ML Module Diagnostics for Lithographic Apparatus
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Diagnosing faults in complex lithographic apparatuses is challenging due to high-dimensional sensor data, limited datasets, non-normal signal distributions, and complex interactions between modules, leading to inaccurate identification of deviating modules and difficulties in prioritizing faults.
Innovation Solution
A method using machine learning to distinguish between sensor data from a faulty production apparatus and data from healthy apparatuses by training a model on feature sets from both sources, allowing for accurate classification and identification of faulty modules based on evaluated accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional two-step diagnostic process with aggregated statistics is used, then diagnostic process is simple, but identification accuracy of deviating modules is low
Solution Approach 1:
The patent replaces the traditional mechanical/statistical diagnostic approach with a machine learning-based system. A trained classification model automatically analyzes sensor data to identify deviating modules, substituting the manual two-step process (aggregated statistics followed by expert assessment) with an automated intelligent system that achieves higher identification accuracy while managing complexity through algorithmic processing.
Solution Approach 2:
The patent transforms the diagnostic approach by changing from analyzing aggregated statistical parameters to using raw or minimally processed sensor data fed into a machine learning model. The model learns complex patterns and relationships in the data that traditional aggregation methods miss, enabling more accurate identification of deviating modules through transformed parameter representations.
2Measurement precision
If machine learning model is trained on limited sensor data, then training speed is fast, but model accuracy is reduced
Solution Approach 1:
The patent applies preliminary action by collecting and storing sensor data from multiple production apparatuses in advance, creating a comprehensive training dataset before the actual diagnostic task. This pre-collected data pool enables the machine learning model to be trained on sufficient examples, improving accuracy without requiring large quantities of data during the actual diagnostic process.
Solution Approach 2:
The patent creates a universal training dataset by collecting sensor data from multiple different production apparatuses (both healthy and faulty). This multi-source data serves multiple purposes: training the classification model, validating its performance, and enabling it to generalize across different apparatus instances. The same data infrastructure supports both model training and diagnostic evaluation functions.
3Loss of information
If sensor data from multiple modules is analyzed together, then comprehensive diagnosis is achieved, but difficulty in identifying specific deviating module increases
Solution Approach 1:
The patent applies segmentation by training separate machine learning classification models for each individual module type rather than analyzing all modules together in a single model. Each model is specialized to identify deviations in its specific module type, reducing the complexity of inter-module interactions. The system segments the diagnostic task into module-specific sub-tasks while still achieving comprehensive diagnosis across the entire production apparatus.
4Loss of time
If traditional diagnostic approach is used, then expert assessment is required, but diagnostic time and cost increase
Solution Approach 1:
The patent replaces the mechanical process of expert assessment with an automated machine learning classification system. The trained model automatically analyzes sensor data, classifies modules as healthy or deviating, and outputs diagnostic results without requiring human expert intervention. This substitution dramatically reduces diagnostic time while maintaining or improving accuracy, and enables the process to be fully automated.
Solution Approach 2:
The patent enables the diagnostic system to serve itself by using the machine learning model to automatically perform the entire diagnostic task. The model takes sensor data as input, processes it through learned patterns, and generates diagnostic conclusions autonomously. This self-service capability eliminates the need for external expert assessment, reducing both time and cost while increasing automation.
Data Source
AI summary
A fault in a subject production apparatus which is suspected of being a deviating machine, is identified based on whether it is possible to train a machine learning model to distinguish between first sensor data derived from the subject production apparatus, and second sensor data derived from one or more other production apparatuses which are assumed to be behaving normally. Thus, the discriminative ability of the machine learning model is used as an indicator to discriminate between a faulty machine and the population of healthy machines.