"Real men have Fabs" - Jerry Saunders, former AMD CEO.
After all the dodging and lying about not having any plans going fabless, AMD finally admits that it is in fact going to sell its fabs. They intend to remain a minority share holder and team up with a company who haven’t got a single clue about semiconductors. But none of that matters because they have lots of money to spend. And in a time when the usual lenders are looking for a some kind of a bailout themselves, it’s either this arrangement or Chapter 11 for AMD.
"On Oct. 7, 2008, AMD and the Advanced Technology Investment Company announced the intention to create a new global enterprise, The Foundry Company, to address the growing global demand for independent, leading-edge semiconductor manufacturing. There is a strong shift to foundries occurring – particularly to foundries with the capacity to produce devices using leading-edge process technologies. With The Foundry Company, AMD will be able to unlock the value of its world-class manufacturing capability – by making it available to a growing community of fabless semiconductor companies. " - AMD's New Global Foundry page.
There's a lot already said about AMD's move. The best one that highlights the concerns more clearly is from Fabtech:
Another aspect that concerned me was the notion pumped out by AMD that demand for leading-edge foundry capacity was something that was in strong demand. Ask SMIC, Chartered or UMC how much of its capacity is allocated to 65nm-and-below production and you will find it is very small... Also, if demand for foundry capacity was that strong, then why are wafer ASPs in decline and the major foundries cutting CapEx each year?... Even worse is the fact that SMIC has struggled since birth to actually turn a profit, so why should we think that a new foundry start-up in Europe (and the U.S.) will fare any better?
The "growing trend" to go fabless is a decision forced upon companies due to the rising cost of running a Fab. This alarming trend have been identified by Intel in the late 90's. AMD knew this day was coming and set a goal of 30% market share just to avoid ending up where it is now. It's entertaining to see how AMD is making it all appear as if it was an advantageous choice.
How can it be advantageous when AMD will now have to ask another company to spend $Billions every time it wants new equipment for a new process technology. How can it not be disadvantageous when your main competitor do not have such bureaucratic problems. Short term, this move by AMD will buy them time for the next year or two. Long term, I can only see AMD becoming the next Transmeta. When The Foundry Company starts losing large sums of money they will become cost conscious and that's when problems begin. Against Intel's "tick-tock" execution, AMD will only struggle to keep up.
10.08.2008
AMD comes out the closet, says no longer a real man
Subscribe to:
Post Comments (Atom)
236 comments:
«Oldest ‹Older 201 – 236 of 236Please, someone, tell me that I don’t some unexplored chemical imbalance;
Sparks, let me simplify it for you:
- First was performance (then that went away)
- Then it was performance per watt (as raw performance couldn't keep up)
- Now it is Price/performance per watt (meaning performance/watt wasn't keeping up so now they just slash prices accordingly)
Now performance per watt I can somewhat understand (even though it is difficult to quantify absolutely), but price per performance per watt? "We are cheaper" (which is all this new metric is), is not a solid marketing plan.
If something is 4 times slower but 5 times cheaper, it is "better" in the price/performance metric... but would you buy it based simply on the ratio? I know when I went looking for TV's I looked at price/per square inch of display and found a cheap 13" TV to be the best for my entertainment setup!
As there is a limit to how much you can cut pricing I predict this new "metric" will soon be replaced by "per technology node" (this way they can normalize out Intel's tech node lead) or per area of silicon or per the amount of R&D spent or days spent developing the chip or....
"Hey, JumpingJack! Ed at overclockers deleted your excellent comment! You've hit the nail on the head, and the chicken shit deleted it." ...
Yeah, I noticed that :) :) ... I always found his commentary 'amusing' most of the time ... never really rooted in the data, but most of it was abstract financial stuff about AMD floundering.
Nonetheless, I was confused when he kept complaining about the lack of overclocking, even when a half dozen sites showed a 2.66 GHz 920 hitting 3.8-4.0 GHz -- what more does he want? This is about right in line with Penryn based cores, add to that 20% on average IPC improvement, and even 3.8 GHz will beat a 4.0 GHz Penryn in most all applications ...
Nonetheless, I was confused when he kept complaining about the lack of overclocking, even when a half dozen sites showed a 2.66 GHz 920 hitting 3.8-4.0 GHz -- what more does he want?
He, like many others, have gotten spoiled with overclocking - the Core2 OC'ing is so easy now, folks just assume the next gen should be as easy and better.
I would like my VW to have the acceleration of a Ferrari by changing the gaps on my spark plugs adn amybe doing a few simple tweaks, but I've learned if I want that performance I'll need to pony up.
