For chips, the next step is a great leap
NIST explores the next dimension for chips<@VM>Sidebar | Natural order
- By William Jackson
- Dec 07, 2007
As technologies go, CMOS is a tough act to follow. The complementary metal-oxide semiconductor, first used in digital watches almost 40 years ago, is the most widely used integrated circuit in information technology products and has enabled the rapid development of everything from modern PCs to increasingly powerful cell phones and other handheld devices.
'The reason CMOS has been valuable to us is because we have been able to improve it every year,' putting more and faster circuits and transistors into each square inch of chip, said Jeff Welser, director of the Semiconductor Research Corp.'s Nanoelectronics Research Initiative.
This progress in processor performance has allowed industry to follow Moore's Law so far. Postulated by Intel cofounder and engineer Gordon Moore in 1965, Moore's Law states that, because of steady improvements in chip fabrication, the number of transistors on a chip will double roughly every 18 months, with computer processing power increasing correspondingly. This observation has held remarkably true across the decades.
But nothing lasts forever.
'We've always been worried about the physical limits of how small CMOS can get,' Welser said.
SRC, an industry consortium created to fund university research into semiconductors, established the Focus Center Research Program in the 1990s to support research to advance CMOS semiconductors.
In 2004, with the end of the CMOS road coming into view, SRC established the Nanoelectronics Research Initiative (NRI) to develop a new generation of technology to replace CMOS by 2020.
'Getting something by 2020 is a challenge, but reasonably attainable,' Welser said. 'We think we probably have another good decade for advancing CMOS' through scaling techniques such as multicore processors before a replacement technology will be needed.
SRC's NRI recently acquired a partner in its quest, one with deep pockets and broad expertise in nanotechnology research.
The National Institute of Standards and Technology announced in September that it would provide $2.76 million in research grants to NRI projects this year, the first step in a projected five-year program to provide more than $18 million in semiconductor research funding. NIST scientists also will be collaborating with industry and university researchers.
This is the kind of long-term basic research that NIST wants to be involved in, said David Seiler, chief of NIST's semiconductor electronics division.
'They are coming up against the limits of what can be done with current semiconductors, so that there are serious concerns about what happens 10 or 15 years from now,' Seiler said. If the IT industry cannot sustain current progress, there could be serious repercussions on the industry and U.S. economy as a whole, he said.
'The entire semiconductor industry has highlighted the problem as one that needs to be solved,' said Jason Boehm, senior analyst at the NIST program office.
Joaquin Martinez, senior scientist at NIST's Office of Microelectronics Programs, said now is the right time for NIST to get involved.
The program is at a precompetitive stage, when the basic research needed is too expensive for any one company or university to undertake by itself. This gives NIST a chance to advance the state of the entire industry.
Welser called NIST's participation in the program absolutely crucial. The research requires the long-term vision and funding available from a government program, and turning a theory into a commercial product requires expertise in extremely refined testing and measurement. 'This is very much what NIST is good at,' he said.
'NIST is all about measurement,' Seiler said.Positive and negative
CMOS semiconductors enable information processing and calculations by shepherding electrons along circuits and through gates or switches. The complementary part of CMOS refers to the fact that a CMOS chip has an equal number of transistors that switch from positive and negative charges. Switches can be either on or off, which allows digital processing of ones and zeroes. The smaller and more compact the circuits and switches can be made, the more powerful the processors are.
One advantage of CMOS technology has always been its low static power drain; that is, the processor uses power only when switching between off and on, reducing both power consumption and the amount of heat generated.
But as the circuits approach the molecular and atomic scale, they are reaching the limits of miniaturization and power advantages are beginning to disappear too. The current leaking from the switches when not in use almost equals that needed to operate the switches, a situation Welser likened to a car that uses as much gas when parked as when running. Additional heat accompanies the wasted power.
'We are reaching the limits of air cooling,' Welser said. Other techniques, such as water cooling, can work for large pieces of equipment but are not feasible in the small devices such as laptop and handheld computers that CMOS has enabled. 'So we need to find a new way to extend scaling. We need to find something that is better than CMOS.'
This is not the government's first involvement in processor development.
The Defense Advanced Research Projects Agency is the largest single financial supporter of SRC's Focus Center Research Program, which is advancing CMOS technology.
And NIST has a long history of working with the semiconductor industry to develop testing and measurement techniques.
Understanding a technology and being able to reproduce it commercially requires the ability to measure accurately, Martinez said.
Although a goal and a deadline for the program are in place, nobody knows yet what they are looking for. 'This is a long-term, basic research goal,' Boehm said. 'We are starting from scratch, at least in the realm of nanoelectronics.'
That is not to say there are no ideas on how to approach the problem.
'There are some good leads,' Martinez said, such as carbon nanotubes.
