Machine Learning Screening for Integrated Circuit Die Fallout
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
High-volume manufacturing of integrated circuit (IC) dies fails to effectively screen marginal dies during external loopback testing, leading to costly bench failures due to a gap in the screening process.
Innovation Solution
Implementing a machine learning (ML) model, specifically an XGBoost algorithm, trained on bench-failing IC dies data to predict bench fallout during manufacturing by utilizing gradient-boosted decision trees and combining HVM sort and class logged parameters, enabling accurate prediction of IC die failures when coupled to customer circuit boards.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If external loopback testing is used during high-volume manufacturing, then productivity is improved, but reliability deteriorates due to marginal dies not being effectively screened
Solution Approach 1:
The patent applies preliminary action by using machine learning models to predict bench fallout before the actual bench testing occurs. The ML models analyze sort and class test data to identify marginal dies in advance, allowing the system to take preventive actions (such as isolating or retesting specific dies) before they fail during customer bench testing, thus maintaining both high productivity and improved reliability.
2Productivity
If traditional sort and class testing is used, then manufacturing efficiency is maintained, but measurement precision deteriorates due to inability to predict bench failures
Solution Approach 1:
The patent introduces machine learning models as an intermediary between the traditional sort and class testing system and the final bench testing outcome. The ML models process the data from sort and class tests and provide predictive insights about bench fallout, enabling the system to maintain efficient manufacturing workflows while achieving precise failure prediction capability through this intermediate analytical layer.
3Reliability
If more comprehensive bench testing is implemented, then reliability is improved, but productivity decreases due to additional testing time
Solution Approach 1:
The patent applies preliminary action by predicting bench fallout before actual bench testing occurs. By using ML models to identify at-risk dies in advance based on sort and class test data, the system can focus comprehensive testing resources only on the marginal dies that are predicted to fail, rather than testing all dies exhaustively, thus maintaining both high reliability and productivity.
Solution Approach 2:
The patent applies local quality by tailoring the testing strategy to individual dies based on their predicted risk. Instead of applying uniform testing to all dies, the system identifies specific local cases (marginal dies with high predicted fallout) and applies enhanced testing or monitoring only to those, optimizing the balance between comprehensive screening and manufacturing efficiency.
Data Source
AI summary
A computing system includes a processor circuit configured to receive test data generated from testing integrated circuit dies in a test flow. The computing system includes a machine learning model that uses the test data generated from the test flow to predict bench results that are indicative of which ones of the integrated circuit dies fail to satisfy a manufacturing protocol when the integrated circuit dies are coupled to circuit boards.