Autoencoder Wafer Type Detection in Optical Film Thickness Measurement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing semiconductor manufacturing processes face challenges in accurately detecting and preventing the polishing of incorrect wafers, as different types of wafers require specific polishing recipes and endpoint detection conditions, leading to improper polishing when a mismatch occurs.
Innovation Solution
A method and apparatus that utilize an autoencoder trained with machine learning to analyze the spectrum of reflected light from wafers, comparing inspection spectrum data to indicator spectra to determine if a wafer is correctly matched to the polishing recipe, by calculating differences and setting a threshold value to identify incorrect wafers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If a polishing recipe is set for a specific wafer type, then proper polishing can be performed on that wafer type, but the system cannot detect when a wrong wafer type is processed
Solution Approach 1:
The system performs preliminary verification of wafer type by analyzing reflected light spectra before initiating the polishing process. The autoencoder model pre-trained on correct wafer spectra detects whether the input wafer spectrum matches the expected pattern, preventing improper polishing before it occurs.
Solution Approach 2:
The system continuously monitors the reflected light spectrum during polishing and compares it against the expected spectrum pattern using the autoencoder model. When deviations exceed a threshold, the system provides feedback to stop the polishing process, ensuring quality control through real-time spectral analysis.
2Measurement precision
If the polishing endpoint detection conditions are changed for different wafer types, then accurate endpoint detection can be achieved for each type, but the system complexity increases
Solution Approach 1:
The autoencoder model serves as a universal endpoint detection system that automatically adapts to different wafer types. Instead of requiring separate detection conditions for each wafer type, the model processes spectra from various wafer types through a unified neural network architecture, reducing system complexity while maintaining detection accuracy.
Solution Approach 2:
The system changes the detection approach from fixed threshold comparisons to dynamic neural network-based classification. The autoencoder learns optimal detection parameters automatically during training, transforming the detection conditions from static, manually-configured values to adaptive, data-driven parameters that adjust based on the input spectrum.
3Reliability
If spectrum analysis is performed during polishing, then wrong wafers can be detected, but the measurement time increases
Solution Approach 1:
The system performs partial spectrum analysis by focusing on key spectral features rather than analyzing the entire spectrum in detail. The autoencoder model is trained to recognize critical wavelength regions that differentiate correct from incorrect wafers, performing measurements only in these essential bands to reduce analysis time while maintaining verification reliability.
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
Effectively detects and prevents the polishing of incorrect wafers by accurately determining whether the wafer's spectrum matches the expected spectra, ensuring proper polishing and endpoint detection based on the established recipe.
Implementation Method 1
analyze a spectrum of light reflected from the wafer to detect a film thickness of the wafer
Data Source
AI summary
A method capable of detecting a wrong workpiece (e.g., wafer), which is not an object to be polished, is disclosed. The method includes: creating inspection spectrum data of reflected light from a workpiece before polishing of the workpiece or after beginning of polishing of the workpiece; inputting the inspection spectrum data to an autoencoder; calculating a difference between output data from the autoencoder and the inspection spectrum data; and determining that, when the difference is larger than a threshold value, the workpiece used to create the inspection spectrum data is a wrong workpiece which is not an object to be polished.