Fig. 2.8 (a) (c) and Fig. 2.9 (b) (d) show multifrequency C-V curves of MOSCAPs with a thick (~11 nm) Al2O3 film on p- and n-type TMA treated (100)-oriented In0.53Ga0.47As with FGA or with PMA, respectively, measured at 300K.
For the p-type, both two samples never reach strong accumulation, which indicates a high Dit close to the valence band edge. The frequency dispersion of MOSCAPs with PMA is 12.34% larger than MOSCAPs with FGA, but much smaller than MOSCAPs without any post-metallization thermal annealing, shown in Table 2.1. A hump is seen at positive bias in both two samples, which may indicate the response of Dit. For the n-type MOSCAPs, accumulation is achieved at positive gate bias. The frequency dispersion of MOSCAPs with PMA is 8.08% larger than MOSCAPs with FGA, also shown in Table 2.1. At negative gate voltages the total capacitance changes with gate bias and frequency, which may also indicate the response of interface states.
Fig. 2.10 (a) (c) and Fig. 2.11 (b) (d) show conductance maps of Al2O3/TMA-treated In0.53Ga0.47As MOSCAPs with FGA or with PMA, on both p- and n-type (100)-oriented In0.53Ga0.47As, respectively, at temperature 300 K. These maps show the magnitude of normalized conductance (G/ω)/Aq as a function of ac frequency f and the gate voltage VG. For the p-type MOSCAPs, both two plots show that a distinct signal exists at the gate voltage between 0 and 1 volt and at frequency between 100 Hz and 10 kHz. We speculate that this energy loss is caused by interface states due to its gate-voltage dependent and frequency-dependent conductance. For the n-type MOSCAPs, there is a conspicuous variation in color appearing at the negative gate voltage in both two graphs, which implies the dramatic energy loss.
16
Similarly, the G-V response to this energy loss is made by interface states owing to its gate-voltage dependent and frequency-dependent conductance. Additionally, the color variation at negative gate bias in Fig. 2.11 (b) is less dramatic than in Fig. 2.11 (d).
The frequency dispersion of the accumulation capacitance can be attributed to tunneling of carriers between the substrate and defect states in the ALD-Al2O3 dielectric. The term “border traps” is referred to near-interfacial oxide traps that can exchange charge with substrate or gate, respectively [27]. While filling and emptying of interface states is a thermally activated process, tunneling of charges into border traps depends only on the measurement frequency and the distance of the defect states from the oxide/semiconductor interface. Some related studies have reported that the dispersion in the accumulation capacitance of such MOSCAPs is consistent with a tunneling mechanism for charge trapping, insensitive to temperature [28, 29]. Border traps that are close to conduction band edge of semiconductor contribute most effectively to the dispersion in accumulation, for n-type channel. Fig. 2.12 shows schematically the tunneling process between border traps and conduction band in an n-doped MOSCAP operated in accumulation. Our results are generally consistent with previous reports of hydrogen passivation of Al2O3/In0.53Ga0.47As, mainly passivating border traps in the Al2O3 layer [30]. Since interface states nearly respond immediately in deep accumulation, the decreased dispersion in accumulation observed in Fig. 2.8 (a) and Fig. 2.9 (c) may result exclusively from passivation of border traps.
Nonetheless, the response of interface states observed at VG=0~ -2 volts still exists but less dramatic after FGA, shown in Fig. 2.11 (b), and also the evidence of interface states observed at VG=0~ 1 volt for p-type MOSCAPs, shown in Fig. 2.10 (a), does not disappear after FGA. Therefore, we suppose that hydrogen is unable to completely reduce all interface states. Their binding energies are 2.52 eV for In-H, < 2.84 eV for Ga-H, and 2.84 eV for As-H [31]. Other researches consider that annealing in
17
hydrogen can remove most of the fixed oxide charge and interface states [32-34]. The presence of positive bulk fixed charge and negative interfacial fixed charge is identified as oxygen and aluminum dangling bonds (DBs), respectively. Fig. 2.13 shows the band alignment between Al2O3 and relevant semiconductors, and position of charge-state transition levels for dangling bonds in the oxide from the reference [32]. After FGA, O DBs and Al DBs are neutralized by hydrogen, which the resulting binding energies are 1.3 eV for O-H and 1.4 eV for Al-H, respectively. However, some groups consider that the previously reports of FGA cannot be solely attributed to hydrogen passivation [35]. Hydrogen passivation of dangling bonds and border traps is responsible for improving the interfacial properties, while the thermal budget is responsible for minimizing the fixed charge. According to their opinions on the thermal effects of annealing, the presence of sufficient thermal energy can reconstruct the bonds to fill vacancies and to passivate dangling bond throughout the oxide.
Another possibility is that PMA may provide enough thermal energy to break the O-H hydroxyl groups existing as a byproduct during the deposition of Al2O3. Hence, free hydrogen dissociated from these O-H groups can also passivate dangling bonds, but the amount of hydrogen available would be limited by the presence of O-H bonds in the film.
In summary, FGA was found to suppress frequency dispersion in accumulation, which suggests that hydrogen could be a promising candidate as the method for passivation of border traps. In addition, it can also reduce some, but not all, interface states. Hydrogen annealing of these devices at 300 oC for 30 min has demonstrated better C-V and G-V characteristics, and we’ll further discuss the impact of PDA on the electrical characteristics of these MOSCAPs in the next section.
18