Friday, September 08, 2006

AMD readies 65nm and 45nm

INQ reports that AMD's 65nm processors will be out next month. Furthermore, 45nm chips are being tested now. We know this already, because IBM+Chartered said a few days ago that they will have 45 chips in 2007 and AMD spends millions in co-development with IBM.

On the architectural front, K8L will be out in 2Q07. But I bet it will be slightly earlier than that.

Also, in the next three months, AMD will use another transistor, which will be much faster and lower power.

Do I need to repeat my projections for Intel again? No. I suppose.

51 Comments:

Blogger N4CR said...

http://www.charteredsemi.com/

Right there on the front page...

"Plus...

* Technology Roadmap to 45nm"


Nice... nice!

7:45 AM, September 08, 2006  
Anonymous Anonymous said...

hi sharikou,
could you give some opinions on how much improvement going to 65nm will give?
tia

7:48 AM, September 08, 2006  
Anonymous enumae said...

You have stated in the past that The Inq is not a reliable source, but I am assuming that only applies to a negative comment about AMD, I guess if its positive its just fine to use them as a source... thats sad, really sad.

Sharikou said...

"INQ reports that AMD's 65nm processors will be out next month."

If they were not planned until December, will this be a paper launch?

"Furthermore, 45nm chips are being tested now."

How long before we see some 45nm from AMD?

How long before they convert Fab 36 to 45nm, or was it on the June roadmap?

"On the architectural front, K8L will be out in 2Q07."

Server and Workstation, no mention of desktop, you should clarify that.

As it has already been stated AMD in the red for years was able to survive, and now you think Intel will just go away after a some bad quarters...at least we know you can beat your comments into the ground.

8:01 AM, September 08, 2006  
Blogger Pop Catalin Sever said...

Great news!
I'm comfident that AMD tranzition to 65nm will be a smooth and seamless one just like before.

And I think AMD will regain the performance crown once they are ready to do so from a capacity and technology point of view and not before becouse AMD's current 90nm proceess seems to be highly efective and they should realy squize as much as posible from it to maximize their profits before going mass 65 nm.

8:21 AM, September 08, 2006  
Anonymous Anonymous said...

AMD began 65nm production in June. That means than this shipment will be made by rev.G CPUs with 90/65nm Enhanced strain, advanced silicide built transistors. Three months later in late December /early Jan there will be 8th generation of transistors made in 90/65nm Embedded Si-Ge technology.
And finally three months later there will be CPUs made from pure 65nm Enhanced Si-Ge transistors.

I guess that 9th generation technology is probably for K8L type CPUs?

Is Bulldozer 65nm Enh. Si-Ge also?

x74

11:06 AM, September 08, 2006  
Anonymous Edward said...

"On the architectural front, K8L will be out in 2Q07. But I bet it will be slightly earlier than that."

K8L will be out as early as AMD's 65nm process yield matures (i.e. reaches about the same level of its 90nm yield). It will be too expensive for AMD to manufacture K8L on 90nm; the resulting chips will probably consume too much power, too. So IMO before AMD's 65nm process matures and stablizes, there is little its design team could do to tweak/improve the physical design even if K8L's microarchitecture is all done and ready.

It's certainly a good news that AMD is releasing 65nm chips earlier than expected (which I believe was the end of this year).

11:23 AM, September 08, 2006  
Anonymous Anonymous said...

You can leave your ridiculous projections to yourself:)

"Furthermore, 45nm chips are being tested now. We know this already, because IBM+Chartered said a few days ago that they will have 45 chips in 2007 and AMD spends millions in co-development with IBM."

That is an assumption, not fact. IBM and friends tested 45nm silicon, not CPUs.

11:41 AM, September 08, 2006  
Anonymous Anonymous said...

Wow even earlier then I even thought. See AMD lets little to none leak out when it involves info. Ppl can hear one thing and say another. But what it really comes down to is news like this that proves prodictions wrong.

65nm on my bday. K8L at the end of the year.

2:46 PM, September 08, 2006  
Anonymous Anonymous said...

Re; the comments about Chartered/IBM/AMD at 45nm,

From and Aug. 29th post under a different thread:

"http://www.intel.com/pressroom/archive/releases/20060125comp.htm

Where the release on the Inq (regarding IBM/Chartered/AMD's announcement) stated verification of library cells and embedded memory now, Intel announced functional 153Mbit SRAM + logic circuit verification 7 months ago. So much for a "leap ahead" on the 45nm front.
"

In other words, the only thing IBM/Chartered/AMD are claiming now is functional verification in silicon of library cells and blocks of memory. This is a great development, but it does not put them ahead on the 45nm node front either.

3:41 PM, September 08, 2006  
Anonymous Anonymous said...

"Furthermore, 45nm chips are being tested now."

Actually these are "test chips" not real chips; to put this in perspective Intel publicly announces this in I think Jan of this year (which means they probably had been doing it for some time before that).

