Modeling of chemical mechanical polishing process using FEM and abductive network

In this paper, modeling of a two-dimensional axisymmetric quasic-static finite element model in conjunction with an abductive network for chemical-mechanical polishing process (CMP) was established. Three prediction models can be achieved, i.e., model for von Mises stress at wafer center, model for maximum von Mises stress and model for nonuniformity on wafer surface under various combinations of process parameters. The data of von Mises stress and nonuniformity on wafer surface can be first achieved under different conditions of the carrier load, pad's elastic modulus and thickness by using the developed finite element model for CMP. Next, an abductive network was applied to synthesize the data sets from the FE simulation. It is a self-organizing adaptive modeling tool that establishes the mathematical relationship between input and output variables based on abductive modeling technique and it can automatically synthesize the optimal network structure, including the optimal network structure, the number of layers and the form of functional nodes. Finally, the results from the three developed abductive networks with test data are compared with those from FE simulation to confirm the feasibility of this approach. The findings verified that the results confirm the feasibility and the proposed prediction models for CMP are acceptable.