Scatterometry Metrology Modeling for Robust Wafer Process Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current semiconductor metrology methods face challenges in achieving accuracy, process robustness, precision, and throughput due to limitations in metrology tool specifications, within-wafer sampling rates, and the difficulty in obtaining external reference data for model-based metrology solutions.
Innovation Solution
The implementation of machine learning methods using supervised learning to establish models for predicting metrology parameters based on optical signals, incorporating spectral variability and noise terms to improve prediction accuracy and robustness, and integrating these models into semiconductor manufacturing processes for real-time control.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If hardware improvements are made to optimize move and measure time and tool stability, then throughput and sampling rates are improved, but device complexity and cost increase
Solution Approach 1:
The patent replaces hardware-based metrology improvements with a software-based machine learning model. Instead of upgrading physical metrology tools to achieve better accuracy and throughput, the system uses ML algorithms to predict critical dimensions from existing measurement data, thereby avoiding the need for more complex and expensive hardware while maintaining or improving productivity
Solution Approach 2:
The patent introduces a machine learning model as an intermediary between the existing metrology tool and the process control system. This ML intermediary processes measurement data to extract accurate critical dimension information without requiring direct hardware upgrades, thus improving throughput without proportionally increasing device complexity
2Measurement precision
If machine learning models are used to predict metrology parameters, then accuracy and repeatability are improved, but computational complexity and training data requirements increase
Solution Approach 1:
The patent performs the computationally intensive model training in advance during a setup phase, separating the complex computational work from the actual production measurement process. Once trained, the model can quickly make predictions during manufacturing without requiring continuous heavy computation, thus improving measurement accuracy while managing computational complexity through time-based separation
Solution Approach 2:
The patent uses a supervised learning approach that leverages existing measurement data and reference measurements to train the model, rather than requiring complete first-principles simulations. This partial action approach uses available data efficiently to achieve good accuracy without the excessive computational burden of more comprehensive modeling approaches
3Measurement precision
If external reference data is obtained for model training, then model accuracy is improved, but data acquisition difficulty and time increase
Solution Approach 1:
The patent enables the system to generate its own training data using existing metrology measurements and process data from the semiconductor manufacturing environment. Instead of relying on external reference data that would require additional time to acquire, the system self-serves by utilizing internally available measurement data to train and validate the machine learning model, thereby improving model accuracy without the time cost of external data collection
Data Source
AI summary
A semiconductor metrology system including a spectrum acquisition tool for collecting, using a first measurement protocol, baseline scatterometric spectra on first semiconductor wafer targets, and for various sources of spectral variability, variability sets of scatterometric spectra on second semiconductor wafer targets, the variability sets embodying the spectral variability, a reference metrology tool for collecting, using a second measurement protocol, parameter values of the first semiconductor wafer targets, and a training unit for training, using the collected spectra and values, a prediction model using machine learning and minimizing an associated loss function incorporating spectral variability terms, the prediction model for predicting values for production semiconductor wafer targets based on their spectra.