Semiconductor Digital Twin Control for Real-Time Process Drift Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Semiconductor wafer processing complexity increases with shrinking device sizes, requiring multiple optimization steps, but manual adjustments are time-consuming and resource-intensive, and it's challenging to identify and address drifts in quality and consistency in manufacturing systems.
Innovation Solution
Implementing a model-based digital simulation for real-time automated control of semiconductor manufacturing processes using trained machine learning models that run concurrently with the physical process, allowing for autonomous parameter adjustments and predictive data-driven corrective actions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If manual process tuning and optimization is performed using design of experiment (DoE) with physical wafers, then process parameters can be optimized to achieve desired specifications, but valuable wafer resources are consumed and the process takes extensive time
Solution Approach 1:
The patent creates a digital twin (virtual model) of the semiconductor manufacturing process that replicates the behavior of the physical process. This digital copy allows for virtual experimentation and optimization without consuming physical wafer resources or manufacturing time, thereby resolving the contradiction between achieving manufacturing precision and reducing time loss.
Solution Approach 2:
The patent performs process optimization and parameter tuning in the digital domain before actual manufacturing. By conducting virtual design of experiment (DoE) simulations beforehand, the optimal process parameters are determined in advance, eliminating the need for time-consuming physical trial-and-error experimentation.
2Manufacturing precision
If multiple process steps and parameters are used to handle shrinking device sizes, then manufacturing capability is maintained, but process complexity increases significantly
Solution Approach 1:
The patent implements a universal digital twin platform that can simulate and optimize multiple different semiconductor manufacturing processes and device types within a single system. This multi-functional approach consolidates the complexity of handling various shrinking device sizes into one integrated tool, reducing the overall process complexity while maintaining manufacturing precision.
Solution Approach 2:
The patent replaces the complex manual mechanical process of tuning multiple physical process parameters with an automated computational system. The digital twin uses algorithms and simulations to automatically optimize process parameters, substituting the complex manual adjustment of multiple knobs with an automated computational approach.
3Productivity
If real-time monitoring and automated control is implemented, then quality consistency and process efficiency are improved, but system complexity and initial resource investment increase
Solution Approach 1:
The patent implements a feedback mechanism where the digital twin continuously receives real-time data from the physical manufacturing process, compares actual outcomes with simulated predictions, and automatically adjusts process parameters to maintain optimal performance. This closed-loop feedback system improves productivity while managing complexity through automation.
Solution Approach 2:
The patent enables the manufacturing system to self-optimize and self-correct by using the digital twin to automatically identify and compensate for process drifts and variations. The system performs self-diagnosis and self-adjustment without requiring external intervention, improving quality consistency while reducing the need for complex manual monitoring systems.
Data Source
AI summary
The disclosure describes methods and systems for operating a manufacturing process with a concurrent real-time simulation of the manufacturing process via a digital twin model. Sensor data indicative of parameters of an ongoing manufacturing process are input into the digital twin model, and used to predict an output of the manufacturing process. The predicted output is compared to a target output. One or more trained machine learning models are used to determine a corrective action to be implemented by a controller of the manufacturing process to minimize any deviation from the target output.