4.1 Introduction
As the previous study in chapter 3 [1], we have attempted to form CoSi2 through silicidation by the OME method but in a different way in that Co was deposited by conventional sputtering and annealing was performed in lower vacuum environment, which was called oxide-mediated silicidation, and is cost-effective and more compatible with current ULSI technology than the OME method. We have shown that not epitaxial CoSi2 but nano-grained (average grain size 5 nm) polycrystalline CoSi2
thin film with homogeneous grain size distribution (5±1.8 nm) can be obtained by properly controlling the annealing process in this method. In addition, this method also retained the advantages of the OME method in that CoSi2 directly formed as the first phase bypassing the CoSi and Co2Si in lower temperature (500-700 °C) annealing with respect to the conventional process, which required > 750 °C annealing temperature to ensure complete replacement of CoSi and Co2Si with CoSi2. The resulted CoSi2 layer from this method exhibited smooth surface and dense bulk
CoSi2 thin film. All the above advantages are significant for ULSI technology because this process can reduce thermal budget, resistance and leakage current. So far, although it was believed that SiOx acts as the diffusion barrier layer to reduce the concentration of Co that diffused to Si leading to CoSi2 as the first phase, there is no direct evidence to prove it and less systematical study was carried out on the diffusion and nucleation mechanism of oxide-mediated CoSi2 formation. In this chapter, we attempt to find out the formation mechanism of oxide-mediated CoSi2 thin films.
4.2 Experimental Details
P-type (8-12 ohm-cm) silicon substrates without or with Si doping by Si implantation at 30 keV with a dose of 1x1013 atoms/cm2 were chemically cleaned and dipped in a boiling HCl:H2O2:H2O=3:1:1 solution for 3 min to form a SiOx layer (Shiraki Oxide) prior to Co deposition by DC magnetron sputtering. The Co target (99.95% purity) was pre-sputtered for 10 minutes after the base pressure of 3x10-6 torr was reached using argon (99.995% purity) as the sputtering gas. Subsequently, TiN of about 10 nm was deposited as the cap layer before exposing the sample to air. Ex-situ annealing was carried out in a vacuum chamber of 10-5 torr. Upon annealing, all layers except the reactive products were stripped off by chemical etching in order to examine the silicide layer on Si in plan-view. The TiN, unreacted Co and SiOx layers can be
stripped off by NH4OH:H2O2: H2O=1:1:4 solution at 50 °C, aluminum etching solution (H3PO4 71 wt%, HNO3 2.5 wt%, CH2COOH 12.5 wt%, others H2O) at 75 °C and HF solution, respectively. The phase of the samples was then examined by transmission electron microscope (TEM). The element distribution of the multilayered thin films was examined by energy-filtered electroscopic imaging (ESI). The phase of reactive products on Si was also examined through binding energy in X-ray Photoelectron Spectroscopy (XPS).
4.3 Results and Discussion
Fig. 4-1 (a) shows a cross-sectional TEM image from the sample of the unimplanted Si substrate after 600 °C 90 sec annealing with the ESI elemental maps of (b) Co and (c) Si. The thickness of the Co and SiOx layers are about 4 and 2 nm, respectively. The blobs of dark contrast on Si side at the Si/SiOx interface as indicated in Fig. 4-1 (a) are CoSi2, which will be shown in fig. 4-2. Fig. 4-1 (b) exhibits that while Co diffuses into the Si substrate after annealing, Si diffuses less out due to inhibition from the SiOx layer as shown in Fig. 4-1 (c). This is a direct evidence in that the SiOx acts as a one-way diffusion barrier to inhibit Si and Co interdiffusion from forming CoSi and Co2Si. According to Vantomme [2] and Pretorius [3], if the Co effective concentration at the cobalt silicide growth interface is low enough, the
biggest negative change in the free energy would be resulted for the CoSi2 formation.
