AMD Announcing Fab Spin-Off?

[Idontcare]

Saw this over on XS forums (link)

Advanced Micro Devices Inc. Tuesday is announcing a broad plan to cut costs by spinning off its manufacturing operations to a new joint venture, according to people familiar with the matter.

The new venture, which will make chips for AMD and other companies, will take over facilities that include large AMD factories in Germany, with funding provided by investment entities associated with the government of Abu Dhabi, these people said.
WSJ: AMD to Spin Off Manufacturing to New Venture

AMD To Make Significant Corporate Announcement

Company to conduct a call for media and industry analysts

Advanced Micro Devices (NYSE:AMD):

WHO: AMD President and CEO Dirk Meyer and additional company executives
Bloomberg: AMD To Make Significant Corporate Announcement

My sources are telling me that indeed AMD will be spinning off ALL its chip manufacturing so it can focus on chip/processor design. If this is the case -- widely expected -- it could lead to thousands of layoffs and the shuddering of some sizeable AMD fabrication facilities around the world.

CNBC: AMD Prepares 'Big' Announcement

We've all been down this rumor road before, many times before, regarding AMD's Asset Lite plans and what it means. It does appear to be true that AMD has called an analyst meeting for tomorrow, so something about something will likely be said

 

[exar333]

Definitely the way to go in these tough economic times, shed all of your valuable assets....oh wait...

 

[myocardia]

Originally posted by: ExarKun333
Definitely the way to go in these tough economic times, shed all of your valuable assets....oh wait...

Haha, I know what you mean. I guess if it keeps you from filing for bankruptcy, it might not sound so bad, though.

and the shuddering of some sizeable AMD fabrication facilities

BTW, has NBC gone so far downhill that they're now hiring reporters who don't know the difference between shudder and shutter, so obviously must not have graduated from high school, much less college?

 

[Idontcare]

Originally posted by: myocardia

Originally posted by: ExarKun333
Definitely the way to go in these tough economic times, shed all of your valuable assets....oh wait...

Haha, I know what you mean. I guess if it keeps you from filing for bankruptcy, it might not sound so bad, though.

and the shuddering of some sizeable AMD fabrication facilities

BTW, has NBC gone so far downhill that they're now hiring reporters who don't know the difference between shudder and shutter, so obviously must not have graduated from high school, much less college?

Finding a mainstream media journalist that knows jack about the semiconductor industry (the technology or the business) has about the same odds as finding common ground regarding Healthcare between McCain and Obama. Such journalists just don't exist.

Now why didn't the editor catch it? That's a good question, the editor should be a little embarrassed about that.

Hey did Viditor do a repost on this AMD stuff and bury my thread? Bad Viditor, but I forgive you.

Mods please go ahead and lock-up this thread, Vditor's has far more content.

 

[Zebo]

Al-AMD, buying high selling low since 2001.

 

[imported_apocalypse]

Originally posted by: Zebo
Al-AMD, buying high selling low since 2001.

lol!

 

[nyker96]

Originally posted by: Zebo
Al-AMD, buying high selling low since 2001.

I think at least they are trying something. Hopefully they'll conserve enough cash now to hire a good team and design a new Super Phenom to dethrone Intel and save me loads of cash when buying a new quad in the process ... or they'll crash and burn and be a footnote in history. Stay tuned.

 

[Idontcare]

Originally posted by: nyker96

Originally posted by: Zebo
Al-AMD, buying high selling low since 2001.

I think at least they are trying something. Hopefully they'll conserve enough cash now to hire a good team and design a new Super Phenom to dethrone Intel and save me loads of cash when buying a new quad in the process ... or they'll crash and burn and be a footnote in history. Stay tuned.

Irony considering AMD owes its life to DEC's collapse resulting in the freeing up of their Alpha design team (including Dirk Meyer the current CEO of AMD) which AMD hired to design the K7 Athlon.

AMD's demise would be someone else's beginning, Nvidia perhaps.

 

[taltamir]

what baffles me is that AMD says that its fabs spend time being offline because they don't need all of their capacity compared to how many chips they are selling... which spinning off should solve... which baffles me, is there some law preventing them from selling fab time? well actually there might be with their TSMC where they cannot compete with them. And splitting it up into a seperate company might bypass that.

 

[Idontcare]

Originally posted by: taltamir
what baffles me is that AMD says that its fabs spend time being offline because they don't need all of their capacity compared to how many chips they are selling... which spinning off should solve... which baffles me, is there some law preventing them from selling fab time? well actually there might be with their TSMC where they cannot compete with them. And splitting it up into a seperate company might bypass that.

Not at all, in fact that is the technical difference between pure-play foundries and open foundries.

Texas Instruments was an open foundry...the fabs were built to produce chips of TI's design (cellphone chips, memory chips, etc) in addition some fab capacity was allocated to fabbing SUN's CPU's.

IBM is an open foundry...the Fishkill fab exists to produce the Power processors, but the extra fab capacity is available to the highest bidder.

Samsung is an open foundry too, although not well known for it.

