A Python application to flag outliers in very high...
summary
A Python app flags outliers in thousand‑dimensional wafer sensor vectors by projecting to low dimensions with UMAP then clustering with HDBSCAN, automatically finding three distinct outlier clusters that map to specific field locations. Removing those points raises model R² substantially, replaces slow SEM inspection with a fast in‑pipeline filter, and runs on production scatterometer data to improve overlay measurement accuracy in practice.
AI-written summary · about
chapters (10)
- 0:00Outlier detection in sensor data
- 3:44Semiconductor manufacturing challenges
- 6:50Photo lithography technology
- 10:23Defining outliers in data
- 13:14Curse of dimensionality
- 15:24Dimensionality reduction methods
- 17:08UMAP for distances
- 19:12Clustering with HDBSCAN
- 22:02Method on wafer sensor data
- 23:27Conclusion
related talks
25mThe Art of Clustering: The Good, The Bad and The Beautiful
24mThe TL;DR of EDA
34mDimensionally reducing data squeezing out the good stuff
37mSpatial Data Analysis With Python
38mData Mapping for Data Exploration
36mExtracting Relevant Metrics with Spectral Clustering
47mAssessing the quality of a clustering
30mWords in Space