Sign in with
Sign up | Sign in
Your question

AMD 65nm Processors in Q1 2007

Last response: in CPUs
Share
April 27, 2006 2:57:23 AM

Here is the latest on AMD's production plans.

http://www.xbitlabs.com/news/cpu/display/20060426181538...

Essentially, the 65nm process will begin ramping in Q3 of this year, however, products won't actually be released until Q1 2007. The 65nm process looks to come equipped with SiGe for that constantly referred to 42% transistor improvement over a non-stressed process. Given that AMD advertises their 2nd generation process as giving a 24% transistor performance increase, the third generation is about 18% better than the current process.

Also, despite insistance by some people, the 42% improvement in transistor performance doesn't translate into 42% faster processors.

Quote:
However, it doesn’t really mean that the frequency potential of the 0.065micron processors will also increase by 42%: the actual frequency boost will be lower, of course.

AMD is also scheduled to begin production on the 45nm process in 2008 although actual products won't ship until later, possibly 6 months later using the current 65nm plans as an example.
April 27, 2006 3:03:45 AM

Sweet, cheap dual cores in 2007.
April 27, 2006 3:22:49 AM

I cant wait it means that dual core sempron is on the way and k10 is coming out in 07 so put it together you get 5 issue sempron at 2.0 ghz dual core!(I'm just dreaming)
Related resources
April 27, 2006 3:27:32 AM

With the ability to make such complex chips becoming so cheap I wonder when we will finally see a a full computer on a chip minus the RAM and hdd.
April 27, 2006 3:54:32 AM

Quote:
Sweet, cheap dual cores in 2007.


By mid next year the FX-60 and the 840 EE would cost around $400 or lower. :D  But I would still buy the new ones. :D 
April 27, 2006 4:30:25 AM

$400 is far from cheap.
April 27, 2006 4:40:50 AM

Quote:
$400 is far from cheap.
From 1000$ ya its cheap but still you have a point i remember when you could buy a p3 at 1.2 ghz for 150 and oc it to the speed and performance of a 1.4 ghz taulatin.
April 27, 2006 4:47:07 AM

I said tualatin not coppermine. :roll:
April 27, 2006 4:56:01 AM

Quote:
I cant wait it means that dual core sempron is on the way and k10 is coming out in 07 so put it together you get 5 issue sempron at 2.0 ghz dual core!(I'm just dreaming)


I don't think this is k10, I think this is K8L, maybe I'm wrong though, maybe AM2 is K8L...
April 27, 2006 4:56:44 AM

Quote:
This is the most intersting post of the day, you should have put it in all caps...

Jack
Nah then action man would've stepped in with the keyboard and made this thread more hostile.
April 27, 2006 4:58:41 AM

Quote:
I cant wait it means that dual core sempron is on the way and k10 is coming out in 07 so put it together you get 5 issue sempron at 2.0 ghz dual core!(I'm just dreaming)


I don't think this is k10, I think this is K8L, maybe I'm wrong though, maybe AM2 is K8L... Yes it is k9 was dual core cpu's but k8l is just k8 with ddr2.
April 27, 2006 5:06:30 AM

Quote:
This is the most intersting post of the day, you should have put it in all caps...

Jack
Nah then action man would've stepped in with the keyboard and made this thread more hostile.

:p  :lol:  :lol:  :p 

Seriously, how many people think come January 2nd (the 1st is a holiday), we will see any 65 nm product out of AMD? We should start a betting pool ;) 

Anybody for Feb 14, nice valentines day present..
How about Feb 28th, perahps March 15th? It's been hard to narrow AMD down on their 65 nm progress of late. Better late then never. :lol: 
April 27, 2006 10:23:10 AM

I wonder at what freq those 65nm SOI with applied DSL + SMT can work sable and on what voltage.
April 27, 2006 1:29:46 PM

Quote:
This is the most intersting post of the day, you should have put it in all caps...

Jack
Nah then action man would've stepped in with the keyboard and made this thread more hostile.

:p  :lol:  :lol:  :p 

Seriously, how many people think come January 2nd (the 1st is a holiday), we will see any 65 nm product out of AMD? We should start a betting pool ;) 

Anybody for Feb 14, nice valentines day present..
How about Feb 28th, perahps March 15th? It's been hard to narrow AMD down on their 65 nm progress of late. Better late then never. :lol: 
Yep so Intel can run away with Conroe will all that waiting time. :D  [/fanboy] :?:
April 27, 2006 1:31:04 PM

Quote:
This is the most intersting post of the day, you should have put it in all caps...

Jack
Nah then action man would've stepped in with the keyboard and made this thread more hostile.

:p  :lol:  :lol:  :p 

Seriously, how many people think come January 2nd (the 1st is a holiday), we will see any 65 nm product out of AMD? We should start a betting pool ;) 

Anybody for Feb 14, nice valentines day present..
How about Feb 28th, perahps March 15th? It's been hard to narrow AMD down on their 65 nm progress of late. Better late then never. :lol: 

Ohhhh, yes very very true.... hmmmm, you know given a choice a) would I want to get 300 mm online or b) would i want to have 65 nm yielding? I would want 65 nm :) , they may have spent too many resources getting Fab 36 online, I wonder if it may have impacted their 65 nm schedule.

The ramp/release timeline is reasonable, Intel ramped 65 nm starting in Q3, built inventory for Yonah release in Q1 06... Between now and then 90 nm will need to suffice, will be tought though.

I think the chorus of questions was valid -- why not bring Fab 36 up on 65 nm? That would have been the most efficient and wisest, I understand the "bring it up on 90 nm" philosophy, but they have a 300 mm 65 nm development line, why not transfer it to a new fab like Intel does?
Simple they had orders 200 mm wafers and needed to use them all up thats why.
April 27, 2006 8:24:37 PM

Quote:
Here is the latest on AMD's production plans.

http://www.xbitlabs.com/news/cpu/display/20060426181538...

