There are so many major, seismic shifts in the computing industry happening at the 32nm process node that it's hard for me to get my mind around it all. I've been covering the story of x86's journey into the ultramobile and embedded space, a journey that starts at 45nm and really gets interesting at 32nm, but that tale is only one thread in a much larger epic that's emerging bit by bit in one press release and news story after another.
For instance, take this week's announcement that Toshiba is now joining the parade of semiconductor companies who've looked at the $4 billion or higher cost of a 32nm fab and decided against going it alone. The Japanese semiconductor giant will be joining IBM's fab alliance at the 32nm node, bumping the number of alliance members (excluding IBM) up to six. IBM and Toshiba had previously been cooperating (along with Sony) on research for the 32nm node, so the pair's newly announced agreement to join forces on 32nm bulk CMOS fabrication is really just an extension of their previous research partnership. Nonetheless, it's an agreement that takes one more major party out of the running at 32nm.
For the real scoop on this IBM-Toshiba announcement and what it means for the semiconductor industry, there's no way that I can top Dave Manners's blog entry on the topic, so I won't even try. I do, however, want to zoom in on one fascinating part of the post, which explains quite a bit of what's driving Intel and others into commercial competition with the OLPC Project.
Finally, for everyone with a 32nm fab there's going to be a new problem. If 450mm wafers are adopted, and the companies which buy most of the world's manufacturing equipment are pushing hard for 450mm manufacturing equipment to be developed, then there's the problem that only seven fabs will be needed to make the world's total demand for transistors.
Manners develops this point in terms of its implications for fab equipment buyers, but I want to take it in a different direction and dwell for a moment on what it means for an Intel or an IBM if only seven fabs can meet the world's (presumably current) demand for transistors. (I'm not sure where Manners got this number, but it sounds feasible and I trust that it's legit.)
If the combination of a 32nm feature size and a 450mm wafer size increases fab output to the point that only seven fabs are needed to meet the total world demand for transistors, there's only one way for the semi industry to see growth in such a scenario: increase demand. This is why Intel would like to see every school-age child, farmer, factory worker, day laborer, and so on from San Francisco to Siberia suddenly discover a pressing need for lots and lots of transistors.
The vast bulk of the history of computing up until the present day has been about moving semiconductors from the server room to the business desktop, then from the business desktop to the first class cabin, and then from first class into coach. At this point, everyone who can afford a cheap plane ticket or a pair of Nikes is already wired to the gills with transistors, and the major bottlenecks in getting those folks to buy even more of them are mostly out of Intel's control (i.e., screen size/quality, battery life, connectivity, usability).
To see real growth in the coming decades, Intel, AMD, and the rest of the semi industry must focus on markets where an iPhone would cost a month's income, and then on markets where it would cost a year's income. This is the reality behind Intel's and AMD's interest in the device category that OLPC represents. It's the reason why Intel has teams of anthropologists running around rural China, and why AMD launched its 50×15 plan.
So when you're trying to imagine how the computing industry will look in ten to fifteen years, you have to forget about the BlackBerry set almost entirely. The bulk of the market will shift from those people, who will remain a very profitable niche, to consist of people who aren't currently all that wired.
Or, to put it another way, when I was out in rural San Salvador two years ago, every cinderblock home had a transistor radio and a color TV. If I make that trip again in 10 years, I'll find that that color TV has been replaced by a device that has at least the horsepower, connectivity, and functionality of the MacBook Pro from which I'm filing this report. And depending on how "green" that ubiquitous, post-32nm computer is, we're either headed for a networked Nirvana or an ecological nightmare. Hence the focus from semi companies on "green technology," a focus that goes hand in hand with selling transistors to the great, unwired masses.