Competing technologies for downsizing the transistor
By IVARS PETERSON
In the realm of microelectronics, smaller means faster.
During the last 2 decades, the number of transistors crammed onto an integrated-circuit chip has doubled roughly every 2 years, while the price per transistor has dropped steadily. That miniaturization of electronic components has led to enormous increases in computer speed and memory.
Nowadays, the channels defining each transistor on a silicon surface are typically 0.35 micrometer (mm) wide, small enough that some 16 million transistors can occupy a single chip. Looking a decade into the future, the semiconductor industry is contemplating the technical challenges of manufacturing chips with 500 million transistors. Those transistors would have features only a fraction of the size of the present-day standard.
Essential to chip fabrication is a process called lithography, which resembles the printing of a photograph by shining light through a negative onto a photosensitive surface. To produce features less than 0.1 mm wide, semiconductor companies face the task of developing methods that involve light or other electromagnetic radiation of wavelengths much shorter than those commonly used today. The shorter the wavelength, the finer the line that can be sharply drawn.
To create sufficiently thin lines for future chips, researchers have explored several schemes: one based on the use of extreme ultraviolet (EUV) radiation, another on X rays, and a third on electron beams. Laboratory demonstrations show that all three technologies are capable of writing features far smaller than those produced by conventional optics. The big question is whether any of them can be scaled up and put on a factory floor to manufacture integrated-circuit chips accurately and inexpensively.
Such a leap in technology would require a major investment in research and development and in equipment and facilities. Whatever the choice, it would also be a gamble, with no guarantee of success.
Earlier this year, Intel Corp. of Santa Clara, Calif., and two collaborators placed their bet on EUV lithography, agreeing to fund research at the Lawrence Livermore (Calif.) National Laboratory and the Sandia National Laboratories in Livermore, where the technology was initially developed, and the Lawrence Berkeley (Calif.) National Laboratory (SN: 9/20/97, p. 180). "We believe that this approach holds the greatest amount of promise," says Intel spokesman Adam Grossberg. "It's been demonstrated in a lab environment, and we believe it's possible to take this technology to a manufacturing capability."
Other companies, such as IBM and Lucent Technologies, continue to pursue alternative approaches. "All of the technologies work," says Richard R. Freeman, who heads the newly formed Virtual National Laboratory, which coordinates the joint national labs effort on EUV lithography. "It's just a question of whether [advanced lithography] can be done efficiently and at sufficiently low cost."
Chip manufacture has become an extremely expensive proposition. Installing and operating a single factory production line can cost more than $1 billion.
Fabrication begins with a silicon wafer topped by a light-sensitive polymer layer and a stencil-like mask made from glass coated with a light-blocking metal. The mask bears the intricate master pattern that chip designers wish to etch onto the silicon wafer.
The system shines light through the mask, then through a series of lenses to focus the image falling on the polymer layer to about one-quarter of its original size. The light causes chemical changes in the polymer, imprinting the circuit design. Affected areas can then be chemically etched, exposing and eating away some of the underlying silicon. This creates the pattern for a single chip, many of which are ultimately inscribed on a silicon wafer.
These lithographic light projectors are called steppers because they step to a chip location, create the required pattern, then step to the next chip site. In this way, one site after another is exposed to the patterned light.
The wavelength of light used cannot be much larger than the desired size of features on the chip. To etch lines as narrow as 0.35 mm, or 350 nanometers, manufacturers rely on lasers that generate ultraviolet light at a wavelength of 365 nm, somewhat shorter than wavelengths of light visible to the human eye. The newest steppers, designed to create lines 0.25 mm wide, make use of 248-nm radiation, or deep ultraviolet light. Various refinements in the mask technology and the optics, along with the use of a laser that generates 193-nm radiation, could bring the line width down to 0.18 mm, perhaps even to 0.13 mm by using further optical tricks.
Each step down poses tough challenges, from making the illumination increasingly uniform and keeping masks from warping to ensuring that different chip layers are aligned with sufficient precision. Nonetheless, the basic setup remains the same.