In addition, direct CoSi2 formation can effectively reduce the formation temperature because the reaction path is Co + 2 Si Æ CoSi2 rather than CoSi + Si Æ CoSi2, in other words, it needs not to break CoSi bonding [4]. In the oxide-mediated silicidation process, SiOx not only acts as a diffusion barrier to reduce the concentration of Co diffusing into Si but also inhibits Si from out-diffusion. Because CoSi2 forms directly bypassing CoSi and Co2Si, the formation temperature is effectively reduced to 600 °C.
Vantomme et al. [2] even reported that CoSi2 formation temperature can be reduced to 360 °C.
Fig. 4-2 is a bright-field plan-view TEM image and the corresponding diffraction pattern of the reactive silicide upon annealing on the unimplanted Si sample with all the other layers removed, where the annealing conditions is the same as in Fig. 4-1.
The diffraction pattern shows that the characteristic rings correspond to polycrystalline CoSi2. Although Co was deposited by sputtering, the resulting phase agrees well with the study of Tung et al. Fig. 4-3 is the corresponding high resolution TEM cross sectional images as in Fig. 4-2. The dark contrast in the Si adjacent to the SiOx has been assigned to CoSi2 and this layer has a smooth interface with the SiOx, confirming that no Si out-diffusion and other chemical reactions happen there. The smooth interface of the CoSi2 thin film can reduce the junction leakage current of
source and drain in Metal-Oxide-Semiconductor transistors. Si atoms still stay at the lattice sites until reacted with Co, which states that Co is a moving species and Si is a passive species. Besides, a dense CoSi2 layer with no voids between Si and silicide also resulted from no Si out-diffusion. If the CoSi2 layer was porous, high resistivity and leakage current would result. These issues are serious problems often encountered in the conventional silicide process, especially when the feature size deceases, which can be avoided by this new method.
But what is the diffusion path of Co in the SiOx layer to Si? Baten and Fedorovich [5,6] have found that cobalt diffuses through SiO2 without any chemical interaction with the SiO2 networks but only occupies interstices of the very open SiO2
structure and migrates along the interstices as diffusion channels without affecting the regular lattices. The SiOx microstructure is more porous than SiO2, thus more interstices channels exist in the SiOx. Detavernier et. al., [7] also found the CoSi2
formation was taken place underneath the weak points of SiO2. Fig. 4-4 is a cross-sectional TEM image of the same unimplanted Si sample annealed at 600 °C for 60 sec. In Fig. 4-4, the microstructure of the SiOx layer is apparently still intact after annealing, which supports the previous results in that Co diffuses through the interstices as diffusion channels without affecting the regular lattices. The CoSi2
grains formed discontinuously underneath the SiOx layer, which implies that the SiOx
networks in these areas are more porous leading to the faster Co diffusion rate here.
These areas are apparently “weak points” of the SiOx layer. The corresponding bright-field plan-view TEM image in Fig. 4-5 shows discontinuous grain distribution, which is also characterized by Co diffusing only through the weak points of the SiOx
layer leading to island morphology rather than a continuous film. Besides, from the Co ESI map in Fig. 4-1 (b), the Co signal over the SiOx layer is not uniform but more concentrates on some diffusion channels, which provide further evidence. Hence a uniform SiOx layer is crucial for a uniform CoSi2 thin film formation. In order to visualize the effect of oxide thickness on Co diffusion, Fig. 4-6 shows a bright-field cross-sectional TEM image of the sample with a thinner SiOx layer of about 0.8 nm annealed at a lower temperature of 460 °C for 120 sec. Even annealed at a lower temperature, the island morphology was enhanced and shown to be bigger (~10 nm) compared to that in Fig. 4-4 (~2 nm). It was estimated that the time required for the same diffusion distance at 460 °C is about 80 times than 600 °C [1]. This justifies that the Co diffusion is non-uniform diffusion only through the weak points. Therefore, from Fig. 4-6, Co diffuses through the SiOx layer and forms a wavier cobalt silicide, more apparently indicating that Co easier diffuses through the weak points in the SiOx
layer. The thicker CoSi2 thin film in Fig. 4-6 also implies that Co can more easily diffuse through a thinner SiOx layer.
