Dice Testing Method Using Statistical Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional IC packaging tests face increased time and cost due to the manual selection of test items, leading to a lack of data for omitted items, which hinders analysis of potential IC issues.
Innovation Solution
A dice testing method utilizing a system that performs initial tests on a subset of ICs, partitions data, cleans it, models using partial least squares, and verifies equations to predict results for untested items, allowing for analysis of IC problems without full retesting.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If manual selection of test items is performed to reduce testing time and cost, then testing efficiency is improved, but test data completeness deteriorates
Solution Approach 1:
The patent creates virtual copies of test data through prediction models. When certain test items are not physically performed on all ICs, the system generates predicted test results by copying and adapting data from similar test cases and historical patterns, thereby maintaining data completeness without requiring actual physical testing of every item on every IC.
Solution Approach 2:
The patent replaces the mechanical system of physical testing with a computational prediction system. Instead of physically executing all test items on all ICs (mechanical testing), the system uses algorithms and statistical models to substitute and predict outcomes, reducing physical testing requirements while maintaining data availability for analysis.
2Loss of information
If all test items are performed on all ICs to ensure data completeness, then test data completeness is improved, but testing time and cost increase
Solution Approach 1:
The patent applies partial action by performing complete testing on a subset of ICs (excessive action on subset) and using that data to infer results for remaining ICs. Rather than performing all tests on all ICs (100% coverage), the system performs 100% testing on a representative sample and uses statistical inference to obtain sufficient data for the entire batch, reducing overall testing time while maintaining analytical capability.
3Ease of manufacture
If test items are omitted based on historical data to reduce testing cost, then testing cost is reduced, but analysis capability deteriorates
Solution Approach 1:
The patent implements feedback mechanisms where historical test data and results continuously inform and update the prediction models. The system learns from past testing outcomes, refining its ability to predict results for omitted test items. This feedback loop ensures that even when tests are omitted to reduce cost, the accumulated historical data and improved prediction accuracy maintain strong analysis capability for troubleshooting and process improvement.
Data Source
AI summary
A dice testing method is provided. The dice testing method is used to determine which test data of test items can be explained by test data of other test items based on statistical analysis. After the test items with the test data that can be explained by the test data of the other test items are found out, corresponding dices will not be tested for those test items.