Machine-Learning Polishing Pad State Prediction from CMP Conditions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for determining the state of a polishing pad in a substrate processing apparatus are inadequate as they only consider the accumulated use time of the pad, failing to account for the complex interactions of operating conditions of the top ring, polishing table, polishing-fluid supply nozzle, dresser, and atomizer, which affect the pad's state.
Innovation Solution
An information processing apparatus and method that utilize machine learning to predict the state of a polishing pad by analyzing operating conditions, including top-ring, polishing-table, polishing-fluid supply nozzle, and dresser information, using a learning model to correlate these conditions with the pad's state.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If accumulated use time is used to manage polishing pad replacement, then management simplicity is improved, but prediction accuracy of polishing pad state deteriorates
Solution Approach 1:
The patent transforms the single parameter of accumulated use time into multiple operating condition parameters (top ring state, polishing table state, polishing-fluid supply nozzle state, dresser state, atomizer state). This parameter expansion allows the system to capture the complex interactions affecting polishing pad wear, thereby improving prediction accuracy while maintaining management simplicity through automated data collection and machine learning processing.
2Measurement precision
If multiple operating conditions are analyzed to predict polishing pad state, then prediction accuracy is improved, but system complexity increases
Solution Approach 1:
The patent introduces a machine learning model as an intermediary between the multiple operating conditions and the polishing pad state prediction. This intermediary automatically processes the complex interactions among top ring, polishing table, polishing-fluid supply nozzle, dresser, and atomizer conditions, reducing the perceived system complexity while maintaining high prediction accuracy. The learning model acts as a black box that handles the complexity internally.
Solution Approach 2:
The patent replaces traditional mechanical or manual analysis methods with machine learning algorithms. Instead of manually analyzing the complex interactions of multiple operating conditions, the system uses automated machine learning models to process the data, substitute human expertise with computational intelligence, and generate predictions without requiring deep understanding of the underlying complex mechanisms.
3Loss of substance
If traditional time-based management is used, then implementation cost is reduced, but maintenance quality deteriorates
Solution Approach 1:
The patent implements a self-service system where the machine learning model automatically collects operating condition data, processes the information, and generates polishing pad state predictions without requiring external expert intervention. The system serves itself by automatically identifying when pad replacement or maintenance is needed, improving maintenance quality while keeping implementation costs low through automated rather than manual processes.
Data Source
AI summary
The information processing apparatus (5) includes an information acquisition section (500) configured to acquire operating condition information including top-ring state information, polishing-table state information, polishing-fluid-supply-nozzle state information, dresser state information, and atomizer state information, as operating conditions under which a substrate processing apparatus performs operation of chemical mechanical polishing of a substrate. The information processing apparatus (5) further includes a state prediction section (501) configured to predict polishing-pad state information for the operating condition information by inputting the operating condition information acquired by the information acquisition section (500) to a learning model (10A) that has been generated by machine learning that causes the learning model (10A) to learn a correlation between the operating condition information and the polishing-pad state information.