Neural Network Estimation of Silicon Dioxide Breakdown Voltage
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for estimating the breakdown voltage of silicon dioxide films in semiconductor devices are time-consuming and destructive, as they require gradually increasing voltages to the point of destruction, leading to defects and reduced productivity.
Innovation Solution
A neural network model is used to estimate breakdown voltage by applying test voltages lower than the actual breakdown voltage and measuring current levels, generating a breakdown voltage estimation model based on filtered test data from first test dies to predict the breakdown voltage of second test dies without direct destruction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If acceleration stress evaluation is performed to detect defects in silicon dioxide film, then defect detection capability is improved, but testing time and sample quantity requirements increase
Solution Approach 1:
The patent applies preliminary action by using neural network models trained on accelerated stress test data to predict breakdown voltages before actual failure occurs. The model learns patterns from stress-test samples and can estimate breakdown characteristics of new samples without subjecting them to full acceleration stress, thereby reducing testing time while maintaining defect detection capability.
Solution Approach 2:
The patent uses copying by creating a virtual replica of the physical testing process through neural network models. Instead of physically testing every sample under acceleration stress, the system creates digital models that simulate and predict breakdown behavior, allowing rapid evaluation of multiple samples without consuming additional physical test time.
2Measurement precision
If breakdown voltage is measured by gradually increasing voltage to destruction point, then measurement precision is improved, but device damage and productivity loss increase
Solution Approach 1:
The patent introduces an intermediary by using neural network models as a mediator between the applied test voltage and the breakdown voltage measurement. The model processes current-voltage characteristics obtained at safe operating points and infers the breakdown voltage without requiring direct measurement at the destruction point, thus avoiding device damage while maintaining measurement accuracy.
Solution Approach 2:
The patent replaces the mechanical/voltage-based direct measurement system with an information-processing system. Instead of physically pushing devices to breakdown to measure the voltage, the system uses neural networks to analyze electrical characteristics and predict breakdown voltage, substituting physical testing with computational analysis to avoid device destruction.
3Reliability
If more samples are tested to ensure statistical accuracy, then measurement reliability is improved, but testing cost and time increase
Solution Approach 1:
The patent applies parameter changes by transforming the testing approach from physical sample quantity to data quality. The neural network model analyzes the quality and characteristics of test data (current-voltage curves, leakage current patterns) to extract reliable breakdown voltage predictions, reducing the need for large numbers of physical samples while maintaining statistical accuracy through improved data utilization.
Data Source
AI summary
A method of estimating a breakdown voltage of a silicon dioxide film includes; generating breakdown voltage information associated with first test dies selected from among test dies, generating a breakdown voltage estimation model by updating a parameter of a neural network model based on the breakdown voltage information, applying test voltages to second test die selected from among the test dies and distinct from the first dies and receiving currents levels for current generated by the second test dies in response to the test voltages, wherein the test voltages have respective levels lower than levels of breakdown voltages for the first test dies, and estimating breakdown voltages of the second test dies using the breakdown voltage estimation model in relation to the currents levels.