5. Summary and Conclusions

THIS DISSERTATION has presented the technology, modeling, and simulation of SiC semiconductor devices. First, the SiC semiconductor device technology was briefly surveyed. Among over 170 polytypes with different physical properties available in SiC, 3C-SiC, 4H-SiC, and 6H-SiC are the most common polytypes presently being developed for device applications. These polytypes are characterized by the stacking sequence of the biatom layers of the SiC structure. 3C-SiC is the only form of SiC with a cubic crystal lattice structure. 4H-SiC consists of an equal number of cubic and hexagonal bonds, while 6H-SiC is composed of two-thirds cubic bonds and one-third hexagonal bonds. The overall symmetry is hexagonal for both 4H- and 6H-SiC polytypes, despite the cubic bonds which are present in each.

Changing of the stacking sequence has a profound effect on the electrical properties. Since hexagonal polytypes are made up of stacked double layers, several material properties are anisotropic. The degree of anisotropy is measured by the quotient of a parameter value parallel and perpendicular to the c-axis. Comparisons of the SiC material properties with the silicon material properties including the anisotropic values are shown and prepared for model development.

Currently only the 4H- and 6H-SiC polytypes are commercially available as substrate material. Key crystal growth and device fabrication issues that presently limit the performance and capability of high-temperature and high-power 4H- and 6H-SiC devices are identified. The differences between SiC device technology and silicon VLSI technology are discussed. Projected performance figure of merits of SiC devices are highlighted for several applications.

The primary objective of the SiC physical modeling part was to perform a comprehensive and systematic model development, and implementation into MINIMOS-NT based on the recent research findings and published data. Due to the anisotropic nature of the SiC crystal structure, the mobility $\mu$, the dielectric permittivity $\varepsilon$, and the conductivity $\kappa$ are tensors, which are diagonal with only two independent components parallel and perpendicular to the c-axis, respectively. A tensorial formulation of Poisson's equation and the current equations have been adapted into MINIMOS-NT applying the same discretization scheme as in the case of conventional current transport equations.

The most common dopants used for 4H- and 6H-SiC are "deep" impurity states because of their energetic position within the bandgap. Inequivalent sites of SiC, one with cubic surrounding and the other with hexagonal surrounding, cause site-dependent impurity levels. Therefore, an appropriate incomplete ionization model which accounts for ionization level dependence on temperature, polytype, and the location of the dopants has been implemented.

A variety of other SiC-specific models, including band structure and bandgap narrowing; SRH and Auger recombination; mobility dependencies on impurity concentration, lattice temperature, carrier concentration, carrier energy, parallel and perpendicular electric fields; breakdown models including stress dependent, leakage current, and temperature- and field-dependent impact ionization and other models have been implemented.

The physical modeling and implementation was followed by the realization and the application of the models to state-of-the-art SiC devices for both semi-conducting and semi-insulating-based devices. Three classes of SiC rectifiers have been investigated. The Schottky barrier diodes, which offer extremely high switching speed but suffer from high leakage current; the PiN diodes, which offer low leakage current but show reverse recovery charge during switching and have a large junction forward voltage drop due to the wide band gap of SiC; and the merged PiN Schottky diodes, which offer Schottky-like on-state and switching characteristics, and PiN-like off-state characteristics.

The models are further tested on a UMOSFET, which was the first unipolar transistor realized in SiC. The simulation has shown that despite the good on-state and off-state characteristics of the device, the UMOSFET suffers from problems including lower inversion layer mobility and high electric field crowding at the trench corners in the blocking state. The former problem is mainly due to the location of the channel which lies along the vertical sidewall of a trench. The current flow in the channel is parallel to the c-axis of the crystal where carriers must cross alternating silicon and carbon planes of the crystal. The latter problem is caused by the sharp edges of the trench corners in the UMOSFET.

Simulation results demonstrated that the trench problems in the UMOSFET can be avoided by the planar DMOSFET structure formed by using a double ion implantation. However, the DMOSFET is not without its own problems. The consideration of the channel mobility set at $\sim$10% of the bulk mobility during the simulation analysis gives rise to low inversion layer mobility comparable to the UMOSFET. This value assumption was made based on the measurement data of the low channel mobility in the DMOSFET which is due to the surface roughness in ion implanted base regions followed by high-temperature annealing. The simulation result has clearly shown that, in addition to the resistance of the MOS inversion layer, the resistance of the JFET region between the implanted p-base regions increases the specific resistance of the DMOSFET. This results in a lower output current.

The new device structure called accumulation-channel or ACCUFET was proposed by incorporating an n-type counter-doped layer along the oxide/semiconductor interface to restore the low inversion layer mobility observed at both UMOSFET and DMOSFET. Simulation based comparison were conducted between this structure and the conventional structure. Significant improvements on the electrical performance and device reliability were achieved.

Finally, the models have been tested for RF transistors. A MESFET fabricated from 4H-SiC has been investigated for both DC and high frequency characteristics. Excellent agreement between the simulated and measured data were obtained. This MESFET produces a maximum channel current of about 250mA/mm while simultaneously providing a high breakdown voltage of 110V. The device has also shown an $f_\mathrm{t}=5.62$GHz, an $f_\mathrm{max}=37.18$GHz, and an RF gain of 15dB at 1GHz. These results clearly demonstrate the advantages of SiC for high power microwave application where its high thermal conductivity, high voltage and high powerdensity capability are very attractive.

As a conclusion, the implemented models and their corresponding parameter values can be applied to wide bandgap semiconductor devices. However, there are some issues which are not yet addressed due to the poor quality of the SiC materials and lack of understanding their detailed physics. Further experimental and theoretical work is required for developing a comprehensive SiC device simulation tool which performs at the same level as that of Si device simulation tools. No experimental data about the carrier-carrier scattering in SiC at high current ratings is available so far. There is a big challenge ahead on the modeling of the SiC-SiO$_2$-interface that is currently limiting the commercialization of SiC MOS devices. However, there is no doubt that, aside from the extensive progress of SiC device technology, the availability of predictive simulation regarding the design and optimization of SiC devices will grant the further success of modern electronics.

T. Ayalew: SiC Semiconductor Devices Technology, Modeling, and Simulation