reader comments 0
Share this storyBack in 2012, I had the pleasure of visiting the IBM Watson research center. Among the people I talked with was George Tulevski, who was working on developing carbon nanotubes as a possible replacement for silicon in some critical parts of transistors. IBM likes to think about developing technology with about a 10-year time window, which puts us about halfway to when the company might expect to be making nanotube-based hardware.
So, how's it going? This week, there was a bit of a progress report published in Nature Nanotechnology (which included Tulevski as one of its authors). In it, IBM researchers describe how they're now able to put together test hardware that pushes a carbon nanotube-based processor up to 2.8GHz. It's not an especially useful processor, but the methods used for assembling it show that some (but not all) of the technology needed to commercialize nanotube-based hardware is nearly ready.
Semiconducting hurdlesThe story of putting together a carbon nanotube processor is largely one of overcoming hurdles. You wouldn't necessarily expect that; given that the nanotubes can be naturally semiconducting, they'd seem like a natural fit for existing processor technology. But it's a real challenge to get the right nanotubes in the right place and play nicely with the rest of the processor. In fact, it's a series of challenges.
Note that above I said that nanotubes can be semiconducting. Unfortunately, they can also be metallic. (Well, not entirely unfortunately—that's quite useful for other applications.) Even more unfortunately, when we make a batch of nanotubes, we can't control whether they're going to be metallic or semiconducting. Instead, you just end up with a random mixture of the two.
There have been two approaches to dealing with this. The first is to just put more carbon nanotubes than you need into place, then identify the metallic ones and destroy them. Needless to say, this isn't especially efficient. The alternative is to take a batch of carbon nanotubes and then separate out the semiconducting ones. There are various ways of doing this, but most of them haven't been 100-percent efficient. Which of course means that, at some level, you're going to be putting a piece of metal where you wanted a semiconductor, shorting part of your processor out.
For the new work, IBM relied on a development pioneered at the National Renewable Energy Lab (a facility targeted for massive cuts by the current administration). Some bright people at NREL realized that semiconducting carbon nanotubes would preferentially interact with complicated organic solvents that have nitrogen-containing rings in their structure.
Researchers at IBM decided this would be very useful indeed, so they tested the technique out. A single extraction with the same technique and, 10,000 individual nanotubes later, they can report that over 99.9 percent of the purified tubes were semiconducting. We can consider NREL's work replicated. And, if 99.9 percent's not good enough, there's no reason that the process couldn't be repeated while upping the purity further.
A sense of placeOf course, those semiconducting nanotubes don't do a processor much good if they're still sitting in solution. Ideally, you want a method of placing them in specific locations on your chip. Here, IBM rolled its own solution. The company developed a system in which polymers would only form on specific material on its chips. These polymers would help guide carbon nanotubes out of solution and in to specific locations.
So, we've now got a basic construction kit for carbon nanotube processors. But it's still not enough to do something useful. Modern processors have a complicated mix of p- and n-type semiconductors (which tend to build up positive or negative charges). Carbon nanotubes are naturally p-type, but they can be converted to n-type if they're placed in proximity to certain metals. Unfortunately, those metals tend to oxidize under normal conditions.
So the people at IBM put a cap over this metal layer to try to protect it. Unfortunately, the metal they used (scandium) turned out to like oxygen so much that they stripped it out of another part of the hardware, a hafnium oxide layer. So, that layer had to be replaced.
With all of the hurdles cleared, the team decided to make some individual transistors. These worked extremely well, with every one of the 192 transistors the researchers tested being operational. So, the team went on to try to build actual circuitry. Not useful circuitry, but instead a typical test case for new processor technology: a ring oscillator. This is a series of gates set up so to flip bits; if the gates get a 1, they convert it to 0 and vice versa. By putting an odd number in a ring-shaped configuration, each individual gate will oscillate between 1 and 0 with a timing that depends on the amount of delay involved in each individual gate changing its state.
The good news is that they produced 55 functional ring oscillators, with a performance of up to 2.8 GHz. This is an important demonstration that the process works. Unfortunately, IBM had to build 160 ring oscillators to get the 55 functional ones. So the process isn't mature. In fact, since ring oscillators only really involve five functional gates, it's a long way off from producing anything that might be considered a product.
But, to return to the point this discussion started with, IBM—and the rest of the material science community—still have a bit of space left in their timeline to get this commercialized. And, five years ago, they were still working on getting pure semiconducting nanotubes. Given the progress since, I wouldn't rule things out.
Nature Nanotechnology, 2017. DOI: 10.1038/NNANO.2017.115 (About DOIs).