IBM has announced that its Common Platform partners have signed on to extend their chip fabrication technology d

evelopment agreements through 2010. So IBM, Chartered, Samsung, Infineon, and Freescale will all be working together on the next-generation 32nm process, with IBM partner AMD also reaping the benefits through an existing deal. (Recall that 32nm is when IBM will introduce its"airgap" technology.) All of these alliance partners will be competing in one form or another against Intel across a wide variety of markets and market segments, and IBM's announcement provides a good occasion to take a brief look at one oft-ignored factor that will affect who comes out on top in 2009-2010.

Most of the reporting that you'll see on these process transitions focuses on who's going to have their next-generation fabs online first. While it's a good thing to be the first kid on the block with the latest fab technology, there are other factors that complicate the Intel vs. IBM alliance picture.

Making the chip design fit the fab

The new Common Platform agreement includes both 32nm CMOS technology development as well as the joint development of process design kits (PDKs). These PDKs provide a standardized set of tools that will let any one of IBM's fab partners design integrated circuits (e.g., microprocessors, digital signal processors, memory modules) that can then be easily manufactured at any of the Common Platform fabs that the group shares. So the PDK is sort of like an API and a compiler that gives multiple vendors uniform access to the same fab technology. (This analogy isn't perfect, but it's good enough.) And like an API/compiler stack, the kind of abstraction that the PDK represents can have some side effects on performance.

The quality of a foundry-supplied PDK, including the amount of critical knowledge about a foundry's fabrication technology that this kit exposes, has a big impact on a chipmaker's ability to get the best yields, clockspeed, and overall performance out of a complex, cutting-edge product like a CPU or GPU. Because Intel, by way of contrast, does both design and fabrication under one roof, the company has the ability to design and optimize processors specifically for its in-house fabrication technology.

So when considering the IBM alliance's competitiveness with Intel, the date at which the two competitors begin 32nm production isn't the only "horserace"-type factor that matters. The other big factor is to consider is how well IBM and its partners are able to fit their product designs to the jointly developed process, and that's where the PDK comes in. Beyond the typical end results (e.g., product announcements and device characteristics), we probably won't be see much external evidence of how well (or not) the alliance members are making the shared process tech work for themselves.

IBM and Chartered Semi have been working together on process technology since 2002, when they collaborated on the 90nm transition. Samsung came on board in 2004, and now the three companies form the core of an extended alliance that includes Infineon, Freescale, and AMD.