New Model Could Vastly Improve CMP Processes

-- Semiconductor International, 5/1/1999

A model developed by Bin Zhao of Conexant Systems (formerly Rockwell Semiconductor Systems, Newport Beach, Calif.) and Frank G. Shi of UC Irvine better predicts the pressure dependence of polishing rate in dielectric and metal chemical-mechanical polishing (CMP). Using various slurries, film types and removal conditions, the authors demonstrated a P^2/3 dependence of applied pressure on removal rate. Presenting their findings at the CMP-MIC Conference in Santa Clara on Feb. 11, Zhao and Shi also identified a critical threshold pressure in IC CMP processes. By raising this threshold pressure using surfactants or other means, planarization efficiency can be significantly improved. The findings have widespread implications to CMP consumables development and CMP process optimization.

By studying the subtle interactions among CMP process parameters, wafer surface properties, slurry properties and pad properties, Zhao and Shi found that Preston's equation, widely used in CMP process control and consumables development, often leads to discrepancies between modeled results and experimental observations. In fact, Preston's equation is only applicable when the polishing pad material is as hard or harder than the abrasive particles of the slurry and the substrate material. The softer, polyurethane polymer-based pads most prevalent in semiconductor CMP processes are typically much softer than the abrasive particles and silicon wafer. The resulting CMP removal rate demonstrates a sub-linear dependence on pressure given by:

RR = K(V)P^2/3

Where RR is the removal rate, P is the pressure applied to the polished wafer and K(V) is a function of the relative velocity between the wafer and pad and other CMP parameters.

Fig. 1. Experimental data correlates well with the P2/3 dependence of removal rate on applied pressure.

Fig. 2. Experimental polishing of copper films indicates a P2/3 dependence and a distinct threshold pressure below which removal rate is extremely low.

Fig. 3. Increasing the threshold pressure improves planarization efficiency for various final step heights (500 nm initial step height), indicating an optimization opportunity not previously explored.

Polishing rate with a hard pad is fundamentally different as increased pressure increases the indentation depth of slurry particles into the wafer surface. Modeling shows the linear dependence of Preston's equation. With soft pads, the pad compresses upon contact between the hard slurry particles and wafer. Increased pressure increases the number of particles in contact with the wafer, but the indentation depth is essentially unchanged. Removal rate depends on the pressure dependence of the total number of particles in contact with the wafer.

Figure 1 shows experimental data demonstrating the P^2/3 dependence of polishing rate for oxide films. Zhao and Shi also found that the pressure dependence of polishing rate significantly influences planarization efficiency and pattern density dependence. For instance, not all particles in contact with the wafer necessarily remove material efficiently. A particle rolling against the wafer surface removes little material, yet a particle held firmly by the pad slides against the wafer surface, changing the surface structurally and chemically. Sliding and removing requires a minimum local pressure, giving a revised polishing rate of:

RR = K(V)(P^2/3 - P_th^2/3) (P>= P_th)

and

RR = 0 (P< P_th)

where P_th is the threshold pressure. Experimental results from dielectric and other CMP processes, including Al CMP, W CMP and Cu CMP (Figure 2), fully support this new equation. In addition, previous studies demonstrated that a surfactant can be added to a cerium oxide slurry for oxide CMP to introduce a polishing threshold pressure. The existence of a threshold pressure further explains negative values obtained when polishing rate at zero pressure is extrapolated from experimental data for many semiconductor films.

Finally, Zhao and Shi measured remaining step heights for features of different pattern densities as a function of polishing time, finding that non-zero threshold pressure enhances the pattern density effect between high and low densities. Polishing threshold pressure is also responsible for the pressure dependence of polishing selectivity (between the material being polished and the underlying material). As a result, threshold pressure influences planarization efficiency (a measure of how much material has to be removed before planarization is achieved). The effect for various final step heights (Figure 3) shows an increase in efficiency with threshold pressure, requiring less material removal to achieve a given planariz ation result. This study indicates that careful characterization of polishing threshold pressure will likely lead to superior CMP processes.