Self-Attention Driven Transformer for Reliable Semiconductor Map Defect Diagnosis System

Authors

  • E. Prashanthi Author
  • Polaka Durga Sai Lakshmi Author
  • Nagareddy Sushma Author
  • Yellu Nikhitha Author

DOI:

https://doi.org/10.62643/ijerst.2026.v22.n2.pp31-39

Keywords:

Wafer Map Defect Diagnosis, Semiconductor Manufacturing, WM811K, Feature Extraction, Yield Optimization

Abstract

More than 70% of yield loss in semiconductor fabrication has been attributed to spatially correlated wafer defects, while the rapid growth of wafer-level data has made traditional inspection methods increasingly inefficient. With thousands of wafer maps generated daily per production line, accurate and automated defect diagnosis has become essential for maintaining yield and reliability. This study aims to develop a robust wafer map defect diagnosis system capable of classifying patterns such as center, donut, edge-local, edge-ring, local, near-full, none, random, and scratch. Accurate identification of these patterns supports faster root-cause analysis, process optimization, and reduced manufacturing costs. Manual inspection, which depends on human expertise and visual interpretation, is timeconsuming, subjective, and difficult to scale, often failing to detect subtle or complex spatial patterns. To address these limitations, a self-attention-driven transformer-based Semiconductor Map Defect Diagnosis (SMDD) framework is proposed using the WaferMap811K (WM811K) dataset. Initially, wafer maps are preprocessed through resizing and normalization to ensure consistent input representation. A transformer-based feature extraction module is then utilized to capture long-range spatial dependencies and global defect characteristics. For classification, multiple models, including Decision Tree Classifier (DTC), Hidden Markov Model (HMM), and Gradient Boosting (GB), are employed alongside a proposed Quantum Conditional Random Field (QCRF). The QCRF effectively models complex spatial relationships between defect regions, enhancing classification robustness. The proposed system achieves accurate and reliable defect pattern classification, demonstrating improved diagnostic performance for semiconductor manufacturing

Downloads

Published

01-04-2026

How to Cite

Self-Attention Driven Transformer for Reliable Semiconductor Map Defect Diagnosis System. (2026). International Journal of Engineering Research and Science & Technology, 22(2), 31-39. https://doi.org/10.62643/ijerst.2026.v22.n2.pp31-39