People have gotten use to turning a $100-300chip into a $1000chip typically by changing one (FSB clock) or maybe 2 (Vcore) variables in the BIOS. It's like getting a real nice gift for your birthday 1 year and then expecting something as nice or better every new birthday and being disappointed if you only get a so-so gift.
"OC'ability" is not an entitlement. That said, the jury is still out on the new architecture and it might be just as good OC'ing as Core2. If it is, folks are lucky and should be happy about it - if it is not, buy a Core2 and shut the heck up!
anon: ""OC'ability" is not an entitlement."
That is what I was thinking as well. I think that Ed typically takes the point of view that as an OC'ing site, their primary interest in any CPU or CPU family is how well it overclocks, and that is understandable.
Treating it as a right, or speaking to the obligations of manufacturers to overclockers, is a whole other kettle of fish. Hobbyists are a small market, one which chooses to deliberately abuse hardware in order to get more than they paid for. They can generate some publicity, but it's most within their own world. There's no real incentive for companies like AMD or Intel to worry about them, IMO.
It's about knowing your place in the scheme of things. The company I work for will never OC a single CPU and they don't care about OCing one bit. And they'll buy more CPUs than I ever will, by orders of magnitude. Who do I expect Intel to be more concerned about?
"OC'ability" is not an entitlement. That said, the jury is still out on the new architecture and it might be just as good OC'ing as Core2.
This is the A4 stepping on Ci7, right? At least that is what I recall from a CPU-Z screenshot I saw on Xtremesystems. I bet there will be another stepping in a few months, which is when I plan to build anyway.
Should be C0.
I bet there will be another stepping in a few months, which is when I plan to build anyway.
In the near term (next 6 month-9 months), you will likely only see minor steppings like C1 or C2 (barring a TL-like bug!?!)
Down the line (mid'09 or a bit later) you will likely see 1 major stepping revision (like the G0 on Core2). This will undoubtedly bring power down, may increase the upper speed bin a bit, but may or may not impact 'OCablity'
“Should be C0.”
Who was that masked stranger? Quite right, C0 it is.
“……but price per performance per watt?”
Thank you, I thought it was some mid 70’s flashback due to my weekend mushroom collecting forays in the Colorado cow paddies.
“Yeah, I noticed that :) :) ...”
Thanks Jack, some of the boys jumped me for not copying/saving the comment. (Yikes!) You didn’t save I copy, did you? It was well done, and worth the read. Reprint it here if you did.
You must have got to him, or else he wouldn't have deleted it. The logic was impeccable and he looked absolutely foolish.
"OC'ability" is not an entitlement. That said, the jury is still out on the new architecture and it might be just as good OC'ing as Core2.”
Well said. The only way I’m getting in the ‘Harvard Club’ is when my buddy takes me there. So too with XE chips, and that’s that.
As far as performance there’s no contest. Nebojsa Novakovic, a very capable and objective enthusiast at the INQ, has an excellent report/review on the new bad boy i7 965 and my bad boy QX9770. 4 gig is the practical limit both (practical meaning 100% stable with all games and apps), but i7 965 does the Donne Della all over QX9770 at the same speed(s).
And, that’s that.
http://www.theinquirer.net/gb/inquirer/news/2008/11/06/nehalem-very-first-inqpressions
SPARKS
"G"!!! Where's the King, where's the King?!?!
http://www.theinquirer.net/gb/inquirer/news/2008/11/07/first-shanghai-benchmark
Woof Woof it's a dog!
SPARKS
That should prove interesting. I am going to take the Inq's words with a grain of salt, because I can't believe that AMD would purposely rig an Intel system to flop in order to show their own CPU in a better light, considering that there are already legit Spec scores available.
If Shanghai scores are that bad, no amount of PR shenanigans will save it. When are they supposed to release this CPU, paper or otherwise?
"Treating it as a right, or speaking to the obligations of manufacturers to overclockers, is a whole other kettle of fish. Hobbyists are a small market, one which chooses to deliberately abuse hardware in order to get more than they paid for. They can generate some publicity, but it's most within their own world. There's no real incentive for companies like AMD or Intel to worry about them, IMO."
Not only this ... but Ed was way out of line with the overclocking 'is a treasure hunt' fiasco. My opinion is that he searched over all the review sites to find one that would make some sorta statement inline with his personal analysis that Intel is trying to create a 'luxury' line of CPUs and block overclocking ... he thinks finding a good i7 OC is some sorta treasure hunt. He chose Tom's Hardware's initial review, even then they got 3.8 GHz -- and Tom's had to print a retraction because someone called them up and explained how to OC a CPU.
