Forgive me if you know this already, but an embedded computer essentially means a PC that doesn't look like a PC. So, for example, a chip at the heart of a kiosk is an embedded processor, even if it look and smells like an X86 chip. The difference between your typical embedded chip and a desktop device is that the former is much cheaper, runs at lower power, and eschews all the bells and whistles of a fancy-pants PC part.
Which is a long-winded way of saying that embedded processors are a volume game, because low cost equals low margin, means you've got to design and manufacture them expertly to make any money, equals you'd better sell a lot of them to justify the whole venture.
Which, in turn, brings us back to Atom. If you look at its original design, Atom is actually an embedded processor dressed up as a low-end PC chip. Indeed, Atom was launched specifically to create a new category of low-end PCs called netbooks. (OK, you could argue that Atom was built to take advantage of an emerging class of devices called netbooks, but that just turns it into a chicken-and-egg question, and my basic point stands.)
Fortunately for Intel, Atom has been spectacularly successful as a netbook processor. As Stephen Smith, VP of Intel's Digital Enterprise Group, told me about Atom in my post on Monday: "Adoption has been faster than our original forecasts and we actually had to react to put in place more assembly and test capacity to get them out the door. Netbooks have moved very quickly."
With the TSMC alliance, Intel is simply looking to turn lemons into lemonade. It's got a successful design in Atom. However, Intel sees a lot of unexploited upside here. That's because Atom isn't appropriate just for netbooks alone. As I explained about, it's useable in a myriad of embedded applications.
So here's where my angle about Intel CEO Paul Otellini comes in. Ever since Otellini took over in 2005, he's been searching for a way to diversify Intel beyond its single-prop reliance on PC processors. (Read my contemporaneous account of his ascension to the CEO-ship, here.) This is a very smart move, however difficult it is to do. First off, Intel has tried this before, with communications (aka cell phone and smartphone) processors, and it didn't work. (Intel ended up selling its ARM-based XScale processor operations in 2006 to Marvell for $600 million.)
Secondly (again, as I explained above), most chips that aren't PC chips don't have the profit margins of PC chips. (Even most PC chips don't have great profit margins, but that's a separate story.) So if you want to diversify, or lay off the risk, or expand (or whatever you want to call it) by building an embedded business alongside your PC-processor business, that embedded business is going to have to be much bigger, volume-wise, than your PC business to succeed. An embedded biz also is going to be much more difficult to operate -- execution is harder on low-margin parts -- and it's going to be way more price sensitive, too. (Can you say, NAND flash?)
I say this all while pointing out yet again that even the PC processor business, in which Intel has been extremely successful, ain't all that easy. (For reasons I've loosely alluded to but won't go into, because it'd double the length of the post. Let me simply say that there's ample downward pricing pressure in PC processors, too.)
OK, so in conclusion, I think my long-winded explication answers at least one of the questions whipping round the industry. That would be, why did Intel take the unprecedented move of partnering with TSMC, in an alliance which is unusual because it gives TSMC much more skin on the design and manufacturing side than Intel usually does (or has ever done)?
I think the answer is that Otellini is smartly adjusting his strategy, to take advantage of a wealth of capabilities TSMC can bring to the effort. He might be thinking that they'll have the responsiveness, customer connections, and incentives to get embedded SoC Atoms out to vendors (many of them local ODMs, or original design manufacturers), where Intel itself might be more removed, response-wise and geographically.
[Here's how Otellini put it in the press release announcing the TSMC deal: "We believe this effort will make it easier for customers with significant design expertise to take advantage of benefits of the Intel architecture in a manner that allows them to customize the implementation precisely to their needs. The combination of the compelling benefits of our Atom processor combined with the experience and technology of TSMC is another step in our long-term strategic relationship."]
As for the cost angle -- remember, you can't make any money on Atom unless you're cranking away on yields -- Steve Smith confirmed for me that they've got the fabrication down pat. Here's a quote on that from Smith, which I left out of yesterday's post:
"We get something like 2,500 die locations on the wafer. We already have very good yields, very low defect densities. We're not concerned about the cost of the Atom processor. "
Another angle I just thought of is, if Intel knows how to get yields up, it can have TSMC produce regular netbook Atoms near the source of their customers, using incremental TSMC fab capacity which would otherwise go unused. So you're talking even bigger cost savings. It's only fair to note, despite my nattering on about embedded applications, that netbooks are likely to remain the biggest volume app for the chip for some time. Especially when you consider that it quickly becomes difficult to extend that list of potential embedded apps beyond a few obvious targets.
OK, usually I wind up my posts with some smart Alex comment. This time, I hope I've given you lots of information, which sheds some light on what I believe is a really interesting, and potentially very fruitful, foray for Intel.
Please leave a comment below or write to me directly at [email protected].
Subscribe to my blog's feed, here.
For a mobile experience, follow my daily observations on Twitter.
Check out my tech videos on this YouTube channel.
Alex Wolfe is editor-in-chief of InformationWeek.com.