Optical Metrology ML for Non-Destructive Feature Characterization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for evaluating semiconductor wafers, such as x-SEM and CD-SAXS, are costly, time-consuming, and prone to operator error, with destructive techniques requiring multiple coupons and high error bars due to line-edge roughness, making process development and optimization challenging.
Innovation Solution
A supervised machine learning approach using optical metrology signals from scatterometry to predict feature characteristics normally obtained by destructive techniques like x-SEM and STEM, reducing the need for costly and time-consuming destructive measurements by training models with optical data from coupon samples.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If destructive techniques like x-SEM and STEM are used to obtain feature characteristics, then measurement precision is improved, but productivity deteriorates due to time-consuming processes and loss of substance
Solution Approach 1:
The patent creates virtual copies of destructive measurement results through machine learning models trained on paired datasets. The optical metrology system captures optical signals that are processed by ML models to generate predicted feature characteristics (profile, CD, contour) that replicate the accuracy of destructive techniques without physical destruction, enabling rapid iterative measurements on the same coupon
Solution Approach 2:
The patent replaces mechanical/physical destructive measurement systems (x-SEM, STEM requiring physical coupon destruction) with an optical-based measurement system combined with machine learning. The optical metrology system uses light interaction with features, and ML models substitute for the complex image analysis and reconstruction processes required in destructive techniques, eliminating the need for physical destruction while maintaining measurement capability
2Measurement precision
If multiple coupons are used for destructive measurements, then measurement precision is improved by reducing operator error, but loss of substance increases
Solution Approach 1:
The machine learning model creates a virtual replica of the destructive measurement process, allowing unlimited repetitions on the same optical data without consuming additional physical coupons. The model learns from training data the relationship between optical signals and feature characteristics, enabling accurate predictions without requiring multiple physical samples
Solution Approach 2:
Instead of discarding destroyed coupons after measurement, the patent recovers value by using the same physical coupon repeatedly for optical measurements. The non-destructive optical metrology allows the same coupon to be measured multiple times at different process stages, maximizing utilization of each physical sample while the ML model maintains measurement accuracy
3Manufacturing precision
If manual adjustment of etch process parameters is performed, then manufacturing precision can be achieved, but ease of operation deteriorates due to technical difficulty and time consumption
Solution Approach 1:
The patent implements a closed-loop feedback system where optical metrology measurements of actual feature characteristics are fed into machine learning models that predict the relationship between process parameters and outcomes. This feedback loop enables automated process tuning by comparing measured features against targets and adjusting parameters based on ML-predicted correlations, replacing manual trial-and-adjustment with data-driven optimization
Solution Approach 2:
The patent replaces manual expert adjustment of etch parameters with an automated machine learning system. The ML model learns optimal parameter settings from training data and automatically recommends or implements parameter adjustments to achieve target features, substituting human expertise and manual tuning with algorithmic optimization that reduces both time and technical barrier
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This method enables accurate and efficient prediction of feature profiles, critical dimensions, and contours, reducing costs and operator variability, while providing non-destructive and quick assessments, thus improving process development and monitoring in semiconductor fabrication.
Implementation Method 1
The optical metrology is a scatterometry technique. For example, the optical metrology output may include a reflectance spectra and/or ellipsometric output data.
Data Source
AI summary
A metrology system may include an optical metrology tool configured to produce an optical metrology output for one or more features on a processed substrate, and a metrology machine learning model that has been trained using a training set of (i) profiles, critical dimensions, and/or contours for a plurality of features, and (ii) optical metrology outputs for the plurality of features. The metrology machine learning model may be configured to: receive the optical metrology output from the optical metrology tool; and output the profile, critical dimension, and/or contour of the one or more features on the processed substrate.