The first section of this chapter collects the main findings across this thesis into a single section, where they are briefly summarized and brought into context. Based on these findings, the second section lists some ideas for future directions of research.
The main topics investigated in this thesis were charge feedback effects on the 
drift of GaN/AlGaN MIS-HEMTs at forward gate-bias stress, different methods for the extraction of the characteristic time-constants, and the extraction of single-defect parameters from nano-scale GaN/AlGaN fin-MIS-HEMTs.
To investigate charge feedback effects on large-area devices, a simulation study on GaN/AlGaN MIS-HEMTs was performed using the NMP four-state model. The effects of different feedback mechanisms on the experimentally observed capture and emission times were estimated, which led to the following conclusions:
• Transient changes in the surface potential caused by the charge feedback of the defects lead to increased defect levels and a decreased active energy area seen by the defects.
• It is not just the amount of trapped charges which changes due to charge feedback, but also the kinetics of charge capture and emission events.
• The different kinetics are caused by the change of the trap levels and the active energy area, but also by local potential perturbations influencing the characteristic time-constants of neighbouring defects.
• Due to these effects, common modeling approaches like the usage of accumulated stress times cannot be used as the observed defect kinetics are a strong function of previous stress and recovery cycles in the device. Thus, for such of devices, it is of utmost importance to use transient simulations following the experimental bias conditions as closely as possible.
To focus on the microscopic properties of individual defects, a theoretical investigation of Markov processes and the stochastic nature of charge emissions was used to put forward two novel methods to extract the characteristic capture and emission times from RTN signals:
• One method is based on spectral maps and is insensitive to long-term drift of the signal, suited best for basic RTN signals with high to medium SNR.
• The other method is based on a modified HMM and capable of handling more complex RTN signals coming from multiple defects. It can handle significantly more noisy signals and was implemented as a Python library which is able to extract the characteristic time constants of a system of defects by fitting the model to a set of measurement data. Furthermore, the library allows simulating the stochastic RTN emissions of a given system of defects.
Both methods were used to identify the bias-dependent capture and emission times of RTN producing defects in nano-scale GaN/AlGaN fin-MIS-HEMTs.
• The spectral method was used to calculate single-defect parameters like trap levels, vertical defect positions, and the apparent activation energy of two sets of coupled defects.
• The hypothesis of having a pair of coupled defects or a more complex defect structure was checked by estimating the required coupling factors of the defects from HMM simulations and comparing them to the factors calculated from a shielded Coulomb potential.
• Finally, two alternative defect structures were deduced from careful examination of the merged RTN traces. Their characteristic time constants were extracted from HMM training and compared to each other. As none of the two could be discarded with absolute certainty, a slightly modified parameter extraction to obtain their vertical positions, trap levels and thermal activation energies was carried out for both of the candidates.
This work contributed to a more profound physical understanding of the defects responsible for BTI in GaN/AlGaN MIS-HEMTs. For large-area devices, the importance of charge feedback effects on a proper interpretation of the observed time constants in BTI measurements in GaN technology was highlighted. Furthermore, two innovative methods to obtain the stochastic capture and emission time constants from defects were introduced in this work. With these methods, for the first time, the microscopic structure, vertical positions, energy levels, and temperature activation of RTN producing defects could be extracted for GaN.
In general, the presented methods for the extraction of the characteristic time constants are formulated universally enough to be useful for single-defect investigations in many different semiconductor technologies. This is mainly justified by the fact that neither the spectral method nor the HMM depends on any physical defect model except the Markov property (i.e., being a memoryless system).
