Moore's Law may get a new lease on life thanks to a discovery jointly announced Friday by researchers at IBM and Purdue University.
Named after Gordon Moore, former Intel co-founder and chairman, the well-known predict posits that the number of transistors on a chip doubles roughly every two years. But while Moore's Law has held true for the last 43 years, scientists say the computer industry is bound to bump up against limits to the shrinkage of conventional silicon transistor dimensions sometime in the next decade.
But that may no longer be true. IBM and Purdue researchers say they have found that the growth pattern of silicon nanowires is sufficiently predictable so as to become part of a manufacturer's design processes.
One nanometer equals one-millionth of a millimeter.
Nanowires are grown out of silicon "nucleate," and then undergo a process to a solid phase forming into wires. They range in size from 10 to 40 nanometers.
Eric Stach, an assistant professor of materials engineering at Purdue, said this was the first time that scientists have been able to measure the nucleation process with any precision.
The findings are reported in a paper appearing Friday in the journal Science. The paper was written by Stach, along with Purdue doctoral student Bong Joong Kim, and IBM materials scientists Frances Ross, Jerry Tersoff, Suneel Kodambaka, and Mark Reuter from the physical sciences department at the Watson Research Center.
Here's a brief description of the process they describe in a summary:
The silicon nanowires begin forming from tiny gold nanoparticles ranging in size from 10 to 40 nanometers, or billionths of a meter. By comparison, a human red blood cell is more than 100 times larger than the gold particles. The gold particles are placed in the microscope's vacuum chamber and then exposed to a gas containing silicon, and the particles act as a catalyst to liberate silicon from the gas to form into solid wires. The particles are heated to about 600 degrees Celsius, or more than 1,100 degrees Fahrenheit, causing them to melt as they fill with silicon from the gas. With increasing exposure, the liquid gold eventually contains too much silicon and is said to become "supersaturated," and the silicon precipitates as a solid, causing the nanowire to begin forming.
In real world terms, the fact that the nucleation process being repeatable on this small a scale means that scientists can measure and predict when the process is going to occur. Stach said that would help companies more confidently design systems to make nanowires for electronics products.
"The key result we report is a very basic and fundamental one, namely that when forming these structures, they start their formation in a very controllable fashion," Stach said. "In short, if one were to place equal sized catalysts over a large silicon wafer (as one would need to in order to integrate into chip architectures) and exposed them to the source gas, silane, they would all start their growth at pretty much the same time.
"In prior work, the IBM group has shown that their rate of growth thereafter is also very uniform," he continued. "This means one can make the same structures over a large scale, just as needed for real devices. If this were not to be the case, there would be substantial hurdles in making these structures part of real device design, from a manufacturing viewpoint."
The hoped for result: transistors consisting of nanowires. But even with this discovery, Stach said that there remain questions related to the optimization of device creation, and device optimization. Another challenge to using nanowires in electronics, he said, will be the replacement of gold as a catalyst with other metals that are better suited for the electronics industry.
"It represents a challenge: there are issues with gold migrating into the silicon wires, which can affect properties, depending on the way that they are used," he said. "There is work by multiple groups worldwide on catalysts, with copper, nickel, and aluminum among the most prominent. Copper and nickel are interesting, as they have been--and are--used routinely in modern devices as metals, and thus are more 'comfortable' for incorporation in a manufacturing environment."
IBM echoed that caution, saying it was too early to predict when manufacturers would be able to port the science to the production lines.
"Gold is not the best metal from an electronics perspective so the next step would be to integrate other metals such as copper, nickel, or aluminum. So until that happens it would be premature to predict what technology node would be ideal for mainstream manufacturing," said a spokesman for the company.