San Francisco (CA) – Intel is set to announce a completed development phase of its 32 nm chip manufacturing process, which is expected to deliver first commercial processors in late 2009. 32 nm chips will now move into the production phase, while a 22 nm technology, due in 2013, comes out of research and moves into development. Since Intel’s process technology development is based on a three-stage strategy, that also means that research on the 16 nm process has already begun.
To clear up any confusion upfront: Intel isn’t ready to mass-produce 32 nm processors just yet. There is still a lot of work to be done to prepare the process technology for the volumes Intel needs and we haven’t even heard much about the fabs that will those processors: Westmere, a Nehalem shrink, will be the first 32 nm CPU in Q4 2009, while the 32 nm Sandy Bridge will introduce a completely new architecture in 2010. What we do know now, however, is that Intel is on track to deliver 32 nm on time next year and at least about one year ahead of AMD.
It appears that Intel has mastered all the challenges of a 32 nm shrink. It will use a second-generation high-K gate material to contain leakage current and it will transition to a 193 nm immersion lithography production technology to print the circuits on its chips (AMD is using immersion lithography for its 45 nm chip generation, but goes not use a high-K material). Since the announcement of the first 32 nm SRAM chip in September of last year, Intel has reduced the cell size in the chip about 6% smaller (from 0.182 μm2 to 0.171 μm2) and claims to have produced “thousands” of test chips already.
What is somewhat strange about this announcement is the fact that, compared to previous process technology announcements (especially 45 nm, 65 nm), we already had received detailed information about power savings and performance improvements at this point in time of the development process. But there is no information on 32 nm yet. We are told that there will be power savings when compared to 45 nm CPUs at comparable clock speeds. We just don’t know how much. Performance is also a bit blurry, but we got the impression that 32 nm CPUs will scale to higher clock speeds than 45 nm chips.
There will be no changes to the wafer size: 32 nm chips will be printed on 300 mm wafers. Talks about 450 mm have begun, but this wafer size are still very much a topic for the next decade.
The obvious question here is: So, if Intel is at 32 nm – where is AMD? Well, we know that 32 nm AMD server processors will debut in 2010 in a best-case scenario, but consumer processors are now scheduled for 2011. Some readers may be surprised by this plan, especially since it seemed that AMD, thanks to its partnership with IBM, could challenge Intel as it was first to announce 22 nm products. However, it is worth noting that IBM was able to create only one functional 22 nm cell so far, while Intel has built millions of functional 32 nm cells – and typically does not announce a product with a single functional cell. The maturity of a new chip is also measured by cell and area size and the resulting density of a chip. And it appears that Intel still has the lead in this field.
With 32 nm now officially in the production phase at Intel, that means that the two stages before this phase will also change: 22 nm moves from the research into the development phase and if everything goes to plan, we should see the first 22 nm SRAM chips by September of 2009. 16 nm is now in the research phase and despite rumors that production processes beyond 22 nm may hit a wall, we are told that Intel does not see an end to this road yet.
Intel will provide more details about its 32 nm process at the International Electron Devices meeting (IEDM), which will open its doors on December 15 in San Francisco. The presentation will include an updated 291 Mb 32 nm SRAM chip with more than 1.9 billion transistors and a 3.8 GHz clock speed. Intel said it will also briefly discuss early findings relating to its 22 nm CMOS technology.