, but then, they may not be "fast")

Or, maybe you *can't* measure anything (say, amount of dopant implanted). The only way you can check this is by having active electrical structures to test, which as ITK was saying, may be very far down the line from the step you are processing.

And I will tell you that once you get into HVM, the *only* bit of data that makes a damn bit of difference is End-Of-Line-Yield. Nobody is going to be making any sort of change without final silicon validation.

Now, if you are talking about processing wafers differently per chamber, then you're into AMD's much touted "APM" territory. As I've said before: good luck with that. Fix the damn chamber. If you can....

Anonymous said...

FUD FUD FUD

Its obivous TSMC has a half baked process. it was checed out on some test structure likely a small SRAM and they also had the stuctures needed to define a spice models for their customers and they started ramping. Suprise suprise the process is pretty complicated at 40nm. There is subtle wafer to wafer, with in wafer and maybe even PM to PM senstivity. You don't get that with a test chip, you don't get that running a few wafers. How do you learn? By running a lot of real product. Problem is chicken and egg for the foundry. Few need that much leading edge, even fewere can pay.

Result the leading edge ramp at foudries in the past few generation have been lots of low yielding suprises. The only people who can do it? Big IDM or those with volume..

GF, IBM, TSMC don't got it anymore.

Anonymous said...

Batch process is bad and good. Good when its going that the whole run is going to be good. Nice vetical furnace with same footprint as single wafer tool doing 100-200 wafers, not bad.

Bad in that you got a lot of wafers going if the shit is off a whole lot of wafers are screwed. You also got to deal with top, middle and bottom of the load thing. But them tool manufactures are getting pretty good with cross load and cross wafer uniformity.

Single wafer tools are a challenge as matching them is critical. If any little thing is off on one then you got a missmatch. Sometimes it can be very very subtle.

Good manufactures who have a robust process can tolerate some tool to tool variability. Those that make execuses about bad chambers or such are simply saying I don't have a good proces with good design rules to tolerate what my tool capability is.

Last I checked the big DRAM manufactures never blame their tools for losing money... only marginal players and bad engineers blame they tools. Maybe they should look in the mirror to see where the real problem is.

SPARKS said...

Ya know, I was thinking about the "chicken and the egg" thing. I thought it may be foolish to say. However, when one who has faced the challenges of:

"Single wafer tools are a challenge as matching them is critical. If any little thing is off on one then you got a missmatch. Sometimes it can be very very subtle."

and........

"Good manufactures who have a robust process can tolerate some tool to tool variability."

He must know something VERY special, which is why I WANT MY 6-CORE GULFTOWN XE, AND I WANT IT NOW!

It seems INTC's problem is releasing 32nM too early thereby cannabilizing 45nM sales. My problem is squeezing in the (i9?) thereby cannabilizing my St. Maarten, St. Thomas cruise in May.

Boy, both INTC and I sure have our share of problems. Life is a bitch.

SPARKS

Anonymous said...

It seems INTC's problem is releasing 32nM too early thereby cannabilizing 45nM sales.

There is no such thing. 32nm is, by definition, cheaper to make. There is never a reason to delay.

You may be writing off more last-generation technology, but you more than make up for that with cost savings.

InTheKnow said...

There is no such thing. 32nm is, by definition, cheaper to make. There is never a reason to delay.

This is an oversimplification. A new generation requires new tooling and adds process steps to the flow. Those costs eat into the cost savings from moving to a new node. At some point it will take more than the two year cycle to pay back the cost of the last node.

When that point is reached there will be two options. First you can slow down the rate of process shrinks and admit Moore's law has finally broken down. Or second you can go to 450mm wafers. That will allow you to put enough die on the chip to offset the increased processing and equipment costs.

Anonymous said...

This is an oversimplification.

Actually, I am specifically NOT oversimplifying it. =)

Given that 32nm is "ready" now, one has to make the assumption that the cost in under control.

You are absolutely right -- there is no guarantee that a smaller node will be less expensive. But that's something you learn and mitigate in early development. Not when you are ramping.

If the cost is too high, you either don't progress OR have a really good design reason for going forward

My point is, if a process is HVM, it is, by definition and necessity, cost-effective.

InTheKnow said...


My point is, if a process is HVM, it is, by definition and necessity, cost-effective.


Fair enough. I took the earlier comment to be more general in nature (though you did state 32nm specifically). And as you say, there is no guarantee the next node will be cost effective.

Anonymous said...

To me scaling ends (or I should say slows dramatically) if 450mm doesn't happen. I know people talk about how scaling lowers cost, but it generally doesn't (bear with me).