5:57 PM, September 08, 2006  
Anonymous Anonymous said...

http://img.oc.com.tw/b60799/20069511501413494030192.jpg

65nm chip! Been out since 1st production of May or June. Final chips will be at much faster clocked and capable of higher ocs because of its vamped construction.

Bet the Intel people are eating their words. "AMD won't have 65nm until 2007. Du...hur...hur..." Ta whatever dudes! It comes out in oct.

7:52 PM, September 08, 2006  
Anonymous Anonymous said...

Which process is better? Intel's 90nm or AMD's 90nm? Why Intel is not using SOI (Sillicon on Insulator)?

Which process will be better? Intel's 65nm or AMD (IBM/alliance) 65nm is better? Intel is still not using SOI for 65nm right? And would the bigger static leakage current at 65nm really need SOI?

-Longan-

9:00 PM, September 08, 2006  
Anonymous BET said...

"On the architectural front, K8L will be out in 2Q07. But I bet it will be slightly earlier than that."

You BET?

OK, this would be interesting.

Do you bet that if K8L is not out before 2Q07, you would stop this blog?

Please, please, please; take this bet.

3:17 AM, September 09, 2006  
Anonymous Anonymous said...

I hope IBM quikly moves the xbox 360's Xenon CPU from 90nm to 65nm fabrication.

I keep coming back to this blog just to see what's going on, though I often force myself to stay away because of the daily infernos.

The blog certainly provokes a lot of very interesting technical, strategic and economics related debate (at least sometimes).

It would be nice if once in a while there could be the odd Playstation 3 vs xbox 360 debate.

e.g. discussions on cpu, gpu, bus, memory, flops, 90nm/65nm, wattage, heat, overall performance, etc.

(Oh, and pondering which console manufacturer will most likely BK in X quarters- kidding:))

I know this is a PC related blog, but both consoles have 64-bit processors from IBM (pity they're both IBM CPUs: the debate would be much more livelier if it was PS3 with IBM versus x360 with Intel; or, even better, PS3 with AMD versus x360 with Intel).

7:35 AM, September 09, 2006  
Anonymous Anonymous said...

Is this funny or what some monkey said "The blog certainly provokes a lot of very interesting technical, strategic and economics related debate (at least sometimes)."

What strategy what related debate..

Yeah intel BK in 6 quarters or need to cut 80% of staff, oh technical.. its the CPU that causes the computer to catch fire..

Ignore physics, ignore data, ignore reality.

For this monkey, more like coming here to get away from reality.

LOL

9:06 AM, September 09, 2006  
Anonymous Anonymous said...

Some one asked "Which process is better? Intel's 90nm or AMD's 90nm? Why Intel is not using SOI (Sillicon on Insulator)?"

No question AMD 90nm is inferior by every performance/power metric to INTEL's 90nm

AMD or should I say IBM's prostituted process at 65nm is also inferior to INTEL's process in every metric availbel to measure.

Go check IEDM technical digests from the past few years where each company brags about their process. What I can't figure is why IBM/AMD even show up... their process is slower and inferior.

Like showing up to run a 100 meter dash and coming in a distand second by 1" but still proudly saying my process is the best and SOI and DSL is the way to go.. but when you look at the only thing that matters they are a distant second.

Any questions you flies?

The Doctor

9:09 AM, September 09, 2006  
Anonymous muziqaz said...

To that anonymous Doctor:

if it is inferior, how come intels 65nm mobile cpu consumes almost the same amount of power(Watts) as AMDs 90nm turions?
or conroe(65nm) compared with a64(90nm).
and in regard to performance, will see in half a year. :)

10:47 AM, September 09, 2006  
Anonymous Edward said...

"Go check IEDM technical digests from the past few years where each company brags about their process. What I can't figure is why IBM/AMD even show up... their process is slower and inferior."

Slower as in clock rate or time to market? Better as in yield, power consumption, cost, or what?

10:53 AM, September 09, 2006  
Anonymous Anonymous said...

SOI VS standard. YOu must be hi on drugs or something. You don't even know why AMD can keep up aready with 90nm. Your stupid, don't even know fab tech at all. Lets teach you something about fab processes.

CMOS is primitive to SOI. The only reason why AMD could ever out perform intel in everything agenst their pentiums was because of SOI.

CMOS: Cheap, make twice in half the time, hot, slower silicon, lots of transistor leakage, not very relyable, can't take over max temp for very long, no clock speed improvment. Runs pure mhz.

Standrad AMD 90nm SOI: Expencive, takes twice as long to make, cooler, no where near the transistor leakage of standard CMOS does, makes a more relyable processor, can take heat better for longer, makes transistors faster allowing improved mhz speeds. Why AMD was 1200mhz faster then a pentium 4. Runs faster cycles per clock.

Perfected 90nm REV F: Uses enhanced CMOS, SOI, DSL, and Cu to improve wattage useage and max temps, also improves anti-transitor leakage by 35% giving almost no leakage at all. This is a very complex process. Allows AMD to bring 35watt dual cores upto X2 4600 so far. And on par with 65nm conroe. 90nm AMD as good as it gets agenst 65nm Intel.

