Slashdot Mirror
AMD To Spin Off Fabrication From Design Work
I.M.O.G. was one of many readers to write with the news that "Advanced Micro Devices plans to announce Tuesday that it will split into two companies — one focused on designing microprocessors and the other on the costly business of manufacturing them — in a drastic effort to maintain its position as the only real rival to Intel. 'This is the biggest announcement in our history,' said AMD's chief executive, Dirk Meyer. 'This will make us a financially stronger company, both in the near term and in the long term, as a result of being out from the capital expense burden we have had to bear.'"
There seems to be some good information here:
IC production facilities are expensive to build and maintain. Unless they can be kept at nearly full utilization, they will become a drain on the finances of the company that owns them. The foundry model uses two methods to avoid these costs: Fabless companies avoid costs by not owning such facilities. Merchant foundries, on the other hand, find work from the worldwide pool of fabless companies, and by careful scheduling, pricing, and contracting keep their plants at full utilization.
So AMD frees its production facilities to accept contracts from other fabless companies. Meanwhile, they can focus on designing and selling chips and chipsets for motherboards and graphics cards.
I think this will turn out well for AMD, if they can maintain a good relationship with their foundry spin off and if the foundry spin off can keep up with the competition in terms of quality and technology. Although, I guess this would also free AMD to find other partners if they need to either expand production or find better production facilities in the future without neglecting parts of their own business.
Funny, my company just did the opposite. Our design department was just recently merged with manufacturing. This was done because:
A) Design would rarely factor in the manufacturability of it's designs, driving up costs.
B) Manufacturing had a tendency to sacrifice quality to reduce costs.
This new corporate structure has only been in place for a few months, but so far has worked quite well. Entire product lines have been eliminated (design didn't know manufacturing was still making the old stuff). Entire processes have been eliminated (manufacturing thought they were needed to meet the final spec, but weren't).
Most of these issues could have been resolved with better management and communication, but when design and manufacturing are a single unit, these issues resolve themselves naturally.
One of our competitors trademarked the term "hypothesis". From now on, we will call them "boneheaded ideas".
If they are a single company, then, internally, the two groups almost have to use each other or else seem bizarre. I.e., if the designers contracted out fabrication of a model even though their own fabrication division was not fully utilized, that would seem unhealthy. By the same token, if the fabrication division pre-empted production in-house designs for a third-party, that would similarly look bad.
With that view, it would be a tad harder for the fabrication portion of the business to attract design companies, with prospective companies knowing they are putting their manufacturing capabilities in the hands of a company that would be both partner and competitor. The conflict of interest is far from appealing.
Few large corporations under typical circumstances preserve in-house at-scale manufacturing. I.e., most x86 system vendors now at most design the system and then feed to another company for fulfillment, potentially even a company spun off from themselves when they reached a similar conclusion.
As consumers, we don't stand to lose, only to gain. For example, if nVidia has been held back in any quality/performance way by inferior fabrication companies, they may now approach AMD fabrication. Same goes for AMD v. Intel, if another fab company can deliver more aggressive process size/yield improvements, then AMD design can go to that company and produce a valid competitor to Intel.
Or it shows that both halves of the company were completely average nowadays even in only the context of their similar competitors, and still doesn't do well, but that isn't different from today.
XML is like violence. If it doesn't solve the problem, use more.
So the bottom line is that the Abu Dhabi Government is buying AMD?
"The ferrets, they're every where I tell you!"
I don't see anything in here that requires two separate companies.
From an organization or technology perspective, no, there isnt any reason to split into two companies. From a financial perspective, this makes a huge amount of sense. You'll note that this new company is receiving an enormous amount of new funds from investors, and taking a lot of the AMD debt. They're effectively splitting off R&D from Manufacturing, and people are free to invest in just the R&D component or just the manufacturing component.
Some folk out there thought the manufacturing side was worth a huge investment of cash. Lots of assets there that are worth a fair bit - but not if they're tied to work purely on AMD products. If the R&D side of AMD failed, then there are a crapload of perfectly good assets that would be lost, in effect. This allowed investors separate AMDs chips from AMDs fabs when investing. There is no inherent value in splitting the fabs off - except when someone is willing to spend $8 billion to fix them up, separate from the R&D side.
This doesn't mean they think AMD R&D is going to fail - its just about risk. Why tie your $8 billion investment to the ADD of the consumer chip market instead of to physical assets that will be worth something regardless of the mood of the x86 CPU market?
.
Transmeta failed because its product sucked.
Absolutely. For those who don't know, the Crusoe uses a VLIW architecture with 128-bit words, and x86 instructions have to be decoded and RE-ORDERED in real-time into those 128-bit words. This is the same brick wall Intel ran into with optimizing compilers for the Itanium, but unlike Intel the Crusoe has to do it in REAL TIME.
Sure, the software translation layer meant that they could run Crusoe any architecture, but in the end it cost them precious performance. The chip itself wasn't much to sneeze at (two integer units and an anemic FPU), so it really didn't have the performance to spare. Then they hobbled the chip by integrating a nortbridge; this meant that ALL Crusoe-based systems would have the same video and I/O performance limitations, all in exchange for saving a buck or two on parts.
It didn't help that they hyped the successor, the Efficion, and then it didn't deliver in clock speeds or promised performance increases.
Man is the animal that laughs.
And occasionally whores for Karma.