Semiconductor Process ML Control for Wafer Uniformity Drift
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Semiconductor wafer processing complexity increases with shrinking device sizes, requiring multiple steps and numerous knobs for optimization, leading to resource consumption and difficulty in maintaining target quality and consistency, with manual optimization being inadequate for large process spaces and process drift detection.
Innovation Solution
Implementing machine learning models for pixelated-control of temperature, gas flow, and localized plasma generation in semiconductor processing systems, enabling real-time prediction and adjustment of process parameters to maintain uniformity and detect drift.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If multiple process steps and knobs are used to optimize semiconductor manufacturing, then manufacturing precision is improved, but device complexity increases
Solution Approach 1:
The patent applies parameter changes by using machine learning models to automatically adjust and optimize multiple process parameters simultaneously. The system learns optimal parameter combinations from historical data and applies them to maintain manufacturing precision without requiring manual tuning of each parameter, thus resolving the contradiction between precision and complexity.
Solution Approach 2:
The machine learning system performs self-service by autonomously optimizing process parameters without continuous human intervention. The model learns from process data and automatically adjusts settings, enabling the system to maintain high manufacturing precision while reducing the operational complexity burden on engineers.
2Manufacturing precision
If design of experiment is used to tailor process, then manufacturing precision is improved, but loss of time increases
Solution Approach 1:
The machine learning model performs preliminary action by pre-learning optimal process configurations from historical data before actual production. This allows the system to quickly recommend optimized processes without requiring extensive new DoE experiments, significantly reducing the time loss while maintaining manufacturing precision.
Solution Approach 2:
The system uses copying by replicating successful process patterns learned from historical DoE data. Instead of performing complete new experiments, the model copies and adapts proven process configurations to new situations, achieving fast process tailoring with minimal time investment.
3Manufacturing precision
If full wafers are dedicated for DoE, then manufacturing precision is improved, but loss of substance increases
Solution Approach 1:
The machine learning approach applies partial action by using only the necessary minimum data for training the model, rather than consuming full wafers for DoE. The system achieves process optimization through intelligent analysis of partial data sets, significantly reducing wafer consumption while maintaining optimization effectiveness.
4Ease of operation
If manual monitoring is used to detect process drift, then ease of operation is maintained, but measurement precision deteriorates
Solution Approach 1:
The patent replaces manual mechanical monitoring with an automated machine learning-based detection system. The ML model continuously analyzes process data to detect drift with high precision, eliminating the limitations of manual monitoring while maintaining ease of operation through automated alerts and recommendations.
Solution Approach 2:
The system implements continuous feedback by automatically monitoring process parameters and comparing them against learned normal patterns. When drift is detected, the system provides feedback to operators or automatically adjusts parameters, achieving high measurement precision for drift detection while keeping operations simple through automated responses.
Data Source
AI summary
The disclosure describes methods and systems for training and deploying a machine learning predictive model for use in a semiconductor manufacturing process. Specifically, the present disclosure provides for training machine learning predictive models for manufacturing components using design data, process parameters, gas flow configurations from a pixelated showerhead, temperature profile across an electrostatic chuck, and measured uniformity profiles of processed wafers. The present disclosure also provides for deploying the machine learning predictive model to effectuate real-time adjustments to a manufacturing process.