Essentially, the 65nm process will begin ramping in Q3 of this year, however, products won't actually be released until Q1 2007. The 65nm process looks to come equipped with SiGe for that constantly referred to 42% transistor improvement over a non-stressed process. Given that AMD advertises their 2nd generation process as giving a 24% transistor performance increase, the third generation is about 18% better than the current process.

Also, despite insistance by some people, the 42% improvement in transistor performance doesn't translate into 42% faster processors.

However, it doesn’t really mean that the frequency potential of the 0.065micron processors will also increase by 42%: the actual frequency boost will be lower, of course.

AMD is also scheduled to begin production on the 45nm process in 2008 although actual products won't ship until later, possibly 6 months later using the current 65nm plans as an example.


I'd like to add that AMD's 90nm process is more advanced than what Intel used. The next revision is more advanced than what AMD is using now. The new chips in June won't be limited to the same clocks as what AMD has now. There is no reason for AMD to announce new clock speeds. If this doesn't make sense then you need to be aware of AMD's behavior.

Basically, AMD has no reason to increase clocks quickly as long as they are ahead of Intel. It would be a mistake to assume that what AMD has out now is the fastest they can produce. It would doubly be a mistake to assume that the next revision won't be any faster. If AMD's behavior is puzzling to you then you only need to look to Intel. Intel has suggested and implied and generally fostered the impression that they will be releasing at high clock/high performance numbers however Intel has not actually promised anything. Now, imagine this scenario: AMD panics at Intel's song and dance and announces a big jump in clock, Intel releases in July at a lower clock than suggested, then AMD is stuck at delivering the higher clocked chips in the same price range. AMD won't do that.

The much more likely scenario is this: AMD releases AM2 in June at the same clock and takes whatever increase in speed is obtained. As long as AM2 shows a reasonable bump in speed the reviewers will be happy. Intel releases in July and then we see what the real comparison is. Then knowing what is needed to be competitive AMD releases at higher clocks.

I know; there are some here who simply assume that the current clock is the fastest that AMD can make and that AM2 isn't any faster and therefore there is nothing AMD can do until they switch to 65nm. However, this is not the way I would bet. I think AMD has one speed grade in its pocket now and will get a second speed grade with the new revision. My best guess is that AMD is capable of bumping two speed grades in June without affecting yield plus the bump it gets from AM2. On top of this AMD is also capable of producing some 65nm chips in 3Q. The 65nm would produce more chips and probably a bit lower power consumption but I wouldn't expect additional clock increases right away. Of course, AMD can revise 65nm each quarter so 4Q could see some improvement perhaps. Personally, I don't think AMD will need 65nm to increase clock until 2007. I could be totally wrong I suppose although I haven't been in the last three years. Maybe it's time for me to really blow a prediction . . . or maybe not.
April 27, 2006 8:44:04 PM

Quote:
The ramp/release timeline is reasonable, Intel ramped 65 nm starting in Q3, built inventory for Yonah release in Q1 06... Between now and then 90 nm will need to suffice, will be tought though.

I think the chorus of questions was valid -- why not bring Fab 36 up on 65 nm? That would have been the most efficient and wisest, I understand the "bring it up on 90 nm" philosophy, but they have a 300 mm 65 nm development line, why not transfer it to a new fab like Intel does?.


It is difficult for Intel, not for AMD. AMD runs its tests on the same assembly line that runs production. Intel doesn't do this. AMD is only capacity limited. Okay, let me state this another way.

Let's just suppose for the sake of argument that AMD decided tomorrow to switch all production to 65nm. I'm sure this would make the naysayers on here happy but this would be disastrous for AMD. FAB36 is no where near full output yet; the bulk of chip production still comes from FAB30. However, if AMD rapidly moved to 65n this would greatly devalue what it can produce at FAB30 and it wouldn't be able to make up the volume at FAB36. This would cause AMD to take a sharp hit in revenue which would be criminally incompetent.

AMD will give 90nm a push. This process is not nearing its limits as one article suggested. This push on 90nm will give AMD time to ramp up 65nm on FAB36 without losing revenue. It makes no sense for AMD to undermine production on its main facility right now. By mid 2007 AMD will be capable of producing about 75% more chips in FAB36 than FAB30. This will drop FAB 30 to just over 1/3rd of production capacity which should be fine for the lower value market. At this point AMD can start looking to upgrade FAB 30 without affecting production.
April 28, 2006 1:34:16 AM

Quote:
This is the most intersting post of the day, you should have put it in all caps...

Jack


Word.
April 28, 2006 1:36:41 AM

Quote:
This is the most intersting post of the day, you should have put it in all caps...

Jack
Nah then action man would've stepped in with the keyboard and made this thread more hostile.

:p  :lol:  :lol:  :p 

Seriously, how many people think come January 2nd (the 1st is a holiday), we will see any 65 nm product out of AMD? We should start a betting pool ;) 

Anybody for Feb 14, nice valentines day present..
How about Feb 28th, perahps March 15th? It's been hard to narrow AMD down on their 65 nm progress of late. Better late then never. :lol: 

Ohhhh, yes very very true.... hmmmm, you know given a choice a) would I want to get 300 mm online or b) would i want to have 65 nm yielding? I would want 65 nm :) , they may have spent too many resources getting Fab 36 online, I wonder if it may have impacted their 65 nm schedule.

The ramp/release timeline is reasonable, Intel ramped 65 nm starting in Q3, built inventory for Yonah release in Q1 06... Between now and then 90 nm will need to suffice, will be tought though.

I think the chorus of questions was valid -- why not bring Fab 36 up on 65 nm? That would have been the most efficient and wisest, I understand the "bring it up on 90 nm" philosophy, but they have a 300 mm 65 nm development line, why not transfer it to a new fab like Intel does?

