In this approach all parameters of the trapezoidal gate pulses (,
,
,
and
) are constant during the course of experiment.
The reverse bias
is constant as well and can be set to zero. The bottom
level
is varied to scan the interface while the top level
is
kept sufficiently high so that the whole interface becomes inverted during the
top level (for
-channel devices)
. In this method, the interface is
scanned along the characteristics
where
is
the coordinate actually scanned. This characteristic is responsible for the
part of the conventional
curve, discussed in
Section 3.5.1, from the maximum of this curve after
of the channel region until the right falling edge representing
of the channel region. Since the trap emission times are
constant during the course of experiment, the trap density is given with a
simple relationship
derived for symmetrical cases in Appendix G. A drawback of this approach is the variable oxide field while filling the traps with electrons during the gate top level. This effect can change the energy position of the interface traps as discussed in Section 3.1.3. The importance of this effect in measurements is, however, not clear at present. Experimental work is necessary here. If the experiments provide the same result by using the constant amplitude method assuming different gate amplitudes (and proportionally changed rise and fall times so that the slopes remain constant) the latter effect can be judged as irrelevant.
Figure 3.37 presents the charge-pumping characteristic
calculated numerically assuming a nonuniform trap distribution
within the source and drain junctions as shown in
Figure 3.38. The trap distribution extracted by
relationship 3.132 is in good agreement with the assumed
distribution, Figure 3.38. When applying the constant
amplitude method on the stressed devices expression 3.132
should be applied on the difference of currents in stressed and virgin devices.
Note that in this technique the scanned energy interval in the band gap is
variable during the course of experiment. The variations are, however, smaller
than those in method II, where the emission levels also vary due to the
variation of
.
In the analysis presented in this section all differentiations are carried out by using the sin-convolution filter, since the numerical data is too noisy to be differentiated by
finite difference formulas. The procedure is clarified in
Appendix G. By changing the filter with from to
it has been confirmed that the differentiation by the sin-convolution does not
influence the results significantly.
In the calculations, the relations or
are used which
are obtained for the virgin device. In the stressed devices a high density of
localized traps change the local potential, thereby leading to an error in
determining the exact position when
for a virgin device is assumed. To
analyze this error we calculate
for virgin and stressed devices
which correspond to the problem in Figure 3.34
(for the peak density of
). As obtained, the errors in
shown in Figure 3.39 and its derivation
are not significant in practice even for high trap
densities.