The problem is scaling is generally 'given back' via more cache, more cores, more logic circuitry (64 bit, hyperthreading, new security technologies, etc.) The die size over time is really just being held constant or decreasing very slowly because generally speaking designers aren't working under a constraint of fixed transistor counts (where you would see the theoretical shrink benefits the technology should give you) Basically it seems designers are allowed to jack up transistor counts and keep the die size similar (or slightly declining) to previous generations.

The problem is the cost per unit area of silicon is going up on each node - additional metal layers, more expensive litho, new technolgies (high K, strain, etc,,). I think it is going up about 10-15% per generation - so if you are merely holding die size relatively constant on new chips your costs are at best flat or maybe slightly going up.

So you either need to slow down the transistor count scaling so you can conitnue to shrink die size on new products or you need another way to justify the higher finished wafer costs - and the price of CPU's ain't exactly going up over time.

The only way to do this (I think) is to change wafer size. 300mm was estimated at ~50% cost reduction (you get 2 -2.25X area scaling, but you also have increased tooling & fab costs that eats into the theoretical benefit you would get). Rounding egregiously and assuming the 50% # is correct that means that wafer scaling essentially buys back the added cost of ~5 generations of scaling.

Anonymous said...

---- continued ----

On 200mm Intel (I think) did 852, 854, 856, 858 and then made the 300mm transition on (or after in most people's via 860/1260 (.13um)... that's basically 5 generations of scaling... which conveniently (or conicidentally) matches up with my 10% per generation cost vs 50% cost gain on the new wafer size made up #'s

On 300mm you started at low volume on 1260 (.13um tech) which most people consider a pilot/verify the 300mm tooling and now has gone 1262 (.09um), 1264 (65m), 1266 (45nm), 1268 (32nm) and eventually 1270 (22nm)... that again puts 22nm at about the 5th (or 6th) generation of 300mm.

This is why you hear Intel and TSMC and Samsung pushing for 22nm 450mm tooling (which won't happen) and really trying for 15nm (which also may not happen). People keep saying it is way too soon - but it's about the same length of time historically between wafer size changes, it is similar to when folks went 200mm to 300mm (don't recall how many generations were between 150mm and 200mm)

Assuming each gen is stacking ~10% each generation on finished wafer costs - you are talking about 50-60% (or more if you actually take the time to compound the #'s) while the actual chip size has not exacty halved durng that time.

While the density has shrunk and cell sizes have compensated for that as I said before transistor counts have gone up via cache, # of cores, core complexity and soon integrated functions like graphics, etc. I bet, but am too lazy to look up, that the die size of a .13um product is not that much larger than a 32nm product (non-Atom) and I doubt it has shrunk to accomodate the 50-60%+ production cost in that time period.

http://www.ieee.org/portal/site/sscs/menuitem.f07ee9e3b2a01d06bb9305765bac26c8/index.jsp?&pName=sscs_level1_article&path=sscs/07Winter&file=Kumar.xml&xsl=article.xsl;jsessionid=JlmLKnJpwSGQYBFTwXJ9t8ZTgk5nxQGPcthMj6tlvHprhb5n58hr!50297216!-575106651

This is a long article but it has a graph showing about a 39% increase in transistors per year (for some normalize chip at that time) That's in the neighborhood of 80+% for the 2 year cycle technology nodes scale on and we all know by now you don't get double density (close probably to ~80%).... so that's basically a zero sum game, until you consider that each new tech is also going up at that 10% rate which now means you are potentially losing on each generation.... unless you can do a wafer size transistion every 5 or 6 generation to compensate.

Without an increase in wafer size, you either have to limit transistor counts more rigidly for designers, such that die size truly does scales with the cost for each generation or costs will start to go up...

Anonymous said...

Hector Ruiz quits Global Foundries.

http://www.theinquirer.net/inquirer/news/1560558/hector-ruiz-quits-global-foundries

He officially leaves Jan4'10 (and is on leave of absence until then apparently), which is an amazingly quick transition for a CEO change (these transitions under normal conditions are often planned a year, if not years in advance).

I don't like the leave of absence going directly into retirement - generally leave of absences are unpaid and if doing it tight before leaving it is a way to extend decisions on any stock options (once you leave a company the clock starts ticking on your stock options and this artificially gives him a longer window). It also may help him on taxes if some money is paid out in 2010.... if the guy is leaving why play games like this to benefit someone who clearly has done some wrong (regardless of whether he is prosecuted, which I don't think he will be)

Anonymous said...

