UV nanosecond laser anneal (UV-NLA) is demonstrating clear benefits in emerging 3D-integration CMOS processes,1,2 where the allowed thermal budget is strictly limited to preserve underlying device performance. One of the major advantages of UV-NLA comes from its non-equilibrium feature related to melting of materials and subsequent rapid solidification. Especially, crystallization from amorphous Si1,3-4 and dopant activation1,4-5 are widely reported. In a real CMOS process, multilayered structures can be found everywhere. A typical one is a semiconductor material encapsuled by a dielectric thin film as found in the gate of transistor (i.e., Si gate with SiO2 hard mask). Melting a part or whole of the underlying Si may induce wrinkles on the SiO2 surface,6 possibly being an issue for following processes steps. It is, therefore, important to understand the emergence of such wrinkles and related key parameters.In this study, SiO2 thin films of different thicknesses (about 5 to 120 nm) were thermally grown on a standard Si(100) substrate (Fig. 1(a)). The SiO2 thickness was measured by spectroscopic ellipsometry. A 308 nm wavelength UV-NLA was performed on this stack in N2 ambient without stage pre-heating to melt the underlying Si. The pulse duration was set at 160 ns. Emerged wrinkles were characterized by different methods to extract evolution of their specific spatial wavelength (λ) and amplitude (Fig. 1(b)). In addition, we simulated the UV-NLA process based on a self-consistent time-harmonic solution of the Maxwell equations7 in order to roughly estimate an effective Si melting time (tmelt ) and molten depth (hSi ) (Fig. 1(c)). We used an analytical model reported in literature6 to fit the experimental data. The model consists of three regimes of wrinkle formation: (I) initial growth, (II) coarsening, and (iii) elastic equilibrium. It should be noted that our process timescale allows only the coarsening regime. In the equation to calculate λ, the SiO2 thickness (hSiO2 ), the Poisson’s ratio of Si (υ Si ) and SiO2 (υ SiO2 ), the shear modulus of SiO2 (μ SiO2 ), and the viscosity of Si (η Si ) are also included. Fitting the experimental λ determined by SEM for all UV-NLA conditions and all SiO2 thicknesses was performed while introducing a fitting coefficient (a) (Fig. 1(d)). The experimental data showed a monotonic enlargement of λ when increasing the applied laser energy density (i.e., Si molten depth). In addition, a thicker SiO2 film resulted in a larger scale of λ. It should be noted that the presented different SiO2 thicknesses theoretically (i.e., from the Fresnel equation) lead to different reflectance values (R) to the 308 nm light for each SiO2/Si system (i.e., R ~ 0.52, 0.49, and 0, 55 for the 5.7 nm, 70.2 nm, and 119.8 nm-thick films respectively), but it cannot fully explain the observed thickness effect on λ. This tendency was well reproduced by the applied model, adjusting the a value for each SiO2 thickness. Interestingly, a was rapidly increased for a thinner SiO2 film (< 40 nm), while it became closer to the unity for a thicker SiO2 film (> 40 nm) (Fig. 1(e)). This might infer a possible discrepancy of the material properties taken from literature and those in the real thin SiO2 films. Also, in our calculation, tmelt was taken as the full width at half maximum of each Si molten depth evolution profile obtained by simulation, and hSi was defined as the maximum Si molten depth. Although this way of defining tmelt and hSi should have an impact on the a value, it is limited to about 18 to 24% at most and as result cannot fully explain the aforementioned discrepancy.In summary, emergence of wrinkles in thermally-grown SiO2/Si stacks during UV-NLA was systematically investigated for different SiO2 film thicknesses. The experimental results showed a good agreement with a theoretical model by introducing a fitting coefficient which reflects a potentiel dispersion of the actual material properties from those taken from literature. Although dedicated thin film property measurements seem necessary, it has been found that the presented classical model is applicable to the non-equilibrium UV-NLA process.1. L. Brunet et al., IEEE IEDM 2018, p. 153.2. C. Cavalcante et al., IEEE VLSI 2020, TH3.3.3. P.-Y. Hsieh et al., IEEE IEDM 2019, p. 46.4. S. Kerdilès et al, IEEE IWJT 2016.5. A. Vandooren et al, IEEE VLSI 2020, TH3.2.6. R. Huang et al., Phys. Rev. E 74, 026214 (2006).7. S.F. Lombardo et al., Appl. Surf. Sci., 467-468, p. 666 (2019). Figure 1
Search from 250 M+ research papersSearch from 250 M+ research papers
Open Access
Oct 7, 2021
Guangbao Shan + 5
