Chapter 5
Pulsed BTI Measurements

In the previous chapter various BTI stress tests were performed using the measurement-stress-measurement (MSM) and the on-the-fly (OTF) technique. Special attention was given to the fitting of the measurement data onto a universal relaxation law, yielding a separation of the degradation into a recoverable and a poorly recoverable or permanent component. Data gathered at different temperatures and stress voltages were found to follow a universal relaxation law. Interestingly both stress polarities, i.e. NBTI and PBTI stress on a pMOS, always resulted in a negative shift of the threshold voltage. Unfortunately, PBTI had been rarely discussed in literature until Liu et al. monitored a positive shift of the threshold voltage due to PBTI-stressed pMOS-devices [24], which contradicts the results presented by Grasser et al. [30].


PIC


Figure 5.1: pMOSFET monitored under 1000s  of NBTI stress followed by 1000s  of relaxation. While ΔVTH   is measured by the FPM (open squares), ΔIcp   is measured by OFIT (solid circles). The fast pulsed ID(VG )  -characteristics reveal a negative shift of VTH   for NBTI, while during PBTI a positive shift is visible (Fig. 5.2). At the end of the recovery phase the ΔI
   cp   curve is scaled to match the value of ΔV
   TH   . According to Liu et al. the difference between two curves (shown in the inset) yields the amount of contributing oxide traps. Data is taken from [24].



PIC


Figure 5.2: pMOSFET monitored under 1000s  of PBTI stress followed by 1000s  of relaxation. While ΔVTH   is measured by the FPM (open squares), ΔIcp   is measured by OFIT (solid circles). In contrast to Fig. 5.1 the fast pulsed ID(VG)  -characteristics reveal a positive shift of VTH   for PBTI. At the end of the recovery phase the ΔIcp   curve is scaled to match the value of ΔV
  TH   . According to Liu et al. the difference between two curves (shown in the inset) yields the amount of contributing oxide traps. Data is taken from [24].


One reason of this discrepancy might be the fact that Grasser et al. used the OTF and the eMSM technique (cf. Chapter 2.3 and 2.1.3), while the two measurement techniques used in [24] are both based on the application of fast gate pulses: The newly developed on-the-fly fast charge pumping (OFIT) technique and the fast pulsed I (V )
 D  G  -characteristics have been discussed in Chapter 2.5 and 2.2.1. The measurement results obtained by those two pulsed setups are only at a first glance interpreted in a correct way, as the ΔIcp   -curve obtained by OFIT is simply scaled to align the ΔVTH   -curve at the end of the recovery phase in [24]. Based on this alignment scheme depicted in Fig. 5.1 and Fig. 5.2, Liu et al. stated a fast oxide trap component (Not   ) corresponding to the difference of ΔVTH (ID(VG))−  ΔIcp(OFIT )  , which is shown in the insets of Fig. 5.1 and Fig. 5.2. Compared to that, the interface states are considered to recover only slowly. It was furthermore concluded that the fast oxide traps are responsible for the predominant part of VTH   -degradation in the fast pulsed ID(VG )  -characteristics only, since their influence during a DC measurement is drastically reduced due to the measurement delay. Consequently, this makes the interface states dominate the DC regime.

When taking a closer look at the pulsed ID (VG)  -characteristics of Fig. 5.1 and Fig. 5.2, a surprisingly huge offset of about 100mV  between the reference value and the first measurement point after 1s  of stress can be detected. As this already accounts for more than 75%  of the total degradation built up after 1000s  of stress, the high initial ΔVTH   seems to be at least questionable.

In order to determine to what extent interface states and oxide charges really contribute to the measurement signal, a more detailed study of the fast pulsed ID(VG )  and the OFIT technique, besides further measurements is needed. Especially the measurement delay of the setup in combination with its accuracy is of particular interest here.

 5.1 Pulsed ID(VG)  -Characteristics
 5.2 Further Data Extraction Options
  5.2.1 Determination of the Fitting Region
  5.2.2 Impact of the Pulse Amplitude
  5.2.3 Varying Pulse Rise/Fall Times
  5.2.4 Consequences
 5.3 Experimental Identification of Defects
 5.4 OFIT versus CP
 5.5 Analysis of the OFIT Technique
  5.5.1 Dependence on Gate Voltage Low-Level
  5.5.2 Hysteresis due to Stress
 5.6 Extrapolation of Oxide Trap Contribution
 5.7 Simulation of the Charge Pumping Current
 5.8 Results
 5.9 Conclusion