ML Substrate Profile Prediction via Spectral Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing methods for determining substrate profile properties in manufacturing systems require substrates to be removed for separate metrology measurements, leading to costly inefficiencies and low sampling rates, resulting in potential defects and equipment damage.
Innovation Solution
A machine learning model is trained using spectral and non-spectral data from prior substrates to predict metrology measurements for current substrates within the manufacturing system, eliminating the need for substrate removal and enabling high sampling rates.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If substrates are removed from the manufacturing tool for separate metrology measurements, then measurement accuracy is improved, but productivity decreases and process efficiency is reduced
Solution Approach 1:
The patent combines the metrology measurement function with the manufacturing tool by integrating a metrology system within the manufacturing tool, allowing substrates to be measured without removal. This merging of functions eliminates the need for separate measurement operations, thereby maintaining measurement accuracy while improving productivity and process efficiency.
Solution Approach 2:
The patent introduces an intermediary machine learning model that predicts substrate profile properties from spectral data. This intermediary approach allows the system to obtain accurate metrology information without physically removing substrates for measurement, thus resolving the contradiction between measurement precision and productivity.
2Loss of information
If substrates are removed for metrology measurements, then substrate profile properties can be evaluated, but the sampling rate decreases and process decisions are based on limited data
Solution Approach 1:
The patent enables continuous metrology measurements by keeping substrates in the manufacturing tool throughout the processing sequence. The integrated metrology system continuously collects spectral data from substrates without interruption, ensuring no loss of substrate profile information while maintaining high sampling rates across all processed substrates.
Solution Approach 2:
The patent replaces the mechanical substrate removal and physical measurement process with an optical spectral analysis system combined with machine learning algorithms. This substitution allows for non-contact, in-situ measurements that capture substrate profile properties without interrupting the manufacturing flow, thereby maximizing data availability and sampling rate.
3Loss of energy
If few substrates are measured due to removal costs, then measurement costs are reduced, but process decisions based on limited measurements can result in substrate defects and equipment damage
Solution Approach 1:
The patent enables the manufacturing tool to perform its own metrology measurements through the integrated system. By measuring all substrates in-situ without requiring external metrology equipment or substrate removal, the system eliminates the costs associated with substrate handling while simultaneously improving the reliability of process decisions through comprehensive data collection from every substrate.
Data Source
AI summary
Spectral data associated with a first prior substrate and/or a second prior substrate is obtained. A metrology measurement value associated with the first portion of the first prior substrate is determined based on one or more metrology measurement values measured for at least one of a second portion of the first prior substrate or a third portion of a second prior substrate. Training data for training a machine learning model to predict metrology measurement values of a current substrate is generated. Generating the training data includes generating a first training input including the spectral data associated with the first prior substrate and generating a first target output for the first training input, the first target output including the determined metrology measurement value associated with the first portion of the first prior substrate. The training data is provided to train the machine learning model.