Linear Prediction for CMP Endpoint Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Chemical mechanical polishing (CMP) processes face challenges in determining the polishing endpoint due to variations in slurry distribution, polishing pad condition, and load on the substrate, leading to unpredictable material removal rates and delays in endpoint detection, which existing in-situ monitoring techniques fail to adequately address.
Innovation Solution
The implementation of a filtering method that uses linear prediction and a modified Kalman filter to generate predicted values from sensor signals, reducing filter delay and enabling more accurate and timely detection of the polishing endpoint by adjusting for noise and systematic disturbances.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a causal filter is used to remove noise from the sensor signal in real time, then the signal can be smoothed and analyzed, but the filtered signal lags behind the raw signal, introducing unacceptable delay in endpoint detection
Solution Approach 1:
The patent applies linear prediction to forecast future signal values based on past and present data points. By predicting what the signal will be at the current moment using historical data, the system compensates for the inherent delay in causal filtering, effectively presenting an 'advance' signal that accounts for the filter lag and enables timely endpoint detection.
2Measurement precision
If polishing time is extended to account for filter delay, then endpoint detection may be more accurate, but the substrate becomes over-polished, reducing manufacturing precision
Solution Approach 1:
The patent implements a feedback mechanism where the filtered and linearly predicted signal is continuously monitored, and when the endpoint criterion is met, the polishing process is immediately halted. This closed-loop control prevents over-polishing by using the predicted signal to trigger the stop command at the precise moment the endpoint is reached, rather than waiting for delayed filtered signal confirmation.
3Reliability
If traditional filtering methods are used, then noise can be removed from the signal, but the material removal rate variations and systematic disturbances cannot be adequately compensated
Solution Approach 1:
The patent transforms the approach by changing from simple noise filtering to linear prediction based endpoint detection. Instead of merely removing noise, the system uses linear prediction to model and compensate for systematic disturbances and material removal rate variations, fundamentally changing the parameter being processed from raw/filtered signal to predicted signal that accounts for process dynamics.
Data Source
AI summary
A method of controlling polishing includes polishing a substrate, during polishing monitoring the substrate with an in-situ monitoring system, the monitoring including generating a signal from a sensor, and filtering the signal to generate a filtered signal. The signal includes a sequence of measured values, and the filtered signal including a sequence of adjusted values. The filtering includes for each adjusted value in the sequence of adjusted values, generating at least one predicted value from the sequence of measured values using linear prediction, and calculating the adjusted value from the sequence of measured values and the predicted value. At least one of a polishing endpoint or an adjustment for a polishing rate is determined from the filtered signal.