In-Situ Spectral Monitoring for CMP Endpoint and Thickness Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Chemical mechanical polishing (CMP) processes face challenges in achieving consistent material removal rates due to variations in substrate thickness, slurry distribution, polishing pad condition, and relative speed, making it difficult to determine the polishing endpoint accurately and achieve uniformity across substrates.
Innovation Solution
A method involving in-situ spectrographic monitoring and an artificial neural network is used to measure and predict substrate thickness by training on a sequence of test spectra, combining data from both optical and non-optical monitoring systems to improve accuracy and uniformity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If constant pressure is applied during polishing, then the polishing process is simple to control, but the material removal rate varies across substrates and cannot achieve desired profile uniformity
Solution Approach 1:
The system transitions from static constant pressure control to dynamic pressure control based on real-time optical monitoring. The controller continuously adjusts carrier head pressure during polishing based on reflected light intensity measurements, enabling adaptive compensation for variations in material removal rate while maintaining simple operator interface.
Solution Approach 2:
An optical monitoring system provides real-time feedback on substrate thickness or reflectivity during polishing. This feedback signal is fed to the controller which adjusts the carrier head pressure accordingly, creating a closed-loop control system that maintains profile uniformity without requiring complex manual control.
2Ease of operation
If polishing time is used to determine endpoint, then the process is easy to manage, but variations in initial thickness and polishing rate cause inaccurate endpoint detection
Solution Approach 1:
The system replaces time-based endpoint determination with optical property-based detection. Instead of relying on chronological polishing duration, the system uses real-time measurement of reflected light intensity or spectral characteristics to detect when the polishing endpoint is reached, significantly improving accuracy while maintaining ease of operation through automated detection.
Solution Approach 2:
The optical monitoring system continuously measures substrate properties during polishing and provides feedback to the controller. When the measured optical properties indicate the endpoint condition is met, the system automatically signals completion, replacing subjective time-based judgment with objective real-time detection.
3Measurement precision
If in-situ spectrographic monitoring with neural network is implemented, then thickness measurement accuracy and endpoint detection reliability improve, but system complexity and training requirements increase
Solution Approach 1:
The system performs preliminary training of the neural network using training spectra obtained during test substrate polishing before actual production use. This pre-training phase establishes the relationship between optical spectra and thickness values, enabling accurate real-time measurements during subsequent polishing operations without requiring complex manual calibration during production.
Solution Approach 2:
The system uses training spectra copied from test substrate polishing to create a digital model (neural network) that replicates the relationship between optical measurements and physical thickness. This digital copy enables accurate thickness determination during production polishing without requiring physical reference standards or complex measurement apparatus.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
The method enhances the accuracy and speed of thickness measurement, reduces within-wafer and wafer-to-wafer non-uniformity, and improves the reliability of endpoint detection in CMP processes.
Implementation Method 1
measuring by an in-situ spectrographic monitoring system a sequence of test spectra of light reflected from the substrate during polishing of the test substrate
Data Source
AI summary
A method of training a neural network for spectrographic monitoring includes polishing a test substrate, measuring by an in-situ spectrographic monitoring system a sequence of test spectra of light reflected from the substrate and measuring by an in-situ non-optical monitoring system a sequence of test values from the substrate during polishing of the test substrate, measuring at least one of an initial characterizing value for the substrate before polishing or a final characterizing value for the substrate after polishing, inputting the sequence of test values and the initial characterizing value and/or final characterizing value into a thickness predictive model that outputs a sequence of training values with each respective training value in the sequence of training values associated with a respective test spectrum from the sequence of test spectra, and training an artificial neural network using the plurality of training spectra and the plurality of training values.