Semiconductor Experiment-Point Recommendation with ML Uncertainty
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The process of developing semiconductor processing requires numerous processing tests to improve the accuracy of predictive models, which is costly and time-consuming due to the large number of control parameters and their combinations.
Innovation Solution
An experiment point recommendation device that evaluates the contribution, stability, and uncertainty of control parameters using machine learning models to identify optimal experimental conditions for learning data collection, reducing the need for extensive trial and error.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a large number of processing tests are conducted to improve the accuracy of the predictive model, then the model accuracy is improved, but the cost and time for process development increase significantly
Solution Approach 1:
The patent performs preliminary analysis of control parameter contributions, stability, and uncertainty before conducting processing tests. By pre-identifying which control parameters are most influential and where uncertainty is highest, the system prioritizes experiments that will most effectively improve model accuracy, reducing the total number of tests needed.
Solution Approach 2:
The patent implements a feedback loop where the predictive model continuously learns from processing test results. The system evaluates the contribution of each control parameter to prediction accuracy, adjusts experiment selection based on this feedback, and iteratively improves the model with each cycle of targeted experiments rather than conducting all possible tests upfront.
2Measurement precision
If a large number of processing tests are conducted to improve the accuracy of the predictive model, then the model accuracy is improved, but the cost of process development increases significantly
Solution Approach 1:
The system performs preliminary evaluation of control parameter importance and uncertainty distribution before executing processing tests. This allows identification of the minimum necessary experiments required to achieve acceptable model accuracy, avoiding wasteful spending on tests that would provide marginal improvements.
Solution Approach 2:
The patent dynamically changes the selection of control parameters for experimentation based on their evaluated contribution to prediction accuracy and their current uncertainty levels. The system focuses resources on parameters that will yield the highest return on investment for model improvement, optimizing the cost-effectiveness of process development.
3Adaptability or versatility
If many control parameters are used in semiconductor processing, then the device performance and processing capability are improved, but the complexity of process development increases
Solution Approach 1:
The patent segments the large set of control parameters into groups based on their contribution to prediction accuracy and their uncertainty levels. By dividing parameters into high-priority and low-priority groups, the system manages the complexity of process development through structured analysis and targeted experimentation rather than treating all parameters uniformly.
Solution Approach 2:
The system extracts and focuses on the most critical control parameters that have the greatest impact on prediction accuracy and highest uncertainty. By isolating these key parameters for detailed analysis and experimentation, the patent reduces process development complexity by concentrating efforts on the few parameters that matter most rather than managing all parameters simultaneously.
4Reliability
If extensive trial and error of processing tests is conducted, then appropriate processing conditions are derived, but the period of process development is extended
Solution Approach 1:
The patent performs preliminary identification of promising processing conditions by analyzing control parameter contributions and uncertainty distributions before conducting extensive trial and error tests. This preliminary guidance directs the experimentation toward conditions most likely to yield appropriate processing conditions, reducing the overall development period.
Solution Approach 2:
The system implements continuous feedback during the processing test sequence, evaluating how each experiment improves prediction accuracy for different control parameters. Based on this feedback, the system adaptively selects the next most valuable experiment, ensuring that processing conditions are derived efficiently through learned insights rather than random trial and error.
Data Source
AI summary
For a machine learning model that receives control parameters of a semiconductor processing device and outputs shape parameters that express a processed shape of a semiconductor sample processed by the semiconductor processing device, an experiment point obtaining learning data is recommended. A contribution of each control parameter to the prediction of the machine learning model is evaluated from feature quantity data that is a value of a control parameter of the learning data used for learning of the machine learning model, and the experiment point is recommended based on a stability evaluation and an uncertainty evaluation of the prediction by the machine learning model in a space defined by the control parameters selected based on the contribution as axes.