Laser Annealing Control Using Reflected-Light ML Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current laser annealing technologies face challenges in accurately adjusting pulse laser light radiation conditions for semiconductor thin films on substrates, particularly for polycrystalline silicon and IGZO films, due to the difficulty in measuring semiconductor properties during the annealing process and the variability of required conditions based on film thickness and crystal grain size.
Innovation Solution
A machine learning method and system that utilize image data from illuminated semiconductor films to generate a learned model, allowing for the prediction of semiconductor properties and real-time adjustment of pulse laser light radiation conditions using a neural network, enabling precise control of the annealing process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If traditional laser annealing methods are used without machine learning, then the annealing process can be performed, but the precision of adjusting radiation conditions is insufficient due to difficulty in measuring semiconductor properties during the process
Solution Approach 1:
The patent introduces image data as an intermediary that indirectly reflects semiconductor properties. Instead of directly measuring difficult-to-obtain semiconductor properties during annealing, the system captures optical images of the semiconductor film under illumination, which serve as proxies containing information about the film's state. These image features are then processed by machine learning models to infer actual semiconductor properties, thereby solving the measurement difficulty while maintaining high precision in radiation condition adjustment.
Solution Approach 2:
The patent replaces traditional direct measurement methods with an optical imaging and machine learning-based indirect measurement system. Instead of using complex physical measurement devices that would require contact or interfere with the annealing process, the system uses non-contact optical imaging combined with computational algorithms to extract semiconductor property information, achieving both non-intrusive measurement and high precision.
2Manufacturing precision
If laser annealing is performed with fixed radiation conditions, then the process is simple to operate, but the quality of semiconductor thin films varies due to variability in film thickness and crystal grain size
Solution Approach 1:
The patent transforms the static, fixed radiation condition approach into a dynamic, adaptive system. The machine learning model continuously analyzes image data captured during annealing and adjusts radiation conditions in real-time based on the actual state of the semiconductor film. This dynamic adjustment ensures optimal film quality across variations in thickness and crystal grain size while maintaining operational simplicity through automated control.
Solution Approach 2:
The patent implements a feedback loop where image data from the semiconductor film during annealing is fed into a machine learning model, which then predicts semiconductor properties and provides feedback for adjusting radiation conditions. This closed-loop control system automatically compensates for variations in film characteristics, ensuring consistent high-quality output without requiring manual intervention or complex operational procedures.
3Manufacturing precision
If machine learning with neural network is implemented for real-time prediction and adjustment, then the quality and precision of annealing is improved, but the device complexity increases
Solution Approach 1:
The patent uses optical images as copies or representations of the actual semiconductor film state. Instead of directly measuring complex physical properties during annealing, the system captures visual copies (images) of the film under illumination, which contain sufficient information for inference. These image copies are then processed by the machine learning model, simplifying the input requirements while maintaining high prediction precision and reducing the need for complex measurement apparatus.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach enables the prediction of semiconductor properties and adjustment of radiation conditions during laser annealing, improving the quality of semiconductor thin films by optimizing the annealing process and overcoming the limitations of traditional methods.
Implementation Method 1
In the laser annealing, pulse ultraviolet laser light absorbed by an upper-layer semiconductor thin film is used to suppress damage to the substrate due to thermal diffusion
Implementation Method 2
pulse ultraviolet laser light absorbed by an upper-layer semiconductor thin film
Implementation Method 3
acquiring image data generated from reflected light of illumination light radiated to a first region of a semiconductor film
Data Source
AI summary
A machine learning method includes acquiring image data generated from reflected light of illumination light radiated to a first region of a semiconductor film on a substrate, the first region annealed by pulse laser light, acquiring data on a measured semiconductor property of the first region, generating training data including the image data as input and the measured data as output associated with each other, and performing machine learning using a neural network based on the training data to generate a learned model.