Fig. 4-7 shows XPS results from two samples with and without Si implantation in the Si substrate with unreacted TiN, Co and SiOx removed, where the annealing condition was 460 °C 60 sec followed by 600 °C 60 sec annealing. The two-step annealing applied here is intended for growing thicker and continuous CoSi2 in order to obtain enough signals for XPS. Given that the annealing conditions are the same for both samples, the qualitative interpretation in the following should be still valid.
The cobalt binding energy in CoSi2 and Co is about 778.7 and 778 eV, respectively [8,9]. Therefore, they tend to overlap as shown in Fig. 4-7. The sample with the substrate implanted with Si should provide a more negative free energy change for CoSi2 formation due to the amorphorized Si-rich surface [10]. Therefore, the peak from the sample with Si implantation becomes sharper and shifts to 778.7 eV for more CoSi2 formation compared to 778 eV from the sample without Si implantation.
If the peaks were de-convoluted by Gaussian distribution fitting, it is more clearly that except CoSi2, excess Co of about 79% was found to be coexisted in the sample without Si implantation (Fig. 4-7 (c)), which can be largely reduced to 38 % Co accumulation in the sample with Si implantation (Fig. 4-7 (b)). This suggests that CoSi2 nucleation rate is lower than Co diffusion rate resulting in Co accumulation at the interface. Vantomme et al. [2] mentioned that the effective Co concentration at the interface is determined by metal supply as well as silicide reaction rate. In other
words, reducing metal supply rate can lower the effective concentration but lower silicide reaction rate can raise the effective concentration. In order to maintain CoSi2
as the first formation phase, silicide reaction rate must be raised. This is the reason why the reactive deposition epitaxy and high temperature sputtering need to be performed at an elevated substrate temperature for higher silicide reaction rate [2,11].
This also explains why the Co thickness more than 3nm in the OME tends to form CoSi and Co2Si instead, because a thicker Co layer increases Co diffusion rate.
In addition, Detavernier et. al. [7] also found that at higher annealing temperature of 850 °C, high resistive CoSi phase forms by a lateral growth phenomenon where Si from substrate diffuses through CoSi2 and reacts with remaining Co to form CoSi following CoSi2 nucleation directly underneath the weak regions of SiO2. Lateral diffusion will enhance larger grain size, which is not favorable for the low resistivity of CoSi2. Lower annealing temperature can reduce Co and Si diffusion rate and then prevent lateral diffusion, which also means lower Co concentration at the interface.
Hence, the control over Co diffusion and CoSi2 reaction rate are predominant in the oxide-mediated cobalt silicidation, in order to maintain smaller grain size and CoSi2 as the only phase. The XPS results in Fig. 4-7 show that a Si-rich Si substrate can enhance CoSi2 reaction rate, while our previous study [1] also showed that two-step annealing can increase nucleation sites and reduce grain size.
4. 4 Conclusions
The diffusion and nucleation mechanisms of the oxide-mediated cobalt silicidation are studied in this chapter. Co diffuses through the weak points of the SiOx
layer and then reacts with Si to form cobalt disilicide. The SiOx layer acts as a one-way diffusion barrier to reduce the Co concentration and increase Si concentration in the cobalt silicide growth interface, which induces CoSi2 formation as the first phase in lower annealing temperature. The Co diffusion rate is higher than the CoSi2 nucleation rate at 600 °C, so that excess Co coexists with the reacted CoSi2. A Si-implanted Si substrate can increase the CoSi2 nucleation rate and reduce the residual Co accumulation. The understanding of diffusion and nucleation mechanisms of the oxide-mediated cobalt silicidation is essential for the formation of a high quality CoSi2 thin film.