Here are some i920 OCs from around the web, it was not a treasure hunt, it was not hard to find:
Bit-tech, i7 920 @ 4.0 GHz
http://www.bit-tech.net/hardware/2008/11/06/overclocking-intel-core-i7-920/1
Xbitlabs i7 920 @ 3.87 GHz
http://www.xbitlabs.com/articles/cpu/display/intel-core-i7_17.html#sect0
Anandtech i7 920 @ 3.89 GHz
http://www.anandtech.com/cpuchipsets/intel/showdoc.aspx?i=3453&p=2
My least favorite site of all, HardOCP 3.8 Ghz ("easiest 1.14 GHz he ever got:") i7 920
http://enthusiast.hardocp.com/article.html?art=MTU4MCwzLCxoZW50aHVzaWFzdA==
1.1 to 1.2 GHz more is not shabby and is behaving better for OCing than the equivalent 2.66 GHz Penryn.
Now a quick google revealed 2 sites that reviewed the Q9450 (2.66 GHz Penryn), the OC's on those were 3.7 Ghz (http://www.overclockersclub.com/reviews/intel_q9450/3.htm)
and 3.6 Ghz
http://techgage.com/article/intel_core_2_quad_q9450_266ghz/13
So far, OCing the i7 920 is actually better than Penryn from the 10 mins. I spent looking.
I don't care if Ed doesn't like the processor or thinks it is some how inferior, what I do care about is when the data is ignored and reality is misrepresented, he is either uber arrogant or uber stupid.
Jack
Follow up ... spent a bit more time, and few pagers deeper into the google search ...
a third Q9450 OC in a review:
http://www.neoseeker.com/resourcelink.html?rid=153480
this got to 3.8 Ghz
So so far,
3.6-3.8 for Penryn 2.66 Ghz stock proc,
3.8-4.0 for i7 2.66 Ghz stock proc.
It would appear the locked proc OCing has indeed improved.
Comment - Feb6'08
45nm should start up reasonably well for AMD - they aren't changing much at all. For all the talk of immersion litho, quite frankly it is just an alternate method to pattern wafers - it is not something that from an integration perspective is going to muck things up. Yes it is not a no-brainer, you will need new resists, work on cleans, have to tune masks in terms of OPC rules, etc...and I don't mean to completely de-value the importance of the technology but frankly how is this realistically different than what has been done in lithography before? ( for example the switch from 248nm to 193nm litho)
As a result 45nm should be reasonably behaved at it will essentially be an optical shrink of 65nm. As has been discussed it will not have great performance gains - you will see active power benefits but likely see minimal clockspeed benefits over what 65nm SHOULD have gotten. You will see clockspeed gains and many will hail this as AMD being great, but it will represent fixing things that they could not on 65nm (so in my view this is a hollow victory)
For example if 45nm churns out a 2.8 or 3.0Ghz Phenom is this demonstration of a great 45nm process... or simply achieving what was originally EXPECTED for the 65nm process?!? I'm 99% certain the ignorant folks will be saying it's because of immersion litho and ULK ILD's! Keep in mind the lowest original expectation for 65nm was 2.8GHZ (most thought at least 3.0 or even 3.2!), so if you believe AMD's 45nm ~25% improvement, it should be in the 3.4-3.8GHz range. In reality 45nm will likely get to where 65nm was originally expected to get, and many with short term, selective memories will be singing AMD's praises.
Thought I'd dig this assessment out 9 months later.... I think it is pretty much is/will be in the ballpark, especially the last part about the AMD fans evaluating the 45nm overall performance.
Comment-Apr23'08
I tell you what - if I were Ballmer right now... I'd threaten to walk away and say 'wow, if he can get such great performance, perhaps we shouldn't take the company oover and then when the stock crashes to the pre-takeover level and crashes again when Yang missed his ridiculous Q2 numbers, Ballmer should step back in and lowball an offer and say "how do you like me now?!?"
Well I was a little off on the timing, but here we are with Yahoo @$12.2/share (vs the MS offer of $33 back in late Jan) and Yang publicly BEGGING Microsoft to do a deal...
