Carbon-Nanotube Wiring Gets Real

Could carbon nanotubes have a shot at replacing the ­copper wires that connect millions of transistors on today’s ­silicon chips? Chip makers replaced aluminum interconnects with better conducting ­copper ones about seven years ago, but now copper’s days are ­numbered too. Higher-performance chips with more-tightly packed transistors, expected as soon as 2012, will need interconnects less than 40 nanometers wide, at which point copper’s resistance will slow signaling down too much.

Late last month, at the Materials Research Society’s spring meeting in San Francisco, a team of ­engineers from Stanford and Toshiba reported that they have used ­carbon ­nanotubes to wire logic-circuit components on a ­conventional silicon CMOS chip. They claim to have shown that nanotubes can shuttle data at speeds of a little faster than 1 gigahertz, close to the range of state-of-the-art microprocessors, which run at speeds of 2 to 3 GHz.

In principle, nanotubes can handle a current density 1000 times as great as that of copper or silver. Accordingly, many chip makers, including Intel, have been trying to figure out whether nanotubes can be practically combined into an integrated circuit and, if so, how their properties hold up.

Stanford electrical-engineering professor H.S. Philip Wong and his collaborators at Toshiba fabricated ­common test circuits, called ring oscillators, on a ­silicon chip. Each oscillator was missing one wire that would complete the circuit. Then researchers laid down ­nanotubes on top of the circuits to make that last connection. Of the 19 ring oscillators, 16 worked at over 800 megahertz, and the best worked at 1.02 GHz. ”This is the first time that a nanotube as a wire is operating in a conventional chip-type environment,” Wong says.

Alexander Tselev, a chemist studying carbon-nanotube interconnects at Duke University, in Durham, N.C., hails this as ”a step from basic science to real application.” Still, a number of big challenges remain, particularly devising a reliable method to make nanotubes with consistent properties and finding a good way to arrange tubes in a pattern.

Today’s ­manufacturing processes result in batches containing nanotubes of ­different sizes and ­electrical properties, some that ­conduct electricity and ­others that are ­semiconductors. Indeed, it is the ­inconsistencies in resistance and in the length of the nanotubes that result in the ­different ­operating speeds of the ring oscillators, Wong says.

There is also no known way to precisely place ­nanotubes on a ­surface. The researchers use a ­standard method called dye electro­phoresis. It involves ­depositing a nanotube ­solution on a surface and applying electric fields to attract the nanotubes to the required spots. The method is unpredictable. Wong and his colleagues started out by ­fabricating 256 ­oscillators; carbon nanotubes ­completed the wiring in only 19.

There are many other problems to be solved. Interconnects would need to be made from bundles of carbon nanotubes, because they conduct current much better than single nanotubes do. But bundles of tubes would be hard to lay down horizontally. Unlike integrated circuits, which have layers of semiconductor, insulator, and other materials ”nicely stacked one on top of the other,” says Vladimir Stojanovic, an electrical-engineering and computer-science professor at MIT, ”tubes are hard to handle, because they don’t stack up very well.” Manufacturers will have to find either a way to place prefabricated nanotubes in the right spots or a way to grow nanotubes where they’re needed at temperatures that match silicon-fabrication temperatures, Stojanovic says.

Clearly, carbon-nanotube researchers have a lot left to do. With copper needing a replacement as soon as 2012, they might have to speed things up.