Extrapolation of Oxide Trap Contribution

5.6 Extrapolation of Oxide Trap Contribution

As demonstrated above, during an OFIT measurement a distortion of $Icp$ due to oxide charges and due to the creation of defects during the low-level is monitored. In order to analyze this distortion, $ˆ VG,low$ is determined to be the lowest value of $VG,low$ at which no hysteresis is observed. The dataset $VG,low > VˆG,low$ is then used to extrapolate the impact of oxide charges $ΔNot$ down to the stress-level. It is not possible to obtain this information from the stress pulse because of the contribution of both parts $ΔNit$ and $ΔNot$ . Quite remarkably, the data [78] can be fit by a quadratic polynomial, consistent with our NBTI experiments where we also observe a quadratic ( $2 E ox$ ) dependence of the hole-trapping component [99, 18, 98]. The hole-trapping theory developed in [98] was applied to our data and excellent agreement was obtained. The difference between the actual signal ( $Icpit + Icpot$ ) and the extrapolated curve in Fig. 5.15 and Fig. 5.16 finally gives $ΔNit$ .

Figure 5.15: Charge pumping current $Icp$ for the stress pulse ( $Vstress$ = $− 17V$ ) and the relaxation pulse ( $Vrelax$ = $− 8V$ ), shown in Fig. 5.13. To decompose the contribution of oxide charges and additional interface states we look at the difference $Ircispe− Ifcapll$ . In the range $− 8V < VG,low < 0V$ , this difference is constant, implying no additional creation of interface states. From this ‘safe window’ we extrapolate to the minimum low-level to estimate the contribution due to oxide charges. Note that the first branches $fall Icp$ of the stress and relaxation pulse differ from each other due to pre-stress pulses between $VG,low = − 8V$ and $VG,low = − 17V$ . In fact, when using fresh devices for each measurement all $Ifcapll$ would coincide.

In Fig. 5.15 and Fig. 5.16 the extraction algorithm for $ΔNot$ and $ΔNit$ is demonstrated. Stress and relaxation pulse responses both consist of two branches, one falling and one rising, as marked by arrows. In the falling branch, $VG,low$ varies from $0V$ to $− 17V$ . In the rising branch, $VG,low$ varies from $− 17V$ to $0V$ . Only pulses with constant $Irise− Ifall cp cp$ (or even without a hysteresis, i.e. $rise fall Icp − Icp = 0$ ) can be used to create an extrapolation guess for higher $VG,low$ . This ‘safe window’ ranges from $0V$ to $− 8V$ , where both branches are indistinguishable.

Figure 5.16: Top: Lower temperatures simplify the extrapolation due to the absence of degradation. Here the full range of pulse amplitudes can be used to verify the extrapolation down to deep inversion. The missing hysteresis indicates the absence of additional oxide states in deep inversion at low temperatures. Bottom: Noise complicates this procedure at low frequencies. Data are scaled to $f = 125kHz ref$ .

The extracted components for different temperatures and frequencies are given in Fig. 5.17. The additionally created oxide traps $ΔNot$ depend on frequency as well as on temperature and clearly show $V2G,low ∼ E2ox$ behavior. The hysteresis due to additionally created traps $ΔNit$ is independent of frequency, but strongly dependent on temperature.

Figure 5.17: The extracted oxide state density (Top) and additional interface state density (Bottom). The change of oxide trap density $ΔNot$ follows $E2ox$ , and depends on the frequency as well as on the temperature. The hystereses displayed in the previous figures are due to additionally created traps, $ΔNit$ , which are independent of the frequency but strongly dependent on the temperature.

[next] [prev] [prev-tail] [front] [up]