Down with the transistor
By Davide Castelvecchi
April 30th, 2008

New component could radically transform computer chips

After going unchallenged for decades, the transistor’s supremacy could come to an end. Researchers have demonstrated a new type of electronic component that could replace transistors as the building blocks of computer chips, and lead to faster, more powerful and less energy-thirsty computers.

Stanley Williams and his collaborators at HP Labs in Palo Alto, Calif., have created a surprisingly simple new device called a memristor. The device is a piece of an electric circuit with baggage: Its history determines its electrical resistance. Depending on the voltage that was recently applied to it, a memristor will switch from acting as an insulator (“off”) to acting as a conductor (“on”) and back.

This on-off capability offers engineers a way to build circuits that manipulate and store information. “All of a sudden, you have a new tool in your toolbox,” Williams says. The most immediate advantage of memristors, Williams adds, is that they could be packed into chips up to 100 times more densely than transistors.

For decades, progress in electronics has relied on shrinking the features of computer chips, roughly doubling the number of transistors per chip every two years — a trend that has become known as Moore’s law, after Intel cofounder Gordon Moore.

But engineers’ ability to shrink transistor-based electronics is rapidly approaching physical limits, and Moore’s law is expected to hit a hard wall in about 10 years. The memristor offers “an alternate way to continue progress,” says Leon Chua, an electrical engineer at the University of California, Berkeley who first proposed the memristor concept in 1971.

Memristors could be shrunk to smaller sizes because they exploit the very physics that makes shrinking transistors hard. Transistors are built out of semiconducting materials. These semiconductors’ electronic properties are finely tuned by adding small amounts of impurities called dopants. But voltages make dopant atoms move within a transistor. At nanometer scales, this effect is strong enough to change the semiconductor’s properties and degrade performance, something that electrical engineers have so far seen as a nuisance.

Williams’ team built the memristor by sandwiching a thin film of titanium dioxide between two platinum layers. Normally, titanium dioxide is an insulator. But applying a voltage between the platinum layers exerts a force on the film’s oxygen atoms, pushing the atoms toward one side. As the atoms move, they leave behind gaps in the titanium dioxide’s crystal structure. Such gaps create an imbalance in the crystal’s distribution of electric charge, simulating the presence of positive ions. The gaps act like the dopants in a semiconductor, and move in the opposite direction as the oxygen atoms do.

Such gap doping turns titanium dioxide into a good conductor, so the memristor switches to “on.” But if the voltage is reversed, the oxygen atoms go back to their places, turning the memristor back to “off,” the researchers describe in the May 1 Nature.

In the past, researchers including Williams had simulated the behavior of a memristor using combinations of transistors, but this is the first “pure memristor,” he says. He and his collaborators also showed last year that memristor-based devices can act like transistors and be used to process information.

Memristors also can store bits of data, representing a “1” in the on state and a “0” in the off state. And the data is not lost when the device is shut down, similar to the non-hard drive memory that’s become ubiquitous in flash drives, cell phones and mp3 players. Williams says that his lab has already built prototypes of memristor-based computer memory that is tens of times more dense than current flash memory or even than state-of-the-art ordinary RAM.

Moreover, he says, inserting memristors into chips would not require substantial changes to current chip production methods, and so mass production should be feasible.

Chua says that while demonstrating memristors, the researchers’ experiments also showed the limits of ordinary transistors, due to the migration of dopants. “They not only built a device. They showed that, as you get smaller and smaller, transistors are going to stop working.”