The effect of bias temperature instability (BTI) occurs, whenever a bias is applied on the gate site of a transistor. The gate bias can reduce or increase the effective energy barrier for a charge transfer reaction between the charge reservoirs and pre-existing or created traps. This process can be accelerated by increasing the temperature, which increases the phonon energy, but also by increasing the applied gate bias. The higher phonon energy leads then to an accelerated energy dissipation of the electron excess energy. This leads to higher charge transition rates via a multiphonon process. BTI degradation has been known since the 1960s [243, 244, 245] and has become one of the main reliability concerns ever since. Historically, NBTI on pMOS devices was the largest concern, however, modern nMOS technologies are eventually affected by PBTI .
The charges that become trapped during a stress phase of a BTI experiment have an impact on the device electrostatics, leading to a change of . This perturbation can be approximated using the charge sheet approximation together with an estimate for the active number of traps . When the stress bias is lowered, the defects eventually release the trapped charge. The threshold voltage drift, which has been accumulated during stress by can recover back to the initial if it is fully recoverable or otherwise to a constant permanent offset added to . However, separating such a permanent and recoverable part experimentally is difficult which makes the analysis and theoretical description complicated. It is widely accepted that the recoverable component is caused by charge captured at defects in the bulk oxide which can be described with the NMP model . The permanent component can be modeled with an empirical double-well model or, as suggested recently, by considering the depassivation of Si-H bonds which leads to the creation of Pb -centers at the interface [246, 247].
Experimentally, a common approach for the characterization of BTI is using elevated gate bias and temperature conditions for characterizing the shift of the threshold voltage . One widely applied method is the extended Measure-Stress-Measure (eMSM)  scheme presented in the following section.
For this work, the extended Measure-Stress-Measure (eMSM) scheme  is the method of choice for the experimental characterization of BTI. The method consists of three phases and is shown in Fig. 8.1:
• Phase 1: First, an () curve is measured on a pristine device. This () curve is used as reference for further conversion from the measured drain current. For that, it is necessary to identify the of the transistor which can be done in different ways as discussed in detail in Section 6.1.3. For the characterization of large area devices, typically a constant current criteria is used for the definition of , e.g. . For scaled devices on SmartArrays, which show a comparable large variability, a constant current criteria is not practical. Instead, is defined using . The first phase is shown in Fig. 8.1 marked by ().
• Phase 2: After measuring the initial () curve, the device is stressed at a well defined condition given by a stress gate voltage and for a stress time . These phases are called stress phases, and are shown in Fig. 8.1 in the red sections. During stress, traps can capture a charge resulting in a decreased drain-source current and an increasing, corresponding , as can be seen in the lower part in Fig. 8.1.
• Phase 3: The third phase is called recovery phase. It is defined by applying a recovery voltage for a defined recovery period . During that period, typically the source current is measured, because it is more stable against oxide degradation than the drain current . The traps, which have captured a charge during the stress phase, can now emit their charge to the reservoir, thus the absolute value of the recovery voltage decreases, as indicated in Fig. 8.1. Typically, is mapped to using the initially measured (). If the device degradation is fully recoverable, then and decreases to zero. However, typically a permanent component, which does not decay anymore, is built up and potentially caused by Pb centers .
Phase 2 and phase 3 are repeated subsequently, which is done typically with increasing stress and recovery times. On large area devices, the measured recovery curves are smooth as they result from a large number of defects which emit their captured charges continuously, as can be seen in Fig. 8.2 (left) for a commercial 180 nm SiON technology node with drawn dimensions [MJJ2]. On scaled devices it is possible to obtain single capture and emission events as can be seen in Fig. 8.2 (right) on the same technology with drawn dimensions [MJC1]. The measurements on scaled devices build the basis for the time-depend defect spectroscopy (TDDS) . The TDDS is based on the observation that capture and emission times are statistically distributed. By measuring a set of recovery curves with the same and , the extracted set of step heights and associated emission times can be plotted in a scatter map resulting in a so-called spectral map. The resulting clusters correspond to different defects which allow further investigations of different properties, as discussed in detail in Section 8.3 and in .