Lithography Control Using Switched Neural Networks for Real-Time Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional control apparatuses using neural networks face long calculation times, which degrade real-time control performance.
Innovation Solution
A control apparatus comprising a plurality of neural networks and a selector to choose the appropriate network for each control pattern, reducing the scale of individual networks and enabling rapid calculation by switching between them.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single large-scale neural network is used for control, then control accuracy is improved, but calculation time increases and real-time performance degrades
Solution Approach 1:
The patent divides a single large-scale neural network into multiple smaller neural networks, each specialized for specific control patterns. This segmentation reduces the computational burden of each individual network while maintaining overall control accuracy through selective deployment of appropriate networks for different operating conditions
Solution Approach 2:
The system dynamically selects which neural network to deploy based on the current control pattern or operating conditions. This dynamic adaptation allows the system to use computationally efficient networks for simple patterns while reserving more complex networks for challenging scenarios, optimizing both accuracy and real-time performance
2Measurement precision
If a single large-scale neural network is used for control, then control accuracy is improved, but real-time control performance degrades
Solution Approach 1:
The patent segments the control system into multiple specialized neural networks, each optimized for specific control patterns. This allows the system to achieve high control accuracy for each pattern while maintaining fast real-time response by selecting the appropriate pre-trained network for the current condition
Solution Approach 2:
Multiple neural networks are pre-trained offline for different control patterns before deployment. This preliminary action transfers complex computation to the training phase, allowing the runtime system to simply select and execute pre-computed networks, thereby achieving both high accuracy and real-time performance
3Loss of time
If the scale of neural network is reduced, then calculation time is shortened, but control accuracy may deteriorate
Solution Approach 1:
Each neural network is specialized for specific control patterns or operating conditions, achieving high local accuracy for its designated domain. This local quality approach ensures that while individual networks are smaller and faster, they maintain high accuracy for their specific control patterns through targeted training and specialization
Data Source
AI summary
A control apparatus for generating a control signal for controlling a control target includes a plurality of neural networks, and a selector configured to select, from the plurality of neural networks, a neural network to be used to generate the control signal, wherein each of the plurality of neural networks is selected by the selector to be used in execution of one corresponding control pattern among a plurality of control patterns for controlling the control target.