11 December 2007—Computer chips have become so dense that packing even more transistors into them is forcing engineers to break long-standing rules of transistor design. Earlier this year, Intel became the first company to get 45-nanometer chips on the market by finding a way to make a transistor with what's known as a high-k dielectric. To make the transistors work, it had to make the first fundamental change to transistor design in nearly 40 years, switching to metal-gate electrodes from silicon. Intel also had to use a new manufacturing process and two different types of metal to make the transistors. Now a team of engineers at Semiconductor Leading Edge Technologies (Selete), a research consortium for Japanese technology companies, says it has worked out a new way of making high-k dielectric transistors that uses the common manufacturing process and only a single type of metal.

A transistor is basically a switch: When you apply voltage to the gate electrode, it opens or closes a channel in the silicon for charge to flow. In an n-type transistor, electrons flow through the channel, whereas in a p-type transistor, it's actually the absence of electrons (usually referred to as holes) that flows. Transistors in microprocessors are usually arranged so that the two varieties work in a complementary fashion. In these transistors, the gate is separated from the chip surface by a silicon-dioxide insulating layer. As transistors have shrunk down over the decades, the oxide eventually became so thin that it leaked electrons through it, wasting power. Since the mid-1990s, chip makers have been experimenting with replacing the silicon dioxide with high-k dielectrics—materials that can be made physically thick (to stop electron leakage) but electrically thin (so that the transistor turns on easily).

But high-k dielectrics have problems of their own. Engineers have found that they work only with metal gates instead of the silicon gates in use since the late 1960s. In a gate made from silicon, engineers dope the silicon with impurities to control whether it is p-type (having excess holes) or n-type (having excess electrons). The gate has the opposite doping as the channel—a p-type gate controls an n-type channel. Metal electrodes can't be doped. Instead, researchers had to search for metals that happen to have an electrical property called “work function,” which makes them behave the same way as the doped silicon. For Intel, this meant two metals: one whose work function matched the p-type silicon substrate and one that matched the n-type.

The Selete group effectively flipped this method on its head, starting with a single metal whose work function falls in between that of p-type and n-type silicon, and using different high-k dielectrics to push the effective work function toward one value or the other. Using only one type of gate metal makes building the transistor easier and helps control important properties of the transistor, according to Yasuo Nara, whose research group presented its findings yesterday at the International Electron Devices Meeting, in Washington, D.C.

Researchers noticed that during one of the critical manufacturing steps in making transistors—annealing—the interaction between the high-k dielectric and the metal gate changed the gate's effective work function. Intel sought to avoid such a change by going against the conventional procedures and adding the gate after annealing.

But researchers at Selete decided to take advantage of the change. Nara and his colleagues found that if they added certain elements to the metal gate, the annealing process drove the elements into the high-k dielectric, changing its composition and predictably modifying the effective work function. For p-type transistors, they added aluminum, and for n-type, they added magnesium. The result was two different high-k dielectrics but only one metal gate.

It's not yet clear if Selete's method is a viable alternative to Intel's technique, but it may be a new way to attack a difficult problem. “We are not sure how Japanese chip makers feel about our results,” writes Nara by e-mail, but “we believe single-metal integration is beneficial for fabrication.”