ML Spectral Endpoint Detection for Substrate Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional endpoint detection systems in microelectronics manufacturing are inadequate for precise and real-time monitoring of substrate processes, leading to substandard devices due to under-processing or over-processing issues, especially with increasing demands for quality and complexity in semiconductor devices.
Innovation Solution
A machine learning-based system that collects and analyzes spectral data to determine endpoint metrics by training models on historical data, allowing for real-time metrology measurement values and confident endpoint detection, thereby preventing under-processing or over-processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional endpoint detection systems are used, then device manufacturing can proceed, but measurement precision and manufacturing precision deteriorate leading to substandard devices
Solution Approach 1:
The patent replaces conventional optical control systems with a machine learning-based detection system. The system uses spectral data collected during substrate processing as input to trained machine learning models that output metrology measurement values, substituting traditional mechanical/optical measurement methods with intelligent algorithms that analyze spectral patterns to determine endpoint metrics with higher precision.
Solution Approach 2:
The patent transforms raw spectral data into normalized spectral data by applying mathematical transformations. The system changes the parameter representation from raw spectral measurements to normalized values that highlight relevant features, enabling the machine learning models to detect endpoint conditions with improved precision and accuracy.
2Manufacturing precision
If real-time monitoring is implemented, then device quality improves, but system complexity increases
Solution Approach 1:
The patent creates a multi-functional system where the machine learning models serve multiple purposes: they analyze spectral data, determine endpoint metrics, provide real-time monitoring, and generate confidence levels. This universal approach consolidates multiple detection functions into a single integrated system, managing complexity through functional integration rather than proliferation of separate components.
Solution Approach 2:
The patent introduces normalized spectral data as an intermediary between raw spectral measurements and final endpoint determination. This intermediate representation simplifies the analysis by preprocessing the data to emphasize relevant features, reducing the complexity of the machine learning models needed while maintaining high detection accuracy.
3Measurement precision
If machine learning models are trained on historical spectral data, then measurement accuracy improves, but data processing time increases
Solution Approach 1:
The patent performs preliminary actions by training machine learning models on historical spectral data before actual substrate processing. The models are pre-trained to recognize patterns and relationships in spectral data, so during real-time processing, they can quickly analyze new spectral measurements and provide endpoint detection with high accuracy without requiring extensive computation during the actual manufacturing process.
Solution Approach 2:
The patent extracts and utilizes only the most relevant features from spectral data for training and analysis. By focusing on specific spectral characteristics that are most indicative of endpoint conditions, the system reduces the dimensionality and complexity of data processing while maintaining high measurement accuracy, thereby reducing processing time.
Data Source
AI summary
Methods and systems for detection of an endpoint of a substrate process are provided. A set of machine learning models are trained to provide a metrology measurement value associated with a particular type of metrology measurement for a substrate based on spectral data collected for the substrate. A respective machine learning model is selected to be applied to future spectral data collected during a future substrate process for a future substrate in view of a performance rating associated with the particular type of metrology measurement. Current spectral data is collected during a current process for a current substrate and provided as input to the respective machine learning model. An indication of a respective metrology measurement value corresponding to the current substrate is extracted from one or more outputs of the trained machine learning model. In response to a determination that the respective metrology measurement satisfies a metrology measurement criterion, an instruction including a command to terminate the current process is generated.