Introduction
The availability of both semiconducting and metallic variants of carbon nanotubes, combined with their intrinsically nanoscale size, has led naturally to ideas of nanotube-based microelectronics. In the most optimistic of visions, the scaling limits now anticipated for silicon microprocessors could, one day, be overcome by the use of self-assembled nanotube components. There is a catch, however, in that all nanotube synthesis methods invariably make a mixture of both metallic and semiconducting tubes (see Box 1). To build reliable devices, some kind of selection process must be made — preferably one that can be performed easily and on a large scale. On page 60 of this issue, Mark Hersam and co-workers1 at Northwestern University show that by coating as-synthesized carbon nanotubes in a combination of surfactants (engineered soaps, if you will), they can be sorted by both diameter and electronic type.
In light of this advance, and now that 15 years have passed since the discovery of nanotubes, it is perhaps instructive to compare the present state of nanotube electronics with the development of conventional solid-state electronics2. The point-contact transistor was invented in 1947. The first transistor radio was unveiled in 1954. By the early 1960s, a serious limit to further complexity of discrete transistor circuits was the overwhelming number of wires needed to interconnect the large number of discrete components (a problem dubbed the 'tyranny of numbers'). The solution, devised in 1958–59, was the monolithic integrated circuit (IC). In 1967, some 10 years after invention of the IC and about 20 years after that of the transistor, the first handheld calculator appeared. In 1971, Intel unveiled the 4004 microprocessor (2,300 transistors), which has evolved over the last 35 years to the Pentium IV (40 million transistors).
So how does the field of nanotube electronics stack up against this record? Given the challenges presented by manipulating objects at this length scale, the pace of advances has been quite remarkable. The first nanotube-based transistors appeared3 in 1998, only five years after synthesis methods for the single-walled nanotubes were first discovered. Logic circuits4 and a ring oscillator5, built from discrete nanotube transistors, first appeared in 2001. And earlier this year saw the demonstration of a five-stage, ten-transistor ring oscillator, built as an IC, on a single nanotube6.
In a bid to replace silicon, however, nanotubes, to put it mildly, still have a long way to go. For one thing, consider the competition. The IC was a truly brilliant solution that has allowed for near-continuous refinement, codified in Moore's law, which has held for over 40 years. The 40-million-transistor Pentium IV presents stiff competition for any newcomer to this arena. The IC also had a major economic advantage. The first commercial IC was a big clunky thing with all of four transistors. Nevertheless, entering as it did the competitive vacuum of its time, it was immediately useful, and most importantly, marketable. This meant that during those 40 years there has been a steady stream of revenue to fuel the research and development necessary for that continuous refinement. This is a luxury that nanotubes may not have.
Remember that 'self assembly' phrase mentioned earlier. In some circles this is code for 'and magic happens here'. What the phrase embodies, however, is an anticipated means of exposing nanotubes to a sequence of physico–chemical environments that constrain their thermodynamic lowest-energy state to be one step along the route to the ultimately desired assembled structure. Note that this is something that synthetic organic chemists, albeit in different systems, do day in and day out. But here's the rub. From the perspective of commercialization, the self-assembly problem (which is just the tyranny of numbers again, only now with really, really small wires) is either solved at the scale of ten million or so transistors or it's not. Few are likely to buy a ten-transistor chip just because it's made of nanotubes. There may be niche markets where possible nanotube-specific performance advantages will make their small-scale integration valuable. However, it seems unlikely that this will generate the level of revenue required to solve the self-assembly problem at the ultra-large-scale-integration level necessary to compete with silicon.
I worry about this because it means that funding for this long-term goal will have to continue to come from government sources. But officials of funding agencies are only human and are subject to the same raised expectations, influenced by the rapid pace of modern life, as the rest of us. And I've heard grumblings of late that make me worry that governments might not stay the course in this most challenging of goals. That would be a terrible shame because, as reported in this issue, great strides continue to be made.
