5.5.2 Hysteresis due to Stress

When VG,low   is lowered towards the stress voltage, as required in the OFIT technique, Icp   extracted from the rising and falling pulse edges start to deviate, introducing a hysteresis. The hysteresis is only visible for larger pulse amplitudes, indicating degradation (marked with ΔNit   and ΔIcp   ) due to stress. While the impact of the oxide traps visible during medium V
 G,low   appears to be fully recoverable, the component causing the hysteresis is not. This can be seen in Fig. 5.13 and Fig. 5.14, where Icp   increases during subsequent measurements performed on the same device. We attribute this hysteresis to the creation of additional interface states due to NBTI stress at VG,low = Vstr   [78]. Starting at − 2V  there is nearly no stress. The deeper the device is stressed into inversion the larger the hysteresis becomes, resulting in an increased offset for the next pulse. The total hysteresis at a certain stress level hence not only consists of the hysteresis of the momentary charge pumping measurement but depends on the previous measurements3 .

As displayed in the inset in Fig. 5.13, the very first pulses are almost free of stress (no hysteresis, ΔIcp = 0  ) and hence the deviation of Icp   from Icp,0it   is entirely due to oxide traps. Only a negligible amount of interface states ΔNit   are created by the measurement process. The hysteresis-free area will be discussed in more detail in the next section.

When the experiment is repeated at a lower frequency (see bottom of Fig. 5.14), one finds that the interface state contribution can be scaled to the reference frequency (fref =  125kHz  ) [44]. This is compatible with the fact that the stress duration is practically independent of frequency. On the other hand, the recoverable oxide trap contribution to Icp   depends on frequency, consistent with the idea that the lower the frequency (corresponding to more time per pulse) the more oxide traps can contribute to Icp   [95].


PIC


Figure 5.14: Top: At low temperatures the hysteresis is negligible (less than 1%) and the contribution of slow oxide traps is reduced. Bottom: At low frequencies and high temperatures the contribution of oxide traps increases due to the increased rise and fall times. A comparable if not equal part of interface states constituting ΔIcp   can be identified for different frequencies but equal temperature when checking against the left figure. Following these results at least part of the defects must vary with temperature or frequency. For better comparability, the data at 12.5kHz  are scaled to the reference frequency (fref = 125kHz  ).


Finally, at a low temperature, displayed at the top of Fig. 5.14, practically no hysteresis is introduced (no NBTI stress) and also the oxide trap contribution is reduced, consistent with the idea that these traps are due to a thermally activated tunneling mechanism [98] rather than elastic (and thus temperature-independent) hole tunneling [94].