Silicon Crystal Pulling Speed Control via ML
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The production of silicon semiconductor wafers faces challenges in controlling the v/G ratio during crystal growth, leading to the formation of agglomerated point defects, which are difficult to manage due to a narrow process window, especially for larger diameters, and require manual intervention, resulting in inconsistent crystal quality.
Innovation Solution
A system that uses simulation calculations to determine the radial course of the axial temperature gradient and adjusts the pulling speed based on machine learning models trained with data from over 1000 crystal pulls to maintain a stable v/G ratio, automatically correcting for system-specific properties and time-dependent changes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual intervention is used to adjust pulling speed, then some correction can be made, but crystal quality remains inconsistent and yield is too low
Solution Approach 1:
The system continuously monitors the actual radial distribution of the v/G ratio during crystal pulling and feeds this information back to an evaluation unit that compares it against target values. Based on this feedback, the control unit automatically adjusts the pulling speed to maintain optimal conditions, eliminating the inconsistency of manual intervention while improving yield through continuous optimization.
Solution Approach 2:
The control system automatically detects deviations in the v/G ratio distribution and self-corrects by adjusting the pulling speed without requiring manual intervention. The system serves itself by autonomously maintaining optimal crystal growth conditions, thereby improving both consistency and productivity.
2Reliability
If the v/G ratio is controlled to prevent agglomeration, then defect formation is reduced, but the process window is extremely narrow and technically inaccessible
Solution Approach 1:
Instead of controlling a single global v/G ratio, the system monitors and controls the radial distribution of v/G ratios across different positions in the crystal growth zone. This allows local optimization at each radial position, enabling defect-free growth across the entire crystal cross-section while effectively expanding the usable process window.
Solution Approach 2:
The system dynamically adjusts the pulling speed based on real-time measurements of the actual v/G ratio distribution. This dynamic control allows the process to adapt to changing conditions and maintain optimal v/G ratios across the entire radial range, effectively widening the accessible process window while preventing defect formation.
3Reliability
If pulling speed is adjusted manually based on operator experience, then some defects can be reduced, but the correction amount depends on operator experience and yield remains low
Solution Approach 1:
The system replaces manual operator intervention with an automated control system that uses sensors, evaluation algorithms, and automated actuators. This substitution eliminates dependence on operator experience and provides consistent, precise control that improves both defect reduction and crystal yield through objective, data-driven decisions.
Solution Approach 2:
The automated system continuously measures the actual v/G ratio distribution, evaluates it against target values, and adjusts the pulling speed accordingly. This closed-loop feedback mechanism ensures consistent defect reduction and maximizes yield by maintaining optimal conditions throughout the crystal growth process, independent of operator skill level.
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 significantly improves the yield of crystals within the desired defect window, surpassing manual correction methods by providing precise and timely adjustments to the pulling speed, ensuring higher quality crystals with reduced defects.
Implementation Method 1
the radial course of the axial temperature gradient G(r) is determined in advance, in particular by means of simulation calculations
Implementation Method 2
Single crystals - such as single crystals of silicon - can, for example, have grown-in defects that are undesirable because they can disrupt the function of electronic components integrated on semiconductor wafers made from the crystals
Implementation Method 3
When such point defects become supersaturated, they tend to form agglomerates
Data Source
AI summary
Method for growing a silicon crystal according to the Czochralski method, wherein a first crystal is produced by applying a target curve on a crystal growing machine, and the produced first crystal is subjected to a measurement with regard to crystal defects, and the result of the measurement with regard to crystal defects is compared with a desired measurement result, and based on the deviation of the result of the measurement from a desired measurement result, a new target curve is calculated which serves for the production of a further crystal, characterized in that a model calculated by a machine learning method is used for calculating the new target curve.