Conroes enhanced CMOS: No big improvments from CMOS besides how the arc uses it to get lower power consumption. Only slight improvments to max case temp and temps in gental by how the processor manages itself. Only minor transistor anti-leakage improvments. Still lots of leakage even on 65nm. Still crapy cheap old ways of making dies is why, they only have 2 fab processes cus they are inferior. 65watts.

Perfected 65nm Rev G: Silicon Germanium (e-SiGe) with Dual Stress Liner (DSL) and Stress Memorization technology (SMT) on Silicon-On-Insulator (SOI) with advanced CMOS and Cu to top off with. This is a die that will improve transistor performance by 40% and fully eleminate transistor leakage and improve heat and wattage. This will also allow for higher clock speeds and AMD to get into a Ghz war again. This type of die allows AMD to make more in less time and cheaper but far more complex then 90nm tech from them. This is far superior then Intels CMOS Cu only fab processes.

AMD is superior in fab processes so don't get them confused with Intels. They could only dream of those fab processes. You prob don't even understand what I'm talking about and are going to say something like whatever Intel is the best or some fag remark like that, so a fanboy statment. Well I don't give a rats ass, this is real fab processes AMD has and Intel doesn't.

Don't make crap up just because you figure one company is better then another. Its by very little trust me. Intel can make a faster cpu by 10% (Not for long) AMD has better fab processes so they are equal in their own ways. How about that?

11:34 AM, September 09, 2006  
Anonymous Anonymous said...

You arn't no doc thats for sure pal. IBM invented 65nm. Without IBM you wouldn't have Intels at 65nm. IBM is bigger then anybody for your info, they are worth 80 billion. Intel is only a fraction of that.

Shows how much the doc knows huh! (rolls eyes) Plus IBM invented 45nm and so on. The POWER6 will destroy CONroe. Don't get us started flaming agenst your "Intel is superior BS talk" So damn fboy!!! So damn cocky just cus intel brings out a 10% faster cpu then AMD you all go nuts. Seriously its all wishful thinking. Even K8L by die shot specs is more advanced then conroe. Give it a rest. Conroe isn't going to be decently better for that much longer.

CONroe is only superior in your minds. I have a X2 3800+ and a E6400. No difference in the real world. I'm not a fan of any company is why I have both. AMD has the better chipsets and features as well as bandwidth. COnroe isn't anything to go god over. Acts like the damn thing is god or something to these intel ppl. What do yya do have a shrine over it or something too? Warship the CPU GOD CONroe... I'm sure these fokes are crazy enough to do it too.

Its only a cpu... go get a girlfriend and a room and stop humping your damn cpus... sick bastards.

11:47 AM, September 09, 2006  
Anonymous Anonymous said...

[b]Anonynous said...[b]
[i]Go check IEDM technical digests from the past few years where each company brags about their process. What I can't figure is why IBM/AMD even show up... their process is slower and inferior.[i]

By your own admission that is OLD data, MORON.You sound just like JumpingJack at THG.How's it going, JackAss? Hey, Sharikou, this guy claims to have 2, yes 2, PHds but won't provide proof and has the cheek to doubt yours.Do you think he claims 2 because he has multiple
personalities? No, can't be, he has ABSOLUTELY NO personality!

2:36 PM, September 09, 2006  
Anonymous Anonymous said...

Wow some fool said "Lets teach you something about fab processes."

Okay here you go?

SOI what do you get with fully depleted SOI?

1) Expensive processed substrate with risk for yield killers.

2) Expensive substrates with limited volume availability

3) For fully depleted SOI you need very good control of the substrate or you will get variable Vts.

4) Only advantage over bulk is S/D to bulk leakage. SOI doesn NOT help gate oxide leakage nor Source-Drain subthreshold leakage which are the dominate mechisms for off transistors. By the way there are simple ways to minimize both S/D leakage and capacitance that require no exotic/espensive substrates.



The fact of the matter the idiot who posted last knows nothing about technology but is spewing the marketing stuff...

Tell me fundamentally why SOI is lower power then bulk CMOS.

Tell me why the cost of the substrate and the extra process to get the SIMOX to work is worth it over the simple strained approach INTEL has chosen?


You don't know shit... lets argue the facts.. SOI requires far more complexity.. Hey is that why they are late? I would suggest you bone up on some technology papers presnted on SOI vs Bulk from some very direct comparison... not some maketing hype you silly moron.

Your typing is very entertaining.. now go back and study.. I hope you aren't a technologist.. or pitty your manager, your group and your company..

To the guy who thinks IBM is bigger.. Tell me how many fabs does IBM has? how big is microelectronics? WHo has the fastest transistors at 90nm, 65nm, 45nm? Are you stupid or what? IBM gets all of its profits and most of its revenue from everything but silicon... You are truely stupid.. Don't waste your time or energy trying to post lies. Its funny to see the mad like the PhD pretender, then ever more funny the stupid who think they are for real... I know you work for IBM if you want to tell me you have the K8L die shots as they haven't been released into the wild.. too bad you don't have a technology nor the fabs to do volume like Samsung, TSMC, INTEL.