I would choose 300mm wafers more CPU's based on a tested process, better yields, more CPU's, and more money.
April 28, 2006 1:38:19 AM

Quote:
This is the most intersting post of the day, you should have put it in all caps...

Jack
Nah then action man would've stepped in with the keyboard and made this thread more hostile.

:p  :lol:  :lol:  :p 

Seriously, how many people think come January 2nd (the 1st is a holiday), we will see any 65 nm product out of AMD? We should start a betting pool ;) 

Anybody for Feb 14, nice valentines day present..
How about Feb 28th, perahps March 15th? It's been hard to narrow AMD down on their 65 nm progress of late.

Between April 21 to July 12.
April 28, 2006 1:46:19 AM

Quote:
Here is the latest on AMD's production plans.

http://www.xbitlabs.com/news/cpu/display/20060426181538...

Essentially, the 65nm process will begin ramping in Q3 of this year, however, products won't actually be released until Q1 2007. The 65nm process looks to come equipped with SiGe for that constantly referred to 42% transistor improvement over a non-stressed process. Given that AMD advertises their 2nd generation process as giving a 24% transistor performance increase, the third generation is about 18% better than the current process.

Also, despite insistance by some people, the 42% improvement in transistor performance doesn't translate into 42% faster processors.

However, it doesn’t really mean that the frequency potential of the 0.065micron processors will also increase by 42%: the actual frequency boost will be lower, of course.

AMD is also scheduled to begin production on the 45nm process in 2008 although actual products won't ship until later, possibly 6 months later using the current 65nm plans as an example.


I'd like to add that AMD's 90nm process is more advanced than what Intel used. The next revision is more advanced than what AMD is using now. The new chips in June won't be limited to the same clocks as what AMD has now. There is no reason for AMD to announce new clock speeds. If this doesn't make sense then you need to be aware of AMD's behavior.

Basically, AMD has no reason to increase clocks quickly as long as they are ahead of Intel. It would be a mistake to assume that what AMD has out now is the fastest they can produce. It would doubly be a mistake to assume that the next revision won't be any faster. If AMD's behavior is puzzling to you then you only need to look to Intel. Intel has suggested and implied and generally fostered the impression that they will be releasing at high clock/high performance numbers however Intel has not actually promised anything. Now, imagine this scenario: AMD panics at Intel's song and dance and announces a big jump in clock, Intel releases in July at a lower clock than suggested, then AMD is stuck at delivering the higher clocked chips in the same price range. AMD won't do that.

The much more likely scenario is this: AMD releases AM2 in June at the same clock and takes whatever increase in speed is obtained. As long as AM2 shows a reasonable bump in speed the reviewers will be happy. Intel releases in July and then we see what the real comparison is. Then knowing what is needed to be competitive AMD releases at higher clocks.

I know; there are some here who simply assume that the current clock is the fastest that AMD can make and that AM2 isn't any faster and therefore there is nothing AMD can do until they switch to 65nm. However, this is not the way I would bet. I think AMD has one speed grade in its pocket now and will get a second speed grade with the new revision. My best guess is that AMD is capable of bumping two speed grades in June without affecting yield plus the bump it gets from AM2. On top of this AMD is also capable of producing some 65nm chips in 3Q. The 65nm would produce more chips and probably a bit lower power consumption but I wouldn't expect additional clock increases right away. Of course, AMD can revise 65nm each quarter so 4Q could see some improvement perhaps. Personally, I don't think AMD will need 65nm to increase clock until 2007. I could be totally wrong I suppose although I haven't been in the last three years. Maybe it's time for me to really blow a prediction . . . or maybe not.

No AMD's .09u is not more advanced or special or anything in comparison to Intel’s .09u, the only notable difference is SOI.

Clock skew is what's stopping AMD from scaling further actually, and you might not want to admit it but AMD needs a new architecture to avoid further issues, switching from .09u to .065u doesn't make the processor clock faster, it gives them more options for IC and lower power requirements (generally).

But if you don't believe me check with AMD they only supply single core 3.0 machines, could there be binning issues or maybe yields?
April 28, 2006 1:49:28 AM

Quote:
The ramp/release timeline is reasonable, Intel ramped 65 nm starting in Q3, built inventory for Yonah release in Q1 06... Between now and then 90 nm will need to suffice, will be tought though.

I think the chorus of questions was valid -- why not bring Fab 36 up on 65 nm? That would have been the most efficient and wisest, I understand the "bring it up on 90 nm" philosophy, but they have a 300 mm 65 nm development line, why not transfer it to a new fab like Intel does?.


It is difficult for Intel, not for AMD. AMD runs its tests on the same assembly line that runs production. Intel doesn't do this. AMD is only capacity limited. Okay, let me state this another way.

Let's just suppose for the sake of argument that AMD decided tomorrow to switch all production to 65nm. I'm sure this would make the naysayers on here happy but this would be disastrous for AMD. FAB36 is no where near full output yet; the bulk of chip production still comes from FAB30. However, if AMD rapidly moved to 65n this would greatly devalue what it can produce at FAB30 and it wouldn't be able to make up the volume at FAB36. This would cause AMD to take a sharp hit in revenue which would be criminally incompetent.

AMD will give 90nm a push. This process is not nearing its limits as one article suggested. This push on 90nm will give AMD time to ramp up 65nm on FAB36 without losing revenue. It makes no sense for AMD to undermine production on its main facility right now. By mid 2007 AMD will be capable of producing about 75% more chips in FAB36 than FAB30. This will drop FAB 30 to just over 1/3rd of production capacity which should be fine for the lower value market. At this point AMD can start looking to upgrade FAB 30 without affecting production.

What makes you think Intel doesn’t test binned CPU's in the same facility?

