In Chapter 3 it was tried to explain BTI by using either the diffusion of
hydrogen or a dispersive bond breaking mechanism. In both cases interface states
are involved. Unfortunately, the theoretical and experimental analysis of
the on-the-fly interface traps (OFIT) technique presented in the last
chapter revealed that the aberrations leading to the assumption of fast
interface state stress and recovery are due to an artifact of the measurement
routine. Since the recovery of BTI, especially its short-term behavior, is not
explicable with interface states only, hole trapping models have been added
[69, 40, 119, 120]. Today the BTI community does still not agree on
how holes contribute in detail. The earliest hole modeling attempts date
back to the 1950s, where McWhorter used hole trapping to describe
-noise at germanium surfaces [121]. More precisely,
-noise was
considered as oscillations of the trap occupancy of individual defects caused by
capture and emission of carriers. McWhorter’s attempt is based on the
Shockley-Read-Hall (SRH) theory which was originally developed to
model the recombination of bulk defects with an energy
inside the
bandgap [122]. He extended this theory to also model oxide defects, which
feature a trap level within the semiconductor bandgap. The local depth of
the oxide defect
, measured from the interface, enters the model
as a tunneling WKB factor
, where
acts as scaling
factor.
When assuming a defect at capturing a hole from the reservoir in the
substrate, e.g. from
, the hole does not have to surmount a barrier because of
. For the opposite process, namely the hole emission from the
defect, the transition probability is reduced by the Boltzmann factor
. However, the application of this approach to a defect level
, which can be assumed for oxide defects, makes the above
Boltzmann factor larger than unity in the simplest picture. The hole
emission barrier rather vanishes in the case of
. In turn the
corresponding capture process is now affected by an additional Boltzmann
factor
[123]. The hole capture
and emission
barriers for both kinds of defect, leveling above
for the simple
SRH and below
for the extended SRH, are all depicted in Fig. 8.1
(left).
When an additional oxide electric field is present, the defect level is
shifted with respect to
. Since the barrier
is
linearly dependent on
, the defect may now effectively lie below or above
, cf. Fig. 8.1 (right). Unfortunately, the McWhorter model was originally
developed for
-noise in thick oxides and not designed to explain the
strong temperature and bias dependence observed during BTI stress in
modern devices with oxide thicknesses of only a few nanometers, e.g.
. In such devices the McWhorter model only gives time constants
smaller than a millisecond, which contradicts the measurement results
[55].
About thirty years later Kirton and Uren used a modified McWhorter model
to explain their random telegraph noise or signal (RTN/RTS) measurements,
which characterize the change in the drain current of small-area MOSFETs as a
function of time. The times where the signal randomly jumps into the high- and
low-current were identified to be Poisson distributed around the expectation
value of the capture and emission
time constants of individual defects
respectively. To link this capture and emission kinetics to the observed
-spectra, Kirton and Uren proposed the existance of many defects with
uniformly distributed time constants on a log scale ranging from milliseconds to
days [124]. Since they expected a multi-phonon emission (MPE) process to be
responsible for their experimental findings, they added a thermal barrier
to the existing SRH model [125, 126, 127, 128]. This approach will
be continued in the next chapter, where a mathematical description is
presented.