Training Spectra Generation for CMP Endpoint Detection
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 conditions, and pressure, making it difficult to determine the polishing endpoint and achieve desired profiles.
Innovation Solution
A method involving the generation of training spectra for neural networks using a combination of sample and dummy substrate data, along with optical monitoring, to accurately measure layer thickness and adjust processing parameters, reducing within-wafer and wafer-to-wafer non-uniformity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If a constant pressure is applied during CMP, then the polishing process is simple to control, but the polishing uniformity deteriorates due to variations in material removal rate
Solution Approach 1:
The system dynamically adjusts the carrier head pressure during polishing based on real-time optical monitoring feedback. The control system modifies pressure parameters in response to measured layer thickness variations, transforming the static pressure application into a dynamic control process that maintains polishing uniformity across the substrate surface.
Solution Approach 2:
The system implements a closed-loop feedback mechanism where optical monitoring continuously measures layer thickness during polishing, and this measurement feedback is used to adjust the carrier head pressure. The control system processes the optical signals and modifies pressure parameters to compensate for thickness variations, achieving uniform polishing through feedback-driven adaptation.
2Measurement precision
If in-situ optical monitoring is implemented, then the polishing endpoint detection accuracy is improved, but the system complexity increases
Solution Approach 1:
The system replaces mechanical endpoint detection methods with optical monitoring. Instead of relying on mechanical measurements or post-process inspection, the system uses optical signals to non-contactly measure layer thickness in real-time during polishing, substituting mechanical systems with optical fields for enhanced measurement precision.
Solution Approach 2:
The optical monitoring system performs self-calibration and adaptive measurement during the polishing process. The system automatically adjusts its measurement parameters and compensates for optical path variations without external intervention, enabling accurate endpoint detection while minimizing the complexity of external calibration equipment.
3Measurement precision
If empirical data from multiple substrates is collected for training, then the neural network accuracy is improved, but the data collection time and cost increase
Solution Approach 1:
The system uses dummy substrates as copies or proxies for actual device substrates during data collection. By measuring optical signals from dummy substrates with known layer structures, the system generates training data without consuming valuable device substrates, reducing both time and cost while maintaining the statistical representativeness needed for accurate neural network training.
Solution Approach 2:
The system performs preliminary data collection using dummy substrates before actual device processing. By gathering training data in advance from dummy substrates with varying layer thicknesses and compositions, the neural network is pre-trained to recognize polishing patterns, reducing the need for extensive data collection during production and accelerating the overall process.
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
This approach reduces the quantity of empirical data required for training, enhances the accuracy and speed of layer thickness measurement, and improves the reliability of endpoint detection, leading to more uniform polishing results.
Implementation Method 1
measuring a first plurality of training spectra from one or more sample substrates
Data Source
AI summary
A method of generating training spectra for training of a neural network includes measuring a first plurality of training spectra from one or more sample substrates, measuring a characterizing value for each training spectra of the plurality of training spectra to generate a plurality of characterizing values with each training spectrum having an associated characterizing value, measuring a plurality of dummy spectra during processing of one or more dummy substrates, and generating a second plurality of training spectra by combining the first plurality of training spectra and the plurality of dummy spectra, there being a greater number of spectra in the second plurality of training spectra than in the first plurality of training spectra. Each training spectrum of the second plurality of training spectra having an associated characterizing value.