Understanding high-k dielectrics easier, thanks to new model

London (England) - Research being carried out at the London Centre for Nanotechnology has revealed a theoretical model which may provide a better understanding of the dielectric layer. The new model predicts flaws and defects in a visual way which might help researchers hone in on just the right materials for future semiconductors. These could significantly decrease wasted power and heat.

The dielectric layer is a very thin layer of insulating material which, up until very recently, has been Silicon Dioxide (SiO2). It provides a necessary electrical barrier which allows transistors to function, but as features get smaller that barrier is becoming less and less efficient. Alternate materials are needed because at the current thickness of only five atomic layers (5 atoms high), it just won’t cut it any longer. Researchers are looking for what’s called a "high-k" dielectric, or something that has a high dielectric constant. This high-k solution will provide the necessary barrier to keep Moore’s Law trucking along for quite some time.

Researchers at Intel were able to achieve an early version of that goal. In 2003 they developed a hafnium-based alternative which was largely kept under wraps. It’s since been revealed, however, that their hafnium solution is being used in use in their upcoming 45nm Penryn microprocessors as a generation one mixture. It is believed that in 2008, following the other half of their tick-tock strategy, that generation two will arrive with even better power saving features. IBM and AMD announced similar findings at the time of Intel’s public disclosure, however they will not be seen in products until the 32nm process node, which will be here in 2009/2010.

According to Intel, their current 45nm hafnium high-k insulating layer, combined with a new type of metal gate, reduces leakage by up to 10x over SiO2. This creates a tremendous power saving advantage for Intel, one which directly equates to faster clock speeds, denser transistor packing, lower power consumption, more cache, more logic, or any combination thereof. All of that means more capability on the same physical package that you or I would buy and stick in our computers. This gives Intel a lot of headroom in terms of design considerations. Should they make a faster processor ? Then use the reduced heat generation in that way by increasing the power budget. Should they make a more efficient, cooler running processor ? Then reduce the power budget and take advantage of the much lower heat generation through wasted energy.

As you can see, the high-k solution is a type of holy grail long sought after in the semiconductor community as there are many advantages. And now, the researchers in London have released a model of a hafnium solution (it might not be the same one Intel is using) which mathematically predicts where new facets of previously undisclosed-to-the-public flaws may still exist. This new data, if validated through real-world observation, would provide researchers with a powerful tool to essentially optimize their hafnium mix. In fact, it might turn out that hafnium isn’t the best solution, just one that works in generation one. It may later be discovered that some exotic, or even common, material is the one which works best. Specifically, the researchers were able to determine that there are "holes" in the structure of the hafnium/oxygen molecule which ultimately create "pockets" where electrons can gather. This is possible even if the material is absolutely pure and there are no similar defects caused by improper construction or impurity. These molecular holes produce the equivalent of bad sections of electrical lines, and they are here to stay. Whereas in copper lines, for example, bits of impurity can impede the signal, these molecular holes create tiny electron vortexes which capture and deflect electrons from their normal wave-like movement. And unlike copper deposition which can be improved through better process handling, even perfect hafnium solutions will be subject to these molecular flaws. And, as process technologies scale down further and further, the effects of these holes will become more and more pronounced. They’ll eventually yielding the same kind of barrier that we saw with SiO2 at the five atomic layer level. And right now it’s just another problem to solve for future applications. And if I may be allowed to quote India Arie, it seems like they’re "headed in the right direction" toward solving it.

Researchers continue to arm themselves with an ever increasing aresenal of mathematical models which accurately describe the real-world products they’re working on. Such models, only made possible through the use of high-speed computers, are literally changing the face of our world today. I find it somewhat ironic that the first basic components were created by man. Man reasoned his way into a foundational level of science which built the basic machine, the first computers. That generation then gave man more ammunition to take to generation two, which made it better. By now we’re beyond generation 10 and you can see how high-speed computer modeling is now making it possible for researchers to make even faster computers, and make them run cooler. The machine itself, guided by man’s insight and direction, is revealing things about itself that would not have been possible otherwise. It’s a pretty amazing cycle. As Willy Wonka might say, " I wonder how far it will go ?"

Funding for this project and work carried out at the London Centre for Nanotechnology and UCL Department of Physics & Astronomy was made possible by the EPSRC. Access to computer time on the HPCx facility was awarded to the Materials Chemistry Consortium with funding from the EPSRC. Full details of this discovery appeared in an October 12 Physical Review Letters article.