Yes the process is reaching its limits, I almost dare say it has but I'm not semi-conductor engineer so I can't say for certain.
April 28, 2006 3:47:46 AM

Quote:
I know; there are some here who simply assume that the current clock is the fastest that AMD can make and that AM2 isn't any faster and therefore there is nothing AMD can do until they switch to 65nm. However, this is not the way I would bet. I think AMD has one speed grade in its pocket now and will get a second speed grade with the new revision. My best guess is that AMD is capable of bumping two speed grades in June without affecting yield plus the bump it gets from AM2. On top of this AMD is also capable of producing some 65nm chips in 3Q. The 65nm would produce more chips and probably a bit lower power consumption but I wouldn't expect additional clock increases right away. Of course, AMD can revise 65nm each quarter so 4Q could see some improvement perhaps. Personally, I don't think AMD will need 65nm to increase clock until 2007. I could be totally wrong I suppose although I haven't been in the last three years. Maybe it's time for me to really blow a prediction . . . or maybe not.


Ahhh, a betting man... Hmm, gambling being statistically favored towards the house generally is not wise in practice. ;)  Two speed grades by June... I would indeed be surprised to see AMD release an Opteron 2.8GHz and 3.0 GHz in June too!!! It would be a real trick for AMD to release dual core 3GHz chips on 90nm.
April 28, 2006 3:48:15 AM

Quote:
This is what I mean by it is reasonable. AMD/IBM conferenced their 65 nm technology Dec 2005 and release parts approximately 1 year later, this is consistent with ITRS roadmaps (page 6, figure 3 - ITRS 2005 Roadmap) The referenced figure is the typical timeline from development to release of final product. When AMD said 2nd half 2006 I was a bit skeptical, a Q1 2007 is right in line with what we have seen. 1 Quarter to build inventory for a product release is about average.


It is difficult for Intel, not for AMD. AMD runs its tests on the same assembly line that runs production. Intel doesn't do this. AMD is only capacity limited. Okay, let me state this another way.


Odd, Intel is producing product material out of D1D, all Yonah for the Jan launch came from D1D , and Conroe will be launched from the ramp that has already started at D1D... last I heard they were also doing 45 nm there, weird that they are running production and developed a funtional 45 nm SRAM 3 months ahead of AMD is it not? Intel also maintains it's own in house mask shop, this is an enormous asset, as resources are dedicated to developing the litho masks in quick fashion for bebug and testing.
Info: 45 nm development
Inte's Inhouse Mask Shop
D1D makes 65 nm processors

Quote:

Let's just suppose for the sake of argument that AMD decided tomorrow to switch all production to 65nm. I'm sure this would make the naysayers on here happy but this would be disastrous for AMD. FAB36 is no where near full output yet; the bulk of chip production still comes from FAB30. However, if AMD rapidly moved to 65n this would greatly devalue what it can produce at FAB30 and it wouldn't be able to make up the volume at FAB36. This would cause AMD to take a sharp hit in revenue which would be criminally incompetent.


About roughly 80% of a fabs current tooling can be reused from technology to technology, the other 20% needs to be upgraded or new tooling purchased. It is not as easy as you are stating, in fact, AMD is " installing kits to begin printing 65 nm in next quarter' a q1 release if you want to believe the inquirer. Moving a fab to 65 nm to "devalue" it is simply ludicrous. If AMD could produce revenue parts from 65 nm tomorrow they would do it. The cost benefit for going smaller is much higher than assuming they are make a strategic move not to release.

Quote:

AMD will give 90nm a push. This process is not nearing its limits as one article suggested.

Explain please?

Quote:

This push on 90nm will give AMD time to ramp up 65nm on FAB36 without losing revenue. It makes no sense for AMD to undermine production on its main facility right now.


I guess this is why we are getting such cheaper dual core chips out of AMD right now.

Quote:

By mid 2007 AMD will be capable of producing about 75% more chips in FAB36 than FAB30. This will drop FAB 30 to just over 1/3rd of production capacity which should be fine for the lower value market. At this point AMD can start looking to upgrade FAB 30 without affecting production.


This is true, if AMD is wise, they would design and produce their own server chips on the 90 nm 200mm fab.

Cheers!
Jack

Quote:
Here is the latest on AMD's production plans.
...
...
AMD is also scheduled to begin production on the 45nm process in 2008 although actual products won't ship until later, possibly 6 months later using the current 65nm plans as an example.



I'd like to add that AMD's 90nm process is more advanced than what Intel used. The next revision is more advanced than what AMD is using now. The new chips in June won't be limited to the same clocks as what AMD has now. There is no reason for AMD to announce new clock speeds. If this doesn't make sense then you need to be aware of AMD's behavior.
....

No AMD's .09u is not more advanced or special or anything in comparison to Intel’s .09u, the only notable difference is SOI.

Clock skew is what's stopping AMD from scaling further actually, and you might not want to admit it but AMD needs a new architecture to avoid further issues, switching from .09u to .065u doesn't make the processor clock faster, it gives them more options for IC and lower power requirements (generally).

But if you don't believe me check with AMD they only supply single core 3.0 machines, could there be binning issues or maybe yields?

:lol:  :lol:  :lol:  :lol:  :lol: 

But SOI is soooooo cool, and easy to spell it has only 3 letters.

I'm not sure that everyone understands the difference between Intel and AMD on process technology.

At Intel they used FAB D1C to hammer out the 90nm process. Intel built D1D to take over as D1C became outdated so D1D was used to create the 65nm process and is currently working on 45nm and preliminary 32nm. These tests produce a recipe that is then taken to an actual production FAB and then the engineers work to replicate it. However, there are differences between testing and volume FABs so the recipe has to be adjusted. Intel also tests its new die designs at these test FABs. For example, the Woodcrest and Conroe test masks and production steps would have been run at D1D.