TSMC is a pure-play foundry. They do not design IC's to produce in their own fabs. 100% of their fab production is generated for customers who designed the IC's.

UMC, SMIC, Chartered...all pure play foundries.

So why then, you might ask, did AMD not pursue being an open foundry like TI and IBM when they admittedly had all this excess capacity? Well for one they saw how badly such a business plan was going for IBM. Another reason was their advanced process tech was SOI, not bulk. Inherent cost increase with that, plus unfamiliar territory for some fabless guys.

Plus it takes a LOT of effort to court customers and give them the confidence they need to convince their board of directors to allow them to invest their company's future in AMD's process tech roadmap. AMD has been the shakiest of fab owners these past couple years, not the best marketing material to convince a fabless company to bet the farm on AMD existing in 2010 or delivering 32nm on time in 2010.

 

[Rhonda the Sly]

Would have been a lot more amusing if they called the second company ATI.

 

[Foxery]

Originally posted by: Rhonda the Sly
Would have been a lot more amusing if they called the second company ATI.

Their financial backing is coming from a company called Advanced Technology Investment... it's close.
(DailyTech link, now that the news is official.)

Originally posted by: Idontcare
So why then, you might ask, did AMD not pursue being an open foundry like TI and IBM when they admittedly had all this excess capacity?
...
AMD has been the shakiest of fab owners these past couple years, not the best marketing material to convince a fabless company to bet the farm on AMD existing in 2010 or delivering 32nm on time in 2010.

Splitting and a huge cash infusion could/should make this a much more realistic opportunity for them. The dedicated "Foundry" company will have that same excess capacity, plus the new fab in New York, plus the free time to court customers, plus newfound stability to woo them with. I hope it works out well for them.

 

[taltamir]

eh, splitting up is just splitting up, although as idontcare mentioned, there is the whole future of the company deal, i guess people might feel more at ease signing a contract with "foundery company" then with AMD... the only biggest benefit I am seeing is the United Arab Emirates government committing billions to this new foundry company.

 

[nyker96]

Originally posted by: Idontcare

Originally posted by: nyker96

Originally posted by: Zebo
Al-AMD, buying high selling low since 2001.

I think at least they are trying something. Hopefully they'll conserve enough cash now to hire a good team and design a new Super Phenom to dethrone Intel and save me loads of cash when buying a new quad in the process ... or they'll crash and burn and be a footnote in history. Stay tuned.

Irony considering AMD owes its life to DEC's collapse resulting in the freeing up of their Alpha design team (including Dirk Meyer the current CEO of AMD) which AMD hired to design the K7 Athlon.

AMD's demise would be someone else's beginning, Nvidia perhaps.

I was dreading the possibility of AMD going out of business for a whuile. But I think you are right, those talented engineers will just move and shake this CPU industry from another company. But god forbid if they all going to Intel ...

 

[Idontcare]

Originally posted by: nyker96
I was dreading the possibility of AMD going out of business for a whuile. But I think you are right, those talented engineers will just move and shake this CPU industry from another company. But god forbid if they all going to Intel ...

If they go to Intel they will still be put to good use. The world will always be a better place because good people exist in it.

Where the negatives come out in this is the short-term turmoil those good people and their families experience as their futures have so much uncertainty injected into them.

The people who have the most to fear in this kind of turmoil are the mediocre employees who have no place in the industry when a 10,000 job market shrinks to a 8,000 job market. Fewer and fewer places for the Wally's to hide and take home a paycheck.

 

[pm]

what baffles me is that AMD says that its fabs spend time being offline because they don't need all of their capacity compared to how many chips they are selling... which spinning off should solve... which baffles me, is there some law preventing them from selling fab time?

I can't imagine that there's a law - in fact, I'm pretty certain that there isn't. But there are some good business reasons why being a foundry is hard. One of the biggest is that when you have spare capacity is usually when the customers aren't that interested in using a lot of it (ie. a recession), and when you need your capacity most is when your customers want it most as well (ie. boom times). So, you need attract your customers by promising that they will get parts on the process even when times are good. Another big issue is one of intellectual property - if a customer wants to use your process technology then you need to give them lots of details about your process technology and this is often a difficult thing for a company to do (technology companies are big into secrets).

Another issue is that customers using say for example, TSMC's fab process, will have a fair bit of work to get their design working on a totally different process - particularly, say, an SOI process. The design rules for the two processes - about how far apart metal corners can be to other metal edges, how close you can put the active material to the edge of the well, etc - will be different and so you need to spend a bunch of engineering time moving things around to get the design to work on the new process (CTho9305 wrote a really nice post here at AT about how hard this is, but I don't have a link handy).

And lastly, another issue is that once you let your customers use your process technology, it limits you from "tweaking" it. So, if your new CPU design comes in a bit higher power than you'd like and you want to lower leakage, then you'll also slow down the transistors (lower IDSat) but you can't do this because the other company's designs on your process are tuned to the process and don't want their designs running slower. So you are limiting yourself greatly from one of the really cool things about owning your own fab which is tweaking the process to fit a given design.


