Wafer Position Reconstruction Using Assignment Uncertainty Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In semiconductor component packaging, the traceability of components to their original wafer position is lost after cutting, leading to a combinatorial assignment problem that is factorial in complexity, especially for PowerMOS components lacking memory, making unambiguous identification impossible.
Innovation Solution
A method using a machine learning system, trained on an assignment rule, to estimate the uncertainty of assignments by calculating a likelihood matrix from inaccuracies and covariance, allowing probabilistic quantification of uncertainty without additional metadata, enabling reliable position reconstruction and traceability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If unambiguous identifiers are stored in memory during wafer level test, then assignment between final test and wafer level test becomes possible, but this solution cannot be applied to PowerMOS components due to lack of memory
Solution Approach 1:
The patent replaces the mechanical/electronic storage system (memory) with a statistical/mathematical system (machine learning model). Instead of storing identifiers in component memory, the system uses test result patterns and probabilistic models to infer assignments, making the solution applicable to components without memory capability.
Solution Approach 2:
The patent introduces test results as an intermediary between the component and its identification. Rather than directly storing or reading identifiers from the component, the system uses test results performed on the component as a mediator to establish the assignment relationship through pattern matching and statistical analysis.
2Productivity
If rough matching is used between loose semiconductor components and wafer components, then some traceability is maintained, but several thousand components cannot be assigned to wafers
Solution Approach 1:
The patent transforms the assignment problem from a deterministic matching task to a probabilistic one by introducing likelihood scores. Instead of seeking exact matches, the system calculates probability distributions over possible assignments and selects the most likely configuration, enabling both speed and reasonable accuracy.
Solution Approach 2:
The patent performs partial assignment rather than requiring complete unambiguous assignment for all components. By accepting that some assignments may be uncertain and providing probabilistic information, the system can process much larger numbers of components efficiently while still providing useful traceability for the majority.
3Measurement precision
If combinatorial assignment methods are used to achieve accurate position reconstruction, then assignment accuracy improves, but the complexity becomes factorial
Solution Approach 1:
The patent segments the combinatorial optimization problem into independent or weakly coupled subproblems by using test result patterns as features. Instead of considering all possible assignments simultaneously, the machine learning model processes local patterns and relationships, breaking down the factorial complexity into manageable computational tasks.
Solution Approach 2:
The patent replaces the combinatorial optimization approach with a machine learning-based probabilistic model. This substitution transforms the computational problem from one requiring exhaustive search or complex optimization to one that can be solved through pattern recognition and statistical inference, dramatically reducing computational complexity.
Data Source
AI summary
A method for estimating an uncertainty of an assignment rule that assigns first variables from a first set of first variables to second variables from a second set of second variables. The method includes: ascertaining inaccuracies of a machine learning system trained with the assignment rule, wherein the inaccuracies are ascertained by means of a difference between the second variables predicted by the machine learning system depending on the first variables and the second variables assigned to the first variables according to the assignment rule; ascertaining a covariance matrix depending on the ascertained inaccuracies; ascertaining a likelihood matrix and normalizing the likelihood matrix by dividing the value of the likelihood matrix by the corresponding column sum.