It must suck to work at IBM micro electronics with your long history but now are relegated to prostituting yourself to foreign companies and second rate AMD

3:18 PM, September 09, 2006  
Anonymous enumae said...

Anonymous said...

"IBM invented 65nm."

Could you link to it?

"Plus IBM invented 45nm and so on."

Could you link to it?

"So damn cocky just cus intel brings out a 10% faster cpu then AMD you all go nuts."

You do remember thats just in gaming, last time I looked (depending upon review) it was closer to 20-30% in other applications.

"Conroe isn't going to be decently better for that much longer."

So you have a relesae date for a "desktop" K8L?

Another anonymous said...

"...This is a very complex process. Allows AMD to bring 35watt dual cores upto X2 4600 so far. And on par with 65nm conroe. 90nm AMD as good as it gets agenst 65nm Intel."

Just so you know the X2 4600+ is 65W.

Are you taking into account the fact that AMD doesn't have as much cache?

I understand they do not need it.

Looking at the release info. about the Athlon 64 X2 5200+ the TDP went from 65W to 89W, and thats only using 2MB (1 MB per core) of cache...thats 24W just by adding 1 MB of cache.

Intel uses 4MB (2 MB per core) of cache and is at 65W, so your comparison of 65nm to 90nm is useless.

65nm to 65nm will be the test.

3:39 PM, September 09, 2006  
Anonymous Edward said...

Lets look at the results and stop this chanting from people sounding like hurt employees. Whose 90nm chips are cheaper, faster, and cooler? AMD's or Intel's?

Nobody cares who invented what or which technology seems better on paper. Just look at the results - the answer to above proves itself.

8:55 PM, September 09, 2006  
Anonymous Anonymous said...

Edward: "K8L will be out as early as AMD's 65nm process yield matures"

I thought you (or maybe it was Sharikou) who said AMD is at "mature Yield" when they start on a new node; you know due to those APM3.0 magic beans.

As AMD is already shipping 65nm product to the channel; their yields must be mature due to your (or maybe it was Sharikou's) excellent fab process technology analysis.

Therefore shouldn't K8L be shipping now? Or is it just possible that either:
A) AMD is not quite at mature yield at the beginning of tech node transition.
B) Maybe there are other limiters on K8L schedule - you know I hear some CPU maufacturers do this whole reliability/burn-in validation and may actual iterate the circuit designs from the initial tapeouts to fix issues.

Although I am pretty sure it was you who was saying "AMD only transitions to the next technology when they are at mature yields" and then pointed incorrectly to some analyst day foils?

8:45 PM, September 10, 2006  
Anonymous Anonymous said...

"Slower as in clock rate or time to market? Better as in yield, power consumption, cost, or what?"

As in Ion/Ioff performance which is the only way to compare different company's technologies apples to apples (this takes out impact of chip design)

Ion/Ioff is a measure of transistor performance. The variables you describe above are not a reasoonable way to compare process technologies as those variables are also heavily dependent on specific chip design.

8:49 PM, September 10, 2006  
Anonymous Anonymous said...

Lets look at the results and stop this chanting from people sounding like hurt employees. Whose 90nm chips are cheaper, faster, and cooler? AMD's or Intel's?

You talking desktop or mobile? How about 65nm Core 2 vs 65nm Athlon (As you can see my questions show how heavily impactful chip design is). The people above were saying AMD had a superior PROCESS TECHNOLOGY on a given node - normal scientific benchmarks prove this wrong.

That said, I completely agree with you - it really doesn't matter how fast a specific transistor or whether the chip usess Si or SOI, Cu, low K, DSL, etc... what matters is the final chip performance. Of course the same can be said of the people who say "Kentsfield is just glued together and not native quad" - really doesn't matter if the performance is there (if the perfomance is not there than that is a whole 'nother story)

9:00 PM, September 10, 2006  
Anonymous Edward said...

"I thought you (or maybe it was Sharikou) who said AMD is at "mature Yield" when they start on a new node; you know due to those APM3.0 magic beans.""

I really hate people with reading and logic-thinking problem distorting what I said. I said AMD releases product at close to mature yield. Plus, do you really know the difference between "as early as" and "as soon as"? If I said K8L will be out as early as AMD's 65nm process matures, that means K8L will be out after that. The process has to stablize, then the design team can tweak/improve the design on that node.

Please stop acting like a smart ass when all you have are half-baked logic and short wits. Thanks.

11:31 PM, September 10, 2006  
Anonymous Edward said...

"As in Ion/Ioff performance which is the only way to compare different company's technologies apples to apples (this takes out impact of chip design)"

1) Do you have any proof showing Intel's 90nm Ion/Ioff better than IBM's?

2) In your opinion, yield and speed (capacitance etc.) don't reflect transistor performance, or they aren't apple-to-apple comparison?