And Balmer says no, just as publicly ('It's in the past')! Perfect! The pressure on Yang will cause him to step down or make the first offer (perhaps thru Icahn) to MS... and MS will be bargaining from a position of strength. And Balmer or may not say "Do you like Apples? I said do you like Applers? Well how do you like them Apples"?
http://seattle.bizjournals.com/seattle/stories/2008/11/03/daily37.html?ana=yfcpc
Despite Balmer's comments, I think there is at least an even chance of doing a deal now - though Yahoo will be lucky to get ~17-19/share out of MS now (or not too far from 1/2 off the original offer!). Or maybe Yahoo needs to find some financial friends in Dubai?
So in the battle for the worst stagerinessment would you rather have Ruiz or Yang running your company?
jack: "and Tom's had to print a retraction because someone called them up and explained how to OC a CPU."
Hehe, I got a chuckle out of their 'update' to their article, wherein they apparently blame an Intel motherboard for their early OC results.
You make an excellent point about research. I am going to assume that Ed was comparing the Nehalem OC results to the hobbyist results that have popped up from time to time, listing overclocks of 4.0 to 4.5GHz on Penryn systems. If Nehalem is OCing as well as Penryn did, then it's likely that we'll see hobbyists hitting those speeds as well. Time will tell, but that was a bit sloppy.
I also find it a bit disingenuous that he is now stating that the primary thing that matters is how much speed you end up with. He uses the example that a 10MHz CPU that overclocks to 100MHz is not that great because after all, 100MHz is very slow.
So now I am wondering... if Intel or AMD releases a 5GHz CPU that only overclocks to 5.2GHz, would he consider that CPU a 'great OC' because 5.2GHz would make it the fastest x86 CPU?
“Ed was comparing the Nehalem OC results to the hobbyist results that have popped up from time to time, listing overclocks of 4.0 to 4.5GHz on Penryn systems.”
Tonus- Be careful with that range. Not that I’m disagreeing with you, in fact, this is exactly what he does. What sticks in my craw is the way most of these ‘overclockers’ throw that 500 MHz range around like Democrat’s at a pork barrel fest.
Ed, being the overclocking expert, (yeah right) should know better. There is a world of hurt waiting for ANYONE in the 4.27 GHz range with Penryn. Weird shit starts happening, and you live with unpredictability. So much so, the difference vary from chip to chip, even at the top bin. I’ve got a cherry picked one and I know this for a fact.
If I learned anything on this site, I've learned the limits to speed are directly proportional to its architectural design and it's material design. Heat dissipation, thermal runaway, material breakdown, leakage, crosstalk, the list is endless, and far beyond my scope.
But, I paid 1500 bucks for the education.
How did GURU know exactly Pheromones would clock to 2.7 ~ 3.0 range? GURU knew how it was made and what it was made of. The similarities to other branches of engineering are remarkable, especially when you’re pushing the envelope. He’s like the Smoky Yunick of chip design.
http://en.wikipedia.org/wiki/Smokey_Yunick
Smokey knew that there’s “no substitute for cubic inches”. GURU knows the limits of chip design and material structure. Similarly, like all branches of engineering, it’s like living on the exponential curve, be it horsepower, speed, power output, whatever.
The broader the base of the “tit” the higher you can go to get to the nipple. The bigger base, the higher nipples, it’s just an exponential FUNCTION of the design, right off the chest.
However, in all cases the closer you get to the tip the easier it is to fall off the other side, and it’s very unstable up there.
Penryn and i7 has Dolly Parton sized hooters (i7 is shaping up to be a bit firmer), whereas Pheromones has ‘Twiggy’ sized pancakes and they just ain’t gonna get there. Call it baseline torque, baseline dynamic scalability, or a huge set of knockers, who knows?
Obviously, Ed doesn’t have a clue, and he’s no engineer. Neither am I. But when I’m groping around in the dark I can assure you I know what I’ve got in my hands.
GURU, however, knows his tits.
SPARKS
How did GURU know exactly Pheromones would clock to 2.7 ~ 3.0 range?
It was a guess, with some working knowledge (the absolute closeness was a bit of luck). You need to look behind the double talk on the 45nm improvement:
1) ULK (the Ultra Low K dielectric used to electircally isolate the Cu wiring) was said to provide up to 25% performance but what was not mentioned that this could only be realized if the transistor was sped up.
Let's say you have water flowing through 100ft of 3/4" pipe and want to increase the water flow. If you simply increase 50 ft of the pipe to 1" diameter (and leave the rest the same), will you get any measurable difference in water flow? But I can in a clever say it offers the potential of a gain and conveniently leave out the fact that the only way to do this would be to change the other things. That's why IBM/AMD said potentially (or up to) a 25% gain and so were not technially wrong, but were clearly misleading the unwitting press.