Well the spin is staring at UAEzone... "well Ruiz had previously planned to step down Jan'10 anyway"

Omitted part? The idiot, ummm CANzone poster, was basing this on a report that Ruiz previously submitted his resignation in Sept, buthe neglected to mention it was FOR UNSPECIFIED REASONS! You think perhaps he may have seen the writing on the wall - the guy he's been involved with has known to be under investigation for some time and while the Ruiniz part just came out, Wrector probably knew this was coming 2 months ago when others were getting ensnared.

Gotta love the spin of blind zealots... it was planned anyway... I needed a good laugh today and GERzone never disapppoints!

SPARKS said...

Wrecktor Ruinz, didn't I say he was a corporate joke?

SPARKS

Anonymous said...

Let me ask you a question, if someone is innocent why would they resign?

The crook, liar, and cheat has been caught with his hand in the cookie jar and has been shown the door!

It will be too funny to watch that dumb clown try and testify during the INTEL and AMD lawsuite. He raises his right hand and swears to tell the a fucking lie

Anonymous said...

Nothing is more powerful then moving to the next generation!

You get double the density in transistors, probably 10-30% frequency speed up, and power reduction. You could choose to keep the same transistor count, performacne and half your cost/double your output and also reduce your power by 30-50%. Or you can keep about the same die size, double the transistors, increase the frequency yet fit in the same power envelope. Nope, not a bad value given only 5-15% increase in cost. ANyone who thinks Moore's law is a waste is probably coming to late to the party. The people there reap all the benifits. Ask any DRAM company how important it is to scale to the next node as early as possible. ONly an AMD idiot would question it.

Its a seperate question as to which way you choose to go. INTEL has chosen to go both paths with high end CPUs that go into severs, desktops, and laptops wiht more cores with more features and more cache. They sell various options with respect to cores and cache at different price points. They also have gone small selling the incredible small atom, of which they get huge revenue from each wafer as there are so many die. Its FUD from people who say its undercutting their profits. ITs a seperate matter as to whether its undercutting the profits of their customers the comptuer manufactures. DO the math!

Introducing a new node before your competition gets you cost and performance leadership as you push your old technology into celeron space. Look at the competitions average ASP compared to INTEL, and look at the gross margins. ANyone who think a new node doesn't have ROI is an idiot.

Moore rocks and because INTEL is ahead it kicks ass in performance, power and cost. AMD has failed and ARM will soon go the way of the doo doo bird

SPARKS said...

That's a Dodo bird. However, it can be said AMD is in the doo doo, and in it deep.

http://www.yenra.com/dodo-bird/

SPARKS

SPARKS said...

ITK responce has givin me the gilhonies to try and debate this one, the risk of losing them.

"There is no such thing. 32nm is, by definition, cheaper to make. There is never a reason to delay."

Yes, INTC has already invested billions in for FABS ready (or will be ready) to crank out 32nM en masse. The quicker they crank, the more die they get per wafer, the cheaper they are, agreed.

However, conversely, there's a lot of inventory out there in Never-Never Land, and this is a depressed market. Further, kickass LGA 775 is dug in like an Alabama tick. Do you piss off OEM's and vendors by an early release BEFORE they clear inventory? Do you lower prices to make LGA 1156 or LGA 1366 more attractive than existing 775 hardware in the field, or do you wait until the market is ready, or all of these? I probably am wrong ,but from where I sit that's exactly what they've done.

Given the performance of current LGA 775, especially rockets like E8600's and Q9550 which are only 10 bucks apart, and are basically the same price as a i920. In fact, currently, most i7 are cheaper than a rough equivalent of 775.

Even a lunatic like myself resisted the change to the new platform (so far). OK, so QX9770 was a special case, but every nut I know is clocking the E8600 into the ground, on existing platforms. Got a LGA 775? Fine, pick a rocket, any rocket, for $275.

Far be it me to ever criticize INTC, but are you telling me they never delayed hardware for the benefit of the industry?

Sorry, INTC is going to play ball with the industry and allow 45nM LGA 775 die a very slow natural death, while 32nM takes a LGA 1156,/1366 32nM foothold in the market with LGA 45nM LGA 775 dying of attrition.

Frankly, they ain't gonna do anything early. Perhaps, and this is a long shot, they may tease the lunatic fringe (like me) with an early 32nm 6-Core behemoth 1Q. This is possible, but it will not be that early, and there will be nothing more till 2Q 2010 when 45nM LGA 775 is done, while simultaneously phasing out 45nM Core i7's.

ITK, those 450mm tools ain't gonna be cheap either, to say the least. That's a big plate, small time operators need not apply, and I'm no expert.

ANON,If you're implying that 22nM is on track on 450mm, perhaps you know something we don't, Bona Salute!

SPARKS

InTheKnow said...

ANON,If you're implying that 22nM is on track on 450mm, perhaps you know something we don't, Bona Salute!

