Intel Looks Beyond Silicon

New nonsilicon transistors could be integrated into existing chip-making processes.

Intel has developed a new kind of transistor, made of a material other than silicon, that has the potential to be faster and use less electricity than today's chips. And, crucially, the new transistors are economical and could be fabricated using existing manufacturing facilities because they can be built directly on top of standard silicon wafers. Such chips made with these nonsilicon components are still at least a decade away, but industry experts believe that they are one of the more promising options to replace silicon in the coming years.

A better buffer: Intel has shown that nonsilicon transistors can be fabricated on a silicon wafer by growing them on top of a thin buffer layer. Previous buffer layers were relatively thick and hence would crack, damaging transistors. Credit: Intel

As transistors get increasingly small, the silicon that composes them doesn't work as well: electricity leaks through the layers, causing excess heat and faulty logic. Researchers at Intel and other chip-making companies such as AMD and IBM, as well as at universities around the world, are scrambling to find a replacement for silicon. Some suspect that carbon nanotubes or another carbon material called graphene could be the answer. (See "Carbon Nanotube Computers" and "New Graphene Transistors Show Promise.") But others are putting money and research into compound semiconductors, a class of semiconductor that is made from a combination of elements from the third and fifth columns of the periodic table. (See "Beyond Silicon" and "Maintaining Moore's Law without Silicon.")

Compound semiconductors are attractive to engineers because electrons move through them easier than they move through silicon. This means that the compound semiconductors can work as fast as, or faster than, a silicon-based transistor, but without needing as large a voltage. And as devices shrink, it's crucial that they require low voltages: otherwise, they overheat and leak electricity--problems that are beginning to plague silicon. However, compound semiconductors aren't easy to grow directly on silicon. The materials are often incompatible with silicon--the atoms are spaced so that they don't layer well. When layered directly on top of one another, the result is a cracked crystal and defective transistors.

Intel has proposed a solution to the atomic-mismatch problem in a paper presented today at the International Electron Devices Meeting, in Washington, DC. To build their new transistors, the researchers layered the compound semiconductors, called indium gallium arsenide and indium aluminum arsenide.

When these materials are stacked, their electronic properties interact to form quantum wells--places where charged particles such as electrons can be confined--that act as transistors, says Michael Mayberry, director of components research and vice president of Intel's technology and manufacturing group. To avoid the strain and cracking, the researchers added buffer layers of the two materials. The trick is to make sure that the buffer layers contain concentrations of atoms that are slightly more compatible with silicon. But as more layers are added, the atomic spacing perfectly matches that of the transistor layers. Mayberry says that the buffer is slightly more than one micrometer thick, and it keeps any defects from affecting the transistors.

To be sure, many researchers have proposed adding buffer layers between silicon and nonsilicon materials. For instance, a company called Amberwave, in Salem, NH, founded by Gene Fitzgerald, a professor of material science at MIT, has an approach that uses germanium on a type of silicon wafer with a layer of silicon dioxide, an insulating material, built in. (See "Adding Speed to Silicon.")

However, notes Jesus del Alamo, who is also a professor of electrical engineering at MIT, Intel's approach is unique in that the researchers have grown the buffer layers out of the same material that they use for the transistors. In addition, he says, Intel has shown that only a thin buffer layer is necessary to get good quality. A thick buffer layer, which can be up to five micrometers thick, is costly and is more prone to cracking, he adds.

Story continues below

The work is "impressive," del Alamo says. "If you can bring the layer structure on silicon, then the substrates feel and look like silicon, and all the tools that have been developed for silicon manufacturing can be reused in this new technology."

Still, Mayberry notes that there remain a handful of problems that need to be addressed before these transistors can appear in consumer electronics. For one, the gate, or the on-off switch, for these new transistors is relatively large at 80 nanometers. Mayberry says that engineers will need to shrink this down so that chips will have a relatively high transistor density. He adds that in the meantime, some of these materials may find their way into the chip-making process, to be used in specific components in microprocessors.