ML Process Condition Search With Reliability for Semiconductor Targets

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

In semiconductor manufacturing, finding a process condition that achieves a target process result is time-consuming due to various parameters and customer-specific target values for process results, which differ among customers.

Innovation Solution

An information processing device creates a machine learning model to select appropriate regression methods, performs optimization calculations to determine process conditions achieving the target result, and displays selected conditions along with their reliability, enabling efficient search for optimal process conditions.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional parameter search methods are used to find process conditions achieving target results, then comprehensive exploration of parameter space is possible, but the time required becomes excessively long

Engineering Contradiction:

Improveaccuracy of process result predictionVSAvoidtime to find optimal process conditions

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The system performs preliminary actions by creating multiple machine learning models using different regression methods in advance, then selecting the most appropriate model before performing optimization calculations. This pre-preparation of multiple predictive models enables faster and more accurate determination of process conditions that achieve target results, significantly reducing the time required compared to traditional sequential parameter search methods

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system changes the parameter of regression methods by creating multiple machine learning models using different regression approaches (e.g., ordinary least squares, ridge regression, lasso regression). This parameter change in the modeling approach allows the system to select the most suitable model for the specific data characteristics, improving prediction accuracy while maintaining computational efficiency

Inventive Principle:

Principle #35Parameter changes

2Measurement precision

If multiple regression methods are used to create machine learning models, then prediction accuracy improves, but model selection complexity increases

Engineering Contradiction:

Improveprediction accuracy of process resultsVSAvoidcomplexity of model selection process

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The system implements feedback by evaluating the performance of multiple machine learning models created with different regression methods and using this evaluation information to select the most appropriate model. The selection is based on feedback from model performance metrics, ensuring that the chosen model is well-suited for the specific dataset and prediction task, thereby maintaining high accuracy without excessive complexity

Inventive Principle:

Principle #23Feedback

3Manufacturing precision

If optimization calculations are performed to determine process conditions achieving target results, then process accuracy improves, but computational complexity increases

Engineering Contradiction:

Improveaccuracy of process condition determinationVSAvoidcomputational complexity of optimization

Core Design Contradiction:

Manufacturing precisionVSDevice complexity

Solution Approach 1:

The system performs preliminary action by pre-creating multiple machine learning models using different regression methods before performing optimization calculations. This pre-preparation enables the optimization process to use ready-trained models for rapid prediction, significantly reducing computational complexity while maintaining high accuracy in determining process conditions that achieve target results

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS12253838B2Information processing device, recording medium, and process condition search method

Publication Date: 2025.03.18 TOKYO ELECTRON LTD

AI summary

An information processing device includes: a machine learning model selection part configured to select a machine learning model appropriate for a data set used for learning of the machine learning model; a calculation part configured to perform an optimization calculation by using the selected machine learning model to calculate process conditions that can achieve a target process result, predicted values of a process result corresponding to each of the process conditions, and reliability of the predicted values; a process condition selection part configured to select, among the process conditions that can achieve the target process result, one or more process conditions according to the predicted values of the process result and the reliability of the predicted values; and a display controller configured to display the selected process conditions, the predicted values of the process result corresponding to each of the selected process conditions, and the reliability of the predicted values.