Charge Pumping Measurements .1 Fixed Amplitude Sweep

From the above-mentioned discussions, we can conclude that post-deposition N2O plasma treatment can considerably enhance the electrical characteristics of pMOSFETs with HfO2/SiON gate stacks in terms of gate leakage current, subthreshold swing, normalized transconductance, and driving current, despite the slight increase in CET.

2.3.2 Charge Pumping Measurements 2.3.2.1 Fixed Amplitude Sweep

To further highlight the origin of the benefits given by the post-deposition treatment, charge pumping measurements were performed. It is well known that the charge pumping measurement is used to quantify the interface state density by monitoring the substrate current.

The basic charge pumping measurement involves the measurement of the substrate current while a series of voltage pulses with fixed amplitude, rise time, fall time, frequency, and duty cycle is being applied to the gate of the transistor (Fig 2-7), with source, drain and substrate connected to ground. However, during the measurement, the substrate current could be affected by the gate leakage current [24, 25], especially at lower frequencies and thinner oxides. Therefore, to accurately analyze interface state densities or bulk traps in the dielectrics from charge pumping measurement results, we need to pay close attention to the leakage current issue. In this work, we employed three conventional types of the voltage pulse train applying to the gate electrode, as depicted in Fig. 2-8, i.e., (a) fixed amplitude sweep, (b) fixed base sweep, and (c) fixed peak sweep [26]. The measured substrate current is actually a recombination current as gate voltage pulses swing the device between inversion and accumulation back and forth. As the transistor goes into inversion, holes from the source and drain fill interface traps. As the transistor goes back to the accumulation, untrapped holes go back to source and drain, while trapped holes recombine with the majority carrier, electrons, in the substrate and can be measured as substrate current, Icp. The results of the devices without post-N2O plasma treatment in this work with three different kinds of pulse trains at various frequencies are shown in Fig. 2-9. The mean interface trap density within an energy range (D_it ) is calculated by:

where Icp and Isd are the measured substrate current and source/drain current, respectively.

A is the area of the gate electrode; f is the frequency of pulses train applied to the gate. q is the electron charge, ∆E is the difference between the inversion Fermi level and the accumulation Fermi level. Fig.2-9(a) shows that the Ncp traces identically with Nsd over the entire sweep

voltage range and both of Ncp and Nsd have negligible frequency dependence, especially at higher frequencies. This is thought to be coming from the poor control over the charge exchange of bulk traps in the HfO2 dielectrics, due to the chosen magnitude of amplitude [26].

Thus, the fixed amplitude measurement with higher frequency will be a suitable approach for quantifying interface state densities.

2.3.2.2 Fixed Base Sweep and Fixed Peak Sweep

However, it is quite different for the other two measurements, i.e., fixed base and fixed peak sweeps. Recently, the charge pumping measurement had been frequently employed to qualify the level of bulk traps in the HfO2 dielectrics using these two methods [22, 26, 29]. For SiO2 gate oxide (close to ideal case), Icp measured from fixed amplitude sweep is almost identical to that from fixed base or fixed peak sweep, and it equals the Isd, implying that the charge pumping current is completely contributed by the recombination current of the interface state densities. However, for the dielectrics with inferior quality, it is commonly seen that the measured Icp increases with increasing amplitude of the gate voltage pulse and decreasing frequency of gate pulses when using the fixed base or fixed peak sweep. This may arise from carriers trapping and detrapping of bulk traps in the high-k films [22, 26]. In order to avoid the influence of geometric effect [30], in the charge pumping measurement, small devices (<10um) were normally used [27]. The devices we used here have channel length of 3um. This means that the geometric effect can be excluded. Therefore, when the measured Icp and Isd currents are not identical, there must be some extra components contributing to the Icp if the presence of bulk traps is not the culprit. Nevertheless, the results for the fixed base sweep shown in Fig. 2-9 (b) show that a remarkable deviation exists between Ncp and Nsd at larger positive Vgh with lower frequencies, which obviously can not be accounted for by the presence of bulk traps.

Because if this is caused by the recombination from the charging/discharging of the bulk traps in the high-k dielectrics, the increment of Icp and Isd should be the same due to the requirement

of equal amount of carriers of opposite polarities for the recombination. So, we believe the rapid increase in difference between Ncp and Nsd with increasing amplitude is ascribed to the larger leakage current. In addition, when the entire charge pumping gate voltage waveform lies at lower (Vgh<-0.5V) region, i.e., the transistor is operating in inversion mode; the difference of Icp between Isd is shown in Fig. 2-10(a). Here, in this region we can see that both of Isd and Icp

are independent of frequency. The leakage component is contributing to Isd and makes Isd two orders larger than Icp, but it really has little impact on the Nit calculation of fixed base sweep (Fig. 2-9 (b)). As the Vgh increases gradually, the transistor spends half time operating in accumulation, and the other leakage component, i.e., electrons from substrate, becomes dominant with increasing Vgh and decreasing frequency.

To further confirm our speculation, the fixed peak sweep (Fig. 2-8(c)) was also performed, with results shown in Fig. 2-9 (c). We can see that the deviation can be observed at 100 kHz and becomes worse when |Vgl| increases and frequency decreases. Again, we have to determine the leakage current component. By examining Fig. 2-10 (b), in the region of Vgl>-0.5V, i.e., when the transistor is operating in accumulation mode, Icp is slightly larger than Isd. As gate pulses sweep to the right (Fig. 2-8 (c)), the transistor spends half time operating in the inversion mode, and half time operating in the accumulation mode. Since the accumulation leakage is larger than the inversion leakage, we can observe in Fig. 2-10 (b), the gate leakage contributes to Icp during the entire period due to the fact that the peak voltage is fixed at 2V.

Therefore, irrespective of whether the fixed peak or fixed base is used, Icp is larger than Isd, especially at lower frequencies and higher Vgh or lower Vgl. A plausible explanation for this phenomenon is that the contribution of the considerably large gate leakage disturbs Icp current at higher positive gate voltages, resulting in the tunneling of electrons from the substrate, as illustrated in Fig. 2-11 [26]. This suggests that in order to precisely analyze the bulk traps in the dielectrics using the fixed base or fixed peak sweep from the charge pumping measurement, we need to monitor the substrate current and source/drain current simultaneously for clarifying the

influence of leakage current. This is especially true at lower frequencies, or when the leakage current is considerably larger. Similar trends also can be observed for the samples with N2O plasma treatment, although the data are not shown here. In should be noted that in our case, the leakage current is found to contribute more significantly to Icp, compared to Isd in pMOSFETs.

This is contrary to other group’s report that the leakage current would contribute to Isd in nMOSFETs [21]. As a result, we chose the fixed base sweep at a relatively low frequency of 5 kHz to examine the behavior of the bulk traps in HfO2 films.

Following the above argument, interface state densities were measured using the fixed amplitude sweep for the pMOSFETs with and without post-deposition N2O plasma treatment as function of the peak voltage. The results are shown in Fig. 2-12. It can be seen that the trend is quite consistent with that in subthreshold swing, as shown in Fig. 2-5. This may be due to the fact that oxygen radicals from the plasma can react with Si substrate, causing a slight higher CET, and thus improving the quality of the interface [31]. Moreover, Fig. 2-13 highlights Ncp and Nsd as determined from the fixed base sweep at a frequency of 5 kHz for devices with and without post-deposition N2O plasma treatment. It indicates that the N2O plasma treatment can not only improve the interface quality but also reduce bulk traps in the HfO2 gate stack effectively.