Going to even shorter wavelengths, however, requires a different approach. For one thing, quartz lenses begin to absorb rather than refract, or bend, light at such wavelengths, so conventional optics can no longer be used.
One advantage of EUV lithography is that the overall setup resembles that required for conventional optical lithography. The main difference is that, in order to reduce absorption, an elaborate set of mirrors replaces the refractive lenses to focus the radiation.
The technology is an outgrowth of research related to the Department of Defense's Strategic Defense Initiative and the development of laser-based antimissile weapons. Starting in the mid-1980s, researchers at Livermore sought ways to control X rays created when high-powered laser beams penetrate a gas. They devised special mirrors coated with alternating layers of materials such as molybdenum and silicon to reflect and direct X rays.
The interaction between an extremely intense laser beam and a gas produces a high-temperature plasma of ions and electrons. The plasma emits radiation spanning a wide range of wavelengths, including some in the extreme ultraviolet that could be useful for lithography when controlled with the Livermore-designed mirrors. That possibility encouraged Sandia engineers to design and build plasma light sources potentially suitable for EUV lithography.
The laser-generated plasmas typically produce radiation at wavelengths between 10 and 20 nm, depending on the target gas. "We chose 13 nm because that's the wavelength at which our mirrors are most highly reflective," Freeman says. Last year, a Sandia team used the technology to make a working transistor and other devices with a minimum feature size of 0.1 mm. "To get even smaller features would require just an incremental improvement in the imaging optics," says Sandia's Richard H. Stulen. "The wavelength would stay the same. That's one of the beauties of this technology."
That's still a long way from putting billions of transistors on a single chip in a factory setting. Researchers face a number of technical issues, such as improving the reflectivity of the mirrors, achieving sufficiently accurate positioning of the wafer and mask, and developing a suitable photosensitive material for coating the wafer.
"You have to get the source right, you have to get the optics right, you have to get the coatings right, and you have to get the mask right," Freeman says. "Everything has to work. It's an incredibly complex system."
X-ray lithography has been under development even longer than EUV technology. The X-ray technique makes use of highly penetrating radiation, ranging from 0.01 to 1.0 nm in wavelength. Normally, that radiation is generated by electrons pushed to a high energy in a particle accelerator and circulating in a storage ring.
As early as 1977, Henry I. Smith and his coworkers at the Massachusetts Institute of Technology used X rays to draw lines less than 0.02 mm wide on a silicon surface. The group later built microelectronic devices with features only 0.05 mm wide.
More recently, John M. Warlaumont and his colleagues at the IBM Thomas J. Watson Research Center in Yorktown Heights, N.Y., demonstrated that X-ray lithography can be used to build many devices side by side.
Several factors have discouraged semiconductor companies from moving this technology out of the laboratory and into the factory. In particular, it requires X rays produced by accelerated electrons, which are available at only a handful of synchrotron facilities throughout the country.
"Semiconductor companies want equipment that can be wheeled on to the factory floor," Freeman says. "They don't want to build a factory around a synchrotron."
The use of high-energy electron beams as "pencils" to write patterns on a silicon surface also has a long history. Such technology is already used to create the masks needed for conventional lithography and to etch tiny features directly on chips built for research purposes.
Although the basic process is too slow for manufacturing, the technique remains promising. High-energy electrons have extremely short wavelengths, making possible the creation of much finer patterns than can be made with other methods. One way to speed up the process is to send multiple electron beams through a mask, allowing each beam to write a separate section of the chip pattern. Alternatively, a wide beam acting as a floodlight could inscribe an entire pattern at once.
Recently, researchers at Lucent Technologies have pushed ahead with an electron-beam process called SCALPEL (scattering with angular limitation projection electron-beam lithography). The method requires a special mask made of thin layers of metals such as chromium and tungsten, which deflect or scatter rather than absorb electrons. The layers sit on a silicon nitride base, which is transparent to electrons.
As electrons pass through the mask, some are scattered and others are unaffected by the coating. A system of magnetic lenses focuses the unscattered electrons on a silicon wafer to create the desired pattern.
