Device independent process control of dielectric chemical mechanical polishing

The use of the chemical-mechanical polishing (CMP) process in the semiconductor industry is growing rapidly, and it is a critical step in the manufacturing of integrated circuits. The CMP process is complicated by many factors, and controlling all of these factors in a single controller has been unrealistic. One of the most significant factors complicating control is the dependency of the polishing process on the pattern layout of the particular device being polished. The interactions between these patterns and the polishing behavior of a CMP tool make monitoring and controlling the process particularly difficult. Current techniques focus on the control of a few sites on a single type of layout being polished on a single tool. We show that using only a few sites does not give insight into what is happening at sites that are not measured and controlled. Further, by restricting attention to a single device, these controllers address only pieces of a larger problem and fail to take into account the effects of different devices being polished on the same tool. This thesis outlines a comprehensive framework for controlling the polishing of multiple devices with arbitrary layout patterns being polished on a single CMP tool. We explore the use of an advanced CMP process model in conjunction with an on-line metrology system and a simple filtering algorithm for controlling the average post-polish thickness and monitoring the global non-uniformity of multiple devices being processed. This framework provides several benefits. First, it allows measurements from any device to be used to update the tool level model that can be used with any other device being processed. Second, it allows us to measure only a few points, while very accurately controlling the average of the true thickness profile. Third, it allows us to monitor the total non-uniformity of the polishing process for each device being processed. Experimental work shows that this approach results in very accurate control of the true average thickness of multiple devices, with a lot to lot variability of only 100A. In addition, we are able to very accurately predict the global non-uniformity of the polished wafers. However, the model used was found to have a minor dependence on the device type, indicating that an improved layout pattern functional model is necessary to achieve truly device independent control. We explore one possible model that might reduce these dependencies, and show this model provides a 50% improvement in fitting errors of both raised and down area thicknesses. Despite this, we find that device dependencies still exist with the improved model, and further work is necessary to make the model and control strategy completely device independent.