The excellent adventure continues
02/01/2004
Creating the foundation for the tools of the Information Age has never been easy, but as the semiconductor industry aims to cross the threshold of billion-transistor chips with features well under 100nm, the challenges grow ever more formidable. What's amazing about this continuous quest is that even though problems seem virtually insoluble (i.e., red brick walls in the ITRS), the technical community somehow finds ways to manipulate the physics and chemistry to solve the problem, or skirts major barriers by finding completely new paths.
Remember how we couldn't print features smaller than the exposure wavelength? Or how copper would poison transistors? Taboos are broken down and things that "could never be done" are now common in fabs all over the world. How do the technologists always manage to find a way?
Pushing frontiers can come from an analytical approach (like physicists predicting new particles based on calculations) or more practical methods (the 1% inspiration, 99% perspiration technique — like Edison trying all kinds of materials to find a suitable lightbulb filament). The first method might be illustrated by Robert Matthews' fanciful exploration of Murphy's Law that won him the 1996 Ig Nobel Prize for physics. The British scientist investigated why dropped toast most often lands on the buttered side. This experiment was described by Marc Abrahams in The Ig Nobel Prizes: The Annals of Improbable Research, Dutton, 2002:
"Matthews explored the behavior of a rigid, rough, homogeneous rectangular lamina, mass m, side 2a, falling from a rigid platform set at a height h above the ground. He considered the dynamics of this body from an initial state where the center of gravity overhangs the table by a distance delta subnought, and analyzed it mercilessly through all stages of its perilous journey to the final resting place at height h = 0."
The second approach is favored by a group of Stanford U. graduates who founded David Kelley Design (now IDEO). An early project was the mouse for Apple's Macintosh. These designers once explained to me that innovations are often approached differently by American and Japanese engineers. Americans decide what will be the biggest problem and do a thorough analysis, while the Japanese tend to build a crude model, discover the problems, and then make steady improvements. For the mouse, the Americans did an in-depth analysis of the sensitivity of the linkage between the moving cursor and a rolling ball. The Stanford group, favoring the Japanese approach, fashioned a crude mouse from items picked up at a hardware store, and found that a much bigger problem came from dust and erasure particles gumming up the works.
At the International Electron Devices Meeting in December, there was a clear example of how our industry, which is now much more globally collaborative, attacks problems from both directions, which can lead to a much fuller picture of the dynamics than either approach would on its own. Two years ago at IEDM, Reisinger et al. suggested that a residual polarization effect in high-k oxides like those of zirconium and hafnium was much larger than in SiO2, meaning that threshold voltage would vary with transistor cycling. At this year's IEDM, Jameson et al. from Stanford U. analyzed the effect, which they called "dielectric relaxation," showing that it was due to tunneling in a double well in amorphous regions of an imperfect crystal. They suggested it couldn't be processed out because it could even occur at grain boundaries. After Jameson's talk, an IMEC researcher commented that his group had put in an interface layer, and the effect had gone away. A number of other reports, such as one on a HfN/HfO2 gate stack from Singapore, Peking U., Jujung of Korea, and U. of Texas, showed how stable VT could be achieved, and in a session on metal gates, Intel (S. Datta et al.) showed how putting TiN gates on HfO2 dielectrics over tensile strained silicon on an SiGe layer not only avoided the problem but also greatly improved mobilities, which could be reduced by nitridization.
That's our magic pixie dust: tackle the problem from numerous creative angles while concurrently doing an in-depth analysis. Next year's IEDM will probably include even more analysis to show the basis of the problems along with a number of clever processing tricks to solve or avoid them. That's why every day is an adventure in this amazing industry.
|
Robert Haavind
Editor in Chief