Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Detecting Defective Wafers Via Modular Networks

Yifeng Zhang, Bryan Baker, Shi Chen, Chao Zhang, Yu Huang, Qi Zhao, Sthitie Bom

TL;DR

This work tackles defect detection in wafer manufacturing by predicting hard-to-measure KQIs from readily available sensor data. It introduces a Neural Module Network-style modular network (MN) that decomposes KQI prediction into stage-specific prototypes and dynamically composed stage modules, trained end-to-end on time-series, stage-wise data. The key contributions are the explicit separation of base functions via implicit prototypes, the staged composition that mirrors manufacturing procedures, and a triad of learning signals that promote interpretability and generalizability across wafer products and environments. Experimental results on real-world data demonstrate improved generalization and interpretability compared with state-of-the-art baselines, highlighting the practical impact for fault diagnosis and rapid prototyping in semiconductor manufacturing.

Abstract

The growing availability of sensors within semiconductor manufacturing processes makes it feasible to detect defective wafers with data-driven models. Without directly measuring the quality of semiconductor devices, they capture the modalities between diverse sensor readings and can be used to predict key quality indicators (KQI, \textit{e.g.}, roughness, resistance) to detect faulty products, significantly reducing the capital and human cost in maintaining physical metrology steps. Nevertheless, existing models pay little attention to the correlations among different processes for diverse wafer products and commonly struggle with generalizability issues. To enable generic fault detection, in this work, we propose a modular network (MN) trained using time series stage-wise datasets that embodies the structure of the manufacturing process. It decomposes KQI prediction as a combination of stage modules to simulate compositional semiconductor manufacturing, universally enhancing faulty wafer detection among different wafer types and manufacturing processes. Extensive experiments demonstrate the usefulness of our approach, and shed light on how the compositional design provides an interpretable interface for more practical applications.

Detecting Defective Wafers Via Modular Networks

TL;DR

This work tackles defect detection in wafer manufacturing by predicting hard-to-measure KQIs from readily available sensor data. It introduces a Neural Module Network-style modular network (MN) that decomposes KQI prediction into stage-specific prototypes and dynamically composed stage modules, trained end-to-end on time-series, stage-wise data. The key contributions are the explicit separation of base functions via implicit prototypes, the staged composition that mirrors manufacturing procedures, and a triad of learning signals that promote interpretability and generalizability across wafer products and environments. Experimental results on real-world data demonstrate improved generalization and interpretability compared with state-of-the-art baselines, highlighting the practical impact for fault diagnosis and rapid prototyping in semiconductor manufacturing.

Abstract

The growing availability of sensors within semiconductor manufacturing processes makes it feasible to detect defective wafers with data-driven models. Without directly measuring the quality of semiconductor devices, they capture the modalities between diverse sensor readings and can be used to predict key quality indicators (KQI, \textit{e.g.}, roughness, resistance) to detect faulty products, significantly reducing the capital and human cost in maintaining physical metrology steps. Nevertheless, existing models pay little attention to the correlations among different processes for diverse wafer products and commonly struggle with generalizability issues. To enable generic fault detection, in this work, we propose a modular network (MN) trained using time series stage-wise datasets that embodies the structure of the manufacturing process. It decomposes KQI prediction as a combination of stage modules to simulate compositional semiconductor manufacturing, universally enhancing faulty wafer detection among different wafer types and manufacturing processes. Extensive experiments demonstrate the usefulness of our approach, and shed light on how the compositional design provides an interpretable interface for more practical applications.
Paper Structure (24 sections, 8 equations, 3 figures, 8 tables)

This paper contains 24 sections, 8 equations, 3 figures, 8 tables.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Key Quality Indicator Predictions
  4. Neural Module Networks
  5. Methodology
  6. Implicit Prototypes
  7. Stage Module
  8. Learning Prototypes and Stages
  9. Dataset Formatting
  10. Schema Specifications.
  11. Data Generation Paradigm.
  12. Handling Missing Entries.
  13. Experimental Results
  14. Implementation Details
  15. Experimental paradigm
  16. ...and 9 more sections

Figures (3)

  • Figure 1: Illustration of the wafer manufacturing process.
  • Figure 2: Overview of stage combinations in modeling wafer manufacturing. Our module-based network is composed of two major components, a collection of implicit prototypes which mimic base functions, and a set of stage modules which compose the manufacturing process. A weighted combination of implicit prototypes in each stage module is selected and recursively updated to mimic the physical functions performed within the manufacturing tools in the corresponding stage.
  • Figure 3: Illustration of Stage Module. (a) Stage Module Overview. (b) Select inter-prototype weights with mod selector $\boldsymbol{S}_j$. (c) Prototype Mappings at Different Mods.