Abstract

To keep up with the growing demand for electronic products, a continous optimization of their mass production is necessary. The semiconductor industry as the main supplier in this market handles this optimization process via miniaturization of microelectronic devices, such as metal-oxide-semiconductor field effect transistors (MOSFETs) which are investigated here. As a consequence of the device shrinkage, reliabiliy issues like the bias temperature instability (BTI) have become a serious topic. BTI happens when the gate is biased while the transistor is exposed to elevated temperatures. This process severely changes some of the transistor parameters, e.g. the threshold voltage. A fundamental challenge in understanding BTI is that the degradation is found to recover when the bias is removed.

In this thesis the characterization of both negative and positive BTI is studied by using different measurement techniques. In addition to commercial measurement tools also equipment conceived and built by Hans Reisinger from Infineon Technologies AG is used. Unfortunately, there is no perfect measurement technique and each one exhibits certain limitations. Based on existing modeling attempts it will be shown that the delay time of the measurement has a huge impact on the characterization of the degradation and therewith on the projected time to failure. This is also the reason why BTI recovery is investigated thoroughly here. Furthermore, BTI is generally not specified in a consistent way because of the different characterization equipments used. A comparison of different measurement routines will show that the postprocessing of measurement output data is a very delicate task.

The most simple way to determine the BTI sensitivity of a device is to first take a reference of the quantity that should be characterized, then stress the device for a well-defined time and afterwards measure the change of the quantity. This is called the measurement-stress-measurement (MSM) method. An extended version thereof alternately stresses the device and then monitors the degradation during recovery with ever increasing stress times. The advantages of the MSM technique compared to others are its very short measurement delay time, its insensitivity to mobility changes and the possibility to obtain an unstressed reference prior to stress. In an extended MSM setup the MSM technique is further combined with the on-the-fly method, which monitors the stress. This allows the observation of both stress and recovery.

A main task of this work is to study both short- and long-term stress and recovery behavior. Though BTI has been known for some decades, the finding that its recovery is spread over many time scales is quite new. This is also the reason why it was thought for a long time that NBTI can be sufficiently described via the diffusion of hydrogen generated at the interface. However, the well-known reaction-diffusion theory is not able to explain the recovery by back diffusion of hydrogen. Furthermore, the temperature, oxide electric field, and frequency dependencies during stress, which are all observed in experiments, can not be modeled by this theory.

Newer modeling approaches are based on faster hole trapping processes and slower interface state generation. It took many attempts to find a possible mechanism that is able to explain the wide time range of the recovery which sometimes exceeds even decades in time. Such wide distributions of time constants have already been observed during the analysis of -noise spectra. Consequently, the models previously used for the explanation of -noise where taken as a starting point. In its extended form the defects are described by adiabatic potentials, which eventually determine the non-radiative multi-phonon (NMP) transitions between the various defect states. Upon the application of BTI stress the initial defect potential is shifted in energy and a transition into another defect configuration is favored. During recovery the transition back into its initial configuration is favored in turn. By such a mechanism it was already possible to explain the step-like recovery behavior of small-area devices by hole emission of single defects.

Large-area devices can also be modeled by using the same NMP theory with the only difference that more defects are necessary to describe BTI. These defects are assumed to exhibit different energies and distances inside the oxide. The correct description of measurement data that includes different temperatures, oxide electric fields, and stress times finally supports the validity of the NMP model for BTI.