Neural Network Correction of Interferometer Fringe Print-Through
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Interferometers in the semiconductor industry face challenges in accurately measuring workpiece surfaces due to fringe print-through (FPT) errors, which affect precision metrics like SFQR, despite existing techniques failing to adequately address these issues.
Innovation Solution
Employing a neural network, specifically a generative adversarial network (GAN), to learn and remove fringe errors from interferogram data by training on synthesized image pairs with and without FPT, enabling correction of phase errors in workpiece surfaces.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional interferometry techniques are used to measure workpiece surfaces, then measurement capability is provided, but fringe print-through errors degrade measurement precision
Solution Approach 1:
A neural network model is introduced as an intermediary processing layer between the interferometer and the measurement analysis. The neural network receives interferogram data, learns to distinguish between true surface features and fringe print-through errors through training on synthesized data pairs, and outputs corrected surface maps that eliminate FPT artifacts while preserving measurement capability
Solution Approach 2:
Traditional mechanical and optical correction methods for fringe print-through are replaced with a computational neural network approach. The system substitutes physical correction mechanisms with a data-driven algorithm that processes interferogram images through convolutional neural network layers to remove FPT errors, achieving superior precision for metrics like SFQR
2Measurement precision
If fringe print-through correction is attempted using existing techniques, then some error reduction is achieved, but measurement precision remains insufficient
Solution Approach 1:
The neural network is pre-trained using synthesized image pairs where interferograms with known fringe print-through errors are paired with their corresponding error-free versions. This preliminary training enables the model to learn the characteristic patterns of FPT errors and develop accurate correction mappings before being applied to actual measurement data, ensuring reliable and precise correction
Solution Approach 2:
The correction process incorporates feedback mechanisms where the neural network continuously refines its predictions based on the relationship between interferogram features and true surface topology. The model uses feedback from training data to adjust its internal representations, ensuring that correction maintains reliability across varying measurement conditions and wafer profiles
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
The GAN-based method significantly improves measurement precision by effectively reducing FPT noise, enhancing SFQR and SBIR metrics, demonstrating robustness across varying environmental conditions and wafer profiles.
Implementation Method 1
Interferometers are used as metrology tools, where workpiece surface information is encoded in the phase of interferogram
Data Source
AI summary
A workpiece is disposed on a stage in an interferometer. Measurements are taken of the workpiece using the interferometer. An image of a surface of the workpiece is generated from the measurements. Phase error is removed from the image with a neural network operated using the processor. The neural network can be a generative adversarial network.