Wafer Test Data Sampling for Faster Anomaly Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for reducing wafer test data sets do not account for the measurement properties of semiconductor chips, making them unsuitable for wafer test data, which hinders effective anomaly detection and increases processing time and costs.
Innovation Solution
A method that defines sub-areas on a wafer, selects a reduced set of measured variables from these sub-areas, and outputs them as a training data set for a machine learning algorithm, ensuring all areas are considered equally and allowing for efficient anomaly detection with reduced data processing time and costs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the complete set of measured variables from all semiconductor chips is used for training, then the quality and completeness of the training data set is improved, but the data processing time and computational costs increase significantly
Solution Approach 1:
The wafer is divided into multiple sub-areas, and from each sub-area, only a subset of semiconductor chips is selected for measurement. This segmentation approach reduces the total number of measured variables while ensuring that all regions of the wafer are represented, thereby maintaining data quality while reducing processing time and costs.
2Quantity of substance
If all semiconductor chips on the wafer are measured, then comprehensive coverage of the wafer is achieved, but the number of measured variables becomes excessively large for efficient processing
Solution Approach 1:
The wafer surface is segmented into multiple sub-areas, and a representative subset of chips is selected from each sub-area. This ensures comprehensive spatial coverage of the wafer while limiting the total number of measured variables to a manageable size for efficient processing.
Solution Approach 2:
Different sub-areas of the wafer are treated independently with localized sampling. Each sub-area contributes a subset of measured variables that reflects the local characteristics of that region, ensuring that the overall data set maintains representative quality without requiring measurement of all chips.
3Productivity
If a reduced data set is used to decrease processing time, then data processing efficiency is improved, but the risk of missing anomalies increases
Solution Approach 1:
By dividing the wafer into sub-areas and sampling from each, the method ensures that anomalies in any region can be detected while keeping the overall data volume reduced. The segmented approach maintains statistical representativeness across the entire wafer surface.
Solution Approach 2:
Instead of measuring all chips (excessive action), the method measures a carefully selected partial set of chips from each sub-area. This partial measurement approach is sufficient to detect anomalies while significantly reducing processing requirements.
Data Source
AI summary
A method for providing wafer test data of at least one wafer with semiconductor chips, in particular for providing a training data set for a machine learning algorithm for anomaly detection, comprising receiving a set of measured variables from the semiconductor chips of the wafer, defining sub-areas on the wafer, each of which comprises a plurality of semiconductor chips of the wafer for which measured variables have been received, and outputting a reduced set of measured variables compared to the received set of measured variables, which comprises only the measured variables of subsets of the semiconductor chips of the respective sub-areas.