Last year, researchers at Bell Laboratories in Murray Hill, N.J., demonstrated that SCALPEL could produce lines as thin as 0.08 mm. A chip built to take advantage of that size could store up to 8 billion bits of information.
"We are confident that SCALPEL will make [electron-beam] lithography viable for high-volume production and decreasing line widths," says Lloyd R. Harriott of Bell Labs. A production-worthy SCALPEL system could be available by the year 2002.
Whatever lithography technique proves viable for manufacturing, the high cost of developing the technology presents a serious obstacle.
"Although nearly $50 million of private and government support has been invested in SCALPEL and the concept has been proven, it will take closer to a $1 billion investment worldwide over the next decade to lead to actual production with SCALPEL," says J. Murray Gibson of the University of Illinois at Urbana-Champaign, who coinvented SCALPEL.
There are also political obstacles. Intel's agreement with the Department of Energy to provide $250 million in funding to develop EUV lithography has been criticized by members of Congress and U.S. companies involved in supplying lithography equipment.
Last month, a letter from four House Democrats to Secretary of Energy Federico F. Peņa noted that the agreement allows Intel to license any developed technology. They argued that the agreement potentially allows the company to grant foreign equipment makers access to sensitive new technology, possibly giving away U.S. trade secrets and compromising national security in the process. Intel replied that licensing agreements were not yet complete and would ultimately include a significant number of U.S. companies. In the meantime, the pursuit of the small and fast continues.
Even the modest decreases in minimum line width permitted by pushing current optical lithography to its limit could enable chip designers to put both memory and logic circuits on the same chip. The resulting increase in the speed and performance of computers could have a tremendous economic impact.
Geppert, L. 1996. Semiconductor lithography for the next millennium. IEEE Spectrum April:33.
Gibson, J.M. 1997. Reading and writing with electron beams. Physics Today October:56.
Goodwin, I. 1997. Intel, Motorola and Advanced Micro Devices enlist 3 DOE labs to develop new computer chip. Physics Today October:85.
Holstein, W.J. 1997. What's good for Intel. . . . U.S. News & World Report Oct. 6:47.
MacIlwain, C. 1997. Chip-making equipment industry warns of threat from Intel deal. Nature 389(Sept.18):215.
Peterson, I. 1997. Tinier transistors for tomorrow's chips. Science News 152(Sept. 20):180.
Schaller, R.R. 1997. Moore's law: Past, present, and future. IEEE Spectrum June:52.
Service, R.F. 1996. Can chip devices keep shrinking? Science 274(Dec. 13):1834.
Spencer, W.J., and C.L. Seitz. 1997. Engines of progress: Semiconductor technology trends and issues. In Defining a Decade: Envisioning CSTB's second 10 years, computer science and telecommunications board. Washington, D.C.: National Academy Press. Available at http://www.nap.edu/reading/room/books/decade.
Wu, C. 1997. Computer chips take a leap forward. Science News 152(Sept. 27):196.
Information about extreme ultraviolet lithography can be found at http://www.ca.sandia.gov/news/euvl/ and at http://xray.eecs.berkeley.edu:8080/EUV/overview.html.
An overview of X-ray lithography is available at http://www.xraylith.wisc.edu/overview/overview.html.
Additional information about the SCALPEL project at Lucent Technologies is available at http://www.bell-labs.com/project/SCALPEL/.
The reactions of four Democratic members of Congress to the agreement between Intel and its partners and DOE is at http://www.house.gov/science_democrats/welcome.htm.
Richard R. Freeman
Virtual National Laboratory
Lawrence Livermore National Laboratory
P.O. Box 808
Livermore, CA 94551
J. Murray Gibson
Department of Physics
University of Illinois at Urbana-Champaign
1110 West Green Street
Urbana, IL 61801
Lloyd R. Harriott
600 Mountain Avenue
Murray Hill, NJ 07974
Richard H. Stulen
Sandia National Laboratories
Mail Stop 9161
Livermore, CA 94550
References - all articlesSources - all articles