IBM's bid to one-up Intel next year with 28 nm processors

Last February, Intel made some changes and adaptations to its processor roadmap in what was generally perceived as a sensible move in light of the current economy: It's expediting its move from the 45 nm to the 32 nm generation of CPUs with increased investments in facilities, but then extending the market lifespan of the 32 nm generation to compensate, and to help reap back the costs incurred. That extension will include the introduction of a "mainstream" 32 nm architecture code-named Westmere, as part of its continued strategy -- successful so far -- to introduce certain elements of its newer designs to a broader market of buyers first.

That strategy was confirmed Monday during Intel's quarterly conference call -- where it also revealed sharply lower profits on much lower revenue: "We have pulled in Westmere, our first 32 nanometer product family, and will now be shipping those products later this year," reported CEO Paul Otellini (our thanks to Seeking Alpha for the transcript). "We have shipped thousands of Westmere samples to over 30 EOM customers already. We also look forward to the launch of our new consumer ultra-low voltage products, which will enable many new...light notebooks at very compelling price points."

The IBM-led Technology Alliance needs any opening it can get, and whether this is the one or not, it's going to try for it. For the group that now includes Global Foundries, the spinoff producer of CPUs for AMD, Intel's move is an alteration of its self-named "tick-tock" timing. The Alliance will try to take advantage of that phase shift, if you will, by accelerating the pace at which it moves its members to a new 28 nm low-power generation, according to an IBM announcement this morning.

"Through this collaboration, IBM and its alliance partners are helping to accelerate development of next-generation technology to achieve high-performance, energy-efficient chips at the 28 nm process level, maintaining our focus on technology leadership for our clients and partners," stated IBM R&D chief Gary Patton in this morning's prepared statement.
The key to this move will be more rapid implementation of high-k-plus-metal-gate (HK+MG) technology, a breakthrough formula obtained in a race-to-the-finish with Intel that resulted, in January 2007, in Intel beating IBM to the announcement by less than two hours. From the beginning, both IBM and Intel promised that their respective formulas would enable foundries to shrink die sizes without overhauling their production processes...but Intel's needed the overhaul anyway with the Nehalem generation (the first to use HK+MG across the board), so it's never had the opportunity to put that theory into play.

IBM, with partners such as ST Microelectronics, Global Foundries, Samsung, and Chartered Semiconductor, will now actually give this theory a try. Today, the Alliance is saying it will introduce partners to a process that lets them go ahead with their 32 nm production plans, while safely transitioning the back end of their 32 nm roadmaps to become 28 nm products.

The Alliance already shipped its first evaluation kits to potential implementers last December, following that up with a public evaluation kit last month. This morning's signing on of Global Foundries means that the producer of AMD's CPUs has already made the evaluation, and could soon announce it's ready to go.

The goal for the Alliance for now is to enable "early risk production" versions of 28 nm processors to be made next year. That's a term that's one step removed from "early adoption tests" -- it means something that could be introduced to high-class customers, maybe not yet the mainstream. Last November, AMD introduced its own revised roadmap, which featured the intention to produce a 32 nm "Liano" architecture CPU in 2011. If this plan goes through as the IBM Alliance suggests it will, that architecture could conceivably become the 28 nm generation without shaking up AMD's roadmap a second time.

And that would mean AMD has a 28 nm CPU for the mainstream while Intel is still burning out its 32 nm generation. It's not a certainty by any means -- success depends on any number of factors turning out in IBM's and AMD's favor. But we know what the tunnel looks like now, because someone, somewhere, just turned a light on.