Bayesian Optimization for CVD Process Parameter Design
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The transition towards higher quality for graphene and transition metal dichalcogenides (TMDs) using chemical vapor deposition (CVD) synthesis is historically slow due to the complex interplay of thermodynamics and chemical kinetics at the growth front, making it challenging to systematically guide experiments towards high-quality conditions.
Innovation Solution
A computer-implemented method is developed to optimize the CVD synthesis process by using machine learning techniques to determine optimal process parameters with a minimum number of experimental trials. This method involves receiving process data, determining a feasible region using classification and regression models, calculating scores for data points, selecting recommended process parameters, and refining these parameters based on experimental results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If traditional trial and error experimentation is used to optimize CVD synthesis parameters, then comprehensive exploration of the design space is achieved, but the time, effort, and cost required increase significantly
Solution Approach 1:
The patent performs preliminary computational actions by training machine learning models (classification and regression models) on existing process data before actual experimentation. The system pre-identifies feasible regions and predicts optimal parameters, allowing experiments to be designed more efficiently from the outset rather than through random trial and error
Solution Approach 2:
The patent implements iterative feedback loops where experimental results are fed back into the machine learning models to refine predictions. The system continuously updates the feasible region identification and parameter optimization based on accumulated experimental data, progressively improving quality while reducing the number of trials needed
2Loss of information
If extensive experimental trials are conducted to map the complex design space, then accurate understanding of parameter relationships is achieved, but resources (time, effort, materials) are consumed excessively
Solution Approach 1:
The patent introduces machine learning models as intermediary components between the complex CVD synthesis process and the experimenter. These models (classification model for feasible region identification and regression model for parameter prediction) act as mediators that process experimental data and provide guided recommendations, reducing the need for direct extensive experimentation
Solution Approach 2:
The patent replaces the mechanical trial-and-error experimentation system with an intelligent computational system. Instead of relying on physical experimentation alone, the system uses machine learning algorithms to predict outcomes and guide experiments, substituting computational intelligence for brute-force experimental exploration
3Reliability
If the feasible region is broadly defined to ensure coverage of successful outcomes, then the probability of finding optimal parameters increases, but the search space becomes larger and more difficult to navigate
Solution Approach 1:
The patent applies local quality by identifying and focusing on specific feasible regions within the broader design space rather than treating all parameter combinations equally. The classification model identifies local regions where successful fabrication is more likely, allowing the system to concentrate computational and experimental resources on promising areas while maintaining high success rates
Data Source
AI summary
Provided herein are methods and systems for a computer implemented method for designing and optimizing a fabrication process characterized by one or more process parameters. The outcome of the fabrication process is characterized by a quality metric used to optimize the fabrication process. The method uses a classification model in conjunction with a regression model, both using artificial intelligence and trained using experimental results, to iteratively recommend process parameter values for performing real world fabrication experiments. The experimental results are used to continuously train the models. The outcome is a set of process parameters that yield an optimal fabrication process with less time and effort than using conventional design of experiment methods.