Time-Series Abnormality Detection Using Adaptive Evaluation Distributions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for detecting abnormalities in time-series data during semiconductor substrate manufacturing lack accuracy due to unsuitable threshold values and failure to consider the passage of time, leading to inadequate detection of anomalies.
Innovation Solution
A data processing method that utilizes evaluation value distributions to assess and update the abnormality judgment, taking into account recent trends in time-series data by standardizing evaluation values and updating distributions accordingly.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional threshold-based abnormality judgment is used, then the detection process is simple, but the accuracy of abnormality detection is insufficient
Solution Approach 1:
The patent applies dynamics by making the threshold values dynamic rather than static. The threshold values are automatically updated based on the distribution of evaluation values from time-series data, allowing the system to adapt to changing conditions in semiconductor manufacturing processes. This resolves the contradiction by improving detection accuracy through adaptive thresholds while maintaining automated processing that doesn't significantly increase operational complexity.
Solution Approach 2:
The patent changes the parameter of threshold values from fixed constants to dynamic values derived from data distributions. By calculating threshold values based on the statistical distribution of evaluation values (e.g., using standard deviations from mean), the system improves abnormality detection accuracy. The parameter change is automated through data processing, balancing improved precision with acceptable processing complexity.
2Measurement precision
If fixed threshold values are used for abnormality judgment, then the judgment process is straightforward, but the detection accuracy deteriorates over time as process conditions change
Solution Approach 1:
The patent implements feedback by continuously updating the distribution of evaluation values based on incoming time-series data. The threshold values are recalculated using the updated distribution, creating a closed-loop system that adapts to process changes over time. This feedback mechanism maintains high detection accuracy without requiring manual intervention, and the automated nature minimizes the effective time loss by operating in the background.
3Measurement precision
If comprehensive time-series data analysis is performed to improve detection accuracy, then the abnormality detection becomes more accurate, but the processing time increases
Solution Approach 1:
The patent extracts only the essential statistical features from comprehensive time-series data - specifically the distribution characteristics (mean, standard deviation, percentiles) of evaluation values. Rather than analyzing all raw data points individually, the system extracts these key distribution parameters to calculate threshold values. This extraction approach maintains high detection accuracy while significantly reducing processing time and computational burden.
Data Source
AI summary
A data processing method that processes a plurality of unit processing data (each unit processing data include plural types of time-series data) includes an evaluation value distribution utilization step, in which processing that uses evaluation value distributions showing degrees of each value of evaluation values obtained by evaluating each time-series datum is carried out (for example, a step in which each time-series datum is compared with reference data and scoring that quantifies results obtained thereby as the evaluation values is carried out, and a step in which judgment of abnormality degrees is carried out using the evaluation value distributions based on results of the scoring); and an evaluation value distribution update step, in which the evaluation value distributions are updated.