Wafer Process Control Factor Selection Across Interacting Fabrication Steps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current control factor selection technologies in semiconductor fabrication processes fail to effectively account for interdependent characteristics across multiple processes, leading to potential quality deterioration due to interactions between processes, as they typically focus on a single process without considering interactions between them.
Innovation Solution
A method using neural networks to generate sequence data from measured process factors, train networks to infer relevant process factors affecting quality, and select control processes and factors based on interactions across multiple processes, employing transformer encoders and multi-layer perceptrons to calculate scores and Shapley values for optimal selection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If control factor selection focuses on a single process using measured values from that process, then the selection process is simple and fast, but it fails to account for interdependent characteristics across multiple processes leading to quality deterioration
Solution Approach 1:
The patent segments the control factor selection process into two distinct neural networks: a first neural network that processes sequence data from multiple processes to identify relevant process factors, and a second neural network that generates quality indices based on selected control factors. This segmentation allows each network to specialize in specific tasks while collectively addressing multi-process interdependencies.
Solution Approach 2:
The patent introduces sequence data as an intermediary representation that captures temporal and contextual relationships across multiple processes. This sequence data serves as a mediator between raw measured values and control factor selection, enabling the system to account for inter-process interactions without requiring direct complex modeling of all process relationships.
2Manufacturing precision
If control factor selection considers interactions between multiple processes, then wafer quality improves, but the complexity of the selection system increases
Solution Approach 1:
The patent replaces traditional mechanical or rule-based control factor selection systems with neural network-based models. The first neural network automatically identifies relevant process factors from sequence data, and the second neural network determines optimal control factors, eliminating the need for manual configuration and reducing system complexity despite handling multi-process interactions.
Solution Approach 2:
The patent transforms the control factor selection problem by changing the parameters from direct process measurements to sequence data representations with embedded temporal relationships. This parameter transformation enables the neural networks to automatically capture inter-process interactions without requiring explicit modeling of complex process relationships.
3Ease of manufacture
If traditional control factor selection methods are used, then the system is easy to implement, but it cannot accurately identify control factors that affect quality across multiple interacting processes
Solution Approach 1:
The patent creates a virtual copy of the multi-process system through sequence data that preserves temporal and contextual relationships. This digital representation allows the neural networks to analyze and identify control factors without requiring physical experimentation or complex sensor networks, making the system easier to implement while maintaining high identification accuracy.
Data Source
AI summary
A method including generating sequence data based on measured values of respective one or more process factors of each of a plurality of processes of a semiconductor fabrication process for a wafer within the semiconductor fabrication process, generating a temporary quality index of the wafer using a second neural network connected to a first neural network that is provided the sequence data, training the first neural network and the second neural network based on a loss between the temporary quality index and a set actual quality index of the wafer, selecting a control process from among the plurality of processes using at least one of the trained first neural network and/or the trained second neural network, and selecting a control factor from among multiple process factors of the selected process using the trained second neural network.