This leads to some unusual aspects of Intel's production. For example, it is entirely possible for D1D to produce chips that are available for testing but not actually in production. D1D is not a volume facility and can't supply the market with chips. After a recipe is created at D1D it still has to be tweaked at the actual production FAB, so there can be a time lag between when Intel has chips ready at D1D and when they are actually available on the market. If Intel has poor yields this process can take even longer. The chips that were shown of Conroe were almost certainly created at D1D and don't reflect real production. Intel's production FABs work best when nothing changes. In other words, during the initial ramping Intel doesn't make any money but once everything is set they are great at producing volume chips for minimal cost. Intel has at least one mixed mode FAB but I don't believe most of them are. Presumably, Intel is currently trying to duplicate the Woodcrest recipe at one of its 65nm production FABs.

AMD is different. The new processes are actually created at IBM. For example, the 130nm SOI, 90nm SOI, 65nm SOI, etc. were all tested at IBM. Now this is the part where it differs from Intel. You see, this recipe is not then given to AMD to try to duplicate. IBM uses AMD's APM software so APM knows what the recipe is in great detail. More importantly, APM knows what the differences are between IBM's FAB and AMD's. When a new manufacturing process is developed at IBM AMD simply ports the recipe and APM figures out how to tweak it so that it is adapted to AMD's FAB. This works to an amazing degree. Although someone suggested that AMD had some problems moving to 90nm this wasn't actually true. Every new process, since 130nm, has been started at revenue producing volumes at AMD. In other words, AMD has not had a low yield, money losing startup since it first moved to 130nm with K7.

There are other differences. AMD does not have a separate FAB to test new masks. Although IBM creates new processes they don't have access to AMD's actual cpu technology. This is why when AMD tests a new process they produce RAM chips instead of cpu's. Intel and AMD do not share cpu technology and none of AMD's dies are run at Intel. AMD tests its new die design at its production FABs. AMD is capable of running new design tests inline with its regular production because of the much greater flexibility of its FAB design. AMD is capable of running multiple processes inline and even multiple process sizes. So, for example, AMD can produce Turion processors along with Athlon 64, FX, Opteron, and Sempron where each one uses a different process recipe and by end of 2006 some will be 90n and others 65nm. AMD's APM is also capable of checking spec on each step of the process and adjusting later steps to bring wafers back in spec. This increases yield. APM is also capable of reducing the cost of each step if producing lower value Sempron steps. APM is also capable of adjusting each tool's actions as it slips away from perfect adjustment. This too increases quality and yield as well as allowing easier maintenance scheduling. Intel is not really capable of doing this.

So, when Intel has test chips available they may not really be available because they are probably from D1D and not in actual production yet. In contrast, when AMD has test chips available they are produced in the same FAB that is used for production and are immediately capable of being produced. This gives AMD about three months head start on Intel.
April 28, 2006 3:48:55 AM

Pass me the soi sauce! :wink:
April 28, 2006 3:50:58 AM

Quote:
Here is the latest on AMD's production plans.
...
...
AMD is also scheduled to begin production on the 45nm process in 2008 although actual products won't ship until later, possibly 6 months later using the current 65nm plans as an example.



I'd like to add that AMD's 90nm process is more advanced than what Intel used. The next revision is more advanced than what AMD is using now. The new chips in June won't be limited to the same clocks as what AMD has now. There is no reason for AMD to announce new clock speeds. If this doesn't make sense then you need to be aware of AMD's behavior.
....

No AMD's .09u is not more advanced or special or anything in comparison to Intel’s .09u, the only notable difference is SOI.

Clock skew is what's stopping AMD from scaling further actually, and you might not want to admit it but AMD needs a new architecture to avoid further issues, switching from .09u to .065u doesn't make the processor clock faster, it gives them more options for IC and lower power requirements (generally).

But if you don't believe me check with AMD they only supply single core 3.0 machines, could there be binning issues or maybe yields?

:lol:  :lol:  :lol:  :lol:  :lol: 

But SOI is soooooo cool, and easy to spell it has only 3 letters. Is it? Silicon On Insulator. Try saying that 5 times fast.
April 28, 2006 4:17:04 AM

I'm not sure why you are suggesting that retail processors aren't produced at D1D, because I'm quite sure that they are. In fact, all the Yonahs available at launch, most of which seemed to have gone to Apple, were produced at D1D.

For example:
http://www.legitreviews.com/article/172/1/

Quote:
They are being produced at D1D Fab, which just happens to be the 65nm development center located in Oregon and is where the other 65nm CPU's will soon be coming from.

This was from the Spring 2005 IDF, and at that point D1D was the only operational 65nm fab. The thing is the next 65nm Fab, Fab 12 only reopened from renovation in November 2005 and didn't likely produce processors until the end of that month. There is no way that Fab 12 could have produced all the 65nm Yonahs, and the 65nm Cedar Mills and Preslers in the month of December to meet the January 2006 launch. D1D must have been contributing production capacity. The other 65nm capable Fab now is Fab 24-2 but that didn't become operational until early 2006 so it would have only contributed production toward the end of Q1 rather than the beginning.

http://www.intel.com/pressroom/kits/manufacturing/manuf...

If you actually look at D1D's available space, you see that it is actually as large as Fab 12. If D1D was only doing development, it wouldn't need all that production space.

I read somewhere that Intel's production systems was that development facilities like D1D test out an optimal process for production which is then exported to all the other Fabs. A chip produced at one Fab should be exactly the same as chips produced in any other Fab because everything is repreduced exactly from the temperature and humidity in the facility to the water used in production. The yields for all the Fabs should therefore be the same.
April 28, 2006 4:30:48 AM

Quote:
I read somewhere that Intel's production systems was that development facilities like D1D test out an optimal process for production which is then exported to all the other Fabs. A chip produced at one Fab should be exactly the same as chips produced in any other Fab because everything is repreduced exactly from the temperature and humidity in the facility to the water used in production. The yields for all the Fabs should therefore be the same.