2) Immersion litho doesn't do squat to performance if you are printing the same feature as a double exposure process. You can argue over the manufacturability, the cost or the impact on design restrictions... But if I print a 40nm gate width with immersion and a 40nm gate width with a double exposure process... they are both 40nm!!!
3) Couple the minimal impact of the above (which were really the only 2 things AMD was touting) with the inability to scale the gate oxide any further (due to leakage issues with SiON), and unless IBM had some breakthrough on implant or strain, there weren't any major transistor improvements coming. And if there were major breakthroughs, IBM would be screaming and printing it all over the press with their corporate culture.
The active power gains were fairly predictable - these are expected due to the lower Vt for the 45nm process (which allows either lower Vcore or the same Vcore with improved clockspeed). It is also likely AMD cleaned up any 65nm issues they were having.
After that post by sparks, I am left wondering if AMD will go tits up!
“the inability to scale the gate oxide any further (due to leakage issues with SiON)”
“The active power gains were fairly predictable - these are expected due to the lower Vt for the 45nm process (which allows either lower Vcore or the same Vcore with improved clockspeed). It is also likely AMD cleaned up any 65nm issues they were having.”
Regarding the move from 65nM to 45nM---
All things being equal (I realize they never are), but for the sake of argument, let’s say because of the shrink, the ULK process offers a thinner insulator to the Cu interconnects and a smaller insulator at the gate, hence a higher (numerical k) dielectric constant. Therefore, process for process, you don’t gain anything and TDP still remains the same, if not worse. Sure, you have a smaller die and a smaller feature set, but you’re still leaking like a pig, you’re interconnects are closer together, and you’re ‘k’ is climbing into the stratosphere, even at lower V.
Smaller die less power, higher K more heat, net gain, peanuts.
Beside the tweaks you mentioned, the gains would be marginal, at best. From your previous explanations, this is what I have come to believe.
They’ve gained nothing, but more real estate and more Triple Cripples. I really don’t know what this “Bangkok” hype is all about.
Tonus, they are tits up, making a living on they’re back, giving it up to the Middle East.
SPARKS
...the ULK process offers a thinner insulator to the Cu interconnects and a smaller insulator at the gate,....
Smaller die less power, higher K more heat, net gain, peanuts.
Sparks, the ULK material shouldn't affect the gate performance. I don't think they are being used at metal 1 (the interconnect layer closest to the transistor), though I could easily be mistaken on that point.
Here is a link to an article on ULK materials. It is more focused on integrating low k materials into the device than the issues you are talking about. The key takeaway for this discussion from the article is
Scaling of integrated circuit (IC) feature sizes has brought tremendous performance improvement and miniaturization in the past few decades. The primary contribution to the IC performance improvement has been coming from the device gate level. However, as device scaling continues into the deep-sub-micron region, metal interconnects become the bottleneck for continued IC performance improvement. The gain in device speed at the gate level is offset by propagation delays at the metal interconnect due to the increased RC time constant. The RC time delay can be reduced by the incorporation of low dielectric constant (k) materials and/or high conductivity metals. The use of low-k dielectric materials also lowers power consumption and reduces crosstalk.
So introduction of ULK materials does not degrade device performance. But as others have posted, it also doesn't help your device if the switching speed isn't fast enough to justify it.
ITK has it dead on... one additional clarification:
The RC time delay can be reduced by the incorporation of low dielectric constant (k) materials and/or high conductivity metals
IBM/AMD has never reported the effective dielectric constant of their process (to my knowledge). In addition to the ULK, ILD film stacks also consist of an etch stop layer (typically SiN or SiC or some derivative thereof) that are needed for patterning. These materials are relatively thin compared to the ULK, but they also have a much higher dielectric constant which has a negative impact on the overall capacitance of the combined films. (think of it effectively as 2 capacitors wired serially)
So while it is nice to report the ULK #, it is meaningless without a description of the etch stop layer IBM is using and its impact on the overall capacitance. I'm sure the combined stack is better than what they previously used (otherwise why switch to it?), but it is my belief IBM continues to quote the bulk film only to get a more 'dramatic' effect.
And it is hard to believe that IBM is running an RC delay limited process, so as I and ITK have mentioned speeding up only the interconnects will not lead to a 1:1 increase in chip speed.
And this is the frustration as AMD waves these itemized performance increases out - they don't always equate to the same actual overall process performance. They also aren't always completely additive (10% + 10% may actually turn out to be 10% overall and not 20%).
