AMD shows The Foundry Company roadmaps

At its lengthy Financial Analyst Day conference yesterday, AMD gave attendees some insight into what The Foundry Company—its soon-to-be spun-off foundry business—will be doing over the next couple of years. If things go as planned, the new manufacturing firm will start cranking out bulk silicon wafers next year, and it will eventually manufacture chipsets and graphics processors for AMD.

Right now, AMD makes its own microprocessors using silicon-on-insulator process (SOI) technology, and it relies on Taiwanese foundries TSMC and UMC to build graphics processors and chipsets using bulk silicon. Likely because few companies aside from IBM and AMD design their chips around SOI tech, The Foundry Company will also have to start offering a bulk silicon process in order to attract other customers (and pad its bottom line). The roadmap above shows that transition will kick off in the second quarter of next year with a 45nm bulk silicon process.

Unsurprisingly, AMD intends to rely on The Foundry Company’s future bulk silicon production capabilities to build GPUs and chipsets. The second slide suggests that move will happen at the 32nm node in 2010, but AMD will apparently stick with 55nm and 40nm processes from TSMC and UMC until then. The second slide also says AMD will use both SOI and bulk silicon for future microprocessors and “APUs” (accelerated processing units). APU is AMD’s shorthand for x86 CPUs that have graphics processors built in, so it’s entirely possible AMD is simply referring to products that have an SOI CPU chip and a bulk GPU chip on the same package. The latest AMD client roadmap pins the launch of the first APUs in 2010.

AMD announced the spinning off of its manufacturing division in early October, and it plans to complete the operation early next year. The Foundry Company will then get its “permanent corporate name and identity,” and AMD will retain a 44.4% stake and equal voting rights with ATIC, the Abu Dhabi-based investment firm that will own the remaining 55.6%. AMD manufacturing chief Doug Grose will be in charge of the company, and former CEO Hector Ruiz will fill in as chairman.

Comments closed
    • ssidbroadcast
    • 11 years ago

    Here’s a thought; How will this effect the supply of AMD processors and GPUs when The Foundry is taking orders from other customers (could be anyone, Samsung, Hitachi, Apple, even nVidia)?

      • UberGerbil
      • 11 years ago

      This was discussed quite a bit back when the Asset Smart plan was first floated (and even earlier actually, whenever outsiders were suggesting AMD spin off its fabs… remember the “real men have fabs” quote is quite old).

      The rationale for Asset Smart, obviously, is that if demand for AMD’s products is insufficient to soak up all of TFC’s capacity, TFC can take on additional customers (thus competing with TMSC and UMC). This unlocks some value in TFC and potentially gives them a more reliable revenue stream (and more clarity in their books), allowing them to better handle the expenses involved with upgrading the fabs (and perhaps having an easier time with the necessary financing). So if AMD is seeing so much demand that it is fully consuming TFC’s capacity, it’s still a win for both of them.

      The problem, as you point out, is if TFC gets overcommitted: who gets bumped to the back of the queue? AMD’s near 50% ownership presumably buys it some kind of priority. But that may make it hard for TFC to pursue some profitable customers, who won’t like the idea of suddenly having their products delayed because AMD happens to be enjoying a spike in demand.

      In practice, of course, production slots get scheduled out in advance and spikes in demand can create shortages of popular chips (with attendant higher prices) for everybody, even companies like Intel that control their own fabs. All merchant foundries have the problem of juggling production capacity to try to please multiple customers. But the others don’t have a primary customer that is a major owner. So it’s probably more of a problem for TFC trying to drum up outside business than it is for AMD. (Another problem for TFC is the SOI process, since it makes it hard for potential customers to migrate an existing design to their production, and then threatens lock-in of anyone who does).

    • Suspenders
    • 11 years ago

    Why are CPU’s and GPU’s built using these different processes? Could they not just build both using one?

      • echo_seven
      • 11 years ago

      As I understand it, it’s because CPU designs emphasize different transistor characteristics vs. GPUs. For GPUs it is easier to solve a problem by throwing more transistors at it, and trying make sure you don’t exceed your TDP. For CPUs you can only throw so many transistors (branch predictor improvements, new instructions, more cache), before everything gets hit with diminishing returns. The CPU at this point must rely on high-performance, high-power transistors to get the absolute highest clockspeed (and thus performance).
      So I suppose this is why they use different processes for CPUs and GPUs.

    • Chillectric
    • 11 years ago

    l[

      • Cyril
      • 11 years ago

      The bars in the roadmap are ramps, AMD said later in the presentation that 32nm production will begin in the first half of 2010 for both bulk and SOI.

        • MadManOriginal
        • 11 years ago

        Ramps in the Roadmap….like Dukes of Hazzard? Yeeehaa!!!

          • Scrotos
          • 11 years ago

          Kew kew kew… them darned Duke boys!

    • sdack
    • 11 years ago

    I wonder who its other customers will be.

      • shaq_mobile
      • 11 years ago

      A little company called “Wayne Enterprise” perhaps…

Pin It on Pinterest

Share This