Me thinking: you've shown nothing, and your "single metric way" of comparison is lame.

11:34 PM, September 10, 2006  
Anonymous Edward said...

"The people above were saying AMD had a superior PROCESS TECHNOLOGY on a given node - normal scientific benchmarks prove this wrong.

... it really doesn't matter how fast a specific transistor or whether the chip usess Si or SOI, Cu, low K, DSL, etc... what matters is the final chip performance. "


You are totally confused, to such degree that I was wondering whether it was worth to reply you. Well, I guess I could be patient at times. ;-)

Process technology is mostly orthogonal to the chip logic design. That's how the fab foundries work and why they work so well. When you compare process technologies from two companies, you should not take into account their chip logic.

There is process technology and performance; there is microarchitecture design and processor core. When we say "AMD readies 65nm and 45nm," we do not care whether it uses those processes to make K8 or K10 or even Core 2. The final CPU benchmark (most of those are not scientific, BTW) really doesn't matter.

11:42 PM, September 10, 2006  
Anonymous Anonymous said...

Alright, I got my answers, thanks. The guy who posted on
11:34 AM, September 09, 2006
gave complete facts and convincing answer. Thank you.

-Longan-

I have the next question: If Intel does not have SOI in their 90nm and 65nm, would they have a major disadvantage in battery life?

In my oppinion, most of the time the laptop is at idle and static current in the CPU will really kill the battery stand-by time?

I would think the lack of SOI will start killing Intel at 65nm for laptop.

11:51 PM, September 10, 2006  
Anonymous Anonymous said...

Of course the same can be said of the people who say "Kentsfield is just glued together and not native quad" - really doesn't matter if the performance is there (if the perfomance is not there than that is a whole 'nother story)

But it can never be viewed as a technology advance. Its not microprocessor design advances. It’s just a packing trick.
Unfortunately it seams that manufacturing capabilities are ahead of time (microprocessor design/innovation halted).
Nothing better can be done by Intel, AMD, IBM, SUN than just copy paste (or mirror the processor if you prefer)

PROS: New performance levels, ahead of time.

CONS:
-Higher power consuming.
-Most user’s (Home and most enterprises) don’t use the applications that get benefits out of it.
-Applications that it does help aren’t used regularly. Encode audio do you really do that, 8 hours a day? Encode video, do you really do that all the time?
-Expensive.
-Brings old problems. Heat level that seem disappeared on the last processor “incarnation” (PD8xx, PD9xx).

3:26 AM, September 11, 2006  
Anonymous Anonymous said...

To the guy who thinks IBM is bigger..

More technology advance doesn’t mean bigger. He did mean invented things...

IBM is behind almost every PC component technology advances. They invented the PC!
Intel and Microsoft should give a big thank to them. Almost every company should be paying royalties to them. Why they don’t that’s the part that I don’t know. They forget/forgot to patent the things?

http://www-03.ibm.com/ibm/history/history/history_intro.html


IBM delivers the world's first copper-based microprocessors, including a PowerPC 740/750 operating on 400MHz. Microprocessors that incorporate copper wiring boost chip performance by about one-third.

IBM scientists announce CMOS 7S, a breakthrough semiconductor manufacturing process that uses copper instead of aluminum to link transistors in chips, the culmination of 30 years of experimentation and inquiry. With copper technology, which conducts electricity 40 percent more efficiently than aluminum, semiconductor devices like microprocessors and memory chips can be more powerful and more energy efficient.

IBM becomes the first company to introduce silicon germanium chip-making technology into mainstream manufacturing

IBM announces the world's highest capacity desktop OPC disk drive. The 16.8-gigabyte drive incorporates breakthrough technology called giant magnetoresistive (GMR) heads. No bigger than the head of a pin, the GMR head is the world's most sensitive sensor for reading and writing computer data on magnetic disks.

During the year, IBM sets repeated records in hard disk drive storage densities

the Microdrive is introduced as the world's smallest and lightest hard disk drive

3:56 AM, September 11, 2006  
Anonymous Dr. Yield, PhD, MBA said...

IBM delivers the world's first copper-based microprocessors, including a PowerPC 740/750 operating on 400MHz. Microprocessors that incorporate copper wiring boost chip performance by about one-third.


No one disputes that IBM brought Cu to production first. IBM is clearly a technical powerhouse in the process department. AMD has had some blockbusters in design in recent years, and their factory management approach clearly has given them some benefits when it comes to factory agility. That said, Intel's forte has been, and continues to be, low cost, high volume, high performance manufacturing. Intel could have used Cu for their 0.18um process, but chose not to. Why? Not because they couldn't pull it off in time, but rather they were able to achieve equivalent or better circuit performance as compared to the competition with the substantially lower risk and cost of aluminum. If your processors are competitive but cost less to make, why take the yield/complexity risks on if you can delay them another node? Bragging rights don't help the bottom line- 0.18um X86 processors from AMD and Intel were in the same ballpark in terms of performance and power consumption, but Intel could produce die cheaper because of yield and process simplification advantages of sticking with Aluminum.

