Processing Condition Learning Using Pseudo States in Vacuum Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In semiconductor manufacturing, optimizing processing procedures with numerous control parameters is challenging due to the difficulty in analyzing correlations between inputs and outputs, and direct observation of sample states is often impossible, especially in vacuum environments like etching chambers.
Innovation Solution
A system utilizing reinforcement learning with a pseudo state calculated from physical quantities correlating with the sample state, allowing for the determination of appropriate processing conditions for each step without direct measurement of the sample state, using a computer-based learning unit that performs value function analysis and adjusts processing conditions based on evaluated pseudo states.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If reinforcement learning is applied to optimize processing procedures with numerous control parameters, then optimization efficiency is improved, but the complexity of analyzing correlations between inputs and outputs increases
Solution Approach 1:
The patent introduces a pseudo-state as an intermediary representation that bridges the gap between complex processing conditions and sample states. Instead of directly analyzing the complex correlations between numerous control parameters and sample states, the system calculates a pseudo-state from physical quantities that serve as a mediator, simplifying the reinforcement learning process while maintaining optimization efficiency
Solution Approach 2:
The patent replaces direct observation and measurement of sample states with a computational model that calculates pseudo-states from physical quantities. This substitution eliminates the need for direct mechanical intervention or complex measurement systems, using information processing instead to achieve the same optimization goal
2Measurement precision
If direct measurement of sample state is performed to achieve accurate control, then control precision is improved, but the ability to operate in vacuum environments without breaking vacuum is deteriorated
Solution Approach 1:
The patent uses physical quantities as intermediaries to infer sample states without direct measurement. By measuring physical quantities that can be obtained in vacuum environments and calculating pseudo-states from these measurements, the system achieves control precision while maintaining vacuum integrity
Solution Approach 2:
The patent replaces direct mechanical measurement of sample states with a computational inference system. Instead of physically accessing or touching the sample to measure its state, the system uses a calculation model that derives sample state information from physical quantity measurements, enabling operation in vacuum environments
3Manufacturing precision
If the number of control parameters is increased to achieve fine and complicated processing, then processing capability is improved, but the time required for process development increases
Solution Approach 1:
The patent enables the system to automatically optimize processing procedures through reinforcement learning without requiring extensive manual experimentation. The system serves itself by using the calculated pseudo-states and value functions to autonomously determine optimal control parameters, significantly reducing the time required for process development while maintaining fine and complicated processing capability
Solution Approach 2:
The patent implements a feedback mechanism where the results of processing are used to update the value function, which in turn guides future processing condition selections. This closed-loop feedback system allows the system to learn from past experiences and continuously improve processing capability over time, reducing the need for extensive preliminary experimentation
Data Source
AI summary
A system for determining a processing procedure including a plurality of processes for controlling an object, the system includes a learning unit for performing a learning process for determining a processing condition of each of a plurality of processes, and the learning unit acquires a physical quantity correlating with a state of the object on which a process has been performed under a predetermined processing condition, from a device for controlling the object on the basis of the processing procedure, calculates a pseudo state corresponding to the state of the object on the basis of the physical quantity, performs a learning process using a value function, and determines a processing condition of each of the plurality of processes to achieve a target state of the object.