They played these games with the 65nm process where they kept comparing the benefit of all their strain technology to a completely unstrained device as opposed to simply comparing it to the previous (90nm) technology node. This is why I discounted a lot of the 65nm gains way back when and saw "the cliff" coming at a lot lower level then I think some expected (and even I was on the high side as I think I said 65nm should top out at 2.6-2.8GHz)
It's another version of the three card monty game, and for folks not working in the field, it is rather easy to misinterpret some of the subtleties (intentionally?) introduced.
"They played these games with the 65nm process where they kept comparing the benefit of all their strain technology to a completely unstrained device as opposed to simply comparing it to the previous (90nm) technology node. This is why I discounted a lot of the 65nm gains way back when and saw "the cliff" coming at a lot lower level then I think some expected (and even I was on the high side as I think I said 65nm should top out at 2.6-2.8GHz)"
In fact, when they originally produced a technical paper on 65 nm it generated the 'IBM/AMD to improve 40%" headline. However, it was, as you said, compared to equivalent transistors without the enhancements.
It was easy to pick out the paper in the library and using s simple old pencil and ruler determine that AMD/IBM's 65 nm parametric were falling about 10-15% short of Intel's....
You need to be careful when you read these grandiose headlines coming from this camp, for them image is everything and a slight of hand in they way you present information can be 'virtually correct' but completely irrelevant.
“Due to the thinning of metal lines and the rise in current density, a large amount of heat is generated at the metal interconnect. This Joule heating in an interconnect system can generate a big localized temperature increase above the silicon junction temperature and thus create reliability problems such as the shortening of electromigration lifetimes. The rise in metal temperature due to Joule heating depends, among other factors, on how efficiently the heat is removed from the metal lines. The heat flows through metal lines and vias as well as dielectric films to the Si substrate. Therefore, the dielectric films not only serve as electrical insulators, but also as thermal conductors to carry away heat from the metal. In some cases, they are the dominant path for the heat transfer.”
ALL- Whoa! Easy partners! I stand corrected! I didn’t know the gates required a different set of rules and proceedures regarding insulation. That,s because I don’t what I talking about. However, the above quote from ITK’s excellent link was on my mind.
Whats more revealing is that things VERY touchy with you guys when someone even smells the transistor. It’s like, “you can over generalize about anything, but watch what you say about the transistor, OK knucklehead?”
A number of things ockurred after your collective replies (aside from you better knowing what you’re talking about):
1. Transistor construction, during process, is given it’s own set of rules and parameters, especially insulation from other components.
2. You bring the rock to Mohamed, Mohamed never goes to the rock. (We give you fast transistor, make sure it stays that way after its planted!) “Some guys build their world around women, you’d better better build your world around this thing.”
3. It really isn’t clear if AMD’s failure was making slow transistors (Guru”s subtile assertions), RC constraints/Joule Heating (in the backend?)(my query), or a combination of both.
4. You build an an entire line of tools specifically orchastrated to construct the whole enchilada, it follows, you pretty much screwed the pooch (down the line) if you commited to those respective design parameters. (subtle tweaks ain’t gonna cut it)
ITK- Excellent link, they don’t sound/read like Patent Attorney’s
Thanks All
SPARKS
Scientia constantly talks out of his ass and "asset-smart" is just one more indication if not proof.
I didn’t know the gates required a different set of rules and proceedures regarding insulation....
Sparks, You'll often see the terms front end and back end thrown around with regards to the process. The front end includes all the steps required to build the transistors. The back end builds the interconnects.
The front end uses a lot of unique tools that are only seen once in the process flow in addition to the wet benches and implanters that wafers are introduced to multiple times in the flow.
The back end has a fairly limited number of tools and tool types. The wafers will loop through a nearly identical process flow (with the obvious exception of using a different mask for each metal layer) multiple times.
Front end and back end processing each have their own unique quirks and issues.
Keeping temps down along with a need for better conductivity (and yes, I do know that temp and conductivity are related) was the big driver for moving from Al to Cu interconnects, as I understand it. That was a huge shift for the industry with a number of technical issues to overcome since a Cu contamination incident in the front end can bring the whole factory to it's knees.
The next Atom?
http://www.eetimes.com/news/semi/showArticle.jhtml?articleID=212001342
What is interesting about this approach is Intel appears to be wanting not only to supply HW, but also provide service - this is key as HW becomes a commodity as you have a continuous revenue stream.
I think this could be the next big thing, it may be a bit early...but then again I thought the same thing about Atom!
BTW - Microsoft is blaming slower future revenue growth on the growth of netbooks. As ~25% of netbooks don't use MS and many use cheaper versions - they will see cannibalization effects that Intel doesn't.
