For more than a decade, engineers have been fretting that they are running out of tricks for continuing to shrink silicon transistors. Intel’s latest chips have transistors with features as small as 14 nanometers, but it is unclear how the industry can keep scaling down silicon transistors much further or what might replace them.
A project at IBM is now aiming to have transistors built using carbon nanotubes ready to take over from silicon transistors soon after 2020. According to the semiconductor industry’s roadmap, transistors at that point must have features as small as five nanometers to keep up with the continuous miniaturization of computer chips. “That’s where silicon scaling runs out of steam, and there really is nothing else,” says Wilfried Haensch, who leads the company’s nanotube project at the company’s T.J. Watson research center in Yorktown Heights, New York. Nanotubes are the only technology that looks capable of keeping the advance of computer power from slowing down, by offering a practical way to make both smaller and faster transistors, he says.
In 1998, researchers at IBM made one of the first working carbon nanotube transistors. And now, after more than a decade of research, IBM is the first major company to commit to getting the technology ready for commercialization.
“We previously worked on it as a sandbox type of thing,” says James Hannon, head of IBM’s molecular assemblies and devices group. Hannon led IBM’s nanotube work before Haensch, who took over in 2011 after a career working on manufacturing conventional chips. “Wilfried joined with a silicon technology background [and] our focus really shifted.”
Haensch’s team chose the target for commercialization based on the timetable of technical improvements the chip industry has mapped out to keep alive Moore’s Law, a prediction originating in 1965 that the number of transistors that could be crammed into a circuit would double every two years. Generations of chip-making technology are known by the size of the smallest structure they can write into a chip. The current best is 14 nanometers, and by 2020, in order to keep up with Moore’s Law, the industry will need to be down to five nanometers. This is the point IBM hopes nanotubes can step in. The most recent report from the microchip industry group the ITRS says the so-called five-nanometer “node” is due in 2019.
IBM has recently made chips with 10,000 nanotube transistors (see “How to Build a Nanotube Computer”). Now it is working on a transistor design that could be built on the silicon wafers used in the industry today with minimal changes to existing design and manufacturing methods. The design was chosen in part based on simulations that evaluated the performance of a chip with billions of transistors. Those simulations suggest that the design chosen should allow a microprocessor to be five times as fast as a silicon one using the same amount of power.
IBM’s chosen design uses six nanotubes lined up in parallel to make a single transistor. Each nanotube is 1.4 nanometers wide, about 30 nanometers long, and spaced roughly eight nanometers apart from its neighbors. Both ends of the six tubes are embedded into electrodes that supply current, leaving around 10 nanometers of their lengths exposed in the middle. A third electrode runs perpendicularly underneath this portion of the tubes and switches the transistor on and off to represent digital 1s and 0s.
The IBM team has tested nanotube transistors with that design, but so far it hasn’t found a way to position the nanotubes closely enough together, because existing chip technology can’t work at that scale. The favored solution is to chemically label the substrate and nanotubes with compounds that would cause them to self-assemble into position. Those compounds could then be stripped away, leaving the nanotubes arranged correctly and ready to have electrodes and other circuitry added to finish a chip.
Haensch’s team buys nanotubes in bulk from industrial suppliers and filters out the tubes with the right properties for transistors using a modified version of a machine used to filter molecules such as proteins in the pharmaceutical industry. It uses electric charge to separate semiconducting nanotubes useful for transistors from those that conduct electricity like metals and can’t be used for transistors.
Last year researchers at Stanford created the first simple computer built using only nanotube transistors (see “The First Nanotube Computer”). But those components were bulky and slow compared to silicon transistors, says Subhasish Mitra, a professor who worked on the project. “We now know that you can build something useful with carbon nanotubes,” he says. “The question is, how do you get the performance that you need?”
Although IBM hasn’t worked out how to make nanotube transistors small enough for mass production, Mirta says it has made concrete steps, and has devised processes that should be amenable to the semiconductor industry.
However, for now IBM’s nanotube effort remains within its research labs, not its semiconductor business unit. And the researchers are open about the fact that success is not guaranteed. In particular, if the nanotube transistors are not ready soon after 2020 when the industry needs them, the window of opportunity might be closed, says IBM’s Hannon.
If nanotubes don’t make it, there’s little else that shows much potential to take over from silicon transistors in that time frame. Devices that manipulate the spin of individual electrons are the closest possible candidate (see “Silicon-Based Spintronics”), but they’re less mature, and unlike carbon nanotubes, they don’t behave similarly to silicon transistors, says Hannon.