NSF preparing for the demise of Moore's Law
- — 13 February, 2008 22:02
In anticipation of Moore's Law becoming irrelevant in the next 10 to 20 years, the National Science Foundation (NSF) wants funding for research that could lead to a replacement for current silicon technology.
The NSF last week requested US$20 million from the U.S. government for fiscal 2009 to start the "Science and Engineering Beyond Moore's Law" effort, which would fund academic research on technologies, including carbon nanotubes, quantum computing and massively multicore computers, that could improve and replace current transistor technology.
Moore's Law states that the number of transistors that can be placed on silicon, and its attendant computational capability, doubles every 18 months.
Human and economic progress in the U.S. over the past 20 years has depended on an increasing ability to do information processing and computing, said Michael Foster, division director of computing and communication foundations at NSF. "If the current technological basis of that ends, we've got to find some way to replace it or we're going to stop moving forward."
The traditional way to improve transistor performance is to decrease the thickness of the gate oxide, or insulator that separates one part of the transistor from the other. The looming barrier is that transistors will be shrunk as small as possible for them to still work effectively, after which they may need to be replaced or somehow improved on, Foster said.
"In the kind-of near future -- in 8 to 10 years -- we will have reduced that gate oxide thickness to the point where it will no longer act as an effective insulator," Foster said. "I don't know of any other proposals to increase the performance of... [current] transistors, which is why we have to look at really radically new structures like transistors based on nanostructures."
Carbon nanotubes could provide a way to create smaller transistors, Foster said. Transistor performance is correlated to transistor size -- the smaller a transistor, the better it performs. "Carbon nanotubes give us the possibility of much smaller transistors than we can make right now," Foster said.
Carbon nanotubes could also be used as interconnect on circuits, Foster said. Nanotubes could be placed one after another on a circuit, though it would require fault-tolerant architectures. That would require new research and improvements on chip architecture as well, Foster said. "We think architecture is going to become an important component of any beyond-Moore's-Law topic," Foster said.
Looking far ahead, quantum computing could be the next answer to delivering massive computing power, Foster said. Quantum computing uses matter -- atoms and molecules -- to process massive amounts of tasks at supercomputing speeds because basic data units, called qubits, hold both the values 0 and 1 simultaneously, and share those values among all qubits. It is based on the laws of quantum mechanics, which look at interaction and behavior of matter on atomic and subatomic -- proton, neutron and electron -- levels.
"The implementations we have so far will, for example, trap single ions at very cold temperatures and use those as qubits. That takes a room full of equipment to make a very small register. We clearly need progress there, but it's promising," Foster said.
The idea behind quantum computing is that it provides inherent parallelism, so its development requires improvements in parallel programming, in which several computers work on the same program together, Foster said. Parallel programming has been researched since the 1970s but progress has been slow, Foster said.
"We could continue to expand our IT abilities without necessarily expanding the capabilities in any one computer. But we need software research to understand how to coordinate the efforts of not just one or two computers, but thousands or millions of computers," Foster said. That poses a research problem, as scientists have been trying to understand parallel programming for decades and haven't succeeded yet, Foster said.
NSF is already doing research in nanotechnology, software and architectural ideas that will contribute to the effort to develop chips that will continually improve computing power beyond Moore's Law.
Ultimately, current transistors may be hard to replace, Foster said, so researchers may look to develop better architectures and chip designs that use current transistor technology to maintain the computing capability growth rate.
"It's not out of the question to me that we could build very large chips with thousands of cores on them -- I hesitate to go too much higher than that just because it will sound crazy -- but if we did that... we [could] use those things to get done the computing and information technology that we need to have done," Foster said.
The $20 million budget request is modest compared to research budgets of companies like Intel and IBM. "The funding is intended to help laboratories assist, not compete, with research from the private sector," Foster said. NSF will seed academic laboratories and industry associations like Semiconductor Research to conduct research, which in turn may work with commercial organizations. The commercial organizations may further invest in the research to drive it into mass production, Foster said.
Expectations from the project are modest, Foster said. "It's very early stage research. Any expectations we have will probably not be accurate." However, it could lead to better fault-tolerant techniques to design and operate systems with small components in them, Foster said.