Semiconductor Process Control Module for Real-Time Recipe Adaptation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional semiconductor manufacturing processes rely on pre-defined recipes that are time-consuming and fail to adapt well to variations in the manufacturing environment, necessitating a more sophisticated and adaptive control system for real-time adjustments.
Innovation Solution
A control module utilizing AI-driven reinforcement learning algorithms, combined with a comprehensive system digital twin and neural networks, autonomously generates and adjusts process recipes based on real-time data from various sensors, incorporating chamber and edge ring digital twins to model plasma exposure and wear.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If pre-defined recipes are used for process control, then manufacturing simplicity is maintained, but adaptability to variations in manufacturing environment deteriorates
Solution Approach 1:
The control system performs self-learning through reinforcement learning algorithms, automatically optimizing process parameters without manual intervention. The system serves itself by generating and refining its own control policies based on real-time sensor data and digital twin simulations, eliminating the need for pre-defined recipes while maintaining ease of operation.
Solution Approach 2:
The control system transitions from static pre-defined recipes to dynamic adaptive control. The policy neural network continuously updates process parameters based on real-time conditions, allowing the system to adapt to manufacturing environment variations while maintaining simplicity through automated decision-making.
2Measurement precision
If traditional digital twins are used for simulating process dynamics, then modeling accuracy is improved, but computational complexity increases
Solution Approach 1:
The patent extracts only the essential features needed for real-time control from complex traditional digital twins. By focusing on the critical process dynamics and using simplified models combined with reinforcement learning, the system achieves adequate modeling accuracy without the full computational burden of traditional digital twins.
Solution Approach 2:
Instead of maintaining a single complex digital twin, the system uses multiple simpler digital twin instances for different subsystems (RF, gas, temperature, plasma) that can be independently trained and updated. This modular approach reduces overall computational complexity while maintaining necessary modeling accuracy.
3Adaptability or versatility
If reinforcement learning algorithms are implemented for autonomous recipe generation, then adaptability to real-time conditions is improved, but computational requirements increase
Solution Approach 1:
The reinforcement learning system is segmented into multiple policy neural networks, each responsible for controlling specific subsystems (RF power, gas flow, temperature, plasma parameters). This segmentation allows parallel processing and reduces the computational energy required for each individual decision, while maintaining overall real-time adaptability.
Solution Approach 2:
The system performs preliminary training of policy neural networks using offline simulations and historical data before real-time operation. This preliminary action prepares the models in advance, reducing the computational energy required during actual real-time control operations.
Data Source
AI summary
Disclosed herein is a control module for a semiconductor process system, utilizing reinforcement learning (RL) algorithms to autonomously generate and adjust process recipes. It features a comprehensive system digital twin, including subsystem, chamber plasma, and process digital twins, and employs neural network models for efficiency. Using a policy neural network and Monte Carlo Tree Search (MCTS), real-time adjustments are based on calibrated state data from various sensors, enhancing precision and adaptability in manufacturing processes.