Multi-Scale Test Result Verification for Fault Reliability Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods lack a quantitative means to reliably verify the reliability of product test results, relying heavily on technician experience and know-how, making it difficult to distinguish between manufacturing errors and test faults.
Innovation Solution
A method and apparatus utilizing neural networks, such as CNN and transformer networks, to analyze test result images at multiple scale levels, determining normality or abnormality through statistical hypothesis testing and feature extraction, enabling reliable verification of test reliability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If technician relies on know-how and experience to analyze reliability of test result, then analysis can be performed, but it is difficult to quantitatively guarantee reliability and accumulation of know-how is essential
Solution Approach 1:
The patent replaces the manual, experience-based analysis method with an automated image processing system using neural networks. The system converts test result data into images and applies deep learning models to objectively determine reliability, eliminating the need for technician know-how while providing quantitative guarantees through statistical hypothesis testing.
Solution Approach 2:
The patent introduces result images as an intermediary representation between raw test data and reliability determination. These images serve as a visual medium that neural networks can process, enabling automated analysis while preserving the essential information needed for reliability assessment.
2Reliability
If low yield is detected in product testing, then manufacturing errors can be identified, but it is difficult to distinguish whether low yield is due to manufacturing error or test fault
Solution Approach 1:
The patent segments the analysis into multiple scale levels (device level, board level, slot level, equipment level) to systematically examine test results at different granularities. This multi-scale approach enables the system to identify patterns that distinguish manufacturing errors from test faults by comparing consistency across different levels.
Solution Approach 2:
The patent implements a feedback mechanism where the neural network analyzes test results and provides reliability verification feedback. This feedback loop allows the system to continuously improve its ability to distinguish between manufacturing errors and test faults by learning from previous analyses and adjusting its determination criteria.
Data Source
AI summary
A method for verifying reliability of a test for products performed by test equipment includes: receiving result images generated from preprocessing of result data of the test, the result data of the test including labels for a plurality of scale levels and the received result images including first result images belonging to a first scale level of the plurality of scale levels and second result images belonging to a second scale level of the plurality of scale levels; making a first determination, from the first result images, whether the first scale level is normal or abnormal; making a second determination, from the second result images, whether the second scale level is normal or abnormal; and determining that no error occurred in the test in response to both the first scale level and the second scale level being determined to be normal; or determining that an error occurred in the test in response to at least one scale level being determined to be abnormal.