Monday, May 29, 2006

Third Factory, 300mm

"As global demand continues to rise for AMD products, we are scaling our manufacturing capacity intelligently to meet our customers' growing needs"
"We expect that AMD's momentum and demand for our products will continue to increase, fueled by our technology leadership, our commitment to customer-centric innovation and the continuing global antitrust scrutiny that will ensure truly competitive markets."
hrough the combination of leading-edge equipment, Automated Precision Manufacturing (APM) and the great people of Dresden, the plant will [reach] full capacity by the end of 2008"

Dr. Héctor de Jesús Ruiz, Chairman of the Board and Chief Executive Officer.

Is it justified a third factory for AMD?
How would the stock market appreciate this 2.6 G$ investment?
Are there other implications about this move?
These questions are what is going to be treated in this article.

1) AMD currently has a 200mm factory running at over 100% designed capacity (it is "borrowing" area from Fab36 among other tricks to accomplish this, according to Charlie Demerjian), by the time Fab36 will be reaching full capacity, substantial investments would be needed in Fab30, including upgrading to 300mm rather than 200mm equipment. It has been demonstrated that there is sinergy between AMD Fabs, look at what Charlie says about Fab 30 doing 150% what it was designed for thanks to spare space in Fab36; that helps to justify to put all the eggs in one basket. Besides, the German government loves this kind of economical activity, and everything seems to indicate that it will continue to underwrite, and even finance, the construction of facilities.

When Dr. Ruiz says "the Great People of Dresden", it is not just a courteous remark, it is true. The great AMD technologies benefit greatly from having top-notch technicians controlling its greatest manufacturing process APM. To be inserted in the heart of Europe is another advantage.

2) This leads me to Wall Street's appreciation: The Street loves when a company puts money directly into the "Capital" section of the Assets ledger, these investments increase the book value of AMD. Better than that, the financial help the German government would provide it going to be seen as a free-bonus. This is no idle or spare production capacity neither, the Street acknowledges that AMD has very good semiconductor manufacturing technology, thus, it may not believe AMD to be able to sell a 100% processor full production, but it doesn't require much imagination to understand that the capacity will be used anyway.

3) The announcement itself is a confirmation that AMD is jacking up its expectations. Fab30 all alone could supply 22% of the market, but let's say that it was using tricks. We can be sure that Fab30 was supplying 18% of the market. Fab36, with 300mm wafers can produce more than double 200mm Fab30, thus, Fab30 + Fab36 are enough for 54% of the market, consequently, the radical upgrading of Fab30 (Fab 38) means that AMD is aiming for more than half of the market (!), significantly above the 30% that it has officially declared to aspire to.

Does AMD has enough steam to aspire to more than half of the market?

A quick overview of Intel's general situation provides a grim outlook for that company:

  1. The transition to 65nm was neutered by AMD's 90nm sSOI; pushing them further down the path toward 42nm. But without SOI, or really deciding whether it is time for immersion technologies or not, the efforts put into 42nm so prematurely may prove rushed and wasted
  2. A significant portion of the benefits to go to smaller scales is lost because of architectural defficiencies. In particular, this thing of the gigantic caches that Intel processors have to try to compensate for not having DCA/ccHTT/HTT can not continue, because the physics of the shrinks determine decreasing power/performance benefits. In particular, Core2 65nm processors are actually larger than the 90nm products they come to replace, thus no economies in production. And just like it was explained before, smaller scales have greater power leakage overall.
  3. Core2 are dauntingly complex µ-procs. This means that there is no architectural improvement coming; Intel decided to rely solely in shrinks to improve speeds and performance. This is not only my opinion, but also that of Ars Technica's eminence Jon Hannibal Stokes: "Core's performance will scale primarily with increases in the number of cores per die (i.e. feature size shrinks) and with the addition of more cache, and secondarily with modest, periodic clockspeed increases". Athough he doesn't says it explicitly, after reading his article one reaches the conclusion that Intel lwon't be able to really do any significant modification to this µ-arch., it would be easier to start over with Banias again.
  4. Also take into account that the complexity may induce nasty problems.
  5. Intel also has a dilemma about Itanium. Should it kill it? Should it keep funding that losses generator?
  6. What about Intel factories? Some say that Intel has worthless factories that it hasn't written off trying to defer the decomissioning expense for better times. In any case, "CopyExactly!" seems so primitive when compared to APM...
  7. Still has to develop the equivalents of SOI, DCA, HTT, ccHTT, APM, etc. to really catch up to AMD instead of "fleeing ahead" to smaller scales.
Having demonstrated that Intel will cede more market share, we can also remember that more and more consumer electronics is being transformed into some sort of personal computer technology. For instance, I watch VHS, DVDs, cable and public TV on my laptop, using my custom VLC Server/Client streaming setup over my wireless network. I also believe that the video game consoles stimulate rather than discourage the market for computers. Because who is going to chat with fellow players through the console rather than a computer?, besides, the entry-level barriers are lower for Computer games than consoles, thus the highest creativity will be expressed in computer games.

