May 15, 2002 10:30 AM PDT
IBM wants a few good nano start-ups
The group, part of the emerging-products organization inside IBM Microelectronics, will serve two main purposes. It will let Big Blue get a better view of the current state of nanotechnology research outside the company, and it could help a handful of companies get beyond the prototype stage, said Thomas Thies, director of physical sciences at IBM, speaking at the Nanotech Planet conference here Tuesday.
The group will offer "deep experience in silicon microelectronics and microelectronic packaging," Thies said--adding quickly that the program is extremely selective. Only companies with working prototypes need apply, and in the end, IBM will work with only a few.
"We don't want to get overwhelmed with proposals," he said.
Nanotechnology is the science of building semiconductors and other computer components by arranging just a dollop of molecules, and proponents such as IBM and Hewlett-Packard say that it will revolutionize computing.
Microprocessors and related gear will not only get smaller and more powerful--for instance, terabits of storage will fit into devices the size of a wristwatch, said Thies--but they will also become far less expensive to manufacture because companies won't have to sink nearly as much into fabrication facilities and lithography equipment.
But even proponents point out that this will be one long and slow revolution. Five years from now, if everything goes perfectly, cutting-edge companies will only be ready to release "passive" nanotechnology products--that is, new manufacturing materials created through molecular manipulation.
In 10 or so years, nanotechnology will begin to appear in sensors and other types of chips, but it will be used to enhance, rather than replace, semiconductors made of silicon. Nanotechnology may be used to make chemical leak detectors or serve as the cache memory on a microprocessor.
"We're where microelectronics was in 1960," said R. Stanley Williams, an HP fellow and director of the company's Quantum Science Research Laboratories. Progress will take massive investment, intensive research and a couple of decades.
Nanotechnology in some ways is the unintended offspring of Moore's Law. That dictum, postulated by Intel founder Gordon Moore in the mid-1960s, holds that the number of transistors on a given chip will double every 18 to 24 months, in turn allowing computers to get progressively smaller, cheaper and more powerful. Until recently, engineers accomplished this by shrinking transistors and packing more on each chip.
Scientists, however, will start hitting the physical barriers of shrinking transistors on silicon wafers over the next 20 years. Transistors won't function the same way after a certain point. When circuit structures get smaller than 100 nanometers, electrons "know" they are in a confined space and behave differently, said Thies.
"We've got a good 10, 15, maybe 20 years left" to see gains from existing technology, said Thies. "Silicon might not be the final way to process information. Magnetic hard-disk drives may not be the final way to story information."
On top of that, the expense of manufacturing on silicon will grow out of control. Circuits are etched onto silicon-based chips through lithography, an expensive process in which light is directed through a series of masks that map the transistor pattern. Fabrication facilities will cost several billion dollars apiece in the near future.
"We want to take the cost of a fab and cut it by 1,000 times or more," said Williams. "We want to change this dramatically because they have to change dramatically."
Ideally, nanotechnology would cure the shrinkage problem because its structures will be far smaller than traditional transistors. Also, manufacturers wouldn't need detailed masks. Instead, molecules will arrange themselves, like crystals in a snowflake, into circuits through their own physical and chemical properties. Later this year, Williams said, HP plans to announce a breakthrough achieved in its labs, hinting that it revolves around how nano wires can assemble themselves into a mesh.
Such self-generation is part of a growing toleration of uncertainty in computing. In the future, designers will have to create systems that work perfectly out of components that don't work perfectly all the time, Williams said.
Similarly, accuracy in complex computing problems will be determined through statistical probability rather than absolute certainty. For instance, memory constructed out of a mesh of nano wires would work correctly, but designers wouldn't be able to trace the exact path of answers.
Nanotechnology should also open up new design dimensions. "The nano scale is very different and unique--the properties of matter change with size or shape," said Williams. "Electrons no longer behave like rocks as they do in current electronic devices. Electrons behave as waves."
Progress is occurring in less glamorous areas. General Motors, for instance, is experimenting on passive nanotechnology to develop new automotive materials that will be hard but not brittle, said Williams. Governments around the world have also begun to pour in research dollars.
After the U.S. dedicated $422 million toward nano research in 2001, the Japanese followed with a $410 million project, he noted. Europe and the rest of the world have sunk $425 million into nano projects.
History may provide some guidance, Williams said. Today, an HP handheld is a thousand times more powerful than ENIAC, which, 50 years ago, was the most powerful computer ever made. Fifty years from now, Williams postulated, "it should be possible to build a handheld device that is more powerful than every computer on the earth combined today." But there is no guarantee--the speed of light was determined over 100 years ago, and no one yet can drive that fast.
Still, it's not as though that will be easy. Researchers will have to invent new manufacturing techniques and products. IBM showed off the first transistorlike structure in 2001 using carbon nanotubes.
To commercialize the technology, Thies said, "we've got to figure out how to do that a trillion times on every chip."