OCD Metrology Up-Sampling via Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current optical critical dimension (OCD) metrology methods rely on sparse sampling schemes due to cycle time constraints and cost considerations, which can lead to insufficient monitoring and delayed detection of wafer production issues.
Innovation Solution
The development of machine learning models that up-sample sparse wafer maps to denser maps, allowing for improved OCD resolution and more extensive sampling of critical dimension parameters without increasing measurement cycle time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If sparse sampling schemes are used due to cycle time constraints, then measurement cycle time is reduced and productivity is improved, but measurement precision and reliability of OCD monitoring deteriorate
Solution Approach 1:
The patent creates a virtual copy of the sparse measurement data by training a machine learning model to generate dense wafer maps that replicate what full wafer mapping would produce. The model learns the relationship between sparse and dense measurements, then generates synthetic dense measurements from new sparse data, effectively copying the information content of extensive sampling without performing the actual extensive measurements.
Solution Approach 2:
The patent performs preliminary action by training the machine learning model in advance using a training set of wafer maps. This pre-training phase allows the model to learn the complex relationships between sparse measurements and actual wafer parameters before production use, so that during actual OCD monitoring, the model can quickly generate accurate dense maps from sparse measurements without time-consuming analysis.
2Measurement precision
If extensive sampling is performed to improve OCD resolution, then measurement precision and reliability are improved, but measurement cycle time increases and productivity decreases
Solution Approach 1:
The machine learning model generates virtual copies of extensive sampling data by predicting wafer parameters at locations where measurements were not actually taken. Instead of physically measuring every location on the wafer, the model copies the information from sparse measurement points to infer parameters across the entire wafer surface, achieving dense map resolution without dense measurement time.
Solution Approach 2:
The patent replaces the mechanical measurement system with an intelligent computational system. Rather than using the metrology tool to physically measure every location on the wafer (mechanical approach), the system uses a trained machine learning model to computationally predict parameters across the wafer based on sparse measurements, substituting computational intelligence for physical measurement at every point.
3Productivity
If sparse sampling is used to maintain fast cycle time, then productivity is maintained, but the ability to detect process issues and hardware performance problems deteriorates
Solution Approach 1:
The model copies the detection capability of extensive sampling by generating virtual measurements across the entire wafer. This allows the system to detect process issues and hardware performance problems with the same sensitivity as full wafer mapping would provide, while maintaining fast cycle times through sparse physical measurements combined with computational prediction.
Solution Approach 2:
The system enhances feedback capability by generating complete wafer maps that provide comprehensive information about wafer parameters across all locations. This dense information feedback enables better detection of process drift, equipment performance degradation, and other manufacturing issues that would be missed with sparse sampling, allowing for more reliable process control.
Data Source
AI summary
A system and methods for OCD metrology are provided including receiving training data for training an OCD machine learning (ML) model, the training data measured from multiple wafers and including multiple pairs of corresponding input and label datasets obtained from each respective wafer. The input dataset of each pair includes multiple scatterometric datasets, measured at multiple respective locations defined by a first map. The label dataset of each pair includes one or more critical dimension (CD) parameters of respective locations defined by a second map, the second map including at least one location not in the first map. The OCD ML model is then applied to a new set of scatterometric datasets, measured from locations of a new wafer, according to the first map, to generate predicted CD parameters of locations of the second map on the new wafer.