Sparks, Intel couldn't keep it a secret if they were going to be doing 450mm wafers on 22nm. The equipment vendors would be hawking their wares trying desperately to recoup their investment and those that could would be saying "Intel chose our 450mm solution."

They would need a new facility to run 450mm in as well as you would need a whole new automation system that would most likely be incompatible with 300mm systems. So you either gut an existing fab, or build a new one. Again, not something Intel could really keep quiet very easily.

Since 22nm has to be in early development now to hit the ground running in 2 years, I think it is a safe bet that it will be on 300mm. The last schedule I saw for 450mm was for initial tool availability in 2012. That would be for development tools, so 15nm would be the earliest you will see 450mm wafers.

Roborat, Ph.D said...

InTheKnow, Sparks, Orthogonal, Khorgano, Tonus, others... can any of you post an e-mail on your blogger profile so I can give permission to publish.

At least one of you can continue on updating the site of with either news or personal opinion (it can get addicting). I'm sure we're all balanced here, unlike our friends over the 'zone. :)

939 posts is way too many.

SPARKS said...

Uh?? Me??? I certainly don't have the experience, to say the absolute very least, as most all of the boys on the site.

If this is legit, I'd like to see GURU take the spot. His analysis has expertise is second to none.

What say you G? If this is legit, of course.

SPARKS

Roborat, Ph.D said...

ITK and Khorgano can now update the site...

the challenge with GURU is he's quite anonymous. some even say there's more than one GURU. or that he or she may be an 8th year old child prodigy... is so if we can find a way to confirm who he is by e-mail, i am all open for it.

SPARKS said...

Kool, ITK, excellent.

SPARKS

SPARKS said...

Speaking of insider information, a few of my buddies who work on Wall Street have heard some rumblings concerning OLE Wreaktor Ruinz. Yes folks, it ain't good.

Although from the 'flavor' of the recent reports who seem to think Wreaktor didn't profit from the leaked information and therefore gets a pass, this is not nessesarily correct. You see, there's a little matter of damages to "fair trading". Remember, for every winner on Wall Street there's an equal number of losers. (It's a buy sell sort of thing) They are not happy. Nobody there likes ANY sort of insider moves, trust me, and these guy don't take prisoners.

After those who were involved in the insider trading give testimony, he still may be libel. Those cases will determine the extent of Wreaktor influence. There's a lot of money people charged in the case with lots of lawyers. Lawyers have a funny way of accessing blame. This will get uglier before it gets prettier, and given today's anti CEO climate, his head is on the chopping block, very much so. The SEC is quite active here, from what I've heard.

All said, these guys want their pound flesh. This, I can tell you for certain, is a fact.

SPARKS

SPARKS said...

Well there it is. New Yorks Attorney General has just taken suit against INTC.

Being 6.2B in debt, I guess he see's it as a windfall. Plus, he's got 1.2B invested in Luther Forest.

NYS in the chip business going after its biggest competitor.

I hate living here.

SPARKS

Anonymous said...

At last, Intel will get some good ol' spankin. I guess this will teach them a lesson once and for all.

Khorgano said...

At last, Intel will get some good ol' spankin. I guess this will teach them a lesson once and for all.

So when the government forces Intel to play "fair" and they sit back on higher priced chips in order to not "hurt" AMD, you'll applaud Intel for making the market pay higher prices?!?! Oh right, I forget, No matter what they do, they're in the wrong.

If you think prices are high, they must be gouging, if prices are too low, then clearly they're using their monopoly to dump on the market at AMD's expense, and if they have their prices somewhere in between then their obviously price fixing.

The story is never kept straight and you always play a semantics game to fit the situation to your agenda. New York state cashing in on an Intel fine will do ZERO for the benefit of the semiconductor industry, but it will do wonder's for Cuomo's career and pad NY's spending defecit.

Anonymous said...

Spending deficit - probably not so much (it may pay for the money stolen out of the taxpayers and handed to Dubai via Abu Dhabi)

Cuomo on the hand - expect a federal run in the future (Senate?). He could also go the Governor --> Pres route.

Anybody in that office in NY has near zero interest in doing good and every interest in tradng up to more power. Clinton went to NY merely to run for president. Many mayors think that can be cashed in to a presidential run. Higher office in NY is merely a stepping stone. Cuomo's office is a stepping stone to move to the upper levels in NY.

The problem is the state is so democratic, folks can typically get away with it - the only challenge when doing a crap job is if you are running for a new office and face a democratic challenger in the primary. Once an incumbent, it is almost like a tenured professor (except maybe governor).

Outside of Illinois, I'm not sure if there is a more corrupt state in the US.