This one just kills me. EETimes has an announcement for IBM's new 45nm SOI foundry.
According to the article the foundry will
fuel a new class of SOI designs
with
Singapore's Chartered Semiconductor Manufacturing Pte. Ltd. act[ing] as a ''second source'' foundry for IBM's 45-nm SOI offering.
But the real kicker is this comment from a supposedly informed publication. They state:
With SOI, IBM appears to have taken the lead in another technology. The company and its foundry partners have recently announced a high-k/metal-gate offering at the 32-nm node, putting the group ahead of its rivals in Taiwan. In comparison, Taiwan Semiconductor Manufacturing Co. Ltd. (TSMC) and United Microelectronics Corp. (UMC) separately will not offer a high-k/metal-gate solution until the 28-nm node.
I wonder what happened to all the processors that Intel has been producing with HK/MG for the past year? Or maybe we are just talking about the race for second place and I missed that part of the article.
Reading between the lines, this tells me that IBM and chartered don't have enough demand to fill their process lines. Not exactly a ringing endorsement of IBM's offerings. Especially when combined with comments I've seen earlier that customers might "skip" IBM's 45nm offering entirely.
Established foundry players are already cutting back expansion plans and now IBM wants to get a piece of the shrinking pie.
And the idea that foundry customers are going to be lining up for 32nm capacity is nothing short of delusional. Foundry customers are looking for a bargain and are seldom willing to pay for cutting edge tech that their designs don't require.
The next Atom?
Intel helped develop the Motion C5 clinical assistant about a year ago. On the 6th of this month, Panasonic announced their atom based version of a similar product.
I remember seeing one of the comments about the motion C5 product that said more about the prospect of the device than I ever could. One of the nurses who was using one said "It's too heavy, it's too hot, and you can't have it back."
But the service angle is something that I've heard rumblings about for a long time. Who wouldn't want to sell something that can generate a long term revenue stream. It sounds like maybe Intel found a model they thought they could make work for them. It will be interesting to see how this plays out.
ITK-An interesting thing happened to me, just today. I was sitting in my train in Penn Station wait to return home after a long day. Four minutes before the train was scheduled to pull out I got a call from my office asking me to go to 590 Madison Ave to start up two Liebert units in a newly constructed server room. (I’m willing to bet GURU knows where exactly where this is.) I left my train, got on the ‘E’ train, got off at the 5th Ave station, walked to 56th, looked up, and low and behold it was the IBM building.
There it was a tall black monolithic structure, almost supernatural in appearance, representing the apex of our technology. I felt small and insignificant, like one of the Apes in ‘2001 A Space Odyssey’. I had never been there before.
On the way up to the 31st floor I asked the elevator operator about IBM. He said IBM sold the building ten years ago, it was bought by a real estate organization, and space was leased to various companies of every sort.
The two A/C units had reversed rotation. I easily corrected the faults. The combination starter for the chiller water supply turned on without a hitch. I finished, got out of the building, turned around, and looked back.
The building seemed less significant and less imposing, somehow an empty shell of an era gone by. It was like an ancient monument to a long ago forgotten great civilization, which was abandoned because they could no longer afford to sustain its grandeur.
The servers were Dell Blades, running Microsoft Operating System, and powered by Intel Corporation Xeon Processors.
……and I own the fastest processor on the planet.
SPARKS
But the service angle is something that I've heard rumblings about for a long time. Who wouldn't want to sell something that can generate a long term revenue stream.
IBM has become very successful at this as now most of their revenue is in services, not HW. Though with this large a piece of pie looming out there I do wonder whether non-HW comapnies will try to hone in on this market as well:
- Micro-Hoo? (the eventual Microsoft-Yahoo merged company)
- Google?
- AMD and anyone they can convince to do this with them? (who also has a pile of cash)
- IBM?
- Then of course there are all the medical service companies.
It looks like Intel is trying to get on this train early (probably not a bad idea given that the competition will likely be pretty fierce)
Heh, I was looking at the model & stepping info in the CPU-Z screenshot, not the revision :). Guess I win the Dweeb-o'-the-Week award, which I hear is a Shanghai CPU.
2nd place is two Shanghais :).
Speaking of jokes, looks like UAEZone fanbois have just now discovered the Ci7 multi-GPU gaming benches, although most of them don't believe it. One even said that Deneb will "destroy" Nehalem because "its bandwidth is larger". Guess he is ignoring the K8 and K10 versions which also used IMC...
