A film thickness virtual measurement method, a model training method and a storage medium
By converting FDC data into a three-dimensional tensor and utilizing a virtual measurement network model, the problem of insufficient utilization of temporal information and metadata in traditional methods is solved, achieving higher accuracy in film thickness prediction and improving production efficiency and product yield.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SKYVERSE TECH CO LTD
- Filing Date
- 2026-01-28
- Publication Date
- 2026-06-05
AI Technical Summary
Existing technologies for wafer film thickness measurement in semiconductor manufacturing suffer from problems such as long measurement cycles, high costs, and inability to fully monitor the film thickness. Furthermore, traditional machine learning algorithms cannot fully utilize the timing information and metadata of FDC data, resulting in low accuracy of virtual film thickness measurement prediction.
By converting FDC process data from a two-dimensional matrix to a three-dimensional tensor, the data embedding function of the virtual measurement network model is used to compile metadata information into features related to film thickness prediction, and long-distance temporal dependencies are captured through an attention mechanism, so as to make full use of temporal information and metadata in FDC data.
This improves the accuracy of virtual measurement of semiconductor wafer film thickness, enables more accurate film thickness prediction, reduces reliance on physical measurement equipment, and enhances production efficiency and product yield.
Smart Images
Figure CN122153281A_ABST