Machine-Learning Run-to-Run Control for Upstream Process Drift
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional run-to-run process control systems in manufacturing systems face challenges in determining which and how much to modify process control variable settings to optimize substrate processing, leading to increased variability, latency, and reduced throughput due to the inability to account for upstream process impacts on subsequent processes.
Innovation Solution
The implementation of a machine learning-based method that uses a trained model to predict metrology measurement drift from target values, allowing for real-time adjustments to process recipes by identifying and correcting for drift caused by upstream processes without removing substrates from the manufacturing system.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If conventional process control systems modify process recipe settings to optimize substrate characteristics, then substrate quality improves, but it becomes difficult to determine which and how much to modify process control variables
Solution Approach 1:
A machine learning model is introduced as an intermediary between process data and control decisions. The model takes process parameters and metrology measurements as input and predicts optimal process recipe modifications, eliminating the need for complex manual determination of which variables to adjust and by how much.
Solution Approach 2:
The system implements a feedback loop where metrology measurements from previous runs are fed into the machine learning model, which then determines process recipe modifications for subsequent runs. This closed-loop feedback enables continuous optimization of substrate characteristics based on actual measurement data.
2Manufacturing precision
If process control systems account for upstream process impacts on subsequent processes, then substrate variability reduces, but processing latency increases due to the need for extensive analysis
Solution Approach 1:
The machine learning model is trained in advance on historical process data and metrology measurements from multiple upstream processes. This preliminary training enables the model to quickly predict the combined impact of upstream processes on subsequent substrate characteristics without requiring real-time analysis of all upstream process interactions.
Solution Approach 2:
The system transforms complex multi-process relationship analysis into a parameter prediction problem. By changing the approach from analyzing process interactions to predicting metrology outcomes based on input parameters, the system achieves fast predictions that account for upstream impacts without extensive real-time computation.
3Productivity
If conventional systems perform run-to-run control without machine learning, then system complexity remains low, but throughput reduces due to inability to optimize process settings effectively
Solution Approach 1:
The machine learning model enables the process control system to self-optimize by automatically analyzing metrology data and determining process recipe modifications without human intervention. This self-service capability allows the system to continuously improve throughput by optimizing process settings based on actual substrate outcomes.
Solution Approach 2:
The patent replaces manual or rule-based process optimization mechanisms with a machine learning-based predictive system. This substitution transforms the control approach from reactive adjustment based on simple rules to proactive optimization driven by data-driven predictions, significantly improving throughput.
Data Source
AI summary
First data associated with a first process performed for a first layer of a substrate is identified. The first layer is to be further processed according to a second process. The first data is provided as input to a machine learning model that is trained to predict metrology measurement values for layers of substrates at the manufacturing system. An amount of drift of a first set of metrology measurement values for the first layer following completion of the first process and/or the second process from target values is determined. Modifications to a recipe for the second process is determined in view of the determined amount of drift and second data associated with a second substrate layer that was previously processed at the manufacturing system. The second process is updated based on the determined one or more modifications.