AUSTIN, Texas Technologists are girding for yet another major change in CMOS gate stack integration: the replacement of polysilicon gate electrodes with two metals for PMOS and NMOS transistors.
Metal gates will be needed because depletion of the polysilicon atoms can add to the thickness of the gate oxide, degrading the transistor switching speed. A switch to high-k oxides, which is expected to begin in earnest in 2005, is part of the effort to cut leakage current for both high-performance and low-power integrated circuits.
Metal gates add to the problem's complexity, however, and will be a focus at the three-day International Electron Devices Meeting starting Dec. 9 in San Francisco.
At IEDM, Motorola Inc. researchers will present experimental work integrating tantalum silicon nitride (TaSiN) as the NMOS gate electrode, and titanium nitride (TiN) as the PMOS electrode, on a hafnium oxide gate dielectric.
Joe Mogab, vice president at Motorola's advanced products research and development laboratory (APRDL), said polysilicon, even when heavily doped, is not a true metal, and must be doped with different materials to create the NMOS and PMOS transistors. Dopants tend to move during high-temperature steps to the interface between the polysilicon gate electrode and the gate oxide. The problem comes when those atoms, particularly the boron atoms used to create p-type transistors, get into the gate and channel regions, causing unpredictable and unacceptable spreads of the threshold voltage.
While metal gates do not need to be doped, there is not one metal that can set the work function the energy required to pull an electron free from the surface of the electrode for both NMOS and PMOS devices.
"To replace polysilicon we need to find two metals, and that presents a major materials and integration challenge to the industry," Mogab said.
Already, budgets for semiconductor process research are being stretched thin by the need to find a replacement for the oxynitride (SiO2 with nitrogen doping) dielectrics used today.
65-nanometer process target
Texas Instruments Inc. has publicly said it plans to use a hafnium-silicate gate oxide, starting in the latter half of the 65-nanometer node. Motorola also is likely to use a hafnium silicate, gradually shifting to a hafnium oxide dielectric, and then introducing metal gate electrodes on the hafnium oxide insulator.
"The need for a new gate stack is there at the 65-nm node," Mogab said. "Our strategy is to approach the problem in a stepwise fashion, because adding the metal gates is harder to do than first appeared. But there is also a strong possibility that we will go right through the throat," Mogab said, combining the metal gate electrodes with the hafnium oxide gate insulator.
One benefit of hafnium oxide, compared with hafnium silicate, is that HfO2 is resistant to the etching steps used to integrate different metals. With hafnium silicate, the ability to etch and stop is much more difficult, and as the insulator becomes extremely thin the chances increase of rupturing through the hafnium silicate during the etch steps.
Phil Tobin, manager of the MOSFET materials development and integration group at Motorola's APRDL, said in terms of the risk of moving away from oxynitrides, "Hafnium silicate is more attractive, and we will definitely need a high-k material at the 65-nm node, both for high-performance and low-power products."
Metal gates could be used at the 45-nm node, "or as soon as we feel comfortable introducing that," Tobin said. Significantly, developers switching from polysilicon to metal can achieve a 2-angstrom to 3-angstrom improvement in the effective or electrical thickness of the dielectric. The improvement occurs largely because the problem of polysilicon depletion at the gate interface is removed entirely.
Motorola's work with two metals takes advantage of the etch selectivity of hafnium oxide, with a process flow that adds only one noncritical mask for removal of the titanium nitride from the n-channel, said Sri Samavedam, a lead researcher on the metal gate electrode effort.
But the search for metals with the proper work function is not over. TaSiN and TiN "are not our final candidates. We have done an extensive amount of modeling on the work function of various metals. As the work function moves away from polysilicon to metals, our team has found a couple of tenths of electron volts of room for improvement" over the metals described in the IEDM paper, said Samavedam.