Elastic Strain Engineering Using ML Bandgap Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current strain engineering methods are limited to uniaxial and biaxial strains due to the complexity of predicting and testing material properties with strains having three or more degrees of freedom, making it impractical to explore the entire range of possible strains for altering material properties.
Innovation Solution
Development of a trained statistical model that predicts the bandgap and energy dispersion of materials under strains with three or more degrees of freedom, using machine learning techniques to generate models from training data and apply them to determine material properties across various strain coordinates.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If strains with three or more degrees of freedom are applied to explore the full parameter space for material property engineering, then the ability to tune material properties (electronic bandgap, electrical, thermal, optical, magnetic characteristics) is significantly improved, but the complexity of predicting and testing the entire range of possible strains becomes prohibitively high
Solution Approach 1:
The patent applies preliminary action by pre-training statistical models using density functional theory calculations across the full strain parameter space before actual material design. The trained models are then reused for predicting material properties under various strain conditions, avoiding the need to perform complex calculations each time a new strain configuration is evaluated. This preliminary preparation resolves the contradiction by enabling rapid property tuning without repeated computational complexity.
2Measurement precision
If the entire range of possible strains is experimentally tested to determine material properties, then accurate property prediction across all strain coordinates is achieved, but the time and resources required for testing become impractical
Solution Approach 1:
The patent employs copying by creating statistical model representations of the complex density functional theory calculations. These trained models serve as simplified copies that can rapidly predict material properties for any strain configuration without requiring actual experimental testing or full computational calculations. This approach achieves accurate property prediction while dramatically reducing the time required compared to exhaustive experimental testing.
3Ease of manufacture
If computational methods are used to predict material properties for all strain configurations, then experimental requirements are reduced, but the computational cost and complexity increase significantly
Solution Approach 1:
The patent resolves this contradiction by performing computational work in advance through training statistical models on density functional theory data. Once trained, these models provide rapid predictions with minimal computational overhead. This preliminary computational action reduces subsequent experimental requirements while avoiding the need for repeated complex calculations, thereby easing the manufacturing and design process.
Data Source
Figure 1A~1B
Figure 2~3
Figure 4A~4B
AI summary
Methods for training statistical models for the bandgap and energy dispersion of materials as a function of an applied strain, as well as uses of these trained statistical models for elastic strain engineering of materials, are described.