SPARKS said...

"(it may pay for the money stolen out of the taxpayers and handed to Dubai via Abu Dhabi)"

Ouch-you're killing me. The truth hurts!

SPARKS

Tonus said...

"At last, Intel will get some good ol' spankin. I guess this will teach them a lesson once and for all."

I doubt that. Assuming that the lawsuit is successful, expect Intel to pay a fine and AMD to see none of the money, just as in the EU. It's amazing to watch governments fine a company after finding that it deliberately harmed another company, then not giving any of the compensation to the company that they built their case around!

Basically, they're squeezing Intel for money while hoping that Intel remains strong enough so that they can do it again later on. If they're really interested in "justice" they should be giving a lot of that money to AMD, since the basis of their claims is that AMD's situation is due to Intel's practices.

Therefore, it isn't just Intel that is being forced to bend over and take it. AMD has to watch while state and national governments say "see! we fined them! But, um... we're keeping the money and are not doing anything about your competitive situation... see you at the bankruptcy proceedings! LOL!!!!"

And guys like you cheer the first part and then try to grin as they slide that massive dildo with "FUCK YOU HAHA" written on the side. I just don't get it.

Anonymous said...

Therefore, it isn't just Intel that is being forced to bend over and take it. AMD has to watch while state and national governments say "see! we fined them! But, um... we're keeping the money and are not doing anything about your competitive situation... see you at the bankruptcy proceedings! LOL!!!!"

I'm not quite with you here. This is a criminal trial (I believe). While there somtimes is an element of restitution in criminal trials, there need not be.

However, criminal trials are quite often followed by civil trials which are MUCH easier to win if you've got one (or two) criminal convictions on your side. After all, there the standard of evidence (reasonable doubt) is much higher than (preponderance of evidence).

So, while this does not help AMD directly, should NY prevail, their civil case just got that much easier.

Anonymous said...

"At last, Intel will get some good ol' spankin. I guess this will teach them a lesson once and for all."

WTF from a dumbfuck!

Who has invested in capacity to bring computing to the masses? Who has innovated and bet the big bucks. Root for AMD and root for a company who's history is about a flamboyant silverhead salesman who drives a rolls royce and a crooked mexican. Then they prostitute themsleves to arabs.

They go to NY state for taxpayer handouts, and surprise surprise a attorney general with asspirations of being govonor sues to protect the weak company.

Last I check prices have only gone down the weaker and weaker AMDs is. The stronger they are suprisingly they don't fall.

Anonymous said...

However, criminal trials are quite often followed by civil trials which are MUCH easier to win if you've got one (or two) criminal convictions on your side. After all, there the standard of evidence (reasonable doubt) is much higher than (preponderance of evidence).

Ummm... nice generalization (and it is generally true), but there is one key thing you omit. AMD has to demonstrate that Intel actually HARMED AMD in terms of revenue... merely demonstrating Intel engaged in anti-competitive behavior is not sufficient - they have to demonstrate these actions specifically resulted in a loss of revenue - that is not going to be easy to do.

The governmental actions around the world (and I believe also in the US) is merely about Intel engaging in anti-competitive and not the "harm to AMD" part.

This is why you saw that ridiculous "study" AMD funded on the impact of Intel's "monopoly" awhile back.

This is also not a criminal trial. It is an action brought by a gov't against a corporation, not against a specific individual. But, hey you seem to have the law down pretty well...

Tonus said...

"This is also not a criminal trial. It is an action brought by a gov't against a corporation, not against a specific individual."

I was going to say the same thing. This is a lawsuit by the state of New York against Intel. It's like a civil suit, since they are alleging criminal acts (bribery and coercion, which are illegal) but not bringing criminal charges against any individuals (a process which would include arrests and arraignments).

The burden of proof in a civil trial is not as high as for a criminal trial. The government could proceed with a RICO case if they wanted to bring criminal charges, but that would be very controversial and highly unpopular and I don't think any AG would go that route over alleged anti-competitive actions. So a lawsuit alleging illegal activities is the option that is both safer and more likely to succeed (though as Guru points out, a civil judgment may be easier to get, but you still need to prove damages in order to get anything).

I believe that the EU did something similar, though I don't know how their laws work in regards to corporations and individuals. The fine may well be the result of a criminal complaint, or it may be a civil suit as well.

My point remains, if these governments are determining that Intel harmed AMD through illegal practices, then why aren't they giving AMD any money? They claim that they want to promote competition, but their actions indicate otherwise. If they are telling the truth, then AMD may be put out of business by a combination of illegal activities by Intel and a lack of redress by those governments.

If I'm AMD, I don't want the EU or USA doing me those kinds of favors.

