Neural Network Control Compensation for Lithography Accuracy Shifts
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional control systems using neural networks face reliability issues when the state of the control object or the disturbance environment changes, leading to decreased control accuracy and productivity due to the time-consuming process of relearning neural network parameters.
Innovation Solution
A control apparatus that includes a first compensator, a corrector with adjustable coefficients, and a neural network-based second compensator, allowing for rapid adjustment of parameter values to maintain control accuracy without the need for extensive relearning, by generating correction signals based on control deviations and using multiple neural networks for different operation patterns.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If neural network parameters are adjusted by machine learning relearning, then control accuracy is improved when state changes occur, but execution time increases considerably and productivity decreases
Solution Approach 1:
The patent pre-calculates and stores correction values for various states of the control object and disturbance environments before actual operation. When state changes occur, the system simply retrieves and applies the pre-computed correction values instead of performing time-consuming relearning, thus maintaining control accuracy while avoiding productivity loss.
Solution Approach 2:
The system dynamically selects appropriate correction values based on the current state of the control object and disturbance environment. By having multiple pre-prepared correction values for different states and dynamically switching between them, the system adapts to changes rapidly without requiring extensive relearning time.
2Device complexity
If a single neural network controller is used, then device complexity is reduced, but adaptability to different operation patterns and state changes deteriorates
Solution Approach 1:
The patent segments the control function by dividing the control object's state space into multiple regions and preparing different correction values for each region. This segmentation allows the system to handle different operation patterns effectively while keeping each individual correction value relatively simple.
Solution Approach 2:
The system achieves multi-functionality by using a single control apparatus that can handle multiple operation patterns and state changes through the use of multiple pre-prepared correction values. The same control structure serves multiple purposes by selecting appropriate correction values based on the current state.
Data Source
AI summary
A control apparatus which generates control signal for controlling a control object, includes a first compensator configured to generate a first signal based on a control deviation of the control object, a corrector configured to generate a correction signal by correcting the control deviation in accordance with an arithmetic expression having an adjustable coefficient, a second compensator configured to generate a second signal by a neural network based on the correction signal, and an arithmetic device configured to generate the control signal based on the first signal and the second signal.