'But nobody knows how to use them to make good transistors consistently.'
One of the first questions to be worked out is how to represent the ones and zeroes used in digital processing.
CMOS processors use on and off switches for this. Researchers are considering ideas such as electron spin and molecular conformational technology, in which atoms are moved in key positions within a molecule to produce different shapes or behaviors.
There still is a lot of work to be done in selecting a method and coupling it with a technology to produce a product, but the 2020 goal is not unreasonable if money and resources are devoted to the project, Seiler said.
'People are very creative,' he said.
'Breakthroughs do happen. Putting the best minds on it will allow us to come up with breakthroughs.'
With the time left before foreseeable improvements in CMOS are exhausted, 'the timing is good,' Seiler said. 'We can be there, hopefully, in 10 years.'
Moving a technology of this complexity from prototype to production typically takes about 10 years, Welser said. NRI hopes to cull some of the theories now being proposed and develop some feasible ideas to work on by 2010.
To accomplish this, NRI in 2006 established three virtual regional research centers made up of groups of cooperating universities. The Western Institute of Nanoelectronics is based in California, the Institute for Nanoelectronic Discovery and Exploration is in New York, and the Southwest Academy for Nanoelectronics is in Texas. NRI has just completed its first annual review of work at these centers.
'I'm surprised at how well we've done' in the first year, he said. There has been good progress in understanding electron spin, but the greatest advance has been getting physicists and chemists to move from theory and experimental science to talking about practical requirements.
'Our biggest challenge has been bridging the communications gap' between scientists and engineers, Welser said. Theories are essential to practice, but schemes that work at a few degrees above absolute zero are a long way from being useful to engineers who have to come up with a process to manufacture products. Fortunately, the scientists see the hurdle as an interesting challenge and are coming up with ideas, he said.
Although NIST intends to spend $18.5 million during the next five years, the money for the next four years of grants has not yet been appropriated. It also has not yet been determined how the initial $2.76 million will be spent, whether it will go to a few big projects or a lot of smaller ones. But the first round of money should be available quickly. A call for proposals is expected to be released this fall, and NIST will assist NRI in evaluating the proposals.
The grants should be in the hands of the researchers by spring.Researchers turn to nature for help in constructing nanoscale circuits
NANOTECHNOLOGY, which involves creating and using tools measured in billionths of a meter, holds great promise for applications such as medicine and quantum computing, but producing the devices in usable quantities in reasonable time remains a challenge.
Researchers at the University of Maryland's A. James Clark School of Engineering are working to enlist nature's help to produce nanocircuits economically.
'While we understand how to make working nanoscale devices, making things out of a countable number of atoms takes a long time,' said Ray Phaneuf, associate professor of materials science and engineering. 'Industry needs to be able to mass-produce them on a practical time scale.'
That's where nature comes in. 'Nature is very good at making many copies of an object' through self-assembly, Phaneuf said. But nature knows how to make only a limited range of patterns for these complex structures, such as shells or crystals. Phaneuf's work focuses on the use of templates to teach nature some new tricks.
'The idea of using templates is not new,' Phaneuf said. 'What is new is the idea of trying to convince nature, based on the topography of the template, that it should assemble objects in a particular place,' atom by atom.
One application for the process could be quantum computing. A host of schemes propose harnessing the quantum states of atomic particles to do complex calculations. One involves assembling pairs of quantum dots ' tiny semiconductors containing from one to 100 particles with elementary electric charges ' to create the qubits used in quantum calculations.
Assembling the billions of dots in the precise patterns needed for massively parallel computing may be possible, but, Phaneuf said, 'it may not be doable within the age of the universe' with current techniques.
'Nature already knows how to assemble quantum dots,' he said. 'We are working on the step before self-assembly, the self-organization of the substrate,' which will act as the template for the dots.
The silicon substrate is etched into steps using lithography, but it is difficult to reach the level of precision required at the atomic scale using lithography alone. Heat and cold can be used to add or subtract atoms on the surfaces and precisely shape the step patterns. The steps can also be shaped. The step patterns, which are stiff, tend to straighten out under heat but are limited by the surrounding patterns in how much they can straighten.
'We play this stiffness off with the repulsive interaction between steps' to create the sizes and shapes needed, Phaneuf said.
The result is a substrate that can be reused many times as a template for growing nanostructures with silicon and gallium arsenide for computer and cell phone components.
'It still is in the development stages,' he said.
'There is still quite a lot to do before we make practical devices out of it. I don't think we're quite ready to make transistors on the chips.'
And the market for the end products has not yet developed. You are not likely to find any deals on quantum computers from Dell or HP in the ads of your Sunday supplements this weekend. More-immediate applications for this technology are likely to be biochips used in biology and medicine.