AI/ML for Yield Learning and Test Optimization in Semiconductor Manufacturing: A Review of Adaptive Testing, Smarter Limits, and Diagnostic Preservation
American Journal of Technology · 19.05.2026 · 5 (6) · 65-90
DOI: 10.58425/ajt.v5i6.544 ↗Kurz gefasst
Fachartikel in der Zeitschrift American Journal of Technology, online erschienen am 19.05.2026. Verfasst von Sriram Nithin Gopal Lanka. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Aim
This review aimed to consolidate fragmented evidence on AI/ML-driven semiconductor test optimization into a single integrated synthesis. It also sought to evaluate the readiness of these methods for high-volume production deployment, focusing on adaptive testing, ML-based outlier detection, yield learning, and diagnostic preservation.
Methods
The study systematically reviewed 27 primary sources published between 2019 and 2025, including IEEE conference papers, journal articles, and industry technical publications. The review also introduced an ML Deployment Readiness Framework to evaluate reviewed methods across seven dimensions, including data requirements, evaluation metrics, uncertainty quantification, explainability, generalizability, concept drift treatment, and deployment evidence.
Results
The findings showed that ML-based methods outperform traditional approaches such as DPAT in outlier detection, yield prediction, and test time reduction. Quantitative results included superior spatial outlier detection using GPR, ensemble GPR achieving 15.6% improved robustness and 98.6% computation-time reduction, PSO-XGBoost achieving 96.4% accuracy and 91.4% AUC, conformal prediction delivering approximately 90% Vmin coverage guarantees, adaptive testing reducing test time by approximately 25%, and SHAP-based explainability reducing yield loss by 21.7%. The review also identified recurring methodological weaknesses, including reliance on proprietary datasets, limited external validation, inadequate treatment of class imbalance, lack of deployment evidence, and insufficient handling of concept drift.
Conclusion
The review concludes that semiconductor testing is evolving toward adaptive, explainable, and data-driven AI/ML systems with strong potential for high-volume manufacturing deployment. Recommendation: The study recommends the development of standardized public benchmark datasets for ML-based outlier detection and controlled studies that quantify diagnostic information loss before and after ML-driven test reduction to support trustworthy and scalable adaptive semiconductor testing.
Zitieren
APA
Lanka, S. N. G. (2026). AI/ML for Yield Learning and Test Optimization in Semiconductor Manufacturing: A Review of Adaptive Testing, Smarter Limits, and Diagnostic Preservation. American Journal of Technology, 5(6), 65-90. https://doi.org/10.58425/ajt.v5i6.544
Vancouver
Lanka SNG. AI/ML for Yield Learning and Test Optimization in Semiconductor Manufacturing: A Review of Adaptive Testing, Smarter Limits, and Diagnostic Preservation. American Journal of Technology. 2026;5(6):65-90. doi:10.58425/ajt.v5i6.544
BibTeX
@article{Lanka2026,
author = {Lanka, Sriram Nithin Gopal},
title = {AI/ML for Yield Learning and Test Optimization in Semiconductor Manufacturing: A Review of Adaptive Testing, Smarter Limits, and Diagnostic Preservation},
journal = {American Journal of Technology},
year = {2026},
volume = {5},
number = {6},
pages = {65--90},
doi = {10.58425/ajt.v5i6.544}
}
EUVIMED