Semiconductor Equipment Monitoring Using Fleet Baseline and Probabilistic Model
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current equipment monitoring strategies in semiconductor manufacturing are more sensitive to dedicated monitoring artifacts rather than production artifacts, leading to delayed detection of equipment excursions and increased potential for false positives, compromising product quality.
Innovation Solution
A method and system that monitor equipment by applying aggregate statistics of a group of tools relative to an individual piece of equipment, using a candidate baseline and probabilistic model to identify outlier equipment with high signal-to-noise, allowing for early detection and correction of issues and reducing false positives through increased sampling and confidence scoring.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional monitoring methods focus on improving measurement uncertainty on dedicated monitoring artifacts, then measurement precision is improved, but sensitivity to production artifacts deteriorates and false positives increase
Solution Approach 1:
The patent segments the monitoring approach by separating dedicated monitoring artifacts from production artifacts, then uses a fleet-based statistical model that processes data from multiple tools measuring production artifacts. This segmentation allows each tool to maintain measurement precision on its own artifacts while the aggregate fleet data provides reliability and sensitivity to production variations.
Solution Approach 2:
The patent merges data from multiple tools in the fleet to create a group candidate baseline. By combining measurements from several tools on production artifacts, the system achieves both the measurement precision of individual tools and the reliability/sensitivity of aggregate fleet performance, resolving the contradiction between precision and reliability.
2Measurement precision
If traditional monitoring methods use dedicated monitoring artifacts, then measurement control is improved, but detection of equipment excursions during steady state operation is delayed
Solution Approach 1:
The patent establishes a group candidate baseline from fleet data before monitoring individual tool excursions. This preliminary aggregation of production artifact measurements creates a reference that enables early detection of deviations, allowing the system to identify equipment excursions before they affect product quality.
Solution Approach 2:
The system continuously updates the group candidate baseline as new fleet data arrives, creating a dynamic feedback mechanism. This allows the monitoring system to adapt to steady-state fleet performance and immediately detect when an individual tool deviates from the established baseline, reducing detection delay while maintaining measurement control.
3Speed
If increased sampling is used to improve early detection, then detection speed is improved, but false positives increase
Solution Approach 1:
The patent combines measurements from multiple tools in the fleet to establish the group candidate baseline. This merging of data sources increases the statistical significance of each measurement, allowing for faster detection speeds through increased sampling while the aggregate nature of the baseline reduces false positives by providing a more robust reference.
Solution Approach 2:
The system changes the parameter of sampling frequency and uses probabilistic models to evaluate excursions against the group baseline. By adjusting these parameters and using statistical evaluation, the system achieves high detection speed through frequent sampling while maintaining reliability by filtering false positives through probabilistic assessment of deviations.
Data Source
AI summary
Systems and methods are provided for monitoring equipment in a semiconductor manufacturing facility. A processing device maintains baseline measurement record generated from measurements performed by a plurality of tools over a first time range. A measurement record is received, the measurement record corresponding to a measurement performed on a wafer, the wafer having been processed by a first tool of the plurality of tools. The measurement record is stored in a drift candidate table corresponding to a second time range. A determination is made as to whether the measurement record satisfy a set of excursion rules in view of the drift candidate table, and, in response to the determination, the measurement record is stored in an alert table.