Anonymous said...

Hector Ruiz raise your right hand do you swear to tell the truth the whole truth and nothing but the truth?

I do.

First I'd like to ask did you ever do anything illegal that would cause you to have to resign your position as the executive of a company?

Objection, not relevant.

Relevance to honesty of witness.

Sustained.

LOL

SPARKS said...

"they have to demonstrate these actions specifically resulted in a loss of revenue - that is not going to be easy to do."

I love to preface your observations. You throw the pass, and I love to run with the ball.

Case in point, the NYS suit, and the EU 'judgment' for that matter, are from 2005 on.
Correct me if Alzheimer's has gotten the better of me, but as I recall AMD did very well during 2005 up until they concocted Wreaktor's wild-eyed ATI deal for 5.2b in mid 2006. In fact, AMD shares doubled from $20 a share in 2005 and to $42 a pop in Jan 2006.

WHY? BECAUSE THEY HAD A SUPERIOR PRODUCT! They should sue Wreaktor for running the company into the ground. He did more to damage AMD than INTC ever did.

Nothing, absolutely nothing does more for computing than the biggest baddest processor on the planet. AMD had it just once, and that's history since 2006.

SPARKS

Anonymous said...

Sparks - compounding AMD's problems a bit was AMD claiming "we're selling everything we can make" back then.

If that were true, and if Intel was found to be limiting AMD's market - what impact is that? If they were selling everything they could make, what difference does it make if there was more demand?

If it is/was true the only potential argument is they may have shifted production mix toward a higher ASP products, but again this is going to be a tough point to prove. This may be a circumstance where boasting and puffing out your chest for the analysts may come back to bite AMD in the ass.

And if Ruiz, et al say they were exaggerating that point to the analysts then they open themselves up wide for a lawsuit from analysts and shareholders and potential criminal charges. Wouldn't that be ironic?

SPARKS said...

It would be very ironic, indeed. There are additional inconsistencies and contradictions which are a matter of history that are still felt by one segment of AMD's market to this very day.

I'm sure you recall when AMD signed on with DELL. They abandoned the 'channel' and the 'whitebox' vendors to supply DELL with a contractually agreed volume of chips. As I recall they couldn't supply both. They were not under any contractual obligation to the channel. Guess who took the hit?

As of late, because of their finacal/market position, they recently started courting the channel on their new "Fusion Platform Program." The channel was less than enthusiastic, to say the least.

"The Dell (NSDQ:Dell) thing is still on a lot of people's minds," said the source, who asked not to be named. AMD's late 2006 deal to supply Dell with desktop processors was seen by many in AMD's whitebox channel as a slap in the face"

Therefore, not only did they, as you pointed out, "sell everything they made," there were also shortages and they couldn't supply everyone. I wonder how they'll twist this one to be INTC's fault?

http://www.crn.com/white-box/220300557;jsessionid=
HOYTAF4PNFQAJQE1GHRSKH4ATMY32JVN

SPARKS

Tonus said...

I am assuming that Intel won't use the reasoning that AMD was its own worst enemy until they address AMD's lawsuit. Until then I expect them to claim that they did no wrong and pursue that angle vigorously. I don't think that they'll admit to wrongdoing unless the case against them is so overwhelming that their best hope is to negotiate a settlement.

Now, for the AMD lawsuit I'd expect them to pursue the strategy that AMD is at fault for their own problems, not Intel. In that context, AMD's statements about sales figures and capacity would be far more helpful to Intel's case. This is probably why the national and state actions against Intel are focused on what Intel did, versus whether or not anyone was harmed.

It's also what makes it look like little more than a money-grab.

But the realities of the industry and CPU technology also play a part in how this all turns out. Governments may milk Intel for cash (literal or political), but they also know that there's no other entity that could pick up where Intel is right now and continue the pace that they're on. Even if they do give AMD any money, they know that harming Intel does not automatically help AMD. More likely, it would just harm the market.

Then again there's a lawsuit against IBM for similar anti-competitive behavior, so maybe it's just an attempt to play god with the technology sector. Government intervention on this level... what could go wrong? :/

Anonymous said...

This is probably why the national and state actions against Intel are focused on what Intel did, versus whether or not anyone was harmed.

There is also different legal standard - the burden for national and state type suits is that Intel violated business laws - they don't specifically have to show harm to a specific competitor.

And in the case of the EU and some other nations, they don't even have to show that the end user was harmed. In fact if the consumer benefited, they can still legally enforce the fine.

While I think Intel will continue to deny the allegations, they do have a second legal prong of: even if it is true, what damage was done to AMD if they were already selling everything they could make? If no damage can be demonstrated, then AMD may have the old $1 lawsuit victory.

