Wafer Position Reconstruction With Assignment Uncertainty Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The traceability of semiconductor components to their original positions on a wafer is lost during the packaging process, leading to a combinatorial problem with factorial complexity, making it difficult to assign semiconductor components correctly after they are cut and packaged, especially for PowerMOS devices lacking memory for unique identifiers.
Innovation Solution
A method using a machine learning system, such as decision trees or neural networks, to estimate the uncertainty of an assignment rule by determining inaccuracies, covariance, and likelihood matrices, allowing probabilistic assessment of assignment reliability without additional metadata, and enabling position reconstruction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a machine learning system is used to assign semiconductor components to wafer positions, then assignment accuracy is improved, but uncertainty quantification becomes necessary to ensure reliability
Solution Approach 1:
The patent implements feedback by using the likelihood matrix to continuously assess and quantify the uncertainty of assignment decisions. The machine learning model's predictions are evaluated against the covariance matrix of measurement errors, providing ongoing feedback about assignment reliability that can be used to adjust confidence levels or trigger re-verification of uncertain assignments.
Solution Approach 2:
The patent transforms the assignment problem from a deterministic classification task into a probabilistic assessment by introducing parameter changes: it calculates covariance matrices from measurement data, derives likelihood matrices from these covariances, and uses these parameters to quantify uncertainty. This transforms fixed assignment decisions into dynamic, confidence-weighted assignments.
2Loss of information
If unique identifiers are stored in semiconductor device memory, then traceability is improved, but this is not possible for PowerMOS devices lacking memory
Solution Approach 1:
The patent introduces an intermediary approach by using measurement data and statistical models as mediators between the semiconductor components and their wafer positions. Instead of direct identification through unique identifiers, the system uses indirect identification through characteristic measurements (electrical, optical, physical properties) that serve as intermediaries to establish traceability without requiring memory storage in the devices themselves.
Solution Approach 2:
The patent replaces the mechanical/electronic system of storing unique identifiers in device memory with a statistical/mathematical system based on measurement data analysis. The identification function is substituted from an active storage-retrieval mechanism to a passive pattern-matching and statistical inference mechanism that works across different device types regardless of their memory capabilities.
3Loss of information
If rough matching between loose components and wafer components is performed, then some traceability is maintained, but several thousand components remain unassignable
Solution Approach 1:
The patent applies dynamics by transforming the static rough matching process into a dynamic, multi-stage refinement process. The system starts with initial matching based on available data, then dynamically improves assignments by incorporating measurement data, calculating likelihoods, and iteratively refining the assignment rule until convergence or maximum confidence is achieved, thereby increasing the assignable component rate.
Solution Approach 2:
The patent implements preliminary action by performing comprehensive measurements and characterizations of both wafer-level components and loose components before the assignment process. This preliminary data collection includes electrical, optical, and physical property measurements that are stored and used to inform the subsequent statistical matching process, enabling more accurate assignments than would be possible with minimal preliminary data.
Data Source
Figure 1
Figure 2
AI summary
The invention relates to a method for estimating an uncertainty for an assignment rule that assigns first variables from a first set of first variables to second variables from a second set of second variables, comprising the following steps: Determining inaccuracies of a machine learning system trained with the assignment rule, wherein the inaccuracies are determined by the machine learning system, by way of a difference between the predicted second variables, depending on the first variables and second variables assigned to the first variables in accordance with the assignment rule. Determining a covariance matrix depending on the determined inaccuracies. Determining a likelihood matrix and normalizing the likelihood matrix by dividing the values of the likelihood matrix by the corresponding column sum.