# The Physics of Non–Equilibrium Reliability Phenomena

#### 5.4 Alternating Stress Modes

To finally stress the conceptual limits of independent degradation mode descriptions, the following experiments and simulations have been performed using alternating stress conditions followed by a final relaxation phase, see [MJJ2]. In total 14 devices, divided into two groups, have been stressed and measured at a temperature of $$T=\SI {443}{K}$$ and bias conditions of $$\Vg =\SI {-2.8}{V},\,\Vd =\SI {0.0}{V}$$ (BTI) and $$\Vg =\SI {-0.5}{V},\,\Vd =\SI {-2.8}{V}$$ (HCD). The first group was subjected to $$\SI {10}{ks}$$ of HCD stress followed by a $$\SI {10}{ks}$$ BTI stress phase and a final relaxation of $$\SI {10}{ks}$$ at recovery conditions ($$\Vg =\SI {-0.5}{V},\,\Vd ={0.0}{V}$$). The stress sequences have been switched for the second group of devices, i.e. BTI with a subsequent HCD cycle and a final relaxation phase. The acquired experimental data set is summarized in Fig. 5.23. After the HCD stress cycle the first group experiences a threshold voltage shift of $$\dvth =\SI {40}{mV}$$, which increases to a drift of $$\dvth =\SI {170}{mV}$$ due to the subsequent BTI cycle. Within the relaxation phase the degradation recovers to $$\dvth =\SI {98}{mV}$$. Interestingly, if BTI is applied prior to HCD, the first stress phase creates $$\dvth =\SI {140}{mV}$$ of total degradation. Even more intriguingly, it seems that the following $$\SI {10}{ks}$$ HCD stress does not trigger additional damage, but rather decreases the $$\dvth$$ shift by $$\SI {65}{mV}$$. Moreover, the final $$\SI {10}{ks}$$ relaxation period does not show any additional recovery effect.

By means of the simulation framework introduced above, see Sec. 5.3, a combination of the hot–carrier degradation model and the NMP$$_\mathrm {eq.+II}$$ model developed in Chapter 4, it is possible to shed light on the mechanisms behind this puzzling phenomenon. The results for the first group of devices (HCD$$\Rightarrow$$BTI$$\Rightarrow$$Relaxation) is illustrated in the top panels of Fig. 5.23. The initial pure hot–carrier stress induces $$\sim \SI {45}{mV}$$ of degradation, mainly due to the presence of energetic electrons at this specific (off–state–stress) bias combination, see Fig. 5.19 (left panels). The subsequent BTI stress and relaxation cycle follows the well known characteristics of NBTI, matching the experimental results very well. One can, therefore, conclude that HCD only adds a pre–existing damage to the device, which slightly perturbs the device electrostatics, but without further implications for oxide defects.

The situation for the second group of devices, exposed to the sequence BTI$$\Rightarrow$$HCD$$\Rightarrow$$Relaxation, is quite different, as is show in the bottom panels of Fig. 5.23. Starting with BTI, the device experiences $$\sim \SI {140}{mV}$$ of drift. Due to the device being fresh, i.e. without any damage at the interface, this value is slightly higher than for the BTI sequence within the first group. The following $$\SI {10}{ks}$$ HCD stress reveals an interesting behaviour. Contrary to the assumption that another portion of damage will be added to the device, the chosen HCD condition strongly influences the discharging dynamics of oxide defects. The high concentration of secondary generated electrons along the channel, see Fig. 5.19, effectively accelerates the discharging dynamics of charged oxide traps which leads to an increased recovery of BTI damage. The competing effect of newly created damage due to HCD, however, masks this mechanism within the experiments and is only accessible by simulations. Note that the impact of HCD is slightly reduced due to charged oxide defects perturbing the device electrostatics. Within the final relaxation phase the simulations predict a total recovery of $$\SI {17}{mV}$$, whereas the experimental data show a flat line for $$\SI {10}{ks}$$. For the sake of completeness the effect of interface state recovery according to the mechanism proposed by Stesmans [221] has also been included within the final relaxation sequence. However, as apparently visible in the measurements for the second device group, this mechanism is absent within the recorded traces. Neglecting this effect would lead to an even better agreement between simulations and experiments for the last recovery phase.