Predictive Metrology Model Weighting for Stable Output Data

Solution Overview

Problem

Machine-learning models used in electronic device manufacturing experience fluctuations in output data, leading to errors in predictive metrology, which can result in defective substrates.

Innovation Solution

The system generates a characteristic sequence defining the relationship between manufacturing parameters and determines weights based on this sequence to apply to features, thereby training a machine-learning model to reduce output data fluctuations.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a machine-learning model is trained using standard sensor data from manufacturing processes, then the model can generate predictive metrology data, but the output data exhibits fluctuations leading to errors

Engineering Contradiction:

Improvepredictive metrology accuracyVSAvoidoutput data consistency

Core Design Contradiction:

Measurement precisionVSReliability

Solution Approach 1:

The system performs preliminary actions by generating characteristic sequences that define relationships between manufacturing parameters before training the machine-learning model. These sequences are used to determine weights for features, which are then applied to the sensor data prior to model training. This preliminary weighting process stabilizes the training data, preventing output fluctuations and improving both measurement precision and reliability of the predictive metrology data.

Inventive Principle:

Principle #10Preliminary action

2Productivity

If sensor data is used directly as input to the machine-learning model, then the model can be trained quickly, but the predictive data contains errors due to unweighted features

Engineering Contradiction:

Improvemodel training efficiencyVSAvoidpredictive metrology accuracy

Core Design Contradiction:

ProductivityVSMeasurement precision

Solution Approach 1:

The system performs preliminary actions by generating characteristic sequences that define relationships between manufacturing parameters before training the machine-learning model. These sequences are used to determine weights for features, which are then applied to the sensor data prior to model training. This preliminary weighting process stabilizes the training data, preventing output fluctuations and improving both measurement precision and reliability of the predictive metrology data.

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS20250189957A1Methods and mechanisms for preventing fluctuation in machine-learning model performance

Publication Date: 2025.06.12 APPLIED MATERIALS INC

AI summary

An electronic device manufacturing system configured to obtain, by a processor, sensor data associated with a substrate manufacturing process performed in a process chamber. The sensor data is provided, as input data, to a machine learning model. The machine learning trained uses a weighted feature that reflects a relationship between one or more variables related to an initial feature and a characteristic sequence that defines a relationship between two or more manufacturing parameters. An output value of the machine learning model is obtained, the output value being indicative of metrology data.