Tackling Cu CMP Dishing and Erosion

Copper chemical mechanical planarization (CMP) is a well-known trouble spot in the fab. It is the final step in the Cu damascene process, where trenches in a dielectric are filled with copper — actually overfilled — and then that Cu (and underlying barrier layer) is polished back to isolate the Cu lines.

The two big problems in Cu CMP are dielectric erosion and metal dishing. Erosion of the dielectric occurs from overpolishing, which is necessary to make sure all the metal is removed from the top of the dielectric (otherwise lines could be shorted together). Dishing occurs on the soft metal, which reduces the thickness of the Cu interconnects and increases the electrical resistance. The overall changes in the planarity caused by dishing and erosion can also lead to challenges in obtaining good focus across the die during subsequent lithographic steps.

To better understand dishing and erosion in an attempt to minimize it, researchers from the Massachusetts Institute of Technology's (Cambridge, Mass.) mechanical engineering department have developed an exhaustive steady-state model of the Cu CMP process, and backed it up with a battery of experiments that looked at the effects of pattern geometry, pad displacement, slurry particle size, slurry pH and other factors. Figure 1 shows the various parameters that entered into calculations made by team members Jiun-Yu Lai, Nannaji Sak and Jun-Hoon Chun.

1. Schematic of the onsets of dishing and erosion with modeling parameters indicated. (Source: MIT)

2. Cross-sectional profiles of the evolution of a pattern with a 5 µm linewidth and 200 µm pitch. (Source: MIT)

As the step-height between the high and low features decreased, the material removal rate on the high features approached that of the low features. This indicates that the pressure distributed more uniformly while the surface was smoothed out. Finally, both material removal rates were close to the blanket Cu polishing rate, ~200 nm/min, and the surface was planarized. Then the Cu surface remained flat until the process endpoint was reached. After passing the endpoint, which was 3-4 min, the Cu lines dished and the amount of dishing increased with polishing time. It is not clear if the Cu started dishing during or after Ta (the diffusion barrier) clearing because the Ta is removed in a very short period. The oxide was also polished, but at a rate much slower than that of soft Cu.

For additional information on wafer processing, go to www.semiconductor.net/wafer.