Etch Stage Measurement System Using Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current etch process control systems face challenges in accurately determining the endpoint of etch processes due to the complexity of optical emission spectroscopy (OES) spectra, particularly in high-density plasma sources like ICP and ECR, and require optimization for real-time measurement and repeatability across semiconductor applications.
Innovation Solution
An etch stage measurement system is designed with multiple optical metrology devices and sensors, coupled with a processor that correlates etch stage measurements to actual data using linear or non-linear relationships, optimizing wavelength selection and algorithm refinement to meet objectives such as accuracy, speed, and reproducibility.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If optical emission spectroscopy (OES) is used to track volatile etch products or reactive species, then endpoint detection capability is improved, but wavelength selection complexity increases due to hundreds of possible wavelengths
Solution Approach 1:
The patent transforms the complex wavelength selection problem into a parameter optimization problem by using machine learning algorithms to automatically identify optimal wavelength combinations. The system changes from manual wavelength selection to algorithm-driven parameter optimization, where the computer system evaluates hundreds of wavelengths and selects the most informative subset based on training data, thereby reducing complexity while maintaining measurement precision.
Solution Approach 2:
The patent introduces machine learning algorithms as an intermediary between the raw OES spectral data and endpoint detection. This intermediary layer processes the complex spectral information, automatically identifying relevant wavelengths and patterns that would be difficult for humans to discern, thus simplifying the overall system operation while improving detection accuracy.
2Manufacturing precision
If high-density plasma sources (ICP, ECR) are used to etch finer line trenches, then manufacturing precision is improved, but endpoint detection reliability deteriorates due to different emission spectra compared to RIE
Solution Approach 1:
The patent creates a universal endpoint detection system that works across multiple plasma types (RIE, ICP, ECR) by using machine learning to learn the distinct spectral characteristics of each plasma source. The system is trained on data from different plasma types and can automatically adapt to the specific emission patterns of each, making the endpoint detection reliable across diverse manufacturing processes while maintaining the precision benefits of high-density plasma sources.
Solution Approach 2:
The system dynamically adjusts detection parameters based on the identified plasma type. When ICP or ECR is detected, the machine learning algorithm modifies the wavelength selection and analysis parameters to match the specific emission characteristics of that plasma source, thereby maintaining reliable endpoint detection across different plasma technologies used for precise trench etching.
3Measurement precision
If multiple wavelengths are analyzed for endpoint detection, then measurement accuracy is improved, but measurement time increases
Solution Approach 1:
Instead of analyzing all hundreds of available wavelengths, the machine learning algorithm identifies and analyzes only the critical subset of wavelengths that provide the most information for endpoint detection. This partial action approach maintains measurement accuracy by focusing on the most relevant wavelengths while dramatically reducing the total analysis time compared to examining the complete spectral range.
Solution Approach 2:
The system performs preliminary training offline to pre-identify the optimal wavelength combinations for different etch scenarios. During actual production measurements, the system only needs to analyze these pre-identified wavelengths, eliminating the need for real-time analysis of all possible wavelengths and thus maintaining high accuracy while achieving fast measurement speeds.
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 system enhances etch process control by improving endpoint detection accuracy, reducing measurement time, and ensuring repeatability and reproducibility, thereby optimizing etch stage measurement objectives in real-time.
Implementation Method 1
An optical emission spectrometer diffracts emissions into its component wavelengths
Implementation Method 2
plasma discharge materials, such as etchant, neutral, and ions in the plasma, are continuously excited by collisions
Data Source
AI summary
Provided is a system for measuring an etch stage of an etch process involving one or more layers in a substrate, the etch stage measurement system configured to meet two or more etch stage measurement objectives. The system includes an etch process tool, the etch process tool having an etch chamber, a controller, and process parameters. The etch process tool is coupled to two or more optical metrology devices and at least one etch sensor device measuring an etch process parameter with high correlation to the etch stage. The processor is coupled to the etch process tool and is configured to extract an etch measurement value using a correlation of etch stage measurements to actual etch stage data and etch stage measurement obtained from the two or more metrology devices and the at least one etch process sensor device.