AMD's "Ecosystem" approach also has a lot of potential: Coprocessors, and partners with vested interests in pushing AMD forward. Remember that AMD leaves a lot more of opportunities to add value to partners than Intel

Confirming that AMD is closer to the bleeding edge of technology than Intel, there is the potential of things such as Z-Ram (Zero Capacitor RAM with 5 times higher densities than regular six-transistor flip-flops). I don't know about the introduction rates for semiconductor technology, but it seems that very soon, like in the scale of one or two years, AMD will be able to include Z-Ram either as L2 or L3 cache memory. All that separates AMD from glory (and Intel from demise) is 8 Mb of cache internal to the µ-proc. Z-Ram may be precisely that. Remember that the gigantic L2 caches is the snorkel that prevents the already shit-submerged Intel from drawning, Z-Ram may nullify that...

AMD does have the package to get to over 50% market share:
  1. Good, proven management
  2. Good, proven, exciting technology
  3. The brand name has appreciated significantly, creating a loyal customer base
  4. Technological leadership assures media exposure, free good publicity, public awareness, toghether with the enthusiasm of partners in the ecosystem
  5. The ecosystem itself
  6. Production Capacity!
The stock market is very myopic, that's why it thinks that a duopoly is not as profitable as Intel's monopoly; but it is not true. Once the technology market gets used to the competitive struggle among AMD and Intel, I predict that an era of technology flourishment will began. It happens every time a key technology gets developed by leaps and bounds because of competitive struggles. That's where the profits will come from to compensate the smaller margins.

My theory is that AMD is very cheap at a market capitalization of less than 23% the combined market caps of Intel and AMD, because that is the true value of AMD given its market share; which today is 121 G$ * 23% ~ 27.8 G$, which today means $57.50 per share.

In the future, I will cease to invest in AMD when it reaches an inflation adjusted value of todays 61 G$, or $120 per share. This is a four-bagger.


Anonymous said...

that's why it thinks that a duopoly is not as profitable as Intel's monopoly; but it is not true

hey..Wall street just stuck in the is easy when all the profits were going to the 3 they have to do some work because the horizons of the AMD ecosystem is broader and allows more players...go to work Wall street!!!


Anonymous said...

Errr... 3rd factory? (I though it was replacing an existing factory - that would make 2)

Your analysis on capacity is way off, several obvious issues:

1) you assume F36 will be double the output of F30 - that would be true IF they both had same wafer starts/month (which they don't) AND 300mm has same yield as 200mm - I haven't seen any data that 300mm yield = 200mm yield for AMD.

2) I believe <10% of AMD's current production is dual core, as they migrate more to dual core they will get fewer die/wafer (unless of course they continue to plan to only make 10% dual core...) And any quad core production will cut into capacity even further. That cuts the legs out of your 300mm will provide 2X the dies of 200mm argument (it will be ~same if you account for migration from single to dual core)

3) The conversion of F30-F38 will actually limit their production during that conversion to <2 fabs (they still may be able to do some 200mm while they start bringing in 300mm tools). Another thing to keep in mind is 300mm automation and material handling systems are completely different so it is not just a matter of swapping 200mm tools for 300mm tools.

FYI - 200mm-300mm conversion is not "radical upgrading"; numerous IC manufacturers have done this.

Anonymous said...

Just a (few) problems with your knowledge of process technology:

"The transition to 65nm was neutered by AMD's 90nm sSOI; pushing them further down the path toward 42nm."

It's 45nm, I thought it may have been a typo until you repeated it again.

Your other link says: "The wires are thinner, the resistance is higher, the lost power greater."

Ummm... the major power deltas between 90nm and 65 nm is off state leakage casued by leakage between the gate and channel and from source to drain (neither of these have anything to do with "tiny wires". Also SOI doesn't impact either of these - the main benfit of SOI is latch up protection which can be minimized much more cheaply with better implant processed. (SOI adds at least 10-15% to production cost of a wafer)

"APM, a robotic tech. for chips production"

Do you actually think APM has or uses robots? It is just a SW control system - all other major manufacturers use their own systems (TSMC, Intel, Samsung, etc) - it is impossibel to compare the systems as none of the companies disclose the specifics of how and what they do.

"First fallacy: It is cheaper to manufacture the same number of transistors at 65nm than 90nm"

As you get roughly double the transistors your yield rate would have to go by a little less than a factor of 2 for it to be cost neutral (the reason it is less than 2 is that there are additional manufacturing steps and 1 metal layer which adds cost to the 65nm process. Your analysis is not congruent with your statment as you are comparing an older technology on single core (or 1/2 cache) to the newer technology with dual core or (double the cache). Your statement above says same number of TRANSISTORS (the 65nm example you give has many more transistors, thus you are comparing apples to oranges)

I'm not sure what you're background is in, but it is clearly not semiconductor process tecnhnology or manufacturing.