SEM-Based Structure Height Estimation for Semiconductor Foreign Matter
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for measuring the height of foreign matter on semiconductor devices are inefficient and inaccurate, particularly in high-throughput semiconductor manufacturing, as they require labor-intensive AFM measurements and struggle to handle the varied shapes and compositions of unintentionally adhering foreign matter.
Innovation Solution
A structure estimation system using a learning device that performs advanced learning with teacher data from charged particle beam devices to estimate the three-dimensional information of foreign matter and its influence on semiconductor structures, utilizing neural networks and other learning models to process SEM data for accurate height estimation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If AFM measurement is used to measure the height of foreign matter, then measurement precision is improved, but productivity deteriorates due to labor-intensive manual measurement
Solution Approach 1:
The patent creates a virtual three-dimensional model (copy) of the foreign matter based on two-dimensional SEM images. The learning device generates a depth map that represents the three-dimensional shape without requiring physical contact measurement, thereby eliminating the need for manual AFM measurement while maintaining measurement capability through digital modeling
Solution Approach 2:
The patent replaces the mechanical AFM measurement system with a computational approach using SEM images and machine learning. Instead of using physical probes to measure height, the system uses image processing and neural networks to estimate three-dimensional information from two-dimensional images, substituting mechanical measurement with computational analysis
2Ease of operation
If a library is prepared in advance to store cross-sectional shape information, then ease of operation is improved, but adaptability deteriorates due to difficulty in generating libraries for various foreign matter shapes and compositions
Solution Approach 1:
The patent creates a universal learning device that can handle various foreign matter shapes and compositions through a single system. The neural network is trained on diverse data and can generalize to different foreign matter types without requiring separate libraries for each shape or composition, making the system both easy to operate and highly adaptable
Solution Approach 2:
The patent changes the approach from storing fixed library data to using a learning device that can adapt its parameters (weights and biases in neural network) based on input data. The system dynamically adjusts its estimation based on the specific characteristics of each foreign matter instance, allowing it to handle various shapes and compositions without pre-defined libraries
3Measurement precision
If cross-sectional SEM or AFM is used for height measurement, then measurement precision is improved, but device complexity increases and productivity decreases
Solution Approach 1:
The patent creates a virtual three-dimensional model (copy) of the foreign matter based on two-dimensional SEM images. The learning device generates a depth map that represents the three-dimensional shape without requiring physical contact measurement, thereby eliminating the need for manual AFM measurement while maintaining measurement capability through digital modeling
Solution Approach 2:
The patent merges the foreign matter inspection function with the existing SEM imaging system. Instead of using separate AFM or cross-sectional SEM devices, the system combines height estimation capabilities with the standard SEM, allowing the same device to perform both imaging and three-dimensional measurement functions
Data Source
AI summary
The present disclosure relates to a system and a non-transitory computer-readable medium for estimating the height of foreign matter, etc. adhering to a sample. In order to achieve the abovementioned purpose, proposed is a system, etc. in which data acquired by a charged particle beam device or features extracted from the data are input to a learning model, which is provided with, in an intermediate layer thereof, a parameter learned using teacher data having data acquired by the charged particle beam device or features extracted from the data as inputs and having the heights or depths of the structures of samples or of foreign matter on the samples as outputs, and height or depth information is output.