Thickness Prediction Network for 3D NAND Layer Measurement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for measuring the thickness of silicon oxide and silicon nitride layers in 3D NAND flash memory manufacturing are destructive, making real-time and total inspection impossible, and are time and cost-intensive.
Innovation Solution
A thickness prediction network learning method that measures optical characteristics of semiconductor structures, generates spectrum measurement data, trains simulation and prediction networks using this data, and calculates layer thicknesses without destructive inspection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If transmission electron microscope (TEM) is used for measuring thickness, then measurement precision is improved, but productivity deteriorates due to destructive inspection requiring time-consuming sample preparation and inability to perform total inspection
Solution Approach 1:
The patent replaces the mechanical destructive inspection method (TEM) with an optical non-destructive measurement method. The optical measurement module uses light reflection characteristics to measure layer thickness without physically cutting or damaging the semiconductor structure, thereby enabling rapid total inspection while maintaining measurement precision through optical property analysis
Solution Approach 2:
The patent creates a virtual model (simulation network) that copies the physical semiconductor structure's optical characteristics. By training the simulation network with TEM data to generate spectrum simulation data, the system creates a digital twin that can be used for thickness prediction without requiring actual physical sampling, thus improving productivity while maintaining measurement accuracy
2Measurement precision
If transmission electron microscope (TEM) is used for measuring thickness, then measurement precision is improved, but loss of time increases due to destructive inspection process
Solution Approach 1:
The patent performs preliminary action by training the simulation network and thickness prediction network in advance using TEM data. Once trained, the system can perform rapid thickness measurements using optical spectra without requiring actual TEM sampling during production, thus reducing inspection time while maintaining precision through the pre-trained predictive models
Solution Approach 2:
The patent substitutes the time-consuming mechanical TEM inspection process with a rapid optical measurement system. The optical measurement module quickly captures reflection spectra, and the pre-trained neural networks instantly predict thickness values, eliminating the lengthy sample preparation and imaging time required by TEM while preserving measurement precision
3Measurement precision
If transmission electron microscope (TEM) is used for measuring thickness, then measurement precision is improved, but device complexity increases due to destructive inspection requirements
Solution Approach 1:
The patent replaces the complex mechanical TEM system with a simpler optical measurement system. The optical measurement module uses light sources and detectors to measure reflection spectra, eliminating the need for complex TEM hardware, sample preparation equipment, and vacuum systems while achieving comparable measurement precision through optical property analysis and neural network prediction
4Measurement precision
If transmission electron microscope (TEM) is used for measuring thickness, then measurement precision is improved, but manufacturing cost increases due to destructive inspection
Solution Approach 1:
The patent substitutes the expensive TEM system with a cost-effective optical measurement system. The optical measurement module uses inexpensive light sources and detectors compared to TEM hardware, and the neural networks process data computationally rather than requiring expensive physical analysis, thereby reducing manufacturing costs while maintaining measurement precision
Solution Approach 2:
The patent uses copying by creating virtual models through simulation networks trained on TEM data. Once the simulation network is trained, it can generate accurate thickness predictions from optical spectra without requiring actual TEM measurements, eliminating the high costs associated with TEM operation, maintenance, and sample preparation while preserving measurement precision
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
Enables real-time, non-destructive thickness measurement and prediction of semiconductor material layers, enhancing the efficiency and accuracy of semiconductor device manufacturing by reducing the need for time-consuming and costly TEM inspections.
Implementation Method 1
measuring spectrums of optical characteristics of a plurality of semiconductor structures
Data Source
AI summary
A thickness prediction network learning method includes measuring spectrums of optical characteristics of a plurality of semiconductor structures each including a substrate and first and second semiconductor material layers alternately stacked thereon to generate sets of spectrum measurement data, measuring thicknesses of the first and second semiconductor material layers to generate sets of thickness data, training a simulation network using the sets of spectrum measurement data and the sets of thickness data, generating sets of spectrum simulation data of spectrums of the optical characteristics of a plurality of virtual semiconductor structures based on thicknesses of first and second virtual semiconductor material layers using the simulation network, each of the first and second virtual semiconductor layers including the same material as the first and second semiconductor material layers, respectively; and training a thickness prediction network by using the sets of spectrum measurement data and the sets of spectrum simulation data.