Yeah it is called "copy exactly" and Intel does that when ramping other fabs for production when they get a good yield process in a test Fab. The copy everything exactly, like even the placement of several-ton machines down to a millimeter tolerance on the floor placement. They even copy stuff that you wouldn't think matters because sometimes really good yields involve a bit-o-black-magic. :) 
April 28, 2006 4:42:34 AM

Quote:
I'm not sure why you are suggesting that retail processors aren't produced at D1D, because I'm quite sure that they are. In fact, all the Yonahs available at launch, most of which seemed to have gone to Apple, were produced at D1D.

For example:
http://www.legitreviews.com/article/172/1/

They are being produced at D1D Fab, which just happens to be the 65nm development center located in Oregon and is where the other 65nm CPU's will soon be coming from.

This was from the Spring 2005 IDF, and at that point D1D was the only operational 65nm fab. The thing is the next 65nm Fab, Fab 12 only reopened from renovation in November 2005 and didn't likely produce processors until the end of that month. There is no way that Fab 12 could have produced all the 65nm Yonahs, and the 65nm Cedar Mills and Preslers in the month of December to meet the January 2006 launch. D1D must have been contributing production capacity. The other 65nm capable Fab now is Fab 24-2 but that didn't become operational until early 2006 so it would have only contributed production toward the end of Q1 rather than the beginning.

http://www.intel.com/pressroom/kits/manufacturing/manuf...

If you actually look at D1D's available space, you see that it is actually as large as Fab 12. If D1D was only doing development, it wouldn't need all that production space.

I read somewhere that Intel's production systems was that development facilities like D1D test out an optimal process for production which is then exported to all the other Fabs. A chip produced at one Fab should be exactly the same as chips produced in any other Fab because everything is repreduced exactly from the temperature and humidity in the facility to the water used in production. The yields for all the Fabs should therefore be the same.
You've answered your own question.

I'll quote myself:
Quote:
Intel also tests its new die designs at these test FABs. For example, the Woodcrest and Conroe test masks and production steps would have been run at D1D.
April 28, 2006 4:47:46 AM

Quote:
You are simply wrong on just about everything you said, I will take this point by point:

I'm not sure that everyone understands the difference between Intel and AMD on process technology.


I, myself, am fully aware. :) 

Quote:

At Intel they used FAB D1C to hammer out the 90nm process. Intel built D1D to take over as D1C became outdated so D1D was used to create the 65nm process and is currently working on 45nm and preliminary 32nm. These tests produce a recipe that is then taken to an actual production FAB and then the engineers work to replicate it. However, there are differences between testing and volume FABs so the recipe has to be adjusted. Intel also tests its new die designs at these test FABs. For example, the Woodcrest and Conroe test masks and production steps would have been run at D1D. This leads to some unusual aspects of Intel's production. For example, it is entirely possible for D1D to produce chips that are available for testing but not actually in production. D1D is not a volume facility and can't supply the market with chips. After a recipe is created at D1D it still has to be tweaked at the actual production FAB, so there can be a time lag between when Intel has chips ready at D1D and when they are actually available on the market. If Intel has poor yields this process can take even longer. The chips that were shown of Conroe were almost certainly created at D1D and don't reflect real production. Intel's production FABs work best when nothing changes. In other words, during the initial ramping Intel doesn't make any money but once everything is set they are great at producing volume chips for minimal cost. Intel has at least one mixed mode FAB but I don't believe most of them are. Presumably, Intel is currently trying to duplicate the Woodcrest recipe at one of its 65nm production FABs.


Ok where to begin, D1D has a floor space of 212 ksq feet, F12 (based on your reasoning a production fab) is 210 ksq feet. (Ref: http://www.intel.com/pressroom/kits/manufacturing/manuf...). Fab 36 (AMD’s fab) is 13,400 sq meters of cleanroom space (Source: http://www.amdboard.com/amd_fab36.html) or in feet is ~ 144 ksq feet. D1C is not out dated, it is running 90 nm and will be receiving 65 nm before the end of the year (source: http://www.tomshardware.com/2005/10/07/a_sneak_peak_at_...) Note: the THG source I quote states they are producing 65 nm. Finally, here is a quote from one of the more famous technologist in the industry, M. Bohr related during a conference call:
Quote:

The D1D fab is also targeted for initial high-volume production at the 45-nm node, which is slated for 2007. The fab will run both 65- and 45-nm devices simultaneously, he said during the conference call.


Source: http://www.eet.com/news/latest/showArticle.jhtml?articl...
Note, this is not an Inquirer link.




Quote:

AMD is different. The new processes are actually created at IBM. For example, the 130nm SOI, 90nm SOI, 65nm SOI, etc. were all tested at IBM. Now this is the part where it differs from Intel. You see, this recipe is not then given to AMD to try to duplicate. IBM uses AMD's APM software so APM knows what the recipe is in great detail. More importantly, APM knows what the differences are between IBM's FAB and AMD's. When a new manufacturing process is developed at IBM AMD simply ports the recipe and APM figures out how to tweak it so that it is adapted to AMD's FAB. This works to an amazing degree. Although someone suggested that AMD had some problems moving to 90nm this wasn't actually true. Every new process, since 130nm, has been started at revenue producing volumes at AMD. In other words, AMD has not had a low yield, money losing startup since it first moved to 130nm with K7.

