Real-time prediction of removal rate and friction coefficient during CMP using motor load currents of polisher

Abstract

CMP process used to polish glass substrates for hard disk drives and sapphire substrates for LEDs is rarely automated, and the success or failure of this process is depended on the skilled technician. In this study, the prediction of the removal rate and friction coefficient was conducted using only the real-time data output from the motors on the polisher. Multiple regression analysis was performed using the acquired training data set with the motor load current as the objective variable and the removal rate and friction coefficient as explanatory variables. The results showed that the predicted removal rate was approximately 1 μm/h lower than the experimental value. Therefore, we performed multiple regression analysis again after compensating for mechanical characteristics by conducting a low pressure polishing test. Multiple regression analysis was also performed by changing from the slurry free introducing method to the circulation method. As a result, it was found that the compensation process can improve the prediction accuracy and it can be applied to monitoring system the consumable condition by real-time prediction.