CMP Polishing Rate and Torque Modeling With Physics-Based Identification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing chemical mechanical polishing (CMP) methods for semiconductor wafers require large amounts of training data and time to create accurate models, and the black-box nature of neural networks makes it difficult to identify and correct errors, leading to inefficient and lengthy model development and polishing processes.
Innovation Solution
A chemical-mechanical-polishing system and method that utilizes a physical model to estimate polishing rates and torques by identifying model parameters using measured polishing physical quantities, allowing for the calculation of accurate polishing rates and torques through a simulation model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning models with neural networks are used for polishing simulation, then polishing estimation accuracy is improved, but model creation time and data requirements increase significantly
Solution Approach 1:
The patent replaces the machine learning/neural network approach with a physics-based simulation model that uses fundamental mechanical and chemical principles of CMP processes. This substitution eliminates the need for extensive training data and long model creation periods while maintaining estimation accuracy through direct physical modeling of polishing mechanics.
Solution Approach 2:
The patent changes the fundamental parameters of the simulation approach from data-driven statistical parameters (in machine learning) to physics-based parameters (material removal rates, abrasive grain characteristics, slurry composition, contact pressure, and relative velocity). This parameter transformation enables accurate polishing estimation without requiring large datasets for training.
2Measurement precision
If complex neural network models are used for polishing simulation, then polishing estimation accuracy is improved, but model correction becomes difficult due to black box nature
Solution Approach 1:
The patent replaces the opaque neural network structure with a transparent physics-based model where all calculations are based on explicit physical equations and measurable parameters. This substitution makes the model fully interpretable and correctable, as each calculation step can be traced back to physical principles and adjusted based on actual process conditions.
3Measurement precision
If more training data is collected to improve model accuracy, then polishing estimation precision is improved, but time and resource requirements increase
Solution Approach 1:
The patent replaces the data-intensive machine learning approach with a physics-based model that requires minimal input data (basic process parameters like pressure, velocity, and material properties) to generate accurate predictions. This substitution eliminates the need for collecting and processing large volumes of training data while maintaining or improving estimation precision.
Data Source
AI summary
A chemical-mechanical-polishing system for polishing a workpiece, such as a wafer, while calculating an estimated polishing rate of the workpiece using a physical model is disclosed. The chemical-mechanical-polishing system is configured to acquire measured polishing physical quantities including a measured polishing rate of a first workpiece and a measured value of torque of a polishing apparatus during or after polishing of the first workpiece; identify the model parameters of the simulation model using the measured polishing physical quantities as variables for identification; and input polishing conditions for a second workpiece into the simulation model to calculate an estimated polishing rate of the second workpiece.