Virtual Chamber Measurement for Substrate Property Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In manufacturing chambers, such as those used for semiconductor wafers, direct measurement of properties proximate to the substrate is often impractical or impossible, leading to inefficient experimentation and frequent recalibration due to chamber drift and variability between identical chambers, resulting in suboptimal processing conditions and increased costs.
Innovation Solution
A physics-informed compressed sensing machine learning model is used to infer property values at the substrate location by measuring properties at other locations within the chamber, utilizing a trained model that combines sensor data with a physics-based model to predict property maps, allowing for virtual measurement and adaptive process control.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If sensors are placed at the substrate location to directly measure properties, then measurement precision is improved, but device complexity and ease of operation deteriorate due to the impracticality and inconvenience of placing sensors at the substrate location
Solution Approach 1:
The patent uses intermediate sensor locations around the chamber as mediators to indirectly measure substrate properties. Sensors are placed at accessible locations (chamber walls, gas distribution plate, susceptor) rather than directly at the substrate, and a machine learning model translates these intermediate measurements into substrate property predictions, resolving the contradiction between measurement accuracy and operational convenience
Solution Approach 2:
The patent creates a virtual copy of the substrate property measurements through machine learning prediction. Instead of physically measuring at the substrate location, the system learns the relationship between sensor readings and substrate properties from training data, then uses this learned model to generate accurate predictions (copies) of substrate properties from remote sensor measurements
2Manufacturing precision
If experimentation is performed frequently to account for chamber drift and variability, then manufacturing precision is improved, but productivity deteriorates due to the time and resource costs of frequent experimentation and recalibration
Solution Approach 1:
The patent performs preliminary action by training the machine learning model offline using historical data that captures chamber drift and variability. Once trained, the model can predict substrate properties in real-time without requiring frequent recalibration or experimentation, thus maintaining manufacturing precision while improving productivity by eliminating the need for frequent re-experimentation
Solution Approach 2:
The system implements feedback by using the machine learning model to continuously predict substrate properties based on real-time sensor measurements. This feedback loop allows the system to adapt to chamber drift and variability automatically, maintaining precision without requiring manual intervention or frequent experimentation, thereby preserving productivity
Data Source
AI summary
A method includes measuring a subset of property values within a manufacturing chamber during a process performed on a substrate within the manufacturing chamber. The method further includes determining property values in the manufacturing chamber at locations removed from the locations the measurements are taken. The method further includes performing a corrective action based on the determined properties.