High-k Gate Dielectrics: No Easy Solution
Peter Singer, Editor-in-Chief -- Semiconductor International, 2/1/2003
Two somewhat conflicting transistor requirements that may ultimately limit how long device scaling can continue are the need for a high drive current to optimize device performance, and low leakage currents to minimize power dissipation. At the heart of this conflict is the gate dielectric: The thinner the dielectric, the higher the drive current, but the higher the leakage currents.
The industry is attacking this problem in several ways. One way is to keep the dielectric thickness the same, but optimize the transistor around it (i.e. through aggressive halo doping or by radical new designs such as the dual-gate FET). Another is to develop alternative materials with higher dielectric constants that would replace the silicon dioxide traditionally used for the gate dielectric. High-k gate dielectrics could be made thinner, yet still provide acceptable protection against leakage currents.
This quest for a suitable high-k gate dielectric is not new, and many different materials have been considered, ranging from aluminum oxide with a k of 8-9, to hafnium and zirconium oxides, which have a k of 20-30 (see Semiconductor International, November 2002 ). HfO2 has been the most heavily researched, as shown by the number of papers presented on different high-k materials at key conferences (Figure).
The challenges are not trivial in that success will hinge on the ability to achieve high layer uniformity, integration with other processes, minimal/controlled reactions with silicon and the gate electrode, and low defect densities. Interface chemistry might also necessitate the use of a metal gate electrode. In addition to physical thickness, high-k dielectrics are often characterized by their equivalent oxide thickness (EOT).
The 2002 update to the ITRS proposes a slightly more aggressive requirement for equivalent physical oxide thickness of 1-1.4 nm (it was 1-1.6 nm). "Although promising high-k candidate materials have been identified, fundamental performance and reliability issues, as well as issues with CMOS integration, are still under investigation," the 2001 ITRS notes.
In a paper presented at the International Electron Devices Meeting last December, Hiroshi Iwai, a professor at the Tokyo Institute of Technology, noted that the following problem areas still exist for high k: interfacial layer formation, microcrystal growth, lateral oxidation at gate edge, lower mobility, fixed charge/flatband shift, higher density of interface states, boron penetration, and contamination from the CVD precursor. IBM researchers added that charge trapping in high-k films causes the threshold voltage to shift with time under a gate bias, and is therefore an important reliability issue for high-k films.
One of the main problems is that common high-k dielectrics will crystallize at a fairly low temperature (much less than 900°C). Crystal grain boundaries then can act as high dopant diffusivity paths and may also be the cause of device failure and high leakage. As reported at IEDM, Toshiba researchers, for one, observed that crystallized -HfO2 portions "strongly degrade" the electron mobility in hafnium-silicate MISFETs.
The good news here is that adding nitrogen to the film helps keep the film amorphous at temperatures of up to 700-800°C, and also helps improve the dielectric constant. Samsung, for example, found that incorporating nitrogen in HfSiON definitely enhanced its dielectric constant. It was also demonstrated that the dielectric constant of ~13 was retained through 1000°C annealing for HfSiON, while it severely degraded without nitrogen. A polysilicon/HfSiON/silicon stack was stable up to 1100°C and showed a reduction in leakage current by more than three orders of magnitude compared with that of SiO2 obtained with a nitrogen concentration of 30%. Researchers said that excellent properties of HfSiON were owed to the suppression of crystallization, phase separation and microscopic inhomogeneity of the film, which can be achieved with nitrogen over 20%.
Progress continues to be made, but it is unclear when a manufacturable solution might be developed. It's likely that nitrogen-doped HfO2 might be the first high-k gate dielectric that goes into production. Longer term, Iwai and co-workers suggest that lanthanide oxides might be the best option.
For additional information on wafer processing, go to www.semiconductor.net/wafer.