* I am not acting as a representative of any company - my opinions are my own

 

[Idontcare]

Originally posted by: pm
And lastly, another issue is that once you let your customers use your process technology, it limits you from "tweaking" it. So, if your new CPU design comes in a bit higher power than you'd like and you want to lower leakage, then you'll also slow down the transistors (lower IDSat) but you can't do this because the other company's designs on your process are tuned to the process and don't want their designs running slower. So you are limiting yourself greatly from one of the really cool things about owning your own fab which is tweaking the process to fit a given design.


* I am not acting as a representative of any company - my opinions are my own

Not wanting to nitpick or detract from the overall correctness of your post but I did want to chime in regarding this last point and say that this isn't necessarily a true limitation.

It can still be a limitation due to corporate policy or fab management, but its a management limitation if it exists.

It is entirely plausible (I know, I saw it in action every day for a decade) to run multiple process node flavors in parallel in the fab. It just takes difference recipes on the tools combined with a management system that tracks the process flow by device.

The Foundry Co can most certainly run AMD's chips with a CTI philosophy, iteratively tweaking the implants conditions, spacer thickness, etc with every wip cycle while maintaining an entirely static process node for other wip in the fab for other customers. SOI and bulk can be ran in the same fab on the same tools as well (with some end-effector calibration of course).

Its also the same manner by which a fab handles a wip mix spanning multiple process nodes (90nm chipsets, 65nm CPU's, 45nm...etc, fab conversions are not binary outside Intel and I suspect they aren't binary inside Intel either).

We perpetually operated this way at Texas Instruments with 3 different process node sub-versions.

But it does elevate the fab wip management complexity, and for that reason the Foundry Co may artificially invoke a policy of static process tech outside of yield-enhancement (d0 reduction) tweaks.

 

[CTho9305]

Another issue is that customers using say for example, TSMC's fab process, will have a fair bit of work to get their design working on a totally different process - particularly, say, an SOI process. The design rules for the two processes - about how far apart metal corners can be to other metal edges, how close you can put the active material to the edge of the well, etc - will be different and so you need to spend a bunch of engineering time moving things around to get the design to work on the new process (CTho9305 wrote a really nice post here at AT about how hard this is, but I don't have a link handy).

Thanks. I've copied my longer posts here; that post in particular is here.

The Foundry Co can most certainly run AMD's chips with a CTI philosophy, iteratively tweaking the implants conditions, spacer thickness, etc with every wip cycle while maintaining an entirely static process node for other wip in the fab for other customers. SOI and bulk can be ran in the same fab on the same tools as well (with some end-effector calibration of course).

Do R&D costs go down if you don't do CTI, or do your costs for developing the next process node go up enough to cancel out any savings?

 

[Idontcare]

Originally posted by: CTho9305

The Foundry Co can most certainly run AMD's chips with a CTI philosophy, iteratively tweaking the implants conditions, spacer thickness, etc with every wip cycle while maintaining an entirely static process node for other wip in the fab for other customers. SOI and bulk can be ran in the same fab on the same tools as well (with some end-effector calibration of course).

Do R&D costs go down if you don't do CTI, or do your costs for developing the next process node go up enough to cancel out any savings?

Yes the R&D costs would decrease if CTI (and any tediously incremental process of improvement) were eliminated...but presumably the overall business cost structure would then become higher than it would have otherwise been as CTI is intended to improve margins (newer SKU's with higher APS's, existing SKU's with lower cost, etc).

CTI doesn't necessarily reduce the cost or time barrier to developing the next iteration on the process tech branch. The timeline here usually thwarts this from being a practical consideration because you are talking about something exiting the hands of the R&D engineer and taken over by the willing hands of the manufacturing engineer...by the time the manufacturing engineer has carried out the process of implementing the improvement the R&F engineer will have long since iterated their version of it on the R&D side and it has likely evolved into something at that point which is markedly different from the manufacturing version spun-out many many months prior.

Some people may be wondering why doesn't everyone do CTI in some form or another. The short answer is that most do but they don't call it CTI. It will be called yield enhancement or transistor targeting, etc. But the magnitude of the allowed changes are no where near what AMD does with CTI.

The primary reason (to my understanding) most IDM's avoid CTI in practice has to do with reliability. Every process tweak carries with it the possibility of critically undoing a reliability metric (NBTI, TDDB, etc) which has a liability risk that usually far outweighs the relatively minor gains made by any given CTI. (think Nvidia and their recent debacle) So to have a safe and effective CTI you need a robust reliability assessment and characterization process.

Not many IDM's have the resources or the corporate culture to implement this to the degree that is necessary for a CTI approach. And some IDM's (Intel) simply maintain enough of a process tech lead that their time-zero release is just good enough to dominate for the lifespan of the node anyway.

AMD's version of CTI is effective in (large) part because of their APM. APM provides the process control (and feedback, and in-situ characterization needed to know quickly when something has not gone to plan, etc) needed to create an environment where risk-aversion is dialed back enough to foster decision maker's willingness to tweak the process flow and not be paranoid concerned that doing so is going to trash their personal annual bonuses from tanked fab yields.