Now this is true, AMD would not exist for not be for IBM. But you are very wrong on APM. APM is a fab automation package that simply employs SEMI standards (in fine quality, and detail mind you, AMD should be commended on the quality of the work) to interface with the equipment. SEMI standards define many things, on the automation side such as substrate tracking, tool tracking, and even feedback loops. APM is a system that will allow feed forward or feed backward mechanisms that will self correct such things as deposition times, etch times, etc based on measurements taken on the wafers. This is the extent of APM’s “recipe” knowledge. The recipe, as you speak, are self contained on the equipment and proprietary code made by the equipment vendor. Here is a link listing all the SEMI standards:
http://downloads.semi.org/pubs/wstdsbal.nsf/NewPubsList. Now, to think that Intel does not automate their fabs is ludicrous. However, to determine which one is better is impossible for I have witnessed neither in action. However, all equipment vendors code their control software to the SEMI standards, AMD’s APM software or Intel’s internal automation software must adhere to this to some degree. Intel has not had a low yield money losing startup since 1985, when they failed to hit DRAM yields during a transfer which born out of that learning Copy Exactly!

Quote:

There are other differences. AMD does not have a separate FAB to test new masks. Although IBM creates new processes they don't have access to AMD's actual cpu technology. This is why when AMD tests a new process they produce RAM chips instead of cpu's. Intel and AMD do not share cpu technology and none of AMD's dies are run at Intel. AMD tests its new die design at its production FABs. AMD is capable of running new design tests inline with its regular production because of the much greater flexibility of its FAB design. AMD is capable of running multiple processes inline and even multiple process sizes. So, for example, AMD can produce Turion processors along with Athlon 64, FX, Opteron, and Sempron where each one uses a different process recipe and by end of 2006 some will be 90n and others 65nm. AMD's APM is also capable of checking spec on each step of the process and adjusting later steps to bring wafers back in spec. This increases yield. APM is also capable of reducing the cost of each step if producing lower value Sempron steps. APM is also capable of adjusting each tool's actions as it slips away from perfect adjustment. This too increases quality and yield as well as allowing easier maintenance scheduling. Intel is not really capable of doing this.

So, when Intel has test chips available they may not really be available because they are probably from D1D and not in actual production yet. In contrast, when AMD has test chips available they are produced in the same FAB that is used for production and are immediately capable of being produced. This gives AMD about three months head start on Intel.


Dude, SRAM is the industry standard for development because it has such a high transistor density and stresses defect performance. You are spouting jibberish above.

“Intel is not capable of doing this”, please provide data that supports this “opinion.” What you are describing here is called advanced process control (APC), and it is not unqiue to AMD (Source: http://lorien.ncl.ac.uk/ming/advcontrl/sect3.htm, http://www.apc-network.com/apc/default.aspx, http://www.hitachi.com/ICSFiles/afieldfile/2004/06/07/r...) In fact, the SEMI standards also have specs on APC just as they about everything else. APM, as AMD likes to call it, is nothing more that an outstanding implementation of what the industry already does (Including Intel).

I don't know why there is any confusion about the manufacturing processes used by AMD and Intel. The two companies are in very different positions in terms of manufacturing capacity so why would they use the same process strategy?

AMD is capacity limited; they could potentially sell more chips than they can make. If you are capacity limited then you can't make more money by making more chips. So, instead, you try to add more value to each chip you do make. It makes perfect sense for AMD to use a more advanced process and try to increase the value of each chip. Everything involving production is more advanced at AMD: the assembly line is more advanced, the control software, APM is more advanced, and the actual deposition process is more advanced. This is what you do when capacity is the bottleneck, you try to get more value out of what you can produce. BTW, it is the more advanced assembly line that enables AMD to make upgrades while in production. Intel can't match this feat with its simpler assembly lines.

Intel is not capacity limited but it also lacks the more advanced assembly lines and production control software that AMD has. To compete properly Intel would to embark on a massive upgrade program to make its FABs equal to AMD's. This would take years and Billions of dollars. So, Intel opts for a simpler process with lower value and goes with higher volume. Intel's volume is so high that it can easily make more money than AMD so this strategy works very well for now.

However, this does bring up a curious action by Intel. Assuming that its chips are lower value it begs the question of whether taking on AMD in the server space first is the best strategy. Server chips are high value and Intel seems to be at a disadvantage here. If this assessment is true then Intel will either try to go low in the server market or face large drops in profit with higher costs as it tries to match AMD. It will be interesting to see what happens. At any rate, if this assessment is accurate it means that Intel will be happiest at a lower range than AMD where its volume can bring in good revenue. This would still leave AMD with the best of the server, workstation, and enthusiast/gamer markets. Notebooks are a tossup because AMD would need a better chipset to be truly competitive.
April 28, 2006 5:09:41 AM

Quote:
I read somewhere that Intel's production systems was that development facilities like D1D test out an optimal process for production which is then exported to all the other Fabs. A chip produced at one Fab should be exactly the same as chips produced in any other Fab because everything is repreduced exactly from the temperature and humidity in the facility to the water used in production. The yields for all the Fabs should therefore be the same.

Yeah it is called "copy exactly" and Intel does that when ramping other fabs for production when they get a good yield process in a test Fab. The copy everything exactly, like even the placement of several-ton machines down to a millimeter tolerance on the floor placement. They even copy stuff that you wouldn't think matters because sometimes really good yields involve a bit-o-black-magic. :) 

This is correct. Been there.
April 28, 2006 5:19:58 AM

Quote:
BTW, it is the more advanced assembly line that enables AMD to make upgrades while in production. Intel can't match this feat with its simpler assembly lines.


I have a different perspective on this. Intel has a rigid process for change - any and all change, period. The Intel process for change is known well throughout the industry. All suppliers to Intel know EXACTLY how rigid and disciplined the policies are. Some see this as adding sluggishness to their production but that fails to see that QC and product reliability are first tier priorities. It's a way of doing business where change is taken very seriously after considerable thought and planning.
April 28, 2006 5:20:41 AM

You've answered your own question.

