3D Structure Prediction Using Image and Spectrum Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Scanning electron microscopes (SEM) struggle to accurately predict the three-dimensional (3D) structure of semiconductor components due to a lack of depth information in the image data obtained during the manufacturing process.
Innovation Solution
A 3D structure prediction device utilizing a combination of a first encoder for 2D image data and a second encoder for spectrum data, along with a feature vector generator employing deep learning models to generate a feature vector corresponding to the 3D structure, integrating sensors like SEM, X-ray, AFM, or TEM for image data and OCD for spectrum data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If SEM is used to obtain surface information of wafer, then surface information can be obtained, but depth information is lacking making it difficult to accurately predict 3D structure
Solution Approach 1:
The patent combines multiple types of data (2D image data from SEM and spectrum data from OCD measurement) into a unified feature representation. The encoder network processes both data types simultaneously, merging their complementary information to reconstruct 3D structure, thereby compensating for the depth information deficiency in SEM images alone.
Solution Approach 2:
The patent introduces spectrum data obtained through optical critical dimension (OCD) measurement as an intermediary that bridges the gap between 2D SEM images and 3D structure. This intermediary data source provides depth-related spectral information that, when processed through the deep learning model, enables accurate 3D reconstruction without directly measuring depth.
2Measurement precision
If multiple types of sensors are used to gather both image data and spectrum data, then 3D structure prediction accuracy is improved, but device complexity increases
Solution Approach 1:
The patent designs a unified deep learning framework that processes multiple data types (2D images and spectrum data) through a single integrated architecture. The encoder network accepts both image and spectrum inputs and generates a unified feature map, allowing one system to perform multiple measurement functions and achieve 3D reconstruction without requiring separate dedicated systems for each data type.
Solution Approach 2:
The patent transforms spectrum data (which has different dimensional characteristics than image data) into a comparable feature representation through the encoder network. By converting both data types into unified feature maps with consistent dimensional properties, the system can process diverse inputs through the same neural network architecture, reducing the complexity of handling multiple data formats.
Data Source
AI summary
An example three-dimensional (3D) structure prediction device includes a first encoder, a second encoder, and a feature vector generator. The first encoder generates a first feature map based on two-dimensional (2D) image data corresponding to a target object. The second encoder generates a second feature map based on spectrum data corresponding to the target object. The feature vector generator receives the first feature map and the second feature map, and output a feature vector corresponding to a 3D structure of the target object based on a deep machine learning model.