Bayesian Parameter Optimization for Manufacturing Variation Robustness
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing design techniques struggle to account for manufacturing variations, leading to inefficiencies in device design and optimization.
Innovation Solution
A parameter optimization method that generates neighboring design value sets to simulate and calculate scores, using a Bayesian estimation to adjust design values based on manufacturing variations, and iteratively refine the design through a proxy model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional design techniques are used, then the design process is simple, but manufacturing variations are not accounted for leading to poor design robustness
Solution Approach 1:
The patent applies preliminary action by generating multiple neighboring design value sets before final design selection. These neighboring sets are created by varying design parameters within expected manufacturing variation ranges, allowing the design to be pre-tested for robustness against manufacturing variations before actual production
Solution Approach 2:
The patent uses copying by creating multiple copies of the design model through neighboring design value sets. Each neighboring set represents a potential manufacturing variation scenario, allowing virtual testing of multiple design copies without requiring physical prototypes for each variation
2Reliability
If manufacturing variations are accounted for through multiple design value sets, then design robustness improves, but computational cost and processing time increase
Solution Approach 1:
The patent introduces a proxy model as an intermediary between the detailed simulation model and the optimization process. The proxy model provides approximate predictions of design performance, allowing rapid evaluation of neighboring design value sets without requiring computationally expensive full simulations for each set
Solution Approach 2:
The patent applies partial action by using a simplified proxy model rather than full detailed simulations for all neighboring design value sets. The proxy model provides sufficient accuracy for optimization purposes while requiring significantly less computational time, with detailed simulations used only for final verification
3Measurement precision
If detailed simulations are performed for all neighboring design value sets, then measurement precision improves, but productivity decreases
Solution Approach 1:
The proxy model serves as an intermediary that provides rapid approximate predictions, allowing the system to screen many neighboring design value sets quickly. Only the most promising designs identified through the proxy model undergo detailed simulation for final accuracy verification
Solution Approach 2:
The patent uses partial action by performing detailed simulations for only a subset of neighboring design value sets identified as promising by the proxy model. This selective approach maintains measurement precision for critical evaluations while improving overall productivity by avoiding unnecessary detailed simulations
Data Source
AI summary
Plural neighboring design value sets are generated, from a representative design value set input to a simulator to acquire a representative characteristic value set. A first neighboring design value set is input from plural neighboring design value sets to acquire a neighboring characteristic value set, and a calculation score is calculated from neighboring characteristic values included in the neighboring characteristic value set. The magnitude relation between the calculation score and the discontinuation threshold is determined, and when the criterion is satisfied, the second neighboring design value set is input to the simulator. If the criterion is not satisfied, the value of the objective function is calculated from the characteristic values included in the neighboring characteristic value set. Finally, an acquisition function is calculated from the proxy model of the objective function by Bayesian estimation, and a new representative design value set is generated based on the acquisition function.