Semiconductor Spectrum Prediction Model for Nanoscale Parameter Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Accurate measurement of nanoscale semiconductor structures is challenging due to their complexity and the need for non-destructive testing, especially in predicting parameters like recess height and critical dimensions, which existing methods struggle to address effectively.
Innovation Solution
A method and apparatus that utilize spectrum data to generate a prediction model by reducing dimensionality and training using grid maps, incorporating artificial intelligence models to predict parameters of interest in semiconductor substrates without destructive testing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If non-destructive testing methods are used to measure semiconductor structures, then sample damage is avoided, but measurement precision is insufficient for nanoscale structures
Solution Approach 1:
The patent introduces spectrum data as an intermediary that bridges non-destructive optical inspection and accurate parameter measurement. The spectrum data captures subtle optical characteristics that correlate with structural parameters, enabling precise predictions without direct physical measurement or sample damage.
Solution Approach 2:
The patent replaces traditional mechanical or direct optical measurement systems with an AI-based prediction system that processes spectrum data. This substitution allows for non-contact, non-destructive measurement while achieving nanoscale precision through machine learning models trained on spectral characteristics.
2Adaptability or versatility
If AI prediction models are trained using limited reference data, then measurement coverage is expanded, but prediction accuracy decreases
Solution Approach 1:
The patent transforms the training approach by changing how reference data is utilized. Instead of requiring extensive diverse reference data, the system changes the parameter extraction focus to capture essential spectral features that generalize across different samples, enabling accurate predictions with limited reference data through enhanced feature representation.
Solution Approach 2:
The patent creates a virtual copy of the measurement process through AI training. The model learns to predict parameter values by analyzing spectral patterns from limited reference samples, effectively copying the measurement outcomes without requiring physical measurement of every possible sample type, thus expanding coverage while maintaining accuracy.
3Device complexity
If spectrum data dimensionality is reduced to improve processing efficiency, then computational complexity decreases, but information loss increases
Solution Approach 1:
The patent extracts only the most informative components from the high-dimensional spectrum data through dimensionality reduction techniques. By identifying and extracting the key spectral features that correlate with structural parameters, the system reduces computational complexity while retaining essential information needed for accurate predictions.
Solution Approach 2:
The patent applies different levels of dimensionality reduction to different aspects of the spectrum data based on their local importance. Critical spectral regions that contain information about specific structural parameters are preserved with higher fidelity, while less informative regions are more aggressively reduced, optimizing the balance between processing efficiency and information retention.
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
Enhances prediction accuracy of 3D microstructures by leveraging gradient patterns in spectrum data, providing high consistency even with limited reference data, and enabling precise parameter estimation.
Implementation Method 1
irradiate incident polarized light onto a semiconductor substrate, obtain at least one polarization of transmission polarization and reflection polarization of the incident polarized light reflected from the semiconductor substrate
Data Source
AI summary
A method and an apparatus of generating a trained prediction model includes obtaining first spectrum data from a target structure of a semiconductor substrate, generating a first grid map for the semiconductor substrate by reducing dimension of the first spectrum data, generating a second grid map for the semiconductor substrate from the first spectrum data by using a prediction model for parameters of interest of the target structure, and training the prediction model based on the first grid map and the second grid map.