Labs, chipmakers work to keep silicon alive

The Extreme Ultra Violet LLC (EUV)--a coalition made up of Lawrence Livermore National Laboratories, Sandia National Laboratories, Intel, Advanced Micro Devices and others--is showing off prototypes of lithography tools that researchers believe will permit chipmakers to pack far more circuits into much more powerful processors.

Chips made with these new techniques will debut at 10 GHz in 2005 and are expected to climb in speed from there.

The new technique for lithography--the science of "drawing" circuits on a wafer--will likely wring out most of the remaining potential left in silicon, said Chuck Gwyn, director of the program. Circuits can be drawn only so small in silicon wafers. After 2011, the industry will likely have to find a new chip material or come up with other creative solutions.

"This will help us reach the edge of silicon," Gwyn said. "This will get us through the first decade of the century and perhaps part of the second."

The new technique, called Extreme Ultra Violet (EUV) lithography, will effectively allow computer makers to draw smaller circuits onto chips. EUV light has a short wavelength, which allows chipmakers to project extremely intricate circuit patterns onto silicon wafers.

"By going to shorter wavelengths, we are able to print smaller features," said Rick Stulen, chief operating officer of the Virtual National Laboratory, the collective name for the national laboratories participating in the project. "In essence, we are just shrinking the wavelength."

Reducing circuit size is the cornerstone of Moore's Law, which states that the number of transistors capable of being put on a processor should double every 18 months. Shrinking circuits allows manufacturers to put more transistors onto a wafer, which in turn increases power. Unfortunately, the current technique, called DUV lithography, will likely hit its limit around 2003.

Controlling small wavelength light, however, is not easy. Current lithography machines depend on lenses to focus light. Because EUV light would be absorbed by glass, the new system will use a series of four specially coated convex mirrors to capture the mask image and reduce it. The mirrors each contain 80 separate metallic layers just 12 atoms thick.

The technology stems from work at Stanford University. The laser-light technique, meanwhile, derived from work on missile defense systems, said Dave Attwood, a professor at the University of California and a researcher on the project.

EUV machines will be able to process about 80 wafers an hour, approximately the same as current lithography machines, making the process economically feasible.

The coalition already is working with equipment makers such as ASML to smooth commercial acceptance. "Our goal is to make the equipment manufacturers as smart as possible," said Stulen, who predicted the development cycle for commercially viable EUV systems is being shortened by half.

Although participation in the EUV project is open to all chip companies and semiconductor equipment makers, a hint of nationalism percolates through the project. An Asian coalition that includes Nikon and Canon is working on an EUV project, while Europeans have banded together on an EUV venture called the Euclides project, said Stulen. And federal officials expressed some wariness when Infineon, a German chip manufacturer, joined the U.S. effort.

Some potential friction has already emerged. Canon has announced it has patents that could supercede the work being performed in the United States, he said.

Demonstrations of EUV lithography will begin soon. Later this year, the researchers will connect the part of the system that creates the laser beam with the part of the system that projects the light onto the wafer. Currently, the two halves of the machine sit in adjacent rooms at Sandia Labs in Livermore, Calif.

In 2001, the coalition will demonstrate how circuits can be printed on their prototype. Beta lithography systems will be shipped to coalition members in 2002, with machines that can be used in the production of chips starting to arrive a year later.

Processors made from the lithographic technique will start to emerge in computers in 2005.

Being a coalition member has its advantages, the companies and researchers pointed out. Only coalition members will be able to get EUV machines the first two years they are available, Gwyn said. Coalition members also will have right of first refusal on equipment in subsequent years. The purchasing rights will give these companies a virtual lock on manufacturing cutting-edge processors. Intel, AMD, Infineon, Motorola and Micron are all members. IBM has not yet joined.

Coalition members also will own the lion's share of the intellectual property that comes out of the venture. Under the program, the five participating chip manufacturers gave $250 million to the laboratories to perform research. When the bulk of the research begins to wind down, most of the patents and intellectual property will be transferred to the chipmakers.

Although other lithographic methods are being touted, EUV has become the likely candidate, said Linley Gwennap, principal analyst at the Linley Group. "They've really knocked over some of the obstacles," he said.

By contrast, the other proposals seem to have come up against major physical stumbling blocks. X-ray lithography, for instance, can project intricate mask images. Unfortunately, X-rays shoot through lenses and mirrors. Manufacturers, therefore, would have to make masks the same size as the chip, a difficult task.

Other researchers have proposed electron beam lithography, but that would force manufacturers to abandon mask patterns and print circuit lines one at a time.

Although EUV lithography can keep silicon going for a while, Gwyn and others also admit the end is near. EUV lithography will allow chipmakers to print features that measure 70 nanometers. By 2009, chips will come with 30 nanometer features--about the width of 150 atoms. After that, continuing to use silicon becomes a dicey proposition.

"Once you get to below 20 nanometers, the devices don't operate the same way," said Gwyn. "In terms of geometric size, 15 to 20 nanometers is the perceived limit."