Research map / Label-efficient and robust learning / Recognition and long-tail
Human Interpretable Virtual Metrology in the Semiconductor Manufacturing
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Doctoral consortium paper proposing a multi-output virtual metrology system for PVD metal layers with a model-agnostic explanation method based on projective operators (ProjSe) and a stability index.
Problem
Virtual metrology in semiconductor manufacturing needs accurate, explainable multi-output prediction of metal-layer physical properties (thickness, resistance at 17 points) from equipment sensor process parameters, which current methods lack.
Approach
Extends the projective selection (ProjSe) variable-selection theory to a model-agnostic explanation method for vector-valued (multi-output) predictions, plus a stability index for explanations, and compares it to SHAP and TreeInterpreter.
Key results
- The proposed projective-operator explanation method outperforms SHAP and TreeInterpreter in computation time.
- The introduced stability index and correlation of the explanations are comparable to SHAP and TreeInterpreter.
- ProjSe-based feature selection results aligned with industry experts' experience and the underlying physics of the process.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use the projective-operator explanatory method instead of SHAP or TreeInterpreter | Projective-operator explanatory method · Real-world datasets | computation time | Helped |
| Use the projective-operator explanatory method instead of SHAP or TreeInterpreter | Projective-operator explanatory method · Real-world datasets | stability index | No clear effect |
| Use the projective-operator explanatory method instead of SHAP or TreeInterpreter | Projective-operator explanatory method · Real-world datasets | correlation | No clear effect |
| Select and rank process parameters with ProjSe | Virtual metrology system for PVD metal-layer properties · Semiconductor manufacturer dataset | feature-selection alignment with expert knowledge | Helped |
Surprising findings
- ProjSe performs well for around 20 outputs but struggles with fewer outputs.
Limitations
- Explanation method is preliminary; user studies and public-dataset tests are planned but not done.
- ProjSe struggles with feature selection when the number of outputs is small.
- Data is proprietary industrial data from a single manufacturer.
Experimental setup
Datasets: industrial PVD process data from a semiconductor manufacturer (Infineon)
Benchmarks: industrial PVD virtual metrology dataset
Metrics: computation time stability index correlation
Baselines: SHAP TreeInterpreter tree-model-based feature selection
Builds on: ProjSe SHAP LIME
Closest papers by meaning
- Adaptive Experimental Design to Accelerate Scientific Discovery and Engineering Design (AAAI 2025 · similarity 0.80)
- PAPM: A Physics-aware Proxy Model for Process Systems (ICML 2024 · similarity 0.79)
- ILILT: Implicit Learning of Inverse Lithography Technologies (ICML 2024 · similarity 0.78)
- CViT: Continuous Vision Transformer for Operator Learning (ICLR 2025 · similarity 0.77)
- Neural Interpretable PDEs: Harmonizing Fourier Insights with Attention for Scalable and Interpretable Physics Discovery (ICML 2025 · similarity 0.77)
- Intelligent OPC Engineer Assistant for Semiconductor Manufacturing (AAAI 2025 · similarity 0.76)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.