Methods and systems for detecting anomalies in manufacturing processes

Gaussian process regression models with a bathtub kernel function enhance anomaly detection in manufacturing systems, addressing the challenge of late defect identification in semiconductor production by enabling proactive process adjustments.

JP7875009B2Active Publication Date: 2026-06-17ACCENTURE GLOBAL SERVICES LTD
View PDF 8 Cites 0 Cited by

Patent Information

Authority / Receiving Office
JP Β· JP
Patent Type
Patents
Current Assignee / Owner
ACCENTURE GLOBAL SERVICES LTD
Filing Date
2022-04-11
Publication Date
2026-06-17

AI Technical Summary

Technical Problem

Existing manufacturing systems struggle to effectively detect anomalies in batch production processes, particularly in complex semiconductor manufacturing, where inline quality control methods fail to identify defects until the final wafer testing phase, leading to significant quality loss and delivery issues.

Method used

Implementing Gaussian process regression models with a bathtub kernel function to analyze data from upstream quality control processes, allowing for predictive anomaly detection and adjustment of manufacturing settings based on target values derived from process quality inspection data.

Benefits of technology

Enables early identification of anomalies, reducing quality loss and improving manufacturing efficiency by adjusting processes proactively, even with limited training data, and providing reliable predictions of product characteristics.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 0007875009000006
    Figure 0007875009000006
  • Figure 0007875009000007
    Figure 0007875009000007
  • Figure 0007875009000008
    Figure 0007875009000008
Patent Text Reader

Abstract

To improve manufacturing efficiency by reducing defects.SOLUTION: In an abnormality detection system 300, a quality prediction unit obtains process quality inspection data for a first production lot, trains a Gaussian process regression model from product characteristic data within the lot after a final step, and uses a Gaussian process regression model using a bathtub kernel function to create a predicted distribution of the product characteristic data. The abnormality detection unit obtains process quality inspection data from a quality control process for a second production lot, identifies an abnormality in the second production lot by using the predicted distribution of the product characteristic data and the process quality inspection data from the second production lot, updates the Gaussian process regression model by using the process quality inspection data from the second production lot if no abnormality is detected in the lot, sets target values for one or more values in the process quality inspection data based on the predicted distribution of the product characteristic, and adjusts one or more manufacturing process settings based on the target values.SELECTED DRAWING: Figure 3
Need to check novelty before this filing date? Find Prior Art