Earlier this week
we covered a deal between Intel
that will combine Intel's Atom with Rockchip's graphics IP and considerable share of the Chinese market. I'm returning to the topic because there are some interesting big-picture facets to this move that we didn't cover with the initial announcement. Over the long term, this partnership could transform Intel's entire mobile division -- or cause it no end of headaches in the future.
First, in order to partner with Rockchip, Intel must have definitionally granted the company some form of x86 license. This wouldn't necessarily be a license to design its own x86 processors (in fact, it almost certainly isn't), but the only way for Rockchip to modify the basic SoFIA design is for Intel to give it permission to do so. The quad-core SoFIA variant that Intel is building with a rumored Rockchip Mali
core will be a fusion of both company's work.
The long-term implications for Intel's mobile division if the Rockchip partnership works are substantial. To date, Intel has had very little luck in smartphones. The company won a few designs with Medfield
and predicted its Merrifield
division would propel it into at least the upper midrange device segment. That didn't happen, partly because the part is late (it originally targeted a late 2013 launch), and, based on what we've heard, partly because Intel is struggling with aspects of its own Android implementation, particularly the GPU. That may or may not be true, but the net effect is certainly real -- eight months after launch, you can't buy a Bay Trail tablet running Android from any vendor, despite a handful of product demonstrations stretching back to last year's IDF
The Rockchip deal is going to take Intel into new markets on the back of a company that's used to competing with bottom-end designers like AllWinner in sockets that typically sell for $10-$20. The fact that Intel is willing to compete for these sockets says something about how desperate it is to build market share, but if the deal actually works it could signal a sea change in the company's business. If Rockchip has better luck bringing Intel products to market than Intel does, how much does Chipzilla need its own mobile division? Long term, this deal could evolve into a foundry-sharing arrangement in which Rockchip would design smartphone products around Intel's CPU and GPU IP, bring those cores to market, but possibly retain its own ARM
business as well.
Why Intel Wants The Phone Business
I think it's fair to ask why Intel, which has always sought to maintain high margins and a halo effect for its processors, is willing to partner with a company that specializes in budget ARM cores. It's not just about winning market share in a new segment -- the old Intel would have wanted to partner with an upscale Chinese company that pushed higher-end devices. Nor is it strictly a matter of competitiveness; matched marketing funds have a competitiveness all their own and Intel's Silvermont CPU is competitive with ARM's Cortex-A15 in power and performance, even if it can't quite match Apple's latest A7.
The reason, I think, for why Intel is willing to push into this space at any segment position is that it needs to claw out some
business to have a long-term chance at moving parts back to its own foundries. Right now, the SoFIA devices Rockchip will build are built at TSMC
but Intel is exploring moving those parts (and its modem business) back to their own fabs. That's a huge commitment of time and resources and management probably doesn't want to do it unless there's good reason. At the same time, Intel needs that phone business to keep its fabs running at full tilt. There are enormous fixed costs to idling a foundry, and right now the PC slump has Intel delaying 450mm
rollouts and pausing its Fab 42 buildout in Arizona. With the cost of next-generation process technology and fab deployments always on the rise, the company needs an entry point for the smartphone mass market if it wants to continue leading the semiconductor industry. It's just proven that it'll bet on $10-$20 SoCs to get there.