The other item I think still at issue is what is viable in the lawsuit. Intel is arguing that as the chips were manufactured in Germany, the US only has jurisdiction on US sales, not global sales (which would be the case if they were manufactured in the US) - this substantially limits potential damages because now you are talking about damages in the form of US lost revenue (where AMD does better) and not global revenue. There is precedence on this from an old pharma case where it was determined that even though the company was American, because they manufactured overseas, only sales in the US were part of the US jurisdiction and lawsuit and not global sales.

SPARKS said...

It looks like 'Anon' who said....

"There is no such thing. 32nm is, by definition, cheaper to make. There is never a reason to delay."

...is on the money, this time.

Bit-Tech is reporting the new replacement for the i920 will be on 32nM. Where the Q6600 got a heathy stepping (GO), I do believe the 920 will get a bit more (or less depending on how you view a smaller die) in the i7 930.

'Anon' did know something we didn't, after all. Yeah, yeah, I know, "We never comment on unreleased products"

In any case, if this thing is on 32, holy moly, they may release Gulftown before 2Q!

http://www.bit-tech.net/news/hardware/2009/10
/28/core-i7-930-arriving-q1-2010/1


SPARKS

InTheKnow said...

The whole Intel case leaves me cold, because in my opinion the US legal system has lost it's way. I think this is best illustrated by the words of no less a personage than Oliver Wendell Holmes.

When exhorted to "do justice" Justice Holmes' reply was "This is a court of law, young man, not a court of justice."

When a pre-eminent supreme court justice has lost sight of the purpose for which courts of law were created, it doesn't leave much hope that the current state of affairs will improve. Until and unless this attitude is changed, we will continue to have a legal system and not a justice system.

I believe that people know what is fundamentally right and wrong, but as long as the legal system insist on encouraging a system based on splitting hairs and precise verbiage, then we will continue to get more of the same.

InTheKnow said...

In any case, if this thing is on 32, holy moly, they may release Gulftown before 2Q!

Gulftown is a six core processor. The reported i-930 is said to have 8 threads. If that is the case, then it isn't going to be Gulftown. I just don't see Intel only putting Hyper-threading on only 2 of the six cores.

So the release of the i-930 really has nothing to do with Gulftown and it's timetable. There is also no performance data or thermal data being shown, so I'm not sure what you can really conclude from that report.

SPARKS said...

"So the release of the i-930 really has nothing to do with Gulftown and it's timetable. There is also no performance data or thermal data being shown, so I'm not sure what you can really conclude from that report."

Thermal data?

OK. If there is such a thing as i930 on 32nM, that means INTC has 32nM up and running, no?

A few of our 'special' members have alluded to the possibility (at least by the end of the year), right?

Gultown will be on 32nM.

INTC does enjoy showcasing their hottest wares on the heels of a new process node, sometimes? After all, a FAT 6-Core XE chip on 32nm makes a bigger marketing splash than a lowly, made for the masses, plain Jane, BORING, Quad on 32nM, hmmm?

INTC still has a marketing department, don't they? Can anyone verify this? Have they forgotten the LuNaTiC fringe?!?!?

Right, 4 cores + H. T. = 8 threads, 6 cores + H. T. = 12 threads, got it, thanks for the tip.

Hey, they are a $40B a year operation. Do ya think they run a couple of thousand disc's through the mill with some six core dies on 'em, or are you suggesting the tools that make the 32nm i930 quads can't make the 32nM 6-Core Gulftowns?

Special 32nM Gulftown tools, is that it? Is that the way it works?

Aren't the XE chips the highest margin chips on the planet? I know at least one crazy bastard who's dropped 1500 bucks on a XE chip, and has absolutely NO problem doing it again. XE chips are (partly) about pomp and circumstance, and why not?

I wasn't "concluding" a goddamned thing. I was speculating, so was the article. I'm allowed to speculate, right?

SPARKS

InTheKnow said...

Jeez, Sparks, take a chill pill. :)

I wasn't trying to insult your intelligence, just point out that the report didn't offer much real data to base speculation on.

Let's look at this from a capacity standpoint.

Intel is committed to i3 and i5, right. Presumably D1D is ramped and D1C is ramping up now. That will give you about 1.5 fabs to supply all those needs until the next fab ramps. Fab 32 is scheduled for a Q2 production ramp according to the reports.

I just don't see Intel throwing yet another chip into the works prior to Q2. Personally, I agree with you that Intel will go with Gulftown earlier rather than later. What I don't think is that you will see the chip in this report prior to Q2'10.

They might pull in Gulftown a bit since it will be low volume initially, and as you say, it's high margin. But if anything, this report would imply that Intel will push Gulftown to make room for this unspecified processor. I think that is less likely.

