At this year's International Electron Devices Meeting in San Francisco, the phrase of the day was 22nm CMOS. Because it is the smallest manufacturable feature of a transistor, and because those transistors have been getting inexorably smaller for the past 40 years, the technology node is the defining characteristic of microchips. I was under the impression that fabrication had barely started on integrated circuits at the 45-nanometer node, which represents the current state of the art. But though I have heard precious little about 32 this year, 22nm invaded the territory before 2008 was even out. 32 nm, in other words, is already chopped liver.
What is this middle child phenomenon of one node announced to great fanfare, followed by a sudden uptick in research activity two nodes downstream? Let's take Intel as an example of this behavior: giving birth to the 45-nm node was a rough delivery for them. They needed an epidural and a high-k gate insulator--the first fundamental redesign of the CMOS transistor in 40 years-- to make it happen. Now it seems like they can coast on their research for another node. Last November the first 45-nm chips rolled out of the fab. A year later (last Wednesday to be exact) Intel announced 32nm, and several papers at IEDM detailed some of the technical aspects of 32 nm. (more on that in a later, dedicated post)
The other major themes at IEDM were "More than Moore," and 3-d stackable chips. The smaller, faster transistor, the staple of Moore's Law, is going to start needing a little help as we approach the physical limits of scaling (and 22nm is pretty close to that limit). The most popular idea seemed to be stacking integrated circuits atop one another, the basic skyscraper idea. Only now they're not doing it to conserve real estate for the traditonal reasons of making transistors smaller. The need to do it to shorten the wiring between the countless points A and points B.
As detailed in an article in this month's Spectrum, IEEE Fellow Peter M. Kogge found that if technology trends continue into the future, exascale computing will still be just a dream in 2015.
Getting to exascale by thenâ''or everâ''requires among other things an intensive focus on reducing the power devoted to moving data around in computers.
One way to do that involves stacking layers of integrated circuits atop each other. Hence the extreme focus on 3D ICs.
One of the best panels was on biomedical applications where USC's James Weiland presented a design for an artificial retina that can enable facial recognition and reading. University of California San Francisco researcher Tejal Desai discussed her research on drug-delivery nanospheres -- microscopic fuzzy "tennis balls" whose nanowire-velcro can bind them inside the gastrointestinal tract for better drug delivery.
The problem in the past has been that for oral drug delivery, a pill delivers only about five percent of its medicinal payload. Five percent! That's because the GI tract was designed explicitly to block absorption of foreign materials. It just keeps passing the stuff downstream, and the combination of the mucosa and the tiny fingerlike appendages inside your GI tract (called villi) works to pass foreign objects down the pipeline as quickly as possible, never letting them stay in one place long enough to effectively deliver drugs to the bloodstream.
For aspirin, that's just fine because aspirin is cheap and you can take another one if you don't feel like it's working. But of course for a cancer drug or some other kind of very expensive medication that needs to be specifically targeted, you really want the most possible medication and the most accurate possible dose.
Nanoparticles filled with a drug payload, and coated with velcro-like nanowires interlock with the GI villi to keep them in one place for long enough to let the drug cross into the blood stream.
One observer, however, was less sanguine about targeted drug delivery. "This is fine for medicine that just has to make it to the bloodstream," he said. But for medicines that are using the bloodstream as a highway to get somewhere else in the body? Once it's in the blood, it's not home free-- there, the drug is attacked by all kinds of biological entities and broken down long before it gets to its target organ or receptor.
Many people thought that real targeted drug delivery will start once we start engineering specific viruses for that purpose. Viruses make the 22 nm node look like Godzilla.