IDF In five years' time, Intel will be selling more system-on-a-chip products than mainstream microprocessors. So said CEO Paul Otellini yesterday. It's a bold claim and one that warrants closer consideration.

Much depends on what Otellini consider an SoC. Most folk will think of them as small-scale chips for handhelds and embedded applications. And, yes, Otellini has his eyes on those roles. He specifically hopes Atom-based SoCs will take the company into new markets.

Incidentally, he hopes Atom will do the same for others. During his Intel Developer Forum keynote, he mentioned the chip giant's move to use Taiwanese foundry TSMC as a production facility for Atom-based products. That's not new, but this time he added that he want to allow third-parties to work with Intel and TSMC to add technology of their own to the platform.

We've discussed Intel's plan to take on ARM with x86 chips many times in the past - even before Intel sold off its own ARM product line to Marvell in 2006. Otellini's comments about the TSMC deal indicate he wants Atom to become a licensed architecture like ARM, one other chip makers can take and adapt to their needs, but retain a core level of compatibility.

That's doubly interesting given the plan, also announced yesterday, to encourage Atom-specific software development.

Looking ahead, Otellini revealed a basic Atom roadmap that will take the platform to 32nm, 22nm then 15nm. The 32nm versions is codenamed 'Saltwell', and is clearly a die-shrink from today's 45nm 'Bonnell'. But the 22nm and 15nm implementations will be new-core products, he revealed.

Bonnell, incidentally, is the overarching core design that led to 'Silverthorne' handheld tablet and netbook Atoms and 'Diamondville' desktop Atoms.

Each of these will feature in SoC platforms. One of the first is the little-mention 'Sodaville', an Atom-based part for consumer electronics kit. Indeed, Intel demo'd a handful of set-top box reference designs at IDF, all based on Sodaville.

Bonnell is also the basis for the upcoming 'Moorestown' MID platform which is now due mid 2010, though it won't appear in shipping devices until H2 2010, Otellini admitted. That's more time for netbooks to really take hold as the mobile browsing platform of choice.

Intel insists Atom isn't cannibalising mainstream processor sales, but it's clear that it wants to be shipping products for a lot more devices than PCs, and unless you count Intel's upcoming CPU+GPU products as SoCs - they are, but only if your definition of 'SoC' is very broad - then Otellini's statement implies that Atom will become a far more important product to Intel over the next five years than the Core i series and possibly Xeon.

We suspect Otellini was including more highly integrated PC CPUs in among 'true' SoCs for his sales forecast, along with new and upcoming integrated Xeon server chips.

But Atom remains the most important component of Intel's SoC strategy. It's the spearhead of Intel's latest push into embedded markets - Otellini claimed Atom has more than 460 design wins in this arena, many of them, we'd note, are the kind of apps you'd expect to find ARM chips in.

That's not to say Atom is beating ARM - manufacturers are conducting many, many more than 460 embedded application projects at the moment - but as these projects increasingly seek to bring internet connectivity on board, these two platforms can't help but become pre-eminent, and not just in the netbook vs smartbook battle.

Intel will be providing more details about its Atom-based SoC roadmap later today. ®