Channel material innovations will be required from the 65nm node onward to overcome the performance limitations of bulk silicon or process-induced uniaxial strain in order to maintain continued commensurate device scaling. Biaxially strained silicon, high germanium content SiGe alloys, and germanium are attractive materials for advanced CMOS technology due to their enhanced carrier mobilities. However, ion implantation process parameters for these target materials, necessary to produce a desired dopant distribution and junction depth, which are typically available for silicon, are not currently available. In this dissertation implantation of boron and arsenic at different energies and doses were investigated in selected high-mobility materials by using experimental results and physics-based simulations with MCIMPL-II.
Starting with an introduction to semiconductor doping theory and ion implantation technology, the general physical models necessary for the calculation of ion implantation distributions in crystalline targets were outlined. The Monte Carlo method is based on computing a large number of individual ion trajectories in the simulation domain by using scaled random numbers. The functionality of the ion implantation simulator MCIMPL-II was explained and the implemented models used for the trajectory calculation in silicon were discussed. The simulator uses locally either a crystalline or an amorphous model for searching the next collision partner of the ion. The application of the amorphous model in crystalline materials depends on a certain probability which is derived from the already produced local damage concentration. Therefore, this dynamic simulation approach considers channeling of ions as well as the transition of a semiconductor region to the amorphous state during the implantation process. However, the recent integration of MCIMPL-II with the Wafer-State-Server simulation environment was extensively tested and some improvements were made to the simulator (e.g. advanced smoothing procedure, addition of a one-dimensional grid generator). The accomplishment of a three-dimensional implantation application was demonstrated with the improved simulator for a non-planar structure in the WSS format. For the extension of the simulator from crystalline silicon to other group-IV materials, the crystalline model was modified and the empirical parameters of the electronic stopping model and of the damage model were calibrated. The accuracy of the obtained simulation results was evaluated by comparison of predicted doping profiles with SIMS measurements.
In this study a shift to shallower doping profiles was found for the implantation
of boron and arsenic into SiGe layers with increasing germanium content at a given
ion energy. The trend to shallower profiles is caused by the larger nuclear and
electronic stopping power of the heavier germanium atoms. It is shown that the
projected range of the boron distribution in Si
Ge
lies
well within the boundaries defined by the
values of the distributions in
silicon and in germanium. While a strong shift to the wafer surface is observed
for dopant implantations in SiGe alloys with low germanium concentrations (Ge
content
50%), the profiles in SiGe with high germanium concentrations are
all very similar. The formation of ultra-shallow junctions in a strained
Si/Si
Ge
system was performed by introducing ions at very low
energies. While the reduced atomic density in strained silicon of
approximately 99% of unstrained silicon (in-plane strain of
) has almost no impact on
the boron profile, the arsenic profile in strained silicon shows a slightly
deeper penetration compared to unstrained silicon. A special focus was put on
the boron implantation in germanium due to the large hole mobility enhancement
of germanium. The simulated point responses revealed that the boron distribution
is significantly reduced in germanium in vertical direction, while the lateral
profile is quite similar in silicon and germanium. An advantage of the Monte
Carlo simulation is that the generated point defects can be estimated, which
are associated with a specific implantation profile. We found that the higher
displacement energy in germanium, the stronger backscattering effect, and the
smaller energy transfer from the ion to the primary recoil of a collision
cascade are mainly responsible for the significantly reduced damage accumulation
in the germanium crystal.
Since the introduction of heavily nitrided gate oxides, NBTI has become a severe
reliability problem for high-performance CMOS technologies. Therefore, the NBTI
reliability was systematically investigated for a 90nm CMOS technology.
Experiments for different gate voltages, frequencies, and duty cycles were
performed in order to analyze the degradation of the p-MOSFET parameters. It
was acceptably verified that the R-D model can explain static and dynamic NBTI
data. The gate voltage and frequency dependence of NBTI was included by means
of an empirical relationship. All measured data could be well reproduced by the
performed numerical simulations. The presented simulation approach allows to
predict the lifetime for advanced CMOS technology, which depends strongly on
the applied stress operating conditions. Numerical simulations of the
degradation were performed with the calibrated model in order to analyze NBTI
degradation and lifetime of a CMOS SRAM cell. It turned out that the SRAM
lifetime is 1.14 times longer than the DC lifetime for using an unsymmetry
of 90% in the stress-split between the two p-MOSFETs of the cell. Thereby,
it is demonstrated that the R-D model is useful to study the NBTI response
not only for driving the transistor gate with periodic signals but also for
storing random bit sequences in an SRAM cell.
Finally, the extended simulator MCIMPL-II was applied to three-dimensional ion implantation problems to facilitate the introduction of high-mobility materials in future silicon-compatible advanced CMOS and optoelectronic device applications. A multi-dimensional TCAD-simulation analysis is advantageous over one-dimensional SIMS profiles, because the dopant distributions in the device structure can be studied. For instance, shadowing effects which arise for self-aligned large-tilt-angle implantations can be visualized, or the resulting dopant distribution after multiple implantation steps can be easily optimized by tuning the process parameters of a single step. However, an accurate ion implantation treatment of complex target structures requires very long simulation runtimes on a modern computer workstation. Future work will concentrate on both the speed-up of the methods used for trajectory calculation and on the parallelization of the trajectory calculation for shared-memory multiprocessor systems. It may be necessary to extend the simulator to a wider range of ion species or target materials. Since the calibration of empirical model parameters is a tremendous task, it would be useful to replace the empirical electronic stopping model by a more physically based model.