An important prerequisite for a nanotube transistor is that the nanotube between the source and drain terminals must be semiconducting, as opposed to metallic. With present synthesis methods, the odds of getting a metallic nanotube, and therefore a failed device, are roughly one in three. At the large-scale integration level that makes for a rather intolerable defect rate. Moreover, as shown by recent work, to yield comparable performance specifications, the nanotubes should also all have very similar diameters7.
Such selectivity for nanotubes by metallic or semiconducting type and diameter is precisely what is reported by Hersam and colleagues. Progress along these lines has been reported previously by others8, 9, 10, 11, 12, but all their methods possess limitations (for example, using nanotube-damaging chemical interactions, or having no clear pathway to scalability). The progress reported in this issue is the first to show selectivity near to the level of the specifically desired nanotube (n,m) index, along with a scalability that could, for the first time, make specific nanotube types widely available.
The specific gravities (densities relative to water) of typical SWNTs range from about 1.1 to 1.4. This means that when nanotubes are centrifuged in water at high g-force they should sink to the bottom of the container. As discovered in 2002, however, surfactant shells can provide additional buoyancy that keeps individual nanotubes and small bundles aloft whereas larger bundles sediment out13. What the Northwestern group has found is that the buoyancy imparted by specific surfactants (and their mixtures) can discriminate SWNTs by their diameters in a density gradient1.
Such gradients are produced by layering successively decreasing concentrations of a dense compound from the bottom to the top of a centrifuge tube (a common technique in biochemical separations). The surfactant-coated nanotubes were seeded near the top of the gradient and centrifuged at rates up to 64,000 revolutions per minute, which corresponds to forces in excess of 170,000 g. Over the course of the centrifugations, which lasted between 9 and 24 hours, the coated nanotubes moved through the gradient to settle at positions where their density matched that of the surrounding medium (Fig. 1). The separated fractions could then be collected and analysed.
Figure 1: Carbon nanotubes wrapped with surfactants can be separated by diameter and electronic type when centrifuged at high g-forces.
In step 1, mixtures of surfactants are used to disperse individual nanotubes. In step 2, these are loaded into a centrifuge tube containing a liquid density gradient. When centrifuged, larger-diameter wrapped nanotubes settle further down the centrifuge tube. Layers can be separated, and repeated centrifugation and optimization of the density gradient leads to highly enriched samples of either metallic or semiconducting nanotubes.
Full size image (16 KB)Surprisingly, the sorting goes in the opposite direction to what would be expected based on the relationship between pure nanotube density and diameter. Here, larger-diameter surfactant-coated tubes settled in higher-density layers of the gradient. More remarkable is that discrimination also occurred by electronic type. Using mixtures of surfactants, such electronic-type sorting was optimized so that by repeated centrifugation, with careful adjustment of the density gradients, highly enriched samples of metallic or semiconducting nanotubes were obtained. Although just micrograms of sorted nanotubes are described in this report, the technique is only limited by the scale of the centrifugation. The same principles apply at increased volumes and the use of a larger-capacity industrial centrifuge could enable gram quantities to be sorted in less than a day.
To put such nanotubes into the hands of synthetic organic chemists and self-assembly-minded researchers, the late Rick Smalley imagined a day when one could open a chemicals catalogue and order, from a broad range of such products, bottles of nanotubes specified by type and length (methods of length sorting have existed for some time). The work by the Northwestern group suggests that this day is nearly upon us. Next big challenge: the (nanoscale) tyranny of numbers.
Postscript: The discussion above was predicated on the most difficult of goals, namely where individual nanotube transistor elements permit the extension of Moore's law to technology nodes beyond what might be allowed in silicon. Long before the time when this might be achieved, nanotubes in the form of films and networks should supplant silicon in a broad array of (hopefully) revenue-generating applications. The work by the Northwestern group should also greatly help those efforts.
