Well, it’s official: Intel is selling off its XScale unit to Marvell Technology Group for US$600 million. That’s quite a steal for Marvell, since previous reports valued the units on the block at US$10 billion. What Marvell gets for all that cash are, among other things, the XScale line of ARM-based embedded processors, the related PXA line of SoCs (system on a chip), and an undisclosed amount of debt associated with the acquired businesses. Marvell was already in the embedded semiconductor business, so in terms of products at least the sale is a good fit. Let’s hope the cultures mesh, as well, because Marvell has stated that they plan to keep most of the jobs in the units they just picked up.
Back when I first covered the rumors of that the ARM-based processor lines were up for sale, I (and everyone else, for that matter) talked about it the move in terms of Intel slimming down the company and focusing on their core x86 business. I’ve since had more thoughts about what that focus could look like.
I realize that what I’m about to suggest is complete heresy for the ARM and PowerPC fanatics out there, but what if Intel has gotten out of the ARM business permanently and the embedded business only temporarily? What if the company plans eventually to move Core down into the consumer side of the embedded market? Now, I’m not necessarily talking about the 65nm node, but what about at 45nm and smaller?
A Core 3 Solo Blackberry?
Intel went on a buying spree and snatched up all these embedded product lines at right around the same time that the company was ruled by what Intel CTO Pat Gelsinger has called the “speed demon” camp. This is the camp that brought us Netburst, and that believed that the process-driven clockspeed party was just going to keep on going. So if you believe that something like Netburst is the future of the desktop and server machine—fast, hot, and huge—then you’re probably also looking for something to stick in very low-power consumer electronic devices.
In fact, during the late 90’s, Intel was all about different architectures for different market segments. They had Itanium for the very high end, Netburst for the desktop, the Pentium M was in the works for laptops, and the XScale and PXA lines were in place for embedded applications and consumer electronics. That’s quite a bit of fragmentation at the level of microarchitecture and ISA, and it shows that Intel had begun to think of themselves primarily as a supplier of silicon in various forms to various markets.
The shift that CEO Paul Otellini is effecting now at the company has them transitioning to being a supplier of not of silicon chips of various kinds but of complete platforms—the processor, and supporting logic like the chipset, (NAND Flash) storage, and I/O (graphics, and wired and wireless networking)—aimed at different segments. Furthermore, the bulk of those platforms are now x86-based, and Intel’s entire x86 roadmap from here on out is 100 percent focused on delivering performance/watt.
With an aggressive architectural refresh strategy that sees a new x86 architecture every two years, and a company-wide commitment to making each new x86 architecture more power efficient than the last, is to too far-fetched to speculate that Intel will at some point pitch a Core-based SoC for handhelds?
I don’t deny that, even in 2008 an ARM-based SoC will provide superior performance/watt to a 45nm Core-based SoC. But the ARM-based SoC won’t run +90 percent of the world’s consumer software, while the Core-based SoC will.
The more I think about it, the more I’m convinced that while Intel may have abandoned XScale and PXA, they haven’t abandoned the vast and growing mobile electronics segment. I think Core is likely to wind up in a future must-have mobile gadget—the 2008 or 2009 equivalent of an iPod or a Blackberry. But feel free to prove me wrong in the discussion thread. After all, my track record for predicting portable gadgetry with Intel inside isn’t so hot.