Just six months ago, researchers from IBM, Intel and Europe's IMEC consortium were warning that transistor scaling might be limited not by lithography, but by the ability to aggressively scale gate oxide to achieve higher performance. A heated debate developed over reliability — whether gate oxides thinner than 22 or 23 angstrom could hold up for the needed 10 years or more of reliable operation.

This week, researchers from Lucent Technologies' Bell Labs reported at the International Reliability Physics Symposium in San Jose, Calif., a growing belief that SiO₂ will hold its ground at 15 angstrom or below, a confidence they said is shared by other research teams at major semiconductor makers.

"Previous results, which said that 23 or 24 angstrom would be the limit, were too pessimistic," said Ashraful Alam, a Bell Labs researcher working at Murray Hill, N.J., who developed a crucial physical model showing that hole damage in very thin oxides is less than once expected.

"We see right now that just in terms of oxide [SiO₂] materials, we can probably grow thinner than 15 angstrom and still maintain a reliable film," Alam said. "There are other manufacturing-related constraints that companies need to solve for, but there are no intrinsic reliability problems with growing very thin oxides."

The Bell Labs results were based on oxides grown at Lucent fabs in Orlando, Fla., using regular manufacturing equipment, Alam added.

"Previously, people were thinking that thicker oxides and thinner oxides degrade similarly as a function of time. This assumption is not correct," he said. In fact, the opposite is true. "As you go to thinner oxides, they age better."

According to the Bell Labs physical model, "the number of trap-producing holes becomes fewer at thinner oxides with lower voltages," he said. "The number of holes that can cause damage becomes fewer, and this was not appreciated before." The 15-angstrom films were tested at 3.2 volts at room temperature for several hours, and this would translate to a 1.3-volt, 10-year operation at room temperature. Alam said similar results have been achieved at voltages as low as 1 V.

Bonnie Weir, another Bell Labs researcher, said researchers at the Reliability Physics Symposium were asking for physical models that would buttress the physical results Bell Labs has been reporting. The match-up of Alam's models with the experimental results achieved by growing ultrathin oxides at Orlando "caused a lot of excitement at the [symposium]," she said.

Moreover, "we may be able to go even further," said Jeff Bude, another Bell Labs researcher. "The 15-angstrom oxides are the thinnest for which we have experimental data, but we may be able to go beyond that. Of course, there are manufacturing constraints and leakage constraints. But the good thing about the leakage constraints is that there may be circuit solutions for those."

Added Alam, "If there were an intrinsic problem with the reliability of silicon dioxide, then there would be no solution, and scaling would have stopped."

If Bell Labs has become a center of optimism regarding oxide scaling, scientists elsewhere appear more cautious. IBM Corp. researchers have argued that actual physical testing of devices will be needed, over longer periods, to ensure that oxide breakdown does not occur. Late last year, Intel Corp. researchers warned that oxide scaling might create problems as soon as the 70-nm process node, expected to debut in 2005 and reach volumes several years later.

"I cannot comment about IBM or Intel," Alam said, "but I can say that at the symposium there was a growing consensus by the major research groups that oxides can be scaled to 15 angstrom. Other groups are seeing results, and that was not the case at last year's International Electron Devices Meeting. There is now a broader consensus, a better understanding, of why thin oxides are showing better intrinsic reliability" than thick ones.

T.P. Ma, a professor at Yale University, said he agreed that researchers are becoming more upbeat about silicon dioxide. Still, he said, "The question is, will oxides that thin be manufacturable? You can get the perfect films, and the reliability, on individual devices. But can you make 12-inch wafers with billions of transistors with gate oxides that are grown with rapid thermal-oxide techniques to 15 angstrom?"

That raises the bar for cleaning native oxides off the wafer, maintaining a precise temperature and gas flow during oxide growth steps, and more stringent gradient cooling, Ma said.

The advantage of using SiO₂ is that it is well-known, with a 30-year history. Physically thicker films, with an effective or electrical thickness that is less than the physical thickness, can be obtained by adding nitrogen to the SiO₂. These thicker films help guard against electron tunneling, while still preserving the fast switching needed by high-performance logic transistors. Texas Instruments Inc., for example, is using these nitrided films in its 130-nm process generation, which moves into manufacturing early next year.

However, Bude warned that adding nitrogen to SiO₂ can lead to electron trapping and mobility degradation.Ma said that using SiO₂ longer may give the industry more time for the transition to a new, and largely untested, class of metal-oxide materials, such as zirconium oxide, hafnium oxide or perhaps titanium oxide, that have sharply higher k values than SiO₂ or the nitrided oxides. (Researchers will simultaneously be traveling in the opposite direction, creating low-k materials to insulate a different part of the chip: the interconnect.)

Rinn Cleavelin, director of front-end processing research at International Sematech, said that time is of the essence. "If the industry has got to do a high-k oxide, then it has got to start building the equipment now," he said. "If some companies want to be ready [to use a high-k] material in 2005, that means our member companies need to have the equipment in their research labs by 2003."

Cleavelin, who also is the chief operating officer at Sematech, said Sematech has formed a cooperative effort with the Semiconductor Research Center, based in North Carolina, and nine universities. High-k gate dielectrics are a major focus, he said, with research spread out over "a very large base of new techniques and materials."

Cleavelin said that while next-generation lithography has absorbed billions of dollars in recent years, few resources have been invested in studying the new techniques required to create thin gate oxides with new and untested materials.

Bude, at Bell Labs, agreed, saying that the Bell Labs work with SiO₂ will give the industry some breathing room. "Going to 15 angstrom with SiO₂ gives us more of a time buffer to move to some of these new materials with a much higher dielectric constant, materials that have a much bigger advantage over silicon nitride or SiO₂," he said. Nitrided oxides provide only a small advantage over SiO₂ in terms of the k value, Bude said. "So, instead of using [them] to get an oxide with the equivalent of 15 angstrom, we can just go ahead and use SiO₂. However, nitrided oxides have been shown to prevent boron penetration and reduce leakage current."