An increasing number of publications is based on the assumption that there are two components responsible for NBTI: A fast, or universally recovering component on top of a slowly recovering or permanent component [6, 49, 68, 30]. However, the origin of the permanent and recoverable component has not yet been identified, as some authors, e.g. [6], claim that state interface states are permanent and oxide charges are recoverable, while others [31, 49] claim interface states to be solely responsible for NBTI. To reveal the responsible defects, two measurement techniques frequently used at present were studied in this chapter, the measurement-stress-measurement (MSM) routine and the on-the-fly (OTF) method. Based on simulations augmented by suitable models for interface and oxide charges published in [39], it was tried to explain the experimental results of both techniques.
As expected, both techniques have their specific drawbacks. While the MSM-sequence suffers from an in-situ measurement delay, the OTF-techniques lack the initial reference measurement with the OTF1 and OTF3 extraction additionally being affected by mobility degradation. Moreover, the conversion routine to the threshold voltage shift introduces inaccuracies due to the simplifications made by the compact modeling, already explained in Chapter 2.3.
Nonetheless, the smaller systematic errors are found within the MSM
routine. Despite its intrinsic delay, the time evolution of the recovery
after BTI stress can be monitored most accurately. By using several
MSM-sequences in a single measurement, the overall stress and relaxation
behavior can be reconstructed as follows: Each recovery sequence can be fitted to
the universal relaxation model by an optimization loop. The extracted
permanent part 
, i.e. the remaining degradation at the end of the
extrapolated relaxation behavior, and the recoverable part 
finally render the
possibility to describe the influence of several stress parameters, like the
temperature acceleration for BTI. While 
seems to exhibit Arrhenius-like
behavior with 
independent of the stress time, 
does
not.
Finally, by using 
and 
it is also possible to explain the various values
of extracted power-law stress exponents reported in literature. After a too long
delay, i.e. 
, mostly 
is left to monitor during the relaxation, while 
has already disappeared. This makes the exponent depend on the delay time.
Since higher delay times yield higher exponents, a lifetime extrapolation via such
an exponent is questionable.