6.7 Conclusions

Ultra-fast short-time NBTI stress and relaxation measurements from the μs  to the seconds regime using different temperatures, stress voltages, and oxide thicknesses have been performed. A large dataset is examined here using well defined extraction parameters (t0,ref   , tskip   , and ϵ  ). Amongst them the reference time t0,ref   is identified as the most crucial one. It can be seen that depending on the range used for the data extraction (ϵ  ) the reference time t0,ref   is also changed. While a settled gate pulse, i.e. a small ϵ  , does not contain the full degradation and relaxation data and may therefore indicate a wrong distribution of time constants, too broad limits of ϵ
 rel   may produce spurious relaxation transients due to a limited resolution of smaller than 1μs  . Comparing the different gate voltage criteria taken for the OTF routine yields that choosing a rather large ϵrel   reflects the completely different fast-VTH   measurement method best.

In the initial degradation phase, which is often explained by elastic hole trapping, the data can be well fit by a logarithmic time dependence [151242]. As this log-dependence is considerably distorted during long-term measurements, alternatively a power-law using an exponent considerably smaller (n ≈ 0.04  ) than generally observed during long-time stress (n ≈ 0.12  ) can be used. However, the main disadvantage of the power-law is that the fit is ill defined for up to medium stress conditions. Only high temperatures and/or high VG,str   show the aforementioned small n  .

Moreover, the extracted activation energy of about 0.1eV  is compatible with the values typically obtained during long-time stress [106]. The temperature and voltage dependencies of stress and relaxation rule out elastic and thus temperature-independent hole tunneling as being responsible for short-time NBTI degradation as proposed by [94104]. A possible explanation could involve an inelastic tunneling process [98].