I'll quote myself:

Quote:
Intel also tests its new die designs at these test FABs. For example, the Woodcrest and Conroe test masks and production steps would have been run at D1D.

I think you are missing the point. With Copy Exact, whether the processors are produced at D1D, Fab 12 or Fab 24-2, they will be the same. Now it's possible that the samples we've seen so far are specially picked ones, but that's simply binning and that can happen at any Fab whether it is during development or mass production.
April 28, 2006 6:53:00 AM

Hmm, interesting.
I find it a little strange, how you read
Quote:
However, it doesn’t really mean that the frequency potential of the 0.065micron processors will also increase by 42%:

I read it that that could almost be an option, but for the added heat. I notice that they say
Quote:
compared with the “pure” 0.065micron process.
which to me, implies shrink before change. It also suggests that the wiggle room made available by the shrink itself, has not yet been factored in. The net result could be higher than a 50% perf/watt improvement.

Jack I'm guessing you missed the part of the article where they said
Quote:
It is a clear indication that the production yields and the transition to 0.065micron technology are so great that it allows AMD to introduce complex “additives” all at once without wasting the time on polishing off the “naked” 0.065micron process.
You no longer have to worry about AMD's ability to transition to 65 nanos. It's a done deal. Sort of makes sense, when you consider that they have been marketing the transition technology for a little over a quarter now (you can google for that yourself, since I cant find the link).
By the way, AMD's 65 nano chips will begin to show availability in late October. It's just the way they do it. Besides, they dont have the spare production to sit on that many chips, for any length of time.
It does look like there are interesting times ahead. It looks like AMD has positioned themselves to outpace the best that Intel can bring forward.
April 28, 2006 8:36:49 AM

I can see that you are not familiar with AMD's sop.
Historicly, they have had a senior executive release a notice of intent. In this case it was Mr. Ostrander (yes, you spelled it wrong). That would be the day that the clock starts ticking. The expectation is that they would have reached mature yields, by the nineth month. It is only at that point that they issue an advice that they are shipping.
Because AMD is afraid of damaging thier reputation with the larger oems, they ship first production tech to retail distributors. As a result select retailers generally have a supply of the chips in month 6.
That is why I predict October as a realistic timeline for availability.
From personal experience, I had 180 and 130 nano chips more than two months before the anouncement. Since I was busy with projects around my "old" new house, I'm not sure when 90 nano chips actually became available.
As far as releasing info on production, I think Amd is trying to hold thier cards tight to thier chest. Remember that fab 36 was officially opened more than 6 months ago. I doubt thier engineers let those new toys sit idle.
Perhaps my faith in AMD is unfounded, but it was not given carelessly. While Intel has been the more pedantic of the two, at least in recent years, that has not been an advantage.
April 28, 2006 12:41:56 PM

Quote:
This is the most intersting post of the day, you should have put it in all caps...

Jack
Nah then action man would've stepped in with the keyboard and made this thread more hostile.

:p  :lol:  :lol:  :p 

Seriously, how many people think come January 2nd (the 1st is a holiday), we will see any 65 nm product out of AMD? We should start a betting pool ;) 

Anybody for Feb 14, nice valentines day present..
How about Feb 28th, perahps March 15th? It's been hard to narrow AMD down on their 65 nm progress of late. Better late then never. :lol: 

Ohhhh, yes very very true.... hmmmm, you know given a choice a) would I want to get 300 mm online or b) would i want to have 65 nm yielding? I would want 65 nm :) , they may have spent too many resources getting Fab 36 online, I wonder if it may have impacted their 65 nm schedule.

The ramp/release timeline is reasonable, Intel ramped 65 nm starting in Q3, built inventory for Yonah release in Q1 06... Between now and then 90 nm will need to suffice, will be tought though.

I think the chorus of questions was valid -- why not bring Fab 36 up on 65 nm? That would have been the most efficient and wisest, I understand the "bring it up on 90 nm" philosophy, but they have a 300 mm 65 nm development line, why not transfer it to a new fab like Intel does?


Because moving a process to a new fab isnt easy. We've been doing it since before Barrett took the helm and it still takes a while.

So bringing it up on 90nm takes less time to get first yeild, rough sighting it in if you will. from there its just a hop skip and a fine tuning jump to 65nm, and in AMD's case it's just a shrink, so it will go quick.
We qualified the first ever 300mm fab D1C on 1.3 gig tualitans, even though we were working on the new 90nm process in the background. Then we shrank the existing product, then moved on to bigger better things.

And to really get good site to site transfers, you need to worry about so much crap that you normally wouldnt think about. Piping and ductwork angles, barometric pressures, door placements in regard to airflow patterns, etc. that is nearly impossible your first few times out. I know AMD is no spring chicken, but this is the first time they've attempted to Copy Exactly =) a process this big and complex to another fab and match yeilds.
April 29, 2006 9:42:53 AM

It's easy for us =P. Besides, why would they shink in NY and transfer to germany? I could see sending matching lots and test wafers over, but the production line will have to be litho tuned to 65nm in either case. . . so why double your efforts when you dont need to?
April 29, 2006 12:13:55 PM

This article says that we will see the 65nm cpus from amd in Q1 2007 but it doesn't say if they are going to be quad cores or dual cores. If intel plans to launch quad cores in Q1 2007 then i find it more logical that this 65nm cpus will be quad cores. Will they be dual cores or quad cores?
April 29, 2006 4:31:31 PM

They will be both. Most likely Socket F will go directly to K8L with the 65nm process and just be available in quad cores. The desktop consumer chips will be available in both dual core and quad cores configurations. I don't think the average consumer would have much need for 4 cores anyways and most software is barely optimized for 2 cores much less 4. I definitely won't be expecting quad core mobile chips from either Intel or AMD anytime soon.
!