Machine Learning Method for Yield Management and Prediction in Compound Semiconductor Manufacturing
Speaker: Tzu-Hsuan Chen, Yield Engineer, PDF Solutions
Abstract:
Compound semiconductor manufacturing faces unique yield challenges driven by material defects originating in crystal growth and epitaxy, late-stage defect manifestation, and highly siloed production data across multi-site operations. These challenges are compounded by the presence of nuisance defects and inconsistent traceability, making early defect detection and root-cause analysis difficult with traditional methods. This presentation introduces a machine learning–driven approach to yield management and prediction, built on PDF Solutions’ end-to-end Exensio Yield Management Platform.
By unifying inline defect, metrology, electrical test, and equipment sensor data into a centralized, analysis-ready data model, the solution enables advanced analytics across the full manufacturing flow. A case study demonstrates how configurable ML workflows—leveraging defect filtering, feature reduction, and XGBoost-based classification—can accurately predict die- and wafer-level yield, distinguish critical “killer” defects from benign variability, and balance underkill versus overkill risk through adjustable probability thresholds. Interactive analytics and visualization tools provide actionable insights, improving early defect identification, substrate grading, and yield optimization. This approach highlights how integrated data infrastructure combined with flexible machine learning analytics can materially improve yield, traceability, and manufacturing decision-making in compound semiconductor production.
See this blog post for even more discussion on this topic: Predicting Yield Loss from Source: Machine Learning for Compound Semiconductor Manufacturing
Keywords: Machine Learning, Yield Management, Compound Semiconductor Manufacturing, Manufacturing Analytics
Download the presentation today!
By downloading this information, I agree that PDF Solutions, Inc. may use my information provided herein to contact me about its products and services, even if I have previously unsubscribed from such communications. For information on our privacy practices and commitment to protecting your privacy, please review our Privacy Policy.