SiC Crystal Growth Control for Unobservable Temperature Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge in manufacturing silicon carbide (SiC) crystals lies in the inability to directly measure critical physical quantities, such as temperature and raw material concentration, within the high-temperature environment of the crystal growth chamber due to interference with the raw material supply path and the use of non-transparent heat insulating materials, which limits the accuracy of existing prediction models and hinders real-time feedback control.
Innovation Solution
A control device employing a learning model generated through machine learning, using simulation results and experimental data to estimate unobservable physical quantities, allowing for real-time feedback control and accurate prediction of conditions within the SiC crystal manufacturing apparatus, even in high-temperature environments where direct measurement is impossible.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If sensors are installed on the crystal growth surface to measure physical quantities, then measurement precision is improved, but device complexity increases and installation becomes impossible due to positional interference with raw material supply path
Solution Approach 1:
The patent introduces an intermediary measurement approach by installing sensors outside the chamber and using optical methods (laser irradiation and reflection detection) to measure physical quantities remotely. This mediator system allows measurement without direct contact with the crystal growth surface, avoiding interference with raw material supply while maintaining measurement capability
Solution Approach 2:
The patent replaces direct mechanical sensor installation on the crystal growth surface with an optical measurement system. Instead of physically placing sensors in the chamber where they would interfere with material supply, the system uses laser beams and optical detectors to measure physical quantities through non-contact methods
2Temperature
If the chamber inner wall is covered with non-transparent heat insulating material, then temperature control is improved, but measurement precision deteriorates because sensors cannot observe the crystal growth surface
Solution Approach 1:
The patent uses optical radiation as an intermediary to transfer information from the crystal growth surface to external detectors. Laser beams penetrate or reflect off the chamber environment to carry measurement data without requiring the chamber walls to be transparent, thus maintaining thermal insulation while enabling remote measurement
Solution Approach 2:
The patent utilizes optical properties and radiation characteristics to detect physical quantities. By measuring light reflection, absorption, or emission properties, the system can infer temperature and other physical parameters without requiring visual transparency of the chamber walls
3Productivity
If machine learning is used to predict unobservable physical quantities, then productivity is improved through real-time feedback control, but manufacturing precision deteriorates due to lack of accurate measurement data for training
Solution Approach 1:
The patent implements a feedback mechanism where measurement data from the optical sensor system is continuously fed into the machine learning model. This real-time feedback loop allows the system to adjust predictions based on actual measurements, improving both the accuracy of unobservable quantity estimation and the overall manufacturing precision
Solution Approach 2:
The system uses the measurement data it collects to continuously improve its own prediction capabilities. The machine learning model is trained and refined using data from the optical measurement system, allowing the system to self-enhance its accuracy without requiring external intervention or additional expensive measurement equipment
Data Source
AI summary
A control device has a learning model that outputs an estimated value of a second physical quantity that is unobservable under a condition of manufacturing a SiC crystal, from a first physical quantity that is observable under the condition of manufacturing the SiC crystal. The control device generates a basic learning model by mechanical learning using, as teacher data, a simulation result of a simulation model based on structural data of a SiC crystal manufacturing apparatus. The control device acquires measured values of the first physical quantity and the second physical quantity measured under a condition that the SiC crystal is unable to be manufactured while the second physical quantity is observable, and generates the learning model that corrects an output of the basic learning model based on the measured values.