Chapter 8
Conclusion and Outlook

Due to the downscaling of semiconductor device geometries, modern MOS technologies are becoming increasingly prone to reliability issues, in particular to negative bias temperature instability (NBTI). This parasitic effect seriously limits the lifetime of the devices and has thus aroused considerable scientific interest. The data obtained by time-dependent defect spectroscopy (TDDS) provide experimental evidence that this NBTI issue is not related to a diffusion-controlled problem but rather to a hole trapping process. However, the exact nature of this process has remained vague and thus the charge transfer mechanism involved in hole trapping has shifted into the focus of interest. In this thesis, different kinds of mechanisms were taken into consideration and studied within the framework of rate equations. For each mechanism, the simulated NBTI degradation was compared to the data extracted from the extended measure-stress-measure (eMSM) technique. In these studies, only the short-term part of the NBTI degradation was considered since it is only weakly obscured by the permanent component of NBTI, which is attributed to another degradation mechanism.

The first charge trapping mechanism investigated was elastic tunneling of the charge carriers from the channel into the defects. The elastic tunneling model predicts a logarithmic time behavior, which is also observed for the short-term degradation of NBTI. Nevertheless, this concept has been ruled out for several reasons. First, the elastic tunneling model predicts a negligible temperature dependence, which is inconsistent with the experimental findings of the eMSM technique. Second, its field acceleration is linear instead of quadratic. It is noted that the temperature as well as the field dependence are inherently given by the physical foundations of this model and thus cannot be adjusted by any fitting parameters. Furthermore, modern device technologies have ever smaller oxide thicknesses so that the tunneling to and from the gate enters in the trapping dynamics. This aspect was paid attention in an extended version of the elastic tunneling model used to study the impact of the gate contact. It was found that elastic tunneling is limited to a time range below for oxide thicknesses smaller than . As a consequence, this concept cannot give an explanation of the wide distribution of time constants seen in NBTI experiments.

Another model attempt starts from the assumption that the defect configuration is closely linked to the position of the trap level. Since defects undergo structural relaxation after each charging event, the trap levels rise or fall within the oxide bandgap. This shift of the trap level was expected to have a strong impact on the trapping dynamics. As a consequence, it was incorporated in a new, rate-based model, which was considered as a possible explanation for hole trapping in NBTI. Using first principles calculations, it was verified that the level shift can be significant. These theoretical studies included the usually suspected defects, such as the oxygen vacancy, the centers, the hydrogen bridge, and the hydrogen interstitial, where each of them features a shift of at least one electron Volt. In a second step, the level shift model was investigated for its time dynamics as well as for its field and temperature dependence. Even though it shows a nearly logarithmic time behavior during stress and recovery, it can explain neither the field nor temperature dependence observed in NBTI experiments. As a consequence, this model must also be abandoned as an explanation for NBTI.

The level shift model was extended to account for the activation over thermal barriers. This modified variant, called nonradiative multi-phonon (NMP) theory, was used to describe the charge capture and emission process within bulk materials and has been employed in the two-stage model. This new model achieves a good match with the complicated NBTI stress and relaxation degradation curves for various different gate voltages and temperatures. As such, it fulfills all criteria inferred from the eMSM data and can thus be regarded as a successful model to describe NBTI. In order to test whether the two-stage model properly reflects the behavior of the microscopic processes, it had to be evaluated against the field and temperature dependent hole capture and emission of single defects. For this purpose, the simulated time constants were compared to results of TDDS experiments. The two-stage model was found to yield the correct temperature activation and successfully reproduce the field dependence of the ‘normal’ as well as the ‘anomalous’ defects. However, this model cannot explain the curvature in the experimentally obtained capture time constants. In an improved description of the NMP process, the heights of the thermal barriers were not assumed to be statistically distributed as in the two-stage model but derived from atomistic quantities, such as the vibrational frequencies of a defect. Furthermore, a second metastable state was introduced into the two-stage model in order to allow for temperature independent emission times and a curvature in the capture times. This refined model, called extended NMP model, was shown to capture all experimental feature seen in the TDDS data and, as such, gives an improved description of the microscopic charge transfer process. The validity of the extended NMP model was further supported by the fact that it can also give an explanation for anomalous and temporary random telegraph noise. Therefore, this new model is an important step forward in the understanding of NBTI.

In this work a new model has been devised, which captures the essential physics of charge trapping in MOSFETs. It depends on a couple of model parameters, which are directly linked to the physical properties, such as equilibrium configurations, barrier heights, and positions of the trap levels. By comparison to TDDS data, the range of these parameters could already be narrowed down. Given that information, a very detailed list of requirements on the microscopic defect properties can be compiled. Unfortunately, a corresponding microscopic defect has still not been identified. The most prominent defect candidate has been the oxygen vacancy, which shows the bistability required in the extended NMP model, as already summarized in the Harry-Diamond-Laboratories model. However, recent first-principles simulations have shown that the trap levels of the oxygen vacancy are located far below the substrate valence band edge so that this defect must be ruled out for this model. Therefore, future investigations should be devoted to a systematic search for defects qualified according to the extended NMP model. These investigations should include defect candidates in the various dielectric materials, including silicon dioxide, silicon oxynitride, and high-k dielectrics. Most of these materials are expected to be amorphous so that the defect properties are subject to statistical variations. Furthermore, it has been realized that the defect properties can be seriously affected by the presence of a nearby interface. All these aspects should be accounted for in future investigations on hole trapping. When the question which defects is involved in charge trapping is solved, their occurrence could even be suppressed during processing.