Also it appears Shanghai might be released Thursday - hope a lot of comparisons to Ci7 are done and then Robo puts up a new post for comments :).
Hey you financial whizzies... what does this mean?
"At chip maker Advanced Micro Devices Inc., the situation has gotten so extreme the company is planning a shareholder meeting to ask permission to reprice 99% of its outstanding options. AMD, whose stock price had fallen about 76% over the past 52 weeks to $3.16 a share at Friday's close, said this is necessary to prevent key employees from leaving.
"The vast majority of the historically granted stock options no longer are effective as incentives to motivate and retain employees," a company proxy statement said. The date of the meeting has yet to be set."
What does a company do when it "reprices options"? And how does this work to make the options valuable enough to remain an incentive to stay at the company?
PS- does anyone else have to stifle a chuckle when, even after all this time, they see Mac supporters hyping Intel processors?
How many extra helpings of Kool-Aid did the diehards need to choke down in order to get rid of the bitter aftertaste when Jobs first announced the deal with Intel a couple of years back? :)
What does a company do when it "reprices options"?
Apparently AMD views options as an employee ENTITLEMENT that must be positive... that is no longer an "option" at that point and you might as well just pay the employees a higher salary. The whole idea of an option is that if a company is successful, the option increases in value and everyone wins, if the stock falls and the company is not successful you shouldn't also win. The option is (supposed to be) a bonus.
Of course keeping things as options has the greatest benefit to AMD in terms of cash flow - instead of paying out additional cash via wages, AMD just prints some more stock... and the stockholders get the shaft AGAIN.
In terms of specifics, AMD will seek to reprice the options either at, below or near the current price so AMD employees see the option as a potential payday and therefor are more likely to stay with the company. If the options are at say $20 or $30, there is really no incentive to stick around and hope the stock price gets back above that level.
Once you leave a company there is a defined lifetime on the options - typically in the 90 day range. So if you have a lot of options and the stock is near the option price, you may be more inclined to stay with the company. The options also has a lifetime on it from when they are granted to an employee (can be on the order of 5-10 years). So if the stock needs to go up 5-10X to get the option positive and the option is expiring in the next couple of years, what are the chances of it paying off and what is the point of sticking around other than your base salary?
The irony is AMD is saying they will have a hard time retaining people because the stock price is so low. The irony comes in as AMD has basically ignored returning value to stockholders and running a profitable business. Some of the things were outside their control, but there is plenty they've done recently to screw stockholders which they are now saying is hurting their ability to keep people as all of the options are under water! They issued a ton of stock (to a certain middle eastern investor) which diluted the share price, they issued all those convertible notes. They went on this market share at all cost strategy which completely destroyed profitability and drove the stock price into the ground. And lets not even talk about the ATI purchase...
Now that senior folks have lots of options expiring, they suddenly CARE about the stock price after screwing stockholders for years. However, instead of trying to actually address the stock price they are saying let's just lower the bar on options... which potentially screws the stockholders yet AGAIN!
Had this company been run as business and not like Captain Ahab trying to get that pesky whale, the stock price would probably be higher and there would be less pressure on the options.
"The vast majority of the historically granted stock options no longer are effective as incentives to motivate and retain employees," a company proxy statement said.
This is more of a joke than you can imagine. They are laying off another 500 employees and think that incentives like stock options are going to retain their employees.
The PCB shop I worked for was pushing 1000 employees in my facility at the time I was hired. We were down to around 400ish when I left 3-1/2 years later. In particular, the engineering department had a total of 27 people in 2 departments when I was hired. When I left, we were down to 6. Those cuts were stretched out over the last 2-1/2 years I was there.
So I can speak from personal experience when I tell you that there is nothing that will break employee morale like an extended series of lay offs.
And AMD management has just sent the message that either they were incompetent and didn't know how many people they should have laid off in the first round, or that they intend to slowly drag this process out and nobody's job is safe.
According to Intel employees I've talked to Intel didn't do themselves any favors either when they did their layoffs. Everyone I've spoken to felt that the process was too long and drawn out. Employees just aren't very happy when they go home each night wondering if they will have a job in the morning.
The Sunlighter is designed the same is roughly too slippery and nigh get baffled
when walking off to purchase owl gifts. Do not worry if your cakes are
baked dim and grow fiddling acrimonious, because on toys for boys and pre-teens from multiple sources allows a lean that is more tailored yet offers great mixed bag.
quartet of the about of import with these limited and unique amorous talent ideas you
will maturate to trust this to be utterly truthful.
Here is my web site - owl kitchen decor
my site - owl kitchen decor
Post a Comment