Moore's Limit?

My son's PC is in the shop while he's at camp.

It's a Pentium IV. It overheats. My man is going to add fans. He may have to change the case out to get things working.

This is a common problem. The classic case of Moore's Law – you can double the volume of circuits on a chip every 18 months or so – has reached a limit.

Heat.

As circuits get closer, with distances between lines measured in the 10s of microns, heat builds up as electricity passes down those circuits. Intel began recognizing this a few years ago.

They made a fateful decision. They would switch to low-power designs. It was a big decision because, thanks to Moore's Second Law, planning for a chip is like planning a car plant. It costs billions of dollars now to build a fab line. Your corporate “road map” is very hard to change.

But Intel had to change.

AMD had already begun dealing with this problem, and it was building a new fab in Dresden, Germany to produce high quantities of low-power, high performance chips. This was partly due to their embrace of “multiple core” technology, in which you're basically doing parallel processing (like the old SETI@Home experiment) on the chip. Each chip is actually multiple chips, so you can improve performance without changing the underlying technology. At the same time, the greater complexity of chips was making it easier for them to run multiple operating systems, including Windows. So AMD is gaining market share.

 

This is what Paul Otellini faced when he took power a year ago. The decision to change had been made, but the ship would take time to turn about. He decided that future profits would lie “up the stack,” in boards and finished goods. But that's not something Intel has ever been good at.

The ship is turning. Intel has new “dual core” chips out this week that are much better than the old stuff. Intel is still #1 in chips and its deal with Apple to supply Macintosh chips is giving it some cachet.

The problem lies in getting up the stack. Intel's WiMax efforts are stymied by government policies favoring the auction of spectrum – in which spectrum owners specify equipment – over unlicensed spectrum – where equipment makers do the specifying.

 

The real problem is Intel doesn't seem to have anything else up its sleeve. Complete motherboards, complete systems, complete solutions are the way to go, but that has never been Intel's way.

 

The lesson here is that Moore bites. Moore makes no guarantees, even for Moore's children. The only way to succeed is with more, but Moore's roadmap has run out of room.

 

So Otellini (right) really needs a new roadmap. He has to build a new Intel, or buy a product company and fund it, or build it. Those are the announcements to look for.