Semiconductor Tool Data Formatting for ML-Based Defect Diagnosis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current semiconductor manufacturing systems face challenges in accessing and analyzing data from substrate processing tools due to varying data formats, inconsistent contextual data, and the difficulty in retrieving and reformatting data for holistic insights, which hinders productivity and defect diagnosis.
Innovation Solution
A data collection system that formats and organizes data from substrate processing tools into a common structured format, allowing for centralized storage and analysis using machine learning, utilizing a data diagnostic services computer to receive and store data in a table-like structure and generate output files for further analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If data from multiple substrate processing tools with different formats is collected and analyzed, then comprehensive data analysis and machine learning capabilities are improved, but data formatting and organization complexity increases
Solution Approach 1:
The patent implements a universal data collection system that receives multiple types of data from different substrate processing tools and formats them into a common structured format. The system uses a standardized data model with consistent field names and data types that can accommodate various input formats, enabling comprehensive data analysis while abstracting away the complexity of format conversion through a unified interface
Solution Approach 2:
The patent introduces an intermediary data collection system that acts as a mediator between diverse data sources and analysis tools. This system includes a data collector that receives data in various formats, a formatter that converts to a standard structure, and a storage mechanism that organizes data consistently. This intermediary layer handles all format conversion complexity, allowing analysis tools to work with uniformly formatted data without dealing with source-specific format variations
2Reliability
If contextual data is included with each data point for holistic insights, then defect diagnosis capability is improved, but data storage and retrieval complexity increases
Solution Approach 1:
The patent segments contextual data into distinct hierarchical fields including tool identifier, chamber identifier, substrate identifier, recipe identifier, and process parameters. Each data point is associated with its contextual information through a structured relationship model where contextual data is organized in a tree-like hierarchy. This segmentation allows comprehensive defect diagnosis by preserving all relevant context while organizing data in a manageable, queryable structure that reduces retrieval complexity
Solution Approach 2:
The patent adds a temporal dimension to data storage by including timestamps and sequence numbers with each data point, enabling chronological analysis of process parameters. It also adds a hierarchical dimension by organizing data from multiple levels (tool level, chamber level, substrate level, process step level) into a multi-dimensional structure. This dimensional organization allows comprehensive defect diagnosis through multi-level context while enabling efficient retrieval through hierarchical querying that reduces the complexity of accessing relevant contextual information
3Loss of time
If data is stored in a standardized format for easy retrieval, then data access efficiency is improved, but flexibility in handling different data types decreases
Solution Approach 1:
The patent uses parameter-based formatting where data is stored in a standardized structure with typed fields that can accommodate different data types through type declarations. The system supports optional fields and extensible data models that can be configured for different substrate processing tools. This parameter-driven approach allows efficient retrieval through consistent field access while maintaining flexibility to handle various data types and tool-specific parameters through configurable schemas
Solution Approach 2:
The patent implements a dynamic data model where the structure and fields of stored data can be configured and extended based on the specific tool and process being monitored. The system allows for runtime configuration of data collection parameters and supports adding new field types without restructuring the entire data storage system. This dynamic adaptability enables efficient standardized retrieval for existing data types while allowing flexible accommodation of new or tool-specific data formats through configurable extensions
Data Source
AI summary
A data collection system for semiconductor manufacturing includes: T substrate processing tools, where each of the T substrate processing tools includes: N processing chambers, where each of the N processing chambers includes a processing chamber controller configured to receive a plurality of different types of data during operating of the corresponding one of the N processing chambers, where the plurality of different types of data have different formats, where the processing chamber controller is further configured to format the plurality of different types of data into formatted data, and where T and N are integers; and a data diagnostic services computer configured to: receive and store the formatted data as categories in a common file having a table-like data structure including rows with contextual data; and in response to a request, generate an output file including a subset of the data from the common file.