Sensor Based Modeling of Chemical Mechanical Planarization (Cmp) of Copper For Semiconductor Applications

Abstract

Material removal rate (MRR) and surface quality in copper CMP (Cu-CMP) process are highly sensitive to slurry chemistry parameters, namely, pH, and concentrations of complexing, corrosion inhibiting, and oxidizing agents. Capturing the effects of these slurry parameters on MRR and surface quality in real-time through the use of sensor signals is key to ensuring an efficient Cu-CMP process. In this investigation vibration sensor signals collected from the Cu-CMP experiments are used to capture the variations in various slurry parameters as well as their influence on the MRR.The study has shown that features from wireless accelerometer signals sampled at 500Hz, and those from wired accelerometer signals sampled at 5 kHz can be used to estimate MRR more accurately than conventional static statistical regression models that relate the input (slurry) parameters to MRR. Here, the sensor features have been related to MRR using principal component regression (PCR) models. The improvement in the accuracy of estimation with sensor-based PCR models (R2 of 97.7% compared to 89.8% with a conventional statistical regression model) is likely because the vibration sensor signal characteristics are not only sensitive to variations in MRR, but also to the relevant variations in the input (slurry) parameters during the operation. The in-process variations in the slurry parameters cannot be tracked in conventional (static) statistical regression models.