You can argue the same for SOI- yes, SOI does offer some electrical advantages over bulk Si. But they come at a price- design practices need to change, simulation models need to reflect the substrate differences, substrate costs are substantially higer, process complexity increases and therefore yields decrease... all of these things have cost associated with them that can not be offset with "bragging rights" about who has the biggest technical you-know-what. They have to be offset with the ability to differentiate your product enough to generate higher ASPs and get a decent margin on the more expensive to produce part (compared to the same company making it without SOI). Intel thinks they can be competitive without for 90 and 65nm node. AMD has it designed in. So what? It's all about the performance in the end- which process is faster for your application, or which mobile platform has better battery life. None of the rest matters to the end user.

12:39 PM, September 11, 2006  
Anonymous Anonymous said...

"1) Do you have any proof showing Intel's 90nm Ion/Ioff better than IBM's?"

IEDM published reports - they do this every year; it also includes 130nm and 65nm performance too... here is just one link (I forget which nodes this has):

http://www.realworldtech.com/page.cfm?ArticleID=RWT123005001504


"2) In your opinion, yield and speed (capacitance etc.) don't reflect transistor performance, or they aren't apple-to-apple comparison?"

These matter but aren't a measure of specific technology node performance ONLY - both yield and speed is significantly impacted by chip design and size - if you have 2 radically different die sizes how can you compare process technology capability vs yield?. For example Intel's yield are different on Core Duo (mobile) and Prescott and Core2 - but they all use the same process technology!

If you want to compare AMD's process technology to Intel's unless you are using the same chip design you have 2 variables - the only way to do a valid comarison is to break it down to the transistor level.

Ion/Ioff is just one of the main metrics to compare transistor performance - perhaps you just don't have the transistor background to understand this? Also in the article linked, CV/I is mentioned, this is another common metric used which is normalized way of looking at transistor switching speed (although I don't think the article I refernced has data on this)

3:05 PM, September 11, 2006  
Anonymous Edward said...

"If your processors are competitive but cost less to make, why take the yield/complexity risks on if you can delay them another node?"

I really have a different view, though. I believe it would be better for Intel the company to have switched to Cu, SOI, IMC, etc., as early as possible. However, Intel chosed not to. Why? Because those are new technologies invented elsewhere; new technologies mean new teams and new people, and why would Intel's current engineers & management choose someone else to challenge his own status inside the company, unless it becomes absolutely necessary?

At the very least, they need the time to learn and mater those new stuff themselves, before they let others come into the company and succeed.

3:41 PM, September 11, 2006  
Anonymous Anonymous said...

Yo PHD pretender ...
Print or discard.... it'll determine if I again bother to add real tidbits so you might educate yourself.

1) Bottom line there is no question INTEL process is superior at 130nm and below based on the cleanest apple to apple comparisons like Ion/ioff for transistors, SRAM sell size etc.

2) IBM may have been first in CMOS # but in the end it went into some irrelevanat boutique process that was never manufacturing in any high volume. Like the space shuttle or putting man on the moon if you can't put it into a high volume money making part then so what. INTEL has done that better then AMD/IBM or any other company. Nobody produces as many logic transistors / day then INTEL.

3) Nobody makes more money then INTEL producing those logic transistors.

4) Technology aside, if you excues the Netbust failure. INTEL with PentiumIII and with Dothan, Yonah and COre2 have leading edge designs. There was no question AMD had a winner.. to bad they didn't have the capacity nor silicon technology in place to crush INTEL in 2002-2004. They missed the window and its gone forever.

5) SiGe that IBM revolutionainzed was for HBT. DOn't confuse this with their bad bet in bi-axial strain CMOS they did.

6) Cu interconnect, so what.. first but didn't make any money

Bottom line what company rules, has the most fabs, and makes the most money, and has all the benchmarks..

IBM and its johns are in Fiskhkill sharing development and everything else. INTEL goes it alone in Hillsboro, Samsung in Korea, TI in Dallas, and TSMC in Taiwan. Who do you think is operating from the point of strength, the prostitute in Skanville or the ones going it alone?

THe company the Phd Pretender says that will go BK in 5 quarters is doing the best.. its so obvious.

5:29 PM, September 11, 2006  
Anonymous Edward said...

"Ion/Ioff is just one of the main metrics to compare transistor performance - perhaps you just don't have the transistor background to understand this? Also in the article linked, CV/I is mentioned, this is another common metric used which is normalized way of looking at transistor switching speed (although I don't think the article I refernced has data on this)"

Your referenced article, being a good undergraduate introductory to modern transistor properties, failed to back up your claim that Intel's process being superior to AMD's.

While IBM's Ioff seem higher than Intel's (their Ion's are about the same level), it's more likely due to the self-heating property of SOI wafer. OTOH, using SOI cuts junction leakage to zero, thus actually help leakage under the increased Ioff.

