The phenomenon that the threshold voltage of metal-oxide-semiconductor field-effect transistors changes, when the device is stressed at elevated temperatures, has been observed first in the 1960s and termed the bias temperature instabilities (BTI). It is commonly accepted that the threshold voltage shifts can be attributed to defects located inside the oxide, so called border states and interface states. By investigating BTI in large area devices the collective response of a vast amount of defects can be measured as a continuous degradation and recovery behavior. However, to model the complex nature of BTI properly, a detailed knowledge of the physical mechanism behind charge trapping of single defects is required. This can be achieved by using nanoscale devices, which in contrast to their large area counterparts, contain only a handful of defects with experimentally resolvable threshold voltage shifts. As a consequence, the intricate charge trapping behavior can thus be studied for each defect individually. For the analysis of single charge trapping, the time-dependent defect spectroscopy (TDDS) has been recently proposed. The substantial amount of manual effort currently necessary to analyze TDDS data call for a more automated TDDS workflow. One particular time-consuming task during the TDDS analysis is to identify clusters in the recorded data. Each cluster is subsequently linked to a particular defect in order to obtain statistical parameters of the charge transition times. A detailed understanding of charge trapping for a certain technology requires the trapping parameters of a large number of defects. To simplify the process and increase the accuracy of the extraction of the charge transition times, a sophisticated data analysis algorithm has been developed. This work describes the implementation of an unsupervised algorithm based on expectationmaximization (EM) to perform an automatic cluster detection. Satisfactory results, compared to manually analyzed data, are achieved using the presented algorithm. In addition, the effort necessary to identify single traps is significantly reduced. Although the algorithm requires further optimization with respect to the assignment of clusters to defects, this work offers the ability to efficiently study numerous single traps.