Scatterometry-Based APC Using Deep Learning for Knob Tuning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional advanced process control (APC) methods in semiconductor manufacturing face challenges in achieving high accuracy due to increased variability in processing steps and multiple manufacturing routes, which traditional methods struggle to address effectively, especially in high dimensional knob spaces.
Innovation Solution
A machine learning-based APC system that uses scatterometric training data to recommend changes in process control knob settings by training a neural network to correlate pre-process and post-process data, allowing for optimized process control without relying on expensive reference parameters.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If traditional advanced process control (APC) methods are used, then process control is maintained, but manufacturing precision and accuracy deteriorate due to increased variability in processing steps and multiple manufacturing routes
Solution Approach 1:
The patent replaces traditional mechanical/statistical APC methods with a machine learning system that uses scatterometric data and neural networks to predict optimal process control knob settings. This substitution enables the system to handle high-dimensional knob spaces and multiple manufacturing routes that traditional methods cannot effectively manage, thereby improving manufacturing precision while maintaining reliability
Solution Approach 2:
The patent transforms the APC approach by changing from fixed traditional control parameters to dynamic machine learning-derived parameters. The system uses scatterometric measurements and trained neural networks to continuously adapt process control settings, allowing it to compensate for variability across different processing steps and manufacturing routes, thus improving both precision and reliability
2Manufacturing precision
If traditional APC methods are applied to high dimensional knob spaces, then process control is attempted, but manufacturing precision deteriorates due to the complexity of the parameter space
Solution Approach 1:
The patent substitutes traditional APC algorithms with machine learning models including neural networks and random forests. These ML systems are inherently capable of handling high-dimensional input spaces and automatically identifying relevant patterns without requiring explicit dimensionality reduction or feature selection, thus maintaining manufacturing precision even in complex knob spaces with many interdependent parameters
Solution Approach 2:
The patent creates virtual models through machine learning that replicate the complex relationships between process parameters and outcomes. Instead of directly managing the complexity of high-dimensional knob spaces, the system trains ML models on historical data to create simplified surrogate models that capture essential relationships, enabling precise control without being overwhelmed by the apparent complexity
3Manufacturing precision
If machine learning techniques are applied to improve APC, then manufacturing precision can be improved, but the system requires extensive training data and computational resources
Solution Approach 1:
The patent designs a multi-functional machine learning system where scatterometric measurements serve multiple purposes: they are used both for their primary metrology function and simultaneously as training data for the ML models. This universal use of measurement data maximizes the value extracted from each measurement, reducing the total quantity of training data needed while improving manufacturing precision through the ML-enhanced APC system
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
The system reduces variations in pattern parameters across wafer sites by providing accurate recommendations for process control knob settings, improving manufacturing consistency and reducing variability, thereby enhancing the precision of semiconductor production.
Implementation Method 1
Scatterometric data (also referred to herein as 'spectra data') is typically acquired as reflected light radiation that is indicative of optical properties of patterns at wafer sites
Implementation Method 2
The multiple processing steps in semiconductor manufacturing generate stacked structures ('stacks'), which, like diffraction gratings, have optical properties
Data Source
AI summary
A system and methods for Advance Process Control (APC) in semiconductor manufacturing include: for each of a plurality of waiter sites, receiving a pre-process set of scatterometric training data, measured before implementation of a processing step, receiving a corresponding post-process set of scatterometric training data measured after implementation of the process step, and receiving a set of process control knob training data indicative of process control knob settings applied during implementation of the process step; and generating a machine learning model correlating variations in the pre-process sets of scatterometric training data and the corresponding process control knob training data with the corresponding post-process sets of scatterometric training data, to train the machine learning model to recommend changes to process control knob settings to compensate for variations in the pre-process scatterometric data.