5.1 Pulsed -Characteristics

Based on the large discrepancy between the initial reference and the very first measurement point visible in Fig. 5.1 and Fig. 5.2, different ways to extract a reference of are compared in Fig. 5.3 and Fig. 5.4. A DC-characteristic and a slow -pulse sweep are both compared to the fast -pulse sweep which is used for the fast pulsed -characteristics, cf. Fig. 5.7.

Figure 5.3: Three different transfer characteristics. A DC-curve originating from a DSO with averaging acts as reference to -characteristics obtained by two gate pulses with and . Due to the limited resolution, especially the subthreshold region of the is affected by quantization noise.

While the DC-curve is averaged and hence very smooth, the slow and fast pulses lack accuracy due to the missing averaging, as can be seen best in the subthreshold regime, which is very noisy. As depicted in Fig. 5.4, setting the threshold current criteria to (linear drain current regime), yields extracted values of differing in around . This error is indicated as in Fig. 5.4.

Figure 5.4: A close observation of the linear regime reveals different values of extracted . They differ by from each other, as marked by which is on the order of the obtained degradation for of PBTI stress, cf. Fig. 5.6.

The impact of the various transfer characteristics used to get an initial undegraded reference to eventually measure during stress is depicted for NBTI and PBTI stress in Fig. 5.5 and Fig. 5.6. Here the fast pulsed -characterization using triangular - and -pulses with zero pulse high-time after Li et al. (cf. Chapter 2.2.1) was applied to pMOS-devices with an provided by IMEC¹ .

Figure 5.5: Fast pulsed -measurements (FPM) performed on pMOS devices provided by IMEC after the method of Liu et al. Before FPM is applied using different pulse widths for NBTI/PBTI stress, is determined in three different ways, cf. Fig. 5.3 and Fig. 5.4. Both stress and recovery are interrupted times within three decades ranging from to for an FPM. Unfortunately, a high level of uncertainty is obtained by extracting the threshold voltage manually. Applying NBTI stress yields sound results because of the higher signal-to-noise ratio and the expected negative shift of .

For PBTI stress Fig. 5.6 the determination of delivers values which are of the same order of magnitude as the following degradation itself, cf. first and second subfigure. Depending on the chosen -reference the determined degradation hence varies by a factor of two. The same holds for the relaxation mode (third subfigure) and its -references taken at , the DC-characteristics and slow -pulse (forth subfigure). Even more important is the fact that in contradiction to [24], the PBTI results do not exhibit a positive -shift at all, they solely show negative -shifts.

Figure 5.6: Fast pulsed -measurements (FPM) performed on pMOS devices provided by IMEC after the method of Liu et al. Before FPM is applied using different pulse widths for NBTI/PBTI stress, is determined in three different ways, cf. Fig. 5.3 and Fig. 5.4. Both stress and recovery are interrupted times within three decades ranging from to for an FPM. Unfortunately, a high level of uncertainty is obtained by extracting the threshold voltage manually. When performing PBTI stress again a negative shift of is found. This qualitatively supports the results of Grasser et al. presented in [30]. The suddenly appearing offset of in-between the last two readout points during stress was assumed to be due to heavy oxide damage.

When the overall degradation becomes larger, as it is the case during NBTI stress (Fig. 5.5), the error induced by the reference decreases as expected. Unfortunately, the reason of the poor agreement of the differently extracted initial and post -values remains unclear. As these references do not indicate any systematic error, but seem to vary randomly, a different approach which is able to explain the deviating measurement results is needed.