All-in-all, it proves my previous point: using Ion/Ioff as the sole metric of transistor performance is just lame. You should've got the idea that transistors could not be evaluated by one number in your first year of college semiconductor class. Perhaps you purposely forget those lessons for the sake of this argument?

Besides, when I said "speed," I didn't mean chip clockrate, but FO4 delay. From your referenced article, you can read CV/Isat is actually the 0th order process-neutral metrics, and they showed that by taking ratio of FO4-delay and CV/Isat to be roughly constant across node generation.

In summary, when you compare transistor performance, in the end, there are really two things: speed and power. Both are affected by a number of things: volage, capacitance, current, leakage. On top of that there's yield and cost. Using Ion & Ioff to "prove" Intel's process being superior is not only lame but likely wrong!

5:39 PM, September 11, 2006  
Anonymous Anonymous said...

Let the Dick comparison continue..

Corrected for self heating which is easy.. go tell the world based on VLSI and IEDM articles who has the best Ion/Ioff curves, CV/I, and SRAM size. Those are the cleanest metrics for raw process capability. PSS... Ion/Ioff are generally at the same stated voltage for same process node, or easily corrected. All companies choose similar voltages so they can show their manhood or in the cas of IBM/AMD how lacking they are.


If someone wants to argue about Yields its the worst kept secret. Go take the total CPU marketshare and divide it up by the number of fabs and make a guess. I think with INTEL having 3 factories running 65nm for the better part of a full year and AMD just starting it would be funny for anyone to article that AMD/IBM is ahead of INTEL for yield learning. Anybody who argues that AMD/IBM is ramping only when Yield is umm "mature" knows nothing about how you drive yield to world class "mature" levels.

Keep the drivel going.. it only shows you are grasping...

It sucks to be second with second class resources. Maybe teach your children to study hard to get the best grades and impress interviewers so you can get a job at a first class place.

9:06 PM, September 11, 2006  
Anonymous Edward said...

"All companies choose similar voltages so they can show their manhood or in the cas of IBM/AMD how lacking they are."

And how lacking they are? As lacking as your words are? What the h*ll were you saying, anyway? Intel's 90nm process superior to IBM/AMD's? Bulk Si superior to SOI?

"If someone wants to argue about Yields its the worst kept secret. Go take the total CPU marketshare and divide it up by the number of fabs and make a guess."

And you call the thing you have to guess a worst-kept secret? You don't have exact product mix, you don't have exact number of wafers per run and number of dies per wafer, tell me how you guess the yield (of what?) by simply dividing market share by number of fabs?

"Anybody who argues that AMD/IBM is ramping only when Yield is umm "mature" knows nothing about how you drive yield to world class "mature" levels."

Again another person with reading disability. Who said AMD/IBM is "ramping" only when yield is mature? I think most people agree that AMD releases a product when its yield is "close to" mature level, at least closer than Intel when it releases its chips. Go argue with AMD about that, or admit you can't read nor think straight - and you better not write nor repeat others before you fix that.

"Maybe teach your children to study hard to get the best grades and impress interviewers so you can get a job at a first class place."

Maybe teach your children how to read properly and don't count on his instinctive manhood (or lack of) to make meaningless arguments.

We know how much a "first-class" place Intel is. After all, it must be pretty "first-class" to let its 10k workforce beaten, in product performance, by a rival 1/10th of its size for 3 years. Maybe that's how you expect your children to do? I praise your courage.

1:13 AM, September 12, 2006  
Anonymous Anonymous said...

Edward.... Self-heating - nice regurgitation; would this be with DC or with AC applied? The 65nm data was corrected for this effect by applying AC! (of course that's just my undergraduate learning speaking)

It is nice to see you actually read the article though...

Out of curiosity, what contribution is junction leakage to Ioff vs say subthreshold and gate leakage? For 90nm and beyond how would these 3 rank in overall impact to Ioff (Hint: junction leakage, which is what SOI addresses is a distant 3rd)

By the way good answer to the the idiot who posted before this. He also might actually consider factoring in WSPW (or WSPM) rather than just dividing by # of fabs...unless he/she thinks every fab is sized the same!

10:59 AM, September 12, 2006  
Anonymous Anonymous said...

"I think most people agree that AMD releases a product when its yield is "close to" mature level, at least closer than Intel when it releases its chips"

You have ZERO evidence to support this statement.

11:01 AM, September 12, 2006  
Anonymous Edward said...

"You have ZERO evidence to support this statement."

No I don't, because I don't work at AMD, and AFAIK product yield is one of the most guarded information of a fab.

Have you read my comment in full at all? Go argue this with AMD's Henri Richard, or call him a lier, whatever makes yourself feel better.

12:13 PM, September 12, 2006  
Anonymous Edward said...

"Self-heating - nice regurgitation; would this be with DC or with AC applied? The 65nm data was corrected for this effect by applying AC! (of course that's just my undergraduate learning speaking)"

AFAIK, self-heating affects Ion in DC operation. As the paper 10.5 in IEDM 2005 showed, after it's corrected, the Ion/Ioff of IBM's SOI matches the that of Intel's bulk Si.

