Computational Modeling of CMP Pads: Analyzing the Pad Geometry’s Impact on Planarization Efficiency

The pad choice has a dominant effect in polish behavior during a shallow trench isolation (STI) process. That pad must be chosen in connection with the pattern to be polished, as a pattern may consist of features that span various scales. An ideal pad will result in high planarization efficiencies across all feature sizes, in a low within-die range (strong uniformity across the die), and in low within-wafer ranges (strong uniformity across the wafer). Computational simulations make it possible to test various candidate pad designs, in search of an optimal pad design. We present our CMP modeling framework and use it to evaluate the planarization efficiency of a feature array, and the within-die range of a particular pattern, for pads with varying segment widths. The contact wear model simulates the contact between an oxide die and the polymeric polishing pad. We predict the contact pressure distribution by representing three modes of pad deformation: surface deformations due to local indentations, bulk pad segment downward displacements due to net pressures on the segments, and bulk pad rotations due to effective moments. We calculate the bulk deformations uses finite element simulations (details provided in [1]). We calculate the deformation due to local indentations using Nogi and Kato’s approach [2]. We then use the predicted pressure distribution, along with Preston’s equation, to incrementally remove oxide on the die. As discussed in [1], the pad’s surface texture has a dominant effect in contact wear simulations. Whereas [1] used theoretically generated pad surfaces, here, we use experimentally measured, used pad surfaces. Figure 1(a) shows the measured pad scan (left: the 95 mm x 77 mm overview of the concentric ring pad; right: a cross-section through the pad scan). Our experiment uses a 1.35 mm concentric ring pad, where the segments have an elastic modulus of 790 MPa and the subpad has an elastic modulus of 55 MPa. Note that for this particular pad, since the elastic modulus of the segment material is only one order of magnitude greater than the elastic modulus of the subpad, the effects of local surface indentations dominate over any bulk segment downward displacements or rotations. Modelling results (shown in Figure 1b) indicate that simulations for this pad track the experiments well. The lines represent the simulation, and the error bars represent experimental measurements (95% confidence intervals of the mean). Simulations track the experiments reasonably well. Simulations lie within the error bars for all 4 of the trench measurements, and they lie within the error bars for 3 of the 4 active measurements. Further work aims to improve this agreement. The simulations show the evolution of the die’s topography throughout the polish. Figure 1c shows the simulated die topography after 0, 25, 50, and 75 s of polish. Simulations do indeed show oxide planarization, as the within-die range decreases as the polish time increases. We use these contact simulations to evaluate neighborhood effects. For the 200 micron checkerboard array, we change the pattern density of the features directly to the right. Figure 1d shows that as the pattern density of the right neighbor increases, the material removal rate of the features being evaluated decreases. Of particular note is that the length scale over which these neighborhood effects acts is about 4 mm. We also use this model to predict the polish behavior when using different pad designs. Here, we focus on the pad ring width’s impact on the polish properties. We simulate polishes with four different ring widths: 0.35 mm, 0.45 mm, 0.70 mm, and 1.35 mm. Simulations indicate that pads with a larger ring width result in a larger planarization efficiency, compared to the smaller ring width pads. Figure 1e shows a plot of the predicted planarization efficiency versus step height for a 200 µm checkerboard array using a specific pad design. As the pad ring width is increased, simulations predict higher planarization efficiencies, as expected. Further, simulations show that pads with a larger ring width result in a smaller within-die range. Figure 1f shows a plot of the simulated within-die range versus pad ring width. The within-die range is calculated at the instance when the average active oxide thickness across the die is 100 Å. The plot shows that pad designs with a larger ring width result in a smaller within-die range, as expected. Looking forward, our die-scale model can be used to optimize pad designs and die layouts. References: [1] B. Salazar et al., ICPT 2019, Hsinchu, Taiwan. [2] T. Nogi and T. Kato, Journal of Tribology, 1997. Figure 1