This work deals with the characterization of the bias temperature instablity (BTI) from a metrological approach as well as from a modeling approach. To resolve the question “what to model”, well-designed measurements are of utmost importance. Numerous measurement techniques have therefore been compared and their applicability to BTI was carefully checked. Although it is seemingly obvious, issues like the measurement delay or the impact of the measurement setup on the device have been ignored for a long time. Just a few years ago the characterization of the negative BTI (NBTI) of pMOS devices as the most serious BTI condition was only focused on the stress phase, i.e. the degradation during BTI. This stress phase was widely modeled via the reaction diffusion theory which assumes the diffusion of hydrogen into the oxide. However, as soon as the stress is removed, a part of the degradation was found to recover, which can not be adequately described by the reaction diffusion theory and extended variants thereof.
The final failure of the long-reigning reaction-diffusion theory has demonstrated that without the use of significant and proper experimental NBTI data, modeling attempts are highly questionable, especially when they are based on incorrect premises, like the previously ignored recovery after NBTI. Upon the emergence of refined measurement techniques, many dependencies of the NBTI mechanism were explored, e.g. voltage, temperature, and frequency dependence to only mention a few, and new modeling attempts were published. While some attempts focused on the existence of interface states only, others assumed faster hole traps and slower interface states to contribute to NBTI. For the latter assumption it was suggested that the recovery behavior can be split into a recovering part due to hole traps and a permanent part due to interface states. In order to analyze the role of the recoverable component, measurement delay times down to the microsecond-regime became important.
The main experimental challenges discussed in this thesis include the synchronization of in- and output data with a high time resolution, the reduction of noise, and the clear distinction between stress phase and recovery phase when a pulsed measurement technique is used. Especially, the transition phase between stress and recovery is undefined. It is not surprising, however, that there is no perfect measurement technique and each of them has its specific drawbacks.
-characteristics. When carefully handled, delay
times of a microsecond can be achieved. A general disadvantage of
pulsed measurement techniques lies in the characterization of PBTI of
pMOSFETs. Although stress is performed in the accumulation regime
in this case, degradation has to be characterized in inversion anyway.
As unwanted side effect this may yield small NBTI stress and hence
obscure the real PBTI-induced degradation.
Using these measurement techniques the following results are obtained:
Irrespective of which stress condition (NBTI or PBTI) is imposed to which
device type (nMOS or pMOS), always a negative shift of the threshold voltage is
observed for thin oxides, with the largest degradation for the NBTI/pMOS case.
Furthermore, the degradation can be split into a recoverable hole trap component
and a permanent interface state component. When the recovery is monitored
with a too large measurement delay time, a fraction of the recoverable part can
be missed. Short-term NBTI stress and recovery measurements in the range of
to 
underline the challenge that the measurement delay time as well as
the settling time of the applied gate pulse have a huge impact on the
monitored recovery. A settled gate pulse for example may miss a part
of the degradation and recovery which whould yield spurious effects.
For this reason a more liberal level for the pulse settling time should
be used instead, involving the risk of mixing the stress and recovery
phase.
To rigorously model BTI, it is important to not only study the short-term but also the long-term recovery behavior on an equal footing. New measurement results indicate that the previously assumed permanent part actually exhibits recovery, albeit on a large time scale. When the same type of pMOS previously subjected to NBTI stress is PBTI-stressed instead, a different recovery behavior is obtained. This can be explained by the assumption of a “general recovery behavior” which covers all aspects of the previous stress conditions. Depending on the oxide electric field and stress time only a part of the recovery is visible in the measurement. The invisible part of the general recovery is either too fast to be observable, which is assumed to be the case for NBTI, or so late that even an experiment lasting for two weeks is too short to capture the full recovery characteristics. The latter case is assumed for PBTI.
The fact that recovery even occurs below a microsecond and continues for at
least weeks requires a model that is able to explain large time scales over more
than 
decades. Modelling attempts using Shockley-Read-Hall-like (SRH)
processes are ruled out because they feature a too small time constant range.
Also the field and temperature dependence can not be explained with the SRH
model. The latest attempt is based on the non-radiative multi-phonon (NMP)
theory, where hole capture and emission time constants depend on the barriers
between two different defect states. Upon the application of stress, one
defect configuration is shifted with respect to the other, which favors a
hole capture process during stress. During recovery the energy levels
between the two defect levels favor hole emission. Depending on the kind
of defect this gives a well-defined pair of a capture and emission time
constant.
The step-like recovery behavior of small devices is a clear demonstration of such hole emission events from single defects. However, larger devices contain a larger number of defects. When in the simplest case a uniform distribution of time constants is assumed, a log-like recovery behavior consistent with the measurement results is obtained. This indicates that the underlying mechanism during BTI is actually based on the superposition of a large number of different defects that each feature different pairs of capture and emission times.
Finally, the recovery of high-k SiGe pMOSFETs was modeled by an extended NMP theory which is able to describe the behavior of switching oxide traps during NBTI: The multi-state defect model features two bistable defect states, each consisting of a stable and a metastable defect level. Effective capture and emission rates between the two stable defect levels determine the occupation of the defect. The complexity of the devices required the implementation of quantum mechanical effects like quantization in the channel to successfully model a large experimental dataset of different stress times, voltages and temperatures. To that end, a distribution of energies, barriers, and positions inside the oxide were assumed. The occupancies of all defects were finally summed up over all subbands. The overall degradation yielded excellent agreement with the experimental data, strongly indicating that the extended NMP theory is valid for NBTI.
However, the microscopic structure of the defect(s) contributing to NBTI and PBTI still remains vague. Recent publications have performed density function theory calculations using the oxygen vacancy and the hydrogen bridge as possible defect configurations. Unfortunately, the obtained thermodynamic energy levels of both defects are not in agreement with the experimental observations made during BTI. A successful defect identification would help the semiconductor industry to alter the manufacturing process in a manner that BTI would not be longer a serious reliability issue. Also, with the successful understanding of BTI it would be finally possible to make clearer predictions on the lifetime of MOSFETs.