A new type of computer memory uses carbon, rather than silicon

WAITING for a computer to turn on is a nuisance. That is why manufacturers have been trying to create "non-volatile" memories. These would be fast, like the random-access memory (RAM) chips that are currently used for often-accessed memory, but they would also continue to store information even without power, like hard drives, which are too slow to use except for long-term storage.

Several technologies have been competing to become the standard for fast, non-volatile memory. The best known is magnetic RAM, which IBM and Motorola are touting. Others are based on polymers or on strange-sounding metal alloys called chalcogenides that change shape when an electric charge is applied to them. But there is now a new entrant to the field: carbon.

Carbon comes in many forms. Diamonds and graphite are two of the most familiar ones. A less familiar variety is the nanotube, also known as a "buckytube" after Richard Buckminster Fuller, whose geodesic domes have a framework similar to the arrangement of the atoms in a nanotube. Nanotubes consist of a cylindrical array of carbon atoms whose diameter is only about 1 nanometre (a billionth of a metre). If Nantero, a firm based in Woburn, Massachusetts, proves correct, such tubes will soon be an integral part of computer memories.

Nantero's memory chips consist of billions of nanotubes, each a few hundred nanometres long, suspended from a silicon wafer. Another wafer sits about 100 nanometres below the first. Because the nanotubes that Nantero uses conduct electricity, a small electric charge at one point on the second wafer will draw several dozen nanotubes towards it. Once they are there, they stay there. That is because they are bound by Van der Waals forces--intermolecular bonds that do not depend on external power for their maintenance. An additional application of current, however, will release the nanotubes. This means that a group of a few dozen nanotubes can act as a memory element, storing a single bit (either a one or a zero) of the binary code that computers use to operate. If the connection between the wafers is live at a particular point, the bit represented is a one. If not, it is a zero.

If nanotubes were not so small, this would not be a big deal. Because they are, though, Nantero's technology can already achieve a data density considerably higher than existing RAMs. And because the wafers are so close together, those data can move rapidly from place to place. Nantero's new memory can read or write a bit in as little as half a nanosecond (billionth of a second). The best RAM chips, by contrast, need ten nanoseconds to perform a similar operation.

At the moment, Nantero has only a working prototype. But the firm aims to have memories on the market within a year. It thinks it will be able to tool up for commercial production quickly, because the fabrication technique it uses, though novel, relies on standard semiconductor-making technology.

The main difficulty faced by others who have tried to go down the buckytube route is getting the tubes to align with each other when they are hung from the first wafer. Until now, the approach has been to try to grow all of the tubes in the correct orientation to start with. But Nantero's founders came up with a simpler, if less elegant, solution. They use established lithographic techniques to get rid of tubes that are pointing in the wrong direction by zapping them with an electron beam. That leaves only those that are hanging down towards the opposite wafer.

Though the recent chip is certainly impressive, the reason for getting excited about Nantero is not so much the present as the future. Unlike silicon, which is pushing against its physical limitations, carbon-nanotube technology is in its infancy. Greg Schmergel, Nantero's boss, says that within the next few years the firm's engineers may be able to achieve data densities of a trillion bits per square centimetre (more than 1,000 times that available on existing RAM) and it will be possible to read those memories 100 times faster than can be done at the moment. The days of silicon-based memory may be numbered.

--From The Economist print edition