Error Cause Estimation Using Adaptive Model Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods struggle to efficiently determine the cause of recipe errors in semiconductor inspection and measurement devices, particularly with increasing complexity and variety of recipes, leading to decreased operational efficiency due to time-consuming manual analysis.
Innovation Solution
An error cause estimation device and method utilizing a feature value generation unit, model database, model evaluation unit, model selection unit, and error prediction model generation unit to automatically generate and select suitable error prediction models, even without prior annotation, by using machine learning algorithms to analyze device and recipe data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual analysis of device internal data is used to specify error causes, then accuracy in identifying error causes can be achieved, but time consumption increases significantly leading to decreased operational efficiency
Solution Approach 1:
The patent replaces manual mechanical analysis with an automated machine learning system. The error cause estimation device uses trained models to automatically analyze device internal data and identify error causes, substituting the engineer's manual work with automated computational processes that maintain accuracy while dramatically reducing time consumption.
Solution Approach 2:
The patent implements preliminary action by pre-training error prediction models with historical error data before actual error occurrence. The system prepares multiple candidate error causes and their corresponding prediction models in advance, so when an error occurs, the system can immediately apply the appropriate model without waiting for manual analysis, thus reducing time loss while maintaining accurate identification.
2Adaptability or versatility
If the number of recipes and recipe setting items increases to handle miniaturization and diversification of semiconductors, then measurement capability is improved, but recipe creation becomes more complicated and error analysis time increases
Solution Approach 1:
The patent segments the complex recipe error analysis into multiple independent error prediction models, each dedicated to specific error types or recipe categories. Instead of creating one complex model for all recipes, the system divides error causes into distinct segments with specialized prediction models, making the overall system more manageable and adaptable to diverse semiconductor measurement scenarios.
Solution Approach 2:
The patent creates a universal error cause estimation framework that can handle multiple recipe types and error scenarios through a common system architecture. The error prediction models are designed to be universally applicable across different recipes and semiconductor types, reducing the need to create entirely new analysis procedures for each new recipe, thus managing complexity while maintaining versatility.
3Loss of information
If service engineers manually analyze device internal data to specify error causes, then detailed error understanding can be achieved, but operational rate of the device decreases due to time consumption
Solution Approach 1:
The patent implements self-service by enabling the error cause estimation device to automatically analyze its own operational data and identify errors without requiring external engineer intervention. The system uses its built-in machine learning models to autonomously process device internal data, maintain error understanding quality, and restore operational rate quickly, making the system self-sufficient in error diagnosis.
Data Source
AI summary
An error cause estimation device comprises a feature value generation unit for using data transmitted from the outside to generate feature values suitable for a machine learning model; a model database having a plurality of error prediction models, for determining whether an error has occurred using the feature values as input data; a model evaluation unit for evaluating the performance of an error prediction model by comparing a prediction result of the error prediction model and an actually measured error; a model selection unit for selecting from the model database an error prediction model for which an evaluation value calculated by the model evaluation unit is greater than or equal to a preset defined value; and an error prediction model generation unit for generating a new error prediction model with respect to the measured error when no corresponding error prediction model has been selected by the model selection unit.