The as-deposited high frequency (1MHz) C-V characteristics of capacitors gated by Hf-Mo binary alloys, pure Mo, and pure Hf films are shown in Fig. 2.1. The negative flat-band voltage shift with the increase of Hf power ratio can be observed.
To eliminate the contribution of oxide fixed charges, C-V measurements of MOSCAP devices with several oxide thicknesses were performed to generate the VFB versus EOT plot as shown in Fig. 2.2. All samples exhibit linear relationships from which work function values of binary alloys can be extracted as listed in Table 2.1.
Figure 2.3 exhibits the (110) morphology for the pure Mo film. The extracted work function value (4.93eV) of the (110) oriented Mo is closely consistent with previous reports [13]. However, the as-deposited Mo with (110) orientation is different from the previous report [15], and this would be attributed to the different deposition conditions. Figure 2.4 exhibits the small the hysteresis for the 50% Hf power ratio co-sputtering sample, which is believed to suffer from the most series sputtering damage. Accordingly, the good process quality can be demonstrated.
The C-V curves of as-deposited and post-400 ℃ sintering pure Hf gated capacitor are shown in Fig. 2.5. Obvious EOT variation along with the VFB shift after annealing illustrates the poor thermal stability of Hf on SiO2 and exclude Hf from gate candidates even in the gate last process [16]. By contrast, HfxMo(1-x) gated capacitors exhibit better thermal stability on SiO2 as shown in Fig. 2.6. The dependence of Φm
and EOT variation on annealing conditions for HfxMo(1-x) also indicates that the thermal stability of all alloy samples can be at least higher than 400℃ as shown in Fig. 2.7.
In Fig. 2.7, the decreases of EOT at certain temperature for alloy samples, except for the one with 25% Hf power ratio, are similar to that for the pure Hf sample. The noticeable EOT decrease and the corresponding work function variation might be attributed to that partial SiO2 gate dielectric was transformed into high-k materials, such as HfO2 or HfSixOy, due to the Hf-SiO2 interaction. For the 25% alloy sample, the abnormal EOT increase along with relative small work function variation can be observed. We speculate that the lower Hf concentration might make the effect of Hf-SiO2 interaction be masked by the extra Si-substrate oxidation due to the oxygen contamination. In the case of sample gated by pure Mo, the small amount of Φm
increase (18meV) and the negligible EOT variation (0.08nm) after 950℃ RTA demonstrate the superior thermal stability of Mo on SiO2 gate dielectric. Although the thermal stability seems to be degraded with the increasing of the Hf atomic fraction in the Hf-Mo binary alloy, HfxMo(1-x) still can be adopted as gate material in a gate-last SiO2 CMOS process. It is worth to note that, alloy samples with Hf power ratio lower than 50% can possess work function value suitable for advanced devices and exhibit thermal stability up to 700℃. In comparison with other reported candidates, the thermal stability of Hf-Mo alloy is lower than that for Ru-Ta alloy [9], but higher than that for Pt-Ta alloy [11].
In 1974, Gelatt and Ehrenreich proposed that the work function of an AxB(1-x)
alloy can be approximately expressed as [17] :
( ) ( ) ( ) ( ) ( )
respectively. ρA and ρB are effective density of states in Fermi level for pure element A and B, respectively. In this equation, the first two terms represent that the work function of the binary alloy is a linear combination of that of each pure element. On the other hand, the last term will lead to a deviation from the linear relationship.According to the theory of heat capacity of metal, the observable Sommerfeld factor γ of a metal is directly proportional to its density of state in Fermi level ρ [18].
The calculated results of eq. (1) are shown in fig. 2.8 where several γ ratios are used and values of Φm,A and Φm,B are set to be 3.93 and 4.93, respectively, for convenience. As expected, the work function modulation will deviate from the linear behavior with the difference in γ values between two metals. For a non-linear behavior, work function modulation can be roughly divided into the flat and sharp regime. In the flatter regime, the alloy system would be less susceptible to the composition and process variation, but the Φm modulation efficiency will be lower.
On the other hand, the alloy system will be more sensitive to the process variation but possess higher Φm modulation efficiency in the sharper regime. By contrast, the linear work function modulation can provide a compromise between the modulation efficiency and immunity to the process variation throughout the whole modulation range.
According to the binary alloy phase diagram of Hf-Mo system [19] shown in fig.
2.9, an abrupt work function modulation can be excluded since no specific compound will be formed under 1000℃. Moreover, the Sommerfeld factors for Hf and Mo are
2.16 and 2.0, respectively [18]. Therefore, a continuous and almost linear work function modulation using the HfxMo(1-x) solid solution can be expected. The calculated and experimental results of work functions of HfxMo(1-x) alloys are shown in fig. 2.10. Compared with the experimental data, a good consistency with only a mildly shift in the lower Hf power ratio regime can be observed. This deviation may be attributed to the difference between the Hf power ratio and the Hf atomic fraction due to a relatively lower deposition rate of Mo in this work.
The XRD spectra and AES depth profile of the co-sputtering sample are shown in fig. 2.11 and fig. 2.12, respectively. An amorphous film with uniform composition and abrupt interface is observed. It is worth to note that the relatively lower composition of Mo compared to that of Hf in the co-sputtering sample as shown in fig.
2.12 also demonstrates our speculation about the deviation observed in Fig. 2.10.