Loading...
Thumbnail Image
Publication

Wireless Sensor Fusion Approach for Monitoring Chemical Mechanical Planarization (CMP) Process

Ohri, Amit
Abstract
Chemical mechanical planarization (CMP) is used in the microelectronics and optical industries for local as well as global planarity and for producing mirror finished surfaces. Roughness (Ra), within- non-uniformity (WIWNU), and material removal rate (MRR) are the major performance variables in polishing. CMP is a complex process involving some 36 input variables. Analysis of the review of the literatures showed that static models that use process parameters are inadequate for estimating and monitoring the performance variables in the CMP process. Pad-level interactions play a major role in polishing. Sensor based monitoring techniques enables monitoring of the CMP process. Additionally, sensor fusion techniques may facilitate in improving the robustness and monitoring the process beyond using one sensor. In this work, wireless vibration (Z-axis) and temperature sensors mounted on a bench top polisher (ECOMET polisher from Buehler) are used to monitor the material removal rate (MRR) and surface finish (Ra). The wireless sensor platform has a sampling rate of 500 Hz for the vibration sensor and 4 Hz for the temperature sensor. Alumina-based alkaline slurry is used in polishing process. The process conditions include two loading conditions (10 lb and 5 lb) and two rotational speeds (500 rpm and 300 rpm). The polishing studies were conducted on a 1.6" copper samples and Microcloth pad (from Buehler). The overall approach used involves relating the various sensors signal features to MRR and Ra from the CMP process. The vibration features were extracted using statistical, frequency, and RQA (non-linear) analysis techniques. The vibration features were combined with temperature features to build multiple linear regression models. The regression fitting accuracy for the roughness model is ~ 93% using the statistical features, such as maximum and kurtosis, time-frequency features, such as energy, nonlinear features such as LAM and Lmax and thermal features such as net temperature rise and temperature rise rate. The regression fitting accuracy for the MRR model is ~ 91% using the statistical features, such as variance and skewness, time-frequency features, such as energy and nonlinear features, such as time delay and temperature rise rate as temperature features. The thermal features are able to increase the coefficient of determination of roughness model by 10%. This wireless sensor fusion based regression models are found to be more efficient compared to a single sensor as it takes care of both the mechanical effects using a vibration sensor and thermal effects using a temperature sensor. It appears that this is the first time that a sensor fusion based technique is attempted for predicting Ra and MRR in the CMP process.
Date
2010-12-01
Collections