"(Hint: junction leakage, which is what SOI addresses is a distant 3rd)"

Junction leakage is not the main thing tht SOI addresses. Junction capacitance, reverse body effect, and short-channel effect when scaling are.

1:06 PM, September 12, 2006  
Anonymous Anonymous said...

"As the paper 10.5 in IEDM 2005 showed, after it's corrected, the Ion/Ioff of IBM's SOI matches the that of Intel's bulk Si."

Not really a math whiz but....

1259/200 is NOT roughly equal to 1210/100 (65nm NMOS, when corrected with AC)

And 735/200 is NOT ROUGHLY equal to 710/100 (not sure if this is AC corrected or not, but it is 2X worse)

"using SOI cuts junction leakage to zero, thus actually help leakage under the increased Ioff."

You were the one stating SOI helps Ioff - my point was (and is) the impact of SOI on Ioff (LEAKAGE) on todays process technologies is small as both gate leakage and subthreshold leakage have a much more significant impact on Ioff (and will grow faster with scaling).

Of course SOI helps in other areas but there are alternative technologies to address the issues you have listed. (and looking at Ion/Ioff metrics they appaer to be at a minimum, competitive to SOI).

Not to mention SOI susbtrates add ~10-15% to overall fab cost per wafer production costs and will likely grow when transitioning to SGOI and moving to thinner active Si layers in order to achieved FD as opposed to the partially depleted technology in use today.

"Have you read my comment in full at all? Go argue this with AMD's Henri Richard, or call him a lier, whatever makes yourself feel better."

He has never stated what you stated above - you have twisted his words (go look back at what he has stated at analyst day). He has stated AMD has been getting to mature yields more quickly with each successive technology node (as has Intel, TSMC and many other IC manufacturers). He never stated AMD only release product when it is at mature yield or AMD only ramps when yield is mature this is just Sharikou's and your interpertations of his words which has been said so many times now, people think it is what Richard stated.

2:02 PM, September 12, 2006  
Anonymous Anonymous said...

Has anyone noticed when you look at IBM/AMD/etc SRAM cell size reports it only shows a ~35% area reduction as opposed to the expected ~50% going from 90nm to 65mn? (Also interesting Intel was showing ~43% reduction so they are not getting full 50% reduction too)

If true, it's something that SHarikou will need to factor into his capacity analysis which assumes double die/wafer from the transition.

And before someone asks I think expected improvement in logic portion of the die is usually similar to SRAM size scaling (if anyone out there is familiar with this please correct me if I'm wrong)

8:27 PM, September 12, 2006  
Anonymous Edward said...

"Not really a math whiz but....

1259/200 is NOT roughly equal to 1210/100 (65nm NMOS, when corrected with AC)"


Please... you're going off-chart with your high-school math. The '/' in "Ion/Ioff" is a slash, not a divison! Read the article I was talking about and you'll find that Ion to Ioff ratio is not constant for one particular process, less could it be used to compare two different process technologies.

"He never stated AMD only release product when it is at mature yield or AMD only ramps when yield is mature this is just Sharikou's and your interpertations"

Gosh... people can really be stubborn on their wrong ideas. As a matter of fact, I NEVER said "AMD only release product when it is at mature yield" or "AMD only ramps when yield is mature" (I always said "close to"). Besides, Henri Richard did said that he believes, after releasing chips on a process node, AMD ramps to mature yield faster than Intel does.

Well, the DigiTimes now requires you to pay for their interview articles with Mr. Richard. But no worry, because Thomas Sonderman, director of APM technology at AMD said right here exactly what you need to read. Just search for the phrase "at mature yield" and read the paragraphs (there are 4 matches, read them all)!

Now stop whining, will you?

12:23 AM, September 13, 2006  
Anonymous Anonymous said...

Corporate Ethics
All this talk about process technology is an attempt by Intel fanboys to deflect attention from Intel's inferior platform and the impending court case against it.
Corporate ethics matter and people will buy accordingly.

11:28 AM, September 13, 2006  
Anonymous Anonymous said...

Corporate ethics matter and people will buy accordingly.

Wal-Mart





Do I really need to say more?

2:53 PM, September 13, 2006  
Anonymous Anonymous said...

Corporate ethics..

When I can't win benchmarks I sue..

Looks like AMD is up to it again...

Got no benchmarks, go no 65nm, got no 300mm fab, got crappy yields, Dude got Dell!..

Who cares if I got technology.. I got laywers that can sue.

hows that for corporate ethics!

8:24 PM, September 13, 2006  
Anonymous Edward said...

"When I can't win benchmarks I sue..

Looks like AMD is up to it again..."


AMD filed the suit in June 2005, after 2 years of performance lead.

Within 2 quarters, AMD's revenue and profit have unprecedented growth.

The company that practices ugly and illegal marketing (let alone behaves morally) is clearly not AMD, but Intel.

10:26 PM, September 13, 2006  

Post a Comment

Links to this post:

Create a Link

<< Home