CMP Monitoring Models for Endpoint and Pressure Control
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 load, making it difficult to determine the polishing endpoint and achieve desired substrate profiles.
Innovation Solution
A method involving machine learning models and physical process models is used to generate characterizing values from training data, allowing for the training of a machine learning model that can control polishing parameters such as carrier head pressure and endpoint detection during CMP, leveraging in-situ monitoring systems and metrology data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If constant pressure is applied during CMP, then the polishing process is simple to control, but the material removal rate varies across substrates and from substrate to substrate
Solution Approach 1:
The patent implements dynamic pressure control by adjusting the carrier head pressure in real-time based on in-situ monitoring data and machine learning predictions. Instead of applying constant pressure, the system continuously modifies pressure levels to compensate for variations in material removal rate, achieving uniform polishing across different substrates while maintaining operational simplicity through automated control.
Solution Approach 2:
The system employs feedback control by using in-situ monitoring systems to measure thickness in real-time during polishing, comparing these measurements against target values, and adjusting carrier head pressure accordingly. This closed-loop feedback mechanism enables the system to maintain consistent material removal rates despite variations in substrate properties, slurry distribution, or pad conditions.
2Ease of operation
If polishing time is used to determine endpoint, then the process is easy to monitor, but the endpoint detection accuracy is insufficient due to variations in polishing rate
Solution Approach 1:
The patent replaces time-based endpoint determination with machine learning-based prediction systems that analyze in-situ monitoring data (optical, acoustic, or other signals) to predict the polishing endpoint. This substitution of mechanical/time-based monitoring with intelligent algorithms enables accurate endpoint detection while maintaining ease of monitoring through automated analysis of process signals.
Solution Approach 2:
The system introduces machine learning models as intermediaries between the raw in-situ monitoring signals and the endpoint determination. These models process and interpret the monitoring data, translating complex signal patterns into accurate predictions of the polishing endpoint, thereby improving detection accuracy without complicating the monitoring process for operators.
3Manufacturing precision
If in-situ monitoring is implemented to adjust pressure and detect endpoint, then the polishing precision is improved, but the system complexity increases
Solution Approach 1:
The patent implements multi-functional in-situ monitoring systems that simultaneously perform multiple tasks: measuring thickness, detecting endpoint, guiding pressure adjustment, and providing data for machine learning training. This universal monitoring approach improves polishing precision while reducing overall system complexity by consolidating multiple functions into a single integrated system rather than requiring separate systems for each function.
Solution Approach 2:
The system employs self-service mechanisms where the in-situ monitoring data automatically feeds into machine learning models that generate control decisions without requiring external intervention. The system self-adjusts pressure and self-detects endpoint based on real-time data, improving precision while minimizing the complexity of external control systems and reducing the need for operator intervention.
Data Source
AI summary
Operating a substrate processing system includes receiving a plurality of sets of training data, storing a plurality of machine learning models, storing a plurality of physical process models, receiving a selection of a machine learning model from the plurality of machine learning models and a selection of a physical process model from the plurality of physical process models, generating an implemented machine learning model according to the selected machine learning model, calculating a characterizing value for each training spectrum in each set of training data thereby generating a plurality of training characterizing values with each training characterizing value associated with one of the plurality of training spectra, training the implemented machine learning model using the plurality of training characterizing values and plurality of training spectra to generate a trained machine learning model, and passing the trained machine learning model to a control system of the substrate processing system.