SPARKS said...

"Jeez, Sparks, take a chill pill. :)"

Ah, don't mind me. I'm just bitching 'cause I haven't had the fastest frigg'm machine on the planet since they started clocking i920's (You read me Tonus?). I don't even want to think about an i975.

From all reports Gulftown is kick'n some serious ass at 2.66 GHz, ~20% over a similarly clocked i920. (INTC schrewedly released ES that had locked multipliers) They're holding this one pretty close to the chest. I'm figuring that the architecture boys have Nehalem pretty well refined and nailed down, they added a few tweaks, and I sense a great deal of confidence with the process boys. (And you know who I'm talking about).

The more hush-hush and coy these guys get, the more goodness they have behind the blue curtain. I'm just pissed I that can't get my hands on one of these things, that's all.

Hey, the Yankees just won the Series, and I want my six core XE behemoth on a new machine with multithreaded OS-----NOW.

Rampage Extreme is an understatement.

http://www.tweaktown.com/reviews/1662/
asus_rampage_ii_extreme_x58_motherboard/
index13.html

Sorry ITK.

SPARKS

SPARKS said...

Then there is this. So much for going fabless, yeah, right,sure, OK.

http://www.theinquirer.net/inquirer
/news/1561517/amd-gpus

SPARKS

Anonymous said...

TSMC yields are down for chamber matching, oops I mean implanter now. Funny how the story keeps changing.

Now I read that them Arabs and GF are gunning for INTEL. They think a consortium of people who individually can't make a process think togather they can better intel. Think about it a room full of Koreans, Germans, Chinese, Taiwanese, Japanese, Arabs, and one Mexican in the corner are going to be able to compromise quickly and decide on design rules, process targets, tool selections.

Yeah right.. sorry arabs even 10 billion now and another couple ever year for the next 10 years can't buy what chipzilla is doing.

JanuskieZ said...

Hi... Looking ways to market your blog? try this: http://bit.ly/instantvisitors

SPARKS said...

I thought it might be interesting to examine two pieces of news that are 180 degrees in diametric opposition. It makes one wonder about AMD's in house market analysis.

"In Q3 2009, Intel earned 81.1% share of the worldwide PC processor market's unit shipments, a share gain of 2.2%, while AMD commanded 18.7%, a loss of 2.0%, and Via Technologies had share of 0.2%"

http://www.xbitlabs.com/news/cpu/display
/20091109155410_Microprocessor_Unit_Shipments_
Grow_23_in_Q3_2009_Market_Tracking_Firm.html



And.......



"The world's No. 2 microprocessor company expects gross margin of 40 percent to 45 percent in 2010. That compares with 38 percent in the third quarter and 27 percent in the second quarter of this year."


http://news.moneycentral.msn.com/ticker/
article.aspx?Feed=AP&Date=20091111&ID=
10694010&Symbol=AMD

WTF???


G, the slash and burn tactics still ain't gonna work.

SPARKS

Khorgano said...

Looks like Intel and AMD have settled the Anti-trust suit for a cool $1.25B.

http://finance.yahoo.com/news/Intel-
settles-AMD-claims-but-apf-
748922183.html?x=0&sec=topStories&
pos=5&asset=&ccode=

The whole ordeal is of course absurd, but this should be enough to shut the whiner's up and AMD gets another hand out to keep chugging along.

I'll probably post something on the front page after I've had some to chew this over ;)

Tonus said...

I was about to make a comment about it. :)

AMD had to have the upper hand here, I can't imagine Intel giving up such a huge sum of money otherwise. The cross-license agreement alone would have been a pretty big notch in AMD's belt, IMO. I expect that the terms of the agreement will be sealed and the two companies will agree not to share any details.

But I can't see how this is anything but a big win for AMD. $1.25 billion? Wow.

Anonymous said...

Hmmm... cross license reneg and lawsuit settlement combined and settled... who could have poredicted this mighr happen...

....about a year ago!

;)

It'll be interesting - obviosuly details will be sealed but expect a bunch of class action lawsuits for money grabs by ambulance chasers - not sure if this stays sealed; though I think wrapping the patent stuff/license reneg into it is probably enough to keep it sealed. Also wonder what happens with Cuomo, et al...

This is a win for both companies - AMD really needed the cash and couldn't roll the dice on losing the x86 license - they were skirting on the edge with the GF acquistion of SMIC acquisition and capital calls coming up for the NY fab buildout, they probably were going to lose the "it's still a subsidiary that we control(wink, wink)" card

«Oldest ‹Older   801 – 973 of 973   Newer› Newest»