Virtual Metrology Data Normalization for Semiconductor Fabrication
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing virtual metrology methods in semiconductor fabrication face constraints due to limited available training data and the need for accurate, quick predictions, which are hindered by the lack of data sharing and compatibility between different semiconductor processes.
Innovation Solution
A method and system that normalize manufacturing data sets to reduce statistical differences, enabling data sharing and creating a compatible database for predicting semiconductor process performance, allowing for quicker and more accurate virtual metrology forecasts by transforming data to align with a selected process, thereby controlling semiconductor processing machines effectively.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If virtual metrology uses data from multiple different semiconductor processes, then the amount of available training data increases, but statistical differences and incompatibility between data sets increase
Solution Approach 1:
The patent transforms process data by changing statistical parameters (mean, standard deviation) to normalize data from different semiconductor processes. This allows data from multiple processes to be combined into a unified training set while maintaining statistical compatibility, thereby increasing the quantity of usable training data without sacrificing reliability.
Solution Approach 2:
The patent introduces an intermediate transformation layer that mediates between raw process data and the virtual metrology model. This intermediary step applies statistical normalization and transformation functions that make data from different processes compatible, enabling data sharing across process boundaries while maintaining model accuracy.
2Reliability
If virtual metrology is implemented with process-specific data, then data compatibility is maintained, but the amount of available training data is limited
Solution Approach 1:
The patent creates a universal data transformation framework that can handle multiple semiconductor processes through a single unified approach. By developing transformation functions that work across different process types (etch, deposition, lithography, etc.), the system achieves universality, allowing one virtual metrology model to benefit from data across all processes rather than being limited to process-specific data.
3Measurement precision
If extensive training data is collected for accurate virtual metrology predictions, then prediction accuracy improves, but the time and resources required for data collection and processing increase
Solution Approach 1:
The patent creates transformed copies of process data that can be reused across different virtual metrology applications. By normalizing and transforming data once, the same transformed dataset can serve multiple prediction models and processes, eliminating the need to collect and process extensive raw data repeatedly for each application.
Data Source
AI summary
The present disclosure provides a method of fabricating a semiconductor device. The method includes collecting a plurality of manufacturing data sets from a plurality of semiconductor processes, respectively. The method includes normalizing each of the manufacturing data sets in a manner so that statistical differences among the manufacturing data sets are reduced. The method includes establishing a database that includes the normalized manufacturing data sets. The method includes normalizing the database in a manner so that the manufacturing data sets in the normalized database are statistically compatible with a selected one of the manufacturing data sets. The method includes predicting performance of a selected one of the semiconductor processes by using the normalized database. The selected semiconductor process corresponds to the selected manufacturing data set. The method includes controlling a semiconductor processing machine in response to the predicted performance.