The limit in today’s technologies isn’t the phase-change material itself but the size of the electrical contacts needed to connect to the islands where information is stored. “The smallest contacts they have been able to make are about 50 nanometers,” says Pop. His devices use narrow, highly conductive carbon nanotubes as the electrical contacts. The nanotubes range from one to five nanometers in diameter, and they connect to pieces of phase-change material that are about 10 nanometers in size.
Pop’s group uses established methods to grow arrays of flat, parallel carbon nanotubes on silicon chips. Then they add metal electrical contacts to the nanotubes and zap the nanotubes with a large pulse of electricity until they snap in two, leaving a tiny gap between the two halves. The device is then coated with a phase-change material called GST (a compound of germanium, antimony, and terbium). Each bit consists of an island of GST in the gap in a nanotube. When a small amount of current is passed through the nanotubes, it heats the material, changing its phase. The nanotubes can also be used to read the bit.
It takes about 0.1 milliamps to switch a conventional phase-change memory bit. This week in the journal Science, Pop’s group reports writing to their phase-change memory using 100th as much current. They’ve shown that they can write and rewrite each bit hundreds of times and have made arrays of about 100 bits. Pop says the next step is to demonstrate millions of read-write cycles and larger arrays.
If the device can be produced in high volume, it could benefit not just portable electronics but also desktops and mainframes, says Zhirnov. In today’s computers, moving information between processors and memory, and reading and writing to memory, takes a lot of energy and generates waste heat. Bringing the logic and memory closer together, onto the same chip if possible, is a major goal in computing, says Zhirnov.
Putting the two on one chip isn’t possible with flash, because writing to flash memory requires voltages about 20 times higher than the voltages needed to operate digital logic. The nanotube devices, on the other hand, operate in the same voltage range as the transistors used for digital logic. In principle, this means they could be used to make chips that integrate memory and logic and operate very efficiently.
The Illinois researchers must also demonstrate that the memory is reliable. There’s a lot of variation in the size and conductive properties of carbon nanotubes. Pop says that so far, these variations don’t seem to matter in the memory devices. But that remains to be proved on a larger scale.