Metal Gates Only: High-k Need Not Apply
Peter Singer, Editor-in-Chief -- Semiconductor International, 7/1/2003
One way to improve transistor performance is to switch from a conventional gate dielectric to a material with a higher dielectric constant (k), such as HfO2.
This serves to increase the drive current and reduce leakage current (see "High-k
Gate Dielectrics: No Easy Solution ," Semiconductor International, February 2003). Along with this move to high-k gate
dielectrics, it is thought that a switch from the polysilicon gates now used to
a metal gate might also be necessary (see "SEMATECH
Develops High-k Removal Process
," Semiconductor International
, May 2003).
New work from Advanced Micro Devices (AMD, Sunnyvale, Calif.) shows that there are performance advantages to be had by going to a metal gate, regardless of whether or not the dielectric is a conventional oxide or a high-k dielectric. At the VLSI Symposium held last month in Kyoto, Japan, AMD researchers showed that a nickel silicide (NiSi) gate provided superior improvements used with fully depleted SOI (FD-SOI) devices, as well as with n-channel devices fabricated on strained silicon. "We believe it is not unlikely that we'll see metal gate implementation before you see the high-k gate dielectrics," said Craig Sanders, vice president of process technology development at AMD. The FD-SOI p-channel transistors showed a 30% increase over previously published transistors. The metal gate/strained silicon combination delivered a 20-25% higher NMOS performance relative to conventional strained-silicon transistors.
According to Sanders, a well-known problem with polysilicon gates is that a depletion region forms not only under the gate, where it's needed, but also on top of the gate. "That depletion region in the gate electrode has the effect of increasing the gate dielectric thickness. So you get an apparently thicker gate but you suffer the leakage currents of a thinner gate," he said. The Figure compares a typical polysilicon gate transistor (left) with the new metal gate transistor (right).
With a metal gate, this depletion region in the gate electrode is avoided. "What that does for us is it gives us an effective gate oxide thickness using the NiSi gate that's 4-5 Å thinner than with the conventional gate. A while back, 4-5 Å might not have sounded like much, but with the work we're talking about, these transistors are between 8-10 Å thick to start with so that's a very big deal."
Sanders said another advantage of the metal gate is that it can improve the mobility of electrons and holes in the channel. "In order to control the threshold voltage in a conventional structure, we need to implant impurities into the channel region. With this nickel silicide process, we're able to control the threshold of the device without implanting impurities into the channel. That once again helps to improve the mobility in the channel region. It gives us higher current levels and better performance."
Yet another advantage of NiSi gates is their manufacturability. "NiSi is certainly a more known material and maybe easier to implement that a high-k gate dielectric," Sander said. "(The move to) high-k gate dielectrics are proving to be one of those fundamental changes in materials that's a tough nut to crack."
Although silicided contacts for the gate, source and drain region have been used for some time to minimize contact resistance, all metal gates have been avoided because they are difficult to integrate. Also, many common metals have work functions that would require excessive channel doping. "You would have to put in a lot more dopant in the channel with many of the gate materials that are being looked at to achieve the result we want to achieve," Sanders said. That's not a problem with nickel.
Traditionally, polysilicon gates are topped with a silicide, which is formed by depositing the metal and then reacting it with the underlying silicon to form a silicide, usually done by rapid thermal processing. Source and drain silicides are formed at the same time. AMD used a similar process except that more nickel was deposited over the gate and then fully reacted. "We completely replace polysilicon with nickel silicide. It's an elegant solution," Sanders explained. "Furthermore, we found that, by appropriately introducing impurities to the polysilicon before we react the nickel, we can change the threshold of the transistor without having to introduce impurities into the channel. That's the nugget that's there for NiSi metal gates that we don't see for some of other gate materials such as tantalum that are being proposed, where you have to do other things, usually involving putting dopants in the channel, in order to set the right threshold voltages."
For additional information on wafer processing, go to www.semiconductor.net/wafer.