Tuesday, 15 February 2011

Intel's ARM replacement?

So Intel has announced their ARM-killer device: the "Medfield".

Medfield is a power-constrained dual core Atom. Although Intel hasn't said as much, it's reasonable to expect that they'll be making system-on-chip (SoC) devices using the Medfield core, particularly given recent acquisitions that enhance their silicon portfolio to include modern wireless and graphics support.

Will it be enough?

Unlike Microsoft, Intel doesn't have a great history of entering markets opened by its competitors
It's a hard one to predict. Unlike Microsoft, Intel doesn't have a great history of entering and then dominating markets opened by its competitors. In fact, it's often struggled to retain control of its own core markets. AMD, for example, has taken and retained a significant portion of Intel's market share with its lower-cost x86 processors. That's despite being a market follower, not a market leader, and hence always a generation behind its larger rival. (It's notable that AMD hasn't attempted to compete in the mobile device space, preferring to concentrate on CPUs for desktop and server products.)

ARM CPUs achieved their popularity because of three killer benefits: great processing power per watt, a really low price:performance ratio, and small physical size, all of which earned them a massive lead in the small embedded systems market.

But that's not why ARM dominated their competitors. The real reason has little to do with those benefits. ARM achieved their lofty position by moving from being a chip manufacturer to being a processor core licensor. At a stroke, they improved their cashflow, as intellectual property (IP), once created, doesn't have a lead time to manufacture, and you don't have to warehouse huge stocks of IP somewhere air-conditioned, waiting for someone to buy it: "dead money" until the customer orders them and pays the invoices. They made those difficulties, endemic to manufacture, someone else's problem.

That's why ARM survived successive boom-busts, but the real success came from the number of companies who licensed ARM cores and embedded them in their own products. It reads like a high-school register call of everyone in the embedded device space. I'm not going to list them here: go to ARM's site and read it. It's long - no less than 158 licencees and foundries - and very impressive.

Intel's biggest problem is going to be that it's going to be fighting a war on 159 fronts
Intel's biggest problem is going to be that it's going to be fighting a war on 159 fronts - including ARM, of course - if it wants to claim for itself some of ARM's market. It's going to be doing so against opponents who are well armed, well-entrenched and fanatical to their cause. It's going to be trying to make alliances with sovereign powers that are well-disposed to ARM, cynical of Intel, and little inclined to switch, unless the benefits over ARM are very, very compelling.

Intel's Atom offerings to date haven't exactly set the world alight - well, unless you count heat dissipation. Potential partners have responded to Intel's bugle with a resounding..."Meh." Those that did venture to make Atom-based products found that their netbooks, supposed to change the portable computing battlefield, instead met a "Meh" from the buying public too.

This has put Intel in a very difficult position. The Atom was specifically designed for portable computing devices, and the indifference that met netbooks has, by association, hit Atom too. Intel is scrambling now to find a niche in which Atom can play successfully, and my feeling is that Medfield is probably their last serious attempt to achieve a real share of a sector in which they've never really sat easily - and their partners, burnt in the netbook fiasco, are going to need a lot of persuading to put big money behind Atom a second time.

Intel's Atom customers have a single point of supply; a single point of failure
And let's not forget the point I made above, the biggest difference between the two companies. ARM is a licensor, with a huge number of partners all making their own devices, and a very healthy cashflow that can only be damaged if manufacturing problems hit all their licensees at the same time. Intel is a chip maker, and - as the recent $700m recall of the "Cougar Point" chipset demonstrates - its customers are very vulnerable to Intel manufacturing issues: a single point of supply (for Atom devices) means a single point of failure, whereas they can buy ARMs from almost anyone.

I have no doubt that Intel will be marketing Medfield aggressively - and maybe, just possibly, that might tip the scales. But I can't help but think that it's the wrong product, too little, too late. AMD seems doomed to be the Intel wannabe that can't win, but won't